1 // SPDX-License-Identifier: BSD-2-Clause 2 /* 3 * Copyright (c) 2016-2021, Linaro Limited 4 * Copyright (c) 2014, STMicroelectronics International N.V. 5 * Copyright (c) 2020-2021, Arm Limited 6 */ 7 8 #include <config.h> 9 #include <kernel/asan.h> 10 #include <kernel/lockdep.h> 11 #include <kernel/misc.h> 12 #include <kernel/panic.h> 13 #include <kernel/spinlock.h> 14 #include <kernel/thread.h> 15 #include <kernel/thread_private.h> 16 #include <mm/mobj.h> 17 18 struct thread_ctx threads[CFG_NUM_THREADS]; 19 20 struct thread_core_local thread_core_local[CFG_TEE_CORE_NB_CORE] __nex_bss; 21 22 /* 23 * Stacks 24 * 25 * [Lower addresses on the left] 26 * 27 * [ STACK_CANARY_SIZE/2 | STACK_CHECK_EXTRA | STACK_XXX_SIZE | STACK_CANARY_SIZE/2 ] 28 * ^ ^ ^ ^ 29 * stack_xxx[n] "hard" top "soft" top bottom 30 */ 31 32 #ifdef CFG_WITH_STACK_CANARIES 33 #define STACK_CANARY_SIZE (4 * sizeof(long)) 34 #define START_CANARY_VALUE 0xdededede 35 #define END_CANARY_VALUE 0xabababab 36 #define GET_START_CANARY(name, stack_num) name[stack_num][0] 37 #define GET_END_CANARY(name, stack_num) \ 38 name[stack_num][sizeof(name[stack_num]) / sizeof(uint32_t) - 1] 39 #else 40 #define STACK_CANARY_SIZE 0 41 #endif 42 43 #define DECLARE_STACK(name, num_stacks, stack_size, linkage) \ 44 linkage uint32_t name[num_stacks] \ 45 [ROUNDUP(stack_size + STACK_CANARY_SIZE + STACK_CHECK_EXTRA, \ 46 STACK_ALIGNMENT) / sizeof(uint32_t)] \ 47 __attribute__((section(".nozi_stack." # name), \ 48 aligned(STACK_ALIGNMENT))) 49 50 #define GET_STACK(stack) ((vaddr_t)(stack) + STACK_SIZE(stack)) 51 52 DECLARE_STACK(stack_tmp, CFG_TEE_CORE_NB_CORE, 53 STACK_TMP_SIZE + CFG_STACK_TMP_EXTRA, static); 54 DECLARE_STACK(stack_abt, CFG_TEE_CORE_NB_CORE, STACK_ABT_SIZE, static); 55 #ifndef CFG_WITH_PAGER 56 DECLARE_STACK(stack_thread, CFG_NUM_THREADS, 57 STACK_THREAD_SIZE + CFG_STACK_THREAD_EXTRA, static); 58 #endif 59 60 #define GET_STACK_TOP_HARD(stack, n) \ 61 ((vaddr_t)&(stack)[n] + STACK_CANARY_SIZE / 2) 62 #define GET_STACK_TOP_SOFT(stack, n) \ 63 (GET_STACK_TOP_HARD(stack, n) + STACK_CHECK_EXTRA) 64 #define GET_STACK_BOTTOM(stack, n) ((vaddr_t)&(stack)[n] + sizeof(stack[n]) - \ 65 STACK_CANARY_SIZE / 2) 66 67 const void *stack_tmp_export __section(".identity_map.stack_tmp_export") = 68 (void *)(GET_STACK_BOTTOM(stack_tmp, 0) - STACK_TMP_OFFS); 69 const uint32_t stack_tmp_stride __section(".identity_map.stack_tmp_stride") = 70 sizeof(stack_tmp[0]); 71 72 /* 73 * These stack setup info are required by secondary boot cores before they 74 * each locally enable the pager (the mmu). Hence kept in pager sections. 