xref: /optee_os/core/kernel/thread.c (revision 59724f223500ec9bf077e8caaa5fc0d8fece39a9)
1 // SPDX-License-Identifier: BSD-2-Clause
2 /*
3  * Copyright (c) 2016-2022, 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 <crypto/crypto.h>
10 #include <kernel/asan.h>
11 #include <kernel/boot.h>
12 #include <kernel/lockdep.h>
13 #include <kernel/misc.h>
14 #include <kernel/panic.h>
15 #include <kernel/spinlock.h>
16 #include <kernel/thread.h>
17 #include <kernel/thread_private.h>
18 #include <mm/mobj.h>
19 #include <mm/page_alloc.h>
20 #include <stdalign.h>
21 
22 struct thread_ctx threads[CFG_NUM_THREADS];
23 
24 #if defined(CFG_DYN_STACK_CONFIG)
25 struct thread_core_local *thread_core_local __nex_bss;
26 size_t thread_core_count __nex_bss;
27 #else
28 static struct thread_core_local
29 	__thread_core_local[CFG_TEE_CORE_NB_CORE] __nex_bss;
30 struct thread_core_local *thread_core_local __nex_data = __thread_core_local;
31 size_t thread_core_count __nex_data = CFG_TEE_CORE_NB_CORE;
32 #endif
33 unsigned long thread_core_local_pa __nex_bss;
34 struct thread_core_local *__thread_core_local_new __nex_bss;
35 size_t __thread_core_count_new __nex_bss;
36 
37 /*
38  * Stacks
39  *
40  * [Lower addresses on the left]
41  *
42  * [ STACK_CANARY_SIZE/2 | STACK_CHECK_EXTRA | STACK_XXX_SIZE | STACK_CANARY_SIZE/2 ]
43  * ^                     ^                   ^                ^
44  * stack_xxx[n]          "hard" top          "soft" top       bottom
45  */
46 
47 static uint32_t start_canary_value = 0xdedede00;
48 static uint32_t end_canary_value = 0xababab00;
49 
50 #define DECLARE_STACK(name, num_stacks, stack_size, linkage) \
51 linkage uint32_t name[num_stacks] \
52 		[ROUNDUP(stack_size + STACK_CANARY_SIZE + STACK_CHECK_EXTRA, \
53 			 STACK_ALIGNMENT) / sizeof(uint32_t)] \
54 		__attribute__((section(".nozi_stack." # name), \
55 			       aligned(STACK_ALIGNMENT)))
56 
57 #ifndef CFG_DYN_STACK_CONFIG
58 DECLARE_STACK(stack_tmp, CFG_TEE_CORE_NB_CORE, STACK_TMP_SIZE,
59 	      /* global linkage */);
60 DECLARE_STACK(stack_abt, CFG_TEE_CORE_NB_CORE, STACK_ABT_SIZE, static);
61 #define GET_STACK_BOTTOM(stack, n) ((vaddr_t)&(stack)[n] + sizeof(stack[n]) - \
62 				    STACK_CANARY_SIZE / 2)
63 #else
64 /* Not used */
65 #define GET_STACK_BOTTOM(stack, n) 0
66 #endif
67 #ifndef CFG_WITH_PAGER
68 DECLARE_STACK(stack_thread, CFG_NUM_THREADS, STACK_THREAD_SIZE, static);
69 #define GET_STACK_THREAD_BOTTOM(n) \
70 	((vaddr_t)&stack_thread[n] +  sizeof(stack_thread[n]) - \
71 	 STACK_CANARY_SIZE / 2)
72 #endif
73 
74 
75 #ifndef CFG_DYN_STACK_CONFIG
76 const uint32_t stack_tmp_stride __section(".identity_map.stack_tmp_stride") =
77 	sizeof(stack_tmp[0]);
78 
79 /*
80  * This stack setup info is required by secondary boot cores before they
81  * each locally enable the pager (the mmu). Hence kept in pager sections.
