• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/ptrace.h>
20#include <asm/asm-offsets.h>
21#include <asm/cache.h>
22#include <asm/cputype.h>
23#include <asm/el2_setup.h>
24#include <asm/elf.h>
25#include <asm/image.h>
26#include <asm/kernel-pgtable.h>
27#include <asm/kvm_arm.h>
28#include <asm/memory.h>
29#include <asm/pgtable-hwdef.h>
30#include <asm/page.h>
31#include <asm/scs.h>
32#include <asm/smp.h>
33#include <asm/sysreg.h>
34#include <asm/thread_info.h>
35#include <asm/virt.h>
36
37#include "efi-header.S"
38
39#define __PHYS_OFFSET	KERNEL_START
40
41#if (PAGE_OFFSET & 0x1fffff) != 0
42#error PAGE_OFFSET must be at least 2MB aligned
43#endif
44
45/*
46 * Kernel startup entry point.
47 * ---------------------------
48 *
49 * The requirements are:
50 *   MMU = off, D-cache = off, I-cache = on or off,
51 *   x0 = physical address to the FDT blob.
52 *
53 * This code is mostly position independent so you call this at
54 * __pa(PAGE_OFFSET).
55 *
56 * Note that the callee-saved registers are used for storing variables
57 * that are useful before the MMU is enabled. The allocations are described
58 * in the entry routines.
59 */
60	__HEAD
61_head:
62	/*
63	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
64	 */
65#ifdef CONFIG_EFI
66	/*
67	 * This add instruction has no meaningful effect except that
68	 * its opcode forms the magic "MZ" signature required by UEFI.
69	 */
70	add	x13, x18, #0x16
71	b	primary_entry
72#else
73	b	primary_entry			// branch to kernel start, magic
74	.long	0				// reserved
75#endif
76	.quad	0				// Image load offset from start of RAM, little-endian
77	le64sym	_kernel_size_le			// Effective size of kernel image, little-endian
78	le64sym	_kernel_flags_le		// Informative flags, little-endian
79	.quad	0				// reserved
80	.quad	0				// reserved
81	.quad	0				// reserved
82	.ascii	ARM64_IMAGE_MAGIC		// Magic number
83#ifdef CONFIG_EFI
84	.long	pe_header - _head		// Offset to the PE header.
85
86pe_header:
87	__EFI_PE_HEADER
88#else
89	.long	0				// reserved
90#endif
91
92	__INIT
93
94	/*
95	 * The following callee saved general purpose registers are used on the
96	 * primary lowlevel boot path:
97	 *
98	 *  Register   Scope                      Purpose
99	 *  x21        primary_entry() .. start_kernel()        FDT pointer passed at boot in x0
100	 *  x23        primary_entry() .. start_kernel()        physical misalignment/KASLR offset
101	 *  x28        __create_page_tables()                   callee preserved temp register
102	 *  x19/x20    __primary_switch()                       callee preserved temp registers
103	 *  x24        __primary_switch() .. relocate_kernel()  current RELR displacement
104	 */
105SYM_CODE_START(primary_entry)
106	bl	preserve_boot_args
107	bl	init_kernel_el			// w0=cpu_boot_mode
108	adrp	x23, __PHYS_OFFSET
109	and	x23, x23, MIN_KIMG_ALIGN - 1	// KASLR offset, defaults to 0
110	bl	set_cpu_boot_mode_flag
111	bl	__create_page_tables
112	/*
113	 * The following calls CPU setup code, see arch/arm64/mm/proc.S for
114	 * details.
115	 * On return, the CPU will be ready for the MMU to be turned on and
116	 * the TCR will have been set.
117	 */
118	bl	__cpu_setup			// initialise processor
119	b	__primary_switch
120SYM_CODE_END(primary_entry)
121
122/*
123 * Preserve the arguments passed by the bootloader in x0 .. x3
124 */
125SYM_CODE_START_LOCAL(preserve_boot_args)
126	mov	x21, x0				// x21=FDT
127
128	adr_l	x0, boot_args			// record the contents of
129	stp	x21, x1, [x0]			// x0 .. x3 at kernel entry
130	stp	x2, x3, [x0, #16]
131
132	dmb	sy				// needed before dc ivac with
133						// MMU off
134
135	mov	x1, #0x20			// 4 x 8 bytes
136	b	__inval_dcache_area		// tail call
137SYM_CODE_END(preserve_boot_args)
138
139/*
140 * Macro to create a table entry to the next page.
