quick_entrypoints_arm64.S revision 40df7c16f591e5ec9a2affe711212fa4855a94a4
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(kSaveAllCalleeSaves) 25 */ 26.macro SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 27 // art::Runtime** xIP0 = &art::Runtime::instance_ 28 adrp xIP0, :got:_ZN3art7Runtime9instance_E 29 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 30 31 // Our registers aren't intermixed - just spill in order. 32 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 33 34 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveAllCalleeSaves]; 35 ldr xIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVES_METHOD_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_SAVES != 176) 42#error "FRAME_SIZE_SAVE_ALL_CALLEE_SAVES(ARM64) size not as expected." 43#endif 44 45 // Stack alignment filler [sp, #8]. 46 // FP callee-saves. 47 stp d8, d9, [sp, #16] 48 stp d10, d11, [sp, #32] 49 stp d12, d13, [sp, #48] 50 stp d14, d15, [sp, #64] 51 52 // GP callee-saves 53 stp x19, x20, [sp, #80] 54 .cfi_rel_offset x19, 80 55 .cfi_rel_offset x20, 88 56 57 stp x21, x22, [sp, #96] 58 .cfi_rel_offset x21, 96 59 .cfi_rel_offset x22, 104 60 61 stp x23, x24, [sp, #112] 62 .cfi_rel_offset x23, 112 63 .cfi_rel_offset x24, 120 64 65 stp x25, x26, [sp, #128] 66 .cfi_rel_offset x25, 128 67 .cfi_rel_offset x26, 136 68 69 stp x27, x28, [sp, #144] 70 .cfi_rel_offset x27, 144 71 .cfi_rel_offset x28, 152 72 73 stp x29, xLR, [sp, #160] 74 .cfi_rel_offset x29, 160 75 .cfi_rel_offset x30, 168 76 77 // Store ArtMethod* Runtime::callee_save_methods_[kSaveAllCalleeSaves]. 78 str xIP0, [sp] 79 // Place sp in Thread::Current()->top_quick_frame. 80 mov xIP0, sp 81 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 82.endm 83 84 /* 85 * Macro that sets up the callee save frame to conform with 86 * Runtime::CreateCalleeSaveMethod(kSaveRefsOnly). 87 */ 88.macro SETUP_SAVE_REFS_ONLY_FRAME 89 // art::Runtime** xIP0 = &art::Runtime::instance_ 90 adrp xIP0, :got:_ZN3art7Runtime9instance_E 91 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 92 93 // Our registers aren't intermixed - just spill in order. 94 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 95 96 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveRefOnly]; 97 ldr xIP0, [xIP0, RUNTIME_SAVE_REFS_ONLY_METHOD_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_SAVE_REFS_ONLY != 96) 104#error "FRAME_SIZE_SAVE_REFS_ONLY(ARM64) size not as expected." 105#endif 106 107 // GP callee-saves. 108 // x20 paired with ArtMethod* - see below. 109 stp x21, x22, [sp, #16] 110 .cfi_rel_offset x21, 16 111 .cfi_rel_offset x22, 24 112 113 stp x23, x24, [sp, #32] 114 .cfi_rel_offset x23, 32 115 .cfi_rel_offset x24, 40 116 117 stp x25, x26, [sp, #48] 118 .cfi_rel_offset x25, 48 119 .cfi_rel_offset x26, 56 120 121 stp x27, x28, [sp, #64] 122 .cfi_rel_offset x27, 64 123 .cfi_rel_offset x28, 72 124 125 stp x29, xLR, [sp, #80] 126 .cfi_rel_offset x29, 80 127 .cfi_rel_offset x30, 88 128 129 // Store ArtMethod* Runtime::callee_save_methods_[kSaveRefsOnly]. 130 stp xIP0, x20, [sp] 131 .cfi_rel_offset x20, 8 132 133 // Place sp in Thread::Current()->top_quick_frame. 134 mov xIP0, sp 135 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 136.endm 137 138// TODO: Probably no need to restore registers preserved by aapcs64. 139.macro RESTORE_SAVE_REFS_ONLY_FRAME 140 // Callee-saves. 141 ldr x20, [sp, #8] 142 .cfi_restore x20 143 144 ldp x21, x22, [sp, #16] 145 .cfi_restore x21 146 .cfi_restore x22 147 148 ldp x23, x24, [sp, #32] 149 .cfi_restore x23 150 .cfi_restore x24 151 152 ldp x25, x26, [sp, #48] 153 .cfi_restore x25 154 .cfi_restore x26 155 156 ldp x27, x28, [sp, #64] 157 .cfi_restore x27 158 .cfi_restore x28 159 160 ldp x29, xLR, [sp, #80] 161 .cfi_restore x29 162 .cfi_restore x30 163 164 add sp, sp, #96 165 .cfi_adjust_cfa_offset -96 166.endm 167 168.macro POP_SAVE_REFS_ONLY_FRAME 169 add sp, sp, #96 170 .cfi_adjust_cfa_offset - 96 171.endm 172 173.macro RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN 174 RESTORE_SAVE_REFS_ONLY_FRAME 175 ret 176.endm 177 178 179.macro SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 180 sub sp, sp, #224 181 .cfi_adjust_cfa_offset 224 182 183 // Ugly compile-time check, but we only have the preprocessor. 184#if (FRAME_SIZE_SAVE_REFS_AND_ARGS != 224) 185#error "FRAME_SIZE_SAVE_REFS_AND_ARGS(ARM64) size not as expected." 186#endif 187 188 // Stack alignment filler [sp, #8]. 189 // FP args. 190 stp d0, d1, [sp, #16] 191 stp d2, d3, [sp, #32] 192 stp d4, d5, [sp, #48] 193 stp d6, d7, [sp, #64] 194 195 // Core args. 196 stp x1, x2, [sp, #80] 197 .cfi_rel_offset x1, 80 198 .cfi_rel_offset x2, 88 199 200 stp x3, x4, [sp, #96] 201 .cfi_rel_offset x3, 96 202 .cfi_rel_offset x4, 104 203 204 stp x5, x6, [sp, #112] 205 .cfi_rel_offset x5, 112 206 .cfi_rel_offset x6, 120 207 208 // x7, Callee-saves. 209 stp x7, x20, [sp, #128] 210 .cfi_rel_offset x7, 128 211 .cfi_rel_offset x20, 136 212 213 stp x21, x22, [sp, #144] 214 .cfi_rel_offset x21, 144 215 .cfi_rel_offset x22, 152 216 217 stp x23, x24, [sp, #160] 218 .cfi_rel_offset x23, 160 219 .cfi_rel_offset x24, 168 220 221 stp x25, x26, [sp, #176] 222 .cfi_rel_offset x25, 176 223 .cfi_rel_offset x26, 184 224 225 stp x27, x28, [sp, #192] 226 .cfi_rel_offset x27, 192 227 .cfi_rel_offset x28, 200 228 229 // x29(callee-save) and LR. 230 stp x29, xLR, [sp, #208] 231 .cfi_rel_offset x29, 208 232 .cfi_rel_offset x30, 216 233 234.endm 235 236 /* 237 * Macro that sets up the callee save frame to conform with 238 * Runtime::CreateCalleeSaveMethod(kSaveRefsAndArgs). 239 * 240 * TODO This is probably too conservative - saving FP & LR. 241 */ 242.macro SETUP_SAVE_REFS_AND_ARGS_FRAME 243 // art::Runtime** xIP0 = &art::Runtime::instance_ 244 adrp xIP0, :got:_ZN3art7Runtime9instance_E 245 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 246 247 // Our registers aren't intermixed - just spill in order. 248 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 249 250 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveRefAndArgs]; 251 ldr xIP0, [xIP0, RUNTIME_SAVE_REFS_AND_ARGS_METHOD_OFFSET] 252 253 SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 254 255 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kSaveRefsAndArgs]. 256 // Place sp in Thread::Current()->top_quick_frame. 257 mov xIP0, sp 258 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 259.endm 260 261.macro SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 262 SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 263 str x0, [sp, #0] // Store ArtMethod* to bottom of stack. 264 // Place sp in Thread::Current()->top_quick_frame. 265 mov xIP0, sp 266 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 267.endm 268 269// TODO: Probably no need to restore registers preserved by aapcs64. 270.macro RESTORE_SAVE_REFS_AND_ARGS_FRAME 271 // FP args. 272 ldp d0, d1, [sp, #16] 273 ldp d2, d3, [sp, #32] 274 ldp d4, d5, [sp, #48] 275 ldp d6, d7, [sp, #64] 276 277 // Core args. 278 ldp x1, x2, [sp, #80] 279 .cfi_restore x1 280 .cfi_restore x2 281 282 ldp x3, x4, [sp, #96] 283 .cfi_restore x3 284 .cfi_restore x4 285 286 ldp x5, x6, [sp, #112] 287 .cfi_restore x5 288 .cfi_restore x6 289 290 // x7, Callee-saves. 291 ldp x7, x20, [sp, #128] 292 .cfi_restore x7 293 .cfi_restore x20 294 295 ldp x21, x22, [sp, #144] 296 .cfi_restore x21 297 .cfi_restore x22 298 299 ldp x23, x24, [sp, #160] 300 .cfi_restore x23 301 .cfi_restore x24 302 303 ldp x25, x26, [sp, #176] 304 .cfi_restore x25 305 .cfi_restore x26 306 307 ldp x27, x28, [sp, #192] 308 .cfi_restore x27 309 .cfi_restore x28 310 311 // x29(callee-save) and LR. 312 ldp x29, xLR, [sp, #208] 313 .cfi_restore x29 314 .cfi_restore x30 315 316 add sp, sp, #224 317 .cfi_adjust_cfa_offset -224 318.endm 319 320 /* 321 * Macro that sets up the callee save frame to conform with 322 * Runtime::CreateCalleeSaveMethod(kSaveEverything) 323 */ 324.macro SETUP_SAVE_EVERYTHING_FRAME 325 sub sp, sp, #512 326 .cfi_adjust_cfa_offset 512 327 328 // Ugly compile-time check, but we only have the preprocessor. 329#if (FRAME_SIZE_SAVE_EVERYTHING != 512) 330#error "FRAME_SIZE_SAVE_EVERYTHING(ARM64) size not as expected." 331#endif 332 333 // Save FP registers. 334 // For better performance, store d0 and d31 separately, so that all STPs are 16-byte aligned. 335 str d0, [sp, #8] 336 stp d1, d2, [sp, #16] 337 stp d3, d4, [sp, #32] 338 stp d5, d6, [sp, #48] 339 stp d7, d8, [sp, #64] 340 stp d9, d10, [sp, #80] 341 stp d11, d12, [sp, #96] 342 stp d13, d14, [sp, #112] 343 stp d15, d16, [sp, #128] 344 stp d17, d18, [sp, #144] 345 stp d19, d20, [sp, #160] 346 stp d21, d22, [sp, #176] 347 stp d23, d24, [sp, #192] 348 stp d25, d26, [sp, #208] 349 stp d27, d28, [sp, #224] 350 stp d29, d30, [sp, #240] 351 str d31, [sp, #256] 352 353 // Save core registers. 354 str x0, [sp, #264] 355 .cfi_rel_offset x0, 264 356 357 stp x1, x2, [sp, #272] 358 .cfi_rel_offset x1, 272 359 .cfi_rel_offset x2, 280 360 361 stp x3, x4, [sp, #288] 362 .cfi_rel_offset x3, 288 363 .cfi_rel_offset x4, 296 364 365 stp x5, x6, [sp, #304] 366 .cfi_rel_offset x5, 304 367 .cfi_rel_offset x6, 312 368 369 stp x7, x8, [sp, #320] 370 .cfi_rel_offset x7, 320 371 .cfi_rel_offset x8, 328 372 373 stp x9, x10, [sp, #336] 374 .cfi_rel_offset x9, 336 375 .cfi_rel_offset x10, 344 376 377 stp x11, x12, [sp, #352] 378 .cfi_rel_offset x11, 352 379 .cfi_rel_offset x12, 360 380 381 stp x13, x14, [sp, #368] 382 .cfi_rel_offset x13, 368 383 .cfi_rel_offset x14, 376 384 385 stp x15, x16, [sp, #384] 386 .cfi_rel_offset x15, 384 387 .cfi_rel_offset x16, 392 388 389 stp x17, x18, [sp, #400] 390 .cfi_rel_offset x17, 400 391 .cfi_rel_offset x18, 408 392 393 stp x19, x20, [sp, #416] 394 .cfi_rel_offset x19, 416 395 .cfi_rel_offset x20, 424 396 397 stp x21, x22, [sp, #432] 398 .cfi_rel_offset x21, 432 399 .cfi_rel_offset x22, 440 400 401 stp x23, x24, [sp, #448] 402 .cfi_rel_offset x23, 448 403 .cfi_rel_offset x24, 456 404 405 stp x25, x26, [sp, #464] 406 .cfi_rel_offset x25, 464 407 .cfi_rel_offset x26, 472 408 409 stp x27, x28, [sp, #480] 410 .cfi_rel_offset x27, 480 411 .cfi_rel_offset x28, 488 412 413 stp x29, xLR, [sp, #496] 414 .cfi_rel_offset x29, 496 415 .cfi_rel_offset x30, 504 416 417 // art::Runtime** xIP0 = &art::Runtime::instance_ 418 adrp xIP0, :got:_ZN3art7Runtime9instance_E 419 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 420 421 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 422 423 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveEverything]; 424 ldr xIP0, [xIP0, RUNTIME_SAVE_EVERYTHING_METHOD_OFFSET] 425 426 // Store ArtMethod* Runtime::callee_save_methods_[kSaveEverything]. 427 str xIP0, [sp] 428 // Place sp in Thread::Current()->top_quick_frame. 429 mov xIP0, sp 430 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 431.endm 432 433.macro RESTORE_SAVE_EVERYTHING_FRAME 434 // Restore FP registers. 435 // For better performance, load d0 and d31 separately, so that all LDPs are 16-byte aligned. 436 ldr d0, [sp, #8] 437 ldp d1, d2, [sp, #16] 438 ldp d3, d4, [sp, #32] 439 ldp d5, d6, [sp, #48] 440 ldp d7, d8, [sp, #64] 441 ldp d9, d10, [sp, #80] 442 ldp d11, d12, [sp, #96] 443 ldp d13, d14, [sp, #112] 444 ldp d15, d16, [sp, #128] 445 ldp d17, d18, [sp, #144] 446 ldp d19, d20, [sp, #160] 447 ldp d21, d22, [sp, #176] 448 ldp d23, d24, [sp, #192] 449 ldp d25, d26, [sp, #208] 450 ldp d27, d28, [sp, #224] 451 ldp d29, d30, [sp, #240] 452 ldr d31, [sp, #256] 453 454 // Restore core registers. 