quick_entrypoints_arm64.S revision 2ee98f2172df27c0e57738e214d7a1f0739ac916
1/* 2 * Copyright (C) 2014 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17#include "asm_support_arm64.S" 18 19#include "arch/quick_alloc_entrypoints.S" 20 21 22 /* 23 * Macro that sets up the callee save frame to conform with 24 * Runtime::CreateCalleeSaveMethod(kSaveAll) 25 */ 26.macro SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 27 adrp xIP0, :got:_ZN3art7Runtime9instance_E 28 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 29 30 // Our registers aren't intermixed - just spill in order. 31 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 32 33 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 34 // Loads appropriate callee-save-method. 35 ldr xIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVE_FRAME_OFFSET ] 36 37 sub sp, sp, #176 38 .cfi_adjust_cfa_offset 176 39 40 // Ugly compile-time check, but we only have the preprocessor. 41#if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVE != 176) 42#error "SAVE_ALL_CALLEE_SAVE_FRAME(ARM64) size not as expected." 43#endif 44 45 // 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_[kRefsAndArgs]. 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(kRefsOnly). 87 */ 88.macro SETUP_REFS_ONLY_CALLEE_SAVE_FRAME 89 adrp xIP0, :got:_ZN3art7Runtime9instance_E 90 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 91 92 // Our registers aren't intermixed - just spill in order. 93 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 94 95 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefOnly] . 96 // Loads appropriate callee-save-method. 97 ldr xIP0, [xIP0, RUNTIME_REFS_ONLY_CALLEE_SAVE_FRAME_OFFSET ] 98 99 sub sp, sp, #96 100 .cfi_adjust_cfa_offset 96 101 102 // Ugly compile-time check, but we only have the preprocessor. 103#if (FRAME_SIZE_REFS_ONLY_CALLEE_SAVE != 96) 104#error "REFS_ONLY_CALLEE_SAVE_FRAME(ARM64) size not as expected." 105#endif 106 107 // 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_[kRefsOnly]. 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_REFS_ONLY_CALLEE_SAVE_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_REFS_ONLY_CALLEE_SAVE_FRAME 169 add sp, sp, #96 170 .cfi_adjust_cfa_offset - 96 171.endm 172 173.macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 174 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 175 ret 176.endm 177 178 179.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_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_REFS_AND_ARGS_CALLEE_SAVE != 224) 185#error "REFS_AND_ARGS_CALLEE_SAVE_FRAME(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(kRefsAndArgs). 239 * 240 * TODO This is probably too conservative - saving FP & LR. 241 */ 242.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 243 adrp xIP0, :got:_ZN3art7Runtime9instance_E 244 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 245 246 // Our registers aren't intermixed - just spill in order. 247 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 248 249 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 250 ldr xIP0, [xIP0, RUNTIME_REFS_AND_ARGS_CALLEE_SAVE_FRAME_OFFSET ] 251 252 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 253 254 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs] 255 // Place sp in Thread::Current()->top_quick_frame. 256 mov xIP0, sp 257 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 258.endm 259 260.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 261 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 262 str x0, [sp, #0] // Store ArtMethod* to bottom of stack. 263 // Place sp in Thread::Current()->top_quick_frame. 264 mov xIP0, sp 265 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 266.endm 267 268// TODO: Probably no need to restore registers preserved by aapcs64. 269.macro RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 270 // FP args. 271 ldp d0, d1, [sp, #16] 272 ldp d2, d3, [sp, #32] 273 ldp d4, d5, [sp, #48] 274 ldp d6, d7, [sp, #64] 275 276 // Core args. 277 ldp x1, x2, [sp, #80] 278 .cfi_restore x1 279 .cfi_restore x2 280 281 ldp x3, x4, [sp, #96] 282 .cfi_restore x3 283 .cfi_restore x4 284 285 ldp x5, x6, [sp, #112] 286 .cfi_restore x5 287 .cfi_restore x6 288 289 // x7, Callee-saves. 290 ldp x7, x20, [sp, #128] 291 .cfi_restore x7 292 .cfi_restore x20 293 294 ldp x21, x22, [sp, #144] 295 .cfi_restore x21 296 .cfi_restore x22 297 298 ldp x23, x24, [sp, #160] 299 .cfi_restore x23 300 .cfi_restore x24 301 302 ldp x25, x26, [sp, #176] 303 .cfi_restore x25 304 .cfi_restore x26 305 306 ldp x27, x28, [sp, #192] 307 .cfi_restore x27 308 .cfi_restore x28 309 310 // x29(callee-save) and LR. 311 ldp x29, xLR, [sp, #208] 312 .cfi_restore x29 313 .cfi_restore x30 314 315 add sp, sp, #224 316 .cfi_adjust_cfa_offset -224 317.endm 318 319 /* 320 * Macro that sets up the callee save frame to conform with 321 * Runtime::CreateCalleeSaveMethod(kSaveEverything) 322 */ 323.macro SETUP_SAVE_EVERYTHING_CALLEE_SAVE_FRAME 324 sub sp, sp, #512 325 .cfi_adjust_cfa_offset 512 326 327 // Ugly compile-time check, but we only have the preprocessor. 328#if (FRAME_SIZE_SAVE_EVERYTHING_CALLEE_SAVE != 512) 329#error "SAVE_EVERYTHING_CALLEE_SAVE_FRAME(ARM64) size not as expected." 330#endif 331 332 // Save FP registers. 333 stp d0, d1, [sp, #8] 334 stp d2, d3, [sp, #24] 335 stp d4, d5, [sp, #40] 336 stp d6, d7, [sp, #56] 337 stp d8, d9, [sp, #72] 338 stp d10, d11, [sp, #88] 339 stp d12, d13, [sp, #104] 340 stp d14, d15, [sp, #120] 341 stp d16, d17, [sp, #136] 342 stp d18, d19, [sp, #152] 343 stp d20, d21, [sp, #168] 344 stp d22, d23, [sp, #184] 345 stp d24, d25, [sp, #200] 346 stp d26, d27, [sp, #216] 347 stp d28, d29, [sp, #232] 348 stp d30, d31, [sp, #248] 349 350 // Save core registers. 351 str x0, [sp, #264] 352 .cfi_rel_offset x0, 264 353 354 stp x1, x2, [sp, #272] 355 .cfi_rel_offset x1, 272 356 .cfi_rel_offset x2, 280 357 358 stp x3, x4, [sp, #288] 359 .cfi_rel_offset x3, 288 360 .cfi_rel_offset x4, 296 361 362 stp x5, x6, [sp, #304] 363 .cfi_rel_offset x5, 304 364 .cfi_rel_offset x6, 312 365 366 stp x7, x8, [sp, #320] 367 .cfi_rel_offset x7, 320 368 .cfi_rel_offset x8, 328 369 370 stp x9, x10, [sp, #336] 371 .cfi_rel_offset x9, 336 372 .cfi_rel_offset x10, 344 373 374 stp x11, x12, [sp, #352] 375 .cfi_rel_offset x11, 352 376 .cfi_rel_offset x12, 360 377 378 stp x13, x14, [sp, #368] 379 .cfi_rel_offset x13, 368 380 .cfi_rel_offset x14, 376 381 382 stp x15, x16, [sp, #384] 383 .cfi_rel_offset x15, 384 384 .cfi_rel_offset x16, 392 385 386 stp x17, x18, [sp, #400] 387 .cfi_rel_offset x17, 400 388 .cfi_rel_offset x18, 408 389 390 stp x19, x20, [sp, #416] 391 .cfi_rel_offset x19, 416 392 .cfi_rel_offset x20, 424 393 394 stp x21, x22, [sp, #432] 395 .cfi_rel_offset x21, 432 396 .cfi_rel_offset x22, 440 397 398 stp x23, x24, [sp, #448] 399 .cfi_rel_offset x23, 448 400 .cfi_rel_offset x24, 456 401 402 stp x25, x26, [sp, #464] 403 .cfi_rel_offset x25, 464 404 .cfi_rel_offset x26, 472 405 406 stp x27, x28, [sp, #480] 407 .cfi_rel_offset x27, 480 408 .cfi_rel_offset x28, 488 409 410 stp x29, xLR, [sp, #496] 411 .cfi_rel_offset x29, 496 412 .cfi_rel_offset x30, 504 413 414 adrp xIP0, :got:_ZN3art7Runtime9instance_E 415 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 416 417 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 418 419 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kSaveEverything] . 420 // Loads appropriate callee-save-method. 421 ldr xIP0, [xIP0, RUNTIME_SAVE_EVERYTHING_CALLEE_SAVE_FRAME_OFFSET ] 422 423 // Store ArtMethod* Runtime::callee_save_methods_[kSaveEverything]. 424 str xIP0, [sp] 425 // Place sp in Thread::Current()->top_quick_frame. 426 mov xIP0, sp 427 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 428.endm 429 430.macro RESTORE_SAVE_EVERYTHING_CALLEE_SAVE_FRAME 431 // Restore FP registers. 432 ldp d0, d1, [sp, #8] 433 ldp d2, d3, [sp, #24] 434 ldp d4, d5, [sp, #40] 435 ldp d6, d7, [sp, #56] 436 ldp d8, d9, [sp, #72] 437 ldp d10, d11, [sp, #88] 438 ldp d12, d13, [sp, #104] 439 ldp d14, d15, [sp, #120] 440 ldp d16, d17, [sp, #136] 441 ldp d18, d19, [sp, #152] 442 ldp d20, d21, [sp, #168] 443 ldp d22, d23, [sp, #184] 444 ldp d24, d25, [sp, #200] 445 ldp d26, d27, [sp, #216] 446 ldp d28, d29, [sp, #232] 447 ldp d30, d31, [sp, #248] 448 449 // Restore core registers. 450 ldr x0, [sp, #264] 451 .cfi_restore x0 452 453 ldp x1, x2, [sp, #272] 454 .cfi_restore x1 455 .cfi_restore x2 456 457 ldp x3, x4, [sp, #288] 458 .cfi_restore x3 459 .cfi_restore x4 460 461 ldp x5, x6, [sp, #304] 462 .cfi_restore x5 463 .cfi_restore x6 464 465 ldp x7, x8, [sp, #320] 466 .cfi_restore x7 467 .cfi_restore x8 468 469 ldp x9, x10, [sp, #336] 470 .cfi_restore x9 471 .cfi_restore x10 472 473 ldp x11, x12, [sp, #352] 474 .cfi_restore x11 475 .cfi_restore x12 476 477 ldp x13, x14, [sp, #368] 478 .cfi_restore x13 479 .cfi_restore x14 480 481 ldp x15, x16, [sp, #384] 482 .cfi_restore x15 483 .cfi_restore x16 484 485 ldp x17, x18, [sp, #400] 486 .cfi_restore x17 487 .cfi_restore x18 488 489 ldp x19, x20, [sp, #416] 490 .cfi_restore x19 491 .cfi_restore x20 492 493 ldp x21, x22, [sp, #432] 494 .cfi_restore x21 495 .cfi_restore x22 496 497 ldp x23, x24, [sp, #448] 498 .cfi_restore x23 499 .cfi_restore x24 500 501 ldp x25, x26, [sp, #464] 502 .cfi_restore x25 503 .cfi_restore x26 504 505 ldp x27, x28, [sp, #480] 506 .cfi_restore x27 507 .cfi_restore x28 508 509 ldp x29, xLR, [sp, #496] 510 .cfi_restore x29 511 .cfi_restore x30 512 513 add sp, sp, #512 514 .cfi_adjust_cfa_offset -512 515.endm 516 517.macro RETURN_IF_RESULT_IS_ZERO 518 cbnz x0, 1f // result non-zero branch over 519 ret // return 5201: 521.endm 522 523.macro RETURN_IF_RESULT_IS_NON_ZERO 524 cbz x0, 1f // result zero branch over 525 ret // return 5261: 527.endm 528 529 /* 530 * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending 531 * exception is Thread::Current()->exception_ 532 */ 533.macro DELIVER_PENDING_EXCEPTION 534 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 535 mov x0, xSELF 536 537 // Point of no return. 538 b artDeliverPendingExceptionFromCode // artDeliverPendingExceptionFromCode(Thread*) 539 brk 0 // Unreached 540.endm 541 542.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg 543 ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET] // Get exception field. 544 cbnz \reg, 1f 545 ret 5461: 547 DELIVER_PENDING_EXCEPTION 548.endm 549 550.macro RETURN_OR_DELIVER_PENDING_EXCEPTION 551 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0 552.endm 553 554// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register. 555.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 556 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1 557.endm 558 559.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER 560 cbnz w0, 1f // result non-zero branch over 561 ret // return 5621: 563 DELIVER_PENDING_EXCEPTION 564.endm 565 566.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 567 .extern \cxx_name 568ENTRY \c_name 569 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 570 mov x0, xSELF // pass Thread::Current 571 b \cxx_name // \cxx_name(Thread*) 572END \c_name 573.endm 574 575.