quick_entrypoints_arm64.S revision 5f404331c0ae5217d35ee8a7be77cde5c54a49de
1/* 2 * Copyright (C) 2014 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17#include "asm_support_arm64.S" 18 19#include "arch/quick_alloc_entrypoints.S" 20 21 22 /* 23 * Macro that sets up the callee save frame to conform with 24 * Runtime::CreateCalleeSaveMethod(kSaveAllCalleeSaves) 25 */ 26.macro SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 27 // art::Runtime** xIP0 = &art::Runtime::instance_ 28 adrp xIP0, :got:_ZN3art7Runtime9instance_E 29 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 30 31 // Our registers aren't intermixed - just spill in order. 32 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 33 34 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveAllCalleeSaves]; 35 ldr xIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVES_METHOD_OFFSET] 36 37 sub sp, sp, #176 38 .cfi_adjust_cfa_offset 176 39 40 // Ugly compile-time check, but we only have the preprocessor. 41#if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVES != 176) 42#error "FRAME_SIZE_SAVE_ALL_CALLEE_SAVES(ARM64) size not as expected." 43#endif 44 45 // Stack alignment filler [sp, #8]. 46 // FP callee-saves. 47 stp d8, d9, [sp, #16] 48 stp d10, d11, [sp, #32] 49 stp d12, d13, [sp, #48] 50 stp d14, d15, [sp, #64] 51 52 // GP callee-saves 53 stp x19, x20, [sp, #80] 54 .cfi_rel_offset x19, 80 55 .cfi_rel_offset x20, 88 56 57 stp x21, x22, [sp, #96] 58 .cfi_rel_offset x21, 96 59 .cfi_rel_offset x22, 104 60 61 stp x23, x24, [sp, #112] 62 .cfi_rel_offset x23, 112 63 .cfi_rel_offset x24, 120 64 65 stp x25, x26, [sp, #128] 66 .cfi_rel_offset x25, 128 67 .cfi_rel_offset x26, 136 68 69 stp x27, x28, [sp, #144] 70 .cfi_rel_offset x27, 144 71 .cfi_rel_offset x28, 152 72 73 stp x29, xLR, [sp, #160] 74 .cfi_rel_offset x29, 160 75 .cfi_rel_offset x30, 168 76 77 // Store ArtMethod* Runtime::callee_save_methods_[kSaveAllCalleeSaves]. 78 str xIP0, [sp] 79 // Place sp in Thread::Current()->top_quick_frame. 80 mov xIP0, sp 81 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 82.endm 83 84 /* 85 * Macro that sets up the callee save frame to conform with 86 * Runtime::CreateCalleeSaveMethod(kSaveRefsOnly). 87 */ 88.macro SETUP_SAVE_REFS_ONLY_FRAME 89 // art::Runtime** xIP0 = &art::Runtime::instance_ 90 adrp xIP0, :got:_ZN3art7Runtime9instance_E 91 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 92 93 // Our registers aren't intermixed - just spill in order. 94 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 95 96 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveRefOnly]; 97 ldr xIP0, [xIP0, RUNTIME_SAVE_REFS_ONLY_METHOD_OFFSET] 98 99 sub sp, sp, #96 100 .cfi_adjust_cfa_offset 96 101 102 // Ugly compile-time check, but we only have the preprocessor. 103#if (FRAME_SIZE_SAVE_REFS_ONLY != 96) 104#error "FRAME_SIZE_SAVE_REFS_ONLY(ARM64) size not as expected." 105#endif 106 107 // GP callee-saves. 108 // x20 paired with ArtMethod* - see below. 109 stp x21, x22, [sp, #16] 110 .cfi_rel_offset x21, 16 111 .cfi_rel_offset x22, 24 112 113 stp x23, x24, [sp, #32] 114 .cfi_rel_offset x23, 32 115 .cfi_rel_offset x24, 40 116 117 stp x25, x26, [sp, #48] 118 .cfi_rel_offset x25, 48 119 .cfi_rel_offset x26, 56 120 121 stp x27, x28, [sp, #64] 122 .cfi_rel_offset x27, 64 123 .cfi_rel_offset x28, 72 124 125 stp x29, xLR, [sp, #80] 126 .cfi_rel_offset x29, 80 127 .cfi_rel_offset x30, 88 128 129 // Store ArtMethod* Runtime::callee_save_methods_[kSaveRefsOnly]. 130 stp xIP0, x20, [sp] 131 .cfi_rel_offset x20, 8 132 133 // Place sp in Thread::Current()->top_quick_frame. 134 mov xIP0, sp 135 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 136.endm 137 138// TODO: Probably no need to restore registers preserved by aapcs64. 139.macro RESTORE_SAVE_REFS_ONLY_FRAME 140 // Callee-saves. 141 ldr x20, [sp, #8] 142 .cfi_restore x20 143 144 ldp x21, x22, [sp, #16] 145 .cfi_restore x21 146 .cfi_restore x22 147 148 ldp x23, x24, [sp, #32] 149 .cfi_restore x23 150 .cfi_restore x24 151 152 ldp x25, x26, [sp, #48] 153 .cfi_restore x25 154 .cfi_restore x26 155 156 ldp x27, x28, [sp, #64] 157 .cfi_restore x27 158 .cfi_restore x28 159 160 ldp x29, xLR, [sp, #80] 161 .cfi_restore x29 162 .cfi_restore x30 163 164 add sp, sp, #96 165 .cfi_adjust_cfa_offset -96 166.endm 167 168.macro POP_SAVE_REFS_ONLY_FRAME 169 add sp, sp, #96 170 .cfi_adjust_cfa_offset - 96 171.endm 172 173.macro RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN 174 RESTORE_SAVE_REFS_ONLY_FRAME 175 ret 176.endm 177 178 179.macro SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 180 sub sp, sp, #224 181 .cfi_adjust_cfa_offset 224 182 183 // Ugly compile-time check, but we only have the preprocessor. 184#if (FRAME_SIZE_SAVE_REFS_AND_ARGS != 224) 185#error "FRAME_SIZE_SAVE_REFS_AND_ARGS(ARM64) size not as expected." 186#endif 187 188 // Stack alignment filler [sp, #8]. 189 // FP args. 190 stp d0, d1, [sp, #16] 191 stp d2, d3, [sp, #32] 192 stp d4, d5, [sp, #48] 193 stp d6, d7, [sp, #64] 194 195 // Core args. 196 stp x1, x2, [sp, #80] 197 .cfi_rel_offset x1, 80 198 .cfi_rel_offset x2, 88 199 200 stp x3, x4, [sp, #96] 201 .cfi_rel_offset x3, 96 202 .cfi_rel_offset x4, 104 203 204 stp x5, x6, [sp, #112] 205 .cfi_rel_offset x5, 112 206 .cfi_rel_offset x6, 120 207 208 // x7, Callee-saves. 209 stp x7, x20, [sp, #128] 210 .cfi_rel_offset x7, 128 211 .cfi_rel_offset x20, 136 212 213 stp x21, x22, [sp, #144] 214 .cfi_rel_offset x21, 144 215 .cfi_rel_offset x22, 152 216 217 stp x23, x24, [sp, #160] 218 .cfi_rel_offset x23, 160 219 .cfi_rel_offset x24, 168 220 221 stp x25, x26, [sp, #176] 222 .cfi_rel_offset x25, 176 223 .cfi_rel_offset x26, 184 224 225 stp x27, x28, [sp, #192] 226 .cfi_rel_offset x27, 192 227 .cfi_rel_offset x28, 200 228 229 // x29(callee-save) and LR. 230 stp x29, xLR, [sp, #208] 231 .cfi_rel_offset x29, 208 232 .cfi_rel_offset x30, 216 233 234.endm 235 236 /* 237 * Macro that sets up the callee save frame to conform with 238 * Runtime::CreateCalleeSaveMethod(kSaveRefsAndArgs). 239 * 240 * TODO This is probably too conservative - saving FP & LR. 241 */ 242.macro SETUP_SAVE_REFS_AND_ARGS_FRAME 243 // art::Runtime** xIP0 = &art::Runtime::instance_ 244 adrp xIP0, :got:_ZN3art7Runtime9instance_E 245 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 246 247 // Our registers aren't intermixed - just spill in order. 248 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 249 250 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveRefAndArgs]; 251 ldr xIP0, [xIP0, RUNTIME_SAVE_REFS_AND_ARGS_METHOD_OFFSET] 252 253 SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 254 255 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kSaveRefsAndArgs]. 256 // Place sp in Thread::Current()->top_quick_frame. 257 mov xIP0, sp 258 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 259.endm 260 261.macro SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 262 SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 263 str x0, [sp, #0] // Store ArtMethod* to bottom of stack. 264 // Place sp in Thread::Current()->top_quick_frame. 265 mov xIP0, sp 266 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 267.endm 268 269// TODO: Probably no need to restore registers preserved by aapcs64. 270.macro RESTORE_SAVE_REFS_AND_ARGS_FRAME 271 // FP args. 272 ldp d0, d1, [sp, #16] 273 ldp d2, d3, [sp, #32] 274 ldp d4, d5, [sp, #48] 275 ldp d6, d7, [sp, #64] 276 277 // Core args. 278 ldp x1, x2, [sp, #80] 279 .cfi_restore x1 280 .cfi_restore x2 281 282 ldp x3, x4, [sp, #96] 283 .cfi_restore x3 284 .cfi_restore x4 285 286 ldp x5, x6, [sp, #112] 287 .cfi_restore x5 288 .cfi_restore x6 289 290 // x7, Callee-saves. 291 ldp x7, x20, [sp, #128] 292 .cfi_restore x7 293 .cfi_restore x20 294 295 ldp x21, x22, [sp, #144] 296 .cfi_restore x21 297 .cfi_restore x22 298 299 ldp x23, x24, [sp, #160] 300 .cfi_restore x23 301 .cfi_restore x24 302 303 ldp x25, x26, [sp, #176] 304 .cfi_restore x25 305 .cfi_restore x26 306 307 ldp x27, x28, [sp, #192] 308 .cfi_restore x27 309 .cfi_restore x28 310 311 // x29(callee-save) and LR. 312 ldp x29, xLR, [sp, #208] 313 .cfi_restore x29 314 .cfi_restore x30 315 316 add sp, sp, #224 317 .cfi_adjust_cfa_offset -224 318.endm 319 320 /* 321 * Macro that sets up the callee save frame to conform with 322 * Runtime::CreateCalleeSaveMethod(kSaveEverything) 323 */ 324.macro SETUP_SAVE_EVERYTHING_FRAME 325 sub sp, sp, #512 326 .cfi_adjust_cfa_offset 512 327 328 // Ugly compile-time check, but we only have the preprocessor. 329#if (FRAME_SIZE_SAVE_EVERYTHING != 512) 330#error "FRAME_SIZE_SAVE_EVERYTHING(ARM64) size not as expected." 331#endif 332 333 // Save FP registers. 334 // For better performance, store d0 and d31 separately, so that all STPs are 16-byte aligned. 335 str d0, [sp, #8] 336 stp d1, d2, [sp, #16] 337 stp d3, d4, [sp, #32] 338 stp d5, d6, [sp, #48] 339 stp d7, d8, [sp, #64] 340 stp d9, d10, [sp, #80] 341 stp d11, d12, [sp, #96] 342 stp d13, d14, [sp, #112] 343 stp d15, d16, [sp, #128] 344 stp d17, d18, [sp, #144] 345 stp d19, d20, [sp, #160] 346 stp d21, d22, [sp, #176] 347 stp d23, d24, [sp, #192] 348 stp d25, d26, [sp, #208] 349 stp d27, d28, [sp, #224] 350 stp d29, d30, [sp, #240] 351 str d31, [sp, #256] 352 353 // Save core registers. 354 str x0, [sp, #264] 355 .cfi_rel_offset x0, 264 356 357 stp x1, x2, [sp, #272] 358 .cfi_rel_offset x1, 272 359 .cfi_rel_offset x2, 280 360 361 stp x3, x4, [sp, #288] 362 .cfi_rel_offset x3, 288 363 .cfi_rel_offset x4, 296 364 365 stp x5, x6, [sp, #304] 366 .cfi_rel_offset x5, 304 367 .cfi_rel_offset x6, 312 368 369 stp x7, x8, [sp, #320] 370 .cfi_rel_offset x7, 320 371 .cfi_rel_offset x8, 328 372 373 stp x9, x10, [sp, #336] 374 .cfi_rel_offset x9, 336 375 .cfi_rel_offset x10, 344 376 377 stp x11, x12, [sp, #352] 378 .cfi_rel_offset x11, 352 379 .cfi_rel_offset x12, 360 380 381 stp x13, x14, [sp, #368] 382 .cfi_rel_offset x13, 368 383 .cfi_rel_offset x14, 376 384 385 stp x15, x16, [sp, #384] 386 .cfi_rel_offset x15, 384 387 .cfi_rel_offset x16, 392 388 389 stp x17, x18, [sp, #400] 390 .cfi_rel_offset x17, 400 391 .cfi_rel_offset x18, 408 392 393 stp x19, x20, [sp, #416] 394 .cfi_rel_offset x19, 416 395 .cfi_rel_offset x20, 424 396 397 stp x21, x22, [sp, #432] 398 .cfi_rel_offset x21, 432 399 .cfi_rel_offset x22, 440 400 401 stp x23, x24, [sp, #448] 402 .cfi_rel_offset x23, 448 403 .cfi_rel_offset x24, 456 404 405 stp x25, x26, [sp, #464] 406 .cfi_rel_offset x25, 464 407 .cfi_rel_offset x26, 472 408 409 stp x27, x28, [sp, #480] 410 .cfi_rel_offset x27, 480 411 .cfi_rel_offset x28, 488 412 413 stp x29, xLR, [sp, #496] 414 .cfi_rel_offset x29, 496 415 .cfi_rel_offset x30, 504 416 417 // art::Runtime** xIP0 = &art::Runtime::instance_ 418 adrp xIP0, :got:_ZN3art7Runtime9instance_E 419 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 420 421 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 422 423 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveEverything]; 424 ldr xIP0, [xIP0, RUNTIME_SAVE_EVERYTHING_METHOD_OFFSET] 425 426 // Store ArtMethod* Runtime::callee_save_methods_[kSaveEverything]. 427 str xIP0, [sp] 428 // Place sp in Thread::Current()->top_quick_frame. 429 mov xIP0, sp 430 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 431.endm 432 433.macro RESTORE_SAVE_EVERYTHING_FRAME 434 // Restore FP registers. 435 // For better performance, load d0 and d31 separately, so that all LDPs are 16-byte aligned. 436 ldr d0, [sp, #8] 437 ldp d1, d2, [sp, #16] 438 ldp d3, d4, [sp, #32] 439 ldp d5, d6, [sp, #48] 440 ldp d7, d8, [sp, #64] 441 ldp d9, d10, [sp, #80] 442 ldp d11, d12, [sp, #96] 443 ldp d13, d14, [sp, #112] 444 ldp d15, d16, [sp, #128] 445 ldp d17, d18, [sp, #144] 446 ldp d19, d20, [sp, #160] 447 ldp d21, d22, [sp, #176] 448 ldp d23, d24, [sp, #192] 449 ldp d25, d26, [sp, #208] 450 ldp d27, d28, [sp, #224] 451 ldp d29, d30, [sp, #240] 452 ldr d31, [sp, #256] 453 454 // Restore core registers. 