455 ldr x0, [sp, #264] 456 .cfi_restore x0 457 458 ldp x1, x2, [sp, #272] 459 .cfi_restore x1 460 .cfi_restore x2 461 462 ldp x3, x4, [sp, #288] 463 .cfi_restore x3 464 .cfi_restore x4 465 466 ldp x5, x6, [sp, #304] 467 .cfi_restore x5 468 .cfi_restore x6 469 470 ldp x7, x8, [sp, #320] 471 .cfi_restore x7 472 .cfi_restore x8 473 474 ldp x9, x10, [sp, #336] 475 .cfi_restore x9 476 .cfi_restore x10 477 478 ldp x11, x12, [sp, #352] 479 .cfi_restore x11 480 .cfi_restore x12 481 482 ldp x13, x14, [sp, #368] 483 .cfi_restore x13 484 .cfi_restore x14 485 486 ldp x15, x16, [sp, #384] 487 .cfi_restore x15 488 .cfi_restore x16 489 490 ldp x17, x18, [sp, #400] 491 .cfi_restore x17 492 .cfi_restore x18 493 494 ldp x19, x20, [sp, #416] 495 .cfi_restore x19 496 .cfi_restore x20 497 498 ldp x21, x22, [sp, #432] 499 .cfi_restore x21 500 .cfi_restore x22 501 502 ldp x23, x24, [sp, #448] 503 .cfi_restore x23 504 .cfi_restore x24 505 506 ldp x25, x26, [sp, #464] 507 .cfi_restore x25 508 .cfi_restore x26 509 510 ldp x27, x28, [sp, #480] 511 .cfi_restore x27 512 .cfi_restore x28 513 514 ldp x29, xLR, [sp, #496] 515 .cfi_restore x29 516 .cfi_restore x30 517 518 add sp, sp, #512 519 .cfi_adjust_cfa_offset -512 520.endm 521 522.macro RETURN_IF_RESULT_IS_ZERO 523 cbnz x0, 1f // result non-zero branch over 524 ret // return 5251: 526.endm 527 528.macro RETURN_IF_RESULT_IS_NON_ZERO 529 cbz x0, 1f // result zero branch over 530 ret // return 5311: 532.endm 533 534 /* 535 * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending 536 * exception is Thread::Current()->exception_ 537 */ 538.macro DELIVER_PENDING_EXCEPTION 539 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 540 mov x0, xSELF 541 542 // Point of no return. 543 b artDeliverPendingExceptionFromCode // artDeliverPendingExceptionFromCode(Thread*) 544 brk 0 // Unreached 545.endm 546 547.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg 548 ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET] // Get exception field. 549 cbnz \reg, 1f 550 ret 5511: 552 DELIVER_PENDING_EXCEPTION 553.endm 554 555.macro RETURN_OR_DELIVER_PENDING_EXCEPTION 556 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0 557.endm 558 559// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register. 560.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 561 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1 562.endm 563 564.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER 565 cbnz w0, 1f // result non-zero branch over 566 ret // return 5671: 568 DELIVER_PENDING_EXCEPTION 569.endm 570 571.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 572 .extern \cxx_name 573ENTRY \c_name 574 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context 575 mov x0, xSELF // pass Thread::Current 576 b \cxx_name // \cxx_name(Thread*) 577END \c_name 578.endm 579 580.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name 581 .extern \cxx_name 582ENTRY \c_name 583 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context. 584 mov x1, xSELF // pass Thread::Current. 585 b \cxx_name // \cxx_name(arg, Thread*). 586 brk 0 587END \c_name 588.endm 589 590.macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 591 .extern \cxx_name 592ENTRY \c_name 593 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context 594 mov x2, xSELF // pass Thread::Current 595 b \cxx_name // \cxx_name(arg1, arg2, Thread*) 596 brk 0 597END \c_name 598.endm 599 600 /* 601 * Called by managed code, saves callee saves and then calls artThrowException 602 * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. 603 */ 604ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode 605 606 /* 607 * Called by managed code to create and deliver a NullPointerException. 608 */ 609NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode 610 611 /* 612 * Call installed by a signal handler to create and deliver a NullPointerException. 613 */ 614ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception_from_signal, artThrowNullPointerExceptionFromSignal 615 616 /* 617 * Called by managed code to create and deliver an ArithmeticException. 618 */ 619NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode 620 621 /* 622 * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds 623 * index, arg2 holds limit. 624 */ 625TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode 626 627 /* 628 * Called by managed code to create and deliver a StringIndexOutOfBoundsException 629 * as if thrown from a call to String.charAt(). Arg1 holds index, arg2 holds limit. 630 */ 631TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_string_bounds, artThrowStringBoundsFromCode 632 633 /* 634 * Called by managed code to create and deliver a StackOverflowError. 635 */ 636NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode 637 638 /* 639 * Called by managed code to create and deliver a NoSuchMethodError. 640 */ 641ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode 642 643 /* 644 * All generated callsites for interface invokes and invocation slow paths will load arguments 645 * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain 646 * the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper. 647 * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1. 648 * 649 * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting 650 * of the target Method* in x0 and method->code_ in x1. 651 * 652 * If unsuccessful, the helper will return null/????. There will be a pending exception in the 653 * thread and we branch to another stub to deliver it. 654 * 655 * On success this wrapper will restore arguments and *jump* to the target, leaving the lr 656 * pointing back to the original caller. 657 * 658 * Adapted from ARM32 code. 659 * 660 * Clobbers xIP0. 661 */ 662.macro INVOKE_TRAMPOLINE_BODY cxx_name 663 .extern \cxx_name 664 SETUP_SAVE_REFS_AND_ARGS_FRAME // save callee saves in case allocation triggers GC 665 // Helper signature is always 666 // (method_idx, *this_object, *caller_method, *self, sp) 667 668 mov x2, xSELF // pass Thread::Current 669 mov x3, sp 670 bl \cxx_name // (method_idx, this, Thread*, SP) 671 mov xIP0, x1 // save Method*->code_ 672 RESTORE_SAVE_REFS_AND_ARGS_FRAME 673 cbz x0, 1f // did we find the target? if not go to exception delivery 674 br xIP0 // tail call to target 6751: 676 DELIVER_PENDING_EXCEPTION 677.endm 678.macro INVOKE_TRAMPOLINE c_name, cxx_name 679ENTRY \c_name 680 INVOKE_TRAMPOLINE_BODY \cxx_name 681END \c_name 682.endm 683 684INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck 685 686INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck 687INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck 688INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck 689INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck 690 691 692.macro INVOKE_STUB_CREATE_FRAME 693 694SAVE_SIZE=15*8 // x4, x5, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 695SAVE_SIZE_AND_METHOD=SAVE_SIZE+8 696 697 698 mov x9, sp // Save stack pointer. 699 .cfi_register sp,x9 700 701 add x10, x2, # SAVE_SIZE_AND_METHOD // calculate size of frame. 702 sub x10, sp, x10 // Calculate SP position - saves + ArtMethod* + args 703 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 704 mov sp, x10 // Set new SP. 705 706 sub x10, x9, #SAVE_SIZE // Calculate new FP (later). Done here as we must move SP 707 .cfi_def_cfa_register x10 // before this. 708 .cfi_adjust_cfa_offset SAVE_SIZE 709 710 str x28, [x10, #112] 711 .cfi_rel_offset x28, 112 712 713 stp x26, x27, [x10, #96] 714 .cfi_rel_offset x26, 96 715 .cfi_rel_offset x27, 104 716 717 stp x24, x25, [x10, #80] 718 .cfi_rel_offset x24, 80 719 .cfi_rel_offset x25, 88 720 721 stp x22, x23, [x10, #64] 722 .cfi_rel_offset x22, 64 723 .cfi_rel_offset x23, 72 724 725 stp x20, x21, [x10, #48] 726 .cfi_rel_offset x20, 48 727 .cfi_rel_offset x21, 56 728 729 stp x9, x19, [x10, #32] // Save old stack pointer and x19. 730 .cfi_rel_offset sp, 32 731 .cfi_rel_offset x19, 40 732 733 stp x4, x5, [x10, #16] // Save result and shorty addresses. 734 .cfi_rel_offset x4, 16 735 .cfi_rel_offset x5, 24 736 737 stp xFP, xLR, [x10] // Store LR & FP. 738 .cfi_rel_offset x29, 0 739 .cfi_rel_offset x30, 8 740 741 mov xFP, x10 // Use xFP now, as it's callee-saved. 742 .cfi_def_cfa_register x29 743 mov xSELF, x3 // Move thread pointer into SELF register. 744 745 // Copy arguments into stack frame. 746 // Use simple copy routine for now. 747 // 4 bytes per slot. 748 // X1 - source address 749 // W2 - args length 750 // X9 - destination address. 751 // W10 - temporary 752 add x9, sp, #8 // Destination address is bottom of stack + null. 753 754 // Copy parameters into the stack. Use numeric label as this is a macro and Clang's assembler 755 // does not have unique-id variables. 7561: 757 cmp w2, #0 758 beq 2f 759 sub w2, w2, #4 // Need 65536 bytes of range. 760 ldr w10, [x1, x2] 761 str w10, [x9, x2] 762 763 b 1b 764 7652: 766 // Store null into ArtMethod* at bottom of frame. 767 str xzr, [sp] 768.endm 769 770.macro INVOKE_STUB_CALL_AND_RETURN 771 772 // load method-> METHOD_QUICK_CODE_OFFSET 773 ldr x9, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 774 // Branch to method. 775 blr x9 776 777 // Restore return value address and shorty address. 778 ldp x4,x5, [xFP, #16] 779 .cfi_restore x4 780 .cfi_restore x5 781 782 ldr x28, [xFP, #112] 783 .cfi_restore x28 784 785 ldp x26, x27, [xFP, #96] 786 .cfi_restore x26 787 .cfi_restore x27 788 789 ldp x24, x25, [xFP, #80] 790 .cfi_restore x24 791 .cfi_restore x25 792 793 ldp x22, x23, [xFP, #64] 794 .cfi_restore x22 795 .cfi_restore x23 796 797 ldp x20, x21, [xFP, #48] 798 .cfi_restore x20 799 .cfi_restore x21 800 801 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 802 ldrb w10, [x5] 803 804 // Check the return type and store the correct register into the jvalue in memory. 805 // Use numeric label as this is a macro and Clang's assembler does not have unique-id variables. 806 807 // Don't set anything for a void type. 808 cmp w10, #'V' 809 beq 3f 810 811 // Is it a double? 812 cmp w10, #'D' 813 bne 1f 814 str d0, [x4] 815 b 3f 816 8171: // Is it a float? 818 cmp w10, #'F' 819 bne 2f 820 str s0, [x4] 821 b 3f 822 8232: // Just store x0. Doesn't matter if it is 64 or 32 bits. 824 str x0, [x4] 825 8263: // Finish up. 827 ldp x2, x19, [xFP, #32] // Restore stack pointer and x19. 828 .cfi_restore x19 829 mov sp, x2 830 .cfi_restore sp 831 832 ldp xFP, xLR, [xFP] // Restore old frame pointer and link register. 833 .cfi_restore x29 834 .cfi_restore x30 835 836 ret 837 838.endm 839 840 841/* 842 * extern"C" void art_quick_invoke_stub(ArtMethod *method, x0 843 * uint32_t *args, x1 844 * uint32_t argsize, w2 845 * Thread *self, x3 846 * JValue *result, x4 847 * char *shorty); x5 848 * +----------------------+ 849 * | | 850 * | C/C++ frame | 851 * | LR'' | 852 * | FP'' | <- SP' 853 * +----------------------+ 854 * +----------------------+ 855 * | x28 | <- TODO: Remove callee-saves. 856 * | : | 857 * | x19 | 858 * | SP' | 859 * | X5 | 860 * | X4 | Saved registers 861 * | LR' | 862 * | FP' | <- FP 863 * +----------------------+ 864 * | uint32_t out[n-1] | 865 * | : : | Outs 866 * | uint32_t out[0] | 867 * | ArtMethod* | <- SP value=null 868 * +----------------------+ 869 * 870 * Outgoing registers: 871 * x0 - Method* 872 * x1-x7 - integer parameters. 873 * d0-d7 - Floating point parameters. 