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_per_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#else 318 mov x0, #0 319#endif 320 321 adr x1, boot_mmu_config 322 bl core_init_mmu_map 323 324#ifdef CFG_CORE_ASLR 325 /* 326 * Process relocation information again updating for the virtual 327 * map offset. We're doing this now before MMU is enabled as some 328 * of the memory will become write protected. 329 */ 330 ldr x0, boot_mmu_config + CORE_MMU_CONFIG_MAP_OFFSET 331 cbz x0, 1f 332 /* 333 * Update cached_mem_end address with load offset since it was 334 * calculated before relocation. 335 */ 336 adr x5, cached_mem_end 337 ldr x6, [x5] 338 add x6, x6, x0 339 str x6, [x5] 340 adr x1, _start /* Load address */ 341 bl relocate 3421: 343#endif 344 345 bl __get_core_pos 346 bl enable_mmu 347#ifdef CFG_CORE_ASLR 348 /* 349 * Reinitialize console, since register_serial_console() has 350 * previously registered a PA and with ASLR the VA is different 351 * from the PA. 352 */ 353 bl console_init 354#endif 355 356#ifdef CFG_NS_VIRTUALIZATION 357 /* 358 * Initialize partition tables for each partition to 359 * default_partition which has been relocated now to a different VA 360 */ 361 bl core_mmu_set_default_prtn_tbl 362#endif 363 364 bl boot_init_primary_early 365 366#ifdef CFG_MEMTAG 367 init_memtag_per_cpu 368#endif 369 370#ifndef CFG_NS_VIRTUALIZATION 371 mov x23, sp 372 adr_l x0, threads 373 ldr x0, [x0, #THREAD_CTX_STACK_VA_END] 374 mov sp, x0 375 bl thread_get_core_local 376 mov x24, x0 377 str wzr, [x24, #THREAD_CORE_LOCAL_FLAGS] 378#endif 379 bl boot_init_primary_late 380#ifdef CFG_CORE_PAUTH 381 init_pauth_per_cpu 382#endif 383 384#ifndef CFG_NS_VIRTUALIZATION 385 mov x0, #THREAD_CLF_TMP 386 str w0, [x24, #THREAD_CORE_LOCAL_FLAGS] 387 mov sp, x23 388#endif 389 390#ifdef _CFG_CORE_STACK_PROTECTOR 391 /* Update stack canary value */ 392 sub sp, sp, #0x10 393 mov x0, sp 394 mov x1, #1 395 mov x2, #0x8 396 bl plat_get_random_stack_canaries 397 ldr x0, [sp] 398 adr_l x5, __stack_chk_guard 399 str x0, [x5] 400 add sp, sp, #0x10 401#endif 402 403 /* 404 * In case we've touched memory that secondary CPUs will use before 405 * they have turned on their D-cache, clean and invalidate the 406 * D-cache before exiting to normal world. 407 */ 408 adr_l x0, __text_start 409 ldr x1, cached_mem_end 410 sub x1, x1, x0 411 bl dcache_cleaninv_range 412 413 414 /* 415 * Clear current thread id now to allow the thread to be reused on 416 * next entry. Matches the thread_init_boot_thread in 417 * boot.c. 418 */ 419#ifndef CFG_NS_VIRTUALIZATION 420 bl thread_clr_boot_thread 421#endif 422 423#ifdef CFG_CORE_FFA 424 adr x0, cpu_on_handler 425 /* 426 * Compensate for the virtual map offset since cpu_on_handler() is 427 * called with MMU off. 428 */ 429 ldr x1, boot_mmu_config + CORE_MMU_CONFIG_MAP_OFFSET 430 sub x0, x0, x1 431 bl thread_spmc_register_secondary_ep 432 b thread_ffa_msg_wait 433#else 434 /* 435 * Pass the vector address returned from main_init Compensate for 436 * the virtual map offset since cpu_on_handler() is called with MMU 437 * off. 438 */ 439 ldr x0, boot_mmu_config + CORE_MMU_CONFIG_MAP_OFFSET 440 adr x1, thread_vector_table 441 sub x1, x1, x0 442 mov x0, #TEESMC_OPTEED_RETURN_ENTRY_DONE 443 smc #0 444 /* SMC should not return */ 445 panic_at_smc_return 446#endif 447END_FUNC _start 448DECLARE_KEEP_INIT _start 449 450 .section .identity_map.data 451 .balign 8 452LOCAL_DATA cached_mem_end , : 453 .skip 8 454END_DATA cached_mem_end 455 456#if defined(CFG_CORE_ASLR) || defined(CFG_CORE_PHYS_RELOCATABLE) 457LOCAL_FUNC relocate , : 458 /* 459 * x0 holds relocate offset 460 * x1 holds load address 461 */ 462#ifdef CFG_WITH_PAGER 463 adr_l x6, __init_end 464#else 465 adr_l x6, __end 466#endif 467 ldp w2, w3, [x6, #BOOT_EMBDATA_RELOC_OFFSET] 468 469 add x2, x2, x6 /* start of relocations */ 470 add x3, x3, x2 /* end of relocations */ 471 472 /* 473 * Relocations are not formatted as Rela64, instead they are in a 474 * compressed format created by get_reloc_bin() in 475 * scripts/gen_tee_bin.py 476 * 477 * All the R_AARCH64_RELATIVE relocations are translated into a 478 * list of 32-bit offsets from TEE_LOAD_ADDR. At each address a 479 * 64-bit value pointed out which increased with the load offset. 480 */ 481 482#ifdef CFG_WITH_PAGER 483 /* 484 * With pager enabled we can only relocate the pager and init 485 * parts, the rest has to be done when a page is populated. 486 */ 487 sub x6, x6, x1 488#endif 489 490 b 2f 491 /* Loop over the relocation addresses and process all entries */ 4921: ldr w4, [x2], #4 493#ifdef CFG_WITH_PAGER 494 /* Skip too large addresses */ 495 cmp x4, x6 496 b.ge 2f 497#endif 498 add x4, x4, x1 499 ldr x5, [x4] 500 add x5, x5, x0 501 str x5, [x4] 502 5032: cmp x2, x3 504 b.ne 1b 505 506 ret 507END_FUNC relocate 508#endif 509 510/* 511 * void enable_mmu(unsigned long core_pos); 512 * 513 * This function depends on being mapped with in the identity map where 514 * physical address and virtual address is the same. After MMU has been 515 * enabled the instruction pointer will be updated to execute as the new 516 * offset instead. Stack pointers and the return address are updated. 517 */ 518LOCAL_FUNC enable_mmu , : , .identity_map 519 adr x1, boot_mmu_config 520 load_xregs x1, 0, 2, 6 521 /* 522 * x0 = core_pos 523 * x2 = tcr_el1 524 * x3 = mair_el1 525 * x4 = ttbr0_el1_base 526 * x5 = ttbr0_core_offset 527 * x6 = load_offset 528 */ 529 msr tcr_el1, x2 530 msr mair_el1, x3 531 532 /* 533 * ttbr0_el1 = ttbr0_el1_base + ttbr0_core_offset * core_pos 534 */ 535 madd x1, x5, x0, x4 536 msr ttbr0_el1, x1 537 msr ttbr1_el1, xzr 538 isb 539 540 /* Invalidate TLB */ 541 tlbi vmalle1 542 543 /* 544 * Make sure translation table writes have drained into memory and 545 * the TLB invalidation is complete. 546 */ 547 dsb sy 548 isb 549 550 /* Enable the MMU */ 551 mrs x1, sctlr_el1 552 orr x1, x1, #SCTLR_M 553 msr sctlr_el1, x1 554 isb 555 556 /* Update vbar */ 557 mrs x1, vbar_el1 558 add x1, x1, x6 559 msr vbar_el1, x1 560 isb 561 562 /* Invalidate instruction cache and branch predictor */ 563 ic iallu 564 isb 565 566 /* Enable I and D cache */ 567 mrs x1, sctlr_el1 568 orr x1, x1, #SCTLR_I 569 orr x1, x1, #SCTLR_C 570 msr sctlr_el1, x1 571 isb 572 573 /* Adjust stack pointers and return address */ 574 msr spsel, #1 575 add sp, sp, x6 576 msr spsel, #0 577 add sp, sp, x6 578 add x30, x30, x6 579 580 ret 581END_FUNC enable_mmu 582 583 .section .identity_map.data 584 .balign 8 585DATA boot_mmu_config , : /* struct core_mmu_config */ 586 .skip CORE_MMU_CONFIG_SIZE 587END_DATA boot_mmu_config 588 589FUNC cpu_on_handler , : 590 mov x19, x0 591 mov x20, x1 592 mov x21, x30 593 594 adr x0, reset_vect_table 595 msr vbar_el1, x0 596 isb 597 598 set_sctlr_el1 599 isb 600 601#ifdef CFG_PAN 602 init_pan 603#endif 604 