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.macro INCREASE_FRAME frame_adjustment 23 sub sp, sp, #(\frame_adjustment) 24 .cfi_adjust_cfa_offset (\frame_adjustment) 25.endm 26 27.macro DECREASE_FRAME frame_adjustment 28 add sp, sp, #(\frame_adjustment) 29 .cfi_adjust_cfa_offset -(\frame_adjustment) 30.endm 31 32.macro SAVE_REG reg, offset 33 str \reg, [sp, #(\offset)] 34 .cfi_rel_offset \reg, (\offset) 35.endm 36 37.macro RESTORE_REG reg, offset 38 ldr \reg, [sp, #(\offset)] 39 .cfi_restore \reg 40.endm 41 42.macro SAVE_TWO_REGS reg1, reg2, offset 43 stp \reg1, \reg2, [sp, #(\offset)] 44 .cfi_rel_offset \reg1, (\offset) 45 .cfi_rel_offset \reg2, (\offset) + 8 46.endm 47 48.macro RESTORE_TWO_REGS reg1, reg2, offset 49 ldp \reg1, \reg2, [sp, #(\offset)] 50 .cfi_restore \reg1 51 .cfi_restore \reg2 52.endm 53 54.macro SAVE_TWO_REGS_INCREASE_FRAME reg1, reg2, frame_adjustment 55 stp \reg1, \reg2, [sp, #-(\frame_adjustment)]! 56 .cfi_adjust_cfa_offset (\frame_adjustment) 57 .cfi_rel_offset \reg1, 0 58 .cfi_rel_offset \reg2, 8 59.endm 60 61.macro RESTORE_TWO_REGS_DECREASE_FRAME reg1, reg2, frame_adjustment 62 ldp \reg1, \reg2, [sp], #(\frame_adjustment) 63 .cfi_restore \reg1 64 .cfi_restore \reg2 65 .cfi_adjust_cfa_offset -(\frame_adjustment) 66.endm 67 68 /* 69 * Macro that sets up the callee save frame to conform with 70 * Runtime::CreateCalleeSaveMethod(kSaveAllCalleeSaves) 71 */ 72.macro SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 73 // art::Runtime** xIP0 = &art::Runtime::instance_ 74 adrp xIP0, :got:_ZN3art7Runtime9instance_E 75 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 76 77 // Our registers aren't intermixed - just spill in order. 78 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 79 80 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveAllCalleeSaves]; 81 ldr xIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVES_METHOD_OFFSET] 82 83 INCREASE_FRAME 176 84 85 // Ugly compile-time check, but we only have the preprocessor. 86#if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVES != 176) 87#error "FRAME_SIZE_SAVE_ALL_CALLEE_SAVES(ARM64) size not as expected." 88#endif 89 90 // Stack alignment filler [sp, #8]. 91 // FP callee-saves. 92 stp d8, d9, [sp, #16] 93 stp d10, d11, [sp, #32] 94 stp d12, d13, [sp, #48] 95 stp d14, d15, [sp, #64] 96 97 // GP callee-saves 98 SAVE_TWO_REGS x19, x20, 80 99 SAVE_TWO_REGS x21, x22, 96 100 SAVE_TWO_REGS x23, x24, 112 101 SAVE_TWO_REGS x25, x26, 128 102 SAVE_TWO_REGS x27, x28, 144 103 SAVE_TWO_REGS x29, xLR, 160 104 105 // Store ArtMethod* Runtime::callee_save_methods_[kSaveAllCalleeSaves]. 106 str xIP0, [sp] 107 // Place sp in Thread::Current()->top_quick_frame. 108 mov xIP0, sp 109 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 110.endm 111 112 /* 113 * Macro that sets up the callee save frame to conform with 114 * Runtime::CreateCalleeSaveMethod(kSaveRefsOnly). 115 */ 116.macro SETUP_SAVE_REFS_ONLY_FRAME 117 // art::Runtime** xIP0 = &art::Runtime::instance_ 118 adrp xIP0, :got:_ZN3art7Runtime9instance_E 119 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 120 121 // Our registers aren't intermixed - just spill in order. 122 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 123 124 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveRefOnly]; 125 ldr xIP0, [xIP0, RUNTIME_SAVE_REFS_ONLY_METHOD_OFFSET] 126 127 INCREASE_FRAME 96 128 129 // Ugly compile-time check, but we only have the preprocessor. 130#if (FRAME_SIZE_SAVE_REFS_ONLY != 96) 131#error "FRAME_SIZE_SAVE_REFS_ONLY(ARM64) size not as expected." 132#endif 133 134 // GP callee-saves. 135 // x20 paired with ArtMethod* - see below. 136 SAVE_TWO_REGS x21, x22, 16 137 SAVE_TWO_REGS x23, x24, 32 138 SAVE_TWO_REGS x25, x26, 48 139 SAVE_TWO_REGS x27, x28, 64 140 SAVE_TWO_REGS x29, xLR, 80 141 142 // Store ArtMethod* Runtime::callee_save_methods_[kSaveRefsOnly]. 143 stp xIP0, x20, [sp] 144 .cfi_rel_offset x20, 8 145 146 // Place sp in Thread::Current()->top_quick_frame. 147 mov xIP0, sp 148 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 149.endm 150 151// TODO: Probably no need to restore registers preserved by aapcs64. 152.macro RESTORE_SAVE_REFS_ONLY_FRAME 153 // Callee-saves. 154 RESTORE_REG x20, 8 155 RESTORE_TWO_REGS x21, x22, 16 156 RESTORE_TWO_REGS x23, x24, 32 157 RESTORE_TWO_REGS x25, x26, 48 158 RESTORE_TWO_REGS x27, x28, 64 159 RESTORE_TWO_REGS x29, xLR, 80 160 161 DECREASE_FRAME 96 162.endm 163 164.macro POP_SAVE_REFS_ONLY_FRAME 165 DECREASE_FRAME 96 166.endm 167 168.macro RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN 169 RESTORE_SAVE_REFS_ONLY_FRAME 170 ret 171.endm 172 173 174.macro SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 175 INCREASE_FRAME 224 176 177 // Ugly compile-time check, but we only have the preprocessor. 178#if (FRAME_SIZE_SAVE_REFS_AND_ARGS != 224) 179#error "FRAME_SIZE_SAVE_REFS_AND_ARGS(ARM64) size not as expected." 180#endif 181 182 // Stack alignment filler [sp, #8]. 183 // FP args. 184 stp d0, d1, [sp, #16] 185 stp d2, d3, [sp, #32] 186 stp d4, d5, [sp, #48] 187 stp d6, d7, [sp, #64] 188 189 // Core args. 190 SAVE_TWO_REGS x1, x2, 80 191 SAVE_TWO_REGS x3, x4, 96 192 SAVE_TWO_REGS x5, x6, 112 193 194 // x7, Callee-saves. 195 SAVE_TWO_REGS x7, x20, 128 196 SAVE_TWO_REGS x21, x22, 144 197 SAVE_TWO_REGS x23, x24, 160 198 SAVE_TWO_REGS x25, x26, 176 199 SAVE_TWO_REGS x27, x28, 192 200 201 // x29(callee-save) and LR. 202 SAVE_TWO_REGS x29, xLR, 208 203 204.endm 205 206 /* 207 * Macro that sets up the callee save frame to conform with 208 * Runtime::CreateCalleeSaveMethod(kSaveRefsAndArgs). 209 * 210 * TODO This is probably too conservative - saving FP & LR. 211 */ 212.macro SETUP_SAVE_REFS_AND_ARGS_FRAME 213 // art::Runtime** xIP0 = &art::Runtime::instance_ 214 adrp xIP0, :got:_ZN3art7Runtime9instance_E 215 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 216 217 // Our registers aren't intermixed - just spill in order. 218 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 219 220 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveRefAndArgs]; 221 ldr xIP0, [xIP0, RUNTIME_SAVE_REFS_AND_ARGS_METHOD_OFFSET] 222 223 SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 224 225 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kSaveRefsAndArgs]. 226 // Place sp in Thread::Current()->top_quick_frame. 227 mov xIP0, sp 228 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 229.endm 230 231.macro SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 232 SETUP_SAVE_REFS_AND_ARGS_FRAME_INTERNAL 233 str x0, [sp, #0] // Store ArtMethod* to bottom of stack. 234 // Place sp in Thread::Current()->top_quick_frame. 235 mov xIP0, sp 236 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 237.endm 238 239// TODO: Probably no need to restore registers preserved by aapcs64. 240.macro RESTORE_SAVE_REFS_AND_ARGS_FRAME 241 // FP args. 242 ldp d0, d1, [sp, #16] 243 ldp d2, d3, [sp, #32] 244 ldp d4, d5, [sp, #48] 245 ldp d6, d7, [sp, #64] 246 247 // Core args. 248 RESTORE_TWO_REGS x1, x2, 80 249 RESTORE_TWO_REGS x3, x4, 96 250 RESTORE_TWO_REGS x5, x6, 112 251 252 // x7, Callee-saves. 253 RESTORE_TWO_REGS x7, x20, 128 254 RESTORE_TWO_REGS x21, x22, 144 255 RESTORE_TWO_REGS x23, x24, 160 256 RESTORE_TWO_REGS x25, x26, 176 257 RESTORE_TWO_REGS x27, x28, 192 258 259 // x29(callee-save) and LR. 260 RESTORE_TWO_REGS x29, xLR, 208 261 262 DECREASE_FRAME 224 263.endm 264 265 /* 266 * Macro that sets up the callee save frame to conform with 267 * Runtime::CreateCalleeSaveMethod(kSaveEverything) 268 * when the SP has already been decremented by FRAME_SIZE_SAVE_EVERYTHING 269 * and saving registers x29 and LR is handled elsewhere. 270 */ 271.macro SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP_SKIP_X29_LR 272 // Ugly compile-time check, but we only have the preprocessor. 273#if (FRAME_SIZE_SAVE_EVERYTHING != 512) 274#error "FRAME_SIZE_SAVE_EVERYTHING(ARM64) size not as expected." 275#endif 276 277 // Save FP registers. 278 // For better performance, store d0 and d31 separately, so that all STPs are 16-byte aligned. 279 str d0, [sp, #8] 280 stp d1, d2, [sp, #16] 281 stp d3, d4, [sp, #32] 282 stp d5, d6, [sp, #48] 283 stp d7, d8, [sp, #64] 284 stp d9, d10, [sp, #80] 285 stp d11, d12, [sp, #96] 286 stp d13, d14, [sp, #112] 287 stp d15, d16, [sp, #128] 288 stp d17, d18, [sp, #144] 289 stp d19, d20, [sp, #160] 290 stp d21, d22, [sp, #176] 291 stp d23, d24, [sp, #192] 292 stp d25, d26, [sp, #208] 293 stp d27, d28, [sp, #224] 294 stp d29, d30, [sp, #240] 295 str d31, [sp, #256] 296 297 // Save core registers. 298 SAVE_REG x0, 264 299 SAVE_TWO_REGS x1, x2, 272 300 SAVE_TWO_REGS x3, x4, 288 301 SAVE_TWO_REGS x5, x6, 304 302 SAVE_TWO_REGS x7, x8, 320 303 SAVE_TWO_REGS x9, x10, 336 304 SAVE_TWO_REGS x11, x12, 352 305 SAVE_TWO_REGS x13, x14, 368 306 SAVE_TWO_REGS x15, x16, 384 307 SAVE_TWO_REGS x17, x18, 400 308 SAVE_TWO_REGS x19, x20, 416 309 SAVE_TWO_REGS x21, x22, 432 310 SAVE_TWO_REGS x23, x24, 448 311 SAVE_TWO_REGS x25, x26, 464 312 SAVE_TWO_REGS x27, x28, 480 313 314 // art::Runtime** xIP0 = &art::Runtime::instance_ 315 adrp xIP0, :got:_ZN3art7Runtime9instance_E 316 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 317 318 ldr xIP0, [xIP0] // art::Runtime* xIP0 = art::Runtime::instance_; 319 320 // ArtMethod* xIP0 = Runtime::instance_->callee_save_methods_[kSaveEverything]; 321 ldr xIP0, [xIP0, RUNTIME_SAVE_EVERYTHING_METHOD_OFFSET] 322 323 // Store ArtMethod* Runtime::callee_save_methods_[kSaveEverything]. 324 str xIP0, [sp] 325 // Place sp in Thread::Current()->top_quick_frame. 326 mov xIP0, sp 327 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 328.endm 329 330 /* 331 * Macro that sets up the callee save frame to conform with 332 * Runtime::CreateCalleeSaveMethod(kSaveEverything) 333 */ 334.macro SETUP_SAVE_EVERYTHING_FRAME 335 INCREASE_FRAME 512 336 SAVE_TWO_REGS x29, xLR, 496 337 SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP_SKIP_X29_LR 338.endm 339 340.macro RESTORE_SAVE_EVERYTHING_FRAME_KEEP_X0 341 // Restore FP registers. 342 // For better performance, load d0 and d31 separately, so that all LDPs are 16-byte aligned. 343 ldr d0, [sp, #8] 344 ldp d1, d2, [sp, #16] 345 ldp d3, d4, [sp, #32] 346 ldp d5, d6, [sp, #48] 347 ldp d7, d8, [sp, #64] 348 ldp d9, d10, [sp, #80] 349 ldp d11, d12, [sp, #96] 350 ldp d13, d14, [sp, #112] 351 ldp d15, d16, [sp, #128] 352 ldp d17, d18, [sp, #144] 353 ldp d19, d20, [sp, #160] 354 ldp d21, d22, [sp, #176] 355 ldp d23, d24, [sp, #192] 356 ldp d25, d26, [sp, #208] 357 ldp d27, d28, [sp, #224] 358 ldp d29, d30, [sp, #240] 359 ldr d31, [sp, #256] 360 361 // Restore core registers. 362 RESTORE_TWO_REGS x1, x2, 272 363 RESTORE_TWO_REGS x3, x4, 288 364 RESTORE_TWO_REGS x5, x6, 304 365 RESTORE_TWO_REGS x7, x8, 320 366 RESTORE_TWO_REGS x9, x10, 336 367 RESTORE_TWO_REGS x11, x12, 352 368 RESTORE_TWO_REGS x13, x14, 368 369 RESTORE_TWO_REGS x15, x16, 384 370 RESTORE_TWO_REGS x17, x18, 400 371 RESTORE_TWO_REGS x19, x20, 416 372 RESTORE_TWO_REGS x21, x22, 432 373 RESTORE_TWO_REGS x23, x24, 448 374 RESTORE_TWO_REGS x25, x26, 464 375 RESTORE_TWO_REGS x27, x28, 480 376 RESTORE_TWO_REGS x29, xLR, 496 377 378 DECREASE_FRAME 512 379.endm 380 381.macro RESTORE_SAVE_EVERYTHING_FRAME 382 RESTORE_REG x0, 264 383 RESTORE_SAVE_EVERYTHING_FRAME_KEEP_X0 384.endm 385 386.macro RETURN_IF_RESULT_IS_ZERO 387 cbnz x0, 1f // result non-zero branch over 388 ret // return 3891: 390.endm 391 392.macro RETURN_IF_RESULT_IS_NON_ZERO 393 cbz x0, 1f // result zero branch over 394 ret // return 3951: 396.endm 397 398 /* 399 * Macro that calls through to artDeliverPendingExceptionFromCode, where the pending 400 * exception is Thread::Current()->exception_ when the runtime method frame is ready. 401 */ 402.macro DELIVER_PENDING_EXCEPTION_FRAME_READY 403 mov x0, xSELF 404 405 // Point of no return. 406 bl artDeliverPendingExceptionFromCode // artDeliverPendingExceptionFromCode(Thread*) 407 brk 0 // Unreached 408.endm 409 410 /* 411 * Macro that calls through to artDeliverPendingExceptionFromCode, where the pending 412 * exception is Thread::Current()->exception_. 413 */ 414.macro DELIVER_PENDING_EXCEPTION 415 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 416 DELIVER_PENDING_EXCEPTION_FRAME_READY 417.endm 418 419.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg 420 ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET] // Get exception field. 421 cbnz \reg, 1f 422 ret 4231: 424 DELIVER_PENDING_EXCEPTION 425.endm 426 427.macro RETURN_OR_DELIVER_PENDING_EXCEPTION 428 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0 429.endm 430 431// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register. 