xref: /linux/arch/arm64/kernel/head.S (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * Low-level CPU initialisation
4 * Based on arch/arm/kernel/head.S
5 *
6 * Copyright (C) 1994-2002 Russell King
7 * Copyright (C) 2003-2012 ARM Ltd.
8 * Authors:	Catalin Marinas <catalin.marinas@arm.com>
9 *		Will Deacon <will.deacon@arm.com>
10 */
11
12#include <linux/linkage.h>
13#include <linux/init.h>
14#include <linux/pgtable.h>
15
16#include <asm/asm_pointer_auth.h>
17#include <asm/assembler.h>
18#include <asm/boot.h>
19#include <asm/bug.h>
20#include <asm/ptrace.h>
21#include <asm/asm-offsets.h>
22#include <asm/cache.h>
23#include <asm/cputype.h>
24#include <asm/el2_setup.h>
25#include <asm/elf.h>
26#include <asm/image.h>
27#include <asm/kernel-pgtable.h>
28#include <asm/kvm_arm.h>
29#include <asm/memory.h>
30#include <asm/pgtable-hwdef.h>
31#include <asm/page.h>
32#include <asm/scs.h>
33#include <asm/smp.h>
34#include <asm/sysreg.h>
35#include <asm/thread_info.h>
36#include <asm/virt.h>
37
38#include "efi-header.S"
39
40#define __PHYS_OFFSET	KERNEL_START
41
42#if (PAGE_OFFSET & 0x1fffff) != 0
43#error PAGE_OFFSET must be at least 2MB aligned
44#endif
45
46/*
47 * Kernel startup entry point.
48 * ---------------------------
49 *
50 * The requirements are:
51 *   MMU = off, D-cache = off, I-cache = on or off,
52 *   x0 = physical address to the FDT blob.
53 *
54 * This code is mostly position independent so you call this at
55 * __pa(PAGE_OFFSET).
56 *
57 * Note that the callee-saved registers are used for storing variables
58 * that are useful before the MMU is enabled. The allocations are described
59 * in the entry routines.
60 */
61	__HEAD
62	/*
63	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
64	 */
65	efi_signature_nop			// special NOP to identity as PE/COFF executable
66	b	primary_entry			// branch to kernel start, magic
67	.quad	0				// Image load offset from start of RAM, little-endian
68	le64sym	_kernel_size_le			// Effective size of kernel image, little-endian
69	le64sym	_kernel_flags_le		// Informative flags, little-endian
70	.quad	0				// reserved
71	.quad	0				// reserved
72	.quad	0				// reserved
73	.ascii	ARM64_IMAGE_MAGIC		// Magic number
74	.long	.Lpe_header_offset		// Offset to the PE header.
75
76	__EFI_PE_HEADER
77
78	__INIT
79
80	/*
81	 * The following callee saved general purpose registers are used on the
82	 * primary lowlevel boot path:
83	 *
84	 *  Register   Scope                      Purpose
85	 *  x21        primary_entry() .. start_kernel()        FDT pointer passed at boot in x0
86	 *  x23        primary_entry() .. start_kernel()        physical misalignment/KASLR offset
87	 *  x28        __create_page_tables()                   callee preserved temp register
88	 *  x19/x20    __primary_switch()                       callee preserved temp registers
89	 *  x24        __primary_switch() .. relocate_kernel()  current RELR displacement
90	 */
91SYM_CODE_START(primary_entry)
92	bl	preserve_boot_args
93	bl	init_kernel_el			// w0=cpu_boot_mode
94	adrp	x23, __PHYS_OFFSET
95	and	x23, x23, MIN_KIMG_ALIGN - 1	// KASLR offset, defaults to 0
96	bl	set_cpu_boot_mode_flag
97	bl	__create_page_tables
98	/*
99	 * The following calls CPU setup code, see arch/arm64/mm/proc.S for
100	 * details.
101	 * On return, the CPU will be ready for the MMU to be turned on and
102	 * the TCR will have been set.
103	 */
104	bl	__cpu_setup			// initialise processor
105	b	__primary_switch
106SYM_CODE_END(primary_entry)
107
108/*
109 * Preserve the arguments passed by the bootloader in x0 .. x3
110 */
111SYM_CODE_START_LOCAL(preserve_boot_args)
112	mov	x21, x0				// x21=FDT
113
114	adr_l	x0, boot_args			// record the contents of
115	stp	x21, x1, [x0]			// x0 .. x3 at kernel entry
116	stp	x2, x3, [x0, #16]
117
118	dmb	sy				// needed before dc ivac with
119						// MMU off
120
121	add	x1, x0, #0x20			// 4 x 8 bytes
122	b	dcache_inval_poc		// tail call
123SYM_CODE_END(preserve_boot_args)
124
125/*
126 * Macro to create a table entry to the next page.