455 ldr x0, [sp, #264] 456 .cfi_restore x0 457 458 ldp x1, x2, [sp, #272] 459 .cfi_restore x1 460 .cfi_restore x2 461 462 ldp x3, x4, [sp, #288] 463 .cfi_restore x3 464 .cfi_restore x4 465 466 ldp x5, x6, [sp, #304] 467 .cfi_restore x5 468 .cfi_restore x6 469 470 ldp x7, x8, [sp, #320] 471 .cfi_restore x7 472 .cfi_restore x8 473 474 ldp x9, x10, [sp, #336] 475 .cfi_restore x9 476 .cfi_restore x10 477 478 ldp x11, x12, [sp, #352] 479 .cfi_restore x11 480 .cfi_restore x12 481 482 ldp x13, x14, [sp, #368] 483 .cfi_restore x13 484 .cfi_restore x14 485 486 ldp x15, x16, [sp, #384] 487 .cfi_restore x15 488 .cfi_restore x16 489 490 ldp x17, x18, [sp, #400] 491 .cfi_restore x17 492 .cfi_restore x18 493 494 ldp x19, x20, [sp, #416] 495 .cfi_restore x19 496 .cfi_restore x20 497 498 ldp x21, x22, [sp, #432] 499 .cfi_restore x21 500 .cfi_restore x22 501 502 ldp x23, x24, [sp, #448] 503 .cfi_restore x23 504 .cfi_restore x24 505 506 ldp x25, x26, [sp, #464] 507 .cfi_restore x25 508 .cfi_restore x26 509 510 ldp x27, x28, [sp, #480] 511 .cfi_restore x27 512 .cfi_restore x28 513 514 ldp x29, xLR, [sp, #496] 515 .cfi_restore x29 516 .cfi_restore x30 517 518 add sp, sp, #512 519 .cfi_adjust_cfa_offset -512 520.endm 521 522.macro RETURN_IF_RESULT_IS_ZERO 523 cbnz x0, 1f // result non-zero branch over 524 ret // return 5251: 526.endm 527 528.macro RETURN_IF_RESULT_IS_NON_ZERO 529 cbz x0, 1f // result zero branch over 530 ret // return 5311: 532.endm 533 534 /* 535 * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending 536 * exception is Thread::Current()->exception_ 537 */ 538.macro DELIVER_PENDING_EXCEPTION 539 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 540 mov x0, xSELF 541 542 // Point of no return. 543 b artDeliverPendingExceptionFromCode // artDeliverPendingExceptionFromCode(Thread*) 544 brk 0 // Unreached 545.endm 546 547.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg 548 ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET] // Get exception field. 549 cbnz \reg, 1f 550 ret 5511: 552 DELIVER_PENDING_EXCEPTION 553.endm 554 555.macro RETURN_OR_DELIVER_PENDING_EXCEPTION 556 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0 557.endm 558 559// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register. 560.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 561 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1 562.endm 563 564.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER 565 cbnz w0, 1f // result non-zero branch over 566 ret // return 5671: 568 DELIVER_PENDING_EXCEPTION 569.endm 570 571.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 572 .extern \cxx_name 573ENTRY \c_name 574 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context 575 mov x0, xSELF // pass Thread::Current 576 b \cxx_name // \cxx_name(Thread*) 577END \c_name 578.endm 579 580.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name 581 .extern \cxx_name 582ENTRY \c_name 583 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context. 584 mov x1, xSELF // pass Thread::Current. 585 b \cxx_name // \cxx_name(arg, Thread*). 586 brk 0 587END \c_name 588.endm 589 590.macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 591 .extern \cxx_name 592ENTRY \c_name 593 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context 594 mov x2, xSELF // pass Thread::Current 595 b \cxx_name // \cxx_name(arg1, arg2, Thread*) 596 brk 0 597END \c_name 598.endm 599 600 /* 601 * Called by managed code, saves callee saves and then calls artThrowException 602 * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. 603 */ 604ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode 605 606 /* 607 * Called by managed code to create and deliver a NullPointerException. 608 */ 609NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode 610 611 /* 612 * Call installed by a signal handler to create and deliver a NullPointerException. 613 */ 614ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception_from_signal, artThrowNullPointerExceptionFromSignal 615 616 /* 617 * Called by managed code to create and deliver an ArithmeticException. 618 */ 619NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode 620 621 /* 622 * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds 623 * index, arg2 holds limit. 624 */ 625TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode 626 627 /* 628 * Called by managed code to create and deliver a StringIndexOutOfBoundsException 629 * as if thrown from a call to String.charAt(). Arg1 holds index, arg2 holds limit. 630 */ 631TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_string_bounds, artThrowStringBoundsFromCode 632 633 /* 634 * Called by managed code to create and deliver a StackOverflowError. 635 */ 636NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode 637 638 /* 639 * Called by managed code to create and deliver a NoSuchMethodError. 640 */ 641ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode 642 643 /* 644 * All generated callsites for interface invokes and invocation slow paths will load arguments 645 * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain 646 * the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper. 647 * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1. 648 * 649 * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting 650 * of the target Method* in x0 and method->code_ in x1. 651 * 652 * If unsuccessful, the helper will return null/????. There will be a pending exception in the 653 * thread and we branch to another stub to deliver it. 654 * 655 * On success this wrapper will restore arguments and *jump* to the target, leaving the lr 656 * pointing back to the original caller. 657 * 658 * Adapted from ARM32 code. 659 * 660 * Clobbers xIP0. 661 */ 662.macro INVOKE_TRAMPOLINE_BODY cxx_name 663 .extern \cxx_name 664 SETUP_SAVE_REFS_AND_ARGS_FRAME // save callee saves in case allocation triggers GC 665 // Helper signature is always 666 // (method_idx, *this_object, *caller_method, *self, sp) 667 668 mov x2, xSELF // pass Thread::Current 669 mov x3, sp 670 bl \cxx_name // (method_idx, this, Thread*, SP) 671 mov xIP0, x1 // save Method*->code_ 672 RESTORE_SAVE_REFS_AND_ARGS_FRAME 673 cbz x0, 1f // did we find the target? if not go to exception delivery 674 br xIP0 // tail call to target 6751: 676 DELIVER_PENDING_EXCEPTION 677.endm 678.macro INVOKE_TRAMPOLINE c_name, cxx_name 679ENTRY \c_name 680 INVOKE_TRAMPOLINE_BODY \cxx_name 681END \c_name 682.endm 683 684INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck 685 686INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck 687INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck 688INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck 689INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck 690 691 692.macro INVOKE_STUB_CREATE_FRAME 693 694SAVE_SIZE=15*8 // x4, x5, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 695SAVE_SIZE_AND_METHOD=SAVE_SIZE+8 696 697 698 mov x9, sp // Save stack pointer. 699 .cfi_register sp,x9 700 701 add x10, x2, # SAVE_SIZE_AND_METHOD // calculate size of frame. 702 sub x10, sp, x10 // Calculate SP position - saves + ArtMethod* + args 703 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 704 mov sp, x10 // Set new SP. 705 706 sub x10, x9, #SAVE_SIZE // Calculate new FP (later). Done here as we must move SP 707 .cfi_def_cfa_register x10 // before this. 708 .cfi_adjust_cfa_offset SAVE_SIZE 709 710 str x28, [x10, #112] 711 .cfi_rel_offset x28, 112 712 713 stp x26, x27, [x10, #96] 714 .cfi_rel_offset x26, 96 715 .cfi_rel_offset x27, 104 716 717 stp x24, x25, [x10, #80] 718 .cfi_rel_offset x24, 80 719 .cfi_rel_offset x25, 88 720 721 stp x22, x23, [x10, #64] 722 .cfi_rel_offset x22, 64 723 .cfi_rel_offset x23, 72 724 725 stp x20, x21, [x10, #48] 726 .cfi_rel_offset x20, 48 727 .cfi_rel_offset x21, 56 728 729 stp x9, x19, [x10, #32] // Save old stack pointer and x19. 730 .cfi_rel_offset sp, 32 731 .cfi_rel_offset x19, 40 732 733 stp x4, x5, [x10, #16] // Save result and shorty addresses. 734 .cfi_rel_offset x4, 16 735 .cfi_rel_offset x5, 24 736 737 stp xFP, xLR, [x10] // Store LR & FP. 738 .cfi_rel_offset x29, 0 739 .cfi_rel_offset x30, 8 740 741 mov xFP, x10 // Use xFP now, as it's callee-saved. 742 .cfi_def_cfa_register x29 743 mov xSELF, x3 // Move thread pointer into SELF register. 744 745 // Copy arguments into stack frame. 746 // Use simple copy routine for now. 747 // 4 bytes per slot. 748 // X1 - source address 749 // W2 - args length 750 // X9 - destination address. 751 // W10 - temporary 752 add x9, sp, #8 // Destination address is bottom of stack + null. 753 754 // Copy parameters into the stack. Use numeric label as this is a macro and Clang's assembler 755 // does not have unique-id variables. 7561: 757 cmp w2, #0 758 beq 2f 759 sub w2, w2, #4 // Need 65536 bytes of range. 760 ldr w10, [x1, x2] 761 str w10, [x9, x2] 762 763 b 1b 764 7652: 766 // Store null into ArtMethod* at bottom of frame. 767 str xzr, [sp] 768.endm 769 770.macro INVOKE_STUB_CALL_AND_RETURN 771 772 // load method-> METHOD_QUICK_CODE_OFFSET 773 ldr x9, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 774 // Branch to method. 775 blr x9 776 777 // Restore return value address and shorty address. 778 ldp x4,x5, [xFP, #16] 779 .cfi_restore x4 780 .cfi_restore x5 781 782 ldr x28, [xFP, #112] 783 .cfi_restore x28 784 785 ldp x26, x27, [xFP, #96] 786 .cfi_restore x26 787 .cfi_restore x27 788 789 ldp x24, x25, [xFP, #80] 790 .cfi_restore x24 791 .cfi_restore x25 792 793 ldp x22, x23, [xFP, #64] 794 .cfi_restore x22 795 .cfi_restore x23 796 797 ldp x20, x21, [xFP, #48] 798 .cfi_restore x20 799 .cfi_restore x21 800 801 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 802 ldrb w10, [x5] 803 804 // Check the return type and store the correct register into the jvalue in memory. 805 // Use numeric label as this is a macro and Clang's assembler does not have unique-id variables. 806 807 // Don't set anything for a void type. 808 cmp w10, #'V' 809 beq 3f 810 811 // Is it a double? 812 cmp w10, #'D' 813 bne 1f 814 str d0, [x4] 815 b 3f 816 8171: // Is it a float? 818 cmp w10, #'F' 819 bne 2f 820 str s0, [x4] 821 b 3f 822 8232: // Just store x0. Doesn't matter if it is 64 or 32 bits. 824 str x0, [x4] 825 8263: // Finish up. 827 ldp x2, x19, [xFP, #32] // Restore stack pointer and x19. 828 .cfi_restore x19 829 mov sp, x2 830 .cfi_restore sp 831 832 ldp xFP, xLR, [xFP] // Restore old frame pointer and link register. 833 .cfi_restore x29 834 .cfi_restore x30 835 836 ret 837 838.endm 839 840 841/* 842 * extern"C" void art_quick_invoke_stub(ArtMethod *method, x0 843 * uint32_t *args, x1 844 * uint32_t argsize, w2 845 * Thread *self, x3 846 * JValue *result, x4 847 * char *shorty); x5 848 * +----------------------+ 849 * | | 850 * | C/C++ frame | 851 * | LR'' | 852 * | FP'' | <- SP' 853 * +----------------------+ 854 * +----------------------+ 855 * | x28 | <- TODO: Remove callee-saves. 856 * | : | 857 * | x19 | 858 * | SP' | 859 * | X5 | 860 * | X4 | Saved registers 861 * | LR' | 862 * | FP' | <- FP 863 * +----------------------+ 864 * | uint32_t out[n-1] | 865 * | : : | Outs 866 * | uint32_t out[0] | 867 * | ArtMethod* | <- SP value=null 868 * +----------------------+ 869 * 870 * Outgoing registers: 871 * x0 - Method* 872 * x1-x7 - integer parameters. 873 * d0-d7 - Floating point parameters. 874 * xSELF = self 875 * SP = & of ArtMethod* 876 * x1 = "this" pointer. 