432.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 433 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1 434.endm 435 436.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER 437 cbnz w0, 1f // result non-zero branch over 438 ret // return 4391: 440 DELIVER_PENDING_EXCEPTION 441.endm 442 443.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 444 .extern \cxx_name 445ENTRY \c_name 446 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context 447 mov x0, xSELF // pass Thread::Current 448 bl \cxx_name // \cxx_name(Thread*) 449 brk 0 450END \c_name 451.endm 452 453.macro NO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING c_name, cxx_name 454 .extern \cxx_name 455ENTRY \c_name 456 SETUP_SAVE_EVERYTHING_FRAME // save all registers as basis for long jump context 457 mov x0, xSELF // pass Thread::Current 458 bl \cxx_name // \cxx_name(Thread*) 459 brk 0 460END \c_name 461.endm 462 463.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name 464 .extern \cxx_name 465ENTRY \c_name 466 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context. 467 mov x1, xSELF // pass Thread::Current. 468 bl \cxx_name // \cxx_name(arg, Thread*). 469 brk 0 470END \c_name 471.endm 472 473.macro TWO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING c_name, cxx_name 474 .extern \cxx_name 475ENTRY \c_name 476 SETUP_SAVE_EVERYTHING_FRAME // save all registers as basis for long jump context 477 mov x2, xSELF // pass Thread::Current 478 bl \cxx_name // \cxx_name(arg1, arg2, Thread*) 479 brk 0 480END \c_name 481.endm 482 483 /* 484 * Called by managed code, saves callee saves and then calls artThrowException 485 * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. 486 */ 487ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode 488 489 /* 490 * Called by managed code to create and deliver a NullPointerException. 491 */ 492NO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode 493 494 /* 495 * Call installed by a signal handler to create and deliver a NullPointerException. 496 */ 497 .extern art_quick_throw_null_pointer_exception_from_signal 498ENTRY art_quick_throw_null_pointer_exception_from_signal 499 // The fault handler pushes the gc map address, i.e. "return address", to stack 500 // and passes the fault address in LR. So we need to set up the CFI info accordingly. 501 .cfi_def_cfa_offset __SIZEOF_POINTER__ 502 .cfi_rel_offset lr, 0 503 // Save all registers as basis for long jump context. 504 INCREASE_FRAME (FRAME_SIZE_SAVE_EVERYTHING - __SIZEOF_POINTER__) 505 SAVE_REG x29, (FRAME_SIZE_SAVE_EVERYTHING - 2 * __SIZEOF_POINTER__) // LR already saved. 506 SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP_SKIP_X29_LR 507 mov x0, lr // pass the fault address stored in LR by the fault handler. 508 mov x1, xSELF // pass Thread::Current. 509 bl artThrowNullPointerExceptionFromSignal // (arg, Thread*). 510 brk 0 511END art_quick_throw_null_pointer_exception_from_signal 512 513 /* 514 * Called by managed code to create and deliver an ArithmeticException. 515 */ 516NO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING art_quick_throw_div_zero, artThrowDivZeroFromCode 517 518 /* 519 * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds 520 * index, arg2 holds limit. 521 */ 522TWO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING art_quick_throw_array_bounds, artThrowArrayBoundsFromCode 523 524 /* 525 * Called by managed code to create and deliver a StringIndexOutOfBoundsException 526 * as if thrown from a call to String.charAt(). Arg1 holds index, arg2 holds limit. 527 */ 528TWO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING art_quick_throw_string_bounds, artThrowStringBoundsFromCode 529 530 /* 531 * Called by managed code to create and deliver a StackOverflowError. 532 */ 533NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode 534 535 /* 536 * All generated callsites for interface invokes and invocation slow paths will load arguments 537 * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain 538 * the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper. 539 * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1. 540 * 541 * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting 542 * of the target Method* in x0 and method->code_ in x1. 543 * 544 * If unsuccessful, the helper will return null/????. There will be a pending exception in the 545 * thread and we branch to another stub to deliver it. 546 * 547 * On success this wrapper will restore arguments and *jump* to the target, leaving the lr 548 * pointing back to the original caller. 549 * 550 * Adapted from ARM32 code. 551 * 552 * Clobbers xIP0. 553 */ 554.macro INVOKE_TRAMPOLINE_BODY cxx_name 555 .extern \cxx_name 556 SETUP_SAVE_REFS_AND_ARGS_FRAME // save callee saves in case allocation triggers GC 557 // Helper signature is always 558 // (method_idx, *this_object, *caller_method, *self, sp) 559 560 mov x2, xSELF // pass Thread::Current 561 mov x3, sp 562 bl \cxx_name // (method_idx, this, Thread*, SP) 563 mov xIP0, x1 // save Method*->code_ 564 RESTORE_SAVE_REFS_AND_ARGS_FRAME 565 cbz x0, 1f // did we find the target? if not go to exception delivery 566 br xIP0 // tail call to target 5671: 568 DELIVER_PENDING_EXCEPTION 569.endm 570.macro INVOKE_TRAMPOLINE c_name, cxx_name 571ENTRY \c_name 572 INVOKE_TRAMPOLINE_BODY \cxx_name 573END \c_name 574.endm 575 576INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck 577 578INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck 579INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck 580INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck 581INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck 582 583 584.macro INVOKE_STUB_CREATE_FRAME 585 586SAVE_SIZE=15*8 // x4, x5, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 587SAVE_SIZE_AND_METHOD=SAVE_SIZE+8 588 589 590 mov x9, sp // Save stack pointer. 591 .cfi_register sp,x9 592 593 add x10, x2, # SAVE_SIZE_AND_METHOD // calculate size of frame. 594 sub x10, sp, x10 // Calculate SP position - saves + ArtMethod* + args 595 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 596 mov sp, x10 // Set new SP. 597 598 sub x10, x9, #SAVE_SIZE // Calculate new FP (later). Done here as we must move SP 599 .cfi_def_cfa_register x10 // before this. 600 .cfi_adjust_cfa_offset SAVE_SIZE 601 602 str x28, [x10, #112] 603 .cfi_rel_offset x28, 112 604 605 stp x26, x27, [x10, #96] 606 .cfi_rel_offset x26, 96 607 .cfi_rel_offset x27, 104 608 609 stp x24, x25, [x10, #80] 610 .cfi_rel_offset x24, 80 611 .cfi_rel_offset x25, 88 612 613 stp x22, x23, [x10, #64] 614 .cfi_rel_offset x22, 64 615 .cfi_rel_offset x23, 72 616 617 stp x20, x21, [x10, #48] 618 .cfi_rel_offset x20, 48 619 .cfi_rel_offset x21, 56 620 621 stp x9, x19, [x10, #32] // Save old stack pointer and x19. 622 .cfi_rel_offset sp, 32 623 .cfi_rel_offset x19, 40 624 625 stp x4, x5, [x10, #16] // Save result and shorty addresses. 626 .cfi_rel_offset x4, 16 627 .cfi_rel_offset x5, 24 628 629 stp xFP, xLR, [x10] // Store LR & FP. 630 .cfi_rel_offset x29, 0 631 .cfi_rel_offset x30, 8 632 633 mov xFP, x10 // Use xFP now, as it's callee-saved. 634 .cfi_def_cfa_register x29 635 mov xSELF, x3 // Move thread pointer into SELF register. 636 637 // Copy arguments into stack frame. 638 // Use simple copy routine for now. 639 // 4 bytes per slot. 640 // X1 - source address 641 // W2 - args length 642 // X9 - destination address. 643 // W10 - temporary 644 add x9, sp, #8 // Destination address is bottom of stack + null. 645 646 // Copy parameters into the stack. Use numeric label as this is a macro and Clang's assembler 647 // does not have unique-id variables. 6481: 649 cmp w2, #0 650 beq 2f 651 sub w2, w2, #4 // Need 65536 bytes of range. 652 ldr w10, [x1, x2] 653 str w10, [x9, x2] 654 655 b 1b 656 6572: 658 // Store null into ArtMethod* at bottom of frame. 659 str xzr, [sp] 660.endm 661 662.macro INVOKE_STUB_CALL_AND_RETURN 663 664 // load method-> METHOD_QUICK_CODE_OFFSET 665 ldr x9, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 666 // Branch to method. 667 blr x9 668 669 // Restore return value address and shorty address. 670 ldp x4,x5, [xFP, #16] 671 .cfi_restore x4 672 .cfi_restore x5 673 674 ldr x28, [xFP, #112] 675 .cfi_restore x28 676 677 ldp x26, x27, [xFP, #96] 678 .cfi_restore x26 679 .cfi_restore x27 680 681 ldp x24, x25, [xFP, #80] 682 .cfi_restore x24 683 .cfi_restore x25 684 685 ldp x22, x23, [xFP, #64] 686 .cfi_restore x22 687 .cfi_restore x23 688 689 ldp x20, x21, [xFP, #48] 690 .cfi_restore x20 691 .cfi_restore x21 692 693 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 694 ldrb w10, [x5] 695 696 // Check the return type and store the correct register into the jvalue in memory. 697 // Use numeric label as this is a macro and Clang's assembler does not have unique-id variables. 698 699 // Don't set anything for a void type. 700 cmp w10, #'V' 701 beq 3f 702 703 // Is it a double? 704 cmp w10, #'D' 705 bne 1f 706 str d0, [x4] 707 b 3f 708 7091: // Is it a float? 710 cmp w10, #'F' 711 bne 2f 712 str s0, [x4] 713 b 3f 714 7152: // Just store x0. Doesn't matter if it is 64 or 32 bits. 716 str x0, [x4] 717 7183: // Finish up. 719 ldp x2, x19, [xFP, #32] // Restore stack pointer and x19. 720 .cfi_restore x19 721 mov sp, x2 722 .cfi_restore sp 723 724 ldp xFP, xLR, [xFP] // Restore old frame pointer and link register. 725 .cfi_restore x29 726 .cfi_restore x30 727 728 ret 729 730.endm 731 732 733/* 734 * extern"C" void art_quick_invoke_stub(ArtMethod *method, x0 735 * uint32_t *args, x1 736 * uint32_t argsize, w2 737 * Thread *self, x3 738 * JValue *result, x4 739 * char *shorty); x5 740 * +----------------------+ 741 * | | 742 * | C/C++ frame | 743 * | LR'' | 744 * | FP'' | <- SP' 745 * +----------------------+ 746 * +----------------------+ 747 * | x28 | <- TODO: Remove callee-saves. 748 * | : | 749 * | x19 | 750 * | SP' | 751 * | X5 | 752 * | X4 | Saved registers 753 * | LR' | 754 * | FP' | <- FP 755 * +----------------------+ 756 * | uint32_t out[n-1] | 757 * | : : | Outs 758 * | uint32_t out[0] | 759 * | ArtMethod* | <- SP value=null 760 * +----------------------+ 761 * 762 * Outgoing registers: 763 * x0 - Method* 764 * x1-x7 - integer parameters. 765 * d0-d7 - Floating point parameters. 766 * xSELF = self 767 * SP = & of ArtMethod* 768 * x1 = "this" pointer. 769 * 770 */ 771ENTRY art_quick_invoke_stub 772 // Spill registers as per AACPS64 calling convention. 773 INVOKE_STUB_CREATE_FRAME 774 775 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 776 // Parse the passed shorty to determine which register to load. 777 // Load addresses for routines that load WXSD registers. 778 adr x11, .LstoreW2 779 adr x12, .LstoreX2 780 adr x13, .LstoreS0 781 adr x14, .LstoreD0 782 783 // Initialize routine offsets to 0 for integers and floats. 784 // x8 for integers, x15 for floating point. 785 mov x8, #0 786 mov x15, #0 787 788 add x10, x5, #1 // Load shorty address, plus one to skip return value. 789 ldr w1, [x9],#4 // Load "this" parameter, and increment arg pointer. 790 791 // Loop to fill registers. 792.LfillRegisters: 793 ldrb w17, [x10], #1 // Load next character in signature, and increment. 794 cbz w17, .LcallFunction // Exit at end of signature. Shorty 0 terminated. 795 796 cmp w17, #'F' // is this a float? 797 bne .LisDouble 798 799 cmp x15, # 8*12 // Skip this load if all registers full. 800 beq .Ladvance4 801 802 add x17, x13, x15 // Calculate subroutine to jump to. 803 br x17 804 805.LisDouble: 806 cmp w17, #'D' // is this a double? 807 bne .LisLong 808 809 cmp x15, # 8*12 // Skip this load if all registers full. 810 beq .Ladvance8 811 812 add x17, x14, x15 // Calculate subroutine to jump to. 813 br x17 814 815.LisLong: 816 cmp w17, #'J' // is this a long? 817 bne .LisOther 818 819 cmp x8, # 6*12 // Skip this load if all registers full. 820 beq .Ladvance8 821 822 add x17, x12, x8 // Calculate subroutine to jump to. 823 br x17 824 825.LisOther: // Everything else takes one vReg. 826 cmp x8, # 6*12 // Skip this load if all registers full. 827 beq .Ladvance4 828 829 add x17, x11, x8 // Calculate subroutine to jump to. 830 br x17 831 832.Ladvance4: 833 add x9, x9, #4 834 b .LfillRegisters 835 836.Ladvance8: 837 add x9, x9, #8 838 b .LfillRegisters 839 840// Macro for loading a parameter into a register. 841// counter - the register with offset into these tables 842// size - the size of the register - 4 or 8 bytes. 843// register - the name of the register to be loaded. 