82  */
83 DECLARE_KEEP_PAGER(stack_tmp_stride);
84 #endif
85 
86 static unsigned int thread_global_lock __nex_bss = SPINLOCK_UNLOCK;
87 
88 static size_t stack_size_to_alloc_size(size_t stack_size)
89 {
90 	return ROUNDUP(stack_size + STACK_CANARY_SIZE + STACK_CHECK_EXTRA,
91 		       STACK_ALIGNMENT);
92 }
93 
94 static vaddr_t stack_end_va_to_top_hard(size_t stack_size, vaddr_t end_va)
95 {
96 	size_t l = stack_size_to_alloc_size(stack_size);
97 
98 	return end_va - l + STACK_CANARY_SIZE;
99 }
100 
101 static vaddr_t stack_end_va_to_top_soft(size_t stack_size, vaddr_t end_va)
102 {
103 	return stack_end_va_to_top_hard(stack_size, end_va) + STACK_CHECK_EXTRA;
104 }
105 
106 static vaddr_t stack_end_va_to_bottom(size_t stack_size __unused,
107 				      vaddr_t end_va)
108 {
109 	return end_va;
110 }
111 
112 static uint32_t *stack_end_va_to_start_canary(size_t stack_size, vaddr_t end_va)
113 {
114 	return (uint32_t *)(stack_end_va_to_top_hard(stack_size, end_va) -
115 			    STACK_CANARY_SIZE / 2);
116 }
117 
118 static uint32_t *stack_end_va_to_end_canary(size_t stack_size __unused,
119 					    vaddr_t end_va)
120 {
121 	return (uint32_t *)(end_va + STACK_CANARY_SIZE / 2 - sizeof(uint32_t));
122 }
123 
124 static void init_canaries(size_t stack_size, vaddr_t va_end)
125 {
126 	uint32_t *canary = NULL;
127 
128 	assert(va_end);
129 	canary = stack_end_va_to_start_canary(stack_size, va_end);
130 	*canary = start_canary_value;
131 	canary = stack_end_va_to_end_canary(stack_size, va_end);
132 	*canary = end_canary_value;
133 }
134 
135 void thread_init_canaries(void)
136 {
137 	vaddr_t va = 0;
138 	size_t n = 0;
139 
140 	if (IS_ENABLED(CFG_WITH_STACK_CANARIES)) {
141 		for (n = 0; n < thread_core_count; n++) {
142 			if (thread_core_local[n].tmp_stack_va_end) {
143 				va = thread_core_local[n].tmp_stack_va_end +
144 				     STACK_TMP_OFFS;
145 				init_canaries(STACK_TMP_SIZE, va);
146 			}
147 			va = thread_core_local[n].abt_stack_va_end;
148 			if (va)
149 				init_canaries(STACK_ABT_SIZE, va);
150 		}
151 
152 	}
153 
154 	if (IS_ENABLED(CFG_WITH_STACK_CANARIES) &&
155 	    !IS_ENABLED(CFG_WITH_PAGER) && !IS_ENABLED(CFG_NS_VIRTUALIZATION)) {
156 		for (n = 0; n < CFG_NUM_THREADS; n++) {
157 			va = threads[n].stack_va_end;
158 			if (va)
159 				init_canaries(STACK_THREAD_SIZE, va);
160 		}
161 	}
162 }
163 
164 #if defined(CFG_WITH_STACK_CANARIES)
165 void thread_update_canaries(void)
166 {
167 	uint32_t canary[2] = { };
168 	uint32_t exceptions = 0;
169 
170 	plat_get_random_stack_canaries(canary, ARRAY_SIZE(canary),
171 				       sizeof(canary[0]));
172 
173 	exceptions = thread_mask_exceptions(THREAD_EXCP_ALL);
174 
175 	thread_check_canaries();
176 
177 	start_canary_value = canary[0];
178 	end_canary_value = canary[1];
179 	thread_init_canaries();
180 
181 	thread_unmask_exceptions(exceptions);
182 }
183 #endif
184 
185 static void check_stack_canary(const char *stack_name __maybe_unused,
186 			       size_t n __maybe_unused,
187 			       size_t stack_size, vaddr_t end_va)
188 {
189 	uint32_t *canary = NULL;
190 