141 *
142 *	tbl:	page table address
143 *	virt:	virtual address
144 *	shift:	#imm page table shift
145 *	ptrs:	#imm pointers per table page
146 *
147 * Preserves:	virt
148 * Corrupts:	ptrs, tmp1, tmp2
149 * Returns:	tbl -> next level table page address
150 */
151	.macro	create_table_entry, tbl, virt, shift, ptrs, tmp1, tmp2
152	add	\tmp1, \tbl, #PAGE_SIZE
153	phys_to_pte \tmp2, \tmp1
154	orr	\tmp2, \tmp2, #PMD_TYPE_TABLE	// address of next table and entry type
155	lsr	\tmp1, \virt, #\shift
156	sub	\ptrs, \ptrs, #1
157	and	\tmp1, \tmp1, \ptrs		// table index
158	str	\tmp2, [\tbl, \tmp1, lsl #3]
159	add	\tbl, \tbl, #PAGE_SIZE		// next level table page
160	.endm
161
162/*
163 * Macro to populate page table entries, these entries can be pointers to the next level
164 * or last level entries pointing to physical memory.
165 *
166 *	tbl:	page table address
167 *	rtbl:	pointer to page table or physical memory
168 *	index:	start index to write
169 *	eindex:	end index to write - [index, eindex] written to
170 *	flags:	flags for pagetable entry to or in
171 *	inc:	increment to rtbl between each entry
172 *	tmp1:	temporary variable
173 *
174 * Preserves:	tbl, eindex, flags, inc
175 * Corrupts:	index, tmp1
176 * Returns:	rtbl
177 */
178	.macro populate_entries, tbl, rtbl, index, eindex, flags, inc, tmp1
179.Lpe\@:	phys_to_pte \tmp1, \rtbl
180	orr	\tmp1, \tmp1, \flags	// tmp1 = table entry
181	str	\tmp1, [\tbl, \index, lsl #3]
182	add	\rtbl, \rtbl, \inc	// rtbl = pa next level
183	add	\index, \index, #1
184	cmp	\index, \eindex
185	b.ls	.Lpe\@
186	.endm
187
188/*
189 * Compute indices of table entries from virtual address range. If multiple entries
190 * were needed in the previous page table level then the next page table level is assumed
191 * to be composed of multiple pages. (This effectively scales the end index).
192 *
193 *	vstart:	virtual address of start of range
194 *	vend:	virtual address of end of range - we map [vstart, vend]
195 *	shift:	shift used to transform virtual address into index
196 *	ptrs:	number of entries in page table
197 *	istart:	index in table corresponding to vstart
198 *	iend:	index in table corresponding to vend
199 *	count:	On entry: how many extra entries were required in previous level, scales
200 *			  our end index.
201 *		On exit: returns how many extra entries required for next page table level
202 *
203 * Preserves:	vstart, vend, shift, ptrs
204 * Returns:	istart, iend, count
205 */
206	.macro compute_indices, vstart, vend, shift, ptrs, istart, iend, count
207	lsr	\iend, \vend, \shift
208	mov	\istart, \ptrs
209	sub	\istart, \istart, #1
210	and	\iend, \iend, \istart	// iend = (vend >> shift) & (ptrs - 1)
211	mov	\istart, \ptrs
212	mul	\istart, \istart, \count
213	add	\iend, \iend, \istart	// iend += (count - 1) * ptrs
214					// our entries span multiple tables
215
216	lsr	\istart, \vstart, \shift
217	mov	\count, \ptrs
218	sub	\count, \count, #1
219	and	\istart, \istart, \count
220
221	sub	\count, \iend, \istart
222	.endm
223
224/*
225 * Map memory for specified virtual address range. Each level of page table needed supports
226 * multiple entries. If a level requires n entries the next page table level is assumed to be
227 * formed from n pages.