877 * 878 */ 879ENTRY art_quick_invoke_stub 880 // Spill registers as per AACPS64 calling convention. 881 INVOKE_STUB_CREATE_FRAME 882 883 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 884 // Parse the passed shorty to determine which register to load. 885 // Load addresses for routines that load WXSD registers. 886 adr x11, .LstoreW2 887 adr x12, .LstoreX2 888 adr x13, .LstoreS0 889 adr x14, .LstoreD0 890 891 // Initialize routine offsets to 0 for integers and floats. 892 // x8 for integers, x15 for floating point. 893 mov x8, #0 894 mov x15, #0 895 896 add x10, x5, #1 // Load shorty address, plus one to skip return value. 897 ldr w1, [x9],#4 // Load "this" parameter, and increment arg pointer. 898 899 // Loop to fill registers. 900.LfillRegisters: 901 ldrb w17, [x10], #1 // Load next character in signature, and increment. 902 cbz w17, .LcallFunction // Exit at end of signature. Shorty 0 terminated. 903 904 cmp w17, #'F' // is this a float? 905 bne .LisDouble 906 907 cmp x15, # 8*12 // Skip this load if all registers full. 908 beq .Ladvance4 909 910 add x17, x13, x15 // Calculate subroutine to jump to. 911 br x17 912 913.LisDouble: 914 cmp w17, #'D' // is this a double? 915 bne .LisLong 916 917 cmp x15, # 8*12 // Skip this load if all registers full. 918 beq .Ladvance8 919 920 add x17, x14, x15 // Calculate subroutine to jump to. 921 br x17 922 923.LisLong: 924 cmp w17, #'J' // is this a long? 925 bne .LisOther 926 927 cmp x8, # 6*12 // Skip this load if all registers full. 928 beq .Ladvance8 929 930 add x17, x12, x8 // Calculate subroutine to jump to. 931 br x17 932 933.LisOther: // Everything else takes one vReg. 934 cmp x8, # 6*12 // Skip this load if all registers full. 935 beq .Ladvance4 936 937 add x17, x11, x8 // Calculate subroutine to jump to. 938 br x17 939 940.Ladvance4: 941 add x9, x9, #4 942 b .LfillRegisters 943 944.Ladvance8: 945 add x9, x9, #8 946 b .LfillRegisters 947 948// Macro for loading a parameter into a register. 949// counter - the register with offset into these tables 950// size - the size of the register - 4 or 8 bytes. 951// register - the name of the register to be loaded. 952.macro LOADREG counter size register return 953 ldr \register , [x9], #\size 954 add \counter, \counter, 12 955 b \return 956.endm 957 958// Store ints. 959.LstoreW2: 960 LOADREG x8 4 w2 .LfillRegisters 961 LOADREG x8 4 w3 .LfillRegisters 962 LOADREG x8 4 w4 .LfillRegisters 963 LOADREG x8 4 w5 .LfillRegisters 964 LOADREG x8 4 w6 .LfillRegisters 965 LOADREG x8 4 w7 .LfillRegisters 966 967// Store longs. 968.LstoreX2: 969 LOADREG x8 8 x2 .LfillRegisters 970 LOADREG x8 8 x3 .LfillRegisters 971 LOADREG x8 8 x4 .LfillRegisters 972 LOADREG x8 8 x5 .LfillRegisters 973 LOADREG x8 8 x6 .LfillRegisters 974 LOADREG x8 8 x7 .LfillRegisters 975 976// Store singles. 977.LstoreS0: 978 LOADREG x15 4 s0 .LfillRegisters 979 LOADREG x15 4 s1 .LfillRegisters 980 LOADREG x15 4 s2 .LfillRegisters 981 LOADREG x15 4 s3 .LfillRegisters 982 LOADREG x15 4 s4 .LfillRegisters 983 LOADREG x15 4 s5 .LfillRegisters 984 LOADREG x15 4 s6 .LfillRegisters 985 LOADREG x15 4 s7 .LfillRegisters 986 987// Store doubles. 988.LstoreD0: 989 LOADREG x15 8 d0 .LfillRegisters 990 LOADREG x15 8 d1 .LfillRegisters 991 LOADREG x15 8 d2 .LfillRegisters 992 LOADREG x15 8 d3 .LfillRegisters 993 LOADREG x15 8 d4 .LfillRegisters 994 LOADREG x15 8 d5 .LfillRegisters 995 LOADREG x15 8 d6 .LfillRegisters 996 LOADREG x15 8 d7 .LfillRegisters 997 998 999.LcallFunction: 1000 1001 INVOKE_STUB_CALL_AND_RETURN 1002 1003END art_quick_invoke_stub 1004 1005/* extern"C" 1006 * void art_quick_invoke_static_stub(ArtMethod *method, x0 1007 * uint32_t *args, x1 1008 * uint32_t argsize, w2 1009 * Thread *self, x3 1010 * JValue *result, x4 1011 * char *shorty); x5 1012 */ 1013ENTRY art_quick_invoke_static_stub 1014 // Spill registers as per AACPS64 calling convention. 1015 INVOKE_STUB_CREATE_FRAME 1016 1017 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 1018 // Parse the passed shorty to determine which register to load. 1019 // Load addresses for routines that load WXSD registers. 1020 adr x11, .LstoreW1_2 1021 adr x12, .LstoreX1_2 1022 adr x13, .LstoreS0_2 1023 adr x14, .LstoreD0_2 1024 1025 // Initialize routine offsets to 0 for integers and floats. 1026 // x8 for integers, x15 for floating point. 1027 mov x8, #0 1028 mov x15, #0 1029 1030 add x10, x5, #1 // Load shorty address, plus one to skip return value. 1031 1032 // Loop to fill registers. 1033.LfillRegisters2: 1034 ldrb w17, [x10], #1 // Load next character in signature, and increment. 1035 cbz w17, .LcallFunction2 // Exit at end of signature. Shorty 0 terminated. 1036 1037 cmp w17, #'F' // is this a float? 1038 bne .LisDouble2 1039 1040 cmp x15, # 8*12 // Skip this load if all registers full. 1041 beq .Ladvance4_2 1042 1043 add x17, x13, x15 // Calculate subroutine to jump to. 1044 br x17 1045 1046.LisDouble2: 1047 cmp w17, #'D' // is this a double? 1048 bne .LisLong2 1049 1050 cmp x15, # 8*12 // Skip this load if all registers full. 1051 beq .Ladvance8_2 1052 1053 add x17, x14, x15 // Calculate subroutine to jump to. 1054 br x17 1055 1056.LisLong2: 1057 cmp w17, #'J' // is this a long? 1058 bne .LisOther2 1059 1060 cmp x8, # 7*12 // Skip this load if all registers full. 1061 beq .Ladvance8_2 1062 1063 add x17, x12, x8 // Calculate subroutine to jump to. 1064 br x17 1065 1066.LisOther2: // Everything else takes one vReg. 1067 cmp x8, # 7*12 // Skip this load if all registers full. 1068 beq .Ladvance4_2 1069 1070 add x17, x11, x8 // Calculate subroutine to jump to. 1071 br x17 1072 1073.Ladvance4_2: 1074 add x9, x9, #4 1075 b .LfillRegisters2 1076 1077.Ladvance8_2: 1078 add x9, x9, #8 1079 b .LfillRegisters2 1080 1081// Store ints. 1082.LstoreW1_2: 1083 LOADREG x8 4 w1 .LfillRegisters2 1084 LOADREG x8 4 w2 .LfillRegisters2 1085 LOADREG x8 4 w3 .LfillRegisters2 1086 LOADREG x8 4 w4 .LfillRegisters2 1087 LOADREG x8 4 w5 .LfillRegisters2 1088 LOADREG x8 4 w6 .LfillRegisters2 1089 LOADREG x8 4 w7 .LfillRegisters2 1090 1091// Store longs. 1092.LstoreX1_2: 1093 LOADREG x8 8 x1 .LfillRegisters2 1094 LOADREG x8 8 x2 .LfillRegisters2 1095 LOADREG x8 8 x3 .LfillRegisters2 1096 LOADREG x8 8 x4 .LfillRegisters2 1097 LOADREG x8 8 x5 .LfillRegisters2 1098 LOADREG x8 8 x6 .LfillRegisters2 1099 LOADREG x8 8 x7 .LfillRegisters2 1100 1101// Store singles. 1102.LstoreS0_2: 1103 LOADREG x15 4 s0 .LfillRegisters2 1104 LOADREG x15 4 s1 .LfillRegisters2 1105 LOADREG x15 4 s2 .LfillRegisters2 1106 LOADREG x15 4 s3 .LfillRegisters2 1107 LOADREG x15 4 s4 .LfillRegisters2 1108 LOADREG x15 4 s5 .LfillRegisters2 1109 LOADREG x15 4 s6 .LfillRegisters2 1110 LOADREG x15 4 s7 .LfillRegisters2 1111 1112// Store doubles. 1113.LstoreD0_2: 1114 LOADREG x15 8 d0 .LfillRegisters2 1115 LOADREG x15 8 d1 .LfillRegisters2 1116 LOADREG x15 8 d2 .LfillRegisters2 1117 LOADREG x15 8 d3 .LfillRegisters2 1118 LOADREG x15 8 d4 .LfillRegisters2 1119 LOADREG x15 8 d5 .LfillRegisters2 1120 LOADREG x15 8 d6 .LfillRegisters2 1121 LOADREG x15 8 d7 .LfillRegisters2 1122 1123 1124.LcallFunction2: 1125 1126 INVOKE_STUB_CALL_AND_RETURN 1127 1128END art_quick_invoke_static_stub 1129 1130 1131 1132/* extern"C" void art_quick_osr_stub(void** stack, x0 1133 * size_t stack_size_in_bytes, x1 1134 * const uin8_t* native_pc, x2 1135 * JValue *result, x3 1136 * char *shorty, x4 1137 * Thread *self) x5 1138 */ 1139ENTRY art_quick_osr_stub 1140SAVE_SIZE=15*8 // x3, x4, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 1141 mov x9, sp // Save stack pointer. 1142 .cfi_register sp,x9 1143 1144 sub x10, sp, # SAVE_SIZE 1145 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 1146 mov sp, x10 // Set new SP. 1147 1148 str x28, [sp, #112] 1149 stp x26, x27, [sp, #96] 1150 stp x24, x25, [sp, #80] 1151 stp x22, x23, [sp, #64] 1152 stp x20, x21, [sp, #48] 1153 stp x9, x19, [sp, #32] // Save old stack pointer and x19. 1154 stp x3, x4, [sp, #16] // Save result and shorty addresses. 1155 stp xFP, xLR, [sp] // Store LR & FP. 1156 mov xSELF, x5 // Move thread pointer into SELF register. 1157 1158 sub sp, sp, #16 1159 str xzr, [sp] // Store null for ArtMethod* slot 1160 // Branch to stub. 1161 bl .Losr_entry 1162 add sp, sp, #16 1163 1164 // Restore return value address and shorty address. 1165 ldp x3,x4, [sp, #16] 1166 ldr x28, [sp, #112] 1167 ldp x26, x27, [sp, #96] 1168 ldp x24, x25, [sp, #80] 1169 ldp x22, x23, [sp, #64] 1170 ldp x20, x21, [sp, #48] 1171 1172 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 1173 ldrb w10, [x4] 1174 1175 // Check the return type and store the correct register into the jvalue in memory. 1176 1177 // Don't set anything for a void type. 1178 cmp w10, #'V' 1179 beq .Losr_exit 1180 1181 // Is it a double? 1182 cmp w10, #'D' 1183 bne .Lno_double 1184 str d0, [x3] 1185 b .Losr_exit 1186 1187.Lno_double: // Is it a float? 1188 cmp w10, #'F' 1189 bne .Lno_float 1190 str s0, [x3] 1191 b .Losr_exit 1192 1193.Lno_float: // Just store x0. Doesn't matter if it is 64 or 32 bits. 1194 str x0, [x3] 1195 1196.Losr_exit: // Finish up. 1197 ldp x2, x19, [sp, #32] // Restore stack pointer and x19. 1198 ldp xFP, xLR, [sp] // Restore old frame pointer and link register. 1199 mov sp, x2 1200 ret 1201 1202.Losr_entry: 1203 // Update stack pointer for the callee 1204 sub sp, sp, x1 1205 1206 // Update link register slot expected by the callee. 1207 sub w1, w1, #8 1208 str lr, [sp, x1] 1209 1210 // Copy arguments into stack frame. 1211 // Use simple copy routine for now. 1212 // 4 bytes per slot. 1213 // X0 - source address 1214 // W1 - args length 1215 // SP - destination address. 1216 // W10 - temporary 1217.Losr_loop_entry: 1218 cmp w1, #0 1219 beq .Losr_loop_exit 1220 sub w1, w1, #4 1221 ldr w10, [x0, x1] 1222 str w10, [sp, x1] 1223 b .Losr_loop_entry 1224 1225.Losr_loop_exit: 1226 // Branch to the OSR entry point. 1227 br x2 1228 1229END art_quick_osr_stub 1230 1231 /* 1232 * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_ 1233 */ 1234 1235ENTRY art_quick_do_long_jump 1236 // Load FPRs 1237 ldp d0, d1, [x1], #16 1238 ldp d2, d3, [x1], #16 1239 ldp d4, d5, [x1], #16 1240 ldp d6, d7, [x1], #16 1241 ldp d8, d9, [x1], #16 1242 ldp d10, d11, [x1], #16 1243 ldp d12, d13, [x1], #16 1244 ldp d14, d15, [x1], #16 1245 ldp d16, d17, [x1], #16 1246 ldp d18, d19, [x1], #16 1247 ldp d20, d21, [x1], #16 1248 ldp d22, d23, [x1], #16 1249 ldp d24, d25, [x1], #16 1250 ldp d26, d27, [x1], #16 1251 ldp d28, d29, [x1], #16 1252 ldp d30, d31, [x1] 1253 1254 // Load GPRs 1255 // TODO: lots of those are smashed, could optimize. 1256 add x0, x0, #30*8 1257 ldp x30, x1, [x0], #-16 // LR & SP 1258 ldp x28, x29, [x0], #-16 1259 ldp x26, x27, [x0], #-16 1260 ldp x24, x25, [x0], #-16 1261 ldp x22, x23, [x0], #-16 1262 ldp x20, x21, [x0], #-16 1263 ldp x18, x19, [x0], #-16 1264 ldp x16, x17, [x0], #-16 1265 ldp x14, x15, [x0], #-16 1266 ldp x12, x13, [x0], #-16 1267 ldp x10, x11, [x0], #-16 1268 ldp x8, x9, [x0], #-16 1269 ldp x6, x7, [x0], #-16 1270 ldp x4, x5, [x0], #-16 1271 ldp x2, x3, [x0], #-16 1272 mov sp, x1 1273 1274 // Need to load PC, it's at the end (after the space for the unused XZR). Use x1. 1275 ldr x1, [x0, #33*8] 1276 // And the value of x0. 