75 */ 76 DECLARE_KEEP_PAGER(stack_tmp_export); 77 DECLARE_KEEP_PAGER(stack_tmp_stride); 78 79 static unsigned int thread_global_lock __nex_bss = SPINLOCK_UNLOCK; 80 81 void thread_init_canaries(void) 82 { 83 #ifdef CFG_WITH_STACK_CANARIES 84 size_t n; 85 #define INIT_CANARY(name) \ 86 for (n = 0; n < ARRAY_SIZE(name); n++) { \ 87 uint32_t *start_canary = &GET_START_CANARY(name, n); \ 88 uint32_t *end_canary = &GET_END_CANARY(name, n); \ 89 \ 90 *start_canary = START_CANARY_VALUE; \ 91 *end_canary = END_CANARY_VALUE; \ 92 } 93 94 INIT_CANARY(stack_tmp); 95 INIT_CANARY(stack_abt); 96 #if !defined(CFG_WITH_PAGER) && !defined(CFG_VIRTUALIZATION) 97 INIT_CANARY(stack_thread); 98 #endif 99 #endif/*CFG_WITH_STACK_CANARIES*/ 100 } 101 102 #define CANARY_DIED(stack, loc, n, addr) \ 103 do { \ 104 EMSG_RAW("Dead canary at %s of '%s[%zu]' (%p)", #loc, #stack, \ 105 n, (void *)addr); \ 106 panic(); \ 107 } while (0) 108 109 void thread_check_canaries(void) 110 { 111 #ifdef CFG_WITH_STACK_CANARIES 112 uint32_t *canary = NULL; 113 size_t n = 0; 114 115 for (n = 0; n < ARRAY_SIZE(stack_tmp); n++) { 116 canary = &GET_START_CANARY(stack_tmp, n); 117 if (*canary != START_CANARY_VALUE) 118 CANARY_DIED(stack_tmp, start, n, canary); 119 canary = &GET_END_CANARY(stack_tmp, n); 120 if (*canary != END_CANARY_VALUE) 121 CANARY_DIED(stack_tmp, end, n, canary); 122 } 123 124 for (n = 0; n < ARRAY_SIZE(stack_abt); n++) { 125 canary = &GET_START_CANARY(stack_abt, n); 126 if (*canary != START_CANARY_VALUE) 127 CANARY_DIED(stack_abt, start, n, canary); 128 canary = &GET_END_CANARY(stack_abt, n); 129 if (*canary != END_CANARY_VALUE) 130 CANARY_DIED(stack_abt, end, n, canary); 131 } 132 #if !defined(CFG_WITH_PAGER) && !defined(CFG_VIRTUALIZATION) 133 for (n = 0; n < ARRAY_SIZE(stack_thread); n++) { 134 canary = &GET_START_CANARY(stack_thread, n); 135 if (*canary != START_CANARY_VALUE) 136 CANARY_DIED(stack_thread, start, n, canary); 137 canary = &GET_END_CANARY(stack_thread, n); 138 if (*canary != END_CANARY_VALUE) 139 CANARY_DIED(stack_thread, end, n, canary); 140 } 141 #endif 142 #endif/*CFG_WITH_STACK_CANARIES*/ 143 } 144 145 void thread_lock_global(void) 146 { 147 cpu_spin_lock(&thread_global_lock); 148 } 149 150 void thread_unlock_global(void) 151 { 152 cpu_spin_unlock(&thread_global_lock); 153 } 154 155 static struct thread_core_local * __nostackcheck 156 get_core_local(unsigned int pos) 157 { 158 /* 159 * Foreign interrupts must be disabled before playing with core_local 160 * since we otherwise may be rescheduled to a different core in the 161 * middle of this function. 162 */ 163 assert(thread_get_exceptions() & THREAD_EXCP_FOREIGN_INTR); 164 165 assert(pos < CFG_TEE_CORE_NB_CORE); 166 return &thread_core_local[pos]; 167 } 168 169 struct thread_core_local * __nostackcheck thread_get_core_local(void) 170 { 171 unsigned int pos = get_core_pos(); 172 173 return get_core_local(pos); 174 } 175 176 #ifdef CFG_CORE_DEBUG_CHECK_STACKS 177 static void print_stack_limits(void) 178 { 179 size_t n = 0; 180 vaddr_t __maybe_unused start = 0; 181 vaddr_t __maybe_unused end = 0; 182 183 for (n = 0; n < CFG_TEE_CORE_NB_CORE; n++) { 184 start = GET_STACK_TOP_SOFT(stack_tmp, n); 185 end = GET_STACK_BOTTOM(stack_tmp, n); 186 DMSG("tmp [%zu] 0x%" PRIxVA "..0x%" PRIxVA, n, start, end); 187 } 188 for (n = 0; n < CFG_TEE_CORE_NB_CORE; n++) { 189 start = GET_STACK_TOP_SOFT(stack_abt, n); 190 end = GET_STACK_BOTTOM(stack_abt, n); 191 DMSG("abt [%zu] 0x%" PRIxVA "..0x%" PRIxVA, n, start, end); 192 } 193 for (n = 0; n < CFG_NUM_THREADS; n++) { 194 end = threads[n].stack_va_end; 195 start = end - STACK_THREAD_SIZE; 196 DMSG("thr [%zu] 0x%" PRIxVA "..0x%" PRIxVA, n, start, end); 197 } 198 } 199 200 static void check_stack_limits(void) 201 { 202 vaddr_t stack_start = 0; 203 vaddr_t stack_end = 0; 204 /* Any value in the current stack frame will do */ 205 vaddr_t current_sp = (vaddr_t)&stack_start; 206 207 if (!get_stack_soft_limits(&stack_start, &stack_end)) 208 panic("Unknown stack limits"); 209 if (current_sp < stack_start || current_sp > stack_end) { 210 DMSG("Stack pointer out of range (0x%" PRIxVA ")", current_sp); 211 print_stack_limits(); 212 panic(); 213 } 214 } 215 216 static bool * __nostackcheck get_stackcheck_recursion_flag(void) 217 { 218 uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR); 219 unsigned int pos = get_core_pos(); 220 struct thread_core_local *l = get_core_local(pos); 221 int ct = l->curr_thread; 222 bool *p = NULL; 223 224 if (l->flags & (THREAD_CLF_ABORT | THREAD_CLF_TMP)) 225 p = &l->stackcheck_recursion; 226 else if (!l->flags) 227 p = &threads[ct].tsd.stackcheck_recursion; 228 229 thread_unmask_exceptions(exceptions); 230 return p; 231 } 232 233 void __cyg_profile_func_enter(void *this_fn, void *call_site); 234 void __nostackcheck __cyg_profile_func_enter(void *this_fn __unused, 235 void *call_site __unused) 236 { 237 bool *p = get_stackcheck_recursion_flag(); 238 239 assert(p); 240 if (*p) 241 return; 242 *p = true; 243 check_stack_limits(); 244 *p = false; 245 } 246 247 void __cyg_profile_func_exit(void *this_fn, void *call_site); 248 void __nostackcheck __cyg_profile_func_exit(void *this_fn __unused, 249 void *call_site __unused) 250 { 251 } 252 #else 253 static void print_stack_limits(void) 254 { 255 } 256 #endif 257 258 void thread_init_boot_thread(void) 259 { 260 struct thread_core_local *l = thread_get_core_local(); 261 262 thread_init_threads(); 263 264 l->curr_thread = 0; 265 threads[0].state = THREAD_STATE_ACTIVE; 266 } 267 268 void __nostackcheck thread_clr_boot_thread(void) 269 { 270 struct thread_core_local *l = thread_get_core_local(); 271 272 assert(l->curr_thread >= 0 && l->curr_thread < CFG_NUM_THREADS); 273 assert(threads[l->curr_thread].state == THREAD_STATE_ACTIVE); 274 threads[l->curr_thread].state = THREAD_STATE_FREE; 275 l->curr_thread = THREAD_ID_INVALID; 276 } 277 278 void __nostackcheck *thread_get_tmp_sp(void) 279 { 280 struct thread_core_local *l = thread_get_core_local(); 281 282 /* 283 * Called from assembly when switching to the temporary stack, so flags 284 * need updating 285 */ 286 l->flags |= THREAD_CLF_TMP; 287 288 return (void *)l->tmp_stack_va_end; 289 } 290 291 vaddr_t thread_stack_start(void) 292 { 293 struct thread_ctx *thr; 294 int ct = thread_get_id_may_fail(); 295 296 if (ct == THREAD_ID_INVALID) 297 return 0; 298 299 thr = threads + ct; 300 return thr->stack_va_end - STACK_THREAD_SIZE; 301 } 302 303 size_t thread_stack_size(void) 304 { 305 return STACK_THREAD_SIZE; 306 } 307 308 bool