191 	canary = stack_end_va_to_start_canary(stack_size, end_va);
192 	if (*canary != start_canary_value) {
193 		EMSG_RAW("Dead canary at start of '%s[%zu]' (%p)",
194 			 stack_name, n, (void *)canary);
195 		panic();
196 	}
197 
198 	canary = stack_end_va_to_end_canary(stack_size, end_va);
199 	if (*canary != end_canary_value) {
200 		EMSG_RAW("Dead canary at end of '%s[%zu]' (%p)",
201 			 stack_name, n, (void *)canary);
202 		panic();
203 	}
204 }
205 
206 void thread_check_canaries(void)
207 {
208 	vaddr_t va = 0;
209 	size_t n = 0;
210 
211 	if (IS_ENABLED(CFG_WITH_STACK_CANARIES)) {
212 		for (n = 0; n < thread_core_count; n++) {
213 			if (thread_core_local[n].tmp_stack_va_end) {
214 				va = thread_core_local[n].tmp_stack_va_end +
215 				     STACK_TMP_OFFS;
216 				check_stack_canary("tmp_stack", n,
217 						   STACK_TMP_SIZE, va);
218 			}
219 
220 			va = thread_core_local[n].abt_stack_va_end;
221 			if (va)
222 				check_stack_canary("abt_stack", n,
223 						   STACK_ABT_SIZE, va);
224 		}
225 	}
226 
227 	if (IS_ENABLED(CFG_WITH_STACK_CANARIES) &&
228 	    !IS_ENABLED(CFG_WITH_PAGER) && !IS_ENABLED(CFG_NS_VIRTUALIZATION)) {
229 		for (n = 0; n < CFG_NUM_THREADS; n++) {
230 			va = threads[n].stack_va_end;
231 			if (va)
232 				check_stack_canary("thread_stack", n,
233 						   STACK_THREAD_SIZE, va);
234 		}
235 	}
236 }
237 
238 void thread_lock_global(void)
239 {
240 	cpu_spin_lock(&thread_global_lock);
241 }
242 
243 void thread_unlock_global(void)
244 {
245 	cpu_spin_unlock(&thread_global_lock);
246 }
247 
248 static struct thread_core_local * __nostackcheck
249 get_core_local(unsigned int pos)
250 {
251 	/*
252 	 * Foreign interrupts must be disabled before playing with core_local
253 	 * since we otherwise may be rescheduled to a different core in the
254 	 * middle of this function.
255 	 */
256 	assert(thread_get_exceptions() & THREAD_EXCP_FOREIGN_INTR);
257 
258 	/*
259 	 * With CFG_BOOT_INIT_CURRENT_THREAD_CORE_LOCAL, we boot on a
260 	 * single core and have allocated only one struct thread_core_local
261 	 * so we return that regardless of pos.
262 	 */
263 	if (IS_ENABLED(CFG_DYN_STACK_CONFIG) &&
264 	    thread_core_local != __thread_core_local_new)
265 		return thread_core_local;
266 
267 	assert(pos < thread_core_count);
268 	return &thread_core_local[pos];
269 }
270 
271 struct thread_core_local * __nostackcheck thread_get_core_local(void)
272 {
273 	unsigned int pos = get_core_pos();
274 
275 	return get_core_local(pos);
276 }
277 
278 #ifdef CFG_CORE_DEBUG_CHECK_STACKS
279 static void print_stack_limits(void)
280 {
281 	size_t n = 0;
282 	vaddr_t __maybe_unused start = 0;
283 	vaddr_t __maybe_unused end = 0;
284 	vaddr_t va = 0;
285 
286 	for (n = 0; n < thread_core_count; n++) {
287 		va = thread_core_local[n].tmp_stack_va_end + STACK_TMP_OFFS;
288 		start = stack_end_va_to_top_soft(STACK_TMP_SIZE, va);
289 		end = stack_end_va_to_bottom(STACK_TMP_SIZE, va);
290 		DMSG("tmp [%zu] 0x%" PRIxVA "..0x%" PRIxVA, n, start, end);
291 
292 		va = thread_core_local[n].abt_stack_va_end;
293 		start = stack_end_va_to_top_soft(STACK_ABT_SIZE, va);