127 *
128 *	tbl:	page table address
129 *	virt:	virtual address
130 *	shift:	#imm page table shift
131 *	ptrs:	#imm pointers per table page
132 *
133 * Preserves:	virt
134 * Corrupts:	ptrs, tmp1, tmp2
135 * Returns:	tbl -> next level table page address
136 */
137	.macro	create_table_entry, tbl, virt, shift, ptrs, tmp1, tmp2
138	add	\tmp1, \tbl, #PAGE_SIZE
139	phys_to_pte \tmp2, \tmp1
140	orr	\tmp2, \tmp2, #PMD_TYPE_TABLE	// address of next table and entry type
141	lsr	\tmp1, \virt, #\shift
142	sub	\ptrs, \ptrs, #1
143	and	\tmp1, \tmp1, \ptrs		// table index
144	str	\tmp2, [\tbl, \tmp1, lsl #3]
145	add	\tbl, \tbl, #PAGE_SIZE		// next level table page
146	.endm
147
148/*
149 * Macro to populate page table entries, these entries can be pointers to the next level
150 * or last level entries pointing to physical memory.
151 *
152 *	tbl:	page table address
153 *	rtbl:	pointer to page table or physical memory
154 *	index:	start index to write
155 *	eindex:	end index to write - [index, eindex] written to
156 *	flags:	flags for pagetable entry to or in
157 *	inc:	increment to rtbl between each entry
158 *	tmp1:	temporary variable
159 *
160 * Preserves:	tbl, eindex, flags, inc
161 * Corrupts:	index, tmp1
162 * Returns:	rtbl
163 */
164	.macro populate_entries, tbl, rtbl, index, eindex, flags, inc, tmp1
165.Lpe\@:	phys_to_pte \tmp1, \rtbl
166	orr	\tmp1, \tmp1, \flags	// tmp1 = table entry
167	str	\tmp1, [\tbl, \index, lsl #3]
168	add	\rtbl, \rtbl, \inc	// rtbl = pa next level
169	add	\index, \index, #1
170	cmp	\index, \eindex
171	b.ls	.Lpe\@
172	.endm
173
174/*
175 * Compute indices of table entries from virtual address range. If multiple entries
176 * were needed in the previous page table level then the next page table level is assumed
177 * to be composed of multiple pages. (This effectively scales the end index).
178 *
179 *	vstart:	virtual address of start of range
180 *	vend:	virtual address of end of range
181 *	shift:	shift used to transform virtual address into index
182 *	ptrs:	number of entries in page table
183 *	istart:	index in table corresponding to vstart
184 *	iend:	index in table corresponding to vend
185 *	count:	On entry: how many extra entries were required in previous level, scales
186 *			  our end index.
187 *		On exit: returns how many extra entries required for next page table level
188 *
189 * Preserves:	vstart, vend, shift, ptrs
190 * Returns:	istart, iend, count
191 */
192	.macro compute_indices, vstart, vend, shift, ptrs, istart, iend, count
193	lsr	\iend, \vend, \shift
194	mov	\istart, \ptrs
195	sub	\istart, \istart, #1
196	and	\iend, \iend, \istart	// iend = (vend >> shift) & (ptrs - 1)
197	mov	\istart, \ptrs
198	mul	\istart, \istart, \count
199	add	\iend, \iend, \istart	// iend += count * ptrs
200					// our entries span multiple tables
201
202	lsr	\istart, \vstart, \shift
203	mov	\count, \ptrs
204	sub	\count, \count, #1
205	and	\istart, \istart, \count
206
207	sub	\count, \iend, \istart
208	.endm
209
210/*
211 * Map memory for specified virtual address range. Each level of page table needed supports
212 * multiple entries. If a level requires n entries the next page table level is assumed to be
213 * formed from n pages.