228 *
229 *	tbl:	location of page table
230 *	rtbl:	address to be used for first level page table entry (typically tbl + PAGE_SIZE)
231 *	vstart:	virtual address of start of range
232 *	vend:	virtual address of end of range - we map [vstart, vend - 1]
233 *	flags:	flags to use to map last level entries
234 *	phys:	physical address corresponding to vstart - physical memory is contiguous
235 *	pgds:	the number of pgd entries
236 *
237 * Temporaries:	istart, iend, tmp, count, sv - these need to be different registers
238 * Preserves:	vstart, flags
239 * Corrupts:	tbl, rtbl, vend, istart, iend, tmp, count, sv
240 */
241	.macro map_memory, tbl, rtbl, vstart, vend, flags, phys, pgds, istart, iend, tmp, count, sv
242	sub \vend, \vend, #1
243	add \rtbl, \tbl, #PAGE_SIZE
244	mov \sv, \rtbl
245	mov \count, #0
246	compute_indices \vstart, \vend, #PGDIR_SHIFT, \pgds, \istart, \iend, \count
247	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
248	mov \tbl, \sv
249	mov \sv, \rtbl
250
251#if SWAPPER_PGTABLE_LEVELS > 3
252	compute_indices \vstart, \vend, #PUD_SHIFT, #PTRS_PER_PUD, \istart, \iend, \count
253	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
254	mov \tbl, \sv
255	mov \sv, \rtbl
256#endif
257
258#if SWAPPER_PGTABLE_LEVELS > 2
259	compute_indices \vstart, \vend, #SWAPPER_TABLE_SHIFT, #PTRS_PER_PMD, \istart, \iend, \count
260	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
261	mov \tbl, \sv
262#endif
263
264	compute_indices \vstart, \vend, #SWAPPER_BLOCK_SHIFT, #PTRS_PER_PTE, \istart, \iend, \count
265	bic \count, \phys, #SWAPPER_BLOCK_SIZE - 1
266	populate_entries \tbl, \count, \istart, \iend, \flags, #SWAPPER_BLOCK_SIZE, \tmp
267	.endm
268
269/*
270 * Setup the initial page tables. We only setup the barest amount which is
271 * required to get the kernel running. The following sections are required:
272 *   - identity mapping to enable the MMU (low address, TTBR0)
273 *   - first few MB of the kernel linear mapping to jump to once the MMU has
274 *     been enabled
275 */
276SYM_FUNC_START_LOCAL(__create_page_tables)
277	mov	x28, lr
278
279	/*
280	 * Invalidate the init page tables to avoid potential dirty cache lines
281	 * being evicted. Other page tables are allocated in rodata as part of
282	 * the kernel image, and thus are clean to the PoC per the boot
283	 * protocol.
284	 */
285	adrp	x0, init_pg_dir
286	adrp	x1, init_pg_end
287	sub	x1, x1, x0
288	bl	__inval_dcache_area
289
290	/*
291	 * Clear the init page tables.
292	 */
293	adrp	x0, init_pg_dir
294	adrp	x1, init_pg_end
295	sub	x1, x1, x0
2961:	stp	xzr, xzr, [x0], #16
297	stp	xzr, xzr, [x0], #16
298	stp	xzr, xzr, [x0], #16
299	stp	xzr, xzr, [x0], #16
300	subs	x1, x1, #64
301	b.ne	1b
302
303	mov	x7, SWAPPER_MM_MMUFLAGS
304
305	/*
306	 * Create the identity mapping.
307	 */
308	adrp	x0, idmap_pg_dir
309	adrp	x3, __idmap_text_start		// __pa(__idmap_text_start)
310
311#ifdef CONFIG_ARM64_VA_BITS_52
312	mrs_s	x6, SYS_ID_AA64MMFR2_EL1
313	and	x6, x6, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
314	mov	x5, #52
315	cbnz	x6, 1f
316#endif
317	mov	x5, #VA_BITS_MIN
3181:
319	adr_l	x6, vabits_actual
320	str	x5, [x6]
321	dmb	sy
322	dc	ivac, x6		// Invalidate potentially stale cache line
323
324	/*
325	 * VA_BITS may be too small to allow for an ID mapping to be created
326	 * that covers system RAM if that is located sufficiently high in the
327	 * physical address space. So for the ID map, use an extended virtual
328	 * range in that case, and configure an additional translation level
329	 * if needed.
