xref: /openbmc/linux/arch/arm64/kernel/head.S (revision 8dce88fe)
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 - we map [vstart, vend]
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:	virtual address of start of range
218 *	vend:	virtual address of end of range - we map [vstart, vend - 1]
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, flags
225 * Corrupts:	tbl, rtbl, vend, istart, iend, tmp, count, sv
226 */
227	.macro map_memory, tbl, rtbl, vstart, vend, flags, phys, pgds, istart, iend, tmp, count, sv
228	sub \vend, \vend, #1
229	add \rtbl, \tbl, #PAGE_SIZE
230	mov \sv, \rtbl
231	mov \count, #0
232	compute_indices \vstart, \vend, #PGDIR_SHIFT, \pgds, \istart, \iend, \count
233	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
234	mov \tbl, \sv
235	mov \sv, \rtbl
236
237#if SWAPPER_PGTABLE_LEVELS > 3
238	compute_indices \vstart, \vend, #PUD_SHIFT, #PTRS_PER_PUD, \istart, \iend, \count
239	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
240	mov \tbl, \sv
241	mov \sv, \rtbl
242#endif
243
244#if SWAPPER_PGTABLE_LEVELS > 2
245	compute_indices \vstart, \vend, #SWAPPER_TABLE_SHIFT, #PTRS_PER_PMD, \istart, \iend, \count
246	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
247	mov \tbl, \sv
248#endif
249
250	compute_indices \vstart, \vend, #SWAPPER_BLOCK_SHIFT, #PTRS_PER_PTE, \istart, \iend, \count
251	bic \count, \phys, #SWAPPER_BLOCK_SIZE - 1
252	populate_entries \tbl, \count, \istart, \iend, \flags, #SWAPPER_BLOCK_SIZE, \tmp
253	.endm
254
255/*
256 * Setup the initial page tables. We only setup the barest amount which is
257 * required to get the kernel running. The following sections are required:
258 *   - identity mapping to enable the MMU (low address, TTBR0)
259 *   - first few MB of the kernel linear mapping to jump to once the MMU has
260 *     been enabled
261 */
262SYM_FUNC_START_LOCAL(__create_page_tables)
263	mov	x28, lr
264
265	/*
266	 * Invalidate the init page tables to avoid potential dirty cache lines
267	 * being evicted. Other page tables are allocated in rodata as part of
268	 * the kernel image, and thus are clean to the PoC per the boot
269	 * protocol.
270	 */
271	adrp	x0, init_pg_dir
272	adrp	x1, init_pg_end
273	bl	dcache_inval_poc
274
275	/*
276	 * Clear the init page tables.
277	 */
278	adrp	x0, init_pg_dir
279	adrp	x1, init_pg_end
280	sub	x1, x1, x0
2811:	stp	xzr, xzr, [x0], #16
282	stp	xzr, xzr, [x0], #16
283	stp	xzr, xzr, [x0], #16
284	stp	xzr, xzr, [x0], #16
285	subs	x1, x1, #64
286	b.ne	1b
287
288	mov	x7, SWAPPER_MM_MMUFLAGS
289
290	/*
291	 * Create the identity mapping.
292	 */
293	adrp	x0, idmap_pg_dir
294	adrp	x3, __idmap_text_start		// __pa(__idmap_text_start)
295
296#ifdef CONFIG_ARM64_VA_BITS_52
297	mrs_s	x6, SYS_ID_AA64MMFR2_EL1
298	and	x6, x6, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
299	mov	x5, #52
300	cbnz	x6, 1f
301#endif
302	mov	x5, #VA_BITS_MIN
3031:
304	adr_l	x6, vabits_actual
305	str	x5, [x6]
306	dmb	sy
307	dc	ivac, x6		// Invalidate potentially stale cache line
308
309	/*
310	 * VA_BITS may be too small to allow for an ID mapping to be created
311	 * that covers system RAM if that is located sufficiently high in the
312	 * physical address space. So for the ID map, use an extended virtual
313	 * range in that case, and configure an additional translation level
314	 * if needed.
315	 *
316	 * Calculate the maximum allowed value for TCR_EL1.T0SZ so that the
317	 * entire ID map region can be mapped. As T0SZ == (64 - #bits used),
318	 * this number conveniently equals the number of leading zeroes in
319	 * the physical address of __idmap_text_end.
320	 */
321	adrp	x5, __idmap_text_end
322	clz	x5, x5
323	cmp	x5, TCR_T0SZ(VA_BITS_MIN) // default T0SZ small enough?