874 * xSELF = self 875 * SP = & of ArtMethod* 876 * x1 = "this" pointer. 877 * 878 */ 879ENTRY art_quick_invoke_stub 880 // Spill registers as per AACPS64 calling convention. 881 INVOKE_STUB_CREATE_FRAME 882 883 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 884 // Parse the passed shorty to determine which register to load. 885 // Load addresses for routines that load WXSD registers. 886 adr x11, .LstoreW2 887 adr x12, .LstoreX2 888 adr x13, .LstoreS0 889 adr x14, .LstoreD0 890 891 // Initialize routine offsets to 0 for integers and floats. 892 // x8 for integers, x15 for floating point. 893 mov x8, #0 894 mov x15, #0 895 896 add x10, x5, #1 // Load shorty address, plus one to skip return value. 897 ldr w1, [x9],#4 // Load "this" parameter, and increment arg pointer. 898 899 // Loop to fill registers. 900.LfillRegisters: 901 ldrb w17, [x10], #1 // Load next character in signature, and increment. 902 cbz w17, .LcallFunction // Exit at end of signature. Shorty 0 terminated. 903 904 cmp w17, #'F' // is this a float? 905 bne .LisDouble 906 907 cmp x15, # 8*12 // Skip this load if all registers full. 908 beq .Ladvance4 909 910 add x17, x13, x15 // Calculate subroutine to jump to. 911 br x17 912 913.LisDouble: 914 cmp w17, #'D' // is this a double? 915 bne .LisLong 916 917 cmp x15, # 8*12 // Skip this load if all registers full. 918 beq .Ladvance8 919 920 add x17, x14, x15 // Calculate subroutine to jump to. 921 br x17 922 923.LisLong: 924 cmp w17, #'J' // is this a long? 925 bne .LisOther 926 927 cmp x8, # 6*12 // Skip this load if all registers full. 928 beq .Ladvance8 929 930 add x17, x12, x8 // Calculate subroutine to jump to. 931 br x17 932 933.LisOther: // Everything else takes one vReg. 934 cmp x8, # 6*12 // Skip this load if all registers full. 935 beq .Ladvance4 936 937 add x17, x11, x8 // Calculate subroutine to jump to. 938 br x17 939 940.Ladvance4: 941 add x9, x9, #4 942 b .LfillRegisters 943 944.Ladvance8: 945 add x9, x9, #8 946 b .LfillRegisters 947 948// Macro for loading a parameter into a register. 949// counter - the register with offset into these tables 950// size - the size of the register - 4 or 8 bytes. 951// register - the name of the register to be loaded. 952.macro LOADREG counter size register return 953 ldr \register , [x9], #\size 954 add \counter, \counter, 12 955 b \return 956.endm 957 958// Store ints. 959.LstoreW2: 960 LOADREG x8 4 w2 .LfillRegisters 961 LOADREG x8 4 w3 .LfillRegisters 962 LOADREG x8 4 w4 .LfillRegisters 963 LOADREG x8 4 w5 .LfillRegisters 964 LOADREG x8 4 w6 .LfillRegisters 965 LOADREG x8 4 w7 .LfillRegisters 966 967// Store longs. 968.LstoreX2: 969 LOADREG x8 8 x2 .LfillRegisters 970 LOADREG x8 8 x3 .LfillRegisters 971 LOADREG x8 8 x4 .LfillRegisters 972 LOADREG x8 8 x5 .LfillRegisters 973 LOADREG x8 8 x6 .LfillRegisters 974 LOADREG x8 8 x7 .LfillRegisters 975 976// Store singles. 977.LstoreS0: 978 LOADREG x15 4 s0 .LfillRegisters 979 LOADREG x15 4 s1 .LfillRegisters 980 LOADREG x15 4 s2 .LfillRegisters 981 LOADREG x15 4 s3 .LfillRegisters 982 LOADREG x15 4 s4 .LfillRegisters 983 LOADREG x15 4 s5 .LfillRegisters 984 LOADREG x15 4 s6 .LfillRegisters 985 LOADREG x15 4 s7 .LfillRegisters 986 987// Store doubles. 988.LstoreD0: 989 LOADREG x15 8 d0 .LfillRegisters 990 LOADREG x15 8 d1 .LfillRegisters 991 LOADREG x15 8 d2 .LfillRegisters 992 LOADREG x15 8 d3 .LfillRegisters 993 LOADREG x15 8 d4 .LfillRegisters 994 LOADREG x15 8 d5 .LfillRegisters 995 LOADREG x15 8 d6 .LfillRegisters 996 LOADREG x15 8 d7 .LfillRegisters 997 998 999.LcallFunction: 1000 1001 INVOKE_STUB_CALL_AND_RETURN 1002 1003END art_quick_invoke_stub 1004 1005/* extern"C" 1006 * void art_quick_invoke_static_stub(ArtMethod *method, x0 1007 * uint32_t *args, x1 1008 * uint32_t argsize, w2 1009 * Thread *self, x3 1010 * JValue *result, x4 1011 * char *shorty); x5 1012 */ 1013ENTRY art_quick_invoke_static_stub 1014 // Spill registers as per AACPS64 calling convention. 1015 INVOKE_STUB_CREATE_FRAME 1016 1017 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 1018 // Parse the passed shorty to determine which register to load. 1019 // Load addresses for routines that load WXSD registers. 1020 adr x11, .LstoreW1_2 1021 adr x12, .LstoreX1_2 1022 adr x13, .LstoreS0_2 1023 adr x14, .LstoreD0_2 1024 1025 // Initialize routine offsets to 0 for integers and floats. 1026 // x8 for integers, x15 for floating point. 1027 mov x8, #0 1028 mov x15, #0 1029 1030 add x10, x5, #1 // Load shorty address, plus one to skip return value. 1031 1032 // Loop to fill registers. 1033.LfillRegisters2: 1034 ldrb w17, [x10], #1 // Load next character in signature, and increment. 1035 cbz w17, .LcallFunction2 // Exit at end of signature. Shorty 0 terminated. 1036 1037 cmp w17, #'F' // is this a float? 1038 bne .LisDouble2 1039 1040 cmp x15, # 8*12 // Skip this load if all registers full. 1041 beq .Ladvance4_2 1042 1043 add x17, x13, x15 // Calculate subroutine to jump to. 1044 br x17 1045 1046.LisDouble2: 1047 cmp w17, #'D' // is this a double? 1048 bne .LisLong2 1049 1050 cmp x15, # 8*12 // Skip this load if all registers full. 1051 beq .Ladvance8_2 1052 1053 add x17, x14, x15 // Calculate subroutine to jump to. 1054 br x17 1055 1056.LisLong2: 1057 cmp w17, #'J' // is this a long? 1058 bne .LisOther2 1059 1060 cmp x8, # 7*12 // Skip this load if all registers full. 1061 beq .Ladvance8_2 1062 1063 add x17, x12, x8 // Calculate subroutine to jump to. 1064 br x17 1065 1066.LisOther2: // Everything else takes one vReg. 1067 cmp x8, # 7*12 // Skip this load if all registers full. 1068 beq .Ladvance4_2 1069 1070 add x17, x11, x8 // Calculate subroutine to jump to. 1071 br x17 1072 1073.Ladvance4_2: 1074 add x9, x9, #4 1075 b .LfillRegisters2 1076 1077.Ladvance8_2: 1078 add x9, x9, #8 1079 b .LfillRegisters2 1080 1081// Store ints. 1082.LstoreW1_2: 1083 LOADREG x8 4 w1 .LfillRegisters2 1084 LOADREG x8 4 w2 .LfillRegisters2 1085 LOADREG x8 4 w3 .LfillRegisters2 1086 LOADREG x8 4 w4 .LfillRegisters2 1087 LOADREG x8 4 w5 .LfillRegisters2 1088 LOADREG x8 4 w6 .LfillRegisters2 1089 LOADREG x8 4 w7 .LfillRegisters2 1090 1091// Store longs. 1092.LstoreX1_2: 1093 LOADREG x8 8 x1 .LfillRegisters2 1094 LOADREG x8 8 x2 .LfillRegisters2 1095 LOADREG x8 8 x3 .LfillRegisters2 1096 LOADREG x8 8 x4 .LfillRegisters2 1097 LOADREG x8 8 x5 .LfillRegisters2 1098 LOADREG x8 8 x6 .LfillRegisters2 1099 LOADREG x8 8 x7 .LfillRegisters2 1100 1101// Store singles. 1102.LstoreS0_2: 1103 LOADREG x15 4 s0 .LfillRegisters2 1104 LOADREG x15 4 s1 .LfillRegisters2 1105 LOADREG x15 4 s2 .LfillRegisters2 1106 LOADREG x15 4 s3 .LfillRegisters2 1107 LOADREG x15 4 s4 .LfillRegisters2 1108 LOADREG x15 4 s5 .LfillRegisters2 1109 LOADREG x15 4 s6 .LfillRegisters2 1110 LOADREG x15 4 s7 .LfillRegisters2 1111 1112// Store doubles. 1113.LstoreD0_2: 1114 LOADREG x15 8 d0 .LfillRegisters2 1115 LOADREG x15 8 d1 .LfillRegisters2 1116 LOADREG x15 8 d2 .LfillRegisters2 1117 LOADREG x15 8 d3 .LfillRegisters2 1118 LOADREG x15 8 d4 .LfillRegisters2 1119 LOADREG x15 8 d5 .LfillRegisters2 1120 LOADREG x15 8 d6 .LfillRegisters2 1121 LOADREG x15 8 d7 .LfillRegisters2 1122 1123 1124.LcallFunction2: 1125 1126 INVOKE_STUB_CALL_AND_RETURN 1127 1128END art_quick_invoke_static_stub 1129 1130 1131 1132/* extern"C" void art_quick_osr_stub(void** stack, x0 1133 * size_t stack_size_in_bytes, x1 1134 * const uin8_t* native_pc, x2 1135 * JValue *result, x3 1136 * char *shorty, x4 1137 * Thread *self) x5 1138 */ 1139ENTRY art_quick_osr_stub 1140SAVE_SIZE=15*8 // x3, x4, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 1141 mov x9, sp // Save stack pointer. 1142 .cfi_register sp,x9 1143 1144 sub x10, sp, # SAVE_SIZE 1145 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 1146 mov sp, x10 // Set new SP. 1147 1148 str x28, [sp, #112] 1149 stp x26, x27, [sp, #96] 1150 stp x24, x25, [sp, #80] 1151 stp x22, x23, [sp, #64] 1152 stp x20, x21, [sp, #48] 1153 stp x9, x19, [sp, #32] // Save old stack pointer and x19. 1154 stp x3, x4, [sp, #16] // Save result and shorty addresses. 1155 stp xFP, xLR, [sp] // Store LR & FP. 1156 mov xSELF, x5 // Move thread pointer into SELF register. 1157 1158 sub sp, sp, #16 1159 str xzr, [sp] // Store null for ArtMethod* slot 1160 // Branch to stub. 1161 bl .Losr_entry 1162 add sp, sp, #16 1163 1164 // Restore return value address and shorty address. 1165 ldp x3,x4, [sp, #16] 1166 ldr x28, [sp, #112] 1167 ldp x26, x27, [sp, #96] 1168 ldp x24, x25, [sp, #80] 1169 ldp x22, x23, [sp, #64] 1170 ldp x20, x21, [sp, #48] 1171 1172 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 1173 ldrb w10, [x4] 1174 1175 // Check the return type and store the correct register into the jvalue in memory. 1176 1177 // Don't set anything for a void type. 1178 cmp w10, #'V' 1179 beq .Losr_exit 1180 1181 // Is it a double? 1182 cmp w10, #'D' 1183 bne .Lno_double 1184 str d0, [x3] 1185 b .Losr_exit 1186 1187.Lno_double: // Is it a float? 1188 cmp w10, #'F' 1189 bne .Lno_float 1190 str s0, [x3] 1191 b .Losr_exit 1192 1193.Lno_float: // Just store x0. Doesn't matter if it is 64 or 32 bits. 1194 str x0, [x3] 1195 1196.Losr_exit: // Finish up. 1197 ldp x2, x19, [sp, #32] // Restore stack pointer and x19. 1198 ldp xFP, xLR, [sp] // Restore old frame pointer and link register. 1199 mov sp, x2 1200 ret 1201 1202.Losr_entry: 1203 // Update stack pointer for the callee 1204 sub sp, sp, x1 1205 1206 // Update link register slot expected by the callee. 1207 sub w1, w1, #8 1208 str lr, [sp, x1] 1209 1210 // Copy arguments into stack frame. 1211 // Use simple copy routine for now. 1212 // 4 bytes per slot. 1213 // X0 - source address 1214 // W1 - args length 1215 // SP - destination address. 1216 // W10 - temporary 1217.Losr_loop_entry: 1218 cmp w1, #0 1219 beq .Losr_loop_exit 1220 sub w1, w1, #4 1221 ldr w10, [x0, x1] 1222 str w10, [sp, x1] 1223 b .Losr_loop_entry 1224 1225.Losr_loop_exit: 1226 // Branch to the OSR entry point. 1227 br x2 1228 1229END art_quick_osr_stub 1230 1231 /* 1232 * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_ 1233 */ 1234 1235ENTRY art_quick_do_long_jump 1236 // Load FPRs 1237 ldp d0, d1, [x1], #16 1238 ldp d2, d3, [x1], #16 1239 ldp d4, d5, [x1], #16 1240 ldp d6, d7, [x1], #16 1241 ldp d8, d9, [x1], #16 1242 ldp d10, d11, [x1], #16 1243 ldp d12, d13, [x1], #16 1244 ldp d14, d15, [x1], #16 1245 ldp d16, d17, [x1], #16 1246 ldp d18, d19, [x1], #16 1247 ldp d20, d21, [x1], #16 1248 ldp d22, d23, [x1], #16 1249 ldp d24, d25, [x1], #16 1250 ldp d26, d27, [x1], #16 1251 ldp d28, d29, [x1], #16 1252 ldp d30, d31, [x1] 1253 1254 // Load GPRs 1255 // TODO: lots of those are smashed, could optimize. 1256 add x0, x0, #30*8 1257 ldp x30, x1, [x0], #-16 // LR & SP 1258 ldp x28, x29, [x0], #-16 1259 ldp x26, x27, [x0], #-16 1260 ldp x24, x25, [x0], #-16 1261 ldp x22, x23, [x0], #-16 1262 ldp x20, x21, [x0], #-16 1263 ldp x18, x19, [x0], #-16 1264 ldp x16, x17, [x0], #-16 1265 ldp x14, x15, [x0], #-16 1266 ldp x12, x13, [x0], #-16 1267 ldp x10, x11, [x0], #-16 1268 ldp x8, x9, [x0], #-16 1269 ldp x6, x7, [x0], #-16 1270 ldp x4, x5, [x0], #-16 1271 ldp x2, x3, [x0], #-16 1272 mov sp, x1 1273 1274 // Need to load PC, it's at the end (after the space for the unused XZR). Use x1. 1275 ldr x1, [x0, #33*8] 1276 // And the value of x0. 