xref: /optee_os/core/arch/arm/kernel/thread.c (revision b166fabf3e8c35de38ccc77a870b58941f73cb83)
1 // SPDX-License-Identifier: BSD-2-Clause
2 /*
3  * Copyright (c) 2016, Linaro Limited
4  * Copyright (c) 2014, STMicroelectronics International N.V.
5  */
6 
7 #include <platform_config.h>
8 
9 #include <arm.h>
10 #include <assert.h>
11 #include <io.h>
12 #include <keep.h>
13 #include <kernel/asan.h>
14 #include <kernel/linker.h>
15 #include <kernel/lockdep.h>
16 #include <kernel/misc.h>
17 #include <kernel/panic.h>
18 #include <kernel/spinlock.h>
19 #include <kernel/tee_ta_manager.h>
20 #include <kernel/thread_defs.h>
21 #include <kernel/thread.h>
22 #include <kernel/virtualization.h>
23 #include <mm/core_memprot.h>
24 #include <mm/mobj.h>
25 #include <mm/tee_mm.h>
26 #include <mm/tee_mmu.h>
27 #include <mm/tee_pager.h>
28 #include <smccc.h>
29 #include <sm/sm.h>
30 #include <trace.h>
31 #include <util.h>
32 
33 #include "thread_private.h"
34 
35 #ifdef CFG_WITH_ARM_TRUSTED_FW
36 #define STACK_TMP_OFFS		0
37 #else
38 #define STACK_TMP_OFFS		SM_STACK_TMP_RESERVE_SIZE
39 #endif
40 
41 
42 #ifdef ARM32
43 #ifdef CFG_CORE_SANITIZE_KADDRESS
44 #define STACK_TMP_SIZE		(3072 + STACK_TMP_OFFS)
45 #else
46 #define STACK_TMP_SIZE		(2048 + STACK_TMP_OFFS)
47 #endif
48 #define STACK_THREAD_SIZE	8192
49 
50 #if defined(CFG_CORE_SANITIZE_KADDRESS) || defined(__clang__)
51 #define STACK_ABT_SIZE		3072
52 #else
53 #define STACK_ABT_SIZE		2048
54 #endif
55 
56 #endif /*ARM32*/
57 
58 #ifdef ARM64
59 #if defined(__clang__) && !defined(CFG_CC_OPTIMIZE_FOR_SIZE)
60 #define STACK_TMP_SIZE		(4096 + STACK_TMP_OFFS)
61 #else
62 #define STACK_TMP_SIZE		(2048 + STACK_TMP_OFFS)
63 #endif
64 #define STACK_THREAD_SIZE	8192
65 
66 #if TRACE_LEVEL > 0
67 #define STACK_ABT_SIZE		3072
68 #else
69 #define STACK_ABT_SIZE		1024
70 #endif
71 #endif /*ARM64*/
72 
73 struct thread_ctx threads[CFG_NUM_THREADS];
74 
75 struct thread_core_local thread_core_local[CFG_TEE_CORE_NB_CORE] __nex_bss;
76 
77 #ifdef CFG_WITH_STACK_CANARIES
78 #ifdef ARM32
79 #define STACK_CANARY_SIZE	(4 * sizeof(uint32_t))
80 #endif
81 #ifdef ARM64
82 #define STACK_CANARY_SIZE	(8 * sizeof(uint32_t))
83 #endif
84 #define START_CANARY_VALUE	0xdededede
85 #define END_CANARY_VALUE	0xabababab
86 #define GET_START_CANARY(name, stack_num) name[stack_num][0]
87 #define GET_END_CANARY(name, stack_num) \
88 	name[stack_num][sizeof(name[stack_num]) / sizeof(uint32_t) - 1]
89 #else
90 #define STACK_CANARY_SIZE	0
91 #endif
92 
93 #define DECLARE_STACK(name, num_stacks, stack_size, linkage) \
94 linkage uint32_t name[num_stacks] \
95 		[ROUNDUP(stack_size + STACK_CANARY_SIZE, STACK_ALIGNMENT) / \
96 		sizeof(uint32_t)] \
97 		__attribute__((section(".nozi_stack." # name), \
98 			       aligned(STACK_ALIGNMENT)))
99 
100 #define STACK_SIZE(stack) (sizeof(stack) - STACK_CANARY_SIZE / 2)
101 
102 #define GET_STACK(stack) \
103 	((vaddr_t)(stack) + STACK_SIZE(stack))
104 
105 DECLARE_STACK(stack_tmp, CFG_TEE_CORE_NB_CORE, STACK_TMP_SIZE, static);
106 DECLARE_STACK(stack_abt, CFG_TEE_CORE_NB_CORE, STACK_ABT_SIZE, static);
107 #ifndef CFG_WITH_PAGER
108 DECLARE_STACK(stack_thread, CFG_NUM_THREADS, STACK_THREAD_SIZE, static);
109 #endif
110 
111 const void *stack_tmp_export __section(".identity_map.stack_tmp_export") =
112 	(uint8_t *)stack_tmp + sizeof(stack_tmp[0]) -
113 	(STACK_TMP_OFFS + STACK_CANARY_SIZE / 2);
114 const uint32_t stack_tmp_stride __section(".identity_map.stack_tmp_stride") =
115 	sizeof(stack_tmp[0]);
116 
117 /*
118  * These stack setup info are required by secondary boot cores before they
119  * each locally enable the pager (the mmu). Hence kept in pager sections.