324	b.ge	1f			// .. then skip VA range extension
325
326	adr_l	x6, idmap_t0sz
327	str	x5, [x6]
328	dmb	sy
329	dc	ivac, x6		// Invalidate potentially stale cache line
330
331#if (VA_BITS < 48)
332#define EXTRA_SHIFT	(PGDIR_SHIFT + PAGE_SHIFT - 3)
333#define EXTRA_PTRS	(1 << (PHYS_MASK_SHIFT - EXTRA_SHIFT))
334
335	/*
336	 * If VA_BITS < 48, we have to configure an additional table level.
337	 * First, we have to verify our assumption that the current value of
338	 * VA_BITS was chosen such that all translation levels are fully
339	 * utilised, and that lowering T0SZ will always result in an additional
340	 * translation level to be configured.
341	 */
342#if VA_BITS != EXTRA_SHIFT
343#error "Mismatch between VA_BITS and page size/number of translation levels"
344#endif
345
346	mov	x4, EXTRA_PTRS
347	create_table_entry x0, x3, EXTRA_SHIFT, x4, x5, x6
348#else
349	/*
350	 * If VA_BITS == 48, we don't have to configure an additional
351	 * translation level, but the top-level table has more entries.
352	 */
353	mov	x4, #1 << (PHYS_MASK_SHIFT - PGDIR_SHIFT)
354	str_l	x4, idmap_ptrs_per_pgd, x5
355#endif
3561:
357	ldr_l	x4, idmap_ptrs_per_pgd
358	adr_l	x6, __idmap_text_end		// __pa(__idmap_text_end)
359
360	map_memory x0, x1, x3, x6, x7, x3, x4, x10, x11, x12, x13, x14
361
362	/*
363	 * Map the kernel image (starting with PHYS_OFFSET).
364	 */
365	adrp	x0, init_pg_dir
366	mov_q	x5, KIMAGE_VADDR		// compile time __va(_text)
367	add	x5, x5, x23			// add KASLR displacement
368	mov	x4, PTRS_PER_PGD
369	adrp	x6, _end			// runtime __pa(_end)
370	adrp	x3, _text			// runtime __pa(_text)
371	sub	x6, x6, x3			// _end - _text
372	add	x6, x6, x5			// runtime __va(_end)
373
374	map_memory x0, x1, x5, x6, x7, x3, x4, x10, x11, x12, x13, x14
375
376	/*
377	 * Since the page tables have been populated with non-cacheable
378	 * accesses (MMU disabled), invalidate those tables again to
379	 * remove any speculatively loaded cache lines.
380	 */
381	dmb	sy
382
383	adrp	x0, idmap_pg_dir
384	adrp	x1, idmap_pg_end
385	bl	dcache_inval_poc
386
387	adrp	x0, init_pg_dir
388	adrp	x1, init_pg_end
389	bl	dcache_inval_poc
390
391	ret	x28
392SYM_FUNC_END(__create_page_tables)
393
394	/*
395	 * Initialize CPU registers with task-specific and cpu-specific context.
396	 *
397	 * Create a final frame record at task_pt_regs(current)->stackframe, so
398	 * that the unwinder can identify the final frame record of any task by
399	 * its location in the task stack. We reserve the entire pt_regs space
400	 * for consistency with user tasks and kthreads.
401	 */
402	.macro	init_cpu_task tsk, tmp1, tmp2
403	msr	sp_el0, \tsk
404
405	ldr	\tmp1, [\tsk, #TSK_STACK]
406	add	sp, \tmp1, #THREAD_SIZE
407	sub	sp, sp, #PT_REGS_SIZE
408
409	stp	xzr, xzr, [sp, #S_STACKFRAME]
410	add	x29, sp, #S_STACKFRAME
411
412	scs_load \tsk
413
414	adr_l	\tmp1, __per_cpu_offset
415	ldr	w\tmp2, [\tsk, #TSK_TI_CPU]
416	ldr	\tmp1, [\tmp1, \tmp2, lsl #3]
417	set_this_cpu_offset \tmp1
418	.endm
419
420/*
421 * The following fragment of code is executed with the MMU enabled.
422 *
423 *   x0 = __PHYS_OFFSET
424 */
425SYM_FUNC_START_LOCAL(__primary_switched)
426	adr_l	x4, init_task
427	init_cpu_task x4, x5, x6
428
429	adr_l	x8, vectors			// load VBAR_EL1 with virtual
430	msr	vbar_el1, x8			// vector table address
431	isb
432
433	stp	x29, x30, [sp, #-16]!