1277 ldr x0, [x0] 1278 1279 br x1 1280END art_quick_do_long_jump 1281 1282 /* 1283 * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the 1284 * possibly null object to lock. 1285 * 1286 * Derived from arm32 code. 1287 */ 1288 .extern artLockObjectFromCode 1289ENTRY art_quick_lock_object 1290 cbz w0, .Lslow_lock 1291 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1292.Lretry_lock: 1293 ldr w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop? 1294 ldxr w1, [x4] 1295 mov x3, x1 1296 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1297 cbnz w3, .Lnot_unlocked // already thin locked 1298 // unlocked case - x1: original lock word that's zero except for the read barrier bits. 1299 orr x2, x1, x2 // x2 holds thread id with count of 0 with preserved read barrier bits 1300 stxr w3, w2, [x4] 1301 cbnz w3, .Llock_stxr_fail // store failed, retry 1302 dmb ishld // full (LoadLoad|LoadStore) memory barrier 1303 ret 1304.Lnot_unlocked: // x1: original lock word 1305 lsr w3, w1, LOCK_WORD_STATE_SHIFT 1306 cbnz w3, .Lslow_lock // if either of the top two bits are set, go slow path 1307 eor w2, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 1308 uxth w2, w2 // zero top 16 bits 1309 cbnz w2, .Lslow_lock // lock word and self thread id's match -> recursive lock 1310 // else contention, go to slow path 1311 mov x3, x1 // copy the lock word to check count overflow. 1312 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits. 1313 add w2, w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count in lock word placing in w2 to check overflow 1314 lsr w3, w2, #LOCK_WORD_GC_STATE_SHIFT // if the first gc state bit is set, we overflowed. 1315 cbnz w3, .Lslow_lock // if we overflow the count go slow path 1316 add w2, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count for real 1317 stxr w3, w2, [x4] 1318 cbnz w3, .Llock_stxr_fail // store failed, retry 1319 ret 1320.Llock_stxr_fail: 1321 b .Lretry_lock // retry 1322.Lslow_lock: 1323 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case we block 1324 mov x1, xSELF // pass Thread::Current 1325 bl artLockObjectFromCode // (Object* obj, Thread*) 1326 RESTORE_SAVE_REFS_ONLY_FRAME 1327 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1328END art_quick_lock_object 1329 1330ENTRY art_quick_lock_object_no_inline 1331 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case we block 1332 mov x1, xSELF // pass Thread::Current 1333 bl artLockObjectFromCode // (Object* obj, Thread*) 1334 RESTORE_SAVE_REFS_ONLY_FRAME 1335 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1336END art_quick_lock_object_no_inline 1337 1338 /* 1339 * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. 1340 * x0 holds the possibly null object to lock. 1341 * 1342 * Derived from arm32 code. 1343 */ 1344 .extern artUnlockObjectFromCode 1345ENTRY art_quick_unlock_object 1346 cbz x0, .Lslow_unlock 1347 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1348.Lretry_unlock: 1349#ifndef USE_READ_BARRIER 1350 ldr w1, [x4] 1351#else 1352 ldxr w1, [x4] // Need to use atomic instructions for read barrier 1353#endif 1354 lsr w2, w1, LOCK_WORD_STATE_SHIFT 1355 cbnz w2, .Lslow_unlock // if either of the top two bits are set, go slow path 1356 ldr w2, [xSELF, #THREAD_ID_OFFSET] 1357 mov x3, x1 // copy lock word to check thread id equality 1358 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1359 eor w3, w3, w2 // lock_word.ThreadId() ^ self->ThreadId() 1360 uxth w3, w3 // zero top 16 bits 1361 cbnz w3, .Lslow_unlock // do lock word and self thread id's match? 1362 mov x3, x1 // copy lock word to detect transition to unlocked 1363 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1364 cmp w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE 1365 bpl .Lrecursive_thin_unlock 1366 // transition to unlocked 1367 mov x3, x1 1368 and w3, w3, #LOCK_WORD_GC_STATE_MASK_SHIFTED // w3: zero except for the preserved read barrier bits 1369 dmb ish // full (LoadStore|StoreStore) memory barrier 1370#ifndef USE_READ_BARRIER 1371 str w3, [x4] 1372#else 1373 stxr w2, w3, [x4] // Need to use atomic instructions for read barrier 1374 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1375#endif 1376 ret 1377.Lrecursive_thin_unlock: // w1: original lock word 1378 sub w1, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // decrement count 1379#ifndef USE_READ_BARRIER 1380 str w1, [x4] 1381#else 1382 stxr w2, w1, [x4] // Need to use atomic instructions for read barrier 1383 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1384#endif 1385 ret 1386.Lunlock_stxr_fail: 1387 b .Lretry_unlock // retry 1388.Lslow_unlock: 1389 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case exception allocation triggers GC 1390 mov x1, xSELF // pass Thread::Current 1391 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1392 RESTORE_SAVE_REFS_ONLY_FRAME 1393 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1394END art_quick_unlock_object 1395 1396ENTRY art_quick_unlock_object_no_inline 1397 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case exception allocation triggers GC 1398 mov x1, xSELF // pass Thread::Current 1399 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1400 RESTORE_SAVE_REFS_ONLY_FRAME 1401 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1402END art_quick_unlock_object_no_inline 1403 1404 /* 1405 * Entry from managed code that calls artIsAssignableFromCode and on failure calls 1406 * artThrowClassCastException. 1407 */ 1408 .extern artThrowClassCastException 1409ENTRY art_quick_check_cast 1410 // Store arguments and link register 1411 // Stack needs to be 16B aligned on calls. 1412 stp x0, x1, [sp,#-32]! 1413 .cfi_adjust_cfa_offset 32 1414 .cfi_rel_offset x0, 0 1415 .cfi_rel_offset x1, 8 1416 str xLR, [sp, #24] 1417 .cfi_rel_offset x30, 24 1418 1419 // Call runtime code 1420 bl artIsAssignableFromCode 1421 1422 // Check for exception 1423 cbz x0, .Lthrow_class_cast_exception 1424 1425 // Restore and return 1426 ldr xLR, [sp, #24] 1427 .cfi_restore x30 1428 ldp x0, x1, [sp], #32 1429 .cfi_restore x0 1430 .cfi_restore x1 1431 .cfi_adjust_cfa_offset -32 1432 ret 1433 1434 .cfi_adjust_cfa_offset 32 // Reset unwind info so following code unwinds. 1435 1436.Lthrow_class_cast_exception: 1437 // Restore 1438 ldr xLR, [sp, #24] 1439 .cfi_restore x30 1440 ldp x0, x1, [sp], #32 1441 .cfi_restore x0 1442 .cfi_restore x1 1443 .cfi_adjust_cfa_offset -32 1444 1445 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context 1446 mov x2, xSELF // pass Thread::Current 1447 b artThrowClassCastException // (Class*, Class*, Thread*) 1448 brk 0 // We should not return here... 1449END art_quick_check_cast 1450 1451// Restore xReg's value from [sp, #offset] if xReg is not the same as xExclude. 1452.macro POP_REG_NE xReg, offset, xExclude 1453 .ifnc \xReg, \xExclude 1454 ldr \xReg, [sp, #\offset] // restore xReg 1455 .cfi_restore \xReg 1456 .endif 1457.endm 1458 1459// Restore xReg1's value from [sp, #offset] if xReg1 is not the same as xExclude. 1460// Restore xReg2's value from [sp, #(offset + 8)] if xReg2 is not the same as xExclude. 1461.macro POP_REGS_NE xReg1, xReg2, offset, xExclude 1462 .ifc \xReg1, \xExclude 1463 ldr \xReg2, [sp, #(\offset + 8)] // restore xReg2 1464 .else 1465 .ifc \xReg2, \xExclude 1466 ldr \xReg1, [sp, #\offset] // restore xReg1 1467 .else 1468 ldp \xReg1, \xReg2, [sp, #\offset] // restore xReg1 and xReg2 1469 .endif 1470 .endif 1471 .cfi_restore \xReg1 1472 .cfi_restore \xReg2 1473.endm 1474 1475 /* 1476 * Macro to insert read barrier, only used in art_quick_aput_obj. 1477 * xDest, wDest and xObj are registers, offset is a defined literal such as 1478 * MIRROR_OBJECT_CLASS_OFFSET. Dest needs both x and w versions of the same register to handle 1479 * name mismatch between instructions. This macro uses the lower 32b of register when possible. 1480 * TODO: When read barrier has a fast path, add heap unpoisoning support for the fast path. 1481 */ 1482.macro READ_BARRIER xDest, wDest, xObj, xTemp, wTemp, offset, number 1483#ifdef USE_READ_BARRIER 1484#ifdef USE_BAKER_READ_BARRIER 1485 ldr \wTemp, [\xObj, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 1486 tbnz \wTemp, #LOCK_WORD_READ_BARRIER_STATE_SHIFT, .Lrb_slowpath\number 1487 // False dependency to avoid needing load/load fence. 1488 add \xObj, \xObj, \xTemp, lsr #32 1489 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1490 UNPOISON_HEAP_REF \wDest 1491 b .Lrb_exit\number 1492#endif 1493.Lrb_slowpath\number: 1494 // Store registers used in art_quick_aput_obj (x0-x4, LR), stack is 16B aligned. 1495 stp x0, x1, [sp, #-48]! 1496 .cfi_adjust_cfa_offset 48 1497 .cfi_rel_offset x0, 0 1498 .cfi_rel_offset x1, 8 1499 stp x2, x3, [sp, #16] 1500 .cfi_rel_offset x2, 16 1501 .cfi_rel_offset x3, 24 1502 stp x4, xLR, [sp, #32] 1503 .cfi_rel_offset x4, 32 1504 .cfi_rel_offset x30, 40 1505 1506 // mov x0, \xRef // pass ref in x0 (no-op for now since parameter ref is unused) 1507 .ifnc \xObj, x1 1508 mov x1, \xObj // pass xObj 1509 .endif 1510 mov w2, #\offset // pass offset 1511 bl artReadBarrierSlow // artReadBarrierSlow(ref, xObj, offset) 1512 // No need to unpoison return value in w0, artReadBarrierSlow() would do the unpoisoning. 1513 .ifnc \wDest, w0 1514 mov \wDest, w0 // save return value in wDest 1515 .endif 1516 1517 // Conditionally restore saved registers 1518 POP_REG_NE x0, 0, \xDest 1519 POP_REG_NE x1, 8, \xDest 1520 POP_REG_NE x2, 16, \xDest 1521 POP_REG_NE x3, 24, \xDest 1522 POP_REG_NE x4, 32, \xDest 1523 ldr xLR, [sp, #40] 1524 .cfi_restore x30 1525 add sp, sp, #48 1526 .cfi_adjust_cfa_offset -48 1527.Lrb_exit\number: 1528#else 1529 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1530 UNPOISON_HEAP_REF \wDest 1531#endif // USE_READ_BARRIER 1532.endm 1533 1534 /* 1535 * Entry from managed code for array put operations of objects where the value being stored 1536 * needs to be checked for compatibility. 1537 * x0 = array, x1 = index, x2 = value 1538 * 1539 * Currently all values should fit into w0/w1/w2, and w1 always will as indices are 32b. We 1540 * assume, though, that the upper 32b are zeroed out. At least for x1/w1 we can do better by 1541 * using index-zero-extension in load/stores. 1542 * 1543 * Temporaries: x3, x4 1544 * TODO: x4 OK? ip seems wrong here. 1545 */ 1546ENTRY art_quick_aput_obj_with_null_and_bound_check 1547 tst x0, x0 1548 bne art_quick_aput_obj_with_bound_check 1549 b art_quick_throw_null_pointer_exception 1550END art_quick_aput_obj_with_null_and_bound_check 1551 1552ENTRY art_quick_aput_obj_with_bound_check 1553 ldr w3, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] 1554 cmp w3, w1 1555 bhi art_quick_aput_obj 1556 mov x0, x1 1557 mov x1, x3 1558 b art_quick_throw_array_bounds 1559END art_quick_aput_obj_with_bound_check 1560 1561#ifdef USE_READ_BARRIER 1562 .extern artReadBarrierSlow 1563#endif 1564ENTRY art_quick_aput_obj 1565 cbz x2, .Ldo_aput_null 1566 READ_BARRIER x3, w3, x0, x3, w3, MIRROR_OBJECT_CLASS_OFFSET, 0 // Heap reference = 32b 1567 // This also zero-extends to x3 1568 READ_BARRIER x3, w3, x3, x4, w4, MIRROR_CLASS_COMPONENT_TYPE_OFFSET, 1 // Heap reference = 32b 1569 // This also zero-extends to x3 1570 READ_BARRIER x4, w4, x2, x4, w4, MIRROR_OBJECT_CLASS_OFFSET, 2 // Heap reference = 32b 1571 // This also zero-extends to x4 1572 cmp w3, w4 // value's type == array's component type - trivial assignability 1573 bne .Lcheck_assignability 1574.Ldo_aput: 1575 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1576 // "Compress" = do nothing 1577 POISON_HEAP_REF w2 1578 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1579 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1580 lsr x0, x0, #7 1581 strb w3, [x3, x0] 1582 ret 1583.Ldo_aput_null: 1584 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1585 // "Compress" = do nothing 1586 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1587 ret 1588.Lcheck_assignability: 1589 // Store arguments and link register 1590 stp x0, x1, [sp,#-32]! 