294 		end = stack_end_va_to_bottom(STACK_ABT_SIZE, va);
295 		DMSG("abt [%zu] 0x%" PRIxVA "..0x%" PRIxVA, n, start, end);
296 	}
297 
298 	for (n = 0; n < CFG_NUM_THREADS; n++) {
299 		va = threads[n].stack_va_end;
300 		start = stack_end_va_to_top_soft(STACK_THREAD_SIZE, va);
301 		end = stack_end_va_to_bottom(STACK_THREAD_SIZE, va);
302 		DMSG("thr [%zu] 0x%" PRIxVA "..0x%" PRIxVA, n, start, end);
303 	}
304 }
305 
306 static void check_stack_limits(void)
307 {
308 	vaddr_t stack_start = 0;
309 	vaddr_t stack_end = 0;
310 	/* Any value in the current stack frame will do */
311 	vaddr_t current_sp = (vaddr_t)&stack_start;
312 
313 	if (!get_stack_soft_limits(&stack_start, &stack_end))
314 		panic("Unknown stack limits");
315 	if (current_sp < stack_start || current_sp > stack_end) {
316 		EMSG("Stack pointer out of range: 0x%" PRIxVA " not in [0x%"
317 		     PRIxVA " .. 0x%" PRIxVA "]", current_sp, stack_start,
318 		     stack_end);
319 		print_stack_limits();
320 		panic();
321 	}
322 }
323 
324 static bool * __nostackcheck get_stackcheck_recursion_flag(void)
325 {
326 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
327 	unsigned int pos = get_core_pos();
328 	struct thread_core_local *l = get_core_local(pos);
329 	int ct = l->curr_thread;
330 	bool *p = NULL;
331 
332 	if (l->flags & (THREAD_CLF_ABORT | THREAD_CLF_TMP))
333 		p = &l->stackcheck_recursion;
334 	else if (!l->flags)
335 		p = &threads[ct].tsd.stackcheck_recursion;
336 
337 	thread_unmask_exceptions(exceptions);
338 	return p;
339 }
340 
341 void __cyg_profile_func_enter(void *this_fn, void *call_site);
342 void __nostackcheck __cyg_profile_func_enter(void *this_fn __unused,
343 					     void *call_site __unused)
344 {
345 	bool *p = get_stackcheck_recursion_flag();
346 
347 	assert(p);
348 	if (*p)
349 		return;
350 	*p = true;
351 	check_stack_limits();
352 	*p = false;
353 }
354 
355 void __cyg_profile_func_exit(void *this_fn, void *call_site);
356 void __nostackcheck __cyg_profile_func_exit(void *this_fn __unused,
357 					    void *call_site __unused)
358 {
359 }
360 #else
361 static void print_stack_limits(void)
362 {
363 }
364 #endif
365 
366 void thread_init_boot_thread(void)
367 {
368 	struct thread_core_local *l = thread_get_core_local();
369 
370 	thread_init_threads();
371 
372 	l->curr_thread = 0;
373 	threads[0].state = THREAD_STATE_ACTIVE;
374 }
375 
376 void __nostackcheck thread_clr_boot_thread(void)
377 {
378 	struct thread_core_local *l = thread_get_core_local();
379 
380 	assert(l->curr_thread >= 0 && l->curr_thread < CFG_NUM_THREADS);
381 	assert(threads[l->curr_thread].state == THREAD_STATE_ACTIVE);
382 	threads[l->curr_thread].state = THREAD_STATE_FREE;
383 	l->curr_thread = THREAD_ID_INVALID;
384 	print_stack_limits();
385 }
386 
387 void __nostackcheck *thread_get_tmp_sp(void)
388 {
389 	struct thread_core_local *l = thread_get_core_local();
390 
391 	/*
392 	 * Called from assembly when switching to the temporary stack, so flags
393 	 * need updating
394 	 */
395 	l->flags |= THREAD_CLF_TMP;
396 