120  */
121 DECLARE_KEEP_PAGER(stack_tmp_export);
122 DECLARE_KEEP_PAGER(stack_tmp_stride);
123 
124 thread_pm_handler_t thread_cpu_on_handler_ptr __nex_bss;
125 thread_pm_handler_t thread_cpu_off_handler_ptr __nex_bss;
126 thread_pm_handler_t thread_cpu_suspend_handler_ptr __nex_bss;
127 thread_pm_handler_t thread_cpu_resume_handler_ptr __nex_bss;
128 thread_pm_handler_t thread_system_off_handler_ptr __nex_bss;
129 thread_pm_handler_t thread_system_reset_handler_ptr __nex_bss;
130 
131 #ifdef CFG_CORE_UNMAP_CORE_AT_EL0
132 static vaddr_t thread_user_kcode_va __nex_bss;
133 long thread_user_kcode_offset __nex_bss;
134 static size_t thread_user_kcode_size __nex_bss;
135 #endif
136 
137 #if defined(CFG_CORE_UNMAP_CORE_AT_EL0) && \
138 	defined(CFG_CORE_WORKAROUND_SPECTRE_BP_SEC) && defined(ARM64)
139 long thread_user_kdata_sp_offset __nex_bss;
140 static uint8_t thread_user_kdata_page[
141 	ROUNDUP(sizeof(thread_core_local), SMALL_PAGE_SIZE)]
142 	__aligned(SMALL_PAGE_SIZE)
143 #ifndef CFG_VIRTUALIZATION
144 	__section(".nozi.kdata_page");
145 #else
146 	__section(".nex_nozi.kdata_page");
147 #endif
148 #endif
149 
150 static unsigned int thread_global_lock __nex_bss = SPINLOCK_UNLOCK;
151 
152 static void init_canaries(void)
153 {
154 #ifdef CFG_WITH_STACK_CANARIES
155 	size_t n;
156 #define INIT_CANARY(name)						\
157 	for (n = 0; n < ARRAY_SIZE(name); n++) {			\
158 		uint32_t *start_canary = &GET_START_CANARY(name, n);	\
159 		uint32_t *end_canary = &GET_END_CANARY(name, n);	\
160 									\
161 		*start_canary = START_CANARY_VALUE;			\
162 		*end_canary = END_CANARY_VALUE;				\
163 		DMSG("#Stack canaries for %s[%zu] with top at %p",	\
164 			#name, n, (void *)(end_canary - 1));		\
165 		DMSG("watch *%p", (void *)end_canary);			\
166 	}
167 
168 	INIT_CANARY(stack_tmp);
169 	INIT_CANARY(stack_abt);
170 #if !defined(CFG_WITH_PAGER) && !defined(CFG_VIRTUALIZATION)
171 	INIT_CANARY(stack_thread);
172 #endif
173 #endif/*CFG_WITH_STACK_CANARIES*/
174 }
175 
176 #define CANARY_DIED(stack, loc, n) \
177 	do { \
178 		EMSG_RAW("Dead canary at %s of '%s[%zu]'", #loc, #stack, n); \
179 		panic(); \
180 	} while (0)
181 
182 void thread_check_canaries(void)
183 {
184 #ifdef CFG_WITH_STACK_CANARIES
185 	size_t n;
186 
187 	for (n = 0; n < ARRAY_SIZE(stack_tmp); n++) {
188 		if (GET_START_CANARY(stack_tmp, n) != START_CANARY_VALUE)
189 			CANARY_DIED(stack_tmp, start, n);
190 		if (GET_END_CANARY(stack_tmp, n) != END_CANARY_VALUE)
191 			CANARY_DIED(stack_tmp, end, n);
192 	}
193 
194 	for (n = 0; n < ARRAY_SIZE(stack_abt); n++) {
195 		if (GET_START_CANARY(stack_abt, n) != START_CANARY_VALUE)
196 			CANARY_DIED(stack_abt, start, n);
197 		if (GET_END_CANARY(stack_abt, n) != END_CANARY_VALUE)
198 			CANARY_DIED(stack_abt, end, n);
199 
200 	}
201 #if !defined(CFG_WITH_PAGER) && !defined(CFG_VIRTUALIZATION)
202 	for (n = 0; n < ARRAY_SIZE(stack_thread); n++) {
203 		if (GET_START_CANARY(stack_thread, n) != START_CANARY_VALUE)
204 			CANARY_DIED(stack_thread, start, n);
205 		if (GET_END_CANARY(stack_thread, n) != END_CANARY_VALUE)
206 			CANARY_DIED(stack_thread, end, n);
207 	}
208 #endif
209 #endif/*CFG_WITH_STACK_CANARIES*/
210 }
211 
212 void thread_lock_global(void)
213 {
214 	cpu_spin_lock(&thread_global_lock);
215 }
216 
217 void thread_unlock_global(void)
218 {
219 	cpu_spin_unlock(&thread_global_lock);
220 }
221 
222 #ifdef ARM32
223 uint32_t thread_get_exceptions(void)
224 {
225 	uint32_t cpsr = read_cpsr();
226 
227 	return (cpsr >> CPSR_F_SHIFT) & THREAD_EXCP_ALL;
228 }
229 
230 void thread_set_exceptions(uint32_t exceptions)
231 {
232 	uint32_t cpsr = read_cpsr();
233 
234 	/* Foreign interrupts must not be unmasked while holding a spinlock */
235 	if (!(exceptions & THREAD_EXCP_FOREIGN_INTR))
236 		assert_have_no_spinlock();
237 
238 	cpsr &= ~(THREAD_EXCP_ALL << CPSR_F_SHIFT);
239 	cpsr |= ((exceptions & THREAD_EXCP_ALL) << CPSR_F_SHIFT);
240 	write_cpsr(cpsr);
241 }
242 #endif /*ARM32*/
243 
244 #ifdef ARM64
245 uint32_t thread_get_exceptions(void)
246 {
247 	uint32_t daif = read_daif();
248 
249 	return (daif >> DAIF_F_SHIFT) & THREAD_EXCP_ALL;
250 }
251 
252 void thread_set_exceptions(uint32_t exceptions)
253 {
254 	uint32_t daif = read_daif();
255 
256 	/* Foreign interrupts must not be unmasked while holding a spinlock */
257 	if (!(exceptions & THREAD_EXCP_FOREIGN_INTR))
258 		assert_have_no_spinlock();
259 
260 	daif &= ~(THREAD_EXCP_ALL << DAIF_F_SHIFT);
261 	daif |= ((exceptions & THREAD_EXCP_ALL) << DAIF_F_SHIFT);
262 	write_daif(daif);
263 }
264 #endif /*ARM64*/
265 
266 uint32_t thread_mask_exceptions(uint32_t exceptions)
267 {
268 	uint32_t state = thread_get_exceptions();
269 
270 	thread_set_exceptions(state | (exceptions & THREAD_EXCP_ALL));
271 	return state;
272 }
273 
274 void thread_unmask_exceptions(uint32_t state)
275 {
276 	thread_set_exceptions(state & THREAD_EXCP_ALL);
277 }
278 
279 
280 struct thread_core_local *thread_get_core_local(void)
281 {
282 	uint32_t cpu_id = get_core_pos();
283 
284 	/*
285 	 * Foreign interrupts must be disabled before playing with core_local
286 	 * since we otherwise may be rescheduled to a different core in the
287 	 * middle of this function.
288 	 */
289 	assert(thread_get_exceptions() & THREAD_EXCP_FOREIGN_INTR);
290 
291 	assert(cpu_id < CFG_TEE_CORE_NB_CORE);
292 	return &thread_core_local[cpu_id];
293 }
294 
295 static void thread_lazy_save_ns_vfp(void)
296 {
297 #ifdef CFG_WITH_VFP
298 	struct thread_ctx *thr = threads + thread_get_id();
299 
300 	thr->vfp_state.ns_saved = false;
301 	vfp_lazy_save_state_init(&thr->vfp_state.ns);
302 #endif /*CFG_WITH_VFP*/
303 }
304 
305 static void thread_lazy_restore_ns_vfp(void)
306 {
307 #ifdef CFG_WITH_VFP
308 	struct thread_ctx *thr = threads + thread_get_id();
309 	struct thread_user_vfp_state *tuv = thr->vfp_state.uvfp;
310 
311 	assert(!thr->vfp_state.sec_lazy_saved && !thr->vfp_state.sec_saved);
312 
313 	if (tuv && tuv->lazy_saved && !tuv->saved) {
314 		vfp_lazy_save_state_final(&tuv->vfp, false /*!force_save*/);
315 		tuv->saved = true;
316 	}
317 
318 	vfp_lazy_restore_state(&thr->vfp_state.ns, thr->vfp_state.ns_saved);
319 	thr->vfp_state.ns_saved = false;
320 #endif /*CFG_WITH_VFP*/
321 }
322 
323 #ifdef ARM32
324 static void init_regs(struct thread_ctx *thread, uint32_t a0, uint32_t a1,
325 		      uint32_t a2, uint32_t a3)
326 {
327 	thread->regs.pc = (uint32_t)thread_std_smc_entry;
328 
329 	/*
330 	 * Stdcalls starts in SVC mode with masked foreign interrupts, masked
331 	 * Asynchronous abort and unmasked native interrupts.
