xref: /optee_os/core/arch/arm/kernel/entry_a64.S (revision 65401337ca064f6c2866b6881e3d93edbf7f2921)
1/* SPDX-License-Identifier: BSD-2-Clause */
2/*
3 * Copyright (c) 2015, Linaro Limited
4 */
5
6#include <platform_config.h>
7
8#include <arm64_macros.S>
9#include <arm.h>
10#include <asm.S>
11#include <generated/asm-defines.h>
12#include <keep.h>
13#include <sm/optee_smc.h>
14#include <sm/teesmc_opteed.h>
15#include <sm/teesmc_opteed_macros.h>
16
17	/*
18	 * Setup SP_EL0 and SPEL1, SP will be set to SP_EL0.
19	 * SP_EL0 is assigned stack_tmp_export + cpu_id * stack_tmp_stride
20	 * SP_EL1 is assigned thread_core_local[cpu_id]
21	 */
22	.macro set_sp
23		bl	__get_core_pos
24		cmp	x0, #CFG_TEE_CORE_NB_CORE
25		/* Unsupported CPU, park it before it breaks something */
26		bge	unhandled_cpu
27		adr	x1, stack_tmp_stride
28		ldr	w1, [x1]
29		mul	x1, x0, x1
30		adrp	x0, stack_tmp_export
31		add	x0, x0, :lo12:stack_tmp_export
32		ldr	x0, [x0]
33		msr	spsel, #0
34		add	sp, x1, x0
35		bl	thread_get_core_local
36		msr	spsel, #1
37		mov	sp, x0
38		msr	spsel, #0
39	.endm
40
41	.macro set_sctlr_el1
42		mrs	x0, sctlr_el1
43		orr	x0, x0, #SCTLR_I
44		orr	x0, x0, #SCTLR_SA
45		orr	x0, x0, #SCTLR_SPAN
46#if defined(CFG_CORE_RWDATA_NOEXEC)
47		orr	x0, x0, #SCTLR_WXN
48#endif
49#if defined(CFG_SCTLR_ALIGNMENT_CHECK)
50		orr	x0, x0, #SCTLR_A
51#else
52		bic	x0, x0, #SCTLR_A
53#endif
54		msr	sctlr_el1, x0
55	.endm
56
57FUNC _start , :
58	mov	x19, x0		/* Save pagable part address */
59#if defined(CFG_DT_ADDR)
60	ldr     x20, =CFG_DT_ADDR
61#else
62	mov	x20, x2		/* Save DT address */
63#endif
64
65	adr	x0, reset_vect_table
66	msr	vbar_el1, x0
67	isb
68
69	set_sctlr_el1
70	isb
71
72#ifdef CFG_WITH_PAGER
73	/*
74	 * Move init code into correct location and move hashes to a
75	 * temporary safe location until the heap is initialized.
76	 *
77	 * The binary is built as:
78	 * [Pager code, rodata and data] : In correct location
79	 * [Init code and rodata] : Should be copied to __init_start
80	 * [struct boot_embdata + data] : Should be saved before
81	 * initializing pager, first uint32_t tells the length of the data
82	 */
83	adr	x0, __init_start	/* dst */
84	adr	x1, __data_end		/* src */
85	adr	x2, __init_end
86	sub	x2, x2, x0		/* init len */
87	ldr	w4, [x1, x2]		/* length of hashes etc */
88	add	x2, x2, x4		/* length of init and hashes etc */
89	/* Copy backwards (as memmove) in case we're overlapping */
90	add	x0, x0, x2		/* __init_start + len */
91	add	x1, x1, x2		/* __data_end + len */
92	adr	x3, cached_mem_end
93	str	x0, [x3]
94	adr	x2, __init_start
95copy_init:
96	ldp	x3, x4, [x1, #-16]!
97	stp	x3, x4, [x0, #-16]!
98	cmp	x0, x2
99	b.gt	copy_init
100#else
101	/*
102	 * The binary is built as:
103	 * [Core, rodata and data] : In correct location
104	 * [struct boot_embdata + data] : Should be moved to __end, first
105	 * uint32_t tells the length of the struct + data
106	 */
107	adr_l	x0, __end		/* dst */
108	adr_l	x1, __data_end		/* src */
109	ldr	w2, [x1]		/* struct boot_embdata::total_len */
110	/* Copy backwards (as memmove) in case we're overlapping */
111	add	x0, x0, x2
112	add	x1, x1, x2
113	adr	x3, cached_mem_end
114	str	x0, [x3]
115	adr_l	x2, __end
116
117copy_init:
118	ldp	x3, x4, [x1, #-16]!