1277 ldr x0, [x0] 1278 1279 br x1 1280END art_quick_do_long_jump 1281 1282 /* 1283 * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the 1284 * possibly null object to lock. 1285 * 1286 * Derived from arm32 code. 1287 */ 1288 .extern artLockObjectFromCode 1289ENTRY art_quick_lock_object 1290 cbz w0, .Lslow_lock 1291 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1292.Lretry_lock: 1293 ldr w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop? 1294 ldxr w1, [x4] 1295 mov x3, x1 1296 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1297 cbnz w3, .Lnot_unlocked // already thin locked 1298 // unlocked case - x1: original lock word that's zero except for the read barrier bits. 1299 orr x2, x1, x2 // x2 holds thread id with count of 0 with preserved read barrier bits 1300 stxr w3, w2, [x4] 1301 cbnz w3, .Llock_stxr_fail // store failed, retry 1302 dmb ishld // full (LoadLoad|LoadStore) memory barrier 1303 ret 1304.Lnot_unlocked: // x1: original lock word 1305 lsr w3, w1, LOCK_WORD_STATE_SHIFT 1306 cbnz w3, .Lslow_lock // if either of the top two bits are set, go slow path 1307 eor w2, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 1308 uxth w2, w2 // zero top 16 bits 1309 cbnz w2, .Lslow_lock // lock word and self thread id's match -> recursive lock 1310 // else contention, go to slow path 1311 mov x3, x1 // copy the lock word to check count overflow. 1312 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits. 1313 add w2, w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count in lock word placing in w2 to check overflow 1314 lsr w3, w2, #LOCK_WORD_GC_STATE_SHIFT // if the first gc state bit is set, we overflowed. 1315 cbnz w3, .Lslow_lock // if we overflow the count go slow path 1316 add w2, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count for real 1317 stxr w3, w2, [x4] 1318 cbnz w3, .Llock_stxr_fail // store failed, retry 1319 ret 1320.Llock_stxr_fail: 1321 b .Lretry_lock // retry 1322.Lslow_lock: 1323 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case we block 1324 mov x1, xSELF // pass Thread::Current 1325 bl artLockObjectFromCode // (Object* obj, Thread*) 1326 RESTORE_SAVE_REFS_ONLY_FRAME 1327 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1328END art_quick_lock_object 1329 1330ENTRY art_quick_lock_object_no_inline 1331 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case we block 1332 mov x1, xSELF // pass Thread::Current 1333 bl artLockObjectFromCode // (Object* obj, Thread*) 1334 RESTORE_SAVE_REFS_ONLY_FRAME 1335 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1336END art_quick_lock_object_no_inline 1337 1338 /* 1339 * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. 1340 * x0 holds the possibly null object to lock. 1341 * 1342 * Derived from arm32 code. 1343 */ 1344 .extern artUnlockObjectFromCode 1345ENTRY art_quick_unlock_object 1346 cbz x0, .Lslow_unlock 1347 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1348.Lretry_unlock: 1349#ifndef USE_READ_BARRIER 1350 ldr w1, [x4] 1351#else 1352 ldxr w1, [x4] // Need to use atomic instructions for read barrier 1353#endif 1354 lsr w2, w1, LOCK_WORD_STATE_SHIFT 1355 cbnz w2, .Lslow_unlock // if either of the top two bits are set, go slow path 1356 ldr w2, [xSELF, #THREAD_ID_OFFSET] 1357 mov x3, x1 // copy lock word to check thread id equality 1358 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1359 eor w3, w3, w2 // lock_word.ThreadId() ^ self->ThreadId() 1360 uxth w3, w3 // zero top 16 bits 1361 cbnz w3, .Lslow_unlock // do lock word and self thread id's match? 1362 mov x3, x1 // copy lock word to detect transition to unlocked 1363 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1364 cmp w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE 1365 bpl .Lrecursive_thin_unlock 1366 // transition to unlocked 1367 mov x3, x1 1368 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED // w3: zero except for the preserved read barrier bits 1369 dmb ish // full (LoadStore|StoreStore) memory barrier 1370#ifndef USE_READ_BARRIER 1371 str w3, [x4] 1372#else 1373 stxr w2, w3, [x4] // Need to use atomic instructions for read barrier 1374 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1375#endif 1376 ret 1377.Lrecursive_thin_unlock: // w1: original lock word 1378 sub w1, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // decrement count 1379#ifndef USE_READ_BARRIER 1380 str w1, [x4] 1381#else 1382 stxr w2, w1, [x4] // Need to use atomic instructions for read barrier 1383 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1384#endif 1385 ret 1386.Lunlock_stxr_fail: 1387 b .Lretry_unlock // retry 1388.Lslow_unlock: 1389 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case exception allocation triggers GC 1390 mov x1, xSELF // pass Thread::Current 1391 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1392 RESTORE_SAVE_REFS_ONLY_FRAME 1393 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1394END art_quick_unlock_object 1395 1396ENTRY art_quick_unlock_object_no_inline 1397 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case exception allocation triggers GC 1398 mov x1, xSELF // pass Thread::Current 1399 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1400 RESTORE_SAVE_REFS_ONLY_FRAME 1401 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1402END art_quick_unlock_object_no_inline 1403 1404 /* 1405 * Entry from managed code that calls artIsAssignableFromCode and on failure calls 1406 * artThrowClassCastException. 1407 */ 1408 .extern artThrowClassCastException 1409ENTRY art_quick_check_cast 1410 // Store arguments and link register 1411 // Stack needs to be 16B aligned on calls. 1412 stp x0, x1, [sp,#-32]! 1413 .cfi_adjust_cfa_offset 32 1414 .cfi_rel_offset x0, 0 1415 .cfi_rel_offset x1, 8 1416 str xLR, [sp, #24] 1417 .cfi_rel_offset x30, 24 1418 1419 // Call runtime code 1420 bl artIsAssignableFromCode 1421 1422 // Check for exception 1423 cbz x0, .Lthrow_class_cast_exception 1424 1425 // Restore and return 1426 ldr xLR, [sp, #24] 1427 .cfi_restore x30 1428 ldp x0, x1, [sp], #32 1429 .cfi_restore x0 1430 .cfi_restore x1 1431 .cfi_adjust_cfa_offset -32 1432 ret 1433 1434 .cfi_adjust_cfa_offset 32 // Reset unwind info so following code unwinds. 1435 1436.Lthrow_class_cast_exception: 1437 // Restore 1438 ldr xLR, [sp, #24] 1439 .cfi_restore x30 1440 ldp x0, x1, [sp], #32 1441 .cfi_restore x0 1442 .cfi_restore x1 1443 .cfi_adjust_cfa_offset -32 1444 1445 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context 1446 mov x2, xSELF // pass Thread::Current 1447 b artThrowClassCastException // (Class*, Class*, Thread*) 1448 brk 0 // We should not return here... 1449END art_quick_check_cast 1450 1451// Restore xReg's value from [sp, #offset] if xReg is not the same as xExclude. 1452.macro POP_REG_NE xReg, offset, xExclude 1453 .ifnc \xReg, \xExclude 1454 ldr \xReg, [sp, #\offset] // restore xReg 1455 .cfi_restore \xReg 1456 .endif 1457.endm 1458 1459// Restore xReg1's value from [sp, #offset] if xReg1 is not the same as xExclude. 1460// Restore xReg2's value from [sp, #(offset + 8)] if xReg2 is not the same as xExclude. 1461.macro POP_REGS_NE xReg1, xReg2, offset, xExclude 1462 .ifc \xReg1, \xExclude 1463 ldr \xReg2, [sp, #(\offset + 8)] // restore xReg2 1464 .else 1465 .ifc \xReg2, \xExclude 1466 ldr \xReg1, [sp, #\offset] // restore xReg1 1467 .else 1468 ldp \xReg1, \xReg2, [sp, #\offset] // restore xReg1 and xReg2 1469 .endif 1470 .endif 1471 .cfi_restore \xReg1 1472 .cfi_restore \xReg2 1473.endm 1474 1475 /* 1476 * Macro to insert read barrier, only used in art_quick_aput_obj. 1477 * xDest, wDest and xObj are registers, offset is a defined literal such as 1478 * MIRROR_OBJECT_CLASS_OFFSET. Dest needs both x and w versions of the same register to handle 1479 * name mismatch between instructions. This macro uses the lower 32b of register when possible. 1480 * TODO: When read barrier has a fast path, add heap unpoisoning support for the fast path. 1481 */ 1482.macro READ_BARRIER xDest, wDest, xObj, xTemp, wTemp, offset, number 1483#ifdef USE_READ_BARRIER 1484#ifdef USE_BAKER_READ_BARRIER 1485 ldr \wTemp, [\xObj, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 1486 tbnz \wTemp, #LOCK_WORD_READ_BARRIER_STATE_SHIFT, .Lrb_slowpath\number 1487 // False dependency to avoid needing load/load fence. 1488 add \xObj, \xObj, \xTemp, lsr #32 1489 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1490 UNPOISON_HEAP_REF \wDest 1491 b .Lrb_exit\number 1492#endif 1493.Lrb_slowpath\number: 1494 // Store registers used in art_quick_aput_obj (x0-x4, LR), stack is 16B aligned. 1495 stp x0, x1, [sp, #-48]! 1496 .cfi_adjust_cfa_offset 48 1497 .cfi_rel_offset x0, 0 1498 .cfi_rel_offset x1, 8 1499 stp x2, x3, [sp, #16] 1500 .cfi_rel_offset x2, 16 1501 .cfi_rel_offset x3, 24 1502 stp x4, xLR, [sp, #32] 1503 .cfi_rel_offset x4, 32 1504 .cfi_rel_offset x30, 40 1505 1506 // mov x0, \xRef // pass ref in x0 (no-op for now since parameter ref is unused) 1507 .ifnc \xObj, x1 1508 mov x1, \xObj // pass xObj 1509 .endif 1510 mov w2, #\offset // pass offset 1511 bl artReadBarrierSlow // artReadBarrierSlow(ref, xObj, offset) 1512 // No need to unpoison return value in w0, artReadBarrierSlow() would do the unpoisoning. 1513 .ifnc \wDest, w0 1514 mov \wDest, w0 // save return value in wDest 1515 .endif 1516 1517 // Conditionally restore saved registers 1518 POP_REG_NE x0, 0, \xDest 1519 POP_REG_NE x1, 8, \xDest 1520 POP_REG_NE x2, 16, \xDest 1521 POP_REG_NE x3, 24, \xDest 1522 POP_REG_NE x4, 32, \xDest 1523 ldr xLR, [sp, #40] 1524 .cfi_restore x30 1525 add sp, sp, #48 1526 .cfi_adjust_cfa_offset -48 1527.Lrb_exit\number: 1528#else 1529 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1530 UNPOISON_HEAP_REF \wDest 1531#endif // USE_READ_BARRIER 1532.endm 1533 1534 /* 1535 * Entry from managed code for array put operations of objects where the value being stored 1536 * needs to be checked for compatibility. 1537 * x0 = array, x1 = index, x2 = value 1538 * 1539 * Currently all values should fit into w0/w1/w2, and w1 always will as indices are 32b. We 1540 * assume, though, that the upper 32b are zeroed out. At least for x1/w1 we can do better by 1541 * using index-zero-extension in load/stores. 1542 * 1543 * Temporaries: x3, x4 1544 * TODO: x4 OK? ip seems wrong here. 1545 */ 1546ENTRY art_quick_aput_obj_with_null_and_bound_check 1547 tst x0, x0 1548 bne art_quick_aput_obj_with_bound_check 1549 b art_quick_throw_null_pointer_exception 1550END art_quick_aput_obj_with_null_and_bound_check 1551 1552ENTRY art_quick_aput_obj_with_bound_check 1553 ldr w3, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] 1554 cmp w3, w1 1555 bhi art_quick_aput_obj 1556 mov x0, x1 1557 mov x1, x3 1558 b art_quick_throw_array_bounds 1559END art_quick_aput_obj_with_bound_check 1560 1561#ifdef USE_READ_BARRIER 1562 .extern artReadBarrierSlow 1563#endif 1564ENTRY art_quick_aput_obj 1565 cbz x2, .Ldo_aput_null 1566 READ_BARRIER x3, w3, x0, x3, w3, MIRROR_OBJECT_CLASS_OFFSET, 0 // Heap reference = 32b 1567 // This also zero-extends to x3 1568 READ_BARRIER x3, w3, x3, x4, w4, MIRROR_CLASS_COMPONENT_TYPE_OFFSET, 1 // Heap reference = 32b 1569 // This also zero-extends to x3 1570 READ_BARRIER x4, w4, x2, x4, w4, MIRROR_OBJECT_CLASS_OFFSET, 2 // Heap reference = 32b 1571 // This also zero-extends to x4 1572 cmp w3, w4 // value's type == array's component type - trivial assignability 1573 bne .Lcheck_assignability 1574.Ldo_aput: 1575 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1576 // "Compress" = do nothing 1577 POISON_HEAP_REF w2 1578 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1579 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1580 lsr x0, x0, #7 1581 strb w3, [x3, x0] 1582 ret 1583.Ldo_aput_null: 1584 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1585 // "Compress" = do nothing 1586 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1587 ret 1588.Lcheck_assignability: 1589 // Store arguments and link register 1590 stp x0, x1, [sp,#-32]! 