330	 *
331	 * Calculate the maximum allowed value for TCR_EL1.T0SZ so that the
332	 * entire ID map region can be mapped. As T0SZ == (64 - #bits used),
333	 * this number conveniently equals the number of leading zeroes in
334	 * the physical address of __idmap_text_end.
335	 */
336	adrp	x5, __idmap_text_end
337	clz	x5, x5
338	cmp	x5, TCR_T0SZ(VA_BITS_MIN) // default T0SZ small enough?
339	b.ge	1f			// .. then skip VA range extension
340
341	adr_l	x6, idmap_t0sz
342	str	x5, [x6]
343	dmb	sy
344	dc	ivac, x6		// Invalidate potentially stale cache line
345
346#if (VA_BITS < 48)
347#define EXTRA_SHIFT	(PGDIR_SHIFT + PAGE_SHIFT - 3)
348#define EXTRA_PTRS	(1 << (PHYS_MASK_SHIFT - EXTRA_SHIFT))
349
350	/*
351	 * If VA_BITS < 48, we have to configure an additional table level.
352	 * First, we have to verify our assumption that the current value of
353	 * VA_BITS was chosen such that all translation levels are fully
354	 * utilised, and that lowering T0SZ will always result in an additional
355	 * translation level to be configured.
356	 */
357#if VA_BITS != EXTRA_SHIFT
358#error "Mismatch between VA_BITS and page size/number of translation levels"
359#endif
360
361	mov	x4, EXTRA_PTRS
362	create_table_entry x0, x3, EXTRA_SHIFT, x4, x5, x6
363#else
364	/*
365	 * If VA_BITS == 48, we don't have to configure an additional
366	 * translation level, but the top-level table has more entries.
367	 */
368	mov	x4, #1 << (PHYS_MASK_SHIFT - PGDIR_SHIFT)
369	str_l	x4, idmap_ptrs_per_pgd, x5
370#endif
3711:
372	ldr_l	x4, idmap_ptrs_per_pgd
373	mov	x5, x3				// __pa(__idmap_text_start)
374	adr_l	x6, __idmap_text_end		// __pa(__idmap_text_end)
375
376	map_memory x0, x1, x3, x6, x7, x3, x4, x10, x11, x12, x13, x14
377
378	/*
379	 * Map the kernel image (starting with PHYS_OFFSET).
380	 */
381	adrp	x0, init_pg_dir
382	mov_q	x5, KIMAGE_VADDR		// compile time __va(_text)
383	add	x5, x5, x23			// add KASLR displacement
384	mov	x4, PTRS_PER_PGD
385	adrp	x6, _end			// runtime __pa(_end)
386	adrp	x3, _text			// runtime __pa(_text)
387	sub	x6, x6, x3			// _end - _text
388	add	x6, x6, x5			// runtime __va(_end)
389
390	map_memory x0, x1, x5, x6, x7, x3, x4, x10, x11, x12, x13, x14
391
392	/*
393	 * Since the page tables have been populated with non-cacheable
394	 * accesses (MMU disabled), invalidate those tables again to
395	 * remove any speculatively loaded cache lines.
396	 */
397	dmb	sy
398
399	adrp	x0, idmap_pg_dir
400	adrp	x1, idmap_pg_end
401	sub	x1, x1, x0
402	bl	__inval_dcache_area
403
404	adrp	x0, init_pg_dir
405	adrp	x1, init_pg_end
406	sub	x1, x1, x0
407	bl	__inval_dcache_area
408
409	ret	x28
410SYM_FUNC_END(__create_page_tables)
411
412/*
413 * The following fragment of code is executed with the MMU enabled.