434	mov	x29, sp
435
436	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
437
438	ldr_l	x4, kimage_vaddr		// Save the offset between
439	sub	x4, x4, x0			// the kernel virtual and
440	str_l	x4, kimage_voffset, x5		// physical mappings
441
442	// Clear BSS
443	adr_l	x0, __bss_start
444	mov	x1, xzr
445	adr_l	x2, __bss_stop
446	sub	x2, x2, x0
447	bl	__pi_memset
448	dsb	ishst				// Make zero page visible to PTW
449
450#if defined(CONFIG_KASAN_GENERIC) || defined(CONFIG_KASAN_SW_TAGS)
451	bl	kasan_early_init
452#endif
453	mov	x0, x21				// pass FDT address in x0
454	bl	early_fdt_map			// Try mapping the FDT early
455	bl	init_feature_override		// Parse cpu feature overrides
456#ifdef CONFIG_RANDOMIZE_BASE
457	tst	x23, ~(MIN_KIMG_ALIGN - 1)	// already running randomized?
458	b.ne	0f
459	bl	kaslr_early_init		// parse FDT for KASLR options
460	cbz	x0, 0f				// KASLR disabled? just proceed
461	orr	x23, x23, x0			// record KASLR offset
462	ldp	x29, x30, [sp], #16		// we must enable KASLR, return
463	ret					// to __primary_switch()
4640:
465#endif
466	bl	switch_to_vhe			// Prefer VHE if possible
467	ldp	x29, x30, [sp], #16
468	bl	start_kernel
469	ASM_BUG()
470SYM_FUNC_END(__primary_switched)
471
472	.pushsection ".rodata", "a"
473SYM_DATA_START(kimage_vaddr)
474	.quad		_text
475SYM_DATA_END(kimage_vaddr)
476EXPORT_SYMBOL(kimage_vaddr)
477	.popsection
478
479/*
480 * end early head section, begin head code that is also used for
481 * hotplug and needs to have the same protections as the text region
482 */
483	.section ".idmap.text","awx"
484
485/*
486 * Starting from EL2 or EL1, configure the CPU to execute at the highest
487 * reachable EL supported by the kernel in a chosen default state. If dropping
488 * from EL2 to EL1, configure EL2 before configuring EL1.
489 *
490 * Since we cannot always rely on ERET synchronizing writes to sysregs (e.g. if
491 * SCTLR_ELx.EOS is clear), we place an ISB prior to ERET.
492 *
493 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in w0 if
494 * booted in EL1 or EL2 respectively.
495 */
496SYM_FUNC_START(init_kernel_el)
497	mrs	x0, CurrentEL
498	cmp	x0, #CurrentEL_EL2
499	b.eq	init_el2
500
501SYM_INNER_LABEL(init_el1, SYM_L_LOCAL)
502	mov_q	x0, INIT_SCTLR_EL1_MMU_OFF
503	msr	sctlr_el1, x0
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	/*
524	 * Fruity CPUs seem to have HCR_EL2.E2H set to RES1,
525	 * making it impossible to start in nVHE mode. Is that
526	 * compliant with the architecture? Absolutely not!
527	 */
528	mrs	x0, hcr_el2
529	and	x0, x0, #HCR_E2H
530	cbz	x0, 1f
531
532	/* Switching to VHE requires a sane SCTLR_EL1 as a start */
533	mov_q	x0, INIT_SCTLR_EL1_MMU_OFF
534	msr_s	SYS_SCTLR_EL12, x0
535
536	/*
537	 * Force an eret into a helper "function", and let it return
538	 * to our original caller... This makes sure that we have
539	 * initialised the basic PSTATE state.
540	 */
541	mov	x0, #INIT_PSTATE_EL2
542	msr	spsr_el1, x0
543	adr	x0, __cpu_stick_to_vhe
544	msr	elr_el1, x0
545	eret
546
5471:
548	mov_q	x0, INIT_SCTLR_EL1_MMU_OFF
549	msr	sctlr_el1, x0
550
551	msr	elr_el2, lr
552	mov	w0, #BOOT_CPU_MODE_EL2
553	eret
554
555__cpu_stick_to_vhe:
556	mov	x0, #HVC_VHE_RESTART
557	hvc	#0
558	mov	x0, #BOOT_CPU_MODE_EL2
559	ret
560SYM_FUNC_END(init_kernel_el)
561
562/*
563 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
564 * in w0. See arch/arm64/include/asm/virt.h for more info.