844.macro LOADREG counter size register return 845 ldr \register , [x9], #\size 846 add \counter, \counter, 12 847 b \return 848.endm 849 850// Store ints. 851.LstoreW2: 852 LOADREG x8 4 w2 .LfillRegisters 853 LOADREG x8 4 w3 .LfillRegisters 854 LOADREG x8 4 w4 .LfillRegisters 855 LOADREG x8 4 w5 .LfillRegisters 856 LOADREG x8 4 w6 .LfillRegisters 857 LOADREG x8 4 w7 .LfillRegisters 858 859// Store longs. 860.LstoreX2: 861 LOADREG x8 8 x2 .LfillRegisters 862 LOADREG x8 8 x3 .LfillRegisters 863 LOADREG x8 8 x4 .LfillRegisters 864 LOADREG x8 8 x5 .LfillRegisters 865 LOADREG x8 8 x6 .LfillRegisters 866 LOADREG x8 8 x7 .LfillRegisters 867 868// Store singles. 869.LstoreS0: 870 LOADREG x15 4 s0 .LfillRegisters 871 LOADREG x15 4 s1 .LfillRegisters 872 LOADREG x15 4 s2 .LfillRegisters 873 LOADREG x15 4 s3 .LfillRegisters 874 LOADREG x15 4 s4 .LfillRegisters 875 LOADREG x15 4 s5 .LfillRegisters 876 LOADREG x15 4 s6 .LfillRegisters 877 LOADREG x15 4 s7 .LfillRegisters 878 879// Store doubles. 880.LstoreD0: 881 LOADREG x15 8 d0 .LfillRegisters 882 LOADREG x15 8 d1 .LfillRegisters 883 LOADREG x15 8 d2 .LfillRegisters 884 LOADREG x15 8 d3 .LfillRegisters 885 LOADREG x15 8 d4 .LfillRegisters 886 LOADREG x15 8 d5 .LfillRegisters 887 LOADREG x15 8 d6 .LfillRegisters 888 LOADREG x15 8 d7 .LfillRegisters 889 890 891.LcallFunction: 892 893 INVOKE_STUB_CALL_AND_RETURN 894 895END art_quick_invoke_stub 896 897/* extern"C" 898 * void art_quick_invoke_static_stub(ArtMethod *method, x0 899 * uint32_t *args, x1 900 * uint32_t argsize, w2 901 * Thread *self, x3 902 * JValue *result, x4 903 * char *shorty); x5 904 */ 905ENTRY art_quick_invoke_static_stub 906 // Spill registers as per AACPS64 calling convention. 907 INVOKE_STUB_CREATE_FRAME 908 909 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 910 // Parse the passed shorty to determine which register to load. 911 // Load addresses for routines that load WXSD registers. 912 adr x11, .LstoreW1_2 913 adr x12, .LstoreX1_2 914 adr x13, .LstoreS0_2 915 adr x14, .LstoreD0_2 916 917 // Initialize routine offsets to 0 for integers and floats. 918 // x8 for integers, x15 for floating point. 919 mov x8, #0 920 mov x15, #0 921 922 add x10, x5, #1 // Load shorty address, plus one to skip return value. 923 924 // Loop to fill registers. 925.LfillRegisters2: 926 ldrb w17, [x10], #1 // Load next character in signature, and increment. 927 cbz w17, .LcallFunction2 // Exit at end of signature. Shorty 0 terminated. 928 929 cmp w17, #'F' // is this a float? 930 bne .LisDouble2 931 932 cmp x15, # 8*12 // Skip this load if all registers full. 933 beq .Ladvance4_2 934 935 add x17, x13, x15 // Calculate subroutine to jump to. 936 br x17 937 938.LisDouble2: 939 cmp w17, #'D' // is this a double? 940 bne .LisLong2 941 942 cmp x15, # 8*12 // Skip this load if all registers full. 943 beq .Ladvance8_2 944 945 add x17, x14, x15 // Calculate subroutine to jump to. 946 br x17 947 948.LisLong2: 949 cmp w17, #'J' // is this a long? 950 bne .LisOther2 951 952 cmp x8, # 7*12 // Skip this load if all registers full. 953 beq .Ladvance8_2 954 955 add x17, x12, x8 // Calculate subroutine to jump to. 956 br x17 957 958.LisOther2: // Everything else takes one vReg. 959 cmp x8, # 7*12 // Skip this load if all registers full. 960 beq .Ladvance4_2 961 962 add x17, x11, x8 // Calculate subroutine to jump to. 963 br x17 964 965.Ladvance4_2: 966 add x9, x9, #4 967 b .LfillRegisters2 968 969.Ladvance8_2: 970 add x9, x9, #8 971 b .LfillRegisters2 972 973// Store ints. 974.LstoreW1_2: 975 LOADREG x8 4 w1 .LfillRegisters2 976 LOADREG x8 4 w2 .LfillRegisters2 977 LOADREG x8 4 w3 .LfillRegisters2 978 LOADREG x8 4 w4 .LfillRegisters2 979 LOADREG x8 4 w5 .LfillRegisters2 980 LOADREG x8 4 w6 .LfillRegisters2 981 LOADREG x8 4 w7 .LfillRegisters2 982 983// Store longs. 984.LstoreX1_2: 985 LOADREG x8 8 x1 .LfillRegisters2 986 LOADREG x8 8 x2 .LfillRegisters2 987 LOADREG x8 8 x3 .LfillRegisters2 988 LOADREG x8 8 x4 .LfillRegisters2 989 LOADREG x8 8 x5 .LfillRegisters2 990 LOADREG x8 8 x6 .LfillRegisters2 991 LOADREG x8 8 x7 .LfillRegisters2 992 993// Store singles. 994.LstoreS0_2: 995 LOADREG x15 4 s0 .LfillRegisters2 996 LOADREG x15 4 s1 .LfillRegisters2 997 LOADREG x15 4 s2 .LfillRegisters2 998 LOADREG x15 4 s3 .LfillRegisters2 999 LOADREG x15 4 s4 .LfillRegisters2 1000 LOADREG x15 4 s5 .LfillRegisters2 1001 LOADREG x15 4 s6 .LfillRegisters2 1002 LOADREG x15 4 s7 .LfillRegisters2 1003 1004// Store doubles. 1005.LstoreD0_2: 1006 LOADREG x15 8 d0 .LfillRegisters2 1007 LOADREG x15 8 d1 .LfillRegisters2 1008 LOADREG x15 8 d2 .LfillRegisters2 1009 LOADREG x15 8 d3 .LfillRegisters2 1010 LOADREG x15 8 d4 .LfillRegisters2 1011 LOADREG x15 8 d5 .LfillRegisters2 1012 LOADREG x15 8 d6 .LfillRegisters2 1013 LOADREG x15 8 d7 .LfillRegisters2 1014 1015 1016.LcallFunction2: 1017 1018 INVOKE_STUB_CALL_AND_RETURN 1019 1020END art_quick_invoke_static_stub 1021 1022 1023 1024/* extern"C" void art_quick_osr_stub(void** stack, x0 1025 * size_t stack_size_in_bytes, x1 1026 * const uin8_t* native_pc, x2 1027 * JValue *result, x3 1028 * char *shorty, x4 1029 * Thread *self) x5 1030 */ 1031ENTRY art_quick_osr_stub 1032SAVE_SIZE=15*8 // x3, x4, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 1033 mov x9, sp // Save stack pointer. 1034 .cfi_register sp,x9 1035 1036 sub x10, sp, # SAVE_SIZE 1037 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 1038 mov sp, x10 // Set new SP. 1039 1040 str x28, [sp, #112] 1041 stp x26, x27, [sp, #96] 1042 stp x24, x25, [sp, #80] 1043 stp x22, x23, [sp, #64] 1044 stp x20, x21, [sp, #48] 1045 stp x9, x19, [sp, #32] // Save old stack pointer and x19. 1046 stp x3, x4, [sp, #16] // Save result and shorty addresses. 1047 stp xFP, xLR, [sp] // Store LR & FP. 1048 mov xSELF, x5 // Move thread pointer into SELF register. 1049 1050 sub sp, sp, #16 1051 str xzr, [sp] // Store null for ArtMethod* slot 1052 // Branch to stub. 1053 bl .Losr_entry 1054 add sp, sp, #16 1055 1056 // Restore return value address and shorty address. 1057 ldp x3,x4, [sp, #16] 1058 ldr x28, [sp, #112] 1059 ldp x26, x27, [sp, #96] 1060 ldp x24, x25, [sp, #80] 1061 ldp x22, x23, [sp, #64] 1062 ldp x20, x21, [sp, #48] 1063 1064 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 1065 ldrb w10, [x4] 1066 1067 // Check the return type and store the correct register into the jvalue in memory. 1068 1069 // Don't set anything for a void type. 1070 cmp w10, #'V' 1071 beq .Losr_exit 1072 1073 // Is it a double? 1074 cmp w10, #'D' 1075 bne .Lno_double 1076 str d0, [x3] 1077 b .Losr_exit 1078 1079.Lno_double: // Is it a float? 1080 cmp w10, #'F' 1081 bne .Lno_float 1082 str s0, [x3] 1083 b .Losr_exit 1084 1085.Lno_float: // Just store x0. Doesn't matter if it is 64 or 32 bits. 1086 str x0, [x3] 1087 1088.Losr_exit: // Finish up. 1089 ldp x2, x19, [sp, #32] // Restore stack pointer and x19. 1090 ldp xFP, xLR, [sp] // Restore old frame pointer and link register. 1091 mov sp, x2 1092 ret 1093 1094.Losr_entry: 1095 // Update stack pointer for the callee 1096 sub sp, sp, x1 1097 1098 // Update link register slot expected by the callee. 1099 sub w1, w1, #8 1100 str lr, [sp, x1] 1101 1102 // Copy arguments into stack frame. 1103 // Use simple copy routine for now. 1104 // 4 bytes per slot. 1105 // X0 - source address 1106 // W1 - args length 1107 // SP - destination address. 1108 // W10 - temporary 1109.Losr_loop_entry: 1110 cmp w1, #0 1111 beq .Losr_loop_exit 1112 sub w1, w1, #4 1113 ldr w10, [x0, x1] 1114 str w10, [sp, x1] 1115 b .Losr_loop_entry 1116 1117.Losr_loop_exit: 1118 // Branch to the OSR entry point. 1119 br x2 1120 1121END art_quick_osr_stub 1122 1123 /* 1124 * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_ 1125 */ 1126 1127ENTRY art_quick_do_long_jump 1128 // Load FPRs 1129 ldp d0, d1, [x1], #16 1130 ldp d2, d3, [x1], #16 1131 ldp d4, d5, [x1], #16 1132 ldp d6, d7, [x1], #16 1133 ldp d8, d9, [x1], #16 1134 ldp d10, d11, [x1], #16 1135 ldp d12, d13, [x1], #16 1136 ldp d14, d15, [x1], #16 1137 ldp d16, d17, [x1], #16 1138 ldp d18, d19, [x1], #16 1139 ldp d20, d21, [x1], #16 1140 ldp d22, d23, [x1], #16 1141 ldp d24, d25, [x1], #16 1142 ldp d26, d27, [x1], #16 1143 ldp d28, d29, [x1], #16 1144 ldp d30, d31, [x1] 1145 1146 // Load GPRs 1147 // TODO: lots of those are smashed, could optimize. 1148 add x0, x0, #30*8 1149 ldp x30, x1, [x0], #-16 // LR & SP 1150 ldp x28, x29, [x0], #-16 1151 ldp x26, x27, [x0], #-16 1152 ldp x24, x25, [x0], #-16 1153 ldp x22, x23, [x0], #-16 1154 ldp x20, x21, [x0], #-16 1155 ldp x18, x19, [x0], #-16 1156 ldp x16, x17, [x0], #-16 1157 ldp x14, x15, [x0], #-16 1158 ldp x12, x13, [x0], #-16 1159 ldp x10, x11, [x0], #-16 1160 ldp x8, x9, [x0], #-16 1161 ldp x6, x7, [x0], #-16 1162 ldp x4, x5, [x0], #-16 1163 ldp x2, x3, [x0], #-16 1164 mov sp, x1 1165 1166 // Need to load PC, it's at the end (after the space for the unused XZR). Use x1. 1167 ldr x1, [x0, #33*8] 1168 // And the value of x0. 1169 ldr x0, [x0] 1170 1171 br x1 1172END art_quick_do_long_jump 1173 1174 /* 1175 * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the 1176 * possibly null object to lock. 1177 * 1178 * Derived from arm32 code. 1179 */ 1180 .extern artLockObjectFromCode 1181ENTRY art_quick_lock_object 1182 cbz w0, .Lslow_lock 1183 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1184.Lretry_lock: 1185 ldr w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop? 1186 ldaxr w1, [x4] // acquire needed only in most common case 1187 and w3, w1, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1188 cbnz w3, .Lnot_unlocked // already thin locked 1189 // unlocked case - x1: original lock word that's zero except for the read barrier bits. 1190 orr x2, x1, x2 // x2 holds thread id with count of 0 with preserved read barrier bits 1191 stxr w3, w2, [x4] 1192 cbnz w3, .Llock_stxr_fail // store failed, retry 1193 ret 1194.Lnot_unlocked: // x1: original lock word 1195 lsr w3, w1, LOCK_WORD_STATE_SHIFT 1196 cbnz w3, .Lslow_lock // if either of the top two bits are set, go slow path 1197 eor w2, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 1198 uxth w2, w2 // zero top 16 bits 1199 cbnz w2, .Lslow_lock // lock word and self thread id's match -> recursive lock 1200 // else contention, go to slow path 1201 and w3, w1, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits. 1202 add w2, w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count in lock word placing in w2 to check overflow 1203 lsr w3, w2, #LOCK_WORD_GC_STATE_SHIFT // if the first gc state bit is set, we overflowed. 1204 cbnz w3, .Lslow_lock // if we overflow the count go slow path 1205 add w2, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count for real 1206 stxr w3, w2, [x4] 1207 cbnz w3, .Llock_stxr_fail // store failed, retry 1208 ret 1209.Llock_stxr_fail: 1210 b .Lretry_lock // retry 1211.Lslow_lock: 1212 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case we block 1213 mov x1, xSELF // pass Thread::Current 1214 bl artLockObjectFromCode // (Object* obj, Thread*) 1215 RESTORE_SAVE_REFS_ONLY_FRAME 1216 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1217END art_quick_lock_object 1218 1219ENTRY art_quick_lock_object_no_inline 1220 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case we block 1221 mov x1, xSELF // pass Thread::Current 1222 bl artLockObjectFromCode // (Object* obj, Thread*) 1223 RESTORE_SAVE_REFS_ONLY_FRAME 1224 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1225END art_quick_lock_object_no_inline 1226 1227 /* 1228 * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. 1229 * x0 holds the possibly null object to lock. 1230 * 1231 * Derived from arm32 code. 1232 */ 1233 .extern artUnlockObjectFromCode 1234ENTRY art_quick_unlock_object 1235 cbz x0, .Lslow_unlock 1236 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1237.Lretry_unlock: 1238#ifndef USE_READ_BARRIER 1239 ldr w1, [x4] 1240#else 1241 ldxr w1, [x4] // Need to use atomic instructions for read barrier 1242#endif 1243 lsr w2, w1, LOCK_WORD_STATE_SHIFT 1244 cbnz w2, .Lslow_unlock // if either of the top two bits are set, go slow path 1245 ldr w2, [xSELF, #THREAD_ID_OFFSET] 1246 and w3, w1, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1247 eor w3, w3, w2 // lock_word.ThreadId() ^ self->ThreadId() 1248 uxth w3, w3 // zero top 16 bits 1249 cbnz w3, .Lslow_unlock // do lock word and self thread id's match? 