119	stp	x3, x4, [x0, #-16]!
120	cmp	x0, x2
121	b.gt	copy_init
122#endif
123
124	/*
125	 * Clear .bss, this code obviously depends on the linker keeping
126	 * start/end of .bss at least 8 byte aligned.
127	 */
128	adr_l	x0, __bss_start
129	adr_l	x1, __bss_end
130clear_bss:
131	str	xzr, [x0], #8
132	cmp	x0, x1
133	b.lt	clear_bss
134
135#ifdef CFG_VIRTUALIZATION
136	/*
137	 * Clear .nex_bss, this code obviously depends on the linker keeping
138	 * start/end of .bss at least 8 byte aligned.
139	 */
140	adr	x0, __nex_bss_start
141	adr	x1, __nex_bss_end
142clear_nex_bss:
143	str	xzr, [x0], #8
144	cmp	x0, x1
145	b.lt	clear_nex_bss
146#endif
147
148	/* Setup SP_EL0 and SP_EL1, SP will be set to SP_EL0 */
149	set_sp
150
151	/* Enable aborts now that we can receive exceptions */
152	msr	daifclr, #DAIFBIT_ABT
153
154	/*
155	 * Invalidate dcache for all memory used during initialization to
156	 * avoid nasty surprices when the cache is turned on. We must not
157	 * invalidate memory not used by OP-TEE since we may invalidate
158	 * entries used by for instance ARM Trusted Firmware.
159	 */
160	adr_l	x0, __text_start
161	ldr	x1, cached_mem_end
162	sub	x1, x1, x0
163	bl	dcache_cleaninv_range
164
165	/* Enable Console */
166	bl	console_init
167
168#ifdef CFG_CORE_ASLR
169	mov	x0, x20
170	bl	get_aslr_seed
171#else
172	mov	x0, #0
173#endif
174
175	adr	x1, boot_mmu_config
176	bl	core_init_mmu_map
177
178#ifdef CFG_CORE_ASLR
179	/*
180	 * Process relocation information again updating for the new
181	 * offset. We're doing this now before MMU is enabled as some of
182	 * the memory will become write protected.
183	 */
184	ldr	x0, boot_mmu_config + CORE_MMU_CONFIG_LOAD_OFFSET
185	/*
186	 * Update cached_mem_end address with load offset since it was
187	 * calculated before relocation.
188	 */
189	adr	x5, cached_mem_end
190	ldr	x6, [x5]
191	add	x6, x6, x0
192	str	x6, [x5]
193	bl	relocate
194#endif
195
196	bl	__get_core_pos
197	bl	enable_mmu
198#ifdef CFG_CORE_ASLR
199	/*
200	 * Reinitialize console, since register_serial_console() has
201	 * previously registered a PA and with ASLR the VA is different
202	 * from the PA.
203	 */
204	bl	console_init
205#endif
206
207	mov	x0, x19		/* pagable part address */
208	mov	x1, #-1
209	mov	x2, x20		/* DT address */
210	bl	boot_init_primary
211
212	/*
213	 * In case we've touched memory that secondary CPUs will use before
214	 * they have turned on their D-cache, clean and invalidate the
215	 * D-cache before exiting to normal world.
216	 */
217	adr_l	x0, __text_start
218	ldr	x1, cached_mem_end
219	sub	x1, x1, x0
220	bl	dcache_cleaninv_range
221
222
223	/*
224	 * Clear current thread id now to allow the thread to be reused on
225	 * next entry. Matches the thread_init_boot_thread in
226	 * boot.c.
227	 */
228#ifndef CFG_VIRTUALIZATION
229	bl 	thread_clr_boot_thread
230#endif
231
232	/*
233	 * Pass the vector address returned from main_init
234	 * Compensate for the load offset since cpu_on_handler() is
235	 * called with MMU off.