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name 576 .extern \cxx_name 577ENTRY \c_name 578 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context. 579 mov x1, xSELF // pass Thread::Current. 580 b \cxx_name // \cxx_name(arg, Thread*). 581 brk 0 582END \c_name 583.endm 584 585.macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 586 .extern \cxx_name 587ENTRY \c_name 588 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 589 mov x2, xSELF // pass Thread::Current 590 b \cxx_name // \cxx_name(arg1, arg2, Thread*) 591 brk 0 592END \c_name 593.endm 594 595 /* 596 * Called by managed code, saves callee saves and then calls artThrowException 597 * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. 598 */ 599ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode 600 601 /* 602 * Called by managed code to create and deliver a NullPointerException. 603 */ 604NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode 605 606 /* 607 * Call installed by a signal handler to create and deliver a NullPointerException. 608 */ 609ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception_from_signal, artThrowNullPointerExceptionFromSignal 610 611 /* 612 * Called by managed code to create and deliver an ArithmeticException. 613 */ 614NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode 615 616 /* 617 * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds 618 * index, arg2 holds limit. 619 */ 620TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode 621 622 /* 623 * Called by managed code to create and deliver a StringIndexOutOfBoundsException 624 * as if thrown from a call to String.charAt(). Arg1 holds index, arg2 holds limit. 625 */ 626TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_string_bounds, artThrowStringBoundsFromCode 627 628 /* 629 * Called by managed code to create and deliver a StackOverflowError. 630 */ 631NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode 632 633 /* 634 * Called by managed code to create and deliver a NoSuchMethodError. 635 */ 636ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode 637 638 /* 639 * All generated callsites for interface invokes and invocation slow paths will load arguments 640 * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain 641 * the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper. 642 * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1. 643 * 644 * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting 645 * of the target Method* in x0 and method->code_ in x1. 646 * 647 * If unsuccessful, the helper will return null/????. There will be a pending exception in the 648 * thread and we branch to another stub to deliver it. 649 * 650 * On success this wrapper will restore arguments and *jump* to the target, leaving the lr 651 * pointing back to the original caller. 652 * 653 * Adapted from ARM32 code. 654 * 655 * Clobbers xIP0. 656 */ 657.macro INVOKE_TRAMPOLINE_BODY cxx_name 658 .extern \cxx_name 659 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME // save callee saves in case allocation triggers GC 660 // Helper signature is always 661 // (method_idx, *this_object, *caller_method, *self, sp) 662 663 mov x2, xSELF // pass Thread::Current 664 mov x3, sp 665 bl \cxx_name // (method_idx, this, Thread*, SP) 666 mov xIP0, x1 // save Method*->code_ 667 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 668 cbz x0, 1f // did we find the target? if not go to exception delivery 669 br xIP0 // tail call to target 6701: 671 DELIVER_PENDING_EXCEPTION 672.endm 673.macro INVOKE_TRAMPOLINE c_name, cxx_name 674ENTRY \c_name 675 INVOKE_TRAMPOLINE_BODY \cxx_name 676END \c_name 677.endm 678 679INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck 680 681INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck 682INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck 683INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck 684INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck 685 686 687.macro INVOKE_STUB_CREATE_FRAME 688 689SAVE_SIZE=15*8 // x4, x5, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 690SAVE_SIZE_AND_METHOD=SAVE_SIZE+8 691 692 693 mov x9, sp // Save stack pointer. 694 .cfi_register sp,x9 695 696 add x10, x2, # SAVE_SIZE_AND_METHOD // calculate size of frame. 697 sub x10, sp, x10 // Calculate SP position - saves + ArtMethod* + args 698 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 699 mov sp, x10 // Set new SP. 700 701 sub x10, x9, #SAVE_SIZE // Calculate new FP (later). Done here as we must move SP 702 .cfi_def_cfa_register x10 // before this. 703 .cfi_adjust_cfa_offset SAVE_SIZE 704 705 str x28, [x10, #112] 706 .cfi_rel_offset x28, 112 707 708 stp x26, x27, [x10, #96] 709 .cfi_rel_offset x26, 96 710 .cfi_rel_offset x27, 104 711 712 stp x24, x25, [x10, #80] 713 .cfi_rel_offset x24, 80 714 .cfi_rel_offset x25, 88 715 716 stp x22, x23, [x10, #64] 717 .cfi_rel_offset x22, 64 718 .cfi_rel_offset x23, 72 719 720 stp x20, x21, [x10, #48] 721 .cfi_rel_offset x20, 48 722 .cfi_rel_offset x21, 56 723 724 stp x9, x19, [x10, #32] // Save old stack pointer and x19. 725 .cfi_rel_offset sp, 32 726 .cfi_rel_offset x19, 40 727 728 stp x4, x5, [x10, #16] // Save result and shorty addresses. 729 .cfi_rel_offset x4, 16 730 .cfi_rel_offset x5, 24 731 732 stp xFP, xLR, [x10] // Store LR & FP. 733 .cfi_rel_offset x29, 0 734 .cfi_rel_offset x30, 8 735 736 mov xFP, x10 // Use xFP now, as it's callee-saved. 737 .cfi_def_cfa_register x29 738 mov xSELF, x3 // Move thread pointer into SELF register. 739 740 // Copy arguments into stack frame. 741 // Use simple copy routine for now. 742 // 4 bytes per slot. 743 // X1 - source address 744 // W2 - args length 745 // X9 - destination address. 746 // W10 - temporary 747 add x9, sp, #8 // Destination address is bottom of stack + null. 748 749 // Copy parameters into the stack. Use numeric label as this is a macro and Clang's assembler 750 // does not have unique-id variables. 7511: 752 cmp w2, #0 753 beq 2f 754 sub w2, w2, #4 // Need 65536 bytes of range. 755 ldr w10, [x1, x2] 756 str w10, [x9, x2] 757 758 b 1b 759 7602: 761 // Store null into ArtMethod* at bottom of frame. 762 str xzr, [sp] 763.endm 764 765.macro INVOKE_STUB_CALL_AND_RETURN 766 767 // load method-> METHOD_QUICK_CODE_OFFSET 768 ldr x9, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 769 // Branch to method. 770 blr x9 771 772 // Restore return value address and shorty address. 773 ldp x4,x5, [xFP, #16] 774 .cfi_restore x4 775 .cfi_restore x5 776 777 ldr x28, [xFP, #112] 778 .cfi_restore x28 779 780 ldp x26, x27, [xFP, #96] 781 .cfi_restore x26 782 .cfi_restore x27 783 784 ldp x24, x25, [xFP, #80] 785 .cfi_restore x24 786 .cfi_restore x25 787 788 ldp x22, x23, [xFP, #64] 789 .cfi_restore x22 790 .cfi_restore x23 791 792 ldp x20, x21, [xFP, #48] 793 .cfi_restore x20 794 .cfi_restore x21 795 796 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 797 ldrb w10, [x5] 798 799 // Check the return type and store the correct register into the jvalue in memory. 800 // Use numeric label as this is a macro and Clang's assembler does not have unique-id variables. 801 802 // Don't set anything for a void type. 803 cmp w10, #'V' 804 beq 3f 805 806 // Is it a double? 807 cmp w10, #'D' 808 bne 1f 809 str d0, [x4] 810 b 3f 811 8121: // Is it a float? 813 cmp w10, #'F' 814 bne 2f 815 str s0, [x4] 816 b 3f 817 8182: // Just store x0. Doesn't matter if it is 64 or 32 bits. 819 str x0, [x4] 820 8213: // Finish up. 822 ldp x2, x19, [xFP, #32] // Restore stack pointer and x19. 823 .cfi_restore x19 824 mov sp, x2 825 .cfi_restore sp 826 827 ldp xFP, xLR, [xFP] // Restore old frame pointer and link register. 828 .cfi_restore x29 829 .cfi_restore x30 830 831 ret 832 833.endm 834 835 836/* 837 * extern"C" void art_quick_invoke_stub(ArtMethod *method, x0 838 * uint32_t *args, x1 839 * uint32_t argsize, w2 840 * Thread *self, x3 841 * JValue *result, x4 842 * char *shorty); x5 843 * +----------------------+ 844 * | | 845 * | C/C++ frame | 846 * | LR'' | 847 * | FP'' | <- SP' 848 * +----------------------+ 849 * +----------------------+ 850 * | x28 | <- TODO: Remove callee-saves. 851 * | : | 852 * | x19 | 853 * | SP' | 854 * | X5 | 855 * | X4 | Saved registers 856 * | LR' | 857 * | FP' | <- FP 858 * +----------------------+ 859 * | uint32_t out[n-1] | 860 * | : : | Outs 861 * | uint32_t out[0] | 862 * | ArtMethod* | <- SP value=null 863 * +----------------------+ 864 * 865 * Outgoing registers: 866 * x0 - Method* 867 * x1-x7 - integer parameters. 868 * d0-d7 - Floating point parameters. 869 * xSELF = self 870 * SP = & of ArtMethod* 871 * x1 = "this" pointer. 872 * 873 */ 874ENTRY art_quick_invoke_stub 875 // Spill registers as per AACPS64 calling convention. 876 INVOKE_STUB_CREATE_FRAME 877 878 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 879 // Parse the passed shorty to determine which register to load. 880 // Load addresses for routines that load WXSD registers. 881 adr x11, .LstoreW2 882 adr x12, .LstoreX2 883 adr x13, .LstoreS0 884 adr x14, .LstoreD0 885 886 // Initialize routine offsets to 0 for integers and floats. 887 // x8 for integers, x15 for floating point. 888 mov x8, #0 889 mov x15, #0 890 891 add x10, x5, #1 // Load shorty address, plus one to skip return value. 892 ldr w1, [x9],#4 // Load "this" parameter, and increment arg pointer. 893 894 // Loop to fill registers. 895.LfillRegisters: 896 ldrb w17, [x10], #1 // Load next character in signature, and increment. 897 cbz w17, .LcallFunction // Exit at end of signature. Shorty 0 terminated. 898 899 cmp w17, #'F' // is this a float? 900 bne .LisDouble 901 902 cmp x15, # 8*12 // Skip this load if all registers full. 903 beq .Ladvance4 904 905 add x17, x13, x15 // Calculate subroutine to jump to. 906 br x17 907 908.LisDouble: 909 cmp w17, #'D' // is this a double? 910 bne .LisLong 911 912 cmp x15, # 8*12 // Skip this load if all registers full. 913 beq .Ladvance8 914 915 add x17, x14, x15 // Calculate subroutine to jump to. 916 br x17 917 918.LisLong: 919 cmp w17, #'J' // is this a long? 920 bne .LisOther 921 922 cmp x8, # 6*12 // Skip this load if all registers full. 923 beq .Ladvance8 924 925 add x17, x12, x8 // Calculate subroutine to jump to. 926 br x17 927 928.LisOther: // Everything else takes one vReg. 929 cmp x8, # 6*12 // Skip this load if all registers full. 930 beq .Ladvance4 931 932 add x17, x11, x8 // Calculate subroutine to jump to. 933 br x17 934 935.Ladvance4: 936 add x9, x9, #4 937 b .LfillRegisters 938 939.Ladvance8: 940 add x9, x9, #8 941 b .LfillRegisters 942 943// Macro for loading a parameter into a register. 944// counter - the register with offset into these tables 945// size - the size of the register - 4 or 8 bytes. 946// register - the name of the register to be loaded. 947.macro LOADREG counter size register return 948 ldr \register , [x9], #\size 949 add \counter, \counter, 12 950 b \return 951.endm 952 953// Store ints. 954.LstoreW2: 955 LOADREG x8 4 w2 .LfillRegisters 956 LOADREG x8 4 w3 .LfillRegisters 957 LOADREG x8 4 w4 .LfillRegisters 958 LOADREG x8 4 w5 .LfillRegisters 959 LOADREG x8 4 w6 .LfillRegisters 960 LOADREG x8 4 w7 .LfillRegisters 961 962// Store longs. 