1/* SPDX-License-Identifier: BSD-2-Clause */ 2/* 3 * Copyright (c) 2015-2022, Linaro Limited 4 * Copyright (c) 2021-2023, Arm Limited 5 */ 6 7#include <platform_config.h> 8 9#include <arm64_macros.S> 10#include <arm.h> 11#include <asm.S> 12#include <generated/asm-defines.h> 13#include <keep.h> 14#include <kernel/thread_private.h> 15#include <mm/core_mmu.h> 16#include <sm/optee_smc.h> 17#include <sm/teesmc_opteed.h> 18#include <sm/teesmc_opteed_macros.h> 19 20 /* 21 * Setup SP_EL0 and SPEL1, SP will be set to SP_EL0. 22 * SP_EL0 is assigned: 23 * stack_tmp + (cpu_id + 1) * stack_tmp_stride - STACK_TMP_GUARD 24 * SP_EL1 is assigned thread_core_local[cpu_id] 25 */ 26 .macro set_sp 27 bl __get_core_pos 28 cmp x0, #CFG_TEE_CORE_NB_CORE 29 /* Unsupported CPU, park it before it breaks something */ 30 bge unhandled_cpu 31 add x0, x0, #1 32 adr_l x1, stack_tmp_stride 33 ldr w1, [x1] 34 mul x1, x0, x1 35 36 /* x0 = stack_tmp - STACK_TMP_GUARD */ 37 adr_l x2, stack_tmp_rel 38 ldr w0, [x2] 39 add x0, x0, x2 40 41 msr spsel, #0 42 add sp, x1, x0 43 bl thread_get_core_local 44 msr spsel, #1 45 mov sp, x0 46 msr spsel, #0 47 .endm 48 49 .macro read_feat_mte reg 50 mrs \reg, id_aa64pfr1_el1 51 ubfx \reg, \reg, #ID_AA64PFR1_EL1_MTE_SHIFT, #4 52 .endm 53 54 .macro read_feat_pan reg 55 mrs \reg, id_mmfr3_el1 56 ubfx \reg, \reg, #ID_MMFR3_EL1_PAN_SHIFT, #4 57 .endm 58 59 .macro set_sctlr_el1 60 mrs x0, sctlr_el1 61 orr x0, x0, #SCTLR_I 62 orr x0, x0, #SCTLR_SA 63 orr x0, x0, #SCTLR_SPAN 64#if defined(CFG_CORE_RWDATA_NOEXEC) 65 orr x0, x0, #SCTLR_WXN 66#endif 67#if defined(CFG_SCTLR_ALIGNMENT_CHECK) 68 orr x0, x0, #SCTLR_A 69#else 70 bic x0, x0, #SCTLR_A 71#endif 72#ifdef CFG_MEMTAG 73 read_feat_mte x1 74 cmp w1, #1 75 b.ls 111f 76 orr x0, x0, #(SCTLR_ATA | SCTLR_ATA0) 77 bic x0, x0, #SCTLR_TCF_MASK 78 bic x0, x0, #SCTLR_TCF0_MASK 79111: 80#endif 81#if defined(CFG_TA_PAUTH) && defined(CFG_TA_BTI) 82 orr x0, x0, #SCTLR_BT0 83#endif 84#if defined(CFG_CORE_PAUTH) && defined(CFG_CORE_BTI) 85 orr x0, x0, #SCTLR_BT1 86#endif 87 msr sctlr_el1, x0 88 .endm 89 90 .macro init_memtag_per_cpu 91 read_feat_mte x0 92 cmp w0, #1 93 b.ls 11f 94 95#ifdef CFG_TEE_CORE_DEBUG 96 /* 97 * This together with GCR_EL1.RRND = 0 will make the tags 98 * acquired with the irg instruction deterministic. 99 */ 100 mov_imm x0, 0xcafe00 101 msr rgsr_el1, x0 102 /* Avoid tag = 0x0 and 0xf */ 103 mov x0, #0 104#else 105 /* 106 * Still avoid tag = 0x0 and 0xf as we use that tag for 107 * everything which isn't explicitly tagged. Setting 108 * GCR_EL1.RRND = 1 to allow an implementation specific 109 * method of generating the tags. 110 */ 111 mov x0, #GCR_EL1_RRND 112#endif 113 orr x0, x0, #1 114 orr x0, x0, #(1 << 15) 115 msr gcr_el1, x0 116 117 /* 118 * Enable the tag checks on the current CPU. 119 * 120 * Depends on boot_init_memtag() having cleared tags for 121 * TEE core memory. Well, not really, addresses with the 122 * tag value 0b0000 will use unchecked access due to 123 * TCR_TCMA0. 