1591 .cfi_adjust_cfa_offset 32 1592 .cfi_rel_offset x0, 0 1593 .cfi_rel_offset x1, 8 1594 stp x2, xLR, [sp, #16] 1595 .cfi_rel_offset x2, 16 1596 .cfi_rel_offset x30, 24 1597 1598 // Call runtime code 1599 mov x0, x3 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1600 mov x1, x4 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1601 bl artIsAssignableFromCode 1602 1603 // Check for exception 1604 cbz x0, .Lthrow_array_store_exception 1605 1606 // Restore 1607 ldp x2, x30, [sp, #16] 1608 .cfi_restore x2 1609 .cfi_restore x30 1610 ldp x0, x1, [sp], #32 1611 .cfi_restore x0 1612 .cfi_restore x1 1613 .cfi_adjust_cfa_offset -32 1614 1615 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1616 // "Compress" = do nothing 1617 POISON_HEAP_REF w2 1618 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1619 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1620 lsr x0, x0, #7 1621 strb w3, [x3, x0] 1622 ret 1623 .cfi_adjust_cfa_offset 32 // 4 restores after cbz for unwinding. 1624.Lthrow_array_store_exception: 1625 ldp x2, x30, [sp, #16] 1626 .cfi_restore x2 1627 .cfi_restore x30 1628 ldp x0, x1, [sp], #32 1629 .cfi_restore x0 1630 .cfi_restore x1 1631 .cfi_adjust_cfa_offset -32 1632 1633 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 1634 mov x1, x2 // Pass value. 1635 mov x2, xSELF // Pass Thread::Current. 1636 b artThrowArrayStoreException // (Object*, Object*, Thread*). 1637 brk 0 // Unreached. 1638END art_quick_aput_obj 1639 1640// Macro to facilitate adding new allocation entrypoints. 1641.macro ONE_ARG_DOWNCALL name, entrypoint, return 1642 .extern \entrypoint 1643ENTRY \name 1644 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1645 mov x1, xSELF // pass Thread::Current 1646 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1647 RESTORE_SAVE_REFS_ONLY_FRAME 1648 \return 1649END \name 1650.endm 1651 1652// Macro to facilitate adding new allocation entrypoints. 1653.macro TWO_ARG_DOWNCALL name, entrypoint, return 1654 .extern \entrypoint 1655ENTRY \name 1656 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1657 mov x2, xSELF // pass Thread::Current 1658 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1659 RESTORE_SAVE_REFS_ONLY_FRAME 1660 \return 1661END \name 1662.endm 1663 1664// Macro to facilitate adding new allocation entrypoints. 1665.macro THREE_ARG_DOWNCALL name, entrypoint, return 1666 .extern \entrypoint 1667ENTRY \name 1668 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1669 mov x3, xSELF // pass Thread::Current 1670 bl \entrypoint 1671 RESTORE_SAVE_REFS_ONLY_FRAME 1672 \return 1673END \name 1674.endm 1675 1676// Macro to facilitate adding new allocation entrypoints. 1677.macro FOUR_ARG_DOWNCALL name, entrypoint, return 1678 .extern \entrypoint 1679ENTRY \name 1680 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1681 mov x4, xSELF // pass Thread::Current 1682 bl \entrypoint // 1683 RESTORE_SAVE_REFS_ONLY_FRAME 1684 \return 1685 DELIVER_PENDING_EXCEPTION 1686END \name 1687.endm 1688 1689// Macros taking opportunity of code similarities for downcalls with referrer. 1690.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return 1691 .extern \entrypoint 1692ENTRY \name 1693 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1694 ldr x1, [sp, #FRAME_SIZE_SAVE_REFS_ONLY] // Load referrer 1695 mov x2, xSELF // pass Thread::Current 1696 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*, SP) 1697 RESTORE_SAVE_REFS_ONLY_FRAME 1698 \return 1699END \name 1700.endm 1701 1702.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return 1703 .extern \entrypoint 1704ENTRY \name 1705 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1706 ldr x2, [sp, #FRAME_SIZE_SAVE_REFS_ONLY] // Load referrer 1707 mov x3, xSELF // pass Thread::Current 1708 bl \entrypoint 1709 RESTORE_SAVE_REFS_ONLY_FRAME 1710 \return 1711END \name 1712.endm 1713 1714.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return 1715 .extern \entrypoint 1716ENTRY \name 1717 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1718 ldr x3, [sp, #FRAME_SIZE_SAVE_REFS_ONLY] // Load referrer 1719 mov x4, xSELF // pass Thread::Current 1720 bl \entrypoint 1721 RESTORE_SAVE_REFS_ONLY_FRAME 1722 \return 1723END \name 1724.endm 1725 1726.macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1727 cbz w0, 1f // result zero branch over 1728 ret // return 17291: 1730 DELIVER_PENDING_EXCEPTION 1731.endm 1732 1733 /* 1734 * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on 1735 * failure. 1736 */ 1737TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1738 1739 /* 1740 * Entry from managed code when uninitialized static storage, this stub will run the class 1741 * initializer and deliver the exception on error. On success the static storage base is 1742 * returned. 1743 */ 1744ONE_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1745 1746ONE_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1747ONE_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1748 1749ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1750ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1751ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1752ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1753ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1754ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1755ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1756 1757TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1758TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1759TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1760TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1761TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1762TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1763TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1764 1765TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1766TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1767TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1768TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1769 1770THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1771THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1772THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1773THREE_ARG_REF_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1774THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1775 1776// This is separated out as the argument order is different. 1777 .extern artSet64StaticFromCode 1778ENTRY art_quick_set64_static 1779 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1780 ldr x1, [sp, #FRAME_SIZE_SAVE_REFS_ONLY] // Load referrer 1781 // x2 contains the parameter 1782 mov x3, xSELF // pass Thread::Current 1783 bl artSet64StaticFromCode 1784 RESTORE_SAVE_REFS_ONLY_FRAME 1785 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1786END art_quick_set64_static 1787 1788 /* 1789 * Entry from managed code to resolve a string, this stub will allocate a String and deliver an 1790 * exception on error. On success the String is returned. w0 holds the string index. The fast 1791 * path check for hit in strings cache has already been performed. 1792 */ 1793ONE_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1794 1795// Generate the allocation entrypoints for each allocator. 1796GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_REGION_TLAB_ALLOCATORS 1797// Comment out allocators that have arm64 specific asm. 1798// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_region_tlab, RegionTLAB) implemented in asm 1799// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_tlab, RegionTLAB) 1800// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB) 1801GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1802// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY(_region_tlab, RegionTLAB) implemented in asm 1803// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB) 1804GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1805GENERATE_ALLOC_ENTRYPOINTS_CHECK_AND_ALLOC_ARRAY(_region_tlab, RegionTLAB) 1806GENERATE_ALLOC_ENTRYPOINTS_CHECK_AND_ALLOC_ARRAY_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1807GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB) 1808GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB) 1809GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB) 1810 1811// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_rosalloc, RosAlloc). 1812ENTRY art_quick_alloc_object_rosalloc 1813 // Fast path rosalloc allocation. 1814 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 1815 // x2-x7: free. 1816 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 1817 // Load the class (x2) 1818 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 1819 cbz x2, .Lart_quick_alloc_object_rosalloc_slow_path // Check null class 1820 // Check class status. 1821 ldr w3, [x2, #MIRROR_CLASS_STATUS_OFFSET] 1822 cmp x3, #MIRROR_CLASS_STATUS_INITIALIZED 1823 bne .Lart_quick_alloc_object_rosalloc_slow_path 1824 // Add a fake dependence from the 1825 // following access flag and size 1826 // loads to the status load. 1827 // This is to prevent those loads 1828 // from being reordered above the 1829 // status load and reading wrong 1830 // values (an alternative is to use 1831 // a load-acquire for the status). 1832 eor x3, x3, x3 1833 add x2, x2, x3 1834 // Check access flags has 1835 // kAccClassIsFinalizable 1836 ldr w3, [x2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] 1837 tst x3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE 1838 bne .Lart_quick_alloc_object_rosalloc_slow_path 1839 ldr x3, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] // Check if the thread local 1840 // allocation stack has room. 1841 // ldp won't work due to large offset. 1842 ldr x4, [xSELF, #THREAD_LOCAL_ALLOC_STACK_END_OFFSET] 1843 cmp x3, x4 1844 bhs .Lart_quick_alloc_object_rosalloc_slow_path 1845 ldr w3, [x2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (x3) 1846 cmp x3, #ROSALLOC_MAX_THREAD_LOCAL_BRACKET_SIZE // Check if the size is for a thread 1847 // local allocation 1848 bhs .Lart_quick_alloc_object_rosalloc_slow_path 1849 // Compute the rosalloc bracket index 1850 // from the size. 1851 // Align up the size by the rosalloc 1852 // bracket quantum size and divide 1853 // by the quantum size and subtract 1854 // by 1. This code is a shorter but 1855 // equivalent version. 1856 sub x3, x3, #1 1857 lsr x3, x3, #ROSALLOC_BRACKET_QUANTUM_SIZE_SHIFT 1858 // Load the rosalloc run (x4) 1859 add x4, xSELF, x3, lsl #POINTER_SIZE_SHIFT 1860 ldr x4, [x4, #THREAD_ROSALLOC_RUNS_OFFSET] 1861 // Load the free list head (x3). This 1862 // will be the return val. 1863 ldr x3, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1864 cbz x3, .Lart_quick_alloc_object_rosalloc_slow_path 1865 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1866 ldr x1, [x3, #ROSALLOC_SLOT_NEXT_OFFSET] // Load the next pointer of the head 1867 // and update the list head with the 1868 // next pointer. 