963.LstoreX2: 964 LOADREG x8 8 x2 .LfillRegisters 965 LOADREG x8 8 x3 .LfillRegisters 966 LOADREG x8 8 x4 .LfillRegisters 967 LOADREG x8 8 x5 .LfillRegisters 968 LOADREG x8 8 x6 .LfillRegisters 969 LOADREG x8 8 x7 .LfillRegisters 970 971// Store singles. 972.LstoreS0: 973 LOADREG x15 4 s0 .LfillRegisters 974 LOADREG x15 4 s1 .LfillRegisters 975 LOADREG x15 4 s2 .LfillRegisters 976 LOADREG x15 4 s3 .LfillRegisters 977 LOADREG x15 4 s4 .LfillRegisters 978 LOADREG x15 4 s5 .LfillRegisters 979 LOADREG x15 4 s6 .LfillRegisters 980 LOADREG x15 4 s7 .LfillRegisters 981 982// Store doubles. 983.LstoreD0: 984 LOADREG x15 8 d0 .LfillRegisters 985 LOADREG x15 8 d1 .LfillRegisters 986 LOADREG x15 8 d2 .LfillRegisters 987 LOADREG x15 8 d3 .LfillRegisters 988 LOADREG x15 8 d4 .LfillRegisters 989 LOADREG x15 8 d5 .LfillRegisters 990 LOADREG x15 8 d6 .LfillRegisters 991 LOADREG x15 8 d7 .LfillRegisters 992 993 994.LcallFunction: 995 996 INVOKE_STUB_CALL_AND_RETURN 997 998END art_quick_invoke_stub 999 1000/* extern"C" 1001 * void art_quick_invoke_static_stub(ArtMethod *method, x0 1002 * uint32_t *args, x1 1003 * uint32_t argsize, w2 1004 * Thread *self, x3 1005 * JValue *result, x4 1006 * char *shorty); x5 1007 */ 1008ENTRY art_quick_invoke_static_stub 1009 // Spill registers as per AACPS64 calling convention. 1010 INVOKE_STUB_CREATE_FRAME 1011 1012 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 1013 // Parse the passed shorty to determine which register to load. 1014 // Load addresses for routines that load WXSD registers. 1015 adr x11, .LstoreW1_2 1016 adr x12, .LstoreX1_2 1017 adr x13, .LstoreS0_2 1018 adr x14, .LstoreD0_2 1019 1020 // Initialize routine offsets to 0 for integers and floats. 1021 // x8 for integers, x15 for floating point. 1022 mov x8, #0 1023 mov x15, #0 1024 1025 add x10, x5, #1 // Load shorty address, plus one to skip return value. 1026 1027 // Loop to fill registers. 1028.LfillRegisters2: 1029 ldrb w17, [x10], #1 // Load next character in signature, and increment. 1030 cbz w17, .LcallFunction2 // Exit at end of signature. Shorty 0 terminated. 1031 1032 cmp w17, #'F' // is this a float? 1033 bne .LisDouble2 1034 1035 cmp x15, # 8*12 // Skip this load if all registers full. 1036 beq .Ladvance4_2 1037 1038 add x17, x13, x15 // Calculate subroutine to jump to. 1039 br x17 1040 1041.LisDouble2: 1042 cmp w17, #'D' // is this a double? 1043 bne .LisLong2 1044 1045 cmp x15, # 8*12 // Skip this load if all registers full. 1046 beq .Ladvance8_2 1047 1048 add x17, x14, x15 // Calculate subroutine to jump to. 1049 br x17 1050 1051.LisLong2: 1052 cmp w17, #'J' // is this a long? 1053 bne .LisOther2 1054 1055 cmp x8, # 7*12 // Skip this load if all registers full. 1056 beq .Ladvance8_2 1057 1058 add x17, x12, x8 // Calculate subroutine to jump to. 1059 br x17 1060 1061.LisOther2: // Everything else takes one vReg. 1062 cmp x8, # 7*12 // Skip this load if all registers full. 1063 beq .Ladvance4_2 1064 1065 add x17, x11, x8 // Calculate subroutine to jump to. 1066 br x17 1067 1068.Ladvance4_2: 1069 add x9, x9, #4 1070 b .LfillRegisters2 1071 1072.Ladvance8_2: 1073 add x9, x9, #8 1074 b .LfillRegisters2 1075 1076// Store ints. 1077.LstoreW1_2: 1078 LOADREG x8 4 w1 .LfillRegisters2 1079 LOADREG x8 4 w2 .LfillRegisters2 1080 LOADREG x8 4 w3 .LfillRegisters2 1081 LOADREG x8 4 w4 .LfillRegisters2 1082 LOADREG x8 4 w5 .LfillRegisters2 1083 LOADREG x8 4 w6 .LfillRegisters2 1084 LOADREG x8 4 w7 .LfillRegisters2 1085 1086// Store longs. 1087.LstoreX1_2: 1088 LOADREG x8 8 x1 .LfillRegisters2 1089 LOADREG x8 8 x2 .LfillRegisters2 1090 LOADREG x8 8 x3 .LfillRegisters2 1091 LOADREG x8 8 x4 .LfillRegisters2 1092 LOADREG x8 8 x5 .LfillRegisters2 1093 LOADREG x8 8 x6 .LfillRegisters2 1094 LOADREG x8 8 x7 .LfillRegisters2 1095 1096// Store singles. 1097.LstoreS0_2: 1098 LOADREG x15 4 s0 .LfillRegisters2 1099 LOADREG x15 4 s1 .LfillRegisters2 1100 LOADREG x15 4 s2 .LfillRegisters2 1101 LOADREG x15 4 s3 .LfillRegisters2 1102 LOADREG x15 4 s4 .LfillRegisters2 1103 LOADREG x15 4 s5 .LfillRegisters2 1104 LOADREG x15 4 s6 .LfillRegisters2 1105 LOADREG x15 4 s7 .LfillRegisters2 1106 1107// Store doubles. 1108.LstoreD0_2: 1109 LOADREG x15 8 d0 .LfillRegisters2 1110 LOADREG x15 8 d1 .LfillRegisters2 1111 LOADREG x15 8 d2 .LfillRegisters2 1112 LOADREG x15 8 d3 .LfillRegisters2 1113 LOADREG x15 8 d4 .LfillRegisters2 1114 LOADREG x15 8 d5 .LfillRegisters2 1115 LOADREG x15 8 d6 .LfillRegisters2 1116 LOADREG x15 8 d7 .LfillRegisters2 1117 1118 1119.LcallFunction2: 1120 1121 INVOKE_STUB_CALL_AND_RETURN 1122 1123END art_quick_invoke_static_stub 1124 1125 1126 1127/* extern"C" void art_quick_osr_stub(void** stack, x0 1128 * size_t stack_size_in_bytes, x1 1129 * const uin8_t* native_pc, x2 1130 * JValue *result, x3 1131 * char *shorty, x4 1132 * Thread *self) x5 1133 */ 1134ENTRY art_quick_osr_stub 1135SAVE_SIZE=15*8 // x3, x4, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 1136 mov x9, sp // Save stack pointer. 1137 .cfi_register sp,x9 1138 1139 sub x10, sp, # SAVE_SIZE 1140 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 1141 mov sp, x10 // Set new SP. 1142 1143 str x28, [sp, #112] 1144 stp x26, x27, [sp, #96] 1145 stp x24, x25, [sp, #80] 1146 stp x22, x23, [sp, #64] 1147 stp x20, x21, [sp, #48] 1148 stp x9, x19, [sp, #32] // Save old stack pointer and x19. 1149 stp x3, x4, [sp, #16] // Save result and shorty addresses. 1150 stp xFP, xLR, [sp] // Store LR & FP. 1151 mov xSELF, x5 // Move thread pointer into SELF register. 1152 1153 sub sp, sp, #16 1154 str xzr, [sp] // Store null for ArtMethod* slot 1155 // Branch to stub. 1156 bl .Losr_entry 1157 add sp, sp, #16 1158 1159 // Restore return value address and shorty address. 1160 ldp x3,x4, [sp, #16] 1161 ldr x28, [sp, #112] 1162 ldp x26, x27, [sp, #96] 1163 ldp x24, x25, [sp, #80] 1164 ldp x22, x23, [sp, #64] 1165 ldp x20, x21, [sp, #48] 1166 1167 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 1168 ldrb w10, [x4] 1169 1170 // Check the return type and store the correct register into the jvalue in memory. 1171 1172 // Don't set anything for a void type. 1173 cmp w10, #'V' 1174 beq .Losr_exit 1175 1176 // Is it a double? 1177 cmp w10, #'D' 1178 bne .Lno_double 1179 str d0, [x3] 1180 b .Losr_exit 1181 1182.Lno_double: // Is it a float? 1183 cmp w10, #'F' 1184 bne .Lno_float 1185 str s0, [x3] 1186 b .Losr_exit 1187 1188.Lno_float: // Just store x0. Doesn't matter if it is 64 or 32 bits. 1189 str x0, [x3] 1190 1191.Losr_exit: // Finish up. 1192 ldp x2, x19, [sp, #32] // Restore stack pointer and x19. 1193 ldp xFP, xLR, [sp] // Restore old frame pointer and link register. 1194 mov sp, x2 1195 ret 1196 1197.Losr_entry: 1198 // Update stack pointer for the callee 1199 sub sp, sp, x1 1200 1201 // Update link register slot expected by the callee. 1202 sub w1, w1, #8 1203 str lr, [sp, x1] 1204 1205 // Copy arguments into stack frame. 1206 // Use simple copy routine for now. 1207 // 4 bytes per slot. 1208 // X0 - source address 1209 // W1 - args length 1210 // SP - destination address. 1211 // W10 - temporary 1212.Losr_loop_entry: 1213 cmp w1, #0 1214 beq .Losr_loop_exit 1215 sub w1, w1, #4 1216 ldr w10, [x0, x1] 1217 str w10, [sp, x1] 1218 b .Losr_loop_entry 1219 1220.Losr_loop_exit: 1221 // Branch to the OSR entry point. 1222 br x2 1223 1224END art_quick_osr_stub 1225 1226 /* 1227 * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_ 1228 */ 1229 1230ENTRY art_quick_do_long_jump 1231 // Load FPRs 1232 ldp d0, d1, [x1], #16 1233 ldp d2, d3, [x1], #16 1234 ldp d4, d5, [x1], #16 1235 ldp d6, d7, [x1], #16 1236 ldp d8, d9, [x1], #16 1237 ldp d10, d11, [x1], #16 1238 ldp d12, d13, [x1], #16 1239 ldp d14, d15, [x1], #16 1240 ldp d16, d17, [x1], #16 1241 ldp d18, d19, [x1], #16 1242 ldp d20, d21, [x1], #16 1243 ldp d22, d23, [x1], #16 1244 ldp d24, d25, [x1], #16 1245 ldp d26, d27, [x1], #16 1246 ldp d28, d29, [x1], #16 1247 ldp d30, d31, [x1] 1248 1249 // Load GPRs 1250 // TODO: lots of those are smashed, could optimize. 1251 add x0, x0, #30*8 1252 ldp x30, x1, [x0], #-16 // LR & SP 1253 ldp x28, x29, [x0], #-16 1254 ldp x26, x27, [x0], #-16 1255 ldp x24, x25, [x0], #-16 1256 ldp x22, x23, [x0], #-16 1257 ldp x20, x21, [x0], #-16 1258 ldp x18, x19, [x0], #-16 1259 ldp x16, x17, [x0], #-16 1260 ldp x14, x15, [x0], #-16 1261 ldp x12, x13, [x0], #-16 1262 ldp x10, x11, [x0], #-16 1263 ldp x8, x9, [x0], #-16 1264 ldp x6, x7, [x0], #-16 1265 ldp x4, x5, [x0], #-16 1266 ldp x2, x3, [x0], #-16 1267 mov sp, x1 1268 1269 // Need to load PC, it's at the end (after the space for the unused XZR). Use x1. 1270 ldr x1, [x0, #33*8] 1271 // And the value of x0. 1272 ldr x0, [x0] 1273 1274 br x1 1275END art_quick_do_long_jump 1276 1277 /* 1278 * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the 1279 * possibly null object to lock. 1280 * 1281 * Derived from arm32 code. 1282 */ 1283 .extern artLockObjectFromCode 1284ENTRY art_quick_lock_object 1285 cbz w0, .Lslow_lock 1286 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1287.Lretry_lock: 1288 ldr w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop? 1289 ldxr w1, [x4] 1290 mov x3, x1 1291 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1292 cbnz w3, .Lnot_unlocked // already thin locked 1293 // unlocked case - x1: original lock word that's zero except for the read barrier bits. 1294 orr x2, x1, x2 // x2 holds thread id with count of 0 with preserved read barrier bits 1295 stxr w3, w2, [x4] 1296 cbnz w3, .Llock_stxr_fail // store failed, retry 1297 dmb ishld // full (LoadLoad|LoadStore) memory barrier 1298 ret 1299.Lnot_unlocked: // x1: original lock word 1300 lsr w3, w1, LOCK_WORD_STATE_SHIFT 1301 cbnz w3, .Lslow_lock // if either of the top two bits are set, go slow path 1302 eor w2, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 1303 uxth w2, w2 // zero top 16 bits 1304 cbnz w2, .Lslow_lock // lock word and self thread id's match -> recursive lock 1305 // else contention, go to slow path 1306 mov x3, x1 // copy the lock word to check count overflow. 1307 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits. 1308 add w2, w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count in lock word placing in w2 to check overflow 1309 lsr w3, w2, #LOCK_WORD_GC_STATE_SHIFT // if the first gc state bit is set, we overflowed. 1310 cbnz w3, .Lslow_lock // if we overflow the count go slow path 1311 add w2, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count for real 1312 stxr w3, w2, [x4] 1313 cbnz w3, .Llock_stxr_fail // store failed, retry 1314 ret 1315.Llock_stxr_fail: 1316 b .Lretry_lock // retry 1317.Lslow_lock: 1318 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case we block 1319 mov x1, xSELF // pass Thread::Current 1320 bl artLockObjectFromCode // (Object* obj, Thread*) 1321 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1322 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1323END art_quick_lock_object 1324 1325ENTRY art_quick_lock_object_no_inline 1326 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case we block 1327 mov x1, xSELF // pass Thread::Current 1328 bl artLockObjectFromCode // (Object* obj, Thread*) 1329 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1330 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1331END art_quick_lock_object_no_inline 1332 1333 /* 1334 * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. 