565 */
566SYM_FUNC_START_LOCAL(set_cpu_boot_mode_flag)
567	adr_l	x1, __boot_cpu_mode
568	cmp	w0, #BOOT_CPU_MODE_EL2
569	b.ne	1f
570	add	x1, x1, #4
5711:	str	w0, [x1]			// Save CPU boot mode
572	dmb	sy
573	dc	ivac, x1			// Invalidate potentially stale cache line
574	ret
575SYM_FUNC_END(set_cpu_boot_mode_flag)
576
577/*
578 * These values are written with the MMU off, but read with the MMU on.
579 * Writers will invalidate the corresponding address, discarding up to a
580 * 'Cache Writeback Granule' (CWG) worth of data. The linker script ensures
581 * sufficient alignment that the CWG doesn't overlap another section.
582 */
583	.pushsection ".mmuoff.data.write", "aw"
584/*
585 * We need to find out the CPU boot mode long after boot, so we need to
586 * store it in a writable variable.
587 *
588 * This is not in .bss, because we set it sufficiently early that the boot-time
589 * zeroing of .bss would clobber it.
590 */
591SYM_DATA_START(__boot_cpu_mode)
592	.long	BOOT_CPU_MODE_EL2
593	.long	BOOT_CPU_MODE_EL1
594SYM_DATA_END(__boot_cpu_mode)
595/*
596 * The booting CPU updates the failed status @__early_cpu_boot_status,
597 * with MMU turned off.
598 */
599SYM_DATA_START(__early_cpu_boot_status)
600	.quad 	0
601SYM_DATA_END(__early_cpu_boot_status)
602
603	.popsection
604
605	/*
606	 * This provides a "holding pen" for platforms to hold all secondary
607	 * cores are held until we're ready for them to initialise.
608	 */
609SYM_FUNC_START(secondary_holding_pen)
610	bl	init_kernel_el			// w0=cpu_boot_mode
611	bl	set_cpu_boot_mode_flag
612	mrs	x0, mpidr_el1
613	mov_q	x1, MPIDR_HWID_BITMASK
614	and	x0, x0, x1
615	adr_l	x3, secondary_holding_pen_release
616pen:	ldr	x4, [x3]
617	cmp	x4, x0
618	b.eq	secondary_startup
619	wfe
620	b	pen
621SYM_FUNC_END(secondary_holding_pen)
622
623	/*
624	 * Secondary entry point that jumps straight into the kernel. Only to
625	 * be used where CPUs are brought online dynamically by the kernel.
626	 */
627SYM_FUNC_START(secondary_entry)
628	bl	init_kernel_el			// w0=cpu_boot_mode
629	bl	set_cpu_boot_mode_flag
630	b	secondary_startup
631SYM_FUNC_END(secondary_entry)
632
633SYM_FUNC_START_LOCAL(secondary_startup)
634	/*
635	 * Common entry point for secondary CPUs.
636	 */
637	bl	switch_to_vhe
638	bl	__cpu_secondary_check52bitva
639	bl	__cpu_setup			// initialise processor
640	adrp	x1, swapper_pg_dir
641	bl	__enable_mmu
642	ldr	x8, =__secondary_switched
643	br	x8
644SYM_FUNC_END(secondary_startup)
645
646SYM_FUNC_START_LOCAL(__secondary_switched)
647	adr_l	x5, vectors
648	msr	vbar_el1, x5
649	isb
650
651	adr_l	x0, secondary_data
652	ldr	x2, [x0, #CPU_BOOT_TASK]
653	cbz	x2, __secondary_too_slow
654
655	init_cpu_task x2, x1, x3
656
657#ifdef CONFIG_ARM64_PTR_AUTH
658	ptrauth_keys_init_cpu x2, x3, x4, x5
659#endif
660
661	bl	secondary_start_kernel
662	ASM_BUG()
663SYM_FUNC_END(__secondary_switched)
664
665SYM_FUNC_START_LOCAL(__secondary_too_slow)
666	wfe
667	wfi
668	b	__secondary_too_slow
669SYM_FUNC_END(__secondary_too_slow)
670
671/*
672 * The booting CPU updates the failed status @__early_cpu_boot_status,
673 * with MMU turned off.