214 *
215 *	tbl:	location of page table
216 *	rtbl:	address to be used for first level page table entry (typically tbl + PAGE_SIZE)
217 *	vstart:	start address to map
218 *	vend:	end address to map - we map [vstart, vend]
219 *	flags:	flags to use to map last level entries
220 *	phys:	physical address corresponding to vstart - physical memory is contiguous
221 *	pgds:	the number of pgd entries
222 *
223 * Temporaries:	istart, iend, tmp, count, sv - these need to be different registers
224 * Preserves:	vstart, vend, flags
225 * Corrupts:	tbl, rtbl, istart, iend, tmp, count, sv
226 */
227	.macro map_memory, tbl, rtbl, vstart, vend, flags, phys, pgds, istart, iend, tmp, count, sv
228	add \rtbl, \tbl, #PAGE_SIZE
229	mov \sv, \rtbl
230	mov \count, #0
231	compute_indices \vstart, \vend, #PGDIR_SHIFT, \pgds, \istart, \iend, \count
232	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
233	mov \tbl, \sv
234	mov \sv, \rtbl
235
236#if SWAPPER_PGTABLE_LEVELS > 3
237	compute_indices \vstart, \vend, #PUD_SHIFT, #PTRS_PER_PUD, \istart, \iend, \count
238	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
239	mov \tbl, \sv
240	mov \sv, \rtbl
241#endif
242
243#if SWAPPER_PGTABLE_LEVELS > 2
244	compute_indices \vstart, \vend, #SWAPPER_TABLE_SHIFT, #PTRS_PER_PMD, \istart, \iend, \count
245	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
246	mov \tbl, \sv
247#endif
248
249	compute_indices \vstart, \vend, #SWAPPER_BLOCK_SHIFT, #PTRS_PER_PTE, \istart, \iend, \count
250	bic \count, \phys, #SWAPPER_BLOCK_SIZE - 1
251	populate_entries \tbl, \count, \istart, \iend, \flags, #SWAPPER_BLOCK_SIZE, \tmp
252	.endm
253
254/*
255 * Setup the initial page tables. We only setup the barest amount which is
256 * required to get the kernel running. The following sections are required:
257 *   - identity mapping to enable the MMU (low address, TTBR0)
258 *   - first few MB of the kernel linear mapping to jump to once the MMU has
259 *     been enabled
260 */
261SYM_FUNC_START_LOCAL(__create_page_tables)
262	mov	x28, lr
263
264	/*
265	 * Invalidate the init page tables to avoid potential dirty cache lines
266	 * being evicted. Other page tables are allocated in rodata as part of
267	 * the kernel image, and thus are clean to the PoC per the boot
268	 * protocol.
269	 */
270	adrp	x0, init_pg_dir
271	adrp	x1, init_pg_end
272	bl	dcache_inval_poc
273
274	/*
275	 * Clear the init page tables.
276	 */
277	adrp	x0, init_pg_dir
278	adrp	x1, init_pg_end
279	sub	x1, x1, x0
2801:	stp	xzr, xzr, [x0], #16
281	stp	xzr, xzr, [x0], #16
282	stp	xzr, xzr, [x0], #16
283	stp	xzr, xzr, [x0], #16
284	subs	x1, x1, #64
285	b.ne	1b
286
287	mov	x7, SWAPPER_MM_MMUFLAGS
288
289	/*
290	 * Create the identity mapping.
291	 */
292	adrp	x0, idmap_pg_dir
293	adrp	x3, __idmap_text_start		// __pa(__idmap_text_start)
294
295#ifdef CONFIG_ARM64_VA_BITS_52
296	mrs_s	x6, SYS_ID_AA64MMFR2_EL1
297	and	x6, x6, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
298	mov	x5, #52
299	cbnz	x6, 1f
300#endif
301	mov	x5, #VA_BITS_MIN
3021:
303	adr_l	x6, vabits_actual
304	str	x5, [x6]
305	dmb	sy
306	dc	ivac, x6		// Invalidate potentially stale cache line
307
308	/*
309	 * VA_BITS may be too small to allow for an ID mapping to be created
310	 * that covers system RAM if that is located sufficiently high in the
311	 * physical address space. So for the ID map, use an extended virtual
312	 * range in that case, and configure an additional translation level
313	 * if needed.
314	 *
315	 * Calculate the maximum allowed value for TCR_EL1.T0SZ so that the
316	 * entire ID map region can be mapped. As T0SZ == (64 - #bits used),
317	 * this number conveniently equals the number of leading zeroes in
318	 * the physical address of __idmap_text_end.
319	 */
320	adrp	x5, __idmap_text_end
321	clz	x5, x5
322	cmp	x5, TCR_T0SZ(VA_BITS_MIN) // default T0SZ small enough?