124 */ 125 mrs x0, tcr_el1 126 orr x0, x0, #TCR_TBI0 127 orr x0, x0, #TCR_TCMA0 128 msr tcr_el1, x0 129 130 mrs x0, sctlr_el1 131 orr x0, x0, #SCTLR_TCF_SYNC 132 orr x0, x0, #SCTLR_TCF0_SYNC 133 msr sctlr_el1, x0 134 135 isb 13611: 137 .endm 138 139 .macro init_pauth_secondary_cpu 140 msr spsel, #1 141 ldp x0, x1, [sp, #THREAD_CORE_LOCAL_KEYS] 142 msr spsel, #0 143 write_apiakeyhi x0 144 write_apiakeylo x1 145 mrs x0, sctlr_el1 146 orr x0, x0, #SCTLR_ENIA 147 msr sctlr_el1, x0 148 isb 149 .endm 150 151 .macro init_pan 152 read_feat_pan x0 153 cmp x0, #0 154 b.eq 1f 155 mrs x0, sctlr_el1 156 bic x0, x0, #SCTLR_SPAN 157 msr sctlr_el1, x0 158 write_pan_enable 159 1: 160 .endm 161 162FUNC _start , : 163 /* 164 * Temporary copy of boot argument registers, will be passed to 165 * boot_save_args() further down. 166 */ 167 mov x19, x0 168 mov x20, x1 169 mov x21, x2 170 mov x22, x3 171 172 adr x0, reset_vect_table 173 msr vbar_el1, x0 174 isb 175 176#ifdef CFG_PAN 177 init_pan 178#endif 179 180 set_sctlr_el1 181 isb 182 183#ifdef CFG_WITH_PAGER 184 /* 185 * Move init code into correct location and move hashes to a 186 * temporary safe location until the heap is initialized. 187 * 188 * The binary is built as: 189 * [Pager code, rodata and data] : In correct location 190 * [Init code and rodata] : Should be copied to __init_start 191 * [struct boot_embdata + data] : Should be saved before 192 * initializing pager, first uint32_t tells the length of the data 193 */ 194 adr x0, __init_start /* dst */ 195 adr x1, __data_end /* src */ 196 adr x2, __init_end 197 sub x2, x2, x0 /* init len */ 198 ldr w4, [x1, x2] /* length of hashes etc */ 199 add x2, x2, x4 /* length of init and hashes etc */ 200 /* Copy backwards (as memmove) in case we're overlapping */ 201 add x0, x0, x2 /* __init_start + len */ 202 add x1, x1, x2 /* __data_end + len */ 203 adr x3, cached_mem_end 204 str x0, [x3] 205 adr x2, __init_start 206copy_init: 207 ldp x3, x4, [x1, #-16]! 208 stp x3, x4, [x0, #-16]! 209 cmp x0, x2 210 b.gt copy_init 211#else 212 /* 213 * The binary is built as: 214 * [Core, rodata and data] : In correct location 215 * [struct boot_embdata + data] : Should be moved to __end, first 216 * uint32_t tells the length of the struct + data 217 */ 218 adr_l x0, __end /* dst */ 219 adr_l x1, __data_end /* src */ 220 ldr w2, [x1] /* struct boot_embdata::total_len */ 221 /* Copy backwards (as memmove) in case we're overlapping */ 222 add x0, x0, x2 223 add x1, x1, x2 224 adr x3, cached_mem_end 225 str x0, [x3] 226 adr_l x2, __end 227 228copy_init: 229 ldp x3, x4, [x1, #-16]! 230 stp x3, x4, [x0, #-16]! 231 cmp x0, x2 232 b.gt copy_init 233#endif 234 235 /* 236 * Clear .bss, this code obviously depends on the linker keeping 237 * start/end of .bss at least 8 byte aligned. 238 */ 239 adr_l x0, __bss_start 240 adr_l x1, __bss_end 241clear_bss: 242 str xzr, [x0], #8 243 cmp x0, x1 244 b.lt clear_bss 245 246#ifdef CFG_NS_VIRTUALIZATION 247 /* 248 * Clear .nex_bss, this code obviously depends on the linker keeping 249 * start/end of .bss at least 8 byte aligned. 