674 *
675 * update_early_cpu_boot_status tmp, status
676 *  - Corrupts tmp1, tmp2
677 *  - Writes 'status' to __early_cpu_boot_status and makes sure
678 *    it is committed to memory.
679 */
680
681	.macro	update_early_cpu_boot_status status, tmp1, tmp2
682	mov	\tmp2, #\status
683	adr_l	\tmp1, __early_cpu_boot_status
684	str	\tmp2, [\tmp1]
685	dmb	sy
686	dc	ivac, \tmp1			// Invalidate potentially stale cache line
687	.endm
688
689/*
690 * Enable the MMU.
691 *
692 *  x0  = SCTLR_EL1 value for turning on the MMU.
693 *  x1  = TTBR1_EL1 value
694 *
695 * Returns to the caller via x30/lr. This requires the caller to be covered
696 * by the .idmap.text section.
697 *
698 * Checks if the selected granule size is supported by the CPU.
699 * If it isn't, park the CPU
700 */
701SYM_FUNC_START(__enable_mmu)
702	mrs	x2, ID_AA64MMFR0_EL1
703	ubfx	x2, x2, #ID_AA64MMFR0_TGRAN_SHIFT, 4
704	cmp     x2, #ID_AA64MMFR0_TGRAN_SUPPORTED_MIN
705	b.lt    __no_granule_support
706	cmp     x2, #ID_AA64MMFR0_TGRAN_SUPPORTED_MAX
707	b.gt    __no_granule_support
708	update_early_cpu_boot_status 0, x2, x3
709	adrp	x2, idmap_pg_dir
710	phys_to_ttbr x1, x1
711	phys_to_ttbr x2, x2
712	msr	ttbr0_el1, x2			// load TTBR0
713	offset_ttbr1 x1, x3
714	msr	ttbr1_el1, x1			// load TTBR1
715	isb
716
717	set_sctlr_el1	x0
718
719	ret
720SYM_FUNC_END(__enable_mmu)
721
722SYM_FUNC_START(__cpu_secondary_check52bitva)
723#ifdef CONFIG_ARM64_VA_BITS_52
724	ldr_l	x0, vabits_actual
725	cmp	x0, #52
726	b.ne	2f
727
728	mrs_s	x0, SYS_ID_AA64MMFR2_EL1
729	and	x0, x0, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
730	cbnz	x0, 2f
731
732	update_early_cpu_boot_status \
733		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_52_BIT_VA, x0, x1
7341:	wfe
735	wfi
736	b	1b
737
738#endif
7392:	ret
740SYM_FUNC_END(__cpu_secondary_check52bitva)
741
742SYM_FUNC_START_LOCAL(__no_granule_support)
743	/* Indicate that this CPU can't boot and is stuck in the kernel */
744	update_early_cpu_boot_status \
745		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_NO_GRAN, x1, x2
7461:
747	wfe
748	wfi
749	b	1b
750SYM_FUNC_END(__no_granule_support)
751
752#ifdef CONFIG_RELOCATABLE
753SYM_FUNC_START_LOCAL(__relocate_kernel)
754	/*
755	 * Iterate over each entry in the relocation table, and apply the
756	 * relocations in place.
757	 */
758	ldr	w9, =__rela_offset		// offset to reloc table
759	ldr	w10, =__rela_size		// size of reloc table
760
761	mov_q	x11, KIMAGE_VADDR		// default virtual offset
762	add	x11, x11, x23			// actual virtual offset
763	add	x9, x9, x11			// __va(.rela)
764	add	x10, x9, x10			// __va(.rela) + sizeof(.rela)
765
7660:	cmp	x9, x10
767	b.hs	1f
768	ldp	x12, x13, [x9], #24
769	ldr	x14, [x9, #-8]
770	cmp	w13, #R_AARCH64_RELATIVE
771	b.ne	0b
772	add	x14, x14, x23			// relocate
773	str	x14, [x12, x23]
774	b	0b
775
7761:
777#ifdef CONFIG_RELR
778	/*
779	 * Apply RELR relocations.
780	 *
781	 * RELR is a compressed format for storing relative relocations. The
782	 * encoded sequence of entries looks like:
783	 * [ AAAAAAAA BBBBBBB1 BBBBBBB1 ... AAAAAAAA BBBBBB1 ... ]
784	 *
785	 * i.e. start with an address, followed by any number of bitmaps. The
786	 * address entry encodes 1 relocation. The subsequent bitmap entries
787	 * encode up to 63 relocations each, at subsequent offsets following
788	 * the last address entry.