323	b.ge	1f			// .. then skip VA range extension
324
325	adr_l	x6, idmap_t0sz
326	str	x5, [x6]
327	dmb	sy
328	dc	ivac, x6		// Invalidate potentially stale cache line
329
330#if (VA_BITS < 48)
331#define EXTRA_SHIFT	(PGDIR_SHIFT + PAGE_SHIFT - 3)
332#define EXTRA_PTRS	(1 << (PHYS_MASK_SHIFT - EXTRA_SHIFT))
333
334	/*
335	 * If VA_BITS < 48, we have to configure an additional table level.
336	 * First, we have to verify our assumption that the current value of
337	 * VA_BITS was chosen such that all translation levels are fully
338	 * utilised, and that lowering T0SZ will always result in an additional
339	 * translation level to be configured.
340	 */
341#if VA_BITS != EXTRA_SHIFT
342#error "Mismatch between VA_BITS and page size/number of translation levels"
343#endif
344
345	mov	x4, EXTRA_PTRS
346	create_table_entry x0, x3, EXTRA_SHIFT, x4, x5, x6
347#else
348	/*
349	 * If VA_BITS == 48, we don't have to configure an additional
350	 * translation level, but the top-level table has more entries.
351	 */
352	mov	x4, #1 << (PHYS_MASK_SHIFT - PGDIR_SHIFT)
353	str_l	x4, idmap_ptrs_per_pgd, x5
354#endif
3551:
356	ldr_l	x4, idmap_ptrs_per_pgd
357	adr_l	x6, __idmap_text_end		// __pa(__idmap_text_end)
358
359	map_memory x0, x1, x3, x6, x7, x3, x4, x10, x11, x12, x13, x14
360
361	/*
362	 * Map the kernel image (starting with PHYS_OFFSET).
363	 */
364	adrp	x0, init_pg_dir
365	mov_q	x5, KIMAGE_VADDR		// compile time __va(_text)
366	add	x5, x5, x23			// add KASLR displacement
367	mov	x4, PTRS_PER_PGD
368	adrp	x6, _end			// runtime __pa(_end)
369	adrp	x3, _text			// runtime __pa(_text)
370	sub	x6, x6, x3			// _end - _text
371	add	x6, x6, x5			// runtime __va(_end)
372
373	map_memory x0, x1, x5, x6, x7, x3, x4, x10, x11, x12, x13, x14
374
375	/*
376	 * Since the page tables have been populated with non-cacheable
377	 * accesses (MMU disabled), invalidate those tables again to
378	 * remove any speculatively loaded cache lines.
379	 */
380	dmb	sy
381
382	adrp	x0, idmap_pg_dir
383	adrp	x1, idmap_pg_end
384	bl	dcache_inval_poc
385
386	adrp	x0, init_pg_dir
387	adrp	x1, init_pg_end
388	bl	dcache_inval_poc
389
390	ret	x28
391SYM_FUNC_END(__create_page_tables)
392
393	/*
394	 * Initialize CPU registers with task-specific and cpu-specific context.
395	 *
396	 * Create a final frame record at task_pt_regs(current)->stackframe, so
397	 * that the unwinder can identify the final frame record of any task by
398	 * its location in the task stack. We reserve the entire pt_regs space
399	 * for consistency with user tasks and kthreads.
400	 */
401	.macro	init_cpu_task tsk, tmp1, tmp2
402	msr	sp_el0, \tsk
403
404	ldr	\tmp1, [\tsk, #TSK_STACK]
405	add	sp, \tmp1, #THREAD_SIZE
406	sub	sp, sp, #PT_REGS_SIZE
407
408	stp	xzr, xzr, [sp, #S_STACKFRAME]
409	add	x29, sp, #S_STACKFRAME
410
411	scs_load \tsk
412
413	adr_l	\tmp1, __per_cpu_offset
414	ldr	w\tmp2, [\tsk, #TSK_CPU]
415	ldr	\tmp1, [\tmp1, \tmp2, lsl #3]
416	set_this_cpu_offset \tmp1
417	.endm
418
419/*
420 * The following fragment of code is executed with the MMU enabled.
421 *
422 *   x0 = __PHYS_OFFSET
423 */
424SYM_FUNC_START_LOCAL(__primary_switched)
425	adr_l	x4, init_task
426	init_cpu_task x4, x5, x6
427
428	adr_l	x8, vectors			// load VBAR_EL1 with virtual
429	msr	vbar_el1, x8			// vector table address
430	isb
431
432	stp	x29, x30, [sp, #-16]!