1869 str x1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1870 // Store the class pointer in the 1871 // header. This also overwrites the 1872 // next pointer. The offsets are 1873 // asserted to match. 1874#if ROSALLOC_SLOT_NEXT_OFFSET != MIRROR_OBJECT_CLASS_OFFSET 1875#error "Class pointer needs to overwrite next pointer." 1876#endif 1877 POISON_HEAP_REF w2 1878 str w2, [x3, #MIRROR_OBJECT_CLASS_OFFSET] 1879 // Fence. This is "ish" not "ishst" so 1880 // that it also ensures ordering of 1881 // the class status load with respect 1882 // to later accesses to the class 1883 // object. Alternatively we could use 1884 // "ishst" if we use load-acquire for 1885 // the class status load.) 1886 // Needs to be done before pushing on 1887 // allocation since Heap::VisitObjects 1888 // relies on seeing the class pointer. 1889 // b/28790624 1890 dmb ish 1891 // Push the new object onto the thread 1892 // local allocation stack and 1893 // increment the thread local 1894 // allocation stack top. 1895 ldr x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1896 str w3, [x1], #COMPRESSED_REFERENCE_SIZE // (Increment x1 as a side effect.) 1897 str x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1898 // Decrement the size of the free list 1899 ldr w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1900 sub x1, x1, #1 1901 // TODO: consider combining this store 1902 // and the list head store above using 1903 // strd. 1904 str w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1905 1906 mov x0, x3 // Set the return value and return. 1907 ret 1908.Lart_quick_alloc_object_rosalloc_slow_path: 1909 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1910 mov x2, xSELF // pass Thread::Current 1911 bl artAllocObjectFromCodeRosAlloc // (uint32_t type_idx, Method* method, Thread*) 1912 RESTORE_SAVE_REFS_ONLY_FRAME 1913 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1914END art_quick_alloc_object_rosalloc 1915 1916 1917// The common fast path code for art_quick_alloc_array_region_tlab. 1918.macro ALLOC_ARRAY_TLAB_FAST_PATH slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1919 // Check null class 1920 cbz \wClass, \slowPathLabel 1921 ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED \slowPathLabel, \xClass, \wClass, \xCount, \wCount, \xTemp0, \wTemp0, \xTemp1, \wTemp1, \xTemp2, \wTemp2 1922.endm 1923 1924// The common fast path code for art_quick_alloc_array_region_tlab. 1925.macro ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1926 // Array classes are never finalizable or uninitialized, no need to check. 1927 ldr \wTemp0, [\xClass, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Load component type 1928 UNPOISON_HEAP_REF \wTemp0 1929 ldr \wTemp0, [\xTemp0, #MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET] 1930 lsr \xTemp0, \xTemp0, #PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT // Component size shift is in high 16 1931 // bits. 1932 // xCount is holding a 32 bit value, 1933 // it can not overflow. 1934 lsl \xTemp1, \xCount, \xTemp0 // Calculate data size 1935 // Add array data offset and alignment. 1936 add \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) 1937#if MIRROR_LONG_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4 1938#error Long array data offset must be 4 greater than int array data offset. 1939#endif 1940 1941 add \xTemp0, \xTemp0, #1 // Add 4 to the length only if the 1942 // component size shift is 3 1943 // (for 64 bit alignment). 1944 and \xTemp0, \xTemp0, #4 1945 add \xTemp1, \xTemp1, \xTemp0 1946 and \xTemp1, \xTemp1, #OBJECT_ALIGNMENT_MASK_TOGGLED64 // Apply alignemnt mask 1947 // (addr + 7) & ~7. The mask must 1948 // be 64 bits to keep high bits in 1949 // case of overflow. 1950 // Negative sized arrays are handled here since xCount holds a zero extended 32 bit value. 1951 // Negative ints become large 64 bit unsigned ints which will always be larger than max signed 1952 // 32 bit int. Since the max shift for arrays is 3, it can not become a negative 64 bit int. 1953 cmp \xTemp1, #MIN_LARGE_OBJECT_THRESHOLD // Possibly a large object, go slow 1954 bhs \slowPathLabel // path. 1955 1956 ldr \xTemp0, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Check tlab for space, note that 1957 // we use (end - begin) to handle 1958 // negative size arrays. It is 1959 // assumed that a negative size will 1960 // always be greater unsigned than 1961 // region size. 1962 ldr \xTemp2, [xSELF, #THREAD_LOCAL_END_OFFSET] 1963 sub \xTemp2, \xTemp2, \xTemp0 1964 cmp \xTemp1, \xTemp2 1965 bhi \slowPathLabel 1966 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1967 // Move old thread_local_pos to x0 1968 // for the return value. 1969 mov x0, \xTemp0 1970 add \xTemp0, \xTemp0, \xTemp1 1971 str \xTemp0, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 1972 ldr \xTemp0, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 1973 add \xTemp0, \xTemp0, #1 1974 str \xTemp0, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 1975 POISON_HEAP_REF \wClass 1976 str \wClass, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 1977 str \wCount, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] // Store the array length. 1978 // Fence. 1979 dmb ishst 1980 ret 1981.endm 1982 1983// The common fast path code for art_quick_alloc_object_tlab and art_quick_alloc_object_region_tlab. 1984// 1985// x0: type_idx/return value, x1: ArtMethod*, x2: Class*, xSELF(x19): Thread::Current 1986// x3-x7: free. 1987// Need to preserve x0 and x1 to the slow path. 1988.macro ALLOC_OBJECT_TLAB_FAST_PATH slowPathLabel 1989 cbz x2, \slowPathLabel // Check null class 1990 ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED \slowPathLabel 1991.endm 1992 1993.macro ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED slowPathLabel 1994 ldr w3, [x2, #MIRROR_CLASS_STATUS_OFFSET] // Check class status. 1995 cmp x3, #MIRROR_CLASS_STATUS_INITIALIZED 1996 bne \slowPathLabel 1997 // Add a fake dependence from the 1998 // following access flag and size 1999 // loads to the status load. 2000 // This is to prevent those loads 2001 // from being reordered above the 2002 // status load and reading wrong 2003 // values (an alternative is to use 2004 // a load-acquire for the status). 2005 eor x3, x3, x3 2006 add x2, x2, x3 2007 ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED \slowPathLabel 2008.endm 2009 2010.macro ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED slowPathLabel 2011 // Check access flags has 2012 // kAccClassIsFinalizable. 2013 ldr w3, [x2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] 2014 tbnz x3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE_BIT, \slowPathLabel 2015 // Load thread_local_pos (x4) and 2016 // thread_local_end (x5). 2017 ldr x4, [xSELF, #THREAD_LOCAL_POS_OFFSET] 2018 ldr x5, [xSELF, #THREAD_LOCAL_END_OFFSET] 2019 sub x6, x5, x4 // Compute the remaining buf size. 2020 ldr w7, [x2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (x7). 2021 cmp x7, x6 // Check if it fits. OK to do this 2022 // before rounding up the object size 2023 // assuming the buf size alignment. 2024 bhi \slowPathLabel 2025 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 2026 // Round up the object size by the 2027 // object alignment. (addr + 7) & ~7. 2028 add x7, x7, #OBJECT_ALIGNMENT_MASK 2029 and x7, x7, #OBJECT_ALIGNMENT_MASK_TOGGLED 2030 // Move old thread_local_pos to x0 2031 // for the return value. 2032 mov x0, x4 2033 add x5, x0, x7 2034 str x5, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 2035 ldr x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 2036 add x5, x5, #1 2037 str x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 2038 POISON_HEAP_REF w2 2039 str w2, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 2040 // Fence. This is "ish" not "ishst" so 2041 // that the code after this allocation 2042 // site will see the right values in 2043 // the fields of the class. 2044 // Alternatively we could use "ishst" 2045 // if we use load-acquire for the 2046 // class status load.) 2047 dmb ish 2048 ret 2049.endm 2050 2051// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_tlab, TLAB). 2052ENTRY art_quick_alloc_object_tlab 2053 // Fast path tlab allocation. 2054 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 2055 // x2-x7: free. 2056#if defined(USE_READ_BARRIER) 2057 mvn x0, xzr // Read barrier not supported here. 2058 ret // Return -1. 2059#endif 2060 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2061 // Load the class (x2) 2062 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2063 ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_tlab_slow_path 2064.Lart_quick_alloc_object_tlab_slow_path: 2065 SETUP_SAVE_REFS_ONLY_FRAME // Save callee saves in case of GC. 2066 mov x2, xSELF // Pass Thread::Current. 2067 bl artAllocObjectFromCodeTLAB // (uint32_t type_idx, Method* method, Thread*) 2068 RESTORE_SAVE_REFS_ONLY_FRAME 2069 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2070END art_quick_alloc_object_tlab 2071 2072// The common code for art_quick_alloc_object_*region_tlab 2073.macro GENERATE_ALLOC_OBJECT_REGION_TLAB name, entrypoint, fast_path, is_resolved 2074ENTRY \name 2075 // Fast path region tlab allocation. 2076 // x0: type_idx/resolved class/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 2077 // If is_resolved is 1 then x0 is the resolved type, otherwise it is the index. 2078 // x2-x7: free. 2079#if !defined(USE_READ_BARRIER) 2080 mvn x0, xzr // Read barrier must be enabled here. 2081 ret // Return -1. 2082#endif 2083.if \is_resolved 2084 mov x2, x0 // class is actually stored in x0 already 2085.else 2086 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2087 // Load the class (x2) 2088 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2089.endif 2090 // Most common case: GC is not marking. 2091 ldr w3, [xSELF, #THREAD_IS_GC_MARKING_OFFSET] 2092 cbnz x3, .Lmarking\name 2093.Ldo_allocation\name: 2094 \fast_path .Lslow_path\name 2095.Lmarking\name: 2096 // GC is marking, check the lock word of the class for the mark bit. 2097 // If the class is null, go slow path. The check is required to read the lock word. 2098 cbz w2, .Lslow_path\name 2099 // Class is not null, check mark bit in lock word. 2100 ldr w3, [x2, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2101 // If the bit is not zero, do the allocation. 2102 tbnz w3, #LOCK_WORD_MARK_BIT_SHIFT, .Ldo_allocation\name 2103 // The read barrier slow path. Mark 2104 // the class. 2105 stp x0, x1, [sp, #-32]! // Save registers (x0, x1, lr). 2106 str xLR, [sp, #16] // Align sp by 16 bytes. 2107 mov x0, x2 // Pass the class as the first param. 2108 bl artReadBarrierMark 2109 mov x2, x0 // Get the (marked) class back. 2110 ldp x0, x1, [sp, #0] // Restore registers. 2111 ldr xLR, [sp, #16] 2112 add sp, sp, #32 2113 b .Ldo_allocation\name 2114.Lslow_path\name: 2115 SETUP_SAVE_REFS_ONLY_FRAME // Save callee saves in case of GC. 2116 mov x2, xSELF // Pass Thread::Current. 