414 *
415 *   x0 = __PHYS_OFFSET
416 */
417SYM_FUNC_START_LOCAL(__primary_switched)
418	adrp	x4, init_thread_union
419	add	sp, x4, #THREAD_SIZE
420	adr_l	x5, init_task
421	msr	sp_el0, x5			// Save thread_info
422
423	adr_l	x8, vectors			// load VBAR_EL1 with virtual
424	msr	vbar_el1, x8			// vector table address
425	isb
426
427	stp	xzr, x30, [sp, #-16]!
428	mov	x29, sp
429
430#ifdef CONFIG_SHADOW_CALL_STACK
431	adr_l	scs_sp, init_shadow_call_stack	// Set shadow call stack
432#endif
433
434	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
435
436	ldr_l	x4, kimage_vaddr		// Save the offset between
437	sub	x4, x4, x0			// the kernel virtual and
438	str_l	x4, kimage_voffset, x5		// physical mappings
439
440	// Clear BSS
441	adr_l	x0, __bss_start
442	mov	x1, xzr
443	adr_l	x2, __bss_stop
444	sub	x2, x2, x0
445	bl	__pi_memset
446	dsb	ishst				// Make zero page visible to PTW
447
448#if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
449	bl	kasan_early_init
450#endif
451	mov	x0, x21				// pass FDT address in x0
452	bl	early_fdt_map			// Try mapping the FDT early
453	bl	init_feature_override		// Parse cpu feature overrides
454#ifdef CONFIG_RANDOMIZE_BASE
455	tst	x23, ~(MIN_KIMG_ALIGN - 1)	// already running randomized?
456	b.ne	0f
457	bl	kaslr_early_init		// parse FDT for KASLR options
458	cbz	x0, 0f				// KASLR disabled? just proceed
459	orr	x23, x23, x0			// record KASLR offset
460	ldp	x29, x30, [sp], #16		// we must enable KASLR, return
461	ret					// to __primary_switch()
4620:
463#endif
464	bl	switch_to_vhe			// Prefer VHE if possible
465	add	sp, sp, #16
466	mov	x29, #0
467	mov	x30, #0
468	b	start_kernel
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	mov_q	x0, INIT_SCTLR_EL1_MMU_OFF
497	msr	sctlr_el1, x0
498
499	mrs	x0, CurrentEL
500	cmp	x0, #CurrentEL_EL2
501	b.eq	init_el2
502
503SYM_INNER_LABEL(init_el1, SYM_L_LOCAL)
504	isb
505	mov_q	x0, INIT_PSTATE_EL1
506	msr	spsr_el1, x0
507	msr	elr_el1, lr
508	mov	w0, #BOOT_CPU_MODE_EL1
509	eret
510
511SYM_INNER_LABEL(init_el2, SYM_L_LOCAL)
512	mov_q	x0, HCR_HOST_NVHE_FLAGS
513	msr	hcr_el2, x0
514	isb
515
516	init_el2_state
517
518	/* Hypervisor stub */
519	adr_l	x0, __hyp_stub_vectors
520	msr	vbar_el2, x0
521	isb
522
523	msr	elr_el2, lr
524	mov	w0, #BOOT_CPU_MODE_EL2
525	eret
526SYM_FUNC_END(init_kernel_el)
527
528/*
529 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
530 * in w0. See arch/arm64/include/asm/virt.h for more info.
531 */
532SYM_FUNC_START_LOCAL(set_cpu_boot_mode_flag)
533	adr_l	x1, __boot_cpu_mode
534	cmp	w0, #BOOT_CPU_MODE_EL2
535	b.ne	1f
536	add	x1, x1, #4
5371:	str	w0, [x1]			// This CPU has booted in EL1
538	dmb	sy
539	dc	ivac, x1			// Invalidate potentially stale cache line
540	ret
541SYM_FUNC_END(set_cpu_boot_mode_flag)
542
543/*
544 * These values are written with the MMU off, but read with the MMU on.
545 * Writers will invalidate the corresponding address, discarding up to a
546 * 'Cache Writeback Granule' (CWG) worth of data. The linker script ensures
547 * sufficient alignment that the CWG doesn't overlap another section.
548 */
549	.pushsection ".mmuoff.data.write", "aw"
550/*
551 * We need to find out the CPU boot mode long after boot, so we need to
552 * store it in a writable variable.