236	 */
237	ldr	x0, boot_mmu_config + CORE_MMU_CONFIG_LOAD_OFFSET
238	adr	x1, thread_vector_table
239	sub	x1, x1, x0
240	mov	x0, #TEESMC_OPTEED_RETURN_ENTRY_DONE
241	smc	#0
242	b	.	/* SMC should not return */
243END_FUNC _start
244DECLARE_KEEP_INIT _start
245
246	.balign	8
247LOCAL_DATA cached_mem_end , :
248	.skip	8
249END_DATA cached_mem_end
250
251#ifdef CFG_CORE_ASLR
252LOCAL_FUNC relocate , :
253	/* x0 holds load offset */
254#ifdef CFG_WITH_PAGER
255	adr_l	x6, __init_end
256#else
257	adr_l	x6, __end
258#endif
259	ldp	w2, w3, [x6, #BOOT_EMBDATA_RELOC_OFFSET]
260
261	mov_imm	x1, TEE_RAM_START
262	add	x2, x2, x6	/* start of relocations */
263	add	x3, x3, x2	/* end of relocations */
264
265	/*
266	 * Relocations are not formatted as Rela64, instead they are in a
267	 * compressed format created by get_reloc_bin() in
268	 * scripts/gen_tee_bin.py
269	 *
270	 * All the R_AARCH64_RELATIVE relocations are translated into a
271	 * list list of 32-bit offsets from TEE_RAM_START. At each address
272	 * a 64-bit value pointed out which increased with the load offset.
273	 */
274
275#ifdef CFG_WITH_PAGER
276	/*
277	 * With pager enabled we can only relocate the pager and init
278	 * parts, the rest has to be done when a page is populated.
279	 */
280	sub	x6, x6, x1
281#endif
282
283	b	2f
284	/* Loop over the relocation addresses and process all entries */
2851:	ldr	w4, [x2], #4
286#ifdef CFG_WITH_PAGER
287	/* Skip too large addresses */
288	cmp	x4, x6
289	b.ge	2f
290#endif
291	add	x4, x4, x1
292	ldr	x5, [x4]
293	add	x5, x5, x0
294	str	x5, [x4]
295
2962:	cmp	x2, x3
297	b.ne	1b
298
299	ret
300END_FUNC relocate
301#endif
302
303/*
304 * void enable_mmu(unsigned long core_pos);
305 *
306 * This function depends on being mapped with in the identity map where
307 * physical address and virtual address is the same. After MMU has been
308 * enabled the instruction pointer will be updated to execute as the new
309 * offset instead. Stack pointers and the return address are updated.
310 */
311LOCAL_FUNC enable_mmu , : , .identity_map
312	adr	x1, boot_mmu_config
313	load_xregs x1, 0, 2, 6
314	/*
315	 * x0 = core_pos
316	 * x2 = tcr_el1
317	 * x3 = mair_el1
318	 * x4 = ttbr0_el1_base
319	 * x5 = ttbr0_core_offset
320	 * x6 = load_offset
321	 */
322	msr	tcr_el1, x2
323	msr	mair_el1, x3
324
325	/*
326	 * ttbr0_el1 = ttbr0_el1_base + ttbr0_core_offset * core_pos
327	 */
328	madd	x1, x5, x0, x4
329	msr	ttbr0_el1, x1
330	msr	ttbr1_el1, xzr
331	isb
332
333	/* Invalidate TLB */
334	tlbi	vmalle1
335
336	/*
337	 * Make sure translation table writes have drained into memory and
338	 * the TLB invalidation is complete.