2117 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 2118 RESTORE_SAVE_REFS_ONLY_FRAME 2119 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2120END \name 2121.endm 2122 2123GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_region_tlab, artAllocObjectFromCodeRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH, 0 2124GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_resolved_region_tlab, artAllocObjectFromCodeResolvedRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED, 1 2125GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_initialized_region_tlab, artAllocObjectFromCodeInitializedRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED, 1 2126 2127// The common code for art_quick_alloc_array_*region_tlab 2128.macro GENERATE_ALLOC_ARRAY_REGION_TLAB name, entrypoint, fast_path, is_resolved 2129ENTRY \name 2130 // Fast path array allocation for region tlab allocation. 2131 // x0: uint32_t type_idx 2132 // x1: int32_t component_count 2133 // x2: ArtMethod* method 2134 // x3-x7: free. 2135#if !defined(USE_READ_BARRIER) 2136 mvn x0, xzr // Read barrier must be enabled here. 2137 ret // Return -1. 2138#endif 2139.if \is_resolved 2140 mov x3, x0 2141 // If already resolved, class is stored in x0 2142.else 2143 ldr x3, [x2, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2144 // Load the class (x2) 2145 ldr w3, [x3, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2146.endif 2147 // Most common case: GC is not marking. 2148 ldr w4, [xSELF, #THREAD_IS_GC_MARKING_OFFSET] 2149 cbnz x4, .Lmarking\name 2150.Ldo_allocation\name: 2151 \fast_path .Lslow_path\name, x3, w3, x1, w1, x4, w4, x5, w5, x6, w6 2152.Lmarking\name: 2153 // GC is marking, check the lock word of the class for the mark bit. 2154 // If the class is null, go slow path. The check is required to read the lock word. 2155 cbz w3, .Lslow_path\name 2156 // Class is not null, check mark bit in lock word. 2157 ldr w4, [x3, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2158 // If the bit is not zero, do the allocation. 2159 tbnz w4, #LOCK_WORD_MARK_BIT_SHIFT, .Ldo_allocation\name 2160 // The read barrier slow path. Mark 2161 // the class. 2162 stp x0, x1, [sp, #-32]! // Save registers (x0, x1, x2, lr). 2163 stp x2, xLR, [sp, #16] 2164 mov x0, x3 // Pass the class as the first param. 2165 bl artReadBarrierMark 2166 mov x3, x0 // Get the (marked) class back. 2167 ldp x2, xLR, [sp, #16] 2168 ldp x0, x1, [sp], #32 // Restore registers. 2169 b .Ldo_allocation\name 2170.Lslow_path\name: 2171 // x0: uint32_t type_idx / mirror::Class* klass (if resolved) 2172 // x1: int32_t component_count 2173 // x2: ArtMethod* method 2174 // x3: Thread* self 2175 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 2176 mov x3, xSELF // pass Thread::Current 2177 bl \entrypoint 2178 RESTORE_SAVE_REFS_ONLY_FRAME 2179 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2180END \name 2181.endm 2182 2183GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_region_tlab, artAllocArrayFromCodeRegionTLAB, ALLOC_ARRAY_TLAB_FAST_PATH, 0 2184// TODO: art_quick_alloc_array_resolved_region_tlab seems to not get called. Investigate compiler. 2185GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED, 1 2186 2187 /* 2188 * Called by managed code when the thread has been asked to suspend. 2189 */ 2190 .extern artTestSuspendFromCode 2191ENTRY art_quick_test_suspend 2192 SETUP_SAVE_EVERYTHING_FRAME // save callee saves for stack crawl 2193 mov x0, xSELF 2194 bl artTestSuspendFromCode // (Thread*) 2195 RESTORE_SAVE_EVERYTHING_FRAME 2196 ret 2197END art_quick_test_suspend 2198 2199ENTRY art_quick_implicit_suspend 2200 mov x0, xSELF 2201 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves for stack crawl 2202 bl artTestSuspendFromCode // (Thread*) 2203 RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN 2204END art_quick_implicit_suspend 2205 2206 /* 2207 * Called by managed code that is attempting to call a method on a proxy class. On entry 2208 * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy 2209 * method agrees with a ref and args callee save frame. 2210 */ 2211 .extern artQuickProxyInvokeHandler 2212ENTRY art_quick_proxy_invoke_handler 2213 SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 2214 mov x2, xSELF // pass Thread::Current 2215 mov x3, sp // pass SP 2216 bl artQuickProxyInvokeHandler // (Method* proxy method, receiver, Thread*, SP) 2217 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 2218 cbnz x2, .Lexception_in_proxy // success if no exception is pending 2219 RESTORE_SAVE_REFS_AND_ARGS_FRAME // Restore frame 2220 fmov d0, x0 // Store result in d0 in case it was float or double 2221 ret // return on success 2222.Lexception_in_proxy: 2223 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2224 DELIVER_PENDING_EXCEPTION 2225END art_quick_proxy_invoke_handler 2226 2227 /* 2228 * Called to resolve an imt conflict. 2229 * x0 is the conflict ArtMethod. 2230 * xIP1 is a hidden argument that holds the target interface method's dex method index. 2231 * 2232 * Note that this stub writes to xIP0, xIP1, and x0. 2233 */ 2234 .extern artInvokeInterfaceTrampoline 2235ENTRY art_quick_imt_conflict_trampoline 2236 ldr xIP0, [sp, #0] // Load referrer 2237 ldr xIP0, [xIP0, #ART_METHOD_DEX_CACHE_METHODS_OFFSET_64] // Load dex cache methods array 2238 ldr xIP0, [xIP0, xIP1, lsl #POINTER_SIZE_SHIFT] // Load interface method 2239 ldr xIP1, [x0, #ART_METHOD_JNI_OFFSET_64] // Load ImtConflictTable 2240 ldr x0, [xIP1] // Load first entry in ImtConflictTable. 2241.Limt_table_iterate: 2242 cmp x0, xIP0 2243 // Branch if found. Benchmarks have shown doing a branch here is better. 2244 beq .Limt_table_found 2245 // If the entry is null, the interface method is not in the ImtConflictTable. 2246 cbz x0, .Lconflict_trampoline 2247 // Iterate over the entries of the ImtConflictTable. 2248 ldr x0, [xIP1, #(2 * __SIZEOF_POINTER__)]! 2249 b .Limt_table_iterate 2250.Limt_table_found: 2251 // We successfully hit an entry in the table. Load the target method 2252 // and jump to it. 2253 ldr x0, [xIP1, #__SIZEOF_POINTER__] 2254 ldr xIP0, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 2255 br xIP0 2256.Lconflict_trampoline: 2257 // Call the runtime stub to populate the ImtConflictTable and jump to the 2258 // resolved method. 2259 INVOKE_TRAMPOLINE_BODY artInvokeInterfaceTrampoline 2260END art_quick_imt_conflict_trampoline 2261 2262ENTRY art_quick_resolution_trampoline 2263 SETUP_SAVE_REFS_AND_ARGS_FRAME 2264 mov x2, xSELF 2265 mov x3, sp 2266 bl artQuickResolutionTrampoline // (called, receiver, Thread*, SP) 2267 cbz x0, 1f 2268 mov xIP0, x0 // Remember returned code pointer in xIP0. 2269 ldr x0, [sp, #0] // artQuickResolutionTrampoline puts called method in *SP. 2270 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2271 br xIP0 22721: 2273 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2274 DELIVER_PENDING_EXCEPTION 2275END art_quick_resolution_trampoline 2276 2277/* 2278 * Generic JNI frame layout: 2279 * 2280 * #-------------------# 2281 * | | 2282 * | caller method... | 2283 * #-------------------# <--- SP on entry 2284 * | Return X30/LR | 2285 * | X29/FP | callee save 2286 * | X28 | callee save 2287 * | X27 | callee save 2288 * | X26 | callee save 2289 * | X25 | callee save 2290 * | X24 | callee save 2291 * | X23 | callee save 2292 * | X22 | callee save 2293 * | X21 | callee save 2294 * | X20 | callee save 2295 * | X19 | callee save 2296 * | X7 | arg7 2297 * | X6 | arg6 2298 * | X5 | arg5 2299 * | X4 | arg4 2300 * | X3 | arg3 2301 * | X2 | arg2 2302 * | X1 | arg1 2303 * | D7 | float arg 8 2304 * | D6 | float arg 7 2305 * | D5 | float arg 6 2306 * | D4 | float arg 5 2307 * | D3 | float arg 4 2308 * | D2 | float arg 3 2309 * | D1 | float arg 2 2310 * | D0 | float arg 1 2311 * | Method* | <- X0 2312 * #-------------------# 2313 * | local ref cookie | // 4B 2314 * | handle scope size | // 4B 2315 * #-------------------# 2316 * | JNI Call Stack | 2317 * #-------------------# <--- SP on native call 2318 * | | 2319 * | Stack for Regs | The trampoline assembly will pop these values 2320 * | | into registers for native call 2321 * #-------------------# 2322 * | Native code ptr | 2323 * #-------------------# 2324 * | Free scratch | 2325 * #-------------------# 2326 * | Ptr to (1) | <--- SP 2327 * #-------------------# 2328 */ 2329 /* 2330 * Called to do a generic JNI down-call 2331 */ 2332ENTRY art_quick_generic_jni_trampoline 2333 SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 2334 2335 // Save SP , so we can have static CFI info. 2336 mov x28, sp 2337 .cfi_def_cfa_register x28 2338 2339 // This looks the same, but is different: this will be updated to point to the bottom 2340 // of the frame when the handle scope is inserted. 2341 mov xFP, sp 2342 2343 mov xIP0, #5120 2344 sub sp, sp, xIP0 2345 2346 // prepare for artQuickGenericJniTrampoline call 2347 // (Thread*, SP) 2348 // x0 x1 <= C calling convention 2349 // xSELF xFP <= where they are 2350 2351 mov x0, xSELF // Thread* 2352 mov x1, xFP 2353 bl artQuickGenericJniTrampoline // (Thread*, sp) 2354 2355 // The C call will have registered the complete save-frame on success. 2356 // The result of the call is: 2357 // x0: pointer to native code, 0 on error. 2358 // x1: pointer to the bottom of the used area of the alloca, can restore stack till there. 2359 2360 // Check for error = 0. 2361 cbz x0, .Lexception_in_native 2362 2363 // Release part of the alloca. 2364 mov sp, x1 2365 2366 // Save the code pointer 2367 mov xIP0, x0 2368 2369 // Load parameters from frame into registers. 2370 // TODO Check with artQuickGenericJniTrampoline. 2371 // Also, check again APPCS64 - the stack arguments are interleaved. 2372 ldp x0, x1, [sp] 2373 ldp x2, x3, [sp, #16] 2374 ldp x4, x5, [sp, #32] 2375 ldp x6, x7, [sp, #48] 2376 2377 ldp d0, d1, [sp, #64] 2378 ldp d2, d3, [sp, #80] 2379 ldp d4, d5, [sp, #96] 2380 ldp d6, d7, [sp, #112] 2381 2382 add sp, sp, #128 2383 2384 blr xIP0 // native call. 2385 2386 // result sign extension is handled in C code 2387 // prepare for artQuickGenericJniEndTrampoline call 2388 // (Thread*, result, result_f) 2389 // x0 x1 x2 <= C calling convention 2390 mov x1, x0 // Result (from saved). 2391 mov x0, xSELF // Thread register. 2392 fmov x2, d0 // d0 will contain floating point result, but needs to go into x2 2393 2394 bl artQuickGenericJniEndTrampoline 2395 2396 // Pending exceptions possible. 2397 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 2398 cbnz x2, .Lexception_in_native 2399 2400 // Tear down the alloca. 2401 mov sp, x28 2402 .cfi_def_cfa_register sp 2403 2404 // Tear down the callee-save frame. 2405 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2406 2407 // store into fpr, for when it's a fpr return... 2408 fmov d0, x0 2409 ret 2410 2411.Lexception_in_native: 2412 // Move to x1 then sp to please assembler. 2413 ldr x1, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 2414 mov sp, x1 2415 .cfi_def_cfa_register sp 2416 # This will create a new save-all frame, required by the runtime. 2417 DELIVER_PENDING_EXCEPTION 2418END art_quick_generic_jni_trampoline 2419 2420/* 2421 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those 2422 * of a quick call: 2423 * x0 = method being called/to bridge to. 2424 * x1..x7, d0..d7 = arguments to that method. 2425 */ 2426ENTRY art_quick_to_interpreter_bridge 2427 SETUP_SAVE_REFS_AND_ARGS_FRAME // Set up frame and save arguments. 2428 2429 // x0 will contain mirror::ArtMethod* method. 2430 mov x1, xSELF // How to get Thread::Current() ??? 