553 *
554 * This is not in .bss, because we set it sufficiently early that the boot-time
555 * zeroing of .bss would clobber it.
556 */
557SYM_DATA_START(__boot_cpu_mode)
558	.long	BOOT_CPU_MODE_EL2
559	.long	BOOT_CPU_MODE_EL1
560SYM_DATA_END(__boot_cpu_mode)
561/*
562 * The booting CPU updates the failed status @__early_cpu_boot_status,
563 * with MMU turned off.
564 */
565SYM_DATA_START(__early_cpu_boot_status)
566	.quad 	0
567SYM_DATA_END(__early_cpu_boot_status)
568
569	.popsection
570
571	/*
572	 * This provides a "holding pen" for platforms to hold all secondary
573	 * cores are held until we're ready for them to initialise.
574	 */
575SYM_FUNC_START(secondary_holding_pen)
576	bl	init_kernel_el			// w0=cpu_boot_mode
577	bl	set_cpu_boot_mode_flag
578	mrs	x0, mpidr_el1
579	mov_q	x1, MPIDR_HWID_BITMASK
580	and	x0, x0, x1
581	adr_l	x3, secondary_holding_pen_release
582pen:	ldr	x4, [x3]
583	cmp	x4, x0
584	b.eq	secondary_startup
585	wfe
586	b	pen
587SYM_FUNC_END(secondary_holding_pen)
588
589	/*
590	 * Secondary entry point that jumps straight into the kernel. Only to
591	 * be used where CPUs are brought online dynamically by the kernel.
592	 */
593SYM_FUNC_START(secondary_entry)
594	bl	init_kernel_el			// w0=cpu_boot_mode
595	bl	set_cpu_boot_mode_flag
596	b	secondary_startup
597SYM_FUNC_END(secondary_entry)
598
599SYM_FUNC_START_LOCAL(secondary_startup)
600	/*
601	 * Common entry point for secondary CPUs.
602	 */
603	bl	switch_to_vhe
604	bl	__cpu_secondary_check52bitva
605	bl	__cpu_setup			// initialise processor
606	adrp	x1, swapper_pg_dir
607	bl	__enable_mmu
608	ldr	x8, =__secondary_switched
609	br	x8
610SYM_FUNC_END(secondary_startup)
611
612SYM_FUNC_START_LOCAL(__secondary_switched)
613	adr_l	x5, vectors
614	msr	vbar_el1, x5
615	isb
616
617	adr_l	x0, secondary_data
618	ldr	x1, [x0, #CPU_BOOT_STACK]	// get secondary_data.stack
619	cbz	x1, __secondary_too_slow
620	mov	sp, x1
621	ldr	x2, [x0, #CPU_BOOT_TASK]
622	cbz	x2, __secondary_too_slow
623	msr	sp_el0, x2
624	scs_load_current
625	mov	x29, #0
626	mov	x30, #0
627
628#ifdef CONFIG_ARM64_PTR_AUTH
629	ptrauth_keys_init_cpu x2, x3, x4, x5
630#endif
631
632	b	secondary_start_kernel
633SYM_FUNC_END(__secondary_switched)
634
635SYM_FUNC_START_LOCAL(__secondary_too_slow)
636	wfe
637	wfi
638	b	__secondary_too_slow
639SYM_FUNC_END(__secondary_too_slow)
640
641/*
642 * The booting CPU updates the failed status @__early_cpu_boot_status,
643 * with MMU turned off.
644 *
645 * update_early_cpu_boot_status tmp, status
646 *  - Corrupts tmp1, tmp2
647 *  - Writes 'status' to __early_cpu_boot_status and makes sure
648 *    it is committed to memory.
649 */
650
651	.macro	update_early_cpu_boot_status status, tmp1, tmp2
652	mov	\tmp2, #\status
653	adr_l	\tmp1, __early_cpu_boot_status
654	str	\tmp2, [\tmp1]
655	dmb	sy
656	dc	ivac, \tmp1			// Invalidate potentially stale cache line
657	.endm
658
659/*
660 * Enable the MMU.