433	mov	x29, sp
434
435	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
436
437	ldr_l	x4, kimage_vaddr		// Save the offset between
438	sub	x4, x4, x0			// the kernel virtual and
439	str_l	x4, kimage_voffset, x5		// physical mappings
440
441	// Clear BSS
442	adr_l	x0, __bss_start
443	mov	x1, xzr
444	adr_l	x2, __bss_stop
445	sub	x2, x2, x0
446	bl	__pi_memset
447	dsb	ishst				// Make zero page visible to PTW
448
449#if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
450	bl	kasan_early_init
451#endif
452	mov	x0, x21				// pass FDT address in x0
453	bl	early_fdt_map			// Try mapping the FDT early
454	bl	init_feature_override		// Parse cpu feature overrides
455#ifdef CONFIG_RANDOMIZE_BASE
456	tst	x23, ~(MIN_KIMG_ALIGN - 1)	// already running randomized?
457	b.ne	0f
458	bl	kaslr_early_init		// parse FDT for KASLR options
459	cbz	x0, 0f				// KASLR disabled? just proceed
460	orr	x23, x23, x0			// record KASLR offset
461	ldp	x29, x30, [sp], #16		// we must enable KASLR, return
462	ret					// to __primary_switch()
4630:
464#endif
465	bl	switch_to_vhe			// Prefer VHE if possible
466	ldp	x29, x30, [sp], #16
467	bl	start_kernel
468	ASM_BUG()
469SYM_FUNC_END(__primary_switched)
470
471	.pushsection ".rodata", "a"
472SYM_DATA_START(kimage_vaddr)
473	.quad		_text
474SYM_DATA_END(kimage_vaddr)
475EXPORT_SYMBOL(kimage_vaddr)
476	.popsection
477
478/*
479 * end early head section, begin head code that is also used for
480 * hotplug and needs to have the same protections as the text region
481 */
482	.section ".idmap.text","awx"
483
484/*
485 * Starting from EL2 or EL1, configure the CPU to execute at the highest
486 * reachable EL supported by the kernel in a chosen default state. If dropping
487 * from EL2 to EL1, configure EL2 before configuring EL1.
488 *
489 * Since we cannot always rely on ERET synchronizing writes to sysregs (e.g. if
490 * SCTLR_ELx.EOS is clear), we place an ISB prior to ERET.
491 *
492 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in w0 if
493 * booted in EL1 or EL2 respectively.
494 */
495SYM_FUNC_START(init_kernel_el)
496	mrs	x0, CurrentEL
497	cmp	x0, #CurrentEL_EL2
498	b.eq	init_el2
499
500SYM_INNER_LABEL(init_el1, SYM_L_LOCAL)
501	mov_q	x0, INIT_SCTLR_EL1_MMU_OFF
502	msr	sctlr_el1, x0
503	isb
504	mov_q	x0, INIT_PSTATE_EL1
505	msr	spsr_el1, x0
506	msr	elr_el1, lr
507	mov	w0, #BOOT_CPU_MODE_EL1
508	eret
509
510SYM_INNER_LABEL(init_el2, SYM_L_LOCAL)
511	mov_q	x0, HCR_HOST_NVHE_FLAGS
512	msr	hcr_el2, x0
513	isb
514
515	init_el2_state
516
517	/* Hypervisor stub */
518	adr_l	x0, __hyp_stub_vectors
519	msr	vbar_el2, x0
520	isb
521
522	/*
523	 * Fruity CPUs seem to have HCR_EL2.E2H set to RES1,
524	 * making it impossible to start in nVHE mode. Is that
525	 * compliant with the architecture? Absolutely not!
526	 */
527	mrs	x0, hcr_el2
528	and	x0, x0, #HCR_E2H
529	cbz	x0, 1f
530
531	/* Switching to VHE requires a sane SCTLR_EL1 as a start */
532	mov_q	x0, INIT_SCTLR_EL1_MMU_OFF
533	msr_s	SYS_SCTLR_EL12, x0
534
535	/*
536	 * Force an eret into a helper "function", and let it return
537	 * to our original caller... This makes sure that we have
538	 * initialised the basic PSTATE state.