605 /* Enable aborts now that we can receive exceptions */ 606 msr daifclr, #DAIFBIT_ABT 607 608 bl __get_core_pos 609 bl enable_mmu 610 611 /* Setup SP_EL0 and SP_EL1, SP will be set to SP_EL0 */ 612 set_sp 613 614#ifdef CFG_MEMTAG 615 init_memtag_per_cpu 616#endif 617#ifdef CFG_CORE_PAUTH 618 init_pauth_per_cpu 619#endif 620 621 mov x0, x19 622 mov x1, x20 623#ifdef CFG_CORE_FFA 624 bl boot_cpu_on_handler 625 b thread_ffa_msg_wait 626#else 627 mov x30, x21 628 b boot_cpu_on_handler 629#endif 630END_FUNC cpu_on_handler 631DECLARE_KEEP_PAGER cpu_on_handler 632 633LOCAL_FUNC unhandled_cpu , : 634 wfi 635 b unhandled_cpu 636END_FUNC unhandled_cpu 637 638LOCAL_DATA stack_tmp_rel , : 639 .word stack_tmp - stack_tmp_rel - STACK_TMP_GUARD 640END_DATA stack_tmp_rel 641 642 /* 643 * This macro verifies that the a given vector doesn't exceed the 644 * architectural limit of 32 instructions. This is meant to be placed 645 * immedately after the last instruction in the vector. It takes the 646 * vector entry as the parameter 647 */ 648 .macro check_vector_size since 649 .if (. - \since) > (32 * 4) 650 .error "Vector exceeds 32 instructions" 651 .endif 652 .endm 653 654 .section .identity_map, "ax", %progbits 655 .align 11 656LOCAL_FUNC reset_vect_table , :, .identity_map, , nobti 657 /* ----------------------------------------------------- 658 * Current EL with SP0 : 0x0 - 0x180 659 * ----------------------------------------------------- 660 */ 661SynchronousExceptionSP0: 662 b SynchronousExceptionSP0 663 check_vector_size SynchronousExceptionSP0 664 665 .align 7 666IrqSP0: 667 b IrqSP0 668 check_vector_size IrqSP0 669 670 .align 7 671FiqSP0: 672 b FiqSP0 673 check_vector_size FiqSP0 674 675 .align 7 676SErrorSP0: 677 b SErrorSP0 678 check_vector_size SErrorSP0 679 680 /* ----------------------------------------------------- 681 * Current EL with SPx: 0x200 - 0x380 682 * ----------------------------------------------------- 683 */ 684 .align 7 685SynchronousExceptionSPx: 686 b SynchronousExceptionSPx 687 check_vector_size SynchronousExceptionSPx 688 689 .align 7 690IrqSPx: 691 b IrqSPx 692 check_vector_size IrqSPx 693 694 .align 7 695FiqSPx: 696 b FiqSPx 697 check_vector_size FiqSPx 698 699 .align 7 700SErrorSPx: 701 b SErrorSPx 702 check_vector_size SErrorSPx 703 704 /* ----------------------------------------------------- 705 * Lower EL using AArch64 : 0x400 - 0x580 706 * ----------------------------------------------------- 707 */ 708 .align 7 709SynchronousExceptionA64: 710 b SynchronousExceptionA64 711 check_vector_size SynchronousExceptionA64 712 713 .align 7 714IrqA64: 715 b IrqA64 716 check_vector_size IrqA64 717 718 .align 7 719FiqA64: 720 b FiqA64 721 check_vector_size FiqA64 722 723 .align 7 724SErrorA64: 725 b SErrorA64 726 check_vector_size SErrorA64 727 728 /* ----------------------------------------------------- 729 * Lower EL using AArch32 : 0x0 - 0x180 730 * ----------------------------------------------------- 731 */ 732 .align 7 733SynchronousExceptionA32: 734 b SynchronousExceptionA32 735 check_vector_size SynchronousExceptionA32 736 737 .align 7 738IrqA32: 739 b IrqA32 740 check_vector_size IrqA32 741 742 .align 7 743FiqA32: 744 b FiqA32 745 check_vector_size FiqA32 746 747 .align 7 748SErrorA32: 749 b SErrorA32 750 check_vector_size SErrorA32 751 752END_FUNC reset_vect_table 753 754BTI(emit_aarch64_feature_1_and GNU_PROPERTY_AARCH64_FEATURE_1_BTI) 755