get_stack_limits(vaddr_t *start, vaddr_t *end, bool hard) 309 { 310 uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR); 311 unsigned int pos = get_core_pos(); 312 struct thread_core_local *l = get_core_local(pos); 313 int ct = l->curr_thread; 314 bool ret = false; 315 316 if (l->flags & THREAD_CLF_TMP) { 317 if (hard) 318 *start = GET_STACK_TOP_HARD(stack_tmp, pos); 319 else 320 *start = GET_STACK_TOP_SOFT(stack_tmp, pos); 321 *end = GET_STACK_BOTTOM(stack_tmp, pos); 322 ret = true; 323 } else if (l->flags & THREAD_CLF_ABORT) { 324 if (hard) 325 *start = GET_STACK_TOP_HARD(stack_abt, pos); 326 else 327 *start = GET_STACK_TOP_SOFT(stack_abt, pos); 328 *end = GET_STACK_BOTTOM(stack_abt, pos); 329 ret = true; 330 } else if (!l->flags) { 331 if (ct < 0 || ct >= CFG_NUM_THREADS) 332 goto out; 333 334 *end = threads[ct].stack_va_end; 335 *start = *end - STACK_THREAD_SIZE; 336 if (!hard) 337 *start += STACK_CHECK_EXTRA; 338 ret = true; 339 } 340 out: 341 thread_unmask_exceptions(exceptions); 342 return ret; 343 } 344 345 bool thread_is_from_abort_mode(void) 346 { 347 struct thread_core_local *l = thread_get_core_local(); 348 349 return (l->flags >> THREAD_CLF_SAVED_SHIFT) & THREAD_CLF_ABORT; 350 } 351 352 /* 353 * This function should always be accurate, but it might be possible to 354 * implement a more efficient depending on cpu architecture. 355 */ 356 bool __weak thread_is_in_normal_mode(void) 357 { 358 uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR); 359 struct thread_core_local *l = thread_get_core_local(); 360 bool ret; 361 362 /* 363 * If any bit in l->flags is set aside from THREAD_CLF_TMP we're 364 * handling some exception. 365 */ 366 ret = (l->curr_thread != THREAD_ID_INVALID) && 367 !(l->flags & ~THREAD_CLF_TMP); 368 thread_unmask_exceptions(exceptions); 369 370 return ret; 371 } 372 373 short int thread_get_id_may_fail(void) 374 { 375 /* 376 * thread_get_core_local() requires foreign interrupts to be disabled 377 */ 378 uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR); 379 struct thread_core_local *l = thread_get_core_local(); 380 short int ct = l->curr_thread; 381 382 thread_unmask_exceptions(exceptions); 383 return ct; 384 } 385 386 short int thread_get_id(void) 387 { 388 short int ct = thread_get_id_may_fail(); 389 390 /* Thread ID has to fit in a short int */ 391 COMPILE_TIME_ASSERT(CFG_NUM_THREADS <= SHRT_MAX); 392 assert(ct >= 0 && ct < CFG_NUM_THREADS); 393 return ct; 394 } 395 396 #ifdef CFG_WITH_PAGER 397 static void init_thread_stacks(void) 398 { 399 size_t n = 0; 400 401 /* 402 * Allocate virtual memory for thread stacks. 