1591 .cfi_adjust_cfa_offset 32 1592 .cfi_rel_offset x0, 0 1593 .cfi_rel_offset x1, 8 1594 stp x2, xLR, [sp, #16] 1595 .cfi_rel_offset x2, 16 1596 .cfi_rel_offset x30, 24 1597 1598 // Call runtime code 1599 mov x0, x3 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1600 mov x1, x4 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1601 bl artIsAssignableFromCode 1602 1603 // Check for exception 1604 cbz x0, .Lthrow_array_store_exception 1605 1606 // Restore 1607 ldp x2, x30, [sp, #16] 1608 .cfi_restore x2 1609 .cfi_restore x30 1610 ldp x0, x1, [sp], #32 1611 .cfi_restore x0 1612 .cfi_restore x1 1613 .cfi_adjust_cfa_offset -32 1614 1615 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1616 // "Compress" = do nothing 1617 POISON_HEAP_REF w2 1618 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1619 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1620 lsr x0, x0, #7 1621 strb w3, [x3, x0] 1622 ret 1623 .cfi_adjust_cfa_offset 32 // 4 restores after cbz for unwinding. 1624.Lthrow_array_store_exception: 1625 ldp x2, x30, [sp, #16] 1626 .cfi_restore x2 1627 .cfi_restore x30 1628 ldp x0, x1, [sp], #32 1629 .cfi_restore x0 1630 .cfi_restore x1 1631 .cfi_adjust_cfa_offset -32 1632 1633 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 1634 mov x1, x2 // Pass value. 1635 mov x2, xSELF // Pass Thread::Current. 1636 b artThrowArrayStoreException // (Object*, Object*, Thread*). 1637 brk 0 // Unreached. 1638END art_quick_aput_obj 1639 1640// Macro to facilitate adding new allocation entrypoints. 1641.macro ONE_ARG_DOWNCALL name, entrypoint, return 1642 .extern \entrypoint 1643ENTRY \name 1644 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1645 mov x1, xSELF // pass Thread::Current 1646 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1647 RESTORE_SAVE_REFS_ONLY_FRAME 1648 \return 1649END \name 1650.endm 1651 1652// Macro to facilitate adding new allocation entrypoints. 1653.macro TWO_ARG_DOWNCALL name, entrypoint, return 1654 .extern \entrypoint 1655ENTRY \name 1656 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1657 mov x2, xSELF // pass Thread::Current 1658 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1659 RESTORE_SAVE_REFS_ONLY_FRAME 1660 \return 1661END \name 1662.endm 1663 1664// Macro to facilitate adding new allocation entrypoints. 1665.macro THREE_ARG_DOWNCALL name, entrypoint, return 1666 .extern \entrypoint 1667ENTRY \name 1668 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1669 mov x3, xSELF // pass Thread::Current 1670 bl \entrypoint 1671 RESTORE_SAVE_REFS_ONLY_FRAME 1672 \return 1673END \name 1674.endm 1675 1676// Macro to facilitate adding new allocation entrypoints. 1677.macro FOUR_ARG_DOWNCALL name, entrypoint, return 1678 .extern \entrypoint 1679ENTRY \name 1680 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1681 mov x4, xSELF // pass Thread::Current 1682 bl \entrypoint // 1683 RESTORE_SAVE_REFS_ONLY_FRAME 1684 \return 1685 DELIVER_PENDING_EXCEPTION 1686END \name 1687.endm 1688 1689// Macros taking opportunity of code similarities for downcalls with referrer. 1690.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return 1691 .extern \entrypoint 1692ENTRY \name 1693 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1694 ldr x1, [sp, #FRAME_SIZE_SAVE_REFS_ONLY] // Load referrer 1695 mov x2, xSELF // pass Thread::Current 1696 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*, SP) 1697 RESTORE_SAVE_REFS_ONLY_FRAME 1698 \return 1699END \name 1700.endm 1701 1702.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return 1703 .extern \entrypoint 1704ENTRY \name 1705 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1706 ldr x2, [sp, #FRAME_SIZE_SAVE_REFS_ONLY] // Load referrer 1707 mov x3, xSELF // pass Thread::Current 1708 bl \entrypoint 1709 RESTORE_SAVE_REFS_ONLY_FRAME 1710 \return 1711END \name 1712.endm 1713 1714.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return 1715 .extern \entrypoint 1716ENTRY \name 1717 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1718 ldr x3, [sp, #FRAME_SIZE_SAVE_REFS_ONLY] // Load referrer 1719 mov x4, xSELF // pass Thread::Current 1720 bl \entrypoint 1721 RESTORE_SAVE_REFS_ONLY_FRAME 1722 \return 1723END \name 1724.endm 1725 1726.macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1727 cbz w0, 1f // result zero branch over 1728 ret // return 17291: 1730 DELIVER_PENDING_EXCEPTION 1731.endm 1732 1733 /* 1734 * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on 1735 * failure. 1736 */ 1737TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1738 1739 /* 1740 * Entry from managed code when uninitialized static storage, this stub will run the class 1741 * initializer and deliver the exception on error. On success the static storage base is 1742 * returned. 1743 */ 1744ONE_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1745 1746ONE_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1747ONE_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1748 1749ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1750ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1751ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1752ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1753ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1754ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1755ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1756 1757TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1758TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1759TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1760TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1761TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1762TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1763TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1764 1765TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1766TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1767TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1768TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1769 1770THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1771THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1772THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1773THREE_ARG_REF_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1774THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1775 1776// This is separated out as the argument order is different. 1777 .extern artSet64StaticFromCode 1778ENTRY art_quick_set64_static 1779 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1780 ldr x1, [sp, #FRAME_SIZE_SAVE_REFS_ONLY] // Load referrer 1781 // x2 contains the parameter 1782 mov x3, xSELF // pass Thread::Current 1783 bl artSet64StaticFromCode 1784 RESTORE_SAVE_REFS_ONLY_FRAME 1785 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1786END art_quick_set64_static 1787 1788 /* 1789 * Entry from managed code to resolve a string, this stub will 1790 * check the dex cache for a matching string (the fast path), and if not found, 1791 * it will allocate a String and deliver an exception on error. 1792 * On success the String is returned. R0 holds the string index. 1793 */ 1794 1795ENTRY art_quick_resolve_string 1796 ldr x1, [sp] // load referrer 1797 ldr w2, [x1, #ART_METHOD_DECLARING_CLASS_OFFSET] // load declaring class 1798 ldr x1, [x2, #DECLARING_CLASS_DEX_CACHE_STRINGS_OFFSET] // load string dex cache 1799 and x2, x0, #STRING_DEX_CACHE_SIZE_MINUS_ONE // get masked string index into x2 1800 ldr x2, [x1, x2, lsl #STRING_DEX_CACHE_ELEMENT_SIZE_SHIFT] // load dex cache pair into x2 1801 cmp x0, x2, lsr #32 // compare against upper 32 bits 1802 bne .Lart_quick_resolve_string_slow_path 1803 ubfx x0, x2, #0, #32 // extract lower 32 bits into x0 1804#ifdef USE_READ_BARRIER 1805 // Most common case: GC is not marking. 1806 ldr w3, [xSELF, #THREAD_IS_GC_MARKING_OFFSET] 1807 cbnz x3, .Lart_quick_resolve_string_marking 1808#endif 1809 ret 1810 1811// Slow path case, the index did not match. 1812.Lart_quick_resolve_string_slow_path: 1813 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1814 mov x1, xSELF // pass Thread::Current 1815 bl artResolveStringFromCode // (int32_t string_idx, Thread* self) 1816 RESTORE_SAVE_REFS_ONLY_FRAME 1817 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1818 1819// GC is marking case, need to check the mark bit. 1820.Lart_quick_resolve_string_marking: 1821 ldr x3, [x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 1822 tbnz x3, #LOCK_WORD_MARK_BIT_SHIFT, .Lart_quick_resolve_string_no_rb 1823 // Save LR so that we can return, also x1 for alignment purposes. 1824 stp x1, xLR, [sp, #-16]! // Save x1, LR. 1825 bl artReadBarrierMark // Get the marked string back. 1826 ldp x1, xLR, [sp], #16 // Restore registers. 1827.Lart_quick_resolve_string_no_rb: 1828 ret 1829 1830END art_quick_resolve_string 1831 1832// Generate the allocation entrypoints for each allocator. 1833GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_REGION_TLAB_ALLOCATORS 1834// Comment out allocators that have arm64 specific asm. 1835// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_region_tlab, RegionTLAB) implemented in asm 1836// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_tlab, RegionTLAB) 1837// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB) 1838GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1839// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY(_region_tlab, RegionTLAB) implemented in asm 1840// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB) 1841GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1842GENERATE_ALLOC_ENTRYPOINTS_CHECK_AND_ALLOC_ARRAY(_region_tlab, RegionTLAB) 1843GENERATE_ALLOC_ENTRYPOINTS_CHECK_AND_ALLOC_ARRAY_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1844GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB) 1845GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB) 1846GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB) 1847 1848// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_rosalloc, RosAlloc). 1849ENTRY art_quick_alloc_object_rosalloc 1850 // Fast path rosalloc allocation. 1851 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 1852 // x2-x7: free. 1853 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 1854 // Load the class (x2) 1855 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 1856 cbz x2, .Lart_quick_alloc_object_rosalloc_slow_path // Check null class 1857 // Check class status. 1858 ldr w3, [x2, #MIRROR_CLASS_STATUS_OFFSET] 1859 cmp x3, #MIRROR_CLASS_STATUS_INITIALIZED 1860 bne .Lart_quick_alloc_object_rosalloc_slow_path 1861 // Add a fake dependence from the 1862 // following access flag and size 1863 // loads to the status load. 1864 // This is to prevent those loads 1865 // from being reordered above the 1866 // status load and reading wrong 1867 // values (an alternative is to use 1868 // a load-acquire for the status). 1869 eor x3, x3, x3 1870 add x2, x2, x3 1871 // Check access flags has 1872 // kAccClassIsFinalizable 1873 ldr w3, [x2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] 1874 tst x3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE 1875 bne .Lart_quick_alloc_object_rosalloc_slow_path 1876 ldr x3, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] // Check if the thread local 1877 // allocation stack has room. 1878 // ldp won't work due to large offset. 1879 ldr x4, [xSELF, #THREAD_LOCAL_ALLOC_STACK_END_OFFSET] 1880 cmp x3, x4 1881 bhs .Lart_quick_alloc_object_rosalloc_slow_path 1882 ldr w3, [x2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (x3) 1883 cmp x3, #ROSALLOC_MAX_THREAD_LOCAL_BRACKET_SIZE // Check if the size is for a thread 1884 // local allocation 1885 bhs .Lart_quick_alloc_object_rosalloc_slow_path 1886 // Compute the rosalloc bracket index 1887 // from the size. 1888 // Align up the size by the rosalloc 1889 // bracket quantum size and divide 1890 // by the quantum size and subtract 1891 // by 1. This code is a shorter but 1892 // equivalent version. 