1250 and w3, w1, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED // zero the gc bits 1251 cmp w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE 1252 bpl .Lrecursive_thin_unlock 1253 // transition to unlocked 1254 and w3, w1, #LOCK_WORD_GC_STATE_MASK_SHIFTED // w3: zero except for the preserved read barrier bits 1255#ifndef USE_READ_BARRIER 1256 stlr w3, [x4] 1257#else 1258 stlxr w2, w3, [x4] // Need to use atomic instructions for read barrier 1259 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1260#endif 1261 ret 1262.Lrecursive_thin_unlock: // w1: original lock word 1263 sub w1, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // decrement count 1264#ifndef USE_READ_BARRIER 1265 str w1, [x4] 1266#else 1267 stxr w2, w1, [x4] // Need to use atomic instructions for read barrier 1268 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1269#endif 1270 ret 1271.Lunlock_stxr_fail: 1272 b .Lretry_unlock // retry 1273.Lslow_unlock: 1274 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case exception allocation triggers GC 1275 mov x1, xSELF // pass Thread::Current 1276 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1277 RESTORE_SAVE_REFS_ONLY_FRAME 1278 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1279END art_quick_unlock_object 1280 1281ENTRY art_quick_unlock_object_no_inline 1282 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case exception allocation triggers GC 1283 mov x1, xSELF // pass Thread::Current 1284 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1285 RESTORE_SAVE_REFS_ONLY_FRAME 1286 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1287END art_quick_unlock_object_no_inline 1288 1289 /* 1290 * Entry from managed code that calls artInstanceOfFromCode and on failure calls 1291 * artThrowClassCastExceptionForObject. 1292 */ 1293 .extern artInstanceOfFromCode 1294 .extern artThrowClassCastExceptionForObject 1295ENTRY art_quick_check_instance_of 1296 // Store arguments and link register 1297 // Stack needs to be 16B aligned on calls. 1298 SAVE_TWO_REGS_INCREASE_FRAME x0, x1, 32 1299 SAVE_REG xLR, 24 1300 1301 // Call runtime code 1302 bl artInstanceOfFromCode 1303 1304 // Check for exception 1305 cbz x0, .Lthrow_class_cast_exception 1306 1307 // Restore and return 1308 .cfi_remember_state 1309 RESTORE_REG xLR, 24 1310 RESTORE_TWO_REGS_DECREASE_FRAME x0, x1, 32 1311 ret 1312 .cfi_restore_state // Reset unwind info so following code unwinds. 1313 .cfi_def_cfa_offset 32 // workaround for clang bug: 31975598 1314 1315.Lthrow_class_cast_exception: 1316 // Restore 1317 RESTORE_REG xLR, 24 1318 RESTORE_TWO_REGS_DECREASE_FRAME x0, x1, 32 1319 1320 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME // save all registers as basis for long jump context 1321 mov x2, xSELF // pass Thread::Current 1322 bl artThrowClassCastExceptionForObject // (Object*, Class*, Thread*) 1323 brk 0 // We should not return here... 1324END art_quick_check_instance_of 1325 1326// Restore xReg's value from [sp, #offset] if xReg is not the same as xExclude. 1327.macro POP_REG_NE xReg, offset, xExclude 1328 .ifnc \xReg, \xExclude 1329 ldr \xReg, [sp, #\offset] // restore xReg 1330 .cfi_restore \xReg 1331 .endif 1332.endm 1333 1334// Restore xReg1's value from [sp, #offset] if xReg1 is not the same as xExclude. 1335// Restore xReg2's value from [sp, #(offset + 8)] if xReg2 is not the same as xExclude. 1336.macro POP_REGS_NE xReg1, xReg2, offset, xExclude 1337 .ifc \xReg1, \xExclude 1338 ldr \xReg2, [sp, #(\offset + 8)] // restore xReg2 1339 .else 1340 .ifc \xReg2, \xExclude 1341 ldr \xReg1, [sp, #\offset] // restore xReg1 1342 .else 1343 ldp \xReg1, \xReg2, [sp, #\offset] // restore xReg1 and xReg2 1344 .endif 1345 .endif 1346 .cfi_restore \xReg1 1347 .cfi_restore \xReg2 1348.endm 1349 1350 /* 1351 * Macro to insert read barrier, only used in art_quick_aput_obj. 1352 * xDest, wDest and xObj are registers, offset is a defined literal such as 1353 * MIRROR_OBJECT_CLASS_OFFSET. Dest needs both x and w versions of the same register to handle 1354 * name mismatch between instructions. This macro uses the lower 32b of register when possible. 1355 * TODO: When read barrier has a fast path, add heap unpoisoning support for the fast path. 1356 */ 1357.macro READ_BARRIER xDest, wDest, xObj, xTemp, wTemp, offset, number 1358#ifdef USE_READ_BARRIER 1359#ifdef USE_BAKER_READ_BARRIER 1360 ldr \wTemp, [\xObj, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 1361 tbnz \wTemp, #LOCK_WORD_READ_BARRIER_STATE_SHIFT, .Lrb_slowpath\number 1362 // False dependency to avoid needing load/load fence. 1363 add \xObj, \xObj, \xTemp, lsr #32 1364 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1365 UNPOISON_HEAP_REF \wDest 1366 b .Lrb_exit\number 1367#endif 1368.Lrb_slowpath\number: 1369 // Store registers used in art_quick_aput_obj (x0-x4, LR), stack is 16B aligned. 1370 SAVE_TWO_REGS_INCREASE_FRAME x0, x1, 48 1371 SAVE_TWO_REGS x2, x3, 16 1372 SAVE_TWO_REGS x4, xLR, 32 1373 1374 // mov x0, \xRef // pass ref in x0 (no-op for now since parameter ref is unused) 1375 .ifnc \xObj, x1 1376 mov x1, \xObj // pass xObj 1377 .endif 1378 mov w2, #\offset // pass offset 1379 bl artReadBarrierSlow // artReadBarrierSlow(ref, xObj, offset) 1380 // No need to unpoison return value in w0, artReadBarrierSlow() would do the unpoisoning. 1381 .ifnc \wDest, w0 1382 mov \wDest, w0 // save return value in wDest 1383 .endif 1384 1385 // Conditionally restore saved registers 1386 POP_REG_NE x0, 0, \xDest 1387 POP_REG_NE x1, 8, \xDest 1388 POP_REG_NE x2, 16, \xDest 1389 POP_REG_NE x3, 24, \xDest 1390 POP_REG_NE x4, 32, \xDest 1391 RESTORE_REG xLR, 40 1392 DECREASE_FRAME 48 1393.Lrb_exit\number: 1394#else 1395 ldr \wDest, [\xObj, #\offset] // Heap reference = 32b. This also zero-extends to \xDest. 1396 UNPOISON_HEAP_REF \wDest 1397#endif // USE_READ_BARRIER 1398.endm 1399 1400#ifdef USE_READ_BARRIER 1401 .extern artReadBarrierSlow 1402#endif 1403ENTRY art_quick_aput_obj 1404 cbz x2, .Ldo_aput_null 1405 READ_BARRIER x3, w3, x0, x3, w3, MIRROR_OBJECT_CLASS_OFFSET, 0 // Heap reference = 32b 1406 // This also zero-extends to x3 1407 READ_BARRIER x3, w3, x3, x4, w4, MIRROR_CLASS_COMPONENT_TYPE_OFFSET, 1 // Heap reference = 32b 1408 // This also zero-extends to x3 1409 READ_BARRIER x4, w4, x2, x4, w4, MIRROR_OBJECT_CLASS_OFFSET, 2 // Heap reference = 32b 1410 // This also zero-extends to x4 1411 cmp w3, w4 // value's type == array's component type - trivial assignability 1412 bne .Lcheck_assignability 1413.Ldo_aput: 1414 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1415 // "Compress" = do nothing 1416 POISON_HEAP_REF w2 1417 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1418 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1419 lsr x0, x0, #7 1420 strb w3, [x3, x0] 1421 ret 1422.Ldo_aput_null: 1423 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1424 // "Compress" = do nothing 1425 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1426 ret 1427.Lcheck_assignability: 1428 // Store arguments and link register 1429 SAVE_TWO_REGS_INCREASE_FRAME x0, x1, 32 1430 SAVE_TWO_REGS x2, xLR, 16 1431 1432 // Call runtime code 1433 mov x0, x3 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1434 mov x1, x4 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1435 bl artIsAssignableFromCode 1436 1437 // Check for exception 1438 cbz x0, .Lthrow_array_store_exception 1439 1440 // Restore 1441 .cfi_remember_state 1442 RESTORE_TWO_REGS x2, xLR, 16 1443 RESTORE_TWO_REGS_DECREASE_FRAME x0, x1, 32 1444 1445 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1446 // "Compress" = do nothing 1447 POISON_HEAP_REF w2 1448 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1449 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1450 lsr x0, x0, #7 1451 strb w3, [x3, x0] 1452 ret 1453 .cfi_restore_state // Reset unwind info so following code unwinds. 1454 .cfi_def_cfa_offset 32 // workaround for clang bug: 31975598 1455.Lthrow_array_store_exception: 1456 RESTORE_TWO_REGS x2, xLR, 16 1457 RESTORE_TWO_REGS_DECREASE_FRAME x0, x1, 32 1458 1459 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 1460 mov x1, x2 // Pass value. 1461 mov x2, xSELF // Pass Thread::Current. 1462 bl artThrowArrayStoreException // (Object*, Object*, Thread*). 1463 brk 0 // Unreached. 1464END art_quick_aput_obj 1465 1466// Macro to facilitate adding new allocation entrypoints. 1467.macro ONE_ARG_DOWNCALL name, entrypoint, return 1468 .extern \entrypoint 1469ENTRY \name 1470 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1471 mov x1, xSELF // pass Thread::Current 1472 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1473 RESTORE_SAVE_REFS_ONLY_FRAME 1474 \return 1475END \name 1476.endm 1477 1478// Macro to facilitate adding new allocation entrypoints. 1479.macro TWO_ARG_DOWNCALL name, entrypoint, return 1480 .extern \entrypoint 1481ENTRY \name 1482 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1483 mov x2, xSELF // pass Thread::Current 1484 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1485 RESTORE_SAVE_REFS_ONLY_FRAME 1486 \return 1487END \name 1488.endm 1489 1490// Macro to facilitate adding new allocation entrypoints. 1491.macro THREE_ARG_DOWNCALL name, entrypoint, return 1492 .extern \entrypoint 1493ENTRY \name 1494 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1495 mov x3, xSELF // pass Thread::Current 1496 bl \entrypoint 1497 RESTORE_SAVE_REFS_ONLY_FRAME 1498 \return 1499END \name 1500.endm 1501 1502// Macro to facilitate adding new allocation entrypoints. 1503.macro FOUR_ARG_DOWNCALL name, entrypoint, return 1504 .extern \entrypoint 1505ENTRY \name 1506 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1507 mov x4, xSELF // pass Thread::Current 1508 bl \entrypoint // 1509 RESTORE_SAVE_REFS_ONLY_FRAME 1510 \return 1511 DELIVER_PENDING_EXCEPTION 1512END \name 1513.endm 1514 1515// Macros taking opportunity of code similarities for downcalls. 1516.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return 1517 .extern \entrypoint 1518ENTRY \name 1519 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1520 mov x1, xSELF // pass Thread::Current 1521 bl \entrypoint // (uint32_t type_idx, Thread*) 1522 RESTORE_SAVE_REFS_ONLY_FRAME 1523 \return 1524END \name 1525.endm 1526 1527.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return 1528 .extern \entrypoint 1529ENTRY \name 1530 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1531 mov x2, xSELF // pass Thread::Current 1532 bl \entrypoint 1533 RESTORE_SAVE_REFS_ONLY_FRAME 1534 \return 1535END \name 1536.endm 1537 1538.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return 1539 .extern \entrypoint 1540ENTRY \name 1541 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1542 mov x3, xSELF // pass Thread::Current 1543 bl \entrypoint 1544 RESTORE_SAVE_REFS_ONLY_FRAME 1545 \return 1546END \name 1547.endm 1548 1549// Macro for string and type resolution and initialization. 1550.macro ONE_ARG_SAVE_EVERYTHING_DOWNCALL name, entrypoint 1551 .extern \entrypoint 1552ENTRY \name 1553 SETUP_SAVE_EVERYTHING_FRAME // save everything for stack crawl 1554 mov x1, xSELF // pass Thread::Current 1555 bl \entrypoint // (int32_t index, Thread* self) 1556 cbz w0, 1f // If result is null, deliver the OOME. 1557 .cfi_remember_state 1558 RESTORE_SAVE_EVERYTHING_FRAME_KEEP_X0 1559 ret // return 1560 .cfi_restore_state 1561 .cfi_def_cfa_offset FRAME_SIZE_SAVE_EVERYTHING // workaround for clang bug: 31975598 15621: 1563 DELIVER_PENDING_EXCEPTION_FRAME_READY 1564END \name 1565.endm 1566 1567.macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1568 cbz w0, 1f // result zero branch over 1569 ret // return 15701: 1571 DELIVER_PENDING_EXCEPTION 1572.endm 1573 1574 /* 1575 * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on 1576 * failure. 1577 */ 1578TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1579 1580 /* 1581 * Entry from managed code when uninitialized static storage, this stub will run the class 1582 * initializer and deliver the exception on error. On success the static storage base is 1583 * returned. 1584 */ 1585ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode 1586 1587ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode 1588ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode 1589ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_resolve_string, artResolveStringFromCode 1590 1591ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1592ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1593ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1594ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1595ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1596ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1597ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1598 1599TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1600TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1601TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1602TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1603TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1604TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1605TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1606 1607TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1608TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1609TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1610TWO_ARG_REF_DOWNCALL art_quick_set64_static, artSet64StaticFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1611TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1612 1613THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1614THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1615THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1616THREE_ARG_REF_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1617THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCompiledCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1618 1619// Generate the allocation entrypoints for each allocator. 