1335 * x0 holds the possibly null object to lock. 1336 * 1337 * Derived from arm32 code. 1338 */ 1339 .extern artUnlockObjectFromCode 1340ENTRY art_quick_unlock_object 1341 cbz x0, .Lslow_unlock 1342 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1343.Lretry_unlock: 1344#ifndef USE_READ_BARRIER 1345 ldr w1, [x4] 1346#else 1347 ldxr w1, [x4] // Need to use atomic instructions for read barrier 1348#endif 1349 lsr w2, w1, LOCK_WORD_STATE_SHIFT 1350 cbnz w2, .Lslow_unlock // if either of the top two bits are set, go slow path 1351 ldr w2, [xSELF, #THREAD_ID_OFFSET] 1352 mov x3, x1 // copy lock word to check thread id equality 1353 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1354 eor w3, w3, w2 // lock_word.ThreadId() ^ self->ThreadId() 1355 uxth w3, w3 // zero top 16 bits 1356 cbnz w3, .Lslow_unlock // do lock word and self thread id's match? 1357 mov x3, x1 // copy lock word to detect transition to unlocked 1358 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1359 cmp w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE 1360 bpl .Lrecursive_thin_unlock 1361 // transition to unlocked 1362 mov x3, x1 1363 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED // w3: zero except for the preserved read barrier bits 1364 dmb ish // full (LoadStore|StoreStore) memory barrier 1365#ifndef USE_READ_BARRIER 1366 str w3, [x4] 1367#else 1368 stxr w2, w3, [x4] // Need to use atomic instructions for read barrier 1369 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1370#endif 1371 ret 1372.Lrecursive_thin_unlock: // w1: original lock word 1373 sub w1, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // decrement count 1374#ifndef USE_READ_BARRIER 1375 str w1, [x4] 1376#else 1377 stxr w2, w1, [x4] // Need to use atomic instructions for read barrier 1378 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1379#endif 1380 ret 1381.Lunlock_stxr_fail: 1382 b .Lretry_unlock // retry 1383.Lslow_unlock: 1384 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case exception allocation triggers GC 1385 mov x1, xSELF // pass Thread::Current 1386 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1387 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1388 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1389END art_quick_unlock_object 1390 1391ENTRY art_quick_unlock_object_no_inline 1392 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case exception allocation triggers GC 1393 mov x1, xSELF // pass Thread::Current 1394 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1395 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1396 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1397END art_quick_unlock_object_no_inline 1398 1399 /* 1400 * Entry from managed code that calls artIsAssignableFromCode and on failure calls 1401 * artThrowClassCastException. 1402 */ 1403 .extern artThrowClassCastException 1404ENTRY art_quick_check_cast 1405 // Store arguments and link register 1406 // Stack needs to be 16B aligned on calls. 1407 stp x0, x1, [sp,#-32]! 1408 .cfi_adjust_cfa_offset 32 1409 .cfi_rel_offset x0, 0 1410 .cfi_rel_offset x1, 8 1411 str xLR, [sp, #24] 1412 .cfi_rel_offset x30, 24 1413 1414 // Call runtime code 1415 bl artIsAssignableFromCode 1416 1417 // Check for exception 1418 cbz x0, .Lthrow_class_cast_exception 1419 1420 // Restore and return 1421 ldr xLR, [sp, #24] 1422 .cfi_restore x30 1423 ldp x0, x1, [sp], #32 1424 .cfi_restore x0 1425 .cfi_restore x1 1426 .cfi_adjust_cfa_offset -32 1427 ret 1428 1429 .cfi_adjust_cfa_offset 32 // Reset unwind info so following code unwinds. 1430 1431.Lthrow_class_cast_exception: 1432 // Restore 1433 ldr xLR, [sp, #24] 1434 .cfi_restore x30 1435 ldp x0, x1, [sp], #32 1436 .cfi_restore x0 1437 .cfi_restore x1 1438 .cfi_adjust_cfa_offset -32 1439 1440 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 1441 mov x2, xSELF // pass Thread::Current 1442 b artThrowClassCastException // (Class*, Class*, Thread*) 1443 brk 0 // We should not return here... 1444END art_quick_check_cast 1445 1446// Restore xReg's value from [sp, #offset] if xReg is not the same as xExclude. 1447.macro POP_REG_NE xReg, offset, xExclude 1448 .ifnc \xReg, \xExclude 1449 ldr \xReg, [sp, #\offset] // restore xReg 1450 .cfi_restore \xReg 1451 .endif 1452.endm 1453 1454// Restore xReg1's value from [sp, #offset] if xReg1 is not the same as xExclude. 1455// Restore xReg2's value from [sp, #(offset + 8)] if xReg2 is not the same as xExclude. 1456.macro POP_REGS_NE xReg1, xReg2, offset, xExclude 1457 .ifc \xReg1, \xExclude 1458 ldr \xReg2, [sp, #(\offset + 8)] // restore xReg2 1459 .else 1460 .ifc \xReg2, \xExclude 1461 ldr \xReg1, [sp, #\offset] // restore xReg1 1462 .else 1463 ldp \xReg1, \xReg2, [sp, #\offset] // restore xReg1 and xReg2 1464 .endif 1465 .endif 1466 .cfi_restore \xReg1 1467 .cfi_restore \xReg2 1468.endm 1469 1470 /* 1471 * Macro to insert read barrier, only used in art_quick_aput_obj. 1472 * xDest, wDest and xObj are registers, offset is a defined literal such as 1473 * MIRROR_OBJECT_CLASS_OFFSET. Dest needs both x and w versions of the same register to handle 1474 * name mismatch between instructions. This macro uses the lower 32b of register when possible. 1475 * TODO: When read barrier has a fast path, add heap unpoisoning support for the fast path. 1476 */ 1477.macro READ_BARRIER xDest, wDest, xObj, xTemp, wTemp, offset, number 1478#ifdef USE_READ_BARRIER 1479#ifdef USE_BAKER_READ_BARRIER 1480 ldr \wTemp, [\xObj, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 1481 tbnz \wTemp, #LOCK_WORD_READ_BARRIER_STATE_SHIFT, .Lrb_slowpath\number 1482 // False dependency to avoid needing load/load fence. 1483 add \xObj, \xObj, \xTemp, lsr #32 1484 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1485 UNPOISON_HEAP_REF \wDest 1486 b .Lrb_exit\number 1487#endif 1488.Lrb_slowpath\number: 1489 // Store registers used in art_quick_aput_obj (x0-x4, LR), stack is 16B aligned. 1490 stp x0, x1, [sp, #-48]! 1491 .cfi_adjust_cfa_offset 48 1492 .cfi_rel_offset x0, 0 1493 .cfi_rel_offset x1, 8 1494 stp x2, x3, [sp, #16] 1495 .cfi_rel_offset x2, 16 1496 .cfi_rel_offset x3, 24 1497 stp x4, xLR, [sp, #32] 1498 .cfi_rel_offset x4, 32 1499 .cfi_rel_offset x30, 40 1500 1501 // mov x0, \xRef // pass ref in x0 (no-op for now since parameter ref is unused) 1502 .ifnc \xObj, x1 1503 mov x1, \xObj // pass xObj 1504 .endif 1505 mov w2, #\offset // pass offset 1506 bl artReadBarrierSlow // artReadBarrierSlow(ref, xObj, offset) 1507 // No need to unpoison return value in w0, artReadBarrierSlow() would do the unpoisoning. 1508 .ifnc \wDest, w0 1509 mov \wDest, w0 // save return value in wDest 1510 .endif 1511 1512 // Conditionally restore saved registers 1513 POP_REG_NE x0, 0, \xDest 1514 POP_REG_NE x1, 8, \xDest 1515 POP_REG_NE x2, 16, \xDest 1516 POP_REG_NE x3, 24, \xDest 1517 POP_REG_NE x4, 32, \xDest 1518 ldr xLR, [sp, #40] 1519 .cfi_restore x30 1520 add sp, sp, #48 1521 .cfi_adjust_cfa_offset -48 1522.Lrb_exit\number: 1523#else 1524 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1525 UNPOISON_HEAP_REF \wDest 1526#endif // USE_READ_BARRIER 1527.endm 1528 1529 /* 1530 * Entry from managed code for array put operations of objects where the value being stored 1531 * needs to be checked for compatibility. 1532 * x0 = array, x1 = index, x2 = value 1533 * 1534 * Currently all values should fit into w0/w1/w2, and w1 always will as indices are 32b. We 1535 * assume, though, that the upper 32b are zeroed out. At least for x1/w1 we can do better by 1536 * using index-zero-extension in load/stores. 1537 * 1538 * Temporaries: x3, x4 1539 * TODO: x4 OK? ip seems wrong here. 1540 */ 1541ENTRY art_quick_aput_obj_with_null_and_bound_check 1542 tst x0, x0 1543 bne art_quick_aput_obj_with_bound_check 1544 b art_quick_throw_null_pointer_exception 1545END art_quick_aput_obj_with_null_and_bound_check 1546 1547ENTRY art_quick_aput_obj_with_bound_check 1548 ldr w3, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] 1549 cmp w3, w1 1550 bhi art_quick_aput_obj 1551 mov x0, x1 1552 mov x1, x3 1553 b art_quick_throw_array_bounds 1554END art_quick_aput_obj_with_bound_check 1555 1556#ifdef USE_READ_BARRIER 1557 .extern artReadBarrierSlow 1558#endif 1559ENTRY art_quick_aput_obj 1560 cbz x2, .Ldo_aput_null 1561 READ_BARRIER x3, w3, x0, x3, w3, MIRROR_OBJECT_CLASS_OFFSET, 0 // Heap reference = 32b 1562 // This also zero-extends to x3 1563 READ_BARRIER x3, w3, x3, x4, w4, MIRROR_CLASS_COMPONENT_TYPE_OFFSET, 1 // Heap reference = 32b 1564 // This also zero-extends to x3 1565 READ_BARRIER x4, w4, x2, x4, w4, MIRROR_OBJECT_CLASS_OFFSET, 2 // Heap reference = 32b 1566 // This also zero-extends to x4 1567 cmp w3, w4 // value's type == array's component type - trivial assignability 1568 bne .Lcheck_assignability 1569.Ldo_aput: 1570 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1571 // "Compress" = do nothing 1572 POISON_HEAP_REF w2 1573 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1574 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1575 lsr x0, x0, #7 1576 strb w3, [x3, x0] 1577 ret 1578.Ldo_aput_null: 1579 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1580 // "Compress" = do nothing 1581 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1582 ret 1583.Lcheck_assignability: 1584 // Store arguments and link register 1585 stp x0, x1, [sp,#-32]! 1586 .cfi_adjust_cfa_offset 32 1587 .cfi_rel_offset x0, 0 1588 .cfi_rel_offset x1, 8 1589 stp x2, xLR, [sp, #16] 1590 .cfi_rel_offset x2, 16 1591 .cfi_rel_offset x30, 24 1592 1593 // Call runtime code 1594 mov x0, x3 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1595 mov x1, x4 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1596 bl artIsAssignableFromCode 1597 1598 // Check for exception 1599 cbz x0, .Lthrow_array_store_exception 1600 1601 // Restore 1602 ldp x2, x30, [sp, #16] 1603 .cfi_restore x2 1604 .cfi_restore x30 1605 ldp x0, x1, [sp], #32 1606 .cfi_restore x0 1607 .cfi_restore x1 1608 .cfi_adjust_cfa_offset -32 1609 1610 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1611 // "Compress" = do nothing 1612 POISON_HEAP_REF w2 1613 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1614 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1615 lsr x0, x0, #7 1616 strb w3, [x3, x0] 1617 ret 1618 .cfi_adjust_cfa_offset 32 // 4 restores after cbz for unwinding. 1619.Lthrow_array_store_exception: 1620 ldp x2, x30, [sp, #16] 1621 .cfi_restore x2 1622 .cfi_restore x30 1623 ldp x0, x1, [sp], #32 1624 .cfi_restore x0 1625 .cfi_restore x1 1626 .cfi_adjust_cfa_offset -32 1627 1628 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 1629 mov x1, x2 // Pass value. 