661 *
662 *  x0  = SCTLR_EL1 value for turning on the MMU.
663 *  x1  = TTBR1_EL1 value
664 *
665 * Returns to the caller via x30/lr. This requires the caller to be covered
666 * by the .idmap.text section.
667 *
668 * Checks if the selected granule size is supported by the CPU.
669 * If it isn't, park the CPU
670 */
671SYM_FUNC_START(__enable_mmu)
672	mrs	x2, ID_AA64MMFR0_EL1
673	ubfx	x2, x2, #ID_AA64MMFR0_TGRAN_SHIFT, 4
674	cmp     x2, #ID_AA64MMFR0_TGRAN_SUPPORTED_MIN
675	b.lt    __no_granule_support
676	cmp     x2, #ID_AA64MMFR0_TGRAN_SUPPORTED_MAX
677	b.gt    __no_granule_support
678	update_early_cpu_boot_status 0, x2, x3
679	adrp	x2, idmap_pg_dir
680	phys_to_ttbr x1, x1
681	phys_to_ttbr x2, x2
682	msr	ttbr0_el1, x2			// load TTBR0
683	offset_ttbr1 x1, x3
684	msr	ttbr1_el1, x1			// load TTBR1
685	isb
686
687	set_sctlr_el1	x0
688
689	ret
690SYM_FUNC_END(__enable_mmu)
691
692SYM_FUNC_START(__cpu_secondary_check52bitva)
693#ifdef CONFIG_ARM64_VA_BITS_52
694	ldr_l	x0, vabits_actual
695	cmp	x0, #52
696	b.ne	2f
697
698	mrs_s	x0, SYS_ID_AA64MMFR2_EL1
699	and	x0, x0, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
700	cbnz	x0, 2f
701
702	update_early_cpu_boot_status \
703		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_52_BIT_VA, x0, x1
7041:	wfe
705	wfi
706	b	1b
707
708#endif
7092:	ret
710SYM_FUNC_END(__cpu_secondary_check52bitva)
711
712SYM_FUNC_START_LOCAL(__no_granule_support)
713	/* Indicate that this CPU can't boot and is stuck in the kernel */
714	update_early_cpu_boot_status \
715		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_NO_GRAN, x1, x2
7161:
717	wfe
718	wfi
719	b	1b
720SYM_FUNC_END(__no_granule_support)
721
722#ifdef CONFIG_RELOCATABLE
723SYM_FUNC_START_LOCAL(__relocate_kernel)
724	/*
725	 * Iterate over each entry in the relocation table, and apply the
726	 * relocations in place.
727	 */
728	ldr	w9, =__rela_offset		// offset to reloc table
729	ldr	w10, =__rela_size		// size of reloc table
730
731	mov_q	x11, KIMAGE_VADDR		// default virtual offset
732	add	x11, x11, x23			// actual virtual offset
733	add	x9, x9, x11			// __va(.rela)
734	add	x10, x9, x10			// __va(.rela) + sizeof(.rela)
735
7360:	cmp	x9, x10
737	b.hs	1f
738	ldp	x12, x13, [x9], #24
739	ldr	x14, [x9, #-8]
740	cmp	w13, #R_AARCH64_RELATIVE
741	b.ne	0b
742	add	x14, x14, x23			// relocate
743	str	x14, [x12, x23]
744	b	0b
745
7461:
747#ifdef CONFIG_RELR
748	/*
749	 * Apply RELR relocations.
750	 *
751	 * RELR is a compressed format for storing relative relocations. The
752	 * encoded sequence of entries looks like:
753	 * [ AAAAAAAA BBBBBBB1 BBBBBBB1 ... AAAAAAAA BBBBBB1 ... ]
754	 *
755	 * i.e. start with an address, followed by any number of bitmaps. The
756	 * address entry encodes 1 relocation. The subsequent bitmap entries
757	 * encode up to 63 relocations each, at subsequent offsets following
758	 * the last address entry.
759	 *
760	 * The bitmap entries must have 1 in the least significant bit. The
761	 * assumption here is that an address cannot have 1 in lsb. Odd
762	 * addresses are not supported. Any odd addresses are stored in the RELA
763	 * section, which is handled above.