1620GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_TLAB_ALLOCATORS 1621// Comment out allocators that have arm64 specific asm. 1622// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_tlab, RegionTLAB) 1623// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB) 1624GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) 1625// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB) 1626// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region_tlab, RegionTLAB) 1627// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region_tlab, RegionTLAB) 1628// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region_tlab, RegionTLAB) 1629// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region_tlab, RegionTLAB) 1630GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB) 1631GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB) 1632GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB) 1633 1634// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_tlab, TLAB) 1635// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_tlab, TLAB) 1636GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_tlab, TLAB) 1637// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_tlab, TLAB) 1638// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_tlab, TLAB) 1639// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_tlab, TLAB) 1640// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_tlab, TLAB) 1641// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_tlab, TLAB) 1642GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_tlab, TLAB) 1643GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_tlab, TLAB) 1644GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_tlab, TLAB) 1645 1646.macro ART_QUICK_ALLOC_OBJECT_ROSALLOC c_name, cxx_name 1647ENTRY \c_name 1648 // Fast path rosalloc allocation. 1649 // x0: type, xSELF(x19): Thread::Current 1650 // x1-x7: free. 1651 ldr x3, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] // Check if the thread local 1652 // allocation stack has room. 1653 // ldp won't work due to large offset. 1654 ldr x4, [xSELF, #THREAD_LOCAL_ALLOC_STACK_END_OFFSET] 1655 cmp x3, x4 1656 bhs .Lslow_path\c_name 1657 ldr w3, [x0, #MIRROR_CLASS_OBJECT_SIZE_ALLOC_FAST_PATH_OFFSET] // Load the object size (x3) 1658 cmp x3, #ROSALLOC_MAX_THREAD_LOCAL_BRACKET_SIZE // Check if the size is for a thread 1659 // local allocation. Also does the 1660 // finalizable and initialization 1661 // checks. 1662 bhs .Lslow_path\c_name 1663 // Compute the rosalloc bracket index 1664 // from the size. Since the size is 1665 // already aligned we can combine the 1666 // two shifts together. 1667 add x4, xSELF, x3, lsr #(ROSALLOC_BRACKET_QUANTUM_SIZE_SHIFT - POINTER_SIZE_SHIFT) 1668 // Subtract pointer size since ther 1669 // are no runs for 0 byte allocations 1670 // and the size is already aligned. 1671 ldr x4, [x4, #(THREAD_ROSALLOC_RUNS_OFFSET - __SIZEOF_POINTER__)] 1672 // Load the free list head (x3). This 1673 // will be the return val. 1674 ldr x3, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1675 cbz x3, .Lslow_path\c_name 1676 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1677 ldr x1, [x3, #ROSALLOC_SLOT_NEXT_OFFSET] // Load the next pointer of the head 1678 // and update the list head with the 1679 // next pointer. 1680 str x1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] 1681 // Store the class pointer in the 1682 // header. This also overwrites the 1683 // next pointer. The offsets are 1684 // asserted to match. 1685#if ROSALLOC_SLOT_NEXT_OFFSET != MIRROR_OBJECT_CLASS_OFFSET 1686#error "Class pointer needs to overwrite next pointer." 1687#endif 1688 POISON_HEAP_REF w0 1689 str w0, [x3, #MIRROR_OBJECT_CLASS_OFFSET] 1690 // Fence. This is "ish" not "ishst" so 1691 // that it also ensures ordering of 1692 // the object size load with respect 1693 // to later accesses to the class 1694 // object. Alternatively we could use 1695 // "ishst" if we use load-acquire for 1696 // the class status load. 1697 // Needs to be done before pushing on 1698 // allocation since Heap::VisitObjects 1699 // relies on seeing the class pointer. 1700 // b/28790624 1701 dmb ish 1702 // Push the new object onto the thread 1703 // local allocation stack and 1704 // increment the thread local 1705 // allocation stack top. 1706 ldr x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1707 str w3, [x1], #COMPRESSED_REFERENCE_SIZE // (Increment x1 as a side effect.) 1708 str x1, [xSELF, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] 1709 // Decrement the size of the free list 1710 ldr w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1711 sub x1, x1, #1 1712 // TODO: consider combining this store 1713 // and the list head store above using 1714 // strd. 1715 str w1, [x4, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] 1716 1717 mov x0, x3 // Set the return value and return. 1718 ret 1719.Lslow_path\c_name: 1720 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1721 mov x1, xSELF // pass Thread::Current 1722 bl \cxx_name 1723 RESTORE_SAVE_REFS_ONLY_FRAME 1724 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1725END \c_name 1726.endm 1727 1728ART_QUICK_ALLOC_OBJECT_ROSALLOC art_quick_alloc_object_resolved_rosalloc, artAllocObjectFromCodeResolvedRosAlloc 1729ART_QUICK_ALLOC_OBJECT_ROSALLOC art_quick_alloc_object_initialized_rosalloc, artAllocObjectFromCodeInitializedRosAlloc 1730 1731.macro ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED slowPathLabel 1732 ldr x4, [xSELF, #THREAD_LOCAL_POS_OFFSET] 1733 ldr x5, [xSELF, #THREAD_LOCAL_END_OFFSET] 1734 ldr w7, [x0, #MIRROR_CLASS_OBJECT_SIZE_ALLOC_FAST_PATH_OFFSET] // Load the object size (x7). 1735 add x6, x4, x7 // Add object size to tlab pos. 1736 cmp x6, x5 // Check if it fits, overflow works 1737 // since the tlab pos and end are 32 1738 // bit values. 1739 bhi \slowPathLabel 1740 str x6, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 1741 ldr x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 1742 add x5, x5, #1 1743 str x5, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 1744 POISON_HEAP_REF w0 1745 str w0, [x4, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 1746 // Fence. This is "ish" not "ishst" so 1747 // that the code after this allocation 1748 // site will see the right values in 1749 // the fields of the class. 1750 // Alternatively we could use "ishst" 1751 // if we use load-acquire for the 1752 // object size load.) 1753 mov x0, x4 1754 dmb ish 1755 ret 1756.endm 1757 1758// The common code for art_quick_alloc_object_*region_tlab 1759.macro GENERATE_ALLOC_OBJECT_RESOLVED_TLAB name, entrypoint 1760ENTRY \name 1761 // Fast path region tlab allocation. 1762 // x0: type, xSELF(x19): Thread::Current 1763 // x1-x7: free. 1764 ALLOC_OBJECT_TLAB_FAST_PATH_RESOLVED .Lslow_path\name 1765.Lslow_path\name: 1766 SETUP_SAVE_REFS_ONLY_FRAME // Save callee saves in case of GC. 1767 mov x1, xSELF // Pass Thread::Current. 1768 bl \entrypoint // (mirror::Class*, Thread*) 1769 RESTORE_SAVE_REFS_ONLY_FRAME 1770 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1771END \name 1772.endm 1773 1774GENERATE_ALLOC_OBJECT_RESOLVED_TLAB art_quick_alloc_object_resolved_region_tlab, artAllocObjectFromCodeResolvedRegionTLAB 1775GENERATE_ALLOC_OBJECT_RESOLVED_TLAB art_quick_alloc_object_initialized_region_tlab, artAllocObjectFromCodeInitializedRegionTLAB 1776GENERATE_ALLOC_OBJECT_RESOLVED_TLAB art_quick_alloc_object_resolved_tlab, artAllocObjectFromCodeResolvedTLAB 1777GENERATE_ALLOC_OBJECT_RESOLVED_TLAB art_quick_alloc_object_initialized_tlab, artAllocObjectFromCodeInitializedTLAB 1778 1779.macro ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE slowPathLabel, xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1780 and \xTemp1, \xTemp1, #OBJECT_ALIGNMENT_MASK_TOGGLED64 // Apply alignemnt mask 1781 // (addr + 7) & ~7. The mask must 1782 // be 64 bits to keep high bits in 1783 // case of overflow. 1784 // Negative sized arrays are handled here since xCount holds a zero extended 32 bit value. 1785 // Negative ints become large 64 bit unsigned ints which will always be larger than max signed 1786 // 32 bit int. Since the max shift for arrays is 3, it can not become a negative 64 bit int. 1787 cmp \xTemp1, #MIN_LARGE_OBJECT_THRESHOLD // Possibly a large object, go slow 1788 bhs \slowPathLabel // path. 1789 1790 ldr \xTemp0, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Check tlab for space, note that 1791 // we use (end - begin) to handle 1792 // negative size arrays. It is 1793 // assumed that a negative size will 1794 // always be greater unsigned than 1795 // region size. 1796 ldr \xTemp2, [xSELF, #THREAD_LOCAL_END_OFFSET] 1797 sub \xTemp2, \xTemp2, \xTemp0 1798 cmp \xTemp1, \xTemp2 1799 bhi \slowPathLabel 1800 // "Point of no slow path". Won't go to the slow path from here on. OK to clobber x0 and x1. 1801 // Move old thread_local_pos to x0 1802 // for the return value. 1803 mov x0, \xTemp0 1804 add \xTemp0, \xTemp0, \xTemp1 1805 str \xTemp0, [xSELF, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. 1806 ldr \xTemp0, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. 1807 add \xTemp0, \xTemp0, #1 1808 str \xTemp0, [xSELF, #THREAD_LOCAL_OBJECTS_OFFSET] 1809 POISON_HEAP_REF \wClass 1810 str \wClass, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. 1811 str \wCount, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] // Store the array length. 1812 // Fence. 1813 dmb ishst 1814 ret 1815.endm 1816 1817.macro GENERATE_ALLOC_ARRAY_TLAB name, entrypoint, size_setup 1818ENTRY \name 1819 // Fast path array allocation for region tlab allocation. 1820 // x0: mirror::Class* type 1821 // x1: int32_t component_count 1822 // x2-x7: free. 1823 mov x3, x0 1824 \size_setup x3, w3, x1, w1, x4, w4, x5, w5, x6, w6 1825 ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE .Lslow_path\name, x3, w3, x1, w1, x4, w4, x5, w5, x6, w6 1826.Lslow_path\name: 1827 // x0: mirror::Class* klass 1828 // x1: int32_t component_count 1829 // x2: Thread* self 1830 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves in case of GC 1831 mov x2, xSELF // pass Thread::Current 1832 bl \entrypoint 1833 RESTORE_SAVE_REFS_ONLY_FRAME 1834 RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1835END \name 1836.endm 1837 1838.macro COMPUTE_ARRAY_SIZE_UNKNOWN xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1839 // Array classes are never finalizable or uninitialized, no need to check. 1840 ldr \wTemp0, [\xClass, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Load component type 1841 UNPOISON_HEAP_REF \wTemp0 1842 ldr \wTemp0, [\xTemp0, #MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET] 1843 lsr \xTemp0, \xTemp0, #PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT // Component size shift is in high 16 1844 // bits. 1845 // xCount is holding a 32 bit value, 1846 // it can not overflow. 1847 lsl \xTemp1, \xCount, \xTemp0 // Calculate data size 1848 // Add array data offset and alignment. 1849 add \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) 1850#if MIRROR_LONG_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4 1851#error Long array data offset must be 4 greater than int array data offset. 1852#endif 1853 1854 add \xTemp0, \xTemp0, #1 // Add 4 to the length only if the 1855 // component size shift is 3 1856 // (for 64 bit alignment). 1857 and \xTemp0, \xTemp0, #4 1858 add \xTemp1, \xTemp1, \xTemp0 1859.endm 1860 1861.macro COMPUTE_ARRAY_SIZE_8 xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1862 // Add array data offset and alignment. 1863 add \xTemp1, \xCount, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) 1864.endm 1865 1866.macro COMPUTE_ARRAY_SIZE_16 xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1867 lsl \xTemp1, \xCount, #1 1868 // Add array data offset and alignment. 