539	 */
540	mov	x0, #INIT_PSTATE_EL2
541	msr	spsr_el1, x0
542	adr	x0, __cpu_stick_to_vhe
543	msr	elr_el1, x0
544	eret
545
5461:
547	mov_q	x0, INIT_SCTLR_EL1_MMU_OFF
548	msr	sctlr_el1, x0
549
550	msr	elr_el2, lr
551	mov	w0, #BOOT_CPU_MODE_EL2
552	eret
553
554__cpu_stick_to_vhe:
555	mov	x0, #HVC_VHE_RESTART
556	hvc	#0
557	mov	x0, #BOOT_CPU_MODE_EL2
558	ret
559SYM_FUNC_END(init_kernel_el)
560
561/*
562 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
563 * in w0. See arch/arm64/include/asm/virt.h for more info.
564 */
565SYM_FUNC_START_LOCAL(set_cpu_boot_mode_flag)
566	adr_l	x1, __boot_cpu_mode
567	cmp	w0, #BOOT_CPU_MODE_EL2
568	b.ne	1f
569	add	x1, x1, #4
5701:	str	w0, [x1]			// Save CPU boot mode
571	dmb	sy
572	dc	ivac, x1			// Invalidate potentially stale cache line
573	ret
574SYM_FUNC_END(set_cpu_boot_mode_flag)
575
576/*
577 * These values are written with the MMU off, but read with the MMU on.
578 * Writers will invalidate the corresponding address, discarding up to a
579 * 'Cache Writeback Granule' (CWG) worth of data. The linker script ensures
580 * sufficient alignment that the CWG doesn't overlap another section.
581 */
582	.pushsection ".mmuoff.data.write", "aw"
583/*
584 * We need to find out the CPU boot mode long after boot, so we need to
585 * store it in a writable variable.
586 *
587 * This is not in .bss, because we set it sufficiently early that the boot-time
588 * zeroing of .bss would clobber it.
589 */
590SYM_DATA_START(__boot_cpu_mode)
591	.long	BOOT_CPU_MODE_EL2
592	.long	BOOT_CPU_MODE_EL1
593SYM_DATA_END(__boot_cpu_mode)
594/*
595 * The booting CPU updates the failed status @__early_cpu_boot_status,
596 * with MMU turned off.
597 */
598SYM_DATA_START(__early_cpu_boot_status)
599	.quad 	0
600SYM_DATA_END(__early_cpu_boot_status)
601
602	.popsection
603
604	/*
605	 * This provides a "holding pen" for platforms to hold all secondary
606	 * cores are held until we're ready for them to initialise.
607	 */
608SYM_FUNC_START(secondary_holding_pen)
609	bl	init_kernel_el			// w0=cpu_boot_mode
610	bl	set_cpu_boot_mode_flag
611	mrs	x0, mpidr_el1
612	mov_q	x1, MPIDR_HWID_BITMASK
613	and	x0, x0, x1
614	adr_l	x3, secondary_holding_pen_release
615pen:	ldr	x4, [x3]
616	cmp	x4, x0
617	b.eq	secondary_startup
618	wfe
619	b	pen
620SYM_FUNC_END(secondary_holding_pen)
621
622	/*
623	 * Secondary entry point that jumps straight into the kernel. Only to
624	 * be used where CPUs are brought online dynamically by the kernel.
625	 */
626SYM_FUNC_START(secondary_entry)
627	bl	init_kernel_el			// w0=cpu_boot_mode
628	bl	set_cpu_boot_mode_flag
629	b	secondary_startup
630SYM_FUNC_END(secondary_entry)
631
632SYM_FUNC_START_LOCAL(secondary_startup)
633	/*
634	 * Common entry point for secondary CPUs.
635	 */
636	bl	switch_to_vhe
637	bl	__cpu_secondary_check52bitva
638	bl	__cpu_setup			// initialise processor
639	adrp	x1, swapper_pg_dir
640	bl	__enable_mmu
641	ldr	x8, =__secondary_switched
642	br	x8
643SYM_FUNC_END(secondary_startup)
644
645SYM_FUNC_START_LOCAL(__secondary_switched)
646	adr_l	x5, vectors
647	msr	vbar_el1, x5
648	isb
649
650	adr_l	x0, secondary_data
651	ldr	x2, [x0, #CPU_BOOT_TASK]
652	cbz	x2, __secondary_too_slow
653
654	init_cpu_task x2, x1, x3
655
656#ifdef CONFIG_ARM64_PTR_AUTH
657	ptrauth_keys_init_cpu x2, x3, x4, x5
658#endif
659
660	bl	secondary_start_kernel
661	ASM_BUG()
662SYM_FUNC_END(__secondary_switched)
663
664SYM_FUNC_START_LOCAL(__secondary_too_slow)
665	wfe
666	wfi
667	b	__secondary_too_slow
668SYM_FUNC_END(__secondary_too_slow)
669
670/*
671 * The booting CPU updates the failed status @__early_cpu_boot_status,
672 * with MMU turned off.