332 	 */
333 	thread->regs.cpsr = read_cpsr() & ARM32_CPSR_E;
334 	thread->regs.cpsr |= CPSR_MODE_SVC | CPSR_A |
335 			(THREAD_EXCP_FOREIGN_INTR << ARM32_CPSR_F_SHIFT);
336 	/* Enable thumb mode if it's a thumb instruction */
337 	if (thread->regs.pc & 1)
338 		thread->regs.cpsr |= CPSR_T;
339 	/* Reinitialize stack pointer */
340 	thread->regs.svc_sp = thread->stack_va_end;
341 
342 	/*
343 	 * Copy arguments into context. This will make the
344 	 * arguments appear in r0-r7 when thread is started.
345 	 */
346 	thread->regs.r0 = a0;
347 	thread->regs.r1 = a1;
348 	thread->regs.r2 = a2;
349 	thread->regs.r3 = a3;
350 	thread->regs.r4 = 0;
351 	thread->regs.r5 = 0;
352 	thread->regs.r6 = 0;
353 	thread->regs.r7 = 0;
354 }
355 #endif /*ARM32*/
356 
357 #ifdef ARM64
358 static void init_regs(struct thread_ctx *thread, uint32_t a0, uint32_t a1,
359 		      uint32_t a2, uint32_t a3)
360 {
361 	thread->regs.pc = (uint64_t)thread_std_smc_entry;
362 
363 	/*
364 	 * Stdcalls starts in SVC mode with masked foreign interrupts, masked
365 	 * Asynchronous abort and unmasked native interrupts.
366 	 */
367 	thread->regs.cpsr = SPSR_64(SPSR_64_MODE_EL1, SPSR_64_MODE_SP_EL0,
368 				THREAD_EXCP_FOREIGN_INTR | DAIFBIT_ABT);
369 	/* Reinitialize stack pointer */
370 	thread->regs.sp = thread->stack_va_end;
371 
372 	/*
373 	 * Copy arguments into context. This will make the
374 	 * arguments appear in x0-x7 when thread is started.
375 	 */
376 	thread->regs.x[0] = a0;
377 	thread->regs.x[1] = a1;
378 	thread->regs.x[2] = a2;
379 	thread->regs.x[3] = a3;
380 	thread->regs.x[4] = 0;
381 	thread->regs.x[5] = 0;
382 	thread->regs.x[6] = 0;
383 	thread->regs.x[7] = 0;
384 
385 	/* Set up frame pointer as per the Aarch64 AAPCS */
386 	thread->regs.x[29] = 0;
387 }
388 #endif /*ARM64*/
389 
390 void thread_init_boot_thread(void)
391 {
392 	struct thread_core_local *l = thread_get_core_local();
393 
394 	thread_init_threads();
395 
396 	l->curr_thread = 0;
397 	threads[0].state = THREAD_STATE_ACTIVE;
398 }
399 
400 void thread_clr_boot_thread(void)
401 {
402 	struct thread_core_local *l = thread_get_core_local();
403 
404 	assert(l->curr_thread >= 0 && l->curr_thread < CFG_NUM_THREADS);
405 	assert(threads[l->curr_thread].state == THREAD_STATE_ACTIVE);
406 	threads[l->curr_thread].state = THREAD_STATE_FREE;
407 	l->curr_thread = -1;
408 }
409 
410 void thread_alloc_and_run(uint32_t a0, uint32_t a1, uint32_t a2, uint32_t a3)
411 {
412 	size_t n;
413 	struct thread_core_local *l = thread_get_core_local();
414 	bool found_thread = false;
415 
416 	assert(l->curr_thread == -1);
417 
418 	thread_lock_global();
419 
420 	for (n = 0; n < CFG_NUM_THREADS; n++) {
421 		if (threads[n].state == THREAD_STATE_FREE) {
422 			threads[n].state = THREAD_STATE_ACTIVE;
423 			found_thread = true;
424 			break;
425 		}
426 	}
427 
428 	thread_unlock_global();
429 
430 	if (!found_thread)
431 		return;
432 
433 	l->curr_thread = n;
434 
435 	threads[n].flags = 0;
436 	init_regs(threads + n, a0, a1, a2, a3);
437 
438 	thread_lazy_save_ns_vfp();
439 	thread_resume(&threads[n].regs);
440 	/*NOTREACHED*/
441 	panic();
442 }
443 
444 #ifdef ARM32
445 static void copy_a0_to_a3(struct thread_ctx_regs *regs, uint32_t a0,
446 			  uint32_t a1, uint32_t a2, uint32_t a3)
447 {
448 	/*
449 	 * Update returned values from RPC, values will appear in
450 	 * r0-r3 when thread is resumed.
451 	 */
452 	regs->r0 = a0;
453 	regs->r1 = a1;
454 	regs->r2 = a2;
455 	regs->r3 = a3;
456 }
457 #endif /*ARM32*/
458 
459 #ifdef ARM64
460 static void copy_a0_to_a3(struct thread_ctx_regs *regs, uint32_t a0,
461 			  uint32_t a1, uint32_t a2, uint32_t a3)
462 {
463 	/*
464 	 * Update returned values from RPC, values will appear in
465 	 * x0-x3 when thread is resumed.