1630 mov x2, xSELF // Pass Thread::Current. 1631 b artThrowArrayStoreException // (Object*, Object*, Thread*). 1632 brk 0 // Unreached. 1633END art_quick_aput_obj 1634 1635// Macro to facilitate adding new allocation entrypoints. 1636.macro ONE_ARG_DOWNCALL name, entrypoint, return 1637 .extern \entrypoint 1638ENTRY \name 1639 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1640 mov x1, xSELF // pass Thread::Current 1641 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1642 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1643 \return 1644END \name 1645.endm 1646 1647// Macro to facilitate adding new allocation entrypoints. 1648.macro TWO_ARG_DOWNCALL name, entrypoint, return 1649 .extern \entrypoint 1650ENTRY \name 1651 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1652 mov x2, xSELF // pass Thread::Current 1653 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1654 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1655 \return 1656END \name 1657.endm 1658 1659// Macro to facilitate adding new allocation entrypoints. 1660.macro THREE_ARG_DOWNCALL name, entrypoint, return 1661 .extern \entrypoint 1662ENTRY \name 1663 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1664 mov x3, xSELF // pass Thread::Current 1665 bl \entrypoint 1666 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1667 \return 1668END \name 1669.endm 1670 1671// Macro to facilitate adding new allocation entrypoints. 1672.macro FOUR_ARG_DOWNCALL name, entrypoint, return 1673 .extern \entrypoint 1674ENTRY \name 1675 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1676 mov x4, xSELF // pass Thread::Current 1677 bl \entrypoint // 1678 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1679 \return 1680 DELIVER_PENDING_EXCEPTION 1681END \name 1682.endm 1683 1684// Macros taking opportunity of code similarities for downcalls with referrer. 1685.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return 1686 .extern \entrypoint 1687ENTRY \name 1688 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1689 ldr x1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1690 mov x2, xSELF // pass Thread::Current 1691 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*, SP) 1692 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1693 \return 1694END \name 1695.endm 1696 1697.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return 1698 .extern \entrypoint 1699ENTRY \name 1700 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1701 ldr x2, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1702 mov x3, xSELF // pass Thread::Current 1703 bl \entrypoint 1704 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1705 \return 1706END \name 1707.endm 1708 1709.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return 1710 .extern \entrypoint 1711ENTRY \name 1712 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1713 ldr x3, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1714 mov x4, xSELF // pass Thread::Current 1715 bl \entrypoint 1716 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1717 \return 1718END \name 1719.endm 1720 1721.macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1722 cbz w0, 1f // result zero branch over 1723 ret // return 17241: 1725 DELIVER_PENDING_EXCEPTION 1726.endm 1727 1728 /* 1729 * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on 1730 * failure. 1731 */ 1732TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1733 1734 /* 1735 * Entry from managed code when uninitialized static storage, this stub will run the class 1736 * initializer and deliver the exception on error. On success the static storage base is 1737 * returned. 1738 */ 1739ONE_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1740 1741ONE_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1742ONE_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1743 1744ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1745ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1746ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1747ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1748ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1749ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1750ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1751 1752TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1753TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1754TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1755TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1756TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1757TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1758TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1759 1760TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1761TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1762TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1763TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1764 1765THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1766THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1767THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1768THREE_ARG_REF_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1769THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1770 1771// This is separated out as the argument order is different. 1772 .extern artSet64StaticFromCode 1773ENTRY art_quick_set64_static 1774 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1775 ldr x1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1776 // x2 contains the parameter 1777 mov x3, xSELF // pass Thread::Current 1778 bl artSet64StaticFromCode 1779 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1780 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1781END art_quick_set64_static 1782 1783 /* 1784 * Entry from managed code to resolve a string, this stub will allocate a String and deliver an 1785 * exception on error. On success the String is returned. w0 holds the string index. The fast 1786 * path check for hit in strings cache has already been performed. 1787 */ 1788ONE_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1789 1790// Generate the allocation entrypoints for each allocator. 1791GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_REGION_TLAB_ALLOCATORS 1792// Comment out allocators that have arm64 specific asm. 1793// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_region_tlab, RegionTLAB) implemented in asm 1794// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_tlab, RegionTLAB) 1795// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB) 1796GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1797// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY(_region_tlab, RegionTLAB) implemented in asm 1798// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB) 1799GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1800GENERATE_ALLOC_ENTRYPOINTS_CHECK_AND_ALLOC_ARRAY(_region_tlab, RegionTLAB) 1801GENERATE_ALLOC_ENTRYPOINTS_CHECK_AND_ALLOC_ARRAY_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1802GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB) 1803GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB) 1804GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB) 1805 1806// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_rosalloc, RosAlloc). 1807ENTRY art_quick_alloc_object_rosalloc 1808 // Fast path rosalloc allocation. 1809 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 1810 // x2-x7: free. 1811 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 1812 // Load the class (x2) 1813 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 1814 cbz x2, .Lart_quick_alloc_object_rosalloc_slow_path // Check null class 1815 // Check class status. 1816 ldr w3, [x2, #MIRROR_CLASS_STATUS_OFFSET] 1817 cmp x3, #MIRROR_CLASS_STATUS_INITIALIZED 1818 bne .Lart_quick_alloc_object_rosalloc_slow_path 1819 // Add a fake dependence from the 1820 // following access flag and size 1821 // loads to the status load. 1822 // This is to prevent those loads 1823 // from being reordered above the 1824 // status load and reading wrong 1825 // values (an alternative is to use 1826 // a load-acquire for the status). 1827 eor x3, x3, x3 1828 add x2, x2, x3 1829 // Check access flags has 1830 // kAccClassIsFinalizable 1831 ldr w3, [x2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] 1832 tst x3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE 1833 bne .Lart_quick_alloc_object_rosalloc_slow_path 1834 ldr x3, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] // Check if the thread local 1835 // allocation stack has room. 1836 // ldp won't work due to large offset. 1837 ldr x4, [xSELF, #THREAD_LOCAL_ALLOC_STACK_END_OFFSET] 1838 cmp x3, x4 1839 bhs .Lart_quick_alloc_object_rosalloc_slow_path 1840 ldr w3, [x2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (x3) 1841 cmp x3, #ROSALLOC_MAX_THREAD_LOCAL_BRACKET_SIZE // Check if the size is for a thread 1842 // local allocation 1843 bhs .Lart_quick_alloc_object_rosalloc_slow_path 1844 // Compute the rosalloc bracket index 1845 // from the size. 1846 // Align up the size by the rosalloc 1847 // bracket quantum size and divide 1848 // by the quantum size and subtract 1849 // by 1. This code is a shorter but 1850 // equivalent version. 1851 sub x3, x3, #1 1852 lsr x3, x3, #ROSALLOC_BRACKET_QUANTUM_SIZE_SHIFT 1853 // Load the rosalloc run (x4) 1854 add x4, xSELF, x3, lsl #POINTER_SIZE_SHIFT 1855 ldr x4, [x4, #THREAD_ROSALLOC_RUNS_OFFSET] 1856 // Load the free list head (x3). This 1857 // will be the return val. 1858 ldr x3, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1859 cbz x3, .Lart_quick_alloc_object_rosalloc_slow_path 1860 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1861 ldr x1, [x3, #ROSALLOC_SLOT_NEXT_OFFSET] // Load the next pointer of the head 1862 // and update the list head with the 1863 // next pointer. 1864 str x1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1865 // Store the class pointer in the 1866 // header. This also overwrites the 1867 // next pointer. The offsets are 1868 // asserted to match. 1869#if ROSALLOC_SLOT_NEXT_OFFSET != MIRROR_OBJECT_CLASS_OFFSET 1870#error "Class pointer needs to overwrite next pointer." 1871#endif 1872 POISON_HEAP_REF w2 1873 str w2, [x3, #MIRROR_OBJECT_CLASS_OFFSET] 1874 // Fence. This is "ish" not "ishst" so 1875 // that it also ensures ordering of 1876 // the class status load with respect 1877 // to later accesses to the class 1878 // object. Alternatively we could use 1879 // "ishst" if we use load-acquire for 1880 // the class status load.) 1881 // Needs to be done before pushing on 1882 // allocation since Heap::VisitObjects 1883 // relies on seeing the class pointer. 1884 // b/28790624 1885 dmb ish 1886 // Push the new object onto the thread 1887 // local allocation stack and 1888 // increment the thread local 1889 // allocation stack top. 1890 ldr x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1891 str w3, [x1], #COMPRESSED_REFERENCE_SIZE // (Increment x1 as a side effect.) 