403 */ 404 for (n = 0; n < CFG_NUM_THREADS; n++) { 405 tee_mm_entry_t *mm = NULL; 406 vaddr_t sp = 0; 407 size_t num_pages = 0; 408 struct fobj *fobj = NULL; 409 410 /* Find vmem for thread stack and its protection gap */ 411 mm = tee_mm_alloc(&tee_mm_vcore, 412 SMALL_PAGE_SIZE + STACK_THREAD_SIZE); 413 assert(mm); 414 415 /* Claim eventual physical page */ 416 tee_pager_add_pages(tee_mm_get_smem(mm), tee_mm_get_size(mm), 417 true); 418 419 num_pages = tee_mm_get_bytes(mm) / SMALL_PAGE_SIZE - 1; 420 fobj = fobj_locked_paged_alloc(num_pages); 421 422 /* Add the region to the pager */ 423 tee_pager_add_core_region(tee_mm_get_smem(mm) + SMALL_PAGE_SIZE, 424 PAGED_REGION_TYPE_LOCK, fobj); 425 fobj_put(fobj); 426 427 /* init effective stack */ 428 sp = tee_mm_get_smem(mm) + tee_mm_get_bytes(mm); 429 asan_tag_access((void *)tee_mm_get_smem(mm), (void *)sp); 430 if (!thread_init_stack(n, sp)) 431 panic("init stack failed"); 432 } 433 } 434 #else 435 static void init_thread_stacks(void) 436 { 437 size_t n; 438 439 /* Assign the thread stacks */ 440 for (n = 0; n < CFG_NUM_THREADS; n++) { 441 if (!thread_init_stack(n, GET_STACK_BOTTOM(stack_thread, n))) 442 panic("thread_init_stack failed"); 443 } 444 } 445 #endif /*CFG_WITH_PAGER*/ 446 447 void thread_init_threads(void) 448 { 449 size_t n = 0; 450 451 init_thread_stacks(); 452 print_stack_limits(); 453 pgt_init(); 454 455 mutex_lockdep_init(); 456 457 for (n = 0; n < CFG_NUM_THREADS; n++) { 458 TAILQ_INIT(&threads[n].tsd.sess_stack); 459 SLIST_INIT(&threads[n].tsd.pgt_cache); 460 } 461 } 462 463 void __nostackcheck thread_init_thread_core_local(void) 464 { 465 size_t n = 0; 466 struct thread_core_local *tcl = thread_core_local; 467 468 for (n = 0; n < CFG_TEE_CORE_NB_CORE; n++) { 469 tcl[n].curr_thread = THREAD_ID_INVALID; 470 tcl[n].flags = THREAD_CLF_TMP; 471 } 472 tcl[0].tmp_stack_va_end = GET_STACK_BOTTOM(stack_tmp, 0); 473 } 474 475 void thread_init_core_local_stacks(void) 476 { 477 size_t n = 0; 478 struct thread_core_local *tcl = thread_core_local; 479 480 for (n = 0; n < CFG_TEE_CORE_NB_CORE; n++) { 481 tcl[n].tmp_stack_va_end = GET_STACK_BOTTOM(stack_tmp, n) - 482 STACK_TMP_OFFS; 483 tcl[n].abt_stack_va_end = GET_STACK_BOTTOM(stack_abt, n); 484 } 485 } 486 487 struct thread_specific_data *thread_get_tsd(void) 488 { 489 return &threads[thread_get_id()].tsd; 490 } 491 492 struct thread_ctx_regs * __nostackcheck thread_get_ctx_regs(void) 493 { 494 struct thread_core_local *l = thread_get_core_local(); 495 496 assert(l->curr_thread != THREAD_ID_INVALID); 497 return &threads[l->curr_thread].regs; 498 } 499 500 void thread_set_foreign_intr(bool enable) 501 { 502 /* thread_get_core_local() requires foreign interrupts to be disabled */ 503 uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR); 504 struct thread_core_local *l; 505 506 l = thread_get_core_local(); 507 508 assert(l->curr_thread != THREAD_ID_INVALID); 509 510 if (enable) { 511 threads[l->curr_thread].flags |= 512 THREAD_FLAGS_FOREIGN_INTR_ENABLE; 513 thread_set_exceptions(exceptions & ~THREAD_EXCP_FOREIGN_INTR); 514 } else { 515 /* 516 * No need to disable foreign interrupts here since they're 517 * already disabled above. 