1869 add \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) 1870.endm 1871 1872.macro COMPUTE_ARRAY_SIZE_32 xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1873 lsl \xTemp1, \xCount, #2 1874 // Add array data offset and alignment. 1875 add \xTemp1, \xTemp1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) 1876.endm 1877 1878.macro COMPUTE_ARRAY_SIZE_64 xClass, wClass, xCount, wCount, xTemp0, wTemp0, xTemp1, wTemp1, xTemp2, wTemp2 1879 lsl \xTemp1, \xCount, #3 1880 // Add array data offset and alignment. 1881 add \xTemp1, \xTemp1, #(MIRROR_WIDE_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) 1882.endm 1883 1884# TODO(ngeoffray): art_quick_alloc_array_resolved_region_tlab is not used for arm64, remove 1885# the entrypoint once all backends have been updated to use the size variants. 1886GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN 1887GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved8_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_8 1888GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved16_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_16 1889GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved32_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_32 1890GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved64_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_64 1891GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN 1892GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved8_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_8 1893GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved16_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_16 1894GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved32_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_32 1895GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved64_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_64 1896 1897 /* 1898 * Called by managed code when the thread has been asked to suspend. 1899 */ 1900 .extern artTestSuspendFromCode 1901ENTRY art_quick_test_suspend 1902 SETUP_SAVE_EVERYTHING_FRAME // save callee saves for stack crawl 1903 mov x0, xSELF 1904 bl artTestSuspendFromCode // (Thread*) 1905 RESTORE_SAVE_EVERYTHING_FRAME 1906 ret 1907END art_quick_test_suspend 1908 1909ENTRY art_quick_implicit_suspend 1910 mov x0, xSELF 1911 SETUP_SAVE_REFS_ONLY_FRAME // save callee saves for stack crawl 1912 bl artTestSuspendFromCode // (Thread*) 1913 RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN 1914END art_quick_implicit_suspend 1915 1916 /* 1917 * Called by managed code that is attempting to call a method on a proxy class. On entry 1918 * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy 1919 * method agrees with a ref and args callee save frame. 1920 */ 1921 .extern artQuickProxyInvokeHandler 1922ENTRY art_quick_proxy_invoke_handler 1923 SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 1924 mov x2, xSELF // pass Thread::Current 1925 mov x3, sp // pass SP 1926 bl artQuickProxyInvokeHandler // (Method* proxy method, receiver, Thread*, SP) 1927 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 1928 cbnz x2, .Lexception_in_proxy // success if no exception is pending 1929 RESTORE_SAVE_REFS_AND_ARGS_FRAME // Restore frame 1930 fmov d0, x0 // Store result in d0 in case it was float or double 1931 ret // return on success 1932.Lexception_in_proxy: 1933 RESTORE_SAVE_REFS_AND_ARGS_FRAME 1934 DELIVER_PENDING_EXCEPTION 1935END art_quick_proxy_invoke_handler 1936 1937 /* 1938 * Called to resolve an imt conflict. 1939 * x0 is the conflict ArtMethod. 1940 * xIP1 is a hidden argument that holds the target interface method's dex method index. 1941 * 1942 * Note that this stub writes to xIP0, xIP1, and x0. 1943 */ 1944 .extern artInvokeInterfaceTrampoline 1945ENTRY art_quick_imt_conflict_trampoline 1946 ldr xIP0, [sp, #0] // Load referrer 1947 ldr xIP0, [xIP0, #ART_METHOD_DEX_CACHE_METHODS_OFFSET_64] // Load dex cache methods array 1948 ldr xIP0, [xIP0, xIP1, lsl #POINTER_SIZE_SHIFT] // Load interface method 1949 ldr xIP1, [x0, #ART_METHOD_JNI_OFFSET_64] // Load ImtConflictTable 1950 ldr x0, [xIP1] // Load first entry in ImtConflictTable. 1951.Limt_table_iterate: 1952 cmp x0, xIP0 1953 // Branch if found. Benchmarks have shown doing a branch here is better. 1954 beq .Limt_table_found 1955 // If the entry is null, the interface method is not in the ImtConflictTable. 1956 cbz x0, .Lconflict_trampoline 1957 // Iterate over the entries of the ImtConflictTable. 1958 ldr x0, [xIP1, #(2 * __SIZEOF_POINTER__)]! 1959 b .Limt_table_iterate 1960.Limt_table_found: 1961 // We successfully hit an entry in the table. Load the target method 1962 // and jump to it. 1963 ldr x0, [xIP1, #__SIZEOF_POINTER__] 1964 ldr xIP0, [x0, #ART_METHOD_QUICK_CODE_OFFSET_64] 1965 br xIP0 1966.Lconflict_trampoline: 1967 // Call the runtime stub to populate the ImtConflictTable and jump to the 1968 // resolved method. 1969 mov x0, xIP0 // Load interface method 1970 INVOKE_TRAMPOLINE_BODY artInvokeInterfaceTrampoline 1971END art_quick_imt_conflict_trampoline 1972 1973ENTRY art_quick_resolution_trampoline 1974 SETUP_SAVE_REFS_AND_ARGS_FRAME 1975 mov x2, xSELF 1976 mov x3, sp 1977 bl artQuickResolutionTrampoline // (called, receiver, Thread*, SP) 1978 cbz x0, 1f 1979 mov xIP0, x0 // Remember returned code pointer in xIP0. 1980 ldr x0, [sp, #0] // artQuickResolutionTrampoline puts called method in *SP. 1981 RESTORE_SAVE_REFS_AND_ARGS_FRAME 1982 br xIP0 19831: 1984 RESTORE_SAVE_REFS_AND_ARGS_FRAME 1985 DELIVER_PENDING_EXCEPTION 1986END art_quick_resolution_trampoline 1987 1988/* 1989 * Generic JNI frame layout: 1990 * 1991 * #-------------------# 1992 * | | 1993 * | caller method... | 1994 * #-------------------# <--- SP on entry 1995 * | Return X30/LR | 1996 * | X29/FP | callee save 1997 * | X28 | callee save 1998 * | X27 | callee save 1999 * | X26 | callee save 2000 * | X25 | callee save 2001 * | X24 | callee save 2002 * | X23 | callee save 2003 * | X22 | callee save 2004 * | X21 | callee save 2005 * | X20 | callee save 2006 * | X19 | callee save 2007 * | X7 | arg7 2008 * | X6 | arg6 2009 * | X5 | arg5 2010 * | X4 | arg4 2011 * | X3 | arg3 2012 * | X2 | arg2 2013 * | X1 | arg1 2014 * | D7 | float arg 8 2015 * | D6 | float arg 7 2016 * | D5 | float arg 6 2017 * | D4 | float arg 5 2018 * | D3 | float arg 4 2019 * | D2 | float arg 3 2020 * | D1 | float arg 2 2021 * | D0 | float arg 1 2022 * | Method* | <- X0 2023 * #-------------------# 2024 * | local ref cookie | // 4B 2025 * | handle scope size | // 4B 2026 * #-------------------# 2027 * | JNI Call Stack | 2028 * #-------------------# <--- SP on native call 2029 * | | 2030 * | Stack for Regs | The trampoline assembly will pop these values 2031 * | | into registers for native call 2032 * #-------------------# 2033 * | Native code ptr | 2034 * #-------------------# 2035 * | Free scratch | 2036 * #-------------------# 2037 * | Ptr to (1) | <--- SP 2038 * #-------------------# 2039 */ 2040 /* 2041 * Called to do a generic JNI down-call 2042 */ 2043ENTRY art_quick_generic_jni_trampoline 2044 SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_X0 2045 2046 // Save SP , so we can have static CFI info. 2047 mov x28, sp 2048 .cfi_def_cfa_register x28 2049 2050 // This looks the same, but is different: this will be updated to point to the bottom 2051 // of the frame when the handle scope is inserted. 2052 mov xFP, sp 2053 2054 mov xIP0, #5120 2055 sub sp, sp, xIP0 2056 2057 // prepare for artQuickGenericJniTrampoline call 2058 // (Thread*, SP) 2059 // x0 x1 <= C calling convention 2060 // xSELF xFP <= where they are 2061 2062 mov x0, xSELF // Thread* 2063 mov x1, xFP 2064 bl artQuickGenericJniTrampoline // (Thread*, sp) 2065 2066 // The C call will have registered the complete save-frame on success. 2067 // The result of the call is: 2068 // x0: pointer to native code, 0 on error. 2069 // x1: pointer to the bottom of the used area of the alloca, can restore stack till there. 2070 2071 // Check for error = 0. 2072 cbz x0, .Lexception_in_native 2073 2074 // Release part of the alloca. 2075 mov sp, x1 2076 2077 // Save the code pointer 2078 mov xIP0, x0 2079 2080 // Load parameters from frame into registers. 2081 // TODO Check with artQuickGenericJniTrampoline. 2082 // Also, check again APPCS64 - the stack arguments are interleaved. 2083 ldp x0, x1, [sp] 2084 ldp x2, x3, [sp, #16] 2085 ldp x4, x5, [sp, #32] 2086 ldp x6, x7, [sp, #48] 2087 2088 ldp d0, d1, [sp, #64] 2089 ldp d2, d3, [sp, #80] 2090 ldp d4, d5, [sp, #96] 2091 ldp d6, d7, [sp, #112] 2092 2093 add sp, sp, #128 2094 2095 blr xIP0 // native call. 2096 2097 // result sign extension is handled in C code 2098 // prepare for artQuickGenericJniEndTrampoline call 2099 // (Thread*, result, result_f) 2100 // x0 x1 x2 <= C calling convention 2101 mov x1, x0 // Result (from saved). 2102 mov x0, xSELF // Thread register. 2103 fmov x2, d0 // d0 will contain floating point result, but needs to go into x2 2104 2105 bl artQuickGenericJniEndTrampoline 2106 2107 // Pending exceptions possible. 2108 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 2109 cbnz x2, .Lexception_in_native 2110 2111 // Tear down the alloca. 2112 mov sp, x28 2113 .cfi_def_cfa_register sp 2114 2115 // Tear down the callee-save frame. 2116 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2117 2118 // store into fpr, for when it's a fpr return... 2119 fmov d0, x0 2120 ret 2121 2122.Lexception_in_native: 2123 // Move to x1 then sp to please assembler. 2124 ldr x1, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 2125 mov sp, x1 2126 .cfi_def_cfa_register sp 2127 # This will create a new save-all frame, required by the runtime. 2128 DELIVER_PENDING_EXCEPTION 2129END art_quick_generic_jni_trampoline 2130 2131/* 2132 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those 2133 * of a quick call: 2134 * x0 = method being called/to bridge to. 2135 * x1..x7, d0..d7 = arguments to that method. 2136 */ 2137ENTRY art_quick_to_interpreter_bridge 2138 SETUP_SAVE_REFS_AND_ARGS_FRAME // Set up frame and save arguments. 2139 2140 // x0 will contain mirror::ArtMethod* method. 2141 mov x1, xSELF // How to get Thread::Current() ??? 2142 mov x2, sp 2143 2144 // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self, 2145 // mirror::ArtMethod** sp) 2146 bl artQuickToInterpreterBridge 2147 2148 RESTORE_SAVE_REFS_AND_ARGS_FRAME // TODO: no need to restore arguments in this case. 2149 2150 fmov d0, x0 2151 2152 RETURN_OR_DELIVER_PENDING_EXCEPTION 2153END art_quick_to_interpreter_bridge 2154 2155/* 2156 * Called to attempt to execute an obsolete method. 2157 */ 2158ONE_ARG_RUNTIME_EXCEPTION art_invoke_obsolete_method_stub, artInvokeObsoleteMethod 2159 2160 2161// 2162// Instrumentation-related stubs 2163// 2164 .extern artInstrumentationMethodEntryFromCode 2165ENTRY art_quick_instrumentation_entry 2166 SETUP_SAVE_REFS_AND_ARGS_FRAME 2167 2168 mov x20, x0 // Preserve method reference in a callee-save. 2169 2170 mov x2, xSELF 2171 mov x3, xLR 2172 bl artInstrumentationMethodEntryFromCode // (Method*, Object*, Thread*, LR) 2173 2174 mov xIP0, x0 // x0 = result of call. 2175 mov x0, x20 // Reload method reference. 2176 2177 RESTORE_SAVE_REFS_AND_ARGS_FRAME // Note: will restore xSELF 2178 adr xLR, art_quick_instrumentation_exit 2179 br xIP0 // Tail-call method with lr set to art_quick_instrumentation_exit. 2180END art_quick_instrumentation_entry 2181 2182 .extern artInstrumentationMethodExitFromCode 2183ENTRY art_quick_instrumentation_exit 2184 mov xLR, #0 // Clobber LR for later checks. 2185 2186 SETUP_SAVE_REFS_ONLY_FRAME 2187 2188 // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then 2189 // we would need to fully restore it. As there are a lot of callee-save registers, it seems 2190 // easier to have an extra small stack area. 2191 2192 str x0, [sp, #-16]! // Save integer result. 2193 .cfi_adjust_cfa_offset 16 2194 str d0, [sp, #8] // Save floating-point result. 2195 2196 add x1, sp, #16 // Pass SP. 2197 mov x2, x0 // Pass integer result. 2198 fmov x3, d0 // Pass floating-point result. 2199 mov x0, xSELF // Pass Thread. 2200 bl artInstrumentationMethodExitFromCode // (Thread*, SP, gpr_res, fpr_res) 2201 2202 mov xIP0, x0 // Return address from instrumentation call. 2203 mov xLR, x1 // r1 is holding link register if we're to bounce to deoptimize 2204 2205 ldr d0, [sp, #8] // Restore floating-point result. 