764	 *
765	 * Excluding the least significant bit in the bitmap, each non-zero
766	 * bit in the bitmap represents a relocation to be applied to
767	 * a corresponding machine word that follows the base address
768	 * word. The second least significant bit represents the machine
769	 * word immediately following the initial address, and each bit
770	 * that follows represents the next word, in linear order. As such,
771	 * a single bitmap can encode up to 63 relocations in a 64-bit object.
772	 *
773	 * In this implementation we store the address of the next RELR table
774	 * entry in x9, the address being relocated by the current address or
775	 * bitmap entry in x13 and the address being relocated by the current
776	 * bit in x14.
777	 *
778	 * Because addends are stored in place in the binary, RELR relocations
779	 * cannot be applied idempotently. We use x24 to keep track of the
780	 * currently applied displacement so that we can correctly relocate if
781	 * __relocate_kernel is called twice with non-zero displacements (i.e.
782	 * if there is both a physical misalignment and a KASLR displacement).
783	 */
784	ldr	w9, =__relr_offset		// offset to reloc table
785	ldr	w10, =__relr_size		// size of reloc table
786	add	x9, x9, x11			// __va(.relr)
787	add	x10, x9, x10			// __va(.relr) + sizeof(.relr)
788
789	sub	x15, x23, x24			// delta from previous offset
790	cbz	x15, 7f				// nothing to do if unchanged
791	mov	x24, x23			// save new offset
792
7932:	cmp	x9, x10
794	b.hs	7f
795	ldr	x11, [x9], #8
796	tbnz	x11, #0, 3f			// branch to handle bitmaps
797	add	x13, x11, x23
798	ldr	x12, [x13]			// relocate address entry
799	add	x12, x12, x15
800	str	x12, [x13], #8			// adjust to start of bitmap
801	b	2b
802
8033:	mov	x14, x13
8044:	lsr	x11, x11, #1
805	cbz	x11, 6f
806	tbz	x11, #0, 5f			// skip bit if not set
807	ldr	x12, [x14]			// relocate bit
808	add	x12, x12, x15
809	str	x12, [x14]
810
8115:	add	x14, x14, #8			// move to next bit's address
812	b	4b
813
8146:	/*
815	 * Move to the next bitmap's address. 8 is the word size, and 63 is the
816	 * number of significant bits in a bitmap entry.
817	 */
818	add	x13, x13, #(8 * 63)
819	b	2b
820
8217:
822#endif
823	ret
824
825SYM_FUNC_END(__relocate_kernel)
826#endif
827
828SYM_FUNC_START_LOCAL(__primary_switch)
829#ifdef CONFIG_RANDOMIZE_BASE
830	mov	x19, x0				// preserve new SCTLR_EL1 value
831	mrs	x20, sctlr_el1			// preserve old SCTLR_EL1 value
832#endif
833
834	adrp	x1, init_pg_dir
835	bl	__enable_mmu
836#ifdef CONFIG_RELOCATABLE
837#ifdef CONFIG_RELR
838	mov	x24, #0				// no RELR displacement yet
839#endif
840	bl	__relocate_kernel
841#ifdef CONFIG_RANDOMIZE_BASE
842	ldr	x8, =__primary_switched
843	adrp	x0, __PHYS_OFFSET
844	blr	x8
845
846	/*
847	 * If we return here, we have a KASLR displacement in x23 which we need
848	 * to take into account by discarding the current kernel mapping and
849	 * creating a new one.
850	 */
851	pre_disable_mmu_workaround
852	msr	sctlr_el1, x20			// disable the MMU
853	isb
854	bl	__create_page_tables		// recreate kernel mapping
855
856	tlbi	vmalle1				// Remove any stale TLB entries
857	dsb	nsh
858	isb
859
860	set_sctlr_el1	x19			// re-enable the MMU
861
862	bl	__relocate_kernel
863#endif
864#endif
865	ldr	x8, =__primary_switched
866	adrp	x0, __PHYS_OFFSET
867	br	x8
868SYM_FUNC_END(__primary_switch)
869