250 */ 251 adr_l x0, __nex_bss_start 252 adr_l x1, __nex_bss_end 253clear_nex_bss: 254 str xzr, [x0], #8 255 cmp x0, x1 256 b.lt clear_nex_bss 257#endif 258 259 260#if defined(CFG_CORE_PHYS_RELOCATABLE) 261 /* 262 * Save the base physical address, it will not change after this 263 * point. 264 */ 265 adr_l x2, core_mmu_tee_load_pa 266 adr x1, _start /* Load address */ 267 str x1, [x2] 268 269 mov_imm x0, TEE_LOAD_ADDR /* Compiled load address */ 270 sub x0, x1, x0 /* Relocatation offset */ 271 272 cbz x0, 1f 273 bl relocate 2741: 275#endif 276 277 /* Setup SP_EL0 and SP_EL1, SP will be set to SP_EL0 */ 278 set_sp 279 280 bl thread_init_thread_core_local 281 282 /* Enable aborts now that we can receive exceptions */ 283 msr daifclr, #DAIFBIT_ABT 284 285 /* 286 * Invalidate dcache for all memory used during initialization to 287 * avoid nasty surprices when the cache is turned on. We must not 288 * invalidate memory not used by OP-TEE since we may invalidate 289 * entries used by for instance ARM Trusted Firmware. 290 */ 291 adr_l x0, __text_start 292 ldr x1, cached_mem_end 293 sub x1, x1, x0 294 bl dcache_cleaninv_range 295 296 /* Enable Console */ 297 bl console_init 298 299 mov x0, x19 300 mov x1, x20 301 mov x2, x21 302 mov x3, x22 303 mov x4, xzr 304 bl boot_save_args 305 306#ifdef CFG_MEMTAG 307 /* 308 * If FEAT_MTE2 is available, initializes the memtag callbacks. 309 * Tags for OP-TEE core memory are then cleared to make it safe to 310 * enable MEMTAG below. 311 */ 312 bl boot_init_memtag 313#endif 314 315#ifdef CFG_CORE_ASLR 316 bl get_aslr_seed 317#ifdef CFG_CORE_ASLR_SEED 318 mov_imm x0, CFG_CORE_ASLR_SEED 319#endif 320#else 321 mov x0, #0 322#endif 323 324 adr x1, boot_mmu_config 325 bl core_init_mmu_map 326 327#ifdef CFG_CORE_ASLR 328 /* 329 * Process relocation information again updating for the virtual 330 * map offset. We're doing this now before MMU is enabled as some 331 * of the memory will become write protected. 332 */ 333 ldr x0, boot_mmu_config + CORE_MMU_CONFIG_MAP_OFFSET 334 cbz x0, 1f 335 /* 336 * Update cached_mem_end address with load offset since it was 337 * calculated before relocation. 338 */ 339 adr x5, cached_mem_end 340 ldr x6, [x5] 341 add x6, x6, x0 342 str x6, [x5] 343 adr x1, _start /* Load address */ 344 bl relocate 3451: 346#endif 347 348 bl __get_core_pos 349 bl enable_mmu 350#ifdef CFG_CORE_ASLR 351 /* 352 * Reinitialize console, since register_serial_console() has 353 * previously registered a PA and with ASLR the VA is different 354 * from the PA. 355 */ 356 bl console_init 357#endif 358 359#ifdef CFG_MEMTAG 360 bl boot_clear_memtag 361#endif 362 363#ifdef CFG_NS_VIRTUALIZATION 364 /* 365 * Initialize partition tables for each partition to 366 * default_partition which has been relocated now to a different VA 367 */ 368 bl core_mmu_set_default_prtn_tbl 369#endif 370 371 bl boot_init_primary_early 372 373#ifdef CFG_MEMTAG 374 init_memtag_per_cpu 375#endif 376 377#ifndef CFG_NS_VIRTUALIZATION 378 mov x23, sp 379 adr_l x0, threads 380 ldr x0, [x0, #THREAD_CTX_STACK_VA_END] 381 