1893 sub x3, x3, #1 1894 lsr x3, x3, #ROSALLOC_BRACKET_QUANTUM_SIZE_SHIFT 1895 // Load the rosalloc run (x4) 1896 add x4, xSELF, x3, lsl #POINTER_SIZE_SHIFT 1897 ldr x4, [x4, #THREAD_ROSALLOC_RUNS_OFFSET] 1898 // Load the free list head (x3). This 1899 // will be the return val. 1900 ldr x3, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1901 cbz x3, .Lart_quick_alloc_object_rosalloc_slow_path 1902 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1903 ldr x1, [x3, #ROSALLOC_SLOT_NEXT_OFFSET] // Load the next pointer of the head 1904 // and update the list head with the 1905 // next pointer. 1906 str x1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1907 // Store the class pointer in the 1908 // header. This also overwrites the 1909 // next pointer. The offsets are 1910 // asserted to match. 1911#if ROSALLOC_SLOT_NEXT_OFFSET != MIRROR_OBJECT_CLASS_OFFSET 1912#error "Class pointer needs to overwrite next pointer." 1913#endif 1914 POISON_HEAP_REF w2 1915 str w2, [x3, #MIRROR_OBJECT_CLASS_OFFSET] 1916 // Fence. This is "ish" not "ishst" so 1917 // that it also ensures ordering of 1918 // the class status load with respect 1919 // to later accesses to the class 1920 // object. Alternatively we could use 1921 // "ishst" if we use load-acquire for 1922 // the class status load.) 1923 // Needs to be done before pushing on 1924 // allocation since Heap::VisitObjects 1925 // relies on seeing the class pointer. 1926 // b/28790624 1927 dmb ish 1928 // Push the new object onto the thread 1929 // local allocation stack and 1930 // increment the thread local 1931 // allocation stack top. 1932 ldr x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1933 str w3, [x1], #COMPRESSED_REFERENCE_SIZE // (Increment x1 as a side effect.) 1934 str x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1935 // Decrement the size of the free list 1936 ldr w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1937 sub x1, x1, #1 1938 // TODO: consider combining this store 1939 // and the list head store above using 1940 // strd. 1941 str w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1942 1943 mov x0, x3 // Set the return value and return. 1944 ret 1945.Lart_quick_alloc_object_rosalloc_slow_path: 1946 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1947 mov x2, xSELF // pass Thread::Current 1948 bl artAllocObjectFromCodeRosAlloc // (uint32_t type_idx, Method* method, Thread*) 1949 RESTORE_SAVE_REFS_ONLY_FRAME 1950 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1951END art_quick_alloc_object_rosalloc 1952 1953 1954// The common fast path code for art_quick_alloc_array_region_tlab. 1955.macro ALLOC_ARRAY_TLAB_FAST_PATH slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1956 // Check null class 1957 cbz \wClass, \slowPathLabel 1958 ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED \slowPathLabel, \xClass, \wClass, \xCount, \wCount, \xTemp0, \wTemp0, \xTemp1, \wTemp1, \xTemp2, \wTemp2 1959.endm 1960 1961// The common fast path code for art_quick_alloc_array_region_tlab. 1962.macro ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1963 // Array classes are never finalizable or uninitialized, no need to check. 1964 ldr \wTemp0, [\xClass, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Load component type 1965 UNPOISON_HEAP_REF \wTemp0 1966 ldr \wTemp0, [\xTemp0, #MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET] 1967 lsr \xTemp0, \xTemp0, #PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT // Component size shift is in high 16 1968 // bits. 1969 // xCount is holding a 32 bit value, 1970 // it can not overflow. 1971 lsl \xTemp1, \xCount, \xTemp0 // Calculate data size 1972 // Add array data offset and alignment. 1973 add \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) 1974#if MIRROR_LONG_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4 1975#error Long array data offset must be 4 greater than int array data offset. 1976#endif 1977 1978 add \xTemp0, \xTemp0, #1 // Add 4 to the length only if the 1979 // component size shift is 3 1980 // (for 64 bit alignment). 1981 and \xTemp0, \xTemp0, #4 1982 add \xTemp1, \xTemp1, \xTemp0 1983 and \xTemp1, \xTemp1, #OBJECT_ALIGNMENT_MASK_TOGGLED64 // Apply alignemnt mask 1984 // (addr + 7) & ~7. The mask must 1985 // be 64 bits to keep high bits in 1986 // case of overflow. 1987 // Negative sized arrays are handled here since xCount holds a zero extended 32 bit value. 1988 // Negative ints become large 64 bit unsigned ints which will always be larger than max signed 1989 // 32 bit int. Since the max shift for arrays is 3, it can not become a negative 64 bit int. 1990 cmp \xTemp1, #MIN_LARGE_OBJECT_THRESHOLD // Possibly a large object, go slow 1991 bhs \slowPathLabel // path. 1992 1993 ldr \xTemp0, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Check tlab for space, note that 1994 // we use (end - begin) to handle 1995 // negative size arrays. It is 1996 // assumed that a negative size will 1997 // always be greater unsigned than 1998 // region size. 1999 ldr \xTemp2, [xSELF, #THREAD_LOCAL_END_OFFSET] 2000 sub \xTemp2, \xTemp2, \xTemp0 2001 cmp \xTemp1, \xTemp2 2002 bhi \slowPathLabel 2003 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 2004 // Move old thread_local_pos to x0 2005 // for the return value. 2006 mov x0, \xTemp0 2007 add \xTemp0, \xTemp0, \xTemp1 2008 str \xTemp0, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 2009 ldr \xTemp0, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 2010 add \xTemp0, \xTemp0, #1 2011 str \xTemp0, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 2012 POISON_HEAP_REF \wClass 2013 str \wClass, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 2014 str \wCount, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] // Store the array length. 2015 // Fence. 2016 dmb ishst 2017 ret 2018.endm 2019 2020// The common fast path code for art_quick_alloc_object_tlab and art_quick_alloc_object_region_tlab. 2021// 2022// x0: type_idx/return value, x1: ArtMethod*, x2: Class*, xSELF(x19): Thread::Current 2023// x3-x7: free. 2024// Need to preserve x0 and x1 to the slow path. 2025.macro ALLOC_OBJECT_TLAB_FAST_PATH slowPathLabel 2026 cbz x2, \slowPathLabel // Check null class 2027 ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED \slowPathLabel 2028.endm 2029 2030.macro ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED slowPathLabel 2031 ldr w3, [x2, #MIRROR_CLASS_STATUS_OFFSET] // Check class status. 2032 cmp x3, #MIRROR_CLASS_STATUS_INITIALIZED 2033 bne \slowPathLabel 2034 // Add a fake dependence from the 2035 // following access flag and size 2036 // loads to the status load. 2037 // This is to prevent those loads 2038 // from being reordered above the 2039 // status load and reading wrong 2040 // values (an alternative is to use 2041 // a load-acquire for the status). 2042 eor x3, x3, x3 2043 add x2, x2, x3 2044 ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED \slowPathLabel 2045.endm 2046 2047.macro ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED slowPathLabel 2048 // Check access flags has 2049 // kAccClassIsFinalizable. 2050 ldr w3, [x2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] 2051 tbnz x3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE_BIT, \slowPathLabel 2052 // Load thread_local_pos (x4) and 2053 // thread_local_end (x5). 2054 ldr x4, [xSELF, #THREAD_LOCAL_POS_OFFSET] 2055 ldr x5, [xSELF, #THREAD_LOCAL_END_OFFSET] 2056 sub x6, x5, x4 // Compute the remaining buf size. 2057 ldr w7, [x2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (x7). 2058 cmp x7, x6 // Check if it fits. OK to do this 2059 // before rounding up the object size 2060 // assuming the buf size alignment. 2061 bhi \slowPathLabel 2062 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 2063 // Round up the object size by the 2064 // object alignment. (addr + 7) & ~7. 2065 add x7, x7, #OBJECT_ALIGNMENT_MASK 2066 and x7, x7, #OBJECT_ALIGNMENT_MASK_TOGGLED 2067 // Move old thread_local_pos to x0 2068 // for the return value. 2069 mov x0, x4 2070 add x5, x0, x7 2071 str x5, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 2072 ldr x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 2073 add x5, x5, #1 2074 str x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 2075 POISON_HEAP_REF w2 2076 str w2, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 2077 // Fence. This is "ish" not "ishst" so 2078 // that the code after this allocation 2079 // site will see the right values in 2080 // the fields of the class. 2081 // Alternatively we could use "ishst" 2082 // if we use load-acquire for the 2083 // class status load.) 2084 dmb ish 2085 ret 2086.endm 2087 2088// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_tlab, TLAB). 2089ENTRY art_quick_alloc_object_tlab 2090 // Fast path tlab allocation. 2091 // x0: type_idx/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 2092 // x2-x7: free. 2093#if defined(USE_READ_BARRIER) 2094 mvn x0, xzr // Read barrier not supported here. 2095 ret // Return -1. 2096#endif 2097 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2098 // Load the class (x2) 2099 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2100 ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_tlab_slow_path 2101.Lart_quick_alloc_object_tlab_slow_path: 2102 SETUP_SAVE_REFS_ONLY_FRAME // Save callee saves in case of GC. 2103 mov x2, xSELF // Pass Thread::Current. 2104 bl artAllocObjectFromCodeTLAB // (uint32_t type_idx, Method* method, Thread*) 2105 RESTORE_SAVE_REFS_ONLY_FRAME 2106 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2107END art_quick_alloc_object_tlab 2108 2109// The common code for art_quick_alloc_object_*region_tlab 2110.macro GENERATE_ALLOC_OBJECT_REGION_TLAB name, entrypoint, fast_path, is_resolved 2111ENTRY \name 2112 // Fast path region tlab allocation. 2113 // x0: type_idx/resolved class/return value, x1: ArtMethod*, xSELF(x19): Thread::Current 2114 // If is_resolved is 1 then x0 is the resolved type, otherwise it is the index. 2115 // x2-x7: free. 2116#if !defined(USE_READ_BARRIER) 2117 mvn x0, xzr // Read barrier must be enabled here. 2118 ret // Return -1. 2119#endif 2120.if \is_resolved 2121 mov x2, x0 // class is actually stored in x0 already 2122.else 2123 ldr x2, [x1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2124 // Load the class (x2) 2125 ldr w2, [x2, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2126.endif 2127 // Most common case: GC is not marking. 2128 ldr w3, [xSELF, #THREAD_IS_GC_MARKING_OFFSET] 2129 cbnz x3, .Lmarking\name 2130.Ldo_allocation\name: 2131 \fast_path .Lslow_path\name 2132.Lmarking\name: 2133 // GC is marking, check the lock word of the class for the mark bit. 2134 // If the class is null, go slow path. The check is required to read the lock word. 2135 cbz w2, .Lslow_path\name 2136 // Class is not null, check mark bit in lock word. 2137 ldr w3, [x2, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2138 // If the bit is not zero, do the allocation. 2139 tbnz w3, #LOCK_WORD_MARK_BIT_SHIFT, .Ldo_allocation\name 2140 // The read barrier slow path. Mark 2141 // the class. 2142 stp x0, x1, [sp, #-32]! // Save registers (x0, x1, lr). 2143 str xLR, [sp, #16] // Align sp by 16 bytes. 2144 mov x0, x2 // Pass the class as the first param. 2145 bl artReadBarrierMark 2146 mov x2, x0 // Get the (marked) class back. 2147 ldp x0, x1, [sp, #0] // Restore registers. 2148 ldr xLR, [sp, #16] 2149 add sp, sp, #32 2150 b .Ldo_allocation\name 2151.Lslow_path\name: 2152 SETUP_SAVE_REFS_ONLY_FRAME // Save callee saves in case of GC. 2153 mov x2, xSELF // Pass Thread::Current. 