466 	 */
467 	regs->x[0] = a0;
468 	regs->x[1] = a1;
469 	regs->x[2] = a2;
470 	regs->x[3] = a3;
471 }
472 #endif /*ARM64*/
473 
474 #ifdef ARM32
475 static bool is_from_user(uint32_t cpsr)
476 {
477 	return (cpsr & ARM32_CPSR_MODE_MASK) == ARM32_CPSR_MODE_USR;
478 }
479 #endif
480 
481 #ifdef ARM64
482 static bool is_from_user(uint32_t cpsr)
483 {
484 	if (cpsr & (SPSR_MODE_RW_32 << SPSR_MODE_RW_SHIFT))
485 		return true;
486 	if (((cpsr >> SPSR_64_MODE_EL_SHIFT) & SPSR_64_MODE_EL_MASK) ==
487 	     SPSR_64_MODE_EL0)
488 		return true;
489 	return false;
490 }
491 #endif
492 
493 #ifdef CFG_SYSCALL_FTRACE
494 static void __noprof ftrace_suspend(void)
495 {
496 	struct tee_ta_session *s = TAILQ_FIRST(&thread_get_tsd()->sess_stack);
497 
498 	if (!s)
499 		return;
500 
501 	if (s->fbuf)
502 		s->fbuf->syscall_trace_suspended = true;
503 }
504 
505 static void __noprof ftrace_resume(void)
506 {
507 	struct tee_ta_session *s = TAILQ_FIRST(&thread_get_tsd()->sess_stack);
508 
509 	if (!s)
510 		return;
511 
512 	if (s->fbuf)
513 		s->fbuf->syscall_trace_suspended = false;
514 }
515 #else
516 static void __noprof ftrace_suspend(void)
517 {
518 }
519 
520 static void __noprof ftrace_resume(void)
521 {
522 }
523 #endif
524 
525 static bool is_user_mode(struct thread_ctx_regs *regs)
526 {
527 	return is_from_user((uint32_t)regs->cpsr);
528 }
529 
530 void thread_resume_from_rpc(uint32_t thread_id, uint32_t a0, uint32_t a1,
531 			    uint32_t a2, uint32_t a3)
532 {
533 	size_t n = thread_id;
534 	struct thread_core_local *l = thread_get_core_local();
535 	bool found_thread = false;
536 
537 	assert(l->curr_thread == -1);
538 
539 	thread_lock_global();
540 
541 	if (n < CFG_NUM_THREADS && threads[n].state == THREAD_STATE_SUSPENDED) {
542 		threads[n].state = THREAD_STATE_ACTIVE;
543 		found_thread = true;
544 	}
545 
546 	thread_unlock_global();
547 
548 	if (!found_thread)
549 		return;
550 
551 	l->curr_thread = n;
552 
553 	if (threads[n].have_user_map) {
554 		core_mmu_set_user_map(&threads[n].user_map);
555 		if (threads[n].flags & THREAD_FLAGS_EXIT_ON_FOREIGN_INTR)
556 			tee_ta_ftrace_update_times_resume();
557 	}
558 
559 	if (is_user_mode(&threads[n].regs))
560 		tee_ta_update_session_utime_resume();
561 
562 	/*
563 	 * Return from RPC to request service of a foreign interrupt must not
564 	 * get parameters from non-secure world.
565 	 */
566 	if (threads[n].flags & THREAD_FLAGS_COPY_ARGS_ON_RETURN) {
567 		copy_a0_to_a3(&threads[n].regs, a0, a1, a2, a3);
568 		threads[n].flags &= ~THREAD_FLAGS_COPY_ARGS_ON_RETURN;
569 	}
570 
571 	thread_lazy_save_ns_vfp();
572 
573 	if (threads[n].have_user_map)
574 		ftrace_resume();
575 
576 	thread_resume(&threads[n].regs);
577 	/*NOTREACHED*/
578 	panic();
579 }
580 
581 void *thread_get_tmp_sp(void)
582 {
583 	struct thread_core_local *l = thread_get_core_local();
584 
585 	return (void *)l->tmp_stack_va_end;
586 }
587 
588 #ifdef ARM64
589 vaddr_t thread_get_saved_thread_sp(void)
590 {
591 	struct thread_core_local *l = thread_get_core_local();
592 	int ct = l->curr_thread;
593 
594 	assert(ct != -1);
595 	return threads[ct].kern_sp;
596 }
597 #endif /*ARM64*/
598 
599 vaddr_t thread_stack_start(void)
600 {
601 	struct thread_ctx *thr;
602 	int ct = thread_get_id_may_fail();
603 
604 	if (ct == -1)
605 		return 0;
606 
607 	thr = threads + ct;
608 	return thr->stack_va_end - STACK_THREAD_SIZE;
609 }
610 
611 size_t thread_stack_size(void)
612 {
613 	return STACK_THREAD_SIZE;
614 }
615 
616 bool thread_is_from_abort_mode(void)
617 {
618 	struct thread_core_local *l = thread_get_core_local();
619 
620 	return (l->flags >> THREAD_CLF_SAVED_SHIFT) & THREAD_CLF_ABORT;
621 }
622 
623 #ifdef ARM32
624 bool thread_is_in_normal_mode(void)
625 {
626 	return (read_cpsr() & ARM32_CPSR_MODE_MASK) == ARM32_CPSR_MODE_SVC;
627 }
628 #endif
629 
630 #ifdef ARM64
631 bool thread_is_in_normal_mode(void)
632 {
633 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
634 	struct thread_core_local *l = thread_get_core_local();
635 	bool ret;
636 
637 	/* If any bit in l->flags is set we're handling some exception. */
638 	ret = !l->flags;
639 	thread_unmask_exceptions(exceptions);
640 
641 	return ret;
642 }
643 #endif
644 
645 void thread_state_free(void)
646 {
647 	struct thread_core_local *l = thread_get_core_local();
648 	int ct = l->curr_thread;
649 
650 	assert(ct != -1);
651 
652 	thread_lazy_restore_ns_vfp();
653 	tee_pager_release_phys(
654 		(void *)(threads[ct].stack_va_end - STACK_THREAD_SIZE),
655 		STACK_THREAD_SIZE);
656 
657 	thread_lock_global();
658 
659 	assert(threads[ct].state == THREAD_STATE_ACTIVE);
660 	threads[ct].state = THREAD_STATE_FREE;
661 	threads[ct].flags = 0;
662 	l->curr_thread = -1;
663 
664 #ifdef CFG_VIRTUALIZATION
665 	virt_unset_guest();
666 #endif
667 	thread_unlock_global();
668 }
669 
670 #ifdef CFG_WITH_PAGER
671 static void release_unused_kernel_stack(struct thread_ctx *thr,
672 					uint32_t cpsr __maybe_unused)
673 {
674 #ifdef ARM64
675 	/*
676 	 * If we're from user mode then thr->regs.sp is the saved user
677 	 * stack pointer and thr->kern_sp holds the last kernel stack
678 	 * pointer. But if we're from kernel mode then thr->kern_sp isn't
679 	 * up to date so we need to read from thr->regs.sp instead.