mov sp, x0 382 bl thread_get_core_local 383 mov x24, x0 384 str wzr, [x24, #THREAD_CORE_LOCAL_FLAGS] 385#endif 386 bl boot_init_primary_late 387#ifdef CFG_CORE_PAUTH 388 adr_l x0, threads 389 ldp x1, x2, [x0, #THREAD_CTX_KEYS] 390 write_apiakeyhi x1 391 write_apiakeylo x2 392 mrs x0, sctlr_el1 393 orr x0, x0, #SCTLR_ENIA 394 msr sctlr_el1, x0 395 isb 396#endif 397 bl boot_init_primary_final 398 399#ifndef CFG_NS_VIRTUALIZATION 400 mov x0, #THREAD_CLF_TMP 401 str w0, [x24, #THREAD_CORE_LOCAL_FLAGS] 402 mov sp, x23 403#ifdef CFG_CORE_PAUTH 404 ldp x0, x1, [x24, #THREAD_CORE_LOCAL_KEYS] 405 write_apiakeyhi x0 406 write_apiakeylo x1 407 isb 408#endif 409#endif 410 411#ifdef _CFG_CORE_STACK_PROTECTOR 412 /* Update stack canary value */ 413 sub sp, sp, #0x10 414 mov x0, sp 415 mov x1, #1 416 mov x2, #0x8 417 bl plat_get_random_stack_canaries 418 ldr x0, [sp] 419 adr_l x5, __stack_chk_guard 420 str x0, [x5] 421 add sp, sp, #0x10 422#endif 423 424 /* 425 * In case we've touched memory that secondary CPUs will use before 426 * they have turned on their D-cache, clean and invalidate the 427 * D-cache before exiting to normal world. 428 */ 429 adr_l x0, __text_start 430 ldr x1, cached_mem_end 431 sub x1, x1, x0 432 bl dcache_cleaninv_range 433 434 435 /* 436 * Clear current thread id now to allow the thread to be reused on 437 * next entry. Matches the thread_init_boot_thread in 438 * boot.c. 439 */ 440#ifndef CFG_NS_VIRTUALIZATION 441 bl thread_clr_boot_thread 442#endif 443 444#ifdef CFG_CORE_FFA 445 adr x0, cpu_on_handler 446 /* 447 * Compensate for the virtual map offset since cpu_on_handler() is 448 * called with MMU off. 449 */ 450 ldr x1, boot_mmu_config + CORE_MMU_CONFIG_MAP_OFFSET 451 sub x0, x0, x1 452 bl thread_spmc_register_secondary_ep 453 b thread_ffa_msg_wait 454#else 455 /* 456 * Pass the vector address returned from main_init Compensate for 457 * the virtual map offset since cpu_on_handler() is called with MMU 458 * off. 459 */ 460 ldr x0, boot_mmu_config + CORE_MMU_CONFIG_MAP_OFFSET 461 adr x1, thread_vector_table 462 sub x1, x1, x0 463 mov x0, #TEESMC_OPTEED_RETURN_ENTRY_DONE 464 smc #0 465 /* SMC should not return */ 466 panic_at_smc_return 467#endif 468END_FUNC _start 469DECLARE_KEEP_INIT _start 470 471 .section .identity_map.data 472 .balign 8 473LOCAL_DATA cached_mem_end , : 474 .skip 8 475END_DATA cached_mem_end 476 477#if defined(CFG_CORE_ASLR) || defined(CFG_CORE_PHYS_RELOCATABLE) 478LOCAL_FUNC relocate , : 479 /* 480 * x0 holds relocate offset 481 * x1 holds load address 482 */ 483#ifdef CFG_WITH_PAGER 484 adr_l x6, __init_end 485#else 486 adr_l x6, __end 487#endif 488 ldp w2, w3, [x6, #BOOT_EMBDATA_RELOC_OFFSET] 489 490 add x2, x2, x6 /* start of relocations */ 491 add x3, x3, x2 /* end of relocations */ 492 493 /* 494 * Relocations are not formatted as Rela64, instead they are in a 495 * compressed format created by get_reloc_bin() in 496 * scripts/gen_tee_bin.py 497 * 498 * All the R_AARCH64_RELATIVE relocations are translated into a 499 * list of 32-bit offsets from TEE_LOAD_ADDR. At each address a 500 * 64-bit value pointed out which increased with the load offset. 