2154 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 2155 RESTORE_SAVE_REFS_ONLY_FRAME 2156 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2157END \name 2158.endm 2159 2160GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_region_tlab, artAllocObjectFromCodeRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH, 0 2161GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_resolved_region_tlab, artAllocObjectFromCodeResolvedRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED, 1 2162GENERATE_ALLOC_OBJECT_REGION_TLAB art_quick_alloc_object_initialized_region_tlab, artAllocObjectFromCodeInitializedRegionTLAB, ALLOC_OBJECT_TLAB_FAST_PATH_INITIALIZED, 1 2163 2164// The common code for art_quick_alloc_array_*region_tlab 2165.macro GENERATE_ALLOC_ARRAY_REGION_TLAB name, entrypoint, fast_path, is_resolved 2166ENTRY \name 2167 // Fast path array allocation for region tlab allocation. 2168 // x0: uint32_t type_idx 2169 // x1: int32_t component_count 2170 // x2: ArtMethod* method 2171 // x3-x7: free. 2172#if !defined(USE_READ_BARRIER) 2173 mvn x0, xzr // Read barrier must be enabled here. 2174 ret // Return -1. 2175#endif 2176.if \is_resolved 2177 mov x3, x0 2178 // If already resolved, class is stored in x0 2179.else 2180 ldr x3, [x2, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_64] // Load dex cache resolved types array 2181 // Load the class (x2) 2182 ldr w3, [x3, x0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] 2183.endif 2184 // Most common case: GC is not marking. 2185 ldr w4, [xSELF, #THREAD_IS_GC_MARKING_OFFSET] 2186 cbnz x4, .Lmarking\name 2187.Ldo_allocation\name: 2188 \fast_path .Lslow_path\name, x3, w3, x1, w1, x4, w4, x5, w5, x6, w6 2189.Lmarking\name: 2190 // GC is marking, check the lock word of the class for the mark bit. 2191 // If the class is null, go slow path. The check is required to read the lock word. 2192 cbz w3, .Lslow_path\name 2193 // Class is not null, check mark bit in lock word. 2194 ldr w4, [x3, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2195 // If the bit is not zero, do the allocation. 2196 tbnz w4, #LOCK_WORD_MARK_BIT_SHIFT, .Ldo_allocation\name 2197 // The read barrier slow path. Mark 2198 // the class. 2199 stp x0, x1, [sp, #-32]! // Save registers (x0, x1, x2, lr). 2200 stp x2, xLR, [sp, #16] 2201 mov x0, x3 // Pass the class as the first param. 2202 bl artReadBarrierMark 2203 mov x3, x0 // Get the (marked) class back. 2204 ldp x2, xLR, [sp, #16] 2205 ldp x0, x1, [sp], #32 // Restore registers. 2206 b .Ldo_allocation\name 2207.Lslow_path\name: 2208 // x0: uint32_t type_idx / mirror::Class* klass (if resolved) 2209 // x1: int32_t component_count 2210 // x2: ArtMethod* method 2211 // x3: Thread* self 2212 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 2213 mov x3, xSELF // pass Thread::Current 2214 bl \entrypoint 2215 RESTORE_SAVE_REFS_ONLY_FRAME 2216 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 2217END \name 2218.endm 2219 2220GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_region_tlab, artAllocArrayFromCodeRegionTLAB, ALLOC_ARRAY_TLAB_FAST_PATH, 0 2221// TODO: art_quick_alloc_array_resolved_region_tlab seems to not get called. Investigate compiler. 2222GENERATE_ALLOC_ARRAY_REGION_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED, 1 2223 2224 /* 2225 * Called by managed code when the thread has been asked to suspend. 2226 */ 2227 .extern artTestSuspendFromCode 2228ENTRY art_quick_test_suspend 2229 SETUP_SAVE_EVERYTHING_FRAME // save callee saves for stack crawl 2230 mov x0, xSELF 2231 bl artTestSuspendFromCode // (Thread*) 2232 RESTORE_SAVE_EVERYTHING_FRAME 2233 ret 2234END art_quick_test_suspend 2235 2236ENTRY art_quick_implicit_suspend 2237 mov x0, xSELF 2238 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves for stack crawl 2239 bl artTestSuspendFromCode // (Thread*) 2240 RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN 2241END art_quick_implicit_suspend 2242 2243 /* 2244 * Called by managed code that is attempting to call a method on a proxy class. On entry 2245 * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy 2246 * method agrees with a ref and args callee save frame. 2247 */ 2248 .extern artQuickProxyInvokeHandler 2249ENTRY art_quick_proxy_invoke_handler 2250 SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 2251 mov x2, xSELF // pass Thread::Current 2252 mov x3, sp // pass SP 2253 bl artQuickProxyInvokeHandler // (Method* proxy method, receiver, Thread*, SP) 2254 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 2255 cbnz x2, .Lexception_in_proxy // success if no exception is pending 2256 RESTORE_SAVE_REFS_AND_ARGS_FRAME // Restore frame 2257 fmov d0, x0 // Store result in d0 in case it was float or double 2258 ret // return on success 2259.Lexception_in_proxy: 2260 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2261 DELIVER_PENDING_EXCEPTION 2262END art_quick_proxy_invoke_handler 2263 2264 /* 2265 * Called to resolve an imt conflict. 2266 * x0 is the conflict ArtMethod. 2267 * xIP1 is a hidden argument that holds the target interface method's dex method index. 2268 * 2269 * Note that this stub writes to xIP0, xIP1, and x0. 2270 */ 2271 .extern artInvokeInterfaceTrampoline 2272ENTRY art_quick_imt_conflict_trampoline 2273 ldr xIP0, [sp, #0] // Load referrer 2274 ldr xIP0, [xIP0, #ART_METHOD_DEX_CACHE_METHODS_OFFSET_64] // Load dex cache methods array 2275 ldr xIP0, [xIP0, xIP1, lsl #POINTER_SIZE_SHIFT] // Load interface method 2276 ldr xIP1, [x0, #ART_METHOD_JNI_OFFSET_64] // Load ImtConflictTable 2277 ldr x0, [xIP1] // Load first entry in ImtConflictTable. 2278.Limt_table_iterate: 2279 cmp x0, xIP0 2280 // Branch if found. Benchmarks have shown doing a branch here is better. 2281 beq .Limt_table_found 2282 // If the entry is null, the interface method is not in the ImtConflictTable. 2283 cbz x0, .Lconflict_trampoline 2284 // Iterate over the entries of the ImtConflictTable. 2285 ldr x0, [xIP1, #(2 * __SIZEOF_POINTER__)]! 2286 b .Limt_table_iterate 2287.Limt_table_found: 2288 // We successfully hit an entry in the table. Load the target method 2289 // and jump to it. 2290 ldr x0, [xIP1, #__SIZEOF_POINTER__] 2291 ldr xIP0, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 2292 br xIP0 2293.Lconflict_trampoline: 2294 // Call the runtime stub to populate the ImtConflictTable and jump to the 2295 // resolved method. 2296 INVOKE_TRAMPOLINE_BODY artInvokeInterfaceTrampoline 2297END art_quick_imt_conflict_trampoline 2298 2299ENTRY art_quick_resolution_trampoline 2300 SETUP_SAVE_REFS_AND_ARGS_FRAME 2301 mov x2, xSELF 2302 mov x3, sp 2303 bl artQuickResolutionTrampoline // (called, receiver, Thread*, SP) 2304 cbz x0, 1f 2305 mov xIP0, x0 // Remember returned code pointer in xIP0. 2306 ldr x0, [sp, #0] // artQuickResolutionTrampoline puts called method in *SP. 2307 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2308 br xIP0 23091: 2310 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2311 DELIVER_PENDING_EXCEPTION 2312END art_quick_resolution_trampoline 2313 2314/* 2315 * Generic JNI frame layout: 2316 * 2317 * #-------------------# 2318 * | | 2319 * | caller method... | 2320 * #-------------------# <--- SP on entry 2321 * | Return X30/LR | 2322 * | X29/FP | callee save 2323 * | X28 | callee save 2324 * | X27 | callee save 2325 * | X26 | callee save 2326 * | X25 | callee save 2327 * | X24 | callee save 2328 * | X23 | callee save 2329 * | X22 | callee save 2330 * | X21 | callee save 2331 * | X20 | callee save 2332 * | X19 | callee save 2333 * | X7 | arg7 2334 * | X6 | arg6 2335 * | X5 | arg5 2336 * | X4 | arg4 2337 * | X3 | arg3 2338 * | X2 | arg2 2339 * | X1 | arg1 2340 * | D7 | float arg 8 2341 * | D6 | float arg 7 2342 * | D5 | float arg 6 2343 * | D4 | float arg 5 2344 * | D3 | float arg 4 2345 * | D2 | float arg 3 2346 * | D1 | float arg 2 2347 * | D0 | float arg 1 2348 * | Method* | <- X0 2349 * #-------------------# 2350 * | local ref cookie | // 4B 2351 * | handle scope size | // 4B 2352 * #-------------------# 2353 * | JNI Call Stack | 2354 * #-------------------# <--- SP on native call 2355 * | | 2356 * | Stack for Regs | The trampoline assembly will pop these values 2357 * | | into registers for native call 2358 * #-------------------# 2359 * | Native code ptr | 2360 * #-------------------# 2361 * | Free scratch | 2362 * #-------------------# 2363 * | Ptr to (1) | <--- SP 2364 * #-------------------# 2365 */ 2366 /* 2367 * Called to do a generic JNI down-call 2368 */ 2369ENTRY art_quick_generic_jni_trampoline 2370 SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 2371 2372 // Save SP , so we can have static CFI info. 2373 mov x28, sp 2374 .cfi_def_cfa_register x28 2375 2376 // This looks the same, but is different: this will be updated to point to the bottom 2377 // of the frame when the handle scope is inserted. 2378 mov xFP, sp 2379 2380 mov xIP0, #5120 2381 sub sp, sp, xIP0 2382 2383 // prepare for artQuickGenericJniTrampoline call 2384 // (Thread*, SP) 2385 // x0 x1 <= C calling convention 2386 // xSELF xFP <= where they are 2387 2388 mov x0, xSELF // Thread* 2389 mov x1, xFP 2390 bl artQuickGenericJniTrampoline // (Thread*, sp) 2391 2392 // The C call will have registered the complete save-frame on success. 2393 // The result of the call is: 2394 // x0: pointer to native code, 0 on error. 2395 // x1: pointer to the bottom of the used area of the alloca, can restore stack till there. 2396 2397 // Check for error = 0. 2398 cbz x0, .Lexception_in_native 2399 2400 // Release part of the alloca. 2401 mov sp, x1 2402 2403 // Save the code pointer 2404 mov xIP0, x0 2405 2406 // Load parameters from frame into registers. 2407 // TODO Check with artQuickGenericJniTrampoline. 2408 // Also, check again APPCS64 - the stack arguments are interleaved. 2409 ldp x0, x1, [sp] 2410 ldp x2, x3, [sp, #16] 2411 ldp x4, x5, [sp, #32] 2412 ldp x6, x7, [sp, #48] 2413 2414 ldp d0, d1, [sp, #64] 2415 ldp d2, d3, [sp, #80] 2416 ldp d4, d5, [sp, #96] 2417 ldp d6, d7, [sp, #112] 2418 2419 add sp, sp, #128 2420 2421 blr xIP0 // native call. 2422 2423 // result sign extension is handled in C code 2424 // prepare for artQuickGenericJniEndTrampoline call 2425 // (Thread*, result, result_f) 2426 // x0 x1 x2 <= C calling convention 2427 mov x1, x0 // Result (from saved). 2428 mov x0, xSELF // Thread register. 2429 fmov x2, d0 // d0 will contain floating point result, but needs to go into x2 2430 2431 bl artQuickGenericJniEndTrampoline 2432 2433 // Pending exceptions possible. 2434 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 2435 cbnz x2, .Lexception_in_native 2436 2437 // Tear down the alloca. 2438 mov sp, x28 2439 .cfi_def_cfa_register sp 2440 2441 // Tear down the callee-save frame. 2442 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2443 2444 // store into fpr, for when it's a fpr return... 2445 fmov d0, x0 2446 ret 2447 2448.Lexception_in_native: 2449 // Move to x1 then sp to please assembler. 2450 ldr x1, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 2451 mov sp, x1 2452 .cfi_def_cfa_register sp 2453 # This will create a new save-all frame, required by the runtime. 2454 DELIVER_PENDING_EXCEPTION 2455END art_quick_generic_jni_trampoline 2456 2457/* 2458 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those 2459 * of a quick call: 2460 * x0 = method being called/to bridge to. 2461 * x1..x7, d0..d7 = arguments to that method. 2462 */ 2463ENTRY art_quick_to_interpreter_bridge 2464 SETUP_SAVE_REFS_AND_ARGS_FRAME // Set up frame and save arguments. 2465 2466 // x0 will contain mirror::ArtMethod* method. 2467 mov x1, xSELF // How to get Thread::Current() ??? 