1892 str x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1893 // Decrement the size of the free list 1894 ldr w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1895 sub x1, x1, #1 1896 // TODO: consider combining this store 1897 // and the list head store above using 1898 // strd. 1899 str w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1900 1901 mov x0, x3 // Set the return value and return. 1902 ret 1903.Lart_quick_alloc_object_rosalloc_slow_path: 1904 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1905 mov x2, xSELF // pass Thread::Current 1906 bl artAllocObjectFromCodeRosAlloc // (uint32_t type_idx, Method* method, Thread*) 1907 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1908 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1909END art_quick_alloc_object_rosalloc 1910 1911 1912// The common fast path code for art_quick_alloc_array_region_tlab. 1913.macro ALLOC_ARRAY_TLAB_FAST_PATH slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1914 // Check null class 1915 cbz \wClass, \slowPathLabel 1916 ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED \slowPathLabel, \xClass, \wClass, \xCount, \wCount, \xTemp0, \wTemp0, \xTemp1, \wTemp1, \xTemp2, \wTemp2 1917.endm 1918 1919// The common fast path code for art_quick_alloc_array_region_tlab. 1920.macro ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1921 // Array classes are never finalizable or uninitialized, no need to check. 1922 ldr \wTemp0, [\xClass, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Load component type 1923 UNPOISON_HEAP_REF \wTemp0 1924 ldr \wTemp0, [\xTemp0, #MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET] 1925 lsr \xTemp0, \xTemp0, #PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT // Component size shift is in high 16 1926 // bits. 1927 // xCount is holding a 32 bit value, 1928 // it can not overflow. 1929 lsl \xTemp1, \xCount, \xTemp0 // Calculate data size 1930 // Add array data offset and alignment. 1931 add \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) 1932#if MIRROR_LONG_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4 1933#error Long array data offset must be 4 greater than int array data offset. 1934#endif 1935 1936 add \xTemp0, \xTemp0, #1 // Add 4 to the length only if the 1937 // component size shift is 3 1938 // (for 64 bit alignment). 1939 and \xTemp0, \xTemp0, #4 1940 add \xTemp1, \xTemp1, \xTemp0 1941 and \xTemp1, \xTemp1, #OBJECT_ALIGNMENT_MASK_TOGGLED64 // Apply alignemnt mask 1942 // (addr + 7) & ~7. The mask must 1943 // be 64 bits to keep high bits in 1944 // case of overflow. 1945 // Negative sized arrays are handled here since xCount holds a zero extended 32 bit value. 1946 // Negative ints become large 64 bit unsigned ints which will always be larger than max signed 1947 // 32 bit int. Since the max shift for arrays is 3, it can not become a negative 64 bit int. 1948 cmp \xTemp1, #MIN_LARGE_OBJECT_THRESHOLD // Possibly a large object, go slow 1949 bhs \slowPathLabel // path. 1950 1951 ldr \xTemp0, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Check tlab for space, note that 1952 // we use (end - begin) to handle 1953 // negative size arrays. It is 1954 // assumed that a negative size will 1955 // always be greater unsigned than 1956 // region size. 1957 ldr \xTemp2, [xSELF, #THREAD_LOCAL_END_OFFSET] 1958 sub \xTemp2, \xTemp2, \xTemp0 1959 cmp \xTemp1, \xTemp2 1960 bhi \slowPathLabel 1961 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1962 // Move old thread_local_pos to x0 1963 // for the return value. 1964 mov x0, \xTemp0 1965 add \xTemp0, \xTemp0, \xTemp1 1966 str \xTemp0, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 1967 ldr \xTemp0, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 1968 add \xTemp0, \xTemp0, #1 1969 str \xTemp0, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 1970 POISON_HEAP_REF \wClass 1971 str \wClass, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 1972 str \wCount, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] // Store the array length. 1973 // Fence. 1974 dmb ishst 1975 ret 1976.endm 1977 1978// The common fast path code for art_quick_alloc_object_tlab and art_quick_alloc_object_region_tlab. 1979// 1980// x0: type_idx/return value, x1: ArtMethod*, x2: Class*, xSELF(x19): Thread::Current 1981// x3-x7: free. 1982// Need to preserve x0 and x1 to the slow path. 1983.macro ALLOC_OBJECT_TLAB_FAST_PATH slowPathLabel 1984 cbz x2, \slowPathLabel // Check null class 1985 ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED \slowPathLabel 1986.endm 1987 1988.macro ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED slowPathLabel 1989 ldr w3, [x2, #MIRROR_CLASS_STATUS_OFFSET] // Check class status. 1990 cmp x3, #MIRROR_CLASS_STATUS_INITIALIZED 1991 bne \slowPathLabel 1992 // Add a fake dependence from the 1993 // following access flag and size 1994 // loads to the status load. 1995 // This is to prevent those loads 1996 // from being reordered above the 1997 // status load and reading wrong 1998 // values (an alternative is to use 1999 // a load-acquire for the status). 2000 eor x3, x3, x3 2001 add x2, x2, x3 2002 ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED \slowPathLabel 2003.endm 2004 2005.macro ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED slowPathLabel 2006 // Check access flags has 2007 // kAccClassIsFinalizable. 2008 ldr w3, [x2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] 2009 tbnz x3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE_BIT, \slowPathLabel 2010 // Load thread_local_pos (x4) and 2011 // thread_local_end (x5). 2012 ldr x4, [xSELF, #THREAD_LOCAL_POS_OFFSET] 2013 ldr x5, [xSELF, #THREAD_LOCAL_END_OFFSET] 2014 sub x6, x5, x4 // Compute the remaining buf size. 2015 ldr w7, [x2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (x7). 2016 cmp x7, x6 // Check if it fits. OK to do this 2017 // before rounding up the object size 2018 // assuming the buf size alignment. 2019 bhi \slowPathLabel 2020 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 2021 // Round up the object size by the 2022 // object alignment. (addr + 7) & ~7. 2023 add x7, x7, #OBJECT_ALIGNMENT_MASK 2024 and x7, x7, #OBJECT_ALIGNMENT_MASK_TOGGLED 2025 // Move old thread_local_pos to x0 2026 // for the return value. 2027 mov x0, x4 2028 add x5, x0, x7 2029 str x5, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 2030 ldr x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 2031 add x5, x5, #1 2032 str x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 2033 POISON_HEAP_REF w2 2034 str w2, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 2035 // Fence. This is "ish" not "ishst" so 2036 // that the code after this allocation 2037 // site will see the right values in 2038 // the fields of the class. 2039 // Alternatively we could use "ishst" 2040 // if we use load-acquire for the 2041 // class status load.) 2042 dmb ish 2043 ret 2044.endm 2045 2046// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_tlab, TLAB). 2047ENTRY art_quick_alloc_object_tlab 2048 // Fast path tlab allocation. 2049 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 2050 // x2-x7: free. 2051#if defined(USE_READ_BARRIER) 2052 mvn x0, xzr // Read barrier not supported here. 2053 ret // Return -1. 2054#endif 2055 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2056 // Load the class (x2) 2057 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2058 ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_tlab_slow_path 2059.Lart_quick_alloc_object_tlab_slow_path: 2060 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // Save callee saves in case of GC. 2061 mov x2, xSELF // Pass Thread::Current. 2062 bl artAllocObjectFromCodeTLAB // (uint32_t type_idx, Method* method, Thread*) 2063 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 2064 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2065END art_quick_alloc_object_tlab 2066 2067// The common code for art_quick_alloc_object_*region_tlab 2068.macro GENERATE_ALLOC_OBJECT_REGION_TLAB name, entrypoint, fast_path, is_resolved 2069ENTRY \name 2070 // Fast path region tlab allocation. 2071 // x0: type_idx/resolved class/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 2072 // If is_resolved is 1 then x0 is the resolved type, otherwise it is the index. 2073 // x2-x7: free. 2074#if !defined(USE_READ_BARRIER) 2075 mvn x0, xzr // Read barrier must be enabled here. 2076 ret // Return -1. 2077#endif 2078.if \is_resolved 2079 mov x2, x0 // class is actually stored in x0 already 2080.else 2081 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2082 // Load the class (x2) 2083 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2084.endif 2085 // Most common case: GC is not marking. 2086 ldr w3, [xSELF, #THREAD_IS_GC_MARKING_OFFSET] 2087 cbnz x3, .Lmarking\name 2088.Ldo_allocation\name: 2089 \fast_path .Lslow_path\name 2090.Lmarking\name: 2091 // GC is marking, check the lock word of the class for the mark bit. 2092 // If the class is null, go slow path. The check is required to read the lock word. 2093 cbz w2, .Lslow_path\name 2094 // Class is not null, check mark bit in lock word. 2095 ldr w3, [x2, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2096 // If the bit is not zero, do the allocation. 2097 tbnz w3, #LOCK_WORD_MARK_BIT_SHIFT, .Ldo_allocation\name 2098 // The read barrier slow path. Mark 2099 // the class. 2100 stp x0, x1, [sp, #-32]! // Save registers (x0, x1, lr). 2101 str xLR, [sp, #16] // Align sp by 16 bytes. 2102 mov x0, x2 // Pass the class as the first param. 2103 bl artReadBarrierMark 2104 mov x2, x0 // Get the (marked) class back. 2105 ldp x0, x1, [sp, #0] // Restore registers. 2106 ldr xLR, [sp, #16] 2107 add sp, sp, #32 2108 b .Ldo_allocation\name 2109.Lslow_path\name: 2110 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // Save callee saves in case of GC. 2111 mov x2, xSELF // Pass Thread::Current. 2112 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 2113 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 2114 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2115END \name 2116.endm 2117 2118GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_region_tlab, artAllocObjectFromCodeRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH, 0 2119GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_resolved_region_tlab, artAllocObjectFromCodeResolvedRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED, 1 2120GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_initialized_region_tlab, artAllocObjectFromCodeInitializedRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED, 1 2121 2122// The common code for art_quick_alloc_array_*region_tlab 2123.macro GENERATE_ALLOC_ARRAY_REGION_TLAB name, entrypoint, fast_path, is_resolved 2124ENTRY \name 2125 // Fast path array allocation for region tlab allocation. 2126 // x0: uint32_t type_idx 2127 // x1: int32_t component_count 2128 // x2: ArtMethod* method 2129 // x3-x7: free. 2130#if !defined(USE_READ_BARRIER) 2131 mvn x0, xzr // Read barrier must be enabled here. 2132 ret // Return -1. 