673 *
674 * update_early_cpu_boot_status tmp, status
675 *  - Corrupts tmp1, tmp2
676 *  - Writes 'status' to __early_cpu_boot_status and makes sure
677 *    it is committed to memory.
678 */
679
680	.macro	update_early_cpu_boot_status status, tmp1, tmp2
681	mov	\tmp2, #\status
682	adr_l	\tmp1, __early_cpu_boot_status
683	str	\tmp2, [\tmp1]
684	dmb	sy
685	dc	ivac, \tmp1			// Invalidate potentially stale cache line
686	.endm
687
688/*
689 * Enable the MMU.
690 *
691 *  x0  = SCTLR_EL1 value for turning on the MMU.
692 *  x1  = TTBR1_EL1 value
693 *
694 * Returns to the caller via x30/lr. This requires the caller to be covered
695 * by the .idmap.text section.
696 *
697 * Checks if the selected granule size is supported by the CPU.
698 * If it isn't, park the CPU
699 */
700SYM_FUNC_START(__enable_mmu)
701	mrs	x2, ID_AA64MMFR0_EL1
702	ubfx	x2, x2, #ID_AA64MMFR0_TGRAN_SHIFT, 4
703	cmp     x2, #ID_AA64MMFR0_TGRAN_SUPPORTED_MIN
704	b.lt    __no_granule_support
705	cmp     x2, #ID_AA64MMFR0_TGRAN_SUPPORTED_MAX
706	b.gt    __no_granule_support
707	update_early_cpu_boot_status 0, x2, x3
708	adrp	x2, idmap_pg_dir
709	phys_to_ttbr x1, x1
710	phys_to_ttbr x2, x2
711	msr	ttbr0_el1, x2			// load TTBR0
712	offset_ttbr1 x1, x3
713	msr	ttbr1_el1, x1			// load TTBR1
714	isb
715
716	set_sctlr_el1	x0
717
718	ret
719SYM_FUNC_END(__enable_mmu)
720
721SYM_FUNC_START(__cpu_secondary_check52bitva)
722#ifdef CONFIG_ARM64_VA_BITS_52
723	ldr_l	x0, vabits_actual
724	cmp	x0, #52
725	b.ne	2f
726
727	mrs_s	x0, SYS_ID_AA64MMFR2_EL1
728	and	x0, x0, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
729	cbnz	x0, 2f
730
731	update_early_cpu_boot_status \
732		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_52_BIT_VA, x0, x1
7331:	wfe
734	wfi
735	b	1b
736
737#endif
7382:	ret
739SYM_FUNC_END(__cpu_secondary_check52bitva)
740
741SYM_FUNC_START_LOCAL(__no_granule_support)
742	/* Indicate that this CPU can't boot and is stuck in the kernel */
743	update_early_cpu_boot_status \
744		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_NO_GRAN, x1, x2
7451:
746	wfe
747	wfi
748	b	1b
749SYM_FUNC_END(__no_granule_support)
750
751#ifdef CONFIG_RELOCATABLE
752SYM_FUNC_START_LOCAL(__relocate_kernel)
753	/*
754	 * Iterate over each entry in the relocation table, and apply the
755	 * relocations in place.
756	 */
757	ldr	w9, =__rela_offset		// offset to reloc table
758	ldr	w10, =__rela_size		// size of reloc table
759
760	mov_q	x11, KIMAGE_VADDR		// default virtual offset
761	add	x11, x11, x23			// actual virtual offset
762	add	x9, x9, x11			// __va(.rela)
763	add	x10, x9, x10			// __va(.rela) + sizeof(.rela)
764
7650:	cmp	x9, x10
766	b.hs	1f
767	ldp	x12, x13, [x9], #24
768	ldr	x14, [x9, #-8]
769	cmp	w13, #R_AARCH64_RELATIVE
770	b.ne	0b
771	add	x14, x14, x23			// relocate
772	str	x14, [x12, x23]
773	b	0b
774
7751:
776#ifdef CONFIG_RELR
777	/*
778	 * Apply RELR relocations.
779	 *
780	 * RELR is a compressed format for storing relative relocations. The
781	 * encoded sequence of entries looks like:
782	 * [ AAAAAAAA BBBBBBB1 BBBBBBB1 ... AAAAAAAA BBBBBB1 ... ]
783	 *
784	 * i.e. start with an address, followed by any number of bitmaps. The
785	 * address entry encodes 1 relocation. The subsequent bitmap entries
786	 * encode up to 63 relocations each, at subsequent offsets following
787	 * the last address entry.