518 */ 519 threads[l->curr_thread].flags &= 520 ~THREAD_FLAGS_FOREIGN_INTR_ENABLE; 521 } 522 } 523 524 void thread_restore_foreign_intr(void) 525 { 526 /* thread_get_core_local() requires foreign interrupts to be disabled */ 527 uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR); 528 struct thread_core_local *l; 529 530 l = thread_get_core_local(); 531 532 assert(l->curr_thread != THREAD_ID_INVALID); 533 534 if (threads[l->curr_thread].flags & THREAD_FLAGS_FOREIGN_INTR_ENABLE) 535 thread_set_exceptions(exceptions & ~THREAD_EXCP_FOREIGN_INTR); 536 } 537 538 static struct mobj *alloc_shm(enum thread_shm_type shm_type, size_t size) 539 { 540 switch (shm_type) { 541 case THREAD_SHM_TYPE_APPLICATION: 542 return thread_rpc_alloc_payload(size); 543 case THREAD_SHM_TYPE_KERNEL_PRIVATE: 544 return thread_rpc_alloc_kernel_payload(size); 545 case THREAD_SHM_TYPE_GLOBAL: 546 return thread_rpc_alloc_global_payload(size); 547 default: 548 return NULL; 549 } 550 } 551 552 static void clear_shm_cache_entry(struct thread_shm_cache_entry *ce) 553 { 554 if (ce->mobj) { 555 switch (ce->type) { 556 case THREAD_SHM_TYPE_APPLICATION: 557 thread_rpc_free_payload(ce->mobj); 558 break; 559 case THREAD_SHM_TYPE_KERNEL_PRIVATE: 560 thread_rpc_free_kernel_payload(ce->mobj); 561 break; 562 case THREAD_SHM_TYPE_GLOBAL: 563 thread_rpc_free_global_payload(ce->mobj); 564 break; 565 default: 566 assert(0); /* "can't happen" */ 567 break; 568 } 569 } 570 ce->mobj = NULL; 571 ce->size = 0; 572 } 573 574 static struct thread_shm_cache_entry * 575 get_shm_cache_entry(enum thread_shm_cache_user user) 576 { 577 struct thread_shm_cache *cache = &threads[thread_get_id()].shm_cache; 578 struct thread_shm_cache_entry *ce = NULL; 579 580 SLIST_FOREACH(ce, cache, link) 581 if (ce->user == user) 582 return ce; 583 584 ce = calloc(1, sizeof(*ce)); 585 if (ce) { 586 ce->user = user; 587 SLIST_INSERT_HEAD(cache, ce, link); 588 } 589 590 return ce; 591 } 592 593 void *thread_rpc_shm_cache_alloc(enum thread_shm_cache_user user, 594 enum thread_shm_type shm_type, 595 size_t size, struct mobj **mobj) 596 { 597 struct thread_shm_cache_entry *ce = NULL; 598 size_t sz = size; 599 paddr_t p = 0; 600 void *va = NULL; 601 602 if (!size) 603 return NULL; 604 605 ce = get_shm_cache_entry(user); 606 if (!ce) 607 return NULL; 608 609 /* 610 * Always allocate in page chunks as normal world allocates payload 611 * memory as complete pages. 612 */ 613 sz = ROUNDUP(size, SMALL_PAGE_SIZE); 614 615 if (ce->type != shm_type || sz > ce->size) { 616 clear_shm_cache_entry(ce); 617 618 ce->mobj = alloc_shm(shm_type, sz); 619 if (!ce->mobj) 620 return NULL; 621 622 if (mobj_get_pa(ce->mobj, 0, 0, &p)) 623 goto err; 624 625 if (!IS_ALIGNED_WITH_TYPE(p, uint64_t)) 626 goto err; 627 628 va = mobj_get_va(ce->mobj, 0, sz); 629 if (!va) 630 goto err; 631 632 ce->size = sz; 633 ce->type = shm_type; 634 } else { 635 va = mobj_get_va(ce->mobj, 0, sz); 636 if (!va) 637 goto err; 638 } 639 *mobj = ce->mobj; 640 641 return va; 642 err: 643 clear_shm_cache_entry(ce); 644 return NULL; 645 } 646 647 void thread_rpc_shm_cache_clear(struct thread_shm_cache *cache) 648 { 649 while (true) { 650 struct thread_shm_cache_entry *ce = SLIST_FIRST(cache); 651 652 if (!ce) 653 break; 654 SLIST_REMOVE_HEAD(cache, link); 655 clear_shm_cache_entry(ce); 656 free(ce); 657 } 658 } 659