501 */ 502 503#ifdef CFG_WITH_PAGER 504 /* 505 * With pager enabled we can only relocate the pager and init 506 * parts, the rest has to be done when a page is populated. 507 */ 508 sub x6, x6, x1 509#endif 510 511 b 2f 512 /* Loop over the relocation addresses and process all entries */ 5131: ldr w4, [x2], #4 514#ifdef CFG_WITH_PAGER 515 /* Skip too large addresses */ 516 cmp x4, x6 517 b.ge 2f 518#endif 519 add x4, x4, x1 520 ldr x5, [x4] 521 add x5, x5, x0 522 str x5, [x4] 523 5242: cmp x2, x3 525 b.ne 1b 526 527 ret 528END_FUNC relocate 529#endif 530 531/* 532 * void enable_mmu(unsigned long core_pos); 533 * 534 * This function depends on being mapped with in the identity map where 535 * physical address and virtual address is the same. After MMU has been 536 * enabled the instruction pointer will be updated to execute as the new 537 * offset instead. Stack pointers and the return address are updated. 538 */ 539LOCAL_FUNC enable_mmu , : , .identity_map 540 adr x1, boot_mmu_config 541 load_xregs x1, 0, 2, 6 542 /* 543 * x0 = core_pos 544 * x2 = tcr_el1 545 * x3 = mair_el1 546 * x4 = ttbr0_el1_base 547 * x5 = ttbr0_core_offset 548 * x6 = load_offset 549 */ 550 msr tcr_el1, x2 551 msr mair_el1, x3 552 553 /* 554 * ttbr0_el1 = ttbr0_el1_base + ttbr0_core_offset * core_pos 555 */ 556 madd x1, x5, x0, x4 557 msr ttbr0_el1, x1 558 msr ttbr1_el1, xzr 559 isb 560 561 /* Invalidate TLB */ 562 tlbi vmalle1 563 564 /* 565 * Make sure translation table writes have drained into memory and 566 * the TLB invalidation is complete. 567 */ 568 dsb sy 569 isb 570 571 /* Enable the MMU */ 572 mrs x1, sctlr_el1 573 orr x1, x1, #SCTLR_M 574 msr sctlr_el1, x1 575 isb 576 577 /* Update vbar */ 578 mrs x1, vbar_el1 579 add x1, x1, x6 580 msr vbar_el1, x1 581 isb 582 583 /* Invalidate instruction cache and branch predictor */ 584 ic iallu 585 isb 586 587 /* Enable I and D cache */ 588 mrs x1, sctlr_el1 589 orr x1, x1, #SCTLR_I 590 orr x1, x1, #SCTLR_C 591 msr sctlr_el1, x1 592 isb 593 594 /* Adjust stack pointers and return address */ 595 msr spsel, #1 596 add sp, sp, x6 597 msr spsel, #0 598 add sp, sp, x6 599 add x30, x30, x6 600 601 ret 602END_FUNC enable_mmu 603 604 .section .identity_map.data 605 .balign 8 606DATA boot_mmu_config , : /* struct core_mmu_config */ 607 .skip CORE_MMU_CONFIG_SIZE 608END_DATA boot_mmu_config 609 610FUNC cpu_on_handler , : 611 mov x19, x0 612 mov x20, x1 613 mov x21, x30 614 615 adr x0, reset_vect_table 616 msr vbar_el1, x0 617 isb 618 619 set_sctlr_el1 620 isb 621 622#ifdef CFG_PAN 623 init_pan 624#endif 625 626 /* Enable aborts now that we can receive exceptions */ 627 msr daifclr, #DAIFBIT_ABT 628 629 bl __get_core_pos 630 bl enable_mmu 631 632 /* Setup SP_EL0 and SP_EL1, SP will be set to SP_EL0 */ 633 set_sp 634 635#ifdef CFG_MEMTAG 636 init_memtag_per_cpu 637#endif 638#ifdef CFG_CORE_PAUTH 639 init_pauth_secondary_cpu 640#endif 641 642 mov x0, x19 643 mov x1, x20 644#ifdef CFG_CORE_FFA 