680 	 */
681 	vaddr_t sp = is_from_user(cpsr) ?  thr->kern_sp : thr->regs.sp;
682 #else
683 	vaddr_t sp = thr->regs.svc_sp;
684 #endif
685 	vaddr_t base = thr->stack_va_end - STACK_THREAD_SIZE;
686 	size_t len = sp - base;
687 
688 	tee_pager_release_phys((void *)base, len);
689 }
690 #else
691 static void release_unused_kernel_stack(struct thread_ctx *thr __unused,
692 					uint32_t cpsr __unused)
693 {
694 }
695 #endif
696 
697 int thread_state_suspend(uint32_t flags, uint32_t cpsr, vaddr_t pc)
698 {
699 	struct thread_core_local *l = thread_get_core_local();
700 	int ct = l->curr_thread;
701 
702 	assert(ct != -1);
703 
704 	if (core_mmu_user_mapping_is_active())
705 		ftrace_suspend();
706 
707 	thread_check_canaries();
708 
709 	release_unused_kernel_stack(threads + ct, cpsr);
710 
711 	if (is_from_user(cpsr)) {
712 		thread_user_save_vfp();
713 		tee_ta_update_session_utime_suspend();
714 		tee_ta_gprof_sample_pc(pc);
715 	}
716 	thread_lazy_restore_ns_vfp();
717 
718 	thread_lock_global();
719 
720 	assert(threads[ct].state == THREAD_STATE_ACTIVE);
721 	threads[ct].flags |= flags;
722 	threads[ct].regs.cpsr = cpsr;
723 	threads[ct].regs.pc = pc;
724 	threads[ct].state = THREAD_STATE_SUSPENDED;
725 
726 	threads[ct].have_user_map = core_mmu_user_mapping_is_active();
727 	if (threads[ct].have_user_map) {
728 		if (threads[ct].flags & THREAD_FLAGS_EXIT_ON_FOREIGN_INTR)
729 			tee_ta_ftrace_update_times_suspend();
730 		core_mmu_get_user_map(&threads[ct].user_map);
731 		core_mmu_set_user_map(NULL);
732 	}
733 
734 	l->curr_thread = -1;
735 
736 #ifdef CFG_VIRTUALIZATION
737 	virt_unset_guest();
738 #endif
739 
740 	thread_unlock_global();
741 
742 	return ct;
743 }
744 
745 #ifdef ARM32
746 static void set_tmp_stack(struct thread_core_local *l, vaddr_t sp)
747 {
748 	l->tmp_stack_va_end = sp;
749 	thread_set_irq_sp(sp);
750 	thread_set_fiq_sp(sp);
751 }
752 
753 static void set_abt_stack(struct thread_core_local *l, vaddr_t sp)
754 {
755 	l->abt_stack_va_end = sp;
756 	thread_set_abt_sp((vaddr_t)l);
757 	thread_set_und_sp((vaddr_t)l);
758 }
759 #endif /*ARM32*/
760 
761 #ifdef ARM64
762 static void set_tmp_stack(struct thread_core_local *l, vaddr_t sp)
763 {
764 	/*
765 	 * We're already using the tmp stack when this function is called
766 	 * so there's no need to assign it to any stack pointer. However,
767 	 * we'll need to restore it at different times so store it here.
768 	 */
769 	l->tmp_stack_va_end = sp;
770 }
771 
772 static void set_abt_stack(struct thread_core_local *l, vaddr_t sp)
773 {
774 	l->abt_stack_va_end = sp;
775 }
776 #endif /*ARM64*/
777 
778 bool thread_init_stack(uint32_t thread_id, vaddr_t sp)
779 {
780 	if (thread_id >= CFG_NUM_THREADS)
781 		return false;
782 	threads[thread_id].stack_va_end = sp;
783 	return true;
784 }
785 
786 int thread_get_id_may_fail(void)
787 {
788 	/*
789 	 * thread_get_core_local() requires foreign interrupts to be disabled
790 	 */
791 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
792 	struct thread_core_local *l = thread_get_core_local();
793 	int ct = l->curr_thread;
794 
795 	thread_unmask_exceptions(exceptions);
796 	return ct;
797 }
798 
799 int thread_get_id(void)
800 {
801 	int ct = thread_get_id_may_fail();
802 
803 	assert(ct >= 0 && ct < CFG_NUM_THREADS);
804 	return ct;
805 }
806 
807 static void init_handlers(const struct thread_handlers *handlers)
808 {
809 	thread_cpu_on_handler_ptr = handlers->cpu_on;
810 	thread_cpu_off_handler_ptr = handlers->cpu_off;
811 	thread_cpu_suspend_handler_ptr = handlers->cpu_suspend;
812 	thread_cpu_resume_handler_ptr = handlers->cpu_resume;
813 	thread_system_off_handler_ptr = handlers->system_off;
814 	thread_system_reset_handler_ptr = handlers->system_reset;
815 }
816 
817 #ifdef CFG_WITH_PAGER
818 static void init_thread_stacks(void)
819 {
820 	size_t n = 0;
821 
822 	/*
823 	 * Allocate virtual memory for thread stacks.
824 	 */
825 	for (n = 0; n < CFG_NUM_THREADS; n++) {
826 		tee_mm_entry_t *mm = NULL;
827 		vaddr_t sp = 0;
828 		size_t num_pages = 0;
829 		struct fobj *fobj = NULL;
830 
831 		/* Find vmem for thread stack and its protection gap */
832 		mm = tee_mm_alloc(&tee_mm_vcore,
833 				  SMALL_PAGE_SIZE + STACK_THREAD_SIZE);
834 		assert(mm);
835 
836 		/* Claim eventual physical page */
837 		tee_pager_add_pages(tee_mm_get_smem(mm), tee_mm_get_size(mm),
838 				    true);
839 
840 		num_pages = tee_mm_get_bytes(mm) / SMALL_PAGE_SIZE - 1;
841 		fobj = fobj_locked_paged_alloc(num_pages);
842 
843 		/* Add the area to the pager */
844 		tee_pager_add_core_area(tee_mm_get_smem(mm) + SMALL_PAGE_SIZE,
845 					PAGER_AREA_TYPE_LOCK, fobj);
846 		fobj_put(fobj);
847 
848 		/* init effective stack */
849 		sp = tee_mm_get_smem(mm) + tee_mm_get_bytes(mm);
850 		asan_tag_access((void *)tee_mm_get_smem(mm), (void *)sp);
851 		if (!thread_init_stack(n, sp))
852 			panic("init stack failed");
853 	}
854 }
855 #else
856 static void init_thread_stacks(void)
857 {
858 	size_t n;
859 
860 	/* Assign the thread stacks */
861 	for (n = 0; n < CFG_NUM_THREADS; n++) {
862 		if (!thread_init_stack(n, GET_STACK(stack_thread[n])))
863 			panic("thread_init_stack failed");
864 	}
865 }
866 #endif /*CFG_WITH_PAGER*/
867 
868 static void init_user_kcode(void)
869 {
870 #ifdef CFG_CORE_UNMAP_CORE_AT_EL0
871 	vaddr_t v = (vaddr_t)thread_excp_vect;
872 	vaddr_t ve = (vaddr_t)thread_excp_vect_end;
873 
874 	thread_user_kcode_va = ROUNDDOWN(v, CORE_MMU_USER_CODE_SIZE);
875 	ve = ROUNDUP(ve, CORE_MMU_USER_CODE_SIZE);
876 	thread_user_kcode_size = ve - thread_user_kcode_va;
877 
878 	core_mmu_get_user_va_range(&v, NULL);
879 	thread_user_kcode_offset = thread_user_kcode_va - v;
880 
881 #if defined(CFG_CORE_WORKAROUND_SPECTRE_BP_SEC) && defined(ARM64)
882 	/*
883 	 * When transitioning to EL0 subtract SP with this much to point to
884 	 * this special kdata page instead. SP is restored by add this much
885 	 * while transitioning back to EL1.