2468 mov x2, sp 2469 2470 // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self, 2471 // mirror::ArtMethod** sp) 2472 bl artQuickToInterpreterBridge 2473 2474 RESTORE_SAVE_REFS_AND_ARGS_FRAME // TODO: no need to restore arguments in this case. 2475 2476 fmov d0, x0 2477 2478 RETURN_OR_DELIVER_PENDING_EXCEPTION 2479END art_quick_to_interpreter_bridge 2480 2481 2482// 2483// Instrumentation-related stubs 2484// 2485 .extern artInstrumentationMethodEntryFromCode 2486ENTRY art_quick_instrumentation_entry 2487 SETUP_SAVE_REFS_AND_ARGS_FRAME 2488 2489 mov x20, x0 // Preserve method reference in a callee-save. 2490 2491 mov x2, xSELF 2492 mov x3, xLR 2493 bl artInstrumentationMethodEntryFromCode // (Method*, Object*, Thread*, LR) 2494 2495 mov xIP0, x0 // x0 = result of call. 2496 mov x0, x20 // Reload method reference. 2497 2498 RESTORE_SAVE_REFS_AND_ARGS_FRAME // Note: will restore xSELF 2499 adr xLR, art_quick_instrumentation_exit 2500 br xIP0 // Tail-call method with lr set to art_quick_instrumentation_exit. 2501END art_quick_instrumentation_entry 2502 2503 .extern artInstrumentationMethodExitFromCode 2504ENTRY art_quick_instrumentation_exit 2505 mov xLR, #0 // Clobber LR for later checks. 2506 2507 SETUP_SAVE_REFS_ONLY_FRAME 2508 2509 // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then 2510 // we would need to fully restore it. As there are a lot of callee-save registers, it seems 2511 // easier to have an extra small stack area. 2512 2513 str x0, [sp, #-16]! // Save integer result. 2514 .cfi_adjust_cfa_offset 16 2515 str d0, [sp, #8] // Save floating-point result. 2516 2517 add x1, sp, #16 // Pass SP. 2518 mov x2, x0 // Pass integer result. 2519 fmov x3, d0 // Pass floating-point result. 2520 mov x0, xSELF // Pass Thread. 2521 bl artInstrumentationMethodExitFromCode // (Thread*, SP, gpr_res, fpr_res) 2522 2523 mov xIP0, x0 // Return address from instrumentation call. 2524 mov xLR, x1 // r1 is holding link register if we're to bounce to deoptimize 2525 2526 ldr d0, [sp, #8] // Restore floating-point result. 2527 ldr x0, [sp], 16 // Restore integer result, and drop stack area. 2528 .cfi_adjust_cfa_offset 16 2529 2530 POP_SAVE_REFS_ONLY_FRAME 2531 2532 br xIP0 // Tail-call out. 2533END art_quick_instrumentation_exit 2534 2535 /* 2536 * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization 2537 * will long jump to the upcall with a special exception of -1. 2538 */ 2539 .extern artDeoptimize 2540ENTRY art_quick_deoptimize 2541 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 2542 mov x0, xSELF // Pass thread. 2543 bl artDeoptimize // artDeoptimize(Thread*) 2544 brk 0 2545END art_quick_deoptimize 2546 2547 /* 2548 * Compiled code has requested that we deoptimize into the interpreter. The deoptimization 2549 * will long jump to the upcall with a special exception of -1. 2550 */ 2551 .extern artDeoptimizeFromCompiledCode 2552ENTRY art_quick_deoptimize_from_compiled_code 2553 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 2554 mov x0, xSELF // Pass thread. 2555 bl artDeoptimizeFromCompiledCode // artDeoptimizeFromCompiledCode(Thread*) 2556 brk 0 2557END art_quick_deoptimize_from_compiled_code 2558 2559 2560 /* 2561 * String's indexOf. 2562 * 2563 * TODO: Not very optimized. 2564 * On entry: 2565 * x0: string object (known non-null) 2566 * w1: char to match (known <= 0xFFFF) 2567 * w2: Starting offset in string data 2568 */ 2569ENTRY art_quick_indexof 2570 ldr w3, [x0, #MIRROR_STRING_COUNT_OFFSET] 2571 add x0, x0, #MIRROR_STRING_VALUE_OFFSET 2572 2573 /* Clamp start to [0..count] */ 2574 cmp w2, #0 2575 csel w2, wzr, w2, lt 2576 cmp w2, w3 2577 csel w2, w3, w2, gt 2578 2579 /* Save a copy to compute result */ 2580 mov x5, x0 2581 2582 /* Build pointer to start of data to compare and pre-bias */ 2583 add x0, x0, x2, lsl #1 2584 sub x0, x0, #2 2585 2586 /* Compute iteration count */ 2587 sub w2, w3, w2 2588 2589 /* 2590 * At this point we have: 2591 * x0: start of the data to test 2592 * w1: char to compare 2593 * w2: iteration count 2594 * x5: original start of string data 2595 */ 2596 2597 subs w2, w2, #4 2598 b.lt .Lindexof_remainder 2599 2600.Lindexof_loop4: 2601 ldrh w6, [x0, #2]! 2602 ldrh w7, [x0, #2]! 2603 ldrh wIP0, [x0, #2]! 2604 ldrh wIP1, [x0, #2]! 2605 cmp w6, w1 2606 b.eq .Lmatch_0 2607 cmp w7, w1 2608 b.eq .Lmatch_1 2609 cmp wIP0, w1 2610 b.eq .Lmatch_2 2611 cmp wIP1, w1 2612 b.eq .Lmatch_3 2613 subs w2, w2, #4 2614 b.ge .Lindexof_loop4 2615 2616.Lindexof_remainder: 2617 adds w2, w2, #4 2618 b.eq .Lindexof_nomatch 2619 2620.Lindexof_loop1: 2621 ldrh w6, [x0, #2]! 2622 cmp w6, w1 2623 b.eq .Lmatch_3 2624 subs w2, w2, #1 2625 b.ne .Lindexof_loop1 2626 2627.Lindexof_nomatch: 2628 mov x0, #-1 2629 ret 2630 2631.Lmatch_0: 2632 sub x0, x0, #6 2633 sub x0, x0, x5 2634 asr x0, x0, #1 2635 ret 2636.Lmatch_1: 2637 sub x0, x0, #4 2638 sub x0, x0, x5 2639 asr x0, x0, #1 2640 ret 2641.Lmatch_2: 2642 sub x0, x0, #2 2643 sub x0, x0, x5 2644 asr x0, x0, #1 2645 ret 2646.Lmatch_3: 2647 sub x0, x0, x5 2648 asr x0, x0, #1 2649 ret 2650END art_quick_indexof 2651 2652 /* 2653 * Create a function `name` calling the ReadBarrier::Mark routine, 2654 * getting its argument and returning its result through W register 2655 * `wreg` (corresponding to X register `xreg`), saving and restoring 2656 * all caller-save registers. 2657 * 2658 * If `wreg` is different from `w0`, the generated function follows a 2659 * non-standard runtime calling convention: 2660 * - register `wreg` is used to pass the (sole) argument of this 2661 * function (instead of W0); 2662 * - register `wreg` is used to return the result of this function 2663 * (instead of W0); 2664 * - W0 is treated like a normal (non-argument) caller-save register; 2665 * - everything else is the same as in the standard runtime calling 2666 * convention (e.g. standard callee-save registers are preserved). 2667 */ 2668.macro READ_BARRIER_MARK_REG name, wreg, xreg 2669ENTRY \name 2670 // Reference is null, no work to do at all. 2671 cbz \wreg, .Lret_rb_\name 2672 /* 2673 * Allocate 46 stack slots * 8 = 368 bytes: 2674 * - 20 slots for core registers X0-X19 2675 * - 24 slots for floating-point registers D0-D7 and D16-D31 2676 * - 1 slot for return address register XLR 2677 * - 1 padding slot for 16-byte stack alignment 2678 */ 2679 // Use wIP0 as temp and check the mark bit of the reference. wIP0 is not used by the compiler. 2680 ldr wIP0, [\xreg, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2681 tbz wIP0, #LOCK_WORD_MARK_BIT_SHIFT, .Lslow_path_rb_\name 2682 ret 2683.Lslow_path_rb_\name: 2684 // Save all potentially live caller-save core registers. 2685 stp x0, x1, [sp, #-368]! 2686 .cfi_adjust_cfa_offset 368 2687 .cfi_rel_offset x0, 0 2688 .cfi_rel_offset x1, 8 2689 stp x2, x3, [sp, #16] 2690 .cfi_rel_offset x2, 16 2691 .cfi_rel_offset x3, 24 2692 stp x4, x5, [sp, #32] 2693 .cfi_rel_offset x4, 32 2694 .cfi_rel_offset x5, 40 2695 stp x6, x7, [sp, #48] 2696 .cfi_rel_offset x6, 48 2697 .cfi_rel_offset x7, 56 2698 stp x8, x9, [sp, #64] 2699 .cfi_rel_offset x8, 64 2700 .cfi_rel_offset x9, 72 2701 stp x10, x11, [sp, #80] 2702 .cfi_rel_offset x10, 80 2703 .cfi_rel_offset x11, 88 2704 stp x12, x13, [sp, #96] 2705 .cfi_rel_offset x12, 96 2706 .cfi_rel_offset x13, 104 2707 stp x14, x15, [sp, #112] 2708 .cfi_rel_offset x14, 112 2709 .cfi_rel_offset x15, 120 2710 stp x16, x17, [sp, #128] 2711 .cfi_rel_offset x16, 128 2712 .cfi_rel_offset x17, 136 2713 stp x18, x19, [sp, #144] 2714 .cfi_rel_offset x18, 144 2715 .cfi_rel_offset x19, 152 2716 // Save all potentially live caller-save floating-point registers. 2717 stp d0, d1, [sp, #160] 2718 stp d2, d3, [sp, #176] 2719 stp d4, d5, [sp, #192] 2720 stp d6, d7, [sp, #208] 2721 stp d16, d17, [sp, #224] 2722 stp d18, d19, [sp, #240] 2723 stp d20, d21, [sp, #256] 2724 stp d22, d23, [sp, #272] 2725 stp d24, d25, [sp, #288] 2726 stp d26, d27, [sp, #304] 2727 stp d28, d29, [sp, #320] 2728 stp d30, d31, [sp, #336] 2729 // Save return address. 2730 str xLR, [sp, #352] 2731 .cfi_rel_offset x30, 352 2732 // (sp + #360 is a padding slot) 2733 2734 .ifnc \wreg, w0 2735 mov w0, \wreg // Pass arg1 - obj from `wreg` 2736 .endif 2737 bl artReadBarrierMark // artReadBarrierMark(obj) 2738 .ifnc \wreg, w0 2739 mov \wreg, w0 // Return result into `wreg` 2740 .endif 2741 2742 // Restore core regs, except `xreg`, as `wreg` is used to return the 2743 // result of this function (simply remove it from the stack instead). 2744 POP_REGS_NE x0, x1, 0, \xreg 2745 POP_REGS_NE x2, x3, 16, \xreg 2746 POP_REGS_NE x4, x5, 32, \xreg 2747 POP_REGS_NE x6, x7, 48, \xreg 2748 POP_REGS_NE x8, x9, 64, \xreg 2749 POP_REGS_NE x10, x11, 80, \xreg 2750 POP_REGS_NE x12, x13, 96, \xreg 2751 POP_REGS_NE x14, x15, 112, \xreg 2752 POP_REGS_NE x16, x17, 128, \xreg 2753 POP_REGS_NE x18, x19, 144, \xreg 2754 // Restore floating-point registers. 2755 ldp d0, d1, [sp, #160] 2756 ldp d2, d3, [sp, #176] 2757 ldp d4, d5, [sp, #192] 2758 ldp d6, d7, [sp, #208] 2759 ldp d16, d17, [sp, #224] 2760 ldp d18, d19, [sp, #240] 2761 ldp d20, d21, [sp, #256] 2762 ldp d22, d23, [sp, #272] 2763 ldp d24, d25, [sp, #288] 2764 ldp d26, d27, [sp, #304] 2765 ldp d28, d29, [sp, #320] 2766 ldp d30, d31, [sp, #336] 2767 // Restore return address and remove padding. 2768 ldr xLR, [sp, #352] 2769 .cfi_restore x30 2770 add sp, sp, #368 2771 .cfi_adjust_cfa_offset -368 2772.Lret_rb_\name: 2773 ret 2774END \name 2775.endm 2776 2777READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg00, w0, x0 2778READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg01, w1, x1 2779READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg02, w2, x2 2780READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg03, w3, x3 2781READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg04, w4, x4 2782READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg05, w5, x5 2783READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg06, w6, x6 2784READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg07, w7, x7 2785READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg08, w8, x8 2786READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg09, w9, x9 2787READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg10, w10, x10 2788READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg11, w11, x11 2789READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg12, w12, x12 2790READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg13, w13, x13 2791READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg14, w14, x14 2792READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg15, w15, x15 2793// READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg16, w16, x16 ip0 is blocked 2794READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg17, w17, x17 2795READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg18, w18, x18 2796READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg19, w19, x19 2797READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg20, w20, x20 2798READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg21, w21, x21 2799READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg22, w22, x22 2800READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg23, w23, x23 2801READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg24, w24, x24 2802READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg25, w25, x25 2803READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg26, w26, x26 2804READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg27, w27, x27 2805READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg28, w28, x28 2806READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg29, w29, x29 2807