339	 */
340	dsb	sy
341	isb
342
343	/* Enable the MMU */
344	mrs	x1, sctlr_el1
345	orr	x1, x1, #SCTLR_M
346	msr	sctlr_el1, x1
347	isb
348
349	/* Update vbar */
350	mrs	x1, vbar_el1
351	add	x1, x1, x6
352	msr	vbar_el1, x1
353	isb
354
355	/* Invalidate instruction cache and branch predictor */
356	ic	iallu
357	isb
358
359	/* Enable I and D cache */
360	mrs	x1, sctlr_el1
361	orr	x1, x1, #SCTLR_I
362	orr	x1, x1, #SCTLR_C
363	msr	sctlr_el1, x1
364	isb
365
366	/* Adjust stack pointers and return address */
367	msr	spsel, #1
368	add	sp, sp, x6
369	msr	spsel, #0
370	add	sp, sp, x6
371	add	x30, x30, x6
372
373	ret
374END_FUNC enable_mmu
375
376	.balign	8
377DATA boot_mmu_config , : /* struct core_mmu_config */
378	.skip	CORE_MMU_CONFIG_SIZE
379END_DATA boot_mmu_config
380
381FUNC cpu_on_handler , :
382	mov	x19, x0
383	mov	x20, x1
384	mov	x21, x30
385
386	adr	x0, reset_vect_table
387	msr	vbar_el1, x0
388	isb
389
390	set_sctlr_el1
391	isb
392
393	/* Enable aborts now that we can receive exceptions */
394	msr	daifclr, #DAIFBIT_ABT
395
396	bl	__get_core_pos
397	bl	enable_mmu
398
399	/* Setup SP_EL0 and SP_EL1, SP will be set to SP_EL0 */
400	set_sp
401
402	mov	x0, x19
403	mov	x1, x20
404	mov	x30, x21
405	b	boot_cpu_on_handler
406END_FUNC cpu_on_handler
407DECLARE_KEEP_PAGER cpu_on_handler
408
409LOCAL_FUNC unhandled_cpu , :
410	wfi
411	b	unhandled_cpu
412END_FUNC unhandled_cpu
413
414	/*
415	 * This macro verifies that the a given vector doesn't exceed the
416	 * architectural limit of 32 instructions. This is meant to be placed
417	 * immedately after the last instruction in the vector. It takes the
418	 * vector entry as the parameter
419	 */
420	.macro check_vector_size since
421	  .if (. - \since) > (32 * 4)
422	    .error "Vector exceeds 32 instructions"
423	  .endif
424	.endm
425
426	.section .identity_map, "ax", %progbits
427	.align	11
428LOCAL_FUNC reset_vect_table , :, .identity_map
429	/* -----------------------------------------------------
430	 * Current EL with SP0 : 0x0 - 0x180
431	 * -----------------------------------------------------
432	 */
433SynchronousExceptionSP0:
434	b	SynchronousExceptionSP0
435	check_vector_size SynchronousExceptionSP0
436
437	.align	7
438IrqSP0:
439	b	IrqSP0
440	check_vector_size IrqSP0
441
442	.align	7
443FiqSP0:
444	b	FiqSP0
445	check_vector_size FiqSP0
446
447	.align	7
448SErrorSP0:
449	b	SErrorSP0
450	check_vector_size SErrorSP0
451
452	/* -----------------------------------------------------
453	 * Current EL with SPx: 0x200 - 0x380
454	 * -----------------------------------------------------
455	 */
456	.align	7
457SynchronousExceptionSPx:
458	b	SynchronousExceptionSPx
459	check_vector_size SynchronousExceptionSPx
460
461	.align	7
462IrqSPx:
463	b	IrqSPx
464	check_vector_size IrqSPx
465
466	.align	7
467FiqSPx:
468	b	FiqSPx
469	check_vector_size FiqSPx
470
471	.align	7
472SErrorSPx:
473	b	SErrorSPx
474	check_vector_size SErrorSPx
475
476	/* -----------------------------------------------------
477	 * Lower EL using AArch64 : 0x400 - 0x580
478	 * -----------------------------------------------------
479	 */
480	.align	7
481SynchronousExceptionA64:
482	b	SynchronousExceptionA64
483	check_vector_size SynchronousExceptionA64
484
485	.align	7
486IrqA64:
487	b	IrqA64
488	check_vector_size IrqA64
489
490	.align	7
491FiqA64:
492	b	FiqA64
493	check_vector_size FiqA64
494
495	.align	7
496SErrorA64:
497	b   	SErrorA64
498	check_vector_size SErrorA64
499
500	/* -----------------------------------------------------
501	 * Lower EL using AArch32 : 0x0 - 0x180
502	 * -----------------------------------------------------
503	 */
504	.align	7
505SynchronousExceptionA32:
506	b	SynchronousExceptionA32
507	check_vector_size SynchronousExceptionA32
508
509	.align	7
510IrqA32:
511	b	IrqA32
512	check_vector_size IrqA32
513
514	.align	7
515FiqA32:
516	b	FiqA32
517	check_vector_size FiqA32
518
519	.align	7
520SErrorA32:
521	b	SErrorA32
522	check_vector_size SErrorA32
523
524END_FUNC reset_vect_table
525