886 	 */
887 	v += thread_user_kcode_size;
888 	thread_user_kdata_sp_offset = (vaddr_t)thread_core_local - v;
889 #endif
890 #endif /*CFG_CORE_UNMAP_CORE_AT_EL0*/
891 }
892 
893 void thread_init_threads(void)
894 {
895 	size_t n = 0;
896 
897 	init_thread_stacks();
898 	pgt_init();
899 
900 	mutex_lockdep_init();
901 
902 	for (n = 0; n < CFG_NUM_THREADS; n++) {
903 		TAILQ_INIT(&threads[n].tsd.sess_stack);
904 		SLIST_INIT(&threads[n].tsd.pgt_cache);
905 	}
906 }
907 
908 void thread_clr_thread_core_local(void)
909 {
910 	size_t n = 0;
911 
912 	for (n = 0; n < CFG_TEE_CORE_NB_CORE; n++)
913 		thread_core_local[n].curr_thread = -1;
914 }
915 
916 void thread_init_primary(const struct thread_handlers *handlers)
917 {
918 	init_handlers(handlers);
919 
920 	/* Initialize canaries around the stacks */
921 	init_canaries();
922 
923 	init_user_kcode();
924 }
925 
926 static void init_sec_mon(size_t pos __maybe_unused)
927 {
928 #if !defined(CFG_WITH_ARM_TRUSTED_FW)
929 	/* Initialize secure monitor */
930 	sm_init(GET_STACK(stack_tmp[pos]));
931 #endif
932 }
933 
934 static uint32_t __maybe_unused get_midr_implementer(uint32_t midr)
935 {
936 	return (midr >> MIDR_IMPLEMENTER_SHIFT) & MIDR_IMPLEMENTER_MASK;
937 }
938 
939 static uint32_t __maybe_unused get_midr_primary_part(uint32_t midr)
940 {
941 	return (midr >> MIDR_PRIMARY_PART_NUM_SHIFT) &
942 	       MIDR_PRIMARY_PART_NUM_MASK;
943 }
944 
945 #ifdef ARM64
946 static bool probe_workaround_available(void)
947 {
948 	int32_t r;
949 
950 	r = thread_smc(SMCCC_VERSION, 0, 0, 0);
951 	if (r < 0)
952 		return false;
953 	if (r < 0x10001)	/* compare with version 1.1 */
954 		return false;
955 
956 	/* Version >= 1.1, so SMCCC_ARCH_FEATURES is available */
957 	r = thread_smc(SMCCC_ARCH_FEATURES, SMCCC_ARCH_WORKAROUND_1, 0, 0);
958 	return r >= 0;
959 }
960 
961 static vaddr_t __maybe_unused select_vector(vaddr_t a)
962 {
963 	if (probe_workaround_available()) {
964 		DMSG("SMCCC_ARCH_WORKAROUND_1 (%#08" PRIx32 ") available",
965 		     SMCCC_ARCH_WORKAROUND_1);
966 		DMSG("SMC Workaround for CVE-2017-5715 used");
967 		return a;
968 	}
969 
970 	DMSG("SMCCC_ARCH_WORKAROUND_1 (%#08" PRIx32 ") unavailable",
971 	     SMCCC_ARCH_WORKAROUND_1);
972 	DMSG("SMC Workaround for CVE-2017-5715 not needed (if ARM-TF is up to date)");
973 	return (vaddr_t)thread_excp_vect;
974 }
975 #else
976 static vaddr_t __maybe_unused select_vector(vaddr_t a)
977 {
978 	return a;
979 }
980 #endif
981 
982 static vaddr_t get_excp_vect(void)
983 {
984 #ifdef CFG_CORE_WORKAROUND_SPECTRE_BP_SEC
985 	uint32_t midr = read_midr();
986 
987 	if (get_midr_implementer(midr) != MIDR_IMPLEMENTER_ARM)
988 		return (vaddr_t)thread_excp_vect;
989 
990 	switch (get_midr_primary_part(midr)) {
991 #ifdef ARM32
992 	case CORTEX_A8_PART_NUM:
993 	case CORTEX_A9_PART_NUM:
994 	case CORTEX_A17_PART_NUM:
995 #endif
996 	case CORTEX_A57_PART_NUM:
997 	case CORTEX_A72_PART_NUM:
998 	case CORTEX_A73_PART_NUM:
999 	case CORTEX_A75_PART_NUM:
1000 		return select_vector((vaddr_t)thread_excp_vect_workaround);
1001 #ifdef ARM32
1002 	case CORTEX_A15_PART_NUM:
1003 		return select_vector((vaddr_t)thread_excp_vect_workaround_a15);
1004 #endif
1005 	default:
1006 		return (vaddr_t)thread_excp_vect;
1007 	}
1008 #endif /*CFG_CORE_WORKAROUND_SPECTRE_BP_SEC*/
1009 
1010 	return (vaddr_t)thread_excp_vect;
1011 }
1012 
1013 void thread_init_per_cpu(void)
1014 {
1015 	size_t pos = get_core_pos();
1016 	struct thread_core_local *l = thread_get_core_local();
1017 
1018 	init_sec_mon(pos);
1019 
1020 	set_tmp_stack(l, GET_STACK(stack_tmp[pos]) - STACK_TMP_OFFS);
1021 	set_abt_stack(l, GET_STACK(stack_abt[pos]));
1022 
1023 	thread_init_vbar(get_excp_vect());
1024 
1025 #ifdef CFG_FTRACE_SUPPORT
1026 	/*
1027 	 * Enable accesses to frequency register and physical counter
1028 	 * register in EL0/PL0 required for timestamping during
1029 	 * function tracing.
1030 	 */
1031 	write_cntkctl(read_cntkctl() | CNTKCTL_PL0PCTEN);
1032 #endif
1033 }
1034 
1035 struct thread_specific_data *thread_get_tsd(void)
1036 {
1037 	return &threads[thread_get_id()].tsd;
1038 }
1039 
1040 struct thread_ctx_regs *thread_get_ctx_regs(void)
1041 {
1042 	struct thread_core_local *l = thread_get_core_local();
1043 
1044 	assert(l->curr_thread != -1);
1045 	return &threads[l->curr_thread].regs;
1046 }
1047 
1048 void thread_set_foreign_intr(bool enable)
1049 {
1050 	/* thread_get_core_local() requires foreign interrupts to be disabled */
1051 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
1052 	struct thread_core_local *l;
1053 
1054 	l = thread_get_core_local();
1055 
1056 	assert(l->curr_thread != -1);
1057 
1058 	if (enable) {
1059 		threads[l->curr_thread].flags |=
1060 					THREAD_FLAGS_FOREIGN_INTR_ENABLE;
1061 		thread_set_exceptions(exceptions & ~THREAD_EXCP_FOREIGN_INTR);
1062 	} else {
1063 		/*
1064 		 * No need to disable foreign interrupts here since they're
1065 		 * already disabled above.