397 	return (void *)l->tmp_stack_va_end;
398 }
399 
400 vaddr_t thread_stack_start(void)
401 {
402 	struct thread_ctx *thr;
403 	int ct = thread_get_id_may_fail();
404 
405 	if (ct == THREAD_ID_INVALID)
406 		return 0;
407 
408 	thr = threads + ct;
409 	return stack_end_va_to_top_soft(STACK_THREAD_SIZE, thr->stack_va_end);
410 }
411 
412 size_t thread_stack_size(void)
413 {
414 	return STACK_THREAD_SIZE;
415 }
416 
417 bool get_stack_limits(vaddr_t *start, vaddr_t *end, bool hard)
418 {
419 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
420 	unsigned int pos = get_core_pos();
421 	struct thread_core_local *l = get_core_local(pos);
422 	int ct = l->curr_thread;
423 	size_t stack_size = 0;
424 	bool ret = true;
425 	vaddr_t va = 0;
426 
427 	if (l->flags & THREAD_CLF_TMP) {
428 		va = l->tmp_stack_va_end + STACK_TMP_OFFS;
429 		stack_size = STACK_TMP_SIZE;
430 	} else if (l->flags & THREAD_CLF_ABORT) {
431 		va = l->abt_stack_va_end;
432 		stack_size = STACK_ABT_SIZE;
433 	} else if (!l->flags && ct >= 0 && ct < CFG_NUM_THREADS) {
434 		va = threads[ct].stack_va_end;
435 		stack_size = STACK_THREAD_SIZE;
436 	} else {
437 		ret = false;
438 		goto out;
439 	}
440 
441 	*end = stack_end_va_to_bottom(stack_size, va);
442 	if (hard)
443 		*start = stack_end_va_to_top_hard(stack_size, va);
444 	else
445 		*start = stack_end_va_to_top_soft(stack_size, va);
446 out:
447 	thread_unmask_exceptions(exceptions);
448 	return ret;
449 }
450 
451 bool thread_is_from_abort_mode(void)
452 {
453 	struct thread_core_local *l = thread_get_core_local();
454 
455 	return (l->flags >> THREAD_CLF_SAVED_SHIFT) & THREAD_CLF_ABORT;
456 }
457 
458 /*
459  * This function should always be accurate, but it might be possible to
460  * implement a more efficient depending on cpu architecture.
461  */
462 bool __weak __noprof thread_is_in_normal_mode(void)
463 {
464 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
465 	struct thread_core_local *l = thread_get_core_local();
466 	bool ret;
467 
468 	/*
469 	 * If any bit in l->flags is set aside from THREAD_CLF_TMP we're
470 	 * handling some exception.
471 	 */
472 	ret = (l->curr_thread != THREAD_ID_INVALID) &&
473 	      !(l->flags & ~THREAD_CLF_TMP);
474 	thread_unmask_exceptions(exceptions);
475 
476 	return ret;
477 }
478 
479 short int __noprof thread_get_id_may_fail(void)
480 {
481 	/*
482 	 * thread_get_core_local() requires foreign interrupts to be disabled
483 	 */
484 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
485 	struct thread_core_local *l = thread_get_core_local();
486 	short int ct = l->curr_thread;
487 
488 	thread_unmask_exceptions(exceptions);
489 	return ct;
490 }
491 
492 short int __noprof thread_get_id(void)
493 {
494 	short int ct = thread_get_id_may_fail();
495 
496 	/* Thread ID has to fit in a short int */
497 	COMPILE_TIME_ASSERT(CFG_NUM_THREADS <= SHRT_MAX);
498 	assert(ct >= 0 && ct < CFG_NUM_THREADS);
499 	return ct;
500 }
501 
502 #ifdef CFG_WITH_PAGER
503 static void init_thread_stacks(void)
504 {
505 	size_t n = 0;
506 
507 	/*
508 	 * Allocate virtual memory for thread stacks.