2133#endif 2134.if \is_resolved 2135 mov x3, x0 2136 // If already resolved, class is stored in x0 2137.else 2138 ldr x3, [x2, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2139 // Load the class (x2) 2140 ldr w3, [x3, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2141.endif 2142 // Most common case: GC is not marking. 2143 ldr w4, [xSELF, #THREAD_IS_GC_MARKING_OFFSET] 2144 cbnz x4, .Lmarking\name 2145.Ldo_allocation\name: 2146 \fast_path .Lslow_path\name, x3, w3, x1, w1, x4, w4, x5, w5, x6, w6 2147.Lmarking\name: 2148 // GC is marking, check the lock word of the class for the mark bit. 2149 // If the class is null, go slow path. The check is required to read the lock word. 2150 cbz w3, .Lslow_path\name 2151 // Class is not null, check mark bit in lock word. 2152 ldr w4, [x3, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2153 // If the bit is not zero, do the allocation. 2154 tbnz w4, #LOCK_WORD_MARK_BIT_SHIFT, .Ldo_allocation\name 2155 // The read barrier slow path. Mark 2156 // the class. 2157 stp x0, x1, [sp, #-32]! // Save registers (x0, x1, x2, lr). 2158 stp x2, xLR, [sp, #16] 2159 mov x0, x3 // Pass the class as the first param. 2160 bl artReadBarrierMark 2161 mov x3, x0 // Get the (marked) class back. 2162 ldp x2, xLR, [sp, #16] 2163 ldp x0, x1, [sp], #32 // Restore registers. 2164 b .Ldo_allocation\name 2165.Lslow_path\name: 2166 // x0: uint32_t type_idx / mirror::Class* klass (if resolved) 2167 // x1: int32_t component_count 2168 // x2: ArtMethod* method 2169 // x3: Thread* self 2170 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 2171 mov x3, xSELF // pass Thread::Current 2172 bl \entrypoint 2173 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 2174 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2175END \name 2176.endm 2177 2178GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_region_tlab, artAllocArrayFromCodeRegionTLAB, ALLOC_ARRAY_TLAB_FAST_PATH, 0 2179// TODO: art_quick_alloc_array_resolved_region_tlab seems to not get called. Investigate compiler. 2180GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED, 1 2181 2182 /* 2183 * Called by managed code when the thread has been asked to suspend. 2184 */ 2185 .extern artTestSuspendFromCode 2186ENTRY art_quick_test_suspend 2187 SETUP_SAVE_EVERYTHING_CALLEE_SAVE_FRAME // save callee saves for stack crawl 2188 mov x0, xSELF 2189 bl artTestSuspendFromCode // (Thread*) 2190 RESTORE_SAVE_EVERYTHING_CALLEE_SAVE_FRAME 2191 ret 2192END art_quick_test_suspend 2193 2194ENTRY art_quick_implicit_suspend 2195 mov x0, xSELF 2196 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 2197 bl artTestSuspendFromCode // (Thread*) 2198 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 2199END art_quick_implicit_suspend 2200 2201 /* 2202 * Called by managed code that is attempting to call a method on a proxy class. On entry 2203 * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy 2204 * method agrees with a ref and args callee save frame. 2205 */ 2206 .extern artQuickProxyInvokeHandler 2207ENTRY art_quick_proxy_invoke_handler 2208 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 2209 mov x2, xSELF // pass Thread::Current 2210 mov x3, sp // pass SP 2211 bl artQuickProxyInvokeHandler // (Method* proxy method, receiver, Thread*, SP) 2212 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 2213 cbnz x2, .Lexception_in_proxy // success if no exception is pending 2214 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Restore frame 2215 fmov d0, x0 // Store result in d0 in case it was float or double 2216 ret // return on success 2217.Lexception_in_proxy: 2218 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 2219 DELIVER_PENDING_EXCEPTION 2220END art_quick_proxy_invoke_handler 2221 2222 /* 2223 * Called to resolve an imt conflict. 2224 * x0 is the conflict ArtMethod. 2225 * xIP1 is a hidden argument that holds the target interface method's dex method index. 2226 * 2227 * Note that this stub writes to xIP0, xIP1, and x0. 2228 */ 2229 .extern artInvokeInterfaceTrampoline 2230ENTRY art_quick_imt_conflict_trampoline 2231 ldr xIP0, [sp, #0] // Load referrer 2232 ldr xIP0, [xIP0, #ART_METHOD_DEX_CACHE_METHODS_OFFSET_64] // Load dex cache methods array 2233 ldr xIP0, [xIP0, xIP1, lsl #POINTER_SIZE_SHIFT] // Load interface method 2234 ldr xIP1, [x0, #ART_METHOD_JNI_OFFSET_64] // Load ImtConflictTable 2235 ldr x0, [xIP1] // Load first entry in ImtConflictTable. 2236.Limt_table_iterate: 2237 cmp x0, xIP0 2238 // Branch if found. Benchmarks have shown doing a branch here is better. 2239 beq .Limt_table_found 2240 // If the entry is null, the interface method is not in the ImtConflictTable. 2241 cbz x0, .Lconflict_trampoline 2242 // Iterate over the entries of the ImtConflictTable. 2243 ldr x0, [xIP1, #(2 * __SIZEOF_POINTER__)]! 2244 b .Limt_table_iterate 2245.Limt_table_found: 2246 // We successfully hit an entry in the table. Load the target method 2247 // and jump to it. 2248 ldr x0, [xIP1, #__SIZEOF_POINTER__] 2249 ldr xIP0, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 2250 br xIP0 2251.Lconflict_trampoline: 2252 // Call the runtime stub to populate the ImtConflictTable and jump to the 2253 // resolved method. 2254 INVOKE_TRAMPOLINE_BODY artInvokeInterfaceTrampoline 2255END art_quick_imt_conflict_trampoline 2256 2257ENTRY art_quick_resolution_trampoline 2258 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 2259 mov x2, xSELF 2260 mov x3, sp 2261 bl artQuickResolutionTrampoline // (called, receiver, Thread*, SP) 2262 cbz x0, 1f 2263 mov xIP0, x0 // Remember returned code pointer in xIP0. 2264 ldr x0, [sp, #0] // artQuickResolutionTrampoline puts called method in *SP. 2265 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 2266 br xIP0 22671: 2268 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 2269 DELIVER_PENDING_EXCEPTION 2270END art_quick_resolution_trampoline 2271 2272/* 2273 * Generic JNI frame layout: 2274 * 2275 * #-------------------# 2276 * | | 2277 * | caller method... | 2278 * #-------------------# <--- SP on entry 2279 * | Return X30/LR | 2280 * | X29/FP | callee save 2281 * | X28 | callee save 2282 * | X27 | callee save 2283 * | X26 | callee save 2284 * | X25 | callee save 2285 * | X24 | callee save 2286 * | X23 | callee save 2287 * | X22 | callee save 2288 * | X21 | callee save 2289 * | X20 | callee save 2290 * | X19 | callee save 2291 * | X7 | arg7 2292 * | X6 | arg6 2293 * | X5 | arg5 2294 * | X4 | arg4 2295 * | X3 | arg3 2296 * | X2 | arg2 2297 * | X1 | arg1 2298 * | D7 | float arg 8 2299 * | D6 | float arg 7 2300 * | D5 | float arg 6 2301 * | D4 | float arg 5 2302 * | D3 | float arg 4 2303 * | D2 | float arg 3 2304 * | D1 | float arg 2 2305 * | D0 | float arg 1 2306 * | Method* | <- X0 2307 * #-------------------# 2308 * | local ref cookie | // 4B 2309 * | handle scope size | // 4B 2310 * #-------------------# 2311 * | JNI Call Stack | 2312 * #-------------------# <--- SP on native call 2313 * | | 2314 * | Stack for Regs | The trampoline assembly will pop these values 2315 * | | into registers for native call 2316 * #-------------------# 2317 * | Native code ptr | 2318 * #-------------------# 2319 * | Free scratch | 2320 * #-------------------# 2321 * | Ptr to (1) | <--- SP 2322 * #-------------------# 2323 */ 2324 /* 2325 * Called to do a generic JNI down-call 2326 */ 2327ENTRY art_quick_generic_jni_trampoline 2328 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 2329 2330 // Save SP , so we can have static CFI info. 2331 mov x28, sp 2332 .cfi_def_cfa_register x28 2333 2334 // This looks the same, but is different: this will be updated to point to the bottom 2335 // of the frame when the handle scope is inserted. 2336 mov xFP, sp 2337 2338 mov xIP0, #5120 2339 sub sp, sp, xIP0 2340 2341 // prepare for artQuickGenericJniTrampoline call 2342 // (Thread*, SP) 2343 // x0 x1 <= C calling convention 2344 // xSELF xFP <= where they are 2345 2346 mov x0, xSELF // Thread* 2347 mov x1, xFP 2348 bl artQuickGenericJniTrampoline // (Thread*, sp) 2349 2350 // The C call will have registered the complete save-frame on success. 2351 // The result of the call is: 2352 // x0: pointer to native code, 0 on error. 2353 // x1: pointer to the bottom of the used area of the alloca, can restore stack till there. 2354 2355 // Check for error = 0. 2356 cbz x0, .Lexception_in_native 2357 2358 // Release part of the alloca. 2359 mov sp, x1 2360 2361 // Save the code pointer 2362 mov xIP0, x0 2363 2364 // Load parameters from frame into registers. 2365 // TODO Check with artQuickGenericJniTrampoline. 2366 // Also, check again APPCS64 - the stack arguments are interleaved. 2367 ldp x0, x1, [sp] 2368 ldp x2, x3, [sp, #16] 2369 ldp x4, x5, [sp, #32] 2370 ldp x6, x7, [sp, #48] 2371 2372 ldp d0, d1, [sp, #64] 2373 ldp d2, d3, [sp, #80] 2374 ldp d4, d5, [sp, #96] 2375 ldp d6, d7, [sp, #112] 2376 2377 add sp, sp, #128 2378 2379 blr xIP0 // native call. 2380 2381 // result sign extension is handled in C code 2382 // prepare for artQuickGenericJniEndTrampoline call 2383 // (Thread*, result, result_f) 2384 // x0 x1 x2 <= C calling convention 2385 mov x1, x0 // Result (from saved). 2386 mov x0, xSELF // Thread register. 2387 fmov x2, d0 // d0 will contain floating point result, but needs to go into x2 2388 2389 bl artQuickGenericJniEndTrampoline 2390 2391 // Pending exceptions possible. 2392 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 2393 cbnz x2, .Lexception_in_native 2394 2395 // Tear down the alloca. 2396 mov sp, x28 2397 .cfi_def_cfa_register sp 2398 2399 // Tear down the callee-save frame. 2400 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 2401 2402 // store into fpr, for when it's a fpr return... 2403 fmov d0, x0 2404 ret 2405 2406.Lexception_in_native: 2407 // Move to x1 then sp to please assembler. 2408 ldr x1, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 2409 mov sp, x1 2410 .cfi_def_cfa_register sp 2411 # This will create a new save-all frame, required by the runtime. 2412 DELIVER_PENDING_EXCEPTION 2413END art_quick_generic_jni_trampoline 2414 2415/* 2416 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those 2417 * of a quick call: 2418 * x0 = method being called/to bridge to. 2419 * x1..x7, d0..d7 = arguments to that method. 2420 */ 2421ENTRY art_quick_to_interpreter_bridge 2422 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Set up frame and save arguments. 2423 2424 // x0 will contain mirror::ArtMethod* method. 2425 mov x1, xSELF // How to get Thread::Current() ??? 2426 mov x2, sp 2427 2428 // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self, 2429 // mirror::ArtMethod** sp) 2430 bl artQuickToInterpreterBridge 2431 2432 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // TODO: no need to restore arguments in this case. 2433 2434 fmov d0, x0 2435 2436 RETURN_OR_DELIVER_PENDING_EXCEPTION 2437END art_quick_to_interpreter_bridge 2438 2439 2440// 2441// Instrumentation-related stubs 2442// 2443 .extern artInstrumentationMethodEntryFromCode 2444ENTRY art_quick_instrumentation_entry 2445 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 2446 2447 mov x20, x0 // Preserve method reference in a callee-save. 2448 2449 mov x2, xSELF 2450 mov x3, xLR 2451 bl artInstrumentationMethodEntryFromCode // (Method*, Object*, Thread*, LR) 2452 2453 mov xIP0, x0 // x0 = result of call. 2454 mov x0, x20 // Reload method reference. 