2431 mov x2, sp 2432 2433 // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self, 2434 // mirror::ArtMethod** sp) 2435 bl artQuickToInterpreterBridge 2436 2437 RESTORE_SAVE_REFS_AND_ARGS_FRAME // TODO: no need to restore arguments in this case. 2438 2439 fmov d0, x0 2440 2441 RETURN_OR_DELIVER_PENDING_EXCEPTION 2442END art_quick_to_interpreter_bridge 2443 2444 2445// 2446// Instrumentation-related stubs 2447// 2448 .extern artInstrumentationMethodEntryFromCode 2449ENTRY art_quick_instrumentation_entry 2450 SETUP_SAVE_REFS_AND_ARGS_FRAME 2451 2452 mov x20, x0 // Preserve method reference in a callee-save. 2453 2454 mov x2, xSELF 2455 mov x3, xLR 2456 bl artInstrumentationMethodEntryFromCode // (Method*, Object*, Thread*, LR) 2457 2458 mov xIP0, x0 // x0 = result of call. 2459 mov x0, x20 // Reload method reference. 2460 2461 RESTORE_SAVE_REFS_AND_ARGS_FRAME // Note: will restore xSELF 2462 adr xLR, art_quick_instrumentation_exit 2463 br xIP0 // Tail-call method with lr set to art_quick_instrumentation_exit. 2464END art_quick_instrumentation_entry 2465 2466 .extern artInstrumentationMethodExitFromCode 2467ENTRY art_quick_instrumentation_exit 2468 mov xLR, #0 // Clobber LR for later checks. 2469 2470 SETUP_SAVE_REFS_ONLY_FRAME 2471 2472 // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then 2473 // we would need to fully restore it. As there are a lot of callee-save registers, it seems 2474 // easier to have an extra small stack area. 2475 2476 str x0, [sp, #-16]! // Save integer result. 2477 .cfi_adjust_cfa_offset 16 2478 str d0, [sp, #8] // Save floating-point result. 2479 2480 add x1, sp, #16 // Pass SP. 2481 mov x2, x0 // Pass integer result. 2482 fmov x3, d0 // Pass floating-point result. 2483 mov x0, xSELF // Pass Thread. 2484 bl artInstrumentationMethodExitFromCode // (Thread*, SP, gpr_res, fpr_res) 2485 2486 mov xIP0, x0 // Return address from instrumentation call. 2487 mov xLR, x1 // r1 is holding link register if we're to bounce to deoptimize 2488 2489 ldr d0, [sp, #8] // Restore floating-point result. 2490 ldr x0, [sp], 16 // Restore integer result, and drop stack area. 2491 .cfi_adjust_cfa_offset 16 2492 2493 POP_SAVE_REFS_ONLY_FRAME 2494 2495 br xIP0 // Tail-call out. 2496END art_quick_instrumentation_exit 2497 2498 /* 2499 * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization 2500 * will long jump to the upcall with a special exception of -1. 2501 */ 2502 .extern artDeoptimize 2503ENTRY art_quick_deoptimize 2504 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 2505 mov x0, xSELF // Pass thread. 2506 bl artDeoptimize // artDeoptimize(Thread*) 2507 brk 0 2508END art_quick_deoptimize 2509 2510 /* 2511 * Compiled code has requested that we deoptimize into the interpreter. The deoptimization 2512 * will long jump to the upcall with a special exception of -1. 2513 */ 2514 .extern artDeoptimizeFromCompiledCode 2515ENTRY art_quick_deoptimize_from_compiled_code 2516 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 2517 mov x0, xSELF // Pass thread. 2518 bl artDeoptimizeFromCompiledCode // artDeoptimizeFromCompiledCode(Thread*) 2519 brk 0 2520END art_quick_deoptimize_from_compiled_code 2521 2522 2523 /* 2524 * String's indexOf. 2525 * 2526 * TODO: Not very optimized. 2527 * On entry: 2528 * x0: string object (known non-null) 2529 * w1: char to match (known <= 0xFFFF) 2530 * w2: Starting offset in string data 2531 */ 2532ENTRY art_quick_indexof 2533 ldr w3, [x0, #MIRROR_STRING_COUNT_OFFSET] 2534 add x0, x0, #MIRROR_STRING_VALUE_OFFSET 2535 2536 /* Clamp start to [0..count] */ 2537 cmp w2, #0 2538 csel w2, wzr, w2, lt 2539 cmp w2, w3 2540 csel w2, w3, w2, gt 2541 2542 /* Save a copy to compute result */ 2543 mov x5, x0 2544 2545 /* Build pointer to start of data to compare and pre-bias */ 2546 add x0, x0, x2, lsl #1 2547 sub x0, x0, #2 2548 2549 /* Compute iteration count */ 2550 sub w2, w3, w2 2551 2552 /* 2553 * At this point we have: 2554 * x0: start of the data to test 2555 * w1: char to compare 2556 * w2: iteration count 2557 * x5: original start of string data 2558 */ 2559 2560 subs w2, w2, #4 2561 b.lt .Lindexof_remainder 2562 2563.Lindexof_loop4: 2564 ldrh w6, [x0, #2]! 2565 ldrh w7, [x0, #2]! 2566 ldrh wIP0, [x0, #2]! 2567 ldrh wIP1, [x0, #2]! 2568 cmp w6, w1 2569 b.eq .Lmatch_0 2570 cmp w7, w1 2571 b.eq .Lmatch_1 2572 cmp wIP0, w1 2573 b.eq .Lmatch_2 2574 cmp wIP1, w1 2575 b.eq .Lmatch_3 2576 subs w2, w2, #4 2577 b.ge .Lindexof_loop4 2578 2579.Lindexof_remainder: 2580 adds w2, w2, #4 2581 b.eq .Lindexof_nomatch 2582 2583.Lindexof_loop1: 2584 ldrh w6, [x0, #2]! 2585 cmp w6, w1 2586 b.eq .Lmatch_3 2587 subs w2, w2, #1 2588 b.ne .Lindexof_loop1 2589 2590.Lindexof_nomatch: 2591 mov x0, #-1 2592 ret 2593 2594.Lmatch_0: 2595 sub x0, x0, #6 2596 sub x0, x0, x5 2597 asr x0, x0, #1 2598 ret 2599.Lmatch_1: 2600 sub x0, x0, #4 2601 sub x0, x0, x5 2602 asr x0, x0, #1 2603 ret 2604.Lmatch_2: 2605 sub x0, x0, #2 2606 sub x0, x0, x5 2607 asr x0, x0, #1 2608 ret 2609.Lmatch_3: 2610 sub x0, x0, x5 2611 asr x0, x0, #1 2612 ret 2613END art_quick_indexof 2614 2615 /* 2616 * Create a function `name` calling the ReadBarrier::Mark routine, 2617 * getting its argument and returning its result through W register 2618 * `wreg` (corresponding to X register `xreg`), saving and restoring 2619 * all caller-save registers. 2620 * 2621 * If `wreg` is different from `w0`, the generated function follows a 2622 * non-standard runtime calling convention: 2623 * - register `wreg` is used to pass the (sole) argument of this 2624 * function (instead of W0); 2625 * - register `wreg` is used to return the result of this function 2626 * (instead of W0); 2627 * - W0 is treated like a normal (non-argument) caller-save register; 2628 * - everything else is the same as in the standard runtime calling 2629 * convention (e.g. standard callee-save registers are preserved). 2630 */ 2631.macro READ_BARRIER_MARK_REG name, wreg, xreg 2632ENTRY \name 2633 // Reference is null, no work to do at all. 2634 cbz \wreg, .Lret_rb_\name 2635 /* 2636 * Allocate 46 stack slots * 8 = 368 bytes: 2637 * - 20 slots for core registers X0-X19 2638 * - 24 slots for floating-point registers D0-D7 and D16-D31 2639 * - 1 slot for return address register XLR 2640 * - 1 padding slot for 16-byte stack alignment 2641 */ 2642 // Use wIP0 as temp and check the mark bit of the reference. wIP0 is not used by the compiler. 2643 ldr wIP0, [\xreg, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2644 tbz wIP0, #LOCK_WORD_MARK_BIT_SHIFT, .Lslow_path_rb_\name 2645 ret 2646.Lslow_path_rb_\name: 2647 // Save all potentially live caller-save core registers. 2648 stp x0, x1, [sp, #-368]! 2649 .cfi_adjust_cfa_offset 368 2650 .cfi_rel_offset x0, 0 2651 .cfi_rel_offset x1, 8 2652 stp x2, x3, [sp, #16] 2653 .cfi_rel_offset x2, 16 2654 .cfi_rel_offset x3, 24 2655 stp x4, x5, [sp, #32] 2656 .cfi_rel_offset x4, 32 2657 .cfi_rel_offset x5, 40 2658 stp x6, x7, [sp, #48] 2659 .cfi_rel_offset x6, 48 2660 .cfi_rel_offset x7, 56 2661 stp x8, x9, [sp, #64] 2662 .cfi_rel_offset x8, 64 2663 .cfi_rel_offset x9, 72 2664 stp x10, x11, [sp, #80] 2665 .cfi_rel_offset x10, 80 2666 .cfi_rel_offset x11, 88 2667 stp x12, x13, [sp, #96] 2668 .cfi_rel_offset x12, 96 2669 .cfi_rel_offset x13, 104 2670 stp x14, x15, [sp, #112] 2671 .cfi_rel_offset x14, 112 2672 .cfi_rel_offset x15, 120 2673 stp x16, x17, [sp, #128] 2674 .cfi_rel_offset x16, 128 2675 .cfi_rel_offset x17, 136 2676 stp x18, x19, [sp, #144] 2677 .cfi_rel_offset x18, 144 2678 .cfi_rel_offset x19, 152 2679 // Save all potentially live caller-save floating-point registers. 2680 stp d0, d1, [sp, #160] 2681 stp d2, d3, [sp, #176] 2682 stp d4, d5, [sp, #192] 2683 stp d6, d7, [sp, #208] 2684 stp d16, d17, [sp, #224] 2685 stp d18, d19, [sp, #240] 2686 stp d20, d21, [sp, #256] 2687 stp d22, d23, [sp, #272] 2688 stp d24, d25, [sp, #288] 2689 stp d26, d27, [sp, #304] 2690 stp d28, d29, [sp, #320] 2691 stp d30, d31, [sp, #336] 2692 // Save return address. 2693 str xLR, [sp, #352] 2694 .cfi_rel_offset x30, 352 2695 // (sp + #360 is a padding slot) 2696 2697 .ifnc \wreg, w0 2698 mov w0, \wreg // Pass arg1 - obj from `wreg` 2699 .endif 2700 bl artReadBarrierMark // artReadBarrierMark(obj) 2701 .ifnc \wreg, w0 2702 mov \wreg, w0 // Return result into `wreg` 2703 .endif 2704 2705 // Restore core regs, except `xreg`, as `wreg` is used to return the 2706 // result of this function (simply remove it from the stack instead). 2707 POP_REGS_NE x0, x1, 0, \xreg 2708 POP_REGS_NE x2, x3, 16, \xreg 2709 POP_REGS_NE x4, x5, 32, \xreg 2710 POP_REGS_NE x6, x7, 48, \xreg 2711 POP_REGS_NE x8, x9, 64, \xreg 2712 POP_REGS_NE x10, x11, 80, \xreg 2713 POP_REGS_NE x12, x13, 96, \xreg 2714 POP_REGS_NE x14, x15, 112, \xreg 2715 POP_REGS_NE x16, x17, 128, \xreg 2716 POP_REGS_NE x18, x19, 144, \xreg 2717 // Restore floating-point registers. 2718 ldp d0, d1, [sp, #160] 2719 ldp d2, d3, [sp, #176] 2720 ldp d4, d5, [sp, #192] 2721 ldp d6, d7, [sp, #208] 2722 ldp d16, d17, [sp, #224] 2723 ldp d18, d19, [sp, #240] 2724 ldp d20, d21, [sp, #256] 2725 ldp d22, d23, [sp, #272] 2726 ldp d24, d25, [sp, #288] 2727 ldp d26, d27, [sp, #304] 2728 ldp d28, d29, [sp, #320] 2729 ldp d30, d31, [sp, #336] 2730 // Restore return address and remove padding. 2731 ldr xLR, [sp, #352] 2732 .cfi_restore x30 2733 add sp, sp, #368 2734 .cfi_adjust_cfa_offset -368 2735.Lret_rb_\name: 2736 ret 2737END \name 2738.endm 2739 2740READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg00, w0, x0 2741READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg01, w1, x1 2742READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg02, w2, x2 2743READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg03, w3, x3 2744READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg04, w4, x4 2745READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg05, w5, x5 2746READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg06, w6, x6 2747READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg07, w7, x7 2748READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg08, w8, x8 2749READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg09, w9, x9 2750READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg10, w10, x10 2751READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg11, w11, x11 2752READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg12, w12, x12 2753READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg13, w13, x13 2754READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg14, w14, x14 2755READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg15, w15, x15 2756// READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg16, w16, x16 ip0 is blocked 2757READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg17, w17, x17 2758READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg18, w18, x18 2759READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg19, w19, x19 2760READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg20, w20, x20 2761READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg21, w21, x21 2762READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg22, w22, x22 2763READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg23, w23, x23 2764READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg24, w24, x24 2765READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg25, w25, x25 2766READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg26, w26, x26 2767READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg27, w27, x27 2768READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg28, w28, x28 2769READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg29, w29, x29 2770