509 	 */
510 	for (n = 0; n < CFG_NUM_THREADS; n++) {
511 		tee_mm_entry_t *mm = NULL;
512 		vaddr_t sp = 0;
513 		size_t num_pages = 0;
514 		struct fobj *fobj = NULL;
515 
516 		/* Find vmem for thread stack and its protection gap */
517 		mm = tee_mm_alloc(&core_virt_mem_pool,
518 				  SMALL_PAGE_SIZE + STACK_THREAD_SIZE);
519 		assert(mm);
520 
521 		/* Claim eventual physical page */
522 		tee_pager_add_pages(tee_mm_get_smem(mm), tee_mm_get_size(mm),
523 				    true);
524 
525 		num_pages = tee_mm_get_bytes(mm) / SMALL_PAGE_SIZE - 1;
526 		fobj = fobj_locked_paged_alloc(num_pages);
527 
528 		/* Add the region to the pager */
529 		tee_pager_add_core_region(tee_mm_get_smem(mm) + SMALL_PAGE_SIZE,
530 					  PAGED_REGION_TYPE_LOCK, fobj);
531 		fobj_put(fobj);
532 
533 		/* init effective stack */
534 		sp = tee_mm_get_smem(mm) + tee_mm_get_bytes(mm);
535 		asan_tag_access((void *)tee_mm_get_smem(mm), (void *)sp);
536 		threads[n].stack_va_end = sp;
537 	}
538 }
539 #else
540 static void init_thread_stacks(void)
541 {
542 	vaddr_t va = 0;
543 	size_t n = 0;
544 
545 	/* Assign the thread stacks */
546 	for (n = 0; n < CFG_NUM_THREADS; n++) {
547 		va = GET_STACK_THREAD_BOTTOM(n);
548 		threads[n].stack_va_end = va;
549 		if (IS_ENABLED(CFG_WITH_STACK_CANARIES))
550 			init_canaries(STACK_THREAD_SIZE, va);
551 	}
552 }
553 #endif /*CFG_WITH_PAGER*/
554 
555 void thread_init_threads(void)
556 {
557 	size_t n = 0;
558 
559 	init_thread_stacks();
560 	print_stack_limits();
561 	pgt_init();
562 
563 	mutex_lockdep_init();
564 
565 	for (n = 0; n < CFG_NUM_THREADS; n++)
566 		TAILQ_INIT(&threads[n].tsd.sess_stack);
567 }
568 
569 #ifndef CFG_DYN_STACK_CONFIG
570 vaddr_t __nostackcheck thread_get_abt_stack(void)
571 {
572 	return GET_STACK_BOTTOM(stack_abt, get_core_pos());
573 }
574 #endif
575 
576 #ifdef CFG_BOOT_INIT_CURRENT_THREAD_CORE_LOCAL
577 static vaddr_t alloc_stack(size_t stack_size, bool nex)
578 {
579 	size_t l = stack_size_to_alloc_size(stack_size);
580 	size_t rl = ROUNDUP(l, SMALL_PAGE_SIZE);
581 	uint32_t flags = MAF_GUARD_HEAD;
582 	vaddr_t end_va = 0;
583 	vaddr_t va = 0;
584 
585 	if (nex)
586 		flags |= MAF_NEX;
587 	va = virt_page_alloc(rl / SMALL_PAGE_SIZE, flags);
588 	if (!va)
589 		panic();
590 
591 	end_va = va + l - STACK_CANARY_SIZE / 2;
592 	if (IS_ENABLED(CFG_WITH_STACK_CANARIES))
593 		init_canaries(stack_size, end_va);
594 
595 	return end_va;
596 }
597 
598 void thread_init_thread_core_local(size_t core_count)
599 {
600 	struct thread_core_local *tcl = NULL;
601 	const size_t core_pos = get_core_pos();
602 	vaddr_t va = 0;
603 	size_t n = 0;
604 
605 	if (IS_ENABLED(CFG_DYN_STACK_CONFIG)) {
606 		assert(core_count <= CFG_TEE_CORE_NB_CORE);
607 		tcl = nex_calloc(core_count, sizeof(*tcl));
608 		if (!tcl)
609 			panic();
610 		__thread_core_local_new = tcl;
611 		__thread_core_count_new = core_count;
612 	} else {
613 		tcl = thread_core_local;
614 		assert(core_count == CFG_TEE_CORE_NB_CORE);
615 
616 		for (n = 0; n < thread_core_count; n++) {
617 			init_canaries(STACK_TMP_SIZE,
618 				      GET_STACK_BOTTOM(stack_tmp, n));
619 			init_canaries(STACK_ABT_SIZE,
620 				      GET_STACK_BOTTOM(stack_abt, n));
621 		}
622 	}
623 