1066 		 */
1067 		threads[l->curr_thread].flags &=
1068 					~THREAD_FLAGS_FOREIGN_INTR_ENABLE;
1069 	}
1070 }
1071 
1072 void thread_restore_foreign_intr(void)
1073 {
1074 	/* thread_get_core_local() requires foreign interrupts to be disabled */
1075 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
1076 	struct thread_core_local *l;
1077 
1078 	l = thread_get_core_local();
1079 
1080 	assert(l->curr_thread != -1);
1081 
1082 	if (threads[l->curr_thread].flags & THREAD_FLAGS_FOREIGN_INTR_ENABLE)
1083 		thread_set_exceptions(exceptions & ~THREAD_EXCP_FOREIGN_INTR);
1084 }
1085 
1086 #ifdef CFG_WITH_VFP
1087 uint32_t thread_kernel_enable_vfp(void)
1088 {
1089 	uint32_t exceptions = thread_mask_exceptions(THREAD_EXCP_FOREIGN_INTR);
1090 	struct thread_ctx *thr = threads + thread_get_id();
1091 	struct thread_user_vfp_state *tuv = thr->vfp_state.uvfp;
1092 
1093 	assert(!vfp_is_enabled());
1094 
1095 	if (!thr->vfp_state.ns_saved) {
1096 		vfp_lazy_save_state_final(&thr->vfp_state.ns,
1097 					  true /*force_save*/);
1098 		thr->vfp_state.ns_saved = true;
1099 	} else if (thr->vfp_state.sec_lazy_saved &&
1100 		   !thr->vfp_state.sec_saved) {
1101 		/*
1102 		 * This happens when we're handling an abort while the
1103 		 * thread was using the VFP state.
1104 		 */
1105 		vfp_lazy_save_state_final(&thr->vfp_state.sec,
1106 					  false /*!force_save*/);
1107 		thr->vfp_state.sec_saved = true;
1108 	} else if (tuv && tuv->lazy_saved && !tuv->saved) {
1109 		/*
1110 		 * This can happen either during syscall or abort
1111 		 * processing (while processing a syscall).
1112 		 */
1113 		vfp_lazy_save_state_final(&tuv->vfp, false /*!force_save*/);
1114 		tuv->saved = true;
1115 	}
1116 
1117 	vfp_enable();
1118 	return exceptions;
1119 }
1120 
1121 void thread_kernel_disable_vfp(uint32_t state)
1122 {
1123 	uint32_t exceptions;
1124 
1125 	assert(vfp_is_enabled());
1126 
1127 	vfp_disable();
1128 	exceptions = thread_get_exceptions();
1129 	assert(exceptions & THREAD_EXCP_FOREIGN_INTR);
1130 	exceptions &= ~THREAD_EXCP_FOREIGN_INTR;
1131 	exceptions |= state & THREAD_EXCP_FOREIGN_INTR;
1132 	thread_set_exceptions(exceptions);
1133 }
1134 
1135 void thread_kernel_save_vfp(void)
1136 {
1137 	struct thread_ctx *thr = threads + thread_get_id();
1138 
1139 	assert(thread_get_exceptions() & THREAD_EXCP_FOREIGN_INTR);
1140 	if (vfp_is_enabled()) {
1141 		vfp_lazy_save_state_init(&thr->vfp_state.sec);
1142 		thr->vfp_state.sec_lazy_saved = true;
1143 	}
1144 }
1145 
1146 void thread_kernel_restore_vfp(void)
1147 {
1148 	struct thread_ctx *thr = threads + thread_get_id();
1149 
1150 	assert(thread_get_exceptions() & THREAD_EXCP_FOREIGN_INTR);
1151 	assert(!vfp_is_enabled());
1152 	if (thr->vfp_state.sec_lazy_saved) {
1153 		vfp_lazy_restore_state(&thr->vfp_state.sec,
1154 				       thr->vfp_state.sec_saved);
1155 		thr->vfp_state.sec_saved = false;
1156 		thr->vfp_state.sec_lazy_saved = false;
1157 	}
1158 }
1159 
1160 void thread_user_enable_vfp(struct thread_user_vfp_state *uvfp)
1161 {
1162 	struct thread_ctx *thr = threads + thread_get_id();
1163 	struct thread_user_vfp_state *tuv = thr->vfp_state.uvfp;
1164 
1165 	assert(thread_get_exceptions() & THREAD_EXCP_FOREIGN_INTR);
1166 	assert(!vfp_is_enabled());
1167 
1168 	if (!thr->vfp_state.ns_saved) {
1169 		vfp_lazy_save_state_final(&thr->vfp_state.ns,
1170 					  true /*force_save*/);
1171 		thr->vfp_state.ns_saved = true;
1172 	} else if (tuv && uvfp != tuv) {
1173 		if (tuv->lazy_saved && !tuv->saved) {
1174 			vfp_lazy_save_state_final(&tuv->vfp,
1175 						  false /*!force_save*/);
1176 			tuv->saved = true;
1177 		}
1178 	}
1179 
1180 	if (uvfp->lazy_saved)
1181 		vfp_lazy_restore_state(&uvfp->vfp, uvfp->saved);
1182 	uvfp->lazy_saved = false;
1183 	uvfp->saved = false;
1184 
1185 	thr->vfp_state.uvfp = uvfp;
1186 	vfp_enable();
1187 }
1188 
1189 void thread_user_save_vfp(void)
1190 {
1191 	struct thread_ctx *thr = threads + thread_get_id();
1192 	struct thread_user_vfp_state *tuv = thr->vfp_state.uvfp;
1193 
1194 	assert(thread_get_exceptions() & THREAD_EXCP_FOREIGN_INTR);
1195 	if (!vfp_is_enabled())
1196 		return;
1197 
1198 	assert(tuv && !tuv->lazy_saved && !tuv->saved);
1199 	vfp_lazy_save_state_init(&tuv->vfp);
1200 	tuv->lazy_saved = true;
1201 }
1202 
1203 void thread_user_clear_vfp(struct thread_user_vfp_state *uvfp)
1204 {
1205 	struct thread_ctx *thr = threads + thread_get_id();
1206 
1207 	if (uvfp == thr->vfp_state.uvfp)
1208 		thr->vfp_state.uvfp = NULL;
1209 	uvfp->lazy_saved = false;
1210 	uvfp->saved = false;
1211 }
1212 #endif /*CFG_WITH_VFP*/
1213 
1214 #ifdef ARM32
1215 static bool get_spsr(bool is_32bit, unsigned long entry_func, uint32_t *spsr)
1216 {
1217 	uint32_t s;
1218 
1219 	if (!is_32bit)
1220 		return false;
1221 
1222 	s = read_cpsr();
1223 	s &= ~(CPSR_MODE_MASK | CPSR_T | CPSR_IT_MASK1 | CPSR_IT_MASK2);
1224 	s |= CPSR_MODE_USR;
1225 	if (entry_func & 1)
1226 		s |= CPSR_T;
1227 	*spsr = s;
1228 	return true;
1229 }
1230 #endif
1231 
1232 #ifdef ARM64
1233 static bool get_spsr(bool is_32bit, unsigned long entry_func, uint32_t *spsr)
1234 {
1235 	uint32_t s;
1236 
1237 	if (is_32bit) {
1238 		s = read_daif() & (SPSR_32_AIF_MASK << SPSR_32_AIF_SHIFT);
1239 		s |= SPSR_MODE_RW_32 << SPSR_MODE_RW_SHIFT;
1240 		s |= (entry_func & SPSR_32_T_MASK) << SPSR_32_T_SHIFT;
1241 	} else {
1242 		s = read_daif() & (SPSR_64_DAIF_MASK << SPSR_64_DAIF_SHIFT);
1243 	}
1244 
1245 	*spsr = s;
1246 	return true;
1247 }
1248 #endif
1249 
1250 static void set_ctx_regs(struct thread_ctx_regs *regs, unsigned long a0,
1251 			 unsigned long a1, unsigned long a2, unsigned long a3,
1252 			 unsigned long user_sp, unsigned long entry_func,
1253 			 uint32_t spsr)
1254 {
1255 	/*
1256 	 * First clear all registers to avoid leaking information from
1257 	 * other TAs or even the Core itself.