789	 *
790	 * The bitmap entries must have 1 in the least significant bit. The
791	 * assumption here is that an address cannot have 1 in lsb. Odd
792	 * addresses are not supported. Any odd addresses are stored in the RELA
793	 * section, which is handled above.
794	 *
795	 * Excluding the least significant bit in the bitmap, each non-zero
796	 * bit in the bitmap represents a relocation to be applied to
797	 * a corresponding machine word that follows the base address
798	 * word. The second least significant bit represents the machine
799	 * word immediately following the initial address, and each bit
800	 * that follows represents the next word, in linear order. As such,
801	 * a single bitmap can encode up to 63 relocations in a 64-bit object.
802	 *
803	 * In this implementation we store the address of the next RELR table
804	 * entry in x9, the address being relocated by the current address or
805	 * bitmap entry in x13 and the address being relocated by the current
806	 * bit in x14.
807	 *
808	 * Because addends are stored in place in the binary, RELR relocations
809	 * cannot be applied idempotently. We use x24 to keep track of the
810	 * currently applied displacement so that we can correctly relocate if
811	 * __relocate_kernel is called twice with non-zero displacements (i.e.
812	 * if there is both a physical misalignment and a KASLR displacement).
813	 */
814	ldr	w9, =__relr_offset		// offset to reloc table
815	ldr	w10, =__relr_size		// size of reloc table
816	add	x9, x9, x11			// __va(.relr)
817	add	x10, x9, x10			// __va(.relr) + sizeof(.relr)
818
819	sub	x15, x23, x24			// delta from previous offset
820	cbz	x15, 7f				// nothing to do if unchanged
821	mov	x24, x23			// save new offset
822
8232:	cmp	x9, x10
824	b.hs	7f
825	ldr	x11, [x9], #8
826	tbnz	x11, #0, 3f			// branch to handle bitmaps
827	add	x13, x11, x23
828	ldr	x12, [x13]			// relocate address entry
829	add	x12, x12, x15
830	str	x12, [x13], #8			// adjust to start of bitmap
831	b	2b
832
8333:	mov	x14, x13
8344:	lsr	x11, x11, #1
835	cbz	x11, 6f
836	tbz	x11, #0, 5f			// skip bit if not set
837	ldr	x12, [x14]			// relocate bit
838	add	x12, x12, x15
839	str	x12, [x14]
840
8415:	add	x14, x14, #8			// move to next bit's address
842	b	4b
843
8446:	/*
845	 * Move to the next bitmap's address. 8 is the word size, and 63 is the
846	 * number of significant bits in a bitmap entry.
847	 */
848	add	x13, x13, #(8 * 63)
849	b	2b
850
8517:
852#endif
853	ret
854
855SYM_FUNC_END(__relocate_kernel)
856#endif
857
858SYM_FUNC_START_LOCAL(__primary_switch)
859#ifdef CONFIG_RANDOMIZE_BASE
860	mov	x19, x0				// preserve new SCTLR_EL1 value
861	mrs	x20, sctlr_el1			// preserve old SCTLR_EL1 value
862#endif
863
864	adrp	x1, init_pg_dir
865	bl	__enable_mmu
866#ifdef CONFIG_RELOCATABLE
867#ifdef CONFIG_RELR
868	mov	x24, #0				// no RELR displacement yet
869#endif
870	bl	__relocate_kernel
871#ifdef CONFIG_RANDOMIZE_BASE
872	ldr	x8, =__primary_switched
873	adrp	x0, __PHYS_OFFSET
874	blr	x8
875
876	/*
877	 * If we return here, we have a KASLR displacement in x23 which we need
878	 * to take into account by discarding the current kernel mapping and
879	 * creating a new one.
880	 */
881	pre_disable_mmu_workaround
882	msr	sctlr_el1, x20			// disable the MMU
883	isb
884	bl	__create_page_tables		// recreate kernel mapping
885
886	tlbi	vmalle1				// Remove any stale TLB entries
887	dsb	nsh
888	isb
889
890	set_sctlr_el1	x19			// re-enable the MMU
891
892	bl	__relocate_kernel
893#endif
894#endif
895	ldr	x8, =__primary_switched
896	adrp	x0, __PHYS_OFFSET
897	br	x8
898SYM_FUNC_END(__primary_switch)
899