624 	for (n = 0; n < core_count; n++) {
625 		if (n == core_pos) {
626 			if (IS_ENABLED(CFG_DYN_STACK_CONFIG))
627 				tcl[n] = thread_core_local[0];
628 			else
629 				continue;
630 		} else {
631 			tcl[n].curr_thread = THREAD_ID_INVALID;
632 			tcl[n].flags = THREAD_CLF_TMP;
633 		}
634 
635 		if (IS_ENABLED(CFG_DYN_STACK_CONFIG))
636 			va = alloc_stack(STACK_TMP_SIZE, true);
637 		else
638 			va = GET_STACK_BOTTOM(stack_tmp, n);
639 		tcl[n].tmp_stack_va_end = va - STACK_TMP_OFFS;
640 #ifdef ARM32
641 		tcl[n].tmp_stack_pa_end =
642 			vaddr_to_phys(tcl[n].tmp_stack_va_end);
643 #endif
644 
645 		if (IS_ENABLED(CFG_DYN_STACK_CONFIG))
646 			va = alloc_stack(STACK_ABT_SIZE, true);
647 		else
648 			va = GET_STACK_BOTTOM(stack_abt, n);
649 		tcl[n].abt_stack_va_end = va;
650 	}
651 }
652 #else
653 void __nostackcheck
654 thread_init_thread_core_local(size_t core_count __maybe_unused)
655 {
656 	size_t n = 0;
657 	struct thread_core_local *tcl = thread_core_local;
658 
659 	assert(core_count == CFG_TEE_CORE_NB_CORE);
660 	for (n = 0; n < CFG_TEE_CORE_NB_CORE; n++) {
661 		tcl[n].curr_thread = THREAD_ID_INVALID;
662 		tcl[n].flags = THREAD_CLF_TMP;
663 	}
664 	tcl[0].tmp_stack_va_end = GET_STACK_BOTTOM(stack_tmp, 0);
665 }
666 
667 void __nostackcheck thread_init_core_local_stacks(void)
668 {
669 	size_t n = 0;
670 	struct thread_core_local *tcl = thread_core_local;
671 
672 	for (n = 0; n < CFG_TEE_CORE_NB_CORE; n++) {
673 		tcl[n].tmp_stack_va_end = GET_STACK_BOTTOM(stack_tmp, n) -
674 					  STACK_TMP_OFFS;
675 		tcl[n].abt_stack_va_end = GET_STACK_BOTTOM(stack_abt, n);
676 	}
677 }
678 #endif /*CFG_BOOT_INIT_CURRENT_THREAD_CORE_LOCAL*/
679 
680 #if defined(CFG_CORE_PAUTH)
681 void thread_init_thread_pauth_keys(void)
682 {
683 	size_t n = 0;
684 
685 	for (n = 0; n < CFG_NUM_THREADS; n++)
686 		if (crypto_rng_read(&threads[n].keys, sizeof(threads[n].keys)))
687 			panic("Failed to init thread pauth keys");
688 }
689 
690 void thread_init_core_local_pauth_keys(void)
691 {
692 	struct thread_core_local *tcl = thread_core_local;
693 	size_t n = 0;
694 
695 	for (n = 0; n < thread_core_count; n++)
696 		if (crypto_rng_read(&tcl[n].keys, sizeof(tcl[n].keys)))
697 			panic("Failed to init core local pauth keys");
698 }
699 #endif
700 
701 struct thread_specific_data * __noprof thread_get_tsd(void)
702 {
703 	return &threads[thread_get_id()].tsd;
704 }
705 
706 struct thread_ctx_regs * __nostackcheck thread_get_ctx_regs(void)
707 {
708 	struct thread_core_local *l = thread_get_core_local();
709 
710 	assert(l->curr_thread != THREAD_ID_INVALID);
711 	return &threads[l->curr_thread].regs;
712 }
713 
714 void thread_set_foreign_intr(bool enable)
715 {
716 	/* thread_get_core_local() requires foreign interrupts to be disabled */
717 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
718 	struct thread_core_local *l;
719 
720 	l = thread_get_core_local();
721 
722 	assert(l->curr_thread != THREAD_ID_INVALID);
723 
724 	if (enable) {
725 		threads[l->curr_thread].flags |=
726 					THREAD_FLAGS_FOREIGN_INTR_ENABLE;
727 		thread_set_exceptions(exceptions & ~THREAD_EXCP_FOREIGN_INTR);
728 	} else {
729 		/*
730 		 * No need to disable foreign interrupts here since they're
731 		 * already disabled above.