2206 ldr x0, [sp], #16 // Restore integer result, and drop stack area. 2207 .cfi_adjust_cfa_offset 16 2208 2209 POP_SAVE_REFS_ONLY_FRAME 2210 2211 br xIP0 // Tail-call out. 2212END art_quick_instrumentation_exit 2213 2214 /* 2215 * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization 2216 * will long jump to the upcall with a special exception of -1. 2217 */ 2218 .extern artDeoptimize 2219ENTRY art_quick_deoptimize 2220 SETUP_SAVE_ALL_CALLEE_SAVES_FRAME 2221 mov x0, xSELF // Pass thread. 2222 bl artDeoptimize // (Thread*) 2223 brk 0 2224END art_quick_deoptimize 2225 2226 /* 2227 * Compiled code has requested that we deoptimize into the interpreter. The deoptimization 2228 * will long jump to the upcall with a special exception of -1. 2229 */ 2230 .extern artDeoptimizeFromCompiledCode 2231ENTRY art_quick_deoptimize_from_compiled_code 2232 SETUP_SAVE_EVERYTHING_FRAME 2233 mov x1, xSELF // Pass thread. 2234 bl artDeoptimizeFromCompiledCode // (DeoptimizationKind, Thread*) 2235 brk 0 2236END art_quick_deoptimize_from_compiled_code 2237 2238 2239 /* 2240 * String's indexOf. 2241 * 2242 * TODO: Not very optimized. 2243 * On entry: 2244 * x0: string object (known non-null) 2245 * w1: char to match (known <= 0xFFFF) 2246 * w2: Starting offset in string data 2247 */ 2248ENTRY art_quick_indexof 2249#if (STRING_COMPRESSION_FEATURE) 2250 ldr w4, [x0, #MIRROR_STRING_COUNT_OFFSET] 2251#else 2252 ldr w3, [x0, #MIRROR_STRING_COUNT_OFFSET] 2253#endif 2254 add x0, x0, #MIRROR_STRING_VALUE_OFFSET 2255#if (STRING_COMPRESSION_FEATURE) 2256 /* w4 holds count (with flag) and w3 holds actual length */ 2257 lsr w3, w4, #1 2258#endif 2259 /* Clamp start to [0..count] */ 2260 cmp w2, #0 2261 csel w2, wzr, w2, lt 2262 cmp w2, w3 2263 csel w2, w3, w2, gt 2264 2265 /* Save a copy to compute result */ 2266 mov x5, x0 2267 2268#if (STRING_COMPRESSION_FEATURE) 2269 tbz w4, #0, .Lstring_indexof_compressed 2270#endif 2271 /* Build pointer to start of data to compare and pre-bias */ 2272 add x0, x0, x2, lsl #1 2273 sub x0, x0, #2 2274 /* Compute iteration count */ 2275 sub w2, w3, w2 2276 2277 /* 2278 * At this point we have: 2279 * x0: start of the data to test 2280 * w1: char to compare 2281 * w2: iteration count 2282 * x5: original start of string data 2283 */ 2284 2285 subs w2, w2, #4 2286 b.lt .Lindexof_remainder 2287 2288.Lindexof_loop4: 2289 ldrh w6, [x0, #2]! 2290 ldrh w7, [x0, #2]! 2291 ldrh wIP0, [x0, #2]! 2292 ldrh wIP1, [x0, #2]! 2293 cmp w6, w1 2294 b.eq .Lmatch_0 2295 cmp w7, w1 2296 b.eq .Lmatch_1 2297 cmp wIP0, w1 2298 b.eq .Lmatch_2 2299 cmp wIP1, w1 2300 b.eq .Lmatch_3 2301 subs w2, w2, #4 2302 b.ge .Lindexof_loop4 2303 2304.Lindexof_remainder: 2305 adds w2, w2, #4 2306 b.eq .Lindexof_nomatch 2307 2308.Lindexof_loop1: 2309 ldrh w6, [x0, #2]! 2310 cmp w6, w1 2311 b.eq .Lmatch_3 2312 subs w2, w2, #1 2313 b.ne .Lindexof_loop1 2314 2315.Lindexof_nomatch: 2316 mov x0, #-1 2317 ret 2318 2319.Lmatch_0: 2320 sub x0, x0, #6 2321 sub x0, x0, x5 2322 asr x0, x0, #1 2323 ret 2324.Lmatch_1: 2325 sub x0, x0, #4 2326 sub x0, x0, x5 2327 asr x0, x0, #1 2328 ret 2329.Lmatch_2: 2330 sub x0, x0, #2 2331 sub x0, x0, x5 2332 asr x0, x0, #1 2333 ret 2334.Lmatch_3: 2335 sub x0, x0, x5 2336 asr x0, x0, #1 2337 ret 2338#if (STRING_COMPRESSION_FEATURE) 2339 /* 2340 * Comparing compressed string character-per-character with 2341 * input character 2342 */ 2343.Lstring_indexof_compressed: 2344 add x0, x0, x2 2345 sub x0, x0, #1 2346 sub w2, w3, w2 2347.Lstring_indexof_compressed_loop: 2348 subs w2, w2, #1 2349 b.lt .Lindexof_nomatch 2350 ldrb w6, [x0, #1]! 2351 cmp w6, w1 2352 b.eq .Lstring_indexof_compressed_matched 2353 b .Lstring_indexof_compressed_loop 2354.Lstring_indexof_compressed_matched: 2355 sub x0, x0, x5 2356 ret 2357#endif 2358END art_quick_indexof 2359 2360 /* 2361 * Create a function `name` calling the ReadBarrier::Mark routine, 2362 * getting its argument and returning its result through W register 2363 * `wreg` (corresponding to X register `xreg`), saving and restoring 2364 * all caller-save registers. 2365 * 2366 * If `wreg` is different from `w0`, the generated function follows a 2367 * non-standard runtime calling convention: 2368 * - register `wreg` is used to pass the (sole) argument of this 2369 * function (instead of W0); 2370 * - register `wreg` is used to return the result of this function 2371 * (instead of W0); 2372 * - W0 is treated like a normal (non-argument) caller-save register; 2373 * - everything else is the same as in the standard runtime calling 2374 * convention (e.g. standard callee-save registers are preserved). 2375 */ 2376.macro READ_BARRIER_MARK_REG name, wreg, xreg 2377ENTRY \name 2378 // Reference is null, no work to do at all. 2379 cbz \wreg, .Lret_rb_\name 2380 // Use wIP0 as temp and check the mark bit of the reference. wIP0 is not used by the compiler. 2381 ldr wIP0, [\xreg, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2382 tbz wIP0, #LOCK_WORD_MARK_BIT_SHIFT, .Lnot_marked_rb_\name 2383.Lret_rb_\name: 2384 ret 2385.Lnot_marked_rb_\name: 2386 // Check if the top two bits are one, if this is the case it is a forwarding address. 2387 tst wIP0, wIP0, lsl #1 2388 bmi .Lret_forwarding_address\name 2389.Lslow_rb_\name: 2390 /* 2391 * Allocate 44 stack slots * 8 = 352 bytes: 2392 * - 20 slots for core registers X0-15, X17-X19, LR 2393 * - 24 slots for floating-point registers D0-D7 and D16-D31 2394 */ 2395 // We must not clobber IP1 since code emitted for HLoadClass and HLoadString 2396 // relies on IP1 being preserved. 2397 // Save all potentially live caller-save core registers. 2398 SAVE_TWO_REGS_INCREASE_FRAME x0, x1, 352 2399 SAVE_TWO_REGS x2, x3, 16 2400 SAVE_TWO_REGS x4, x5, 32 2401 SAVE_TWO_REGS x6, x7, 48 2402 SAVE_TWO_REGS x8, x9, 64 2403 SAVE_TWO_REGS x10, x11, 80 2404 SAVE_TWO_REGS x12, x13, 96 2405 SAVE_TWO_REGS x14, x15, 112 2406 SAVE_TWO_REGS x17, x18, 128 // Skip x16, i.e. IP0. 2407 SAVE_TWO_REGS x19, xLR, 144 // Save also return address. 2408 // Save all potentially live caller-save floating-point registers. 2409 stp d0, d1, [sp, #160] 2410 stp d2, d3, [sp, #176] 2411 stp d4, d5, [sp, #192] 2412 stp d6, d7, [sp, #208] 2413 stp d16, d17, [sp, #224] 2414 stp d18, d19, [sp, #240] 2415 stp d20, d21, [sp, #256] 2416 stp d22, d23, [sp, #272] 2417 stp d24, d25, [sp, #288] 2418 stp d26, d27, [sp, #304] 2419 stp d28, d29, [sp, #320] 2420 stp d30, d31, [sp, #336] 2421 2422 .ifnc \wreg, w0 2423 mov w0, \wreg // Pass arg1 - obj from `wreg` 2424 .endif 2425 bl artReadBarrierMark // artReadBarrierMark(obj) 2426 .ifnc \wreg, w0 2427 mov \wreg, w0 // Return result into `wreg` 2428 .endif 2429 2430 // Restore core regs, except `xreg`, as `wreg` is used to return the 2431 // result of this function (simply remove it from the stack instead). 2432 POP_REGS_NE x0, x1, 0, \xreg 2433 POP_REGS_NE x2, x3, 16, \xreg 2434 POP_REGS_NE x4, x5, 32, \xreg 2435 POP_REGS_NE x6, x7, 48, \xreg 2436 POP_REGS_NE x8, x9, 64, \xreg 2437 POP_REGS_NE x10, x11, 80, \xreg 2438 POP_REGS_NE x12, x13, 96, \xreg 2439 POP_REGS_NE x14, x15, 112, \xreg 2440 POP_REGS_NE x17, x18, 128, \xreg 2441 POP_REGS_NE x19, xLR, 144, \xreg // Restore also return address. 2442 // Restore floating-point registers. 2443 ldp d0, d1, [sp, #160] 2444 ldp d2, d3, [sp, #176] 2445 ldp d4, d5, [sp, #192] 2446 ldp d6, d7, [sp, #208] 2447 ldp d16, d17, [sp, #224] 2448 ldp d18, d19, [sp, #240] 2449 ldp d20, d21, [sp, #256] 2450 ldp d22, d23, [sp, #272] 2451 ldp d24, d25, [sp, #288] 2452 ldp d26, d27, [sp, #304] 2453 ldp d28, d29, [sp, #320] 2454 ldp d30, d31, [sp, #336] 2455 // Remove frame and return. 2456 DECREASE_FRAME 352 2457 ret 2458.Lret_forwarding_address\name: 2459 // Shift left by the forwarding address shift. This clears out the state bits since they are 2460 // in the top 2 bits of the lock word. 2461 lsl \wreg, wIP0, #LOCK_WORD_STATE_FORWARDING_ADDRESS_SHIFT 2462 ret 2463END \name 2464.endm 2465 2466READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg00, w0, x0 2467READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg01, w1, x1 2468READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg02, w2, x2 2469READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg03, w3, x3 2470READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg04, w4, x4 2471READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg05, w5, x5 2472READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg06, w6, x6 2473READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg07, w7, x7 2474READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg08, w8, x8 2475READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg09, w9, x9 2476READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg10, w10, x10 2477READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg11, w11, x11 2478READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg12, w12, x12 2479READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg13, w13, x13 2480READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg14, w14, x14 2481READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg15, w15, x15 2482// READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg16, w16, x16 ip0 is blocked 2483READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg17, w17, x17 2484READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg18, w18, x18 2485READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg19, w19, x19 2486READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg20, w20, x20 2487READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg21, w21, x21 2488READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg22, w22, x22 2489READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg23, w23, x23 2490READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg24, w24, x24 2491READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg25, w25, x25 2492READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg26, w26, x26 2493READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg27, w27, x27 2494READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg28, w28, x28 2495READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg29, w29, x29 2496 2497 2498.macro SELECT_X_OR_W_FOR_MACRO macro_to_use, x, w, xreg 2499 .if \xreg 2500 \macro_to_use \x 2501 .else 2502 \macro_to_use \w 2503 .endif 2504.endm 2505 2506.macro FOR_REGISTERS macro_for_register, macro_for_reserved_register, xreg 2507 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x0, w0, \xreg 2508 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x1, w1, \xreg 2509 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x2, w2, \xreg 2510 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x3, w3, \xreg 2511 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x4, w4, \xreg 2512 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x5, w5, \xreg 2513 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x6, w6, \xreg 2514 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x7, w7, \xreg 2515 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x8, w8, \xreg 2516 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x9, w9, \xreg 2517 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x10, w10, \xreg 2518 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x11, w11, \xreg 2519 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x12, w12, \xreg 2520 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x13, w13, \xreg 2521 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x14, w14, \xreg 2522 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x15, w15, \xreg 2523 \macro_for_reserved_register // IP0 is reserved 2524 \macro_for_reserved_register // IP1 is reserved 2525 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x18, w18, \xreg 2526 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x19, w19, \xreg 2527 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x20, w20, \xreg 2528 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x21, w21, \xreg 2529 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x22, w22, \xreg 2530 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x23, w23, \xreg 2531 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x24, w24, \xreg 2532 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x25, w25, \xreg 2533 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x26, w26, \xreg 2534 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x27, w27, \xreg 2535 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x28, w28, \xreg 2536 SELECT_X_OR_W_FOR_MACRO \macro_for_register, x29, w29, \xreg 2537 \macro_for_reserved_register // lr is reserved 2538 \macro_for_reserved_register // sp is reserved 2539.endm 2540 2541.macro FOR_XREGISTERS macro_for_register, macro_for_reserved_register 2542 FOR_REGISTERS \macro_for_register, \macro_for_reserved_register, /* xreg */ 1 2543.endm 2544 2545.macro FOR_WREGISTERS macro_for_register, macro_for_reserved_register 2546 FOR_REGISTERS \macro_for_register, \macro_for_reserved_register, /* xreg */ 0 2547.endm 2548 2549.macro BRK0_BRK0 2550 brk 0 2551 brk 0 2552.endm 2553 2554#if BAKER_MARK_INTROSPECTION_FIELD_LDR_OFFSET != BAKER_MARK_INTROSPECTION_ARRAY_LDR_OFFSET 2555#error "Array and field introspection code sharing requires same LDR offset." 