645 bl boot_cpu_on_handler 646 b thread_ffa_msg_wait 647#else 648 mov x30, x21 649 b boot_cpu_on_handler 650#endif 651END_FUNC cpu_on_handler 652DECLARE_KEEP_PAGER cpu_on_handler 653 654LOCAL_FUNC unhandled_cpu , : 655 wfi 656 b unhandled_cpu 657END_FUNC unhandled_cpu 658 659LOCAL_DATA stack_tmp_rel , : 660 .word stack_tmp - stack_tmp_rel - STACK_TMP_GUARD 661END_DATA stack_tmp_rel 662 663 /* 664 * This macro verifies that the a given vector doesn't exceed the 665 * architectural limit of 32 instructions. This is meant to be placed 666 * immedately after the last instruction in the vector. It takes the 667 * vector entry as the parameter 668 */ 669 .macro check_vector_size since 670 .if (. - \since) > (32 * 4) 671 .error "Vector exceeds 32 instructions" 672 .endif 673 .endm 674 675 .section .identity_map, "ax", %progbits 676 .align 11 677LOCAL_FUNC reset_vect_table , :, .identity_map, , nobti 678 /* ----------------------------------------------------- 679 * Current EL with SP0 : 0x0 - 0x180 680 * ----------------------------------------------------- 681 */ 682SynchronousExceptionSP0: 683 b SynchronousExceptionSP0 684 check_vector_size SynchronousExceptionSP0 685 686 .align 7 687IrqSP0: 688 b IrqSP0 689 check_vector_size IrqSP0 690 691 .align 7 692FiqSP0: 693 b FiqSP0 694 check_vector_size FiqSP0 695 696 .align 7 697SErrorSP0: 698 b SErrorSP0 699 check_vector_size SErrorSP0 700 701 /* ----------------------------------------------------- 702 * Current EL with SPx: 0x200 - 0x380 703 * ----------------------------------------------------- 704 */ 705 .align 7 706SynchronousExceptionSPx: 707 b SynchronousExceptionSPx 708 check_vector_size SynchronousExceptionSPx 709 710 .align 7 711IrqSPx: 712 b IrqSPx 713 check_vector_size IrqSPx 714 715 .align 7 716FiqSPx: 717 b FiqSPx 718 check_vector_size FiqSPx 719 720 .align 7 721SErrorSPx: 722 b SErrorSPx 723 check_vector_size SErrorSPx 724 725 /* ----------------------------------------------------- 726 * Lower EL using AArch64 : 0x400 - 0x580 727 * ----------------------------------------------------- 728 */ 729 .align 7 730SynchronousExceptionA64: 731 b SynchronousExceptionA64 732 check_vector_size SynchronousExceptionA64 733 734 .align 7 735IrqA64: 736 b IrqA64 737 check_vector_size IrqA64 738 739 .align 7 740FiqA64: 741 b FiqA64 742 check_vector_size FiqA64 743 744 .align 7 745SErrorA64: 746 b SErrorA64 747 check_vector_size SErrorA64 748 749 /* ----------------------------------------------------- 750 * Lower EL using AArch32 : 0x0 - 0x180 751 * ----------------------------------------------------- 752 */ 753 .align 7 754SynchronousExceptionA32: 755 b SynchronousExceptionA32 756 check_vector_size SynchronousExceptionA32 757 758 .align 7 759IrqA32: 760 b IrqA32 761 check_vector_size IrqA32 762 763 .align 7 764FiqA32: 765 b FiqA32 766 check_vector_size FiqA32 767 768 .align 7 769SErrorA32: 770 b SErrorA32 771 check_vector_size SErrorA32 772 773END_FUNC reset_vect_table 774 775BTI(emit_aarch64_feature_1_and GNU_PROPERTY_AARCH64_FEATURE_1_BTI) 776