732 		 */
733 		threads[l->curr_thread].flags &=
734 					~THREAD_FLAGS_FOREIGN_INTR_ENABLE;
735 	}
736 }
737 
738 void thread_restore_foreign_intr(void)
739 {
740 	/* thread_get_core_local() requires foreign interrupts to be disabled */
741 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
742 	struct thread_core_local *l;
743 
744 	l = thread_get_core_local();
745 
746 	assert(l->curr_thread != THREAD_ID_INVALID);
747 
748 	if (threads[l->curr_thread].flags & THREAD_FLAGS_FOREIGN_INTR_ENABLE)
749 		thread_set_exceptions(exceptions & ~THREAD_EXCP_FOREIGN_INTR);
750 }
751 
752 static struct mobj *alloc_shm(enum thread_shm_type shm_type, size_t size)
753 {
754 	switch (shm_type) {
755 	case THREAD_SHM_TYPE_APPLICATION:
756 		return thread_rpc_alloc_payload(size);
757 	case THREAD_SHM_TYPE_KERNEL_PRIVATE:
758 		return thread_rpc_alloc_kernel_payload(size);
759 	case THREAD_SHM_TYPE_GLOBAL:
760 		return thread_rpc_alloc_global_payload(size);
761 	default:
762 		return NULL;
763 	}
764 }
765 
766 static void clear_shm_cache_entry(struct thread_shm_cache_entry *ce)
767 {
768 	if (ce->mobj) {
769 		switch (ce->type) {
770 		case THREAD_SHM_TYPE_APPLICATION:
771 			thread_rpc_free_payload(ce->mobj);
772 			break;
773 		case THREAD_SHM_TYPE_KERNEL_PRIVATE:
774 			thread_rpc_free_kernel_payload(ce->mobj);
775 			break;
776 		case THREAD_SHM_TYPE_GLOBAL:
777 			thread_rpc_free_global_payload(ce->mobj);
778 			break;
779 		default:
780 			assert(0); /* "can't happen" */
781 			break;
782 		}
783 	}
784 	ce->mobj = NULL;
785 	ce->size = 0;
786 }
787 
788 static struct thread_shm_cache_entry *
789 get_shm_cache_entry(enum thread_shm_cache_user user)
790 {
791 	struct thread_shm_cache *cache = &threads[thread_get_id()].shm_cache;
792 	struct thread_shm_cache_entry *ce = NULL;
793 
794 	SLIST_FOREACH(ce, cache, link)
795 		if (ce->user == user)
796 			return ce;
797 
798 	ce = calloc(1, sizeof(*ce));
799 	if (ce) {
800 		ce->user = user;
801 		SLIST_INSERT_HEAD(cache, ce, link);
802 	}
803 
804 	return ce;
805 }
806 
807 void *thread_rpc_shm_cache_alloc(enum thread_shm_cache_user user,
808 				 enum thread_shm_type shm_type,
809 				 size_t size, struct mobj **mobj)
810 {
811 	struct thread_shm_cache_entry *ce = NULL;
812 	size_t sz = size;
813 	paddr_t p = 0;
814 	void *va = NULL;
815 
816 	if (!size)
817 		return NULL;
818 
819 	ce = get_shm_cache_entry(user);
820 	if (!ce)
821 		return NULL;
822 
823 	/*
824 	 * Always allocate in page chunks as normal world allocates payload
825 	 * memory as complete pages.
826 	 */
827 	sz = ROUNDUP(size, SMALL_PAGE_SIZE);
828 
829 	if (ce->type != shm_type || sz > ce->size) {
830 		clear_shm_cache_entry(ce);
831 
832 		ce->mobj = alloc_shm(shm_type, sz);
833 		if (!ce->mobj)
834 			return NULL;
835 
836 		if (mobj_get_pa(ce->mobj, 0, 0, &p))
837 			goto err;
838 
839 		if (!IS_ALIGNED_WITH_TYPE(p, uint64_t))
840 			goto err;
841 
842 		va = mobj_get_va(ce->mobj, 0, sz);
843 		if (!va)
844 			goto err;
845 
846 		ce->size = sz;
847 		ce->type = shm_type;
848 	} else {
849 		va = mobj_get_va(ce->mobj, 0, sz);
850 		if (!va)
851 			goto err;
852 	}
853 	*mobj = ce->mobj;
854 
855 	return va;
856 err:
857 	clear_shm_cache_entry(ce);
858 	return NULL;
859 }
860 
861 void thread_rpc_shm_cache_clear(struct thread_shm_cache *cache)
862 {
863 	while (true) {
864 		struct thread_shm_cache_entry *ce = SLIST_FIRST(cache);
865 
866 		if (!ce)
867 			break;
868 		SLIST_REMOVE_HEAD(cache, link);
869 		clear_shm_cache_entry(ce);
870 		free(ce);
871 	}
872 }
873