2455 2456 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Note: will restore xSELF 2457 adr xLR, art_quick_instrumentation_exit 2458 br xIP0 // Tail-call method with lr set to art_quick_instrumentation_exit. 2459END art_quick_instrumentation_entry 2460 2461 .extern artInstrumentationMethodExitFromCode 2462ENTRY art_quick_instrumentation_exit 2463 mov xLR, #0 // Clobber LR for later checks. 2464 2465 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME 2466 2467 // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then 2468 // we would need to fully restore it. As there are a lot of callee-save registers, it seems 2469 // easier to have an extra small stack area. 2470 2471 str x0, [sp, #-16]! // Save integer result. 2472 .cfi_adjust_cfa_offset 16 2473 str d0, [sp, #8] // Save floating-point result. 2474 2475 add x1, sp, #16 // Pass SP. 2476 mov x2, x0 // Pass integer result. 2477 fmov x3, d0 // Pass floating-point result. 2478 mov x0, xSELF // Pass Thread. 2479 bl artInstrumentationMethodExitFromCode // (Thread*, SP, gpr_res, fpr_res) 2480 2481 mov xIP0, x0 // Return address from instrumentation call. 2482 mov xLR, x1 // r1 is holding link register if we're to bounce to deoptimize 2483 2484 ldr d0, [sp, #8] // Restore floating-point result. 2485 ldr x0, [sp], 16 // Restore integer result, and drop stack area. 2486 .cfi_adjust_cfa_offset 16 2487 2488 POP_REFS_ONLY_CALLEE_SAVE_FRAME 2489 2490 br xIP0 // Tail-call out. 2491END art_quick_instrumentation_exit 2492 2493 /* 2494 * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization 2495 * will long jump to the upcall with a special exception of -1. 2496 */ 2497 .extern artDeoptimize 2498ENTRY art_quick_deoptimize 2499 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 2500 mov x0, xSELF // Pass thread. 2501 bl artDeoptimize // artDeoptimize(Thread*) 2502 brk 0 2503END art_quick_deoptimize 2504 2505 /* 2506 * Compiled code has requested that we deoptimize into the interpreter. The deoptimization 2507 * will long jump to the upcall with a special exception of -1. 2508 */ 2509 .extern artDeoptimizeFromCompiledCode 2510ENTRY art_quick_deoptimize_from_compiled_code 2511 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 2512 mov x0, xSELF // Pass thread. 2513 bl artDeoptimizeFromCompiledCode // artDeoptimizeFromCompiledCode(Thread*) 2514 brk 0 2515END art_quick_deoptimize_from_compiled_code 2516 2517 2518 /* 2519 * String's indexOf. 2520 * 2521 * TODO: Not very optimized. 2522 * On entry: 2523 * x0: string object (known non-null) 2524 * w1: char to match (known <= 0xFFFF) 2525 * w2: Starting offset in string data 2526 */ 2527ENTRY art_quick_indexof 2528 ldr w3, [x0, #MIRROR_STRING_COUNT_OFFSET] 2529 add x0, x0, #MIRROR_STRING_VALUE_OFFSET 2530 2531 /* Clamp start to [0..count] */ 2532 cmp w2, #0 2533 csel w2, wzr, w2, lt 2534 cmp w2, w3 2535 csel w2, w3, w2, gt 2536 2537 /* Save a copy to compute result */ 2538 mov x5, x0 2539 2540 /* Build pointer to start of data to compare and pre-bias */ 2541 add x0, x0, x2, lsl #1 2542 sub x0, x0, #2 2543 2544 /* Compute iteration count */ 2545 sub w2, w3, w2 2546 2547 /* 2548 * At this point we have: 2549 * x0: start of the data to test 2550 * w1: char to compare 2551 * w2: iteration count 2552 * x5: original start of string data 2553 */ 2554 2555 subs w2, w2, #4 2556 b.lt .Lindexof_remainder 2557 2558.Lindexof_loop4: 2559 ldrh w6, [x0, #2]! 2560 ldrh w7, [x0, #2]! 2561 ldrh wIP0, [x0, #2]! 2562 ldrh wIP1, [x0, #2]! 2563 cmp w6, w1 2564 b.eq .Lmatch_0 2565 cmp w7, w1 2566 b.eq .Lmatch_1 2567 cmp wIP0, w1 2568 b.eq .Lmatch_2 2569 cmp wIP1, w1 2570 b.eq .Lmatch_3 2571 subs w2, w2, #4 2572 b.ge .Lindexof_loop4 2573 2574.Lindexof_remainder: 2575 adds w2, w2, #4 2576 b.eq .Lindexof_nomatch 2577 2578.Lindexof_loop1: 2579 ldrh w6, [x0, #2]! 2580 cmp w6, w1 2581 b.eq .Lmatch_3 2582 subs w2, w2, #1 2583 b.ne .Lindexof_loop1 2584 2585.Lindexof_nomatch: 2586 mov x0, #-1 2587 ret 2588 2589.Lmatch_0: 2590 sub x0, x0, #6 2591 sub x0, x0, x5 2592 asr x0, x0, #1 2593 ret 2594.Lmatch_1: 2595 sub x0, x0, #4 2596 sub x0, x0, x5 2597 asr x0, x0, #1 2598 ret 2599.Lmatch_2: 2600 sub x0, x0, #2 2601 sub x0, x0, x5 2602 asr x0, x0, #1 2603 ret 2604.Lmatch_3: 2605 sub x0, x0, x5 2606 asr x0, x0, #1 2607 ret 2608END art_quick_indexof 2609 2610 /* 2611 * Create a function `name` calling the ReadBarrier::Mark routine, 2612 * getting its argument and returning its result through W register 2613 * `wreg` (corresponding to X register `xreg`), saving and restoring 2614 * all caller-save registers. 2615 * 2616 * If `wreg` is different from `w0`, the generated function follows a 2617 * non-standard runtime calling convention: 2618 * - register `wreg` is used to pass the (sole) argument of this 2619 * function (instead of W0); 2620 * - register `wreg` is used to return the result of this function 2621 * (instead of W0); 2622 * - W0 is treated like a normal (non-argument) caller-save register; 2623 * - everything else is the same as in the standard runtime calling 2624 * convention (e.g. standard callee-save registers are preserved). 2625 */ 2626.macro READ_BARRIER_MARK_REG name, wreg, xreg 2627ENTRY \name 2628 // Reference is null, no work to do at all. 2629 cbz \wreg, .Lret_rb_\name 2630 /* 2631 * Allocate 46 stack slots * 8 = 368 bytes: 2632 * - 20 slots for core registers X0-X19 2633 * - 24 slots for floating-point registers D0-D7 and D16-D31 2634 * - 1 slot for return address register XLR 2635 * - 1 padding slot for 16-byte stack alignment 2636 */ 2637 // Use wIP0 as temp and check the mark bit of the reference. wIP0 is not used by the compiler. 2638 ldr wIP0, [\xreg, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2639 tbz wIP0, #LOCK_WORD_MARK_BIT_SHIFT, .Lslow_path_rb_\name 2640 ret 2641.Lslow_path_rb_\name: 2642 // Save all potentially live caller-save core registers. 2643 stp x0, x1, [sp, #-368]! 2644 .cfi_adjust_cfa_offset 368 2645 .cfi_rel_offset x0, 0 2646 .cfi_rel_offset x1, 8 2647 stp x2, x3, [sp, #16] 2648 .cfi_rel_offset x2, 16 2649 .cfi_rel_offset x3, 24 2650 stp x4, x5, [sp, #32] 2651 .cfi_rel_offset x4, 32 2652 .cfi_rel_offset x5, 40 2653 stp x6, x7, [sp, #48] 2654 .cfi_rel_offset x6, 48 2655 .cfi_rel_offset x7, 56 2656 stp x8, x9, [sp, #64] 2657 .cfi_rel_offset x8, 64 2658 .cfi_rel_offset x9, 72 2659 stp x10, x11, [sp, #80] 2660 .cfi_rel_offset x10, 80 2661 .cfi_rel_offset x11, 88 2662 stp x12, x13, [sp, #96] 2663 .cfi_rel_offset x12, 96 2664 .cfi_rel_offset x13, 104 2665 stp x14, x15, [sp, #112] 2666 .cfi_rel_offset x14, 112 2667 .cfi_rel_offset x15, 120 2668 stp x16, x17, [sp, #128] 2669 .cfi_rel_offset x16, 128 2670 .cfi_rel_offset x17, 136 2671 stp x18, x19, [sp, #144] 2672 .cfi_rel_offset x18, 144 2673 .cfi_rel_offset x19, 152 2674 // Save all potentially live caller-save floating-point registers. 2675 stp d0, d1, [sp, #160] 2676 stp d2, d3, [sp, #176] 2677 stp d4, d5, [sp, #192] 2678 stp d6, d7, [sp, #208] 2679 stp d16, d17, [sp, #224] 2680 stp d18, d19, [sp, #240] 2681 stp d20, d21, [sp, #256] 2682 stp d22, d23, [sp, #272] 2683 stp d24, d25, [sp, #288] 2684 stp d26, d27, [sp, #304] 2685 stp d28, d29, [sp, #320] 2686 stp d30, d31, [sp, #336] 2687 // Save return address. 2688 str xLR, [sp, #352] 2689 .cfi_rel_offset x30, 352 2690 // (sp + #360 is a padding slot) 2691 2692 .ifnc \wreg, w0 2693 mov w0, \wreg // Pass arg1 - obj from `wreg` 2694 .endif 2695 bl artReadBarrierMark // artReadBarrierMark(obj) 2696 .ifnc \wreg, w0 2697 mov \wreg, w0 // Return result into `wreg` 2698 .endif 2699 2700 // Restore core regs, except `xreg`, as `wreg` is used to return the 2701 // result of this function (simply remove it from the stack instead). 2702 POP_REGS_NE x0, x1, 0, \xreg 2703 POP_REGS_NE x2, x3, 16, \xreg 2704 POP_REGS_NE x4, x5, 32, \xreg 2705 POP_REGS_NE x6, x7, 48, \xreg 2706 POP_REGS_NE x8, x9, 64, \xreg 2707 POP_REGS_NE x10, x11, 80, \xreg 2708 POP_REGS_NE x12, x13, 96, \xreg 2709 POP_REGS_NE x14, x15, 112, \xreg 2710 POP_REGS_NE x16, x17, 128, \xreg 2711 POP_REGS_NE x18, x19, 144, \xreg 2712 // Restore floating-point registers. 2713 ldp d0, d1, [sp, #160] 2714 ldp d2, d3, [sp, #176] 2715 ldp d4, d5, [sp, #192] 2716 ldp d6, d7, [sp, #208] 2717 ldp d16, d17, [sp, #224] 2718 ldp d18, d19, [sp, #240] 2719 ldp d20, d21, [sp, #256] 2720 ldp d22, d23, [sp, #272] 2721 ldp d24, d25, [sp, #288] 2722 ldp d26, d27, [sp, #304] 2723 ldp d28, d29, [sp, #320] 2724 ldp d30, d31, [sp, #336] 2725 // Restore return address and remove padding. 2726 ldr xLR, [sp, #352] 2727 .cfi_restore x30 2728 add sp, sp, #368 2729 .cfi_adjust_cfa_offset -368 2730.Lret_rb_\name: 2731 ret 2732END \name 2733.endm 2734 2735READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg00, w0, x0 2736READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg01, w1, x1 2737READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg02, w2, x2 2738READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg03, w3, x3 2739READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg04, w4, x4 2740READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg05, w5, x5 2741READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg06, w6, x6 2742READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg07, w7, x7 2743READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg08, w8, x8 2744READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg09, w9, x9 2745READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg10, w10, x10 2746READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg11, w11, x11 2747READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg12, w12, x12 2748READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg13, w13, x13 2749READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg14, w14, x14 2750READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg15, w15, x15 2751READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg16, w16, x16 2752READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg17, w17, x17 2753READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg18, w18, x18 2754READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg19, w19, x19 2755READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg20, w20, x20 2756READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg21, w21, x21 2757READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg22, w22, x22 2758READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg23, w23, x23 2759READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg24, w24, x24 2760READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg25, w25, x25 2761READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg26, w26, x26 2762READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg27, w27, x27 2763READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg28, w28, x28 2764READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg29, w29, x29 2765