2556#endif 2557.macro INTROSPECTION_ARRAY_LOAD index_reg 2558 ldr wIP0, [xIP0, \index_reg, lsl #2] 2559 b art_quick_read_barrier_mark_introspection 2560.endm 2561 2562.macro MOV_WIP0_TO_WREG_AND_BL_LR reg 2563 mov \reg, wIP0 2564 br lr // Do not use RET as we do not enter the entrypoint with "BL". 2565.endm 2566 2567.macro READ_BARRIER_MARK_INTROSPECTION_SLOW_PATH ldr_offset 2568 /* 2569 * Allocate 44 stack slots * 8 = 352 bytes: 2570 * - 19 slots for core registers X0-15, X18-X19, LR 2571 * - 1 slot padding 2572 * - 24 slots for floating-point registers D0-D7 and D16-D31 2573 */ 2574 // Save all potentially live caller-save core registers. 2575 SAVE_TWO_REGS_INCREASE_FRAME x0, x1, 352 2576 SAVE_TWO_REGS x2, x3, 16 2577 SAVE_TWO_REGS x4, x5, 32 2578 SAVE_TWO_REGS x6, x7, 48 2579 SAVE_TWO_REGS x8, x9, 64 2580 SAVE_TWO_REGS x10, x11, 80 2581 SAVE_TWO_REGS x12, x13, 96 2582 SAVE_TWO_REGS x14, x15, 112 2583 SAVE_TWO_REGS x18, x19, 128 // Skip x16, x17, i.e. IP0, IP1. 2584 SAVE_REG xLR, 144 // Save return address, skip padding at 152. 2585 // Save all potentially live caller-save floating-point registers. 2586 stp d0, d1, [sp, #160] 2587 stp d2, d3, [sp, #176] 2588 stp d4, d5, [sp, #192] 2589 stp d6, d7, [sp, #208] 2590 stp d16, d17, [sp, #224] 2591 stp d18, d19, [sp, #240] 2592 stp d20, d21, [sp, #256] 2593 stp d22, d23, [sp, #272] 2594 stp d24, d25, [sp, #288] 2595 stp d26, d27, [sp, #304] 2596 stp d28, d29, [sp, #320] 2597 stp d30, d31, [sp, #336] 2598 2599 mov x0, xIP0 2600 bl artReadBarrierMark // artReadBarrierMark(obj) 2601 mov xIP0, x0 2602 2603 // Restore core regs, except x0 and x1 as the return register switch case 2604 // address calculation is smoother with an extra register. 2605 RESTORE_TWO_REGS x2, x3, 16 2606 RESTORE_TWO_REGS x4, x5, 32 2607 RESTORE_TWO_REGS x6, x7, 48 2608 RESTORE_TWO_REGS x8, x9, 64 2609 RESTORE_TWO_REGS x10, x11, 80 2610 RESTORE_TWO_REGS x12, x13, 96 2611 RESTORE_TWO_REGS x14, x15, 112 2612 RESTORE_TWO_REGS x18, x19, 128 // Skip x16, x17, i.e. IP0, IP1. 2613 RESTORE_REG xLR, 144 // Restore return address. 2614 // Restore caller-save floating-point registers. 2615 ldp d0, d1, [sp, #160] 2616 ldp d2, d3, [sp, #176] 2617 ldp d4, d5, [sp, #192] 2618 ldp d6, d7, [sp, #208] 2619 ldp d16, d17, [sp, #224] 2620 ldp d18, d19, [sp, #240] 2621 ldp d20, d21, [sp, #256] 2622 ldp d22, d23, [sp, #272] 2623 ldp d24, d25, [sp, #288] 2624 ldp d26, d27, [sp, #304] 2625 ldp d28, d29, [sp, #320] 2626 ldp d30, d31, [sp, #336] 2627 2628 ldr x0, [lr, #\ldr_offset] // Load the instruction. 2629 adr xIP1, .Lmark_introspection_return_switch 2630 bfi xIP1, x0, #3, #5 // Calculate switch case address. 2631 RESTORE_TWO_REGS_DECREASE_FRAME x0, x1, 352 2632 br xIP1 2633.endm 2634 2635 /* 2636 * Use introspection to load a reference from the same address as the LDR 2637 * instruction in generated code would load (unless loaded by the thunk, 2638 * see below), call ReadBarrier::Mark() with that reference if needed 2639 * and return it in the same register as the LDR instruction would load. 2640 * 2641 * The entrypoint is called through a thunk that differs across load kinds. 2642 * For field and array loads the LDR instruction in generated code follows 2643 * the branch to the thunk, i.e. the LDR is at [LR, #-4], and the thunk 2644 * knows the holder and performs the gray bit check, returning to the LDR 2645 * instruction if the object is not gray, so this entrypoint no longer 2646 * needs to know anything about the holder. For GC root loads, the LDR 2647 * instruction in generated code precedes the branch to the thunk (i.e. 2648 * the LDR is at [LR, #-8]) and the thunk does not do the gray bit check. 2649 * 2650 * For field accesses and array loads with a constant index the thunk loads 2651 * the reference into IP0 using introspection and calls the main entrypoint, 2652 * art_quick_read_barrier_mark_introspection. With heap poisoning enabled, 2653 * the passed reference is poisoned. 2654 * 2655 * For array accesses with non-constant index, the thunk inserts the bits 2656 * 16-21 of the LDR instruction to the entrypoint address, effectively 2657 * calculating a switch case label based on the index register (bits 16-20) 2658 * and adding an extra offset (bit 21 is set) to differentiate from the 2659 * main entrypoint, then moves the base register to IP0 and jumps to the 2660 * switch case. Therefore we need to align the main entrypoint to 512 bytes, 2661 * accounting for a 256-byte offset followed by 32 array entrypoints 2662 * starting at art_quick_read_barrier_mark_introspection_arrays, each 2663 * containing an LDR (register) and a branch to the main entrypoint. 2664 * 2665 * For GC root accesses we cannot use the main entrypoint because of the 2666 * different offset where the LDR instruction in generated code is located. 2667 * (And even with heap poisoning enabled, GC roots are not poisoned.) 2668 * To re-use the same entrypoint pointer in generated code, we make sure 2669 * that the gc root entrypoint (a copy of the entrypoint with a different 2670 * offset for introspection loads) is located at a known offset (768 bytes, 2671 * or BAKER_MARK_INTROSPECTION_GC_ROOT_ENTRYPOINT_OFFSET) from the main 2672 * entrypoint and the GC root thunk adjusts the entrypoint pointer, moves 2673 * the root register to IP0 and jumps to the customized entrypoint, 2674 * art_quick_read_barrier_mark_introspection_gc_roots. The thunk also 2675 * performs all the fast-path checks, so we need just the slow path. 2676 * 2677 * The code structure is 2678 * art_quick_read_barrier_mark_introspection: 2679 * Up to 256 bytes for the main entrypoint code. 2680 * Padding to 256 bytes if needed. 2681 * art_quick_read_barrier_mark_introspection_arrays: 2682 * Exactly 256 bytes for array load switch cases (32x2 instructions). 2683 * .Lmark_introspection_return_switch: 2684 * Exactly 256 bytes for return switch cases (32x2 instructions). 2685 * art_quick_read_barrier_mark_introspection_gc_roots: 2686 * GC root entrypoint code. 2687 */ 2688 .balign 512 2689ENTRY art_quick_read_barrier_mark_introspection 2690 // At this point, IP0 contains the reference, IP1 can be freely used. 2691 // For heap poisoning, the reference is poisoned, so unpoison it first. 2692 UNPOISON_HEAP_REF wIP0 2693 // If reference is null, just return it in the right register. 2694 cbz wIP0, .Lmark_introspection_return 2695 // Use wIP1 as temp and check the mark bit of the reference. 2696 ldr wIP1, [xIP0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] 2697 tbz wIP1, #LOCK_WORD_MARK_BIT_SHIFT, .Lmark_introspection_unmarked 2698.Lmark_introspection_return: 2699 // Without an extra register for the return switch case address calculation, 2700 // we exploit the high word of the xIP0 to temporarily store the ref_reg*8, 2701 // so the return switch below must move wIP0 instead of xIP0 to the register. 2702 ldr wIP1, [lr, #BAKER_MARK_INTROSPECTION_FIELD_LDR_OFFSET] // Load the instruction. 2703 bfi xIP0, xIP1, #(32 + 3), #5 // Extract ref_reg*8 to high word in xIP0. 2704 adr xIP1, .Lmark_introspection_return_switch 2705 bfxil xIP1, xIP0, #32, #8 // Calculate return switch case address. 2706 br xIP1 2707.Lmark_introspection_unmarked: 2708 // Check if the top two bits are one, if this is the case it is a forwarding address. 2709 tst wIP1, wIP1, lsl #1 2710 bmi .Lmark_introspection_forwarding_address 2711 READ_BARRIER_MARK_INTROSPECTION_SLOW_PATH BAKER_MARK_INTROSPECTION_FIELD_LDR_OFFSET 2712 2713.Lmark_introspection_forwarding_address: 2714 // Shift left by the forwarding address shift. This clears out the state bits since they are 2715 // in the top 2 bits of the lock word. 2716 lsl wIP0, wIP1, #LOCK_WORD_STATE_FORWARDING_ADDRESS_SHIFT 2717 b .Lmark_introspection_return 2718 2719 // We're very close to the alloted 256B for the entrypoint code before the 2720 // array switch cases. Should we go a little bit over the limit, we can 2721 // move some code after the array switch cases and return switch cases. 2722 .balign 256 2723 .hidden art_quick_read_barrier_mark_introspection_arrays 2724 .global art_quick_read_barrier_mark_introspection_arrays 2725art_quick_read_barrier_mark_introspection_arrays: 2726 FOR_XREGISTERS INTROSPECTION_ARRAY_LOAD, BRK0_BRK0 2727.Lmark_introspection_return_switch: 2728 FOR_WREGISTERS MOV_WIP0_TO_WREG_AND_BL_LR, BRK0_BRK0 2729 .hidden art_quick_read_barrier_mark_introspection_gc_roots 2730 .global art_quick_read_barrier_mark_introspection_gc_roots 2731art_quick_read_barrier_mark_introspection_gc_roots: 2732 READ_BARRIER_MARK_INTROSPECTION_SLOW_PATH BAKER_MARK_INTROSPECTION_GC_ROOT_LDR_OFFSET 2733END art_quick_read_barrier_mark_introspection 2734 2735.extern artInvokePolymorphic 2736ENTRY art_quick_invoke_polymorphic 2737 SETUP_SAVE_REFS_AND_ARGS_FRAME // Save callee saves in case allocation triggers GC. 2738 mov x2, xSELF 2739 mov x3, sp 2740 INCREASE_FRAME 16 // Reserve space for JValue result. 2741 str xzr, [sp, #0] // Initialize result to zero. 2742 mov x0, sp // Set r0 to point to result. 2743 bl artInvokePolymorphic // ArtInvokePolymorphic(result, receiver, thread, save_area) 2744 uxtb w0, w0 // Result is the return type descriptor as a char. 2745 sub w0, w0, 'A' // Convert to zero based index. 2746 cmp w0, 'Z' - 'A' 2747 bhi .Lcleanup_and_return // Clean-up if out-of-bounds. 2748 adrp x1, .Lhandler_table // Compute address of handler table. 2749 add x1, x1, :lo12:.Lhandler_table 2750 ldrb w0, [x1, w0, uxtw] // Lookup handler offset in handler table. 2751 adr x1, .Lstart_of_handlers 2752 add x0, x1, w0, sxtb #2 // Convert relative offset to absolute address. 2753 br x0 // Branch to handler. 2754 2755.Lstart_of_handlers: 2756.Lstore_boolean_result: 2757 ldrb w0, [sp] 2758 b .Lcleanup_and_return 2759.Lstore_char_result: 2760 ldrh w0, [sp] 2761 b .Lcleanup_and_return 2762.Lstore_float_result: 2763 ldr s0, [sp] 2764 str s0, [sp, #32] 2765 b .Lcleanup_and_return 2766.Lstore_double_result: 2767 ldr d0, [sp] 2768 str d0, [sp, #32] 2769 b .Lcleanup_and_return 2770.Lstore_long_result: 2771 ldr x0, [sp] 2772 // Fall-through 2773.Lcleanup_and_return: 2774 DECREASE_FRAME 16 2775 RESTORE_SAVE_REFS_AND_ARGS_FRAME 2776 RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 2777 2778 .section .rodata // Place handler table in read-only section away from text. 2779 .align 2 2780.macro HANDLER_TABLE_OFFSET handler_label 2781 .byte (\handler_label - .Lstart_of_handlers) / 4 2782.endm 2783.Lhandler_table: 2784 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // A 2785 HANDLER_TABLE_OFFSET(.Lstore_long_result) // B (byte) 2786 HANDLER_TABLE_OFFSET(.Lstore_char_result) // C (char) 2787 HANDLER_TABLE_OFFSET(.Lstore_double_result) // D (double) 2788 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // E 2789 HANDLER_TABLE_OFFSET(.Lstore_float_result) // F (float) 2790 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // G 2791 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // H 2792 HANDLER_TABLE_OFFSET(.Lstore_long_result) // I (int) 2793 HANDLER_TABLE_OFFSET(.Lstore_long_result) // J (long) 2794 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // K 2795 HANDLER_TABLE_OFFSET(.Lstore_long_result) // L (object - references are compressed and only 32-bits) 2796 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // M 2797 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // N 2798 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // O 2799 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // P 2800 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // Q 2801 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // R 2802 HANDLER_TABLE_OFFSET(.Lstore_long_result) // S (short) 2803 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // T 2804 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // U 2805 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // V (void) 2806 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // W 2807 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // X 2808 HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // Y 2809 HANDLER_TABLE_OFFSET(.Lstore_boolean_result) // Z (boolean) 2810 .text 2811 2812END art_quick_invoke_polymorphic 2813