1258 	 */
1259 	*regs = (struct thread_ctx_regs){ };
1260 #ifdef ARM32
1261 	regs->r0 = a0;
1262 	regs->r1 = a1;
1263 	regs->r2 = a2;
1264 	regs->r3 = a3;
1265 	regs->usr_sp = user_sp;
1266 	regs->pc = entry_func;
1267 	regs->cpsr = spsr;
1268 #endif
1269 #ifdef ARM64
1270 	regs->x[0] = a0;
1271 	regs->x[1] = a1;
1272 	regs->x[2] = a2;
1273 	regs->x[3] = a3;
1274 	regs->sp = user_sp;
1275 	regs->pc = entry_func;
1276 	regs->cpsr = spsr;
1277 	regs->x[13] = user_sp;	/* Used when running TA in Aarch32 */
1278 	regs->sp = user_sp;	/* Used when running TA in Aarch64 */
1279 	/* Set frame pointer (user stack can't be unwound past this point) */
1280 	regs->x[29] = 0;
1281 #endif
1282 }
1283 
1284 uint32_t thread_enter_user_mode(unsigned long a0, unsigned long a1,
1285 		unsigned long a2, unsigned long a3, unsigned long user_sp,
1286 		unsigned long entry_func, bool is_32bit,
1287 		uint32_t *exit_status0, uint32_t *exit_status1)
1288 {
1289 	uint32_t spsr = 0;
1290 	uint32_t exceptions = 0;
1291 	uint32_t rc = 0;
1292 	struct thread_ctx_regs *regs = NULL;
1293 
1294 	tee_ta_update_session_utime_resume();
1295 
1296 	/* Derive SPSR from current CPSR/PSTATE readout. */
1297 	if (!get_spsr(is_32bit, entry_func, &spsr)) {
1298 		*exit_status0 = 1; /* panic */
1299 		*exit_status1 = 0xbadbadba;
1300 		return 0;
1301 	}
1302 
1303 	exceptions = thread_mask_exceptions(THREAD_EXCP_ALL);
1304 	/*
1305 	 * We're using the per thread location of saved context registers
1306 	 * for temporary storage. Now that exceptions are masked they will
1307 	 * not be used for any thing else until they are eventually
1308 	 * unmasked when user mode has been entered.
1309 	 */
1310 	regs = thread_get_ctx_regs();
1311 	set_ctx_regs(regs, a0, a1, a2, a3, user_sp, entry_func, spsr);
1312 	rc = __thread_enter_user_mode(regs, exit_status0, exit_status1);
1313 	thread_unmask_exceptions(exceptions);
1314 	return rc;
1315 }
1316 
1317 #ifdef CFG_CORE_UNMAP_CORE_AT_EL0
1318 void thread_get_user_kcode(struct mobj **mobj, size_t *offset,
1319 			   vaddr_t *va, size_t *sz)
1320 {
1321 	core_mmu_get_user_va_range(va, NULL);
1322 	*mobj = mobj_tee_ram;
1323 	*offset = thread_user_kcode_va - VCORE_START_VA;
1324 	*sz = thread_user_kcode_size;
1325 }
1326 #endif
1327 
1328 #if defined(CFG_CORE_UNMAP_CORE_AT_EL0) && \
1329 	defined(CFG_CORE_WORKAROUND_SPECTRE_BP_SEC) && defined(ARM64)
1330 void thread_get_user_kdata(struct mobj **mobj, size_t *offset,
1331 			   vaddr_t *va, size_t *sz)
1332 {
1333 	vaddr_t v;
1334 
1335 	core_mmu_get_user_va_range(&v, NULL);
1336 	*va = v + thread_user_kcode_size;
1337 	*mobj = mobj_tee_ram;
1338 	*offset = (vaddr_t)thread_user_kdata_page - VCORE_START_VA;
1339 	*sz = sizeof(thread_user_kdata_page);
1340 }
1341 #endif
1342 
1343 static void setup_unwind_user_mode(struct thread_svc_regs *regs)
1344 {
1345 #ifdef ARM32
1346 	regs->lr = (uintptr_t)thread_unwind_user_mode;
1347 	regs->spsr = read_cpsr();
1348 #endif
1349 #ifdef ARM64
1350 	regs->elr = (uintptr_t)thread_unwind_user_mode;
1351 	regs->spsr = SPSR_64(SPSR_64_MODE_EL1, SPSR_64_MODE_SP_EL0, 0);
1352 	regs->spsr |= read_daif();
1353 	/*
1354 	 * Regs is the value of stack pointer before calling the SVC
1355 	 * handler.  By the addition matches for the reserved space at the
1356 	 * beginning of el0_sync_svc(). This prepares the stack when
1357 	 * returning to thread_unwind_user_mode instead of a normal
1358 	 * exception return.
1359 	 */
1360 	regs->sp_el0 = (uint64_t)(regs + 1);
1361 #endif
1362 }
1363 
1364 /*
1365  * Note: this function is weak just to make it possible to exclude it from
1366  * the unpaged area.
1367  */
1368 void __weak thread_svc_handler(struct thread_svc_regs *regs)
1369 {
1370 	struct tee_ta_session *sess = NULL;
1371 	uint32_t state = 0;
1372 
1373 	/* Enable native interrupts */
1374 	state = thread_get_exceptions();
1375 	thread_unmask_exceptions(state & ~THREAD_EXCP_NATIVE_INTR);
1376 
1377 	thread_user_save_vfp();
1378 
1379 	/* TA has just entered kernel mode */
1380 	tee_ta_update_session_utime_suspend();
1381 
1382 	/* Restore foreign interrupts which are disabled on exception entry */
1383 	thread_restore_foreign_intr();
1384 
1385 	tee_ta_get_current_session(&sess);
1386 	assert(sess && sess->ctx->ops && sess->ctx->ops->handle_svc);
1387 	if (sess->ctx->ops->handle_svc(regs)) {
1388 		/* We're about to switch back to user mode */
1389 		tee_ta_update_session_utime_resume();
1390 	} else {
1391 		/* We're returning from __thread_enter_user_mode() */
1392 		setup_unwind_user_mode(regs);
1393 	}
1394 }
1395