788	 *
789	 * The bitmap entries must have 1 in the least significant bit. The
790	 * assumption here is that an address cannot have 1 in lsb. Odd
791	 * addresses are not supported. Any odd addresses are stored in the RELA
792	 * section, which is handled above.
793	 *
794	 * Excluding the least significant bit in the bitmap, each non-zero
795	 * bit in the bitmap represents a relocation to be applied to
796	 * a corresponding machine word that follows the base address
797	 * word. The second least significant bit represents the machine
798	 * word immediately following the initial address, and each bit
799	 * that follows represents the next word, in linear order. As such,
800	 * a single bitmap can encode up to 63 relocations in a 64-bit object.
801	 *
802	 * In this implementation we store the address of the next RELR table
803	 * entry in x9, the address being relocated by the current address or
804	 * bitmap entry in x13 and the address being relocated by the current
805	 * bit in x14.
806	 *
807	 * Because addends are stored in place in the binary, RELR relocations
808	 * cannot be applied idempotently. We use x24 to keep track of the
809	 * currently applied displacement so that we can correctly relocate if
810	 * __relocate_kernel is called twice with non-zero displacements (i.e.
811	 * if there is both a physical misalignment and a KASLR displacement).
812	 */
813	ldr	w9, =__relr_offset		// offset to reloc table
814	ldr	w10, =__relr_size		// size of reloc table
815	add	x9, x9, x11			// __va(.relr)
816	add	x10, x9, x10			// __va(.relr) + sizeof(.relr)
817
818	sub	x15, x23, x24			// delta from previous offset
819	cbz	x15, 7f				// nothing to do if unchanged
820	mov	x24, x23			// save new offset
821
8222:	cmp	x9, x10
823	b.hs	7f
824	ldr	x11, [x9], #8
825	tbnz	x11, #0, 3f			// branch to handle bitmaps
826	add	x13, x11, x23
827	ldr	x12, [x13]			// relocate address entry
828	add	x12, x12, x15
829	str	x12, [x13], #8			// adjust to start of bitmap
830	b	2b
831
8323:	mov	x14, x13
8334:	lsr	x11, x11, #1
834	cbz	x11, 6f
835	tbz	x11, #0, 5f			// skip bit if not set
836	ldr	x12, [x14]			// relocate bit
837	add	x12, x12, x15
838	str	x12, [x14]
839
8405:	add	x14, x14, #8			// move to next bit's address
841	b	4b
842
8436:	/*
844	 * Move to the next bitmap's address. 8 is the word size, and 63 is the
845	 * number of significant bits in a bitmap entry.
846	 */
847	add	x13, x13, #(8 * 63)
848	b	2b
849
8507:
851#endif
852	ret
853
854SYM_FUNC_END(__relocate_kernel)
855#endif
856
857SYM_FUNC_START_LOCAL(__primary_switch)
858#ifdef CONFIG_RANDOMIZE_BASE
859	mov	x19, x0				// preserve new SCTLR_EL1 value
860	mrs	x20, sctlr_el1			// preserve old SCTLR_EL1 value
861#endif
862
863	adrp	x1, init_pg_dir
864	bl	__enable_mmu
865#ifdef CONFIG_RELOCATABLE
866#ifdef CONFIG_RELR
867	mov	x24, #0				// no RELR displacement yet
868#endif
869	bl	__relocate_kernel
870#ifdef CONFIG_RANDOMIZE_BASE
871	ldr	x8, =__primary_switched
872	adrp	x0, __PHYS_OFFSET
873	blr	x8
874
875	/*
876	 * If we return here, we have a KASLR displacement in x23 which we need
877	 * to take into account by discarding the current kernel mapping and
878	 * creating a new one.
879	 */
880	pre_disable_mmu_workaround
881	msr	sctlr_el1, x20			// disable the MMU
882	isb
883	bl	__create_page_tables		// recreate kernel mapping
884
885	tlbi	vmalle1				// Remove any stale TLB entries
886	dsb	nsh
887	isb
888
889	set_sctlr_el1	x19			// re-enable the MMU
890
891	bl	__relocate_kernel
892#endif
893#endif
894	ldr	x8, =__primary_switched
895	adrp	x0, __PHYS_OFFSET
896	br	x8
897SYM_FUNC_END(__primary_switch)
898