• 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/irqchip/arm-gic-v3.h>
15#include <linux/pgtable.h>
16
17#include <asm/asm_pointer_auth.h>
18#include <asm/assembler.h>
19#include <asm/boot.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/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	el2_setup			// Drop to EL1, 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#ifdef CONFIG_ARM64_PTR_AUTH
424	__ptrauth_keys_init_cpu	x5, x6, x7, x8
425#endif
426
427	adr_l	x8, vectors			// load VBAR_EL1 with virtual
428	msr	vbar_el1, x8			// vector table address
429	isb
430
431	stp	xzr, x30, [sp, #-16]!
432	mov	x29, sp
433
434#ifdef CONFIG_SHADOW_CALL_STACK
435	adr_l	scs_sp, init_shadow_call_stack	// Set shadow call stack
436#endif
437
438	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
439
440	ldr_l	x4, kimage_vaddr		// Save the offset between
441	sub	x4, x4, x0			// the kernel virtual and
442	str_l	x4, kimage_voffset, x5		// physical mappings
443
444	// Clear BSS
445	adr_l	x0, __bss_start
446	mov	x1, xzr
447	adr_l	x2, __bss_stop
448	sub	x2, x2, x0
449	bl	__pi_memset
450	dsb	ishst				// Make zero page visible to PTW
451
452#ifdef CONFIG_KASAN
453	bl	kasan_early_init
454#endif
455#ifdef CONFIG_RANDOMIZE_BASE
456	tst	x23, ~(MIN_KIMG_ALIGN - 1)	// already running randomized?
457	b.ne	0f
458	mov	x0, x21				// pass FDT address in x0
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	add	sp, sp, #16
467	mov	x29, #0
468	mov	x30, #0
469	b	start_kernel
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 * If we're fortunate enough to boot at EL2, ensure that the world is
487 * sane before dropping to EL1.
488 *
489 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in w0 if
490 * booted in EL1 or EL2 respectively.
491 */
492SYM_FUNC_START(el2_setup)
493	msr	SPsel, #1			// We want to use SP_EL{1,2}
494	mrs	x0, CurrentEL
495	cmp	x0, #CurrentEL_EL2
496	b.eq	1f
497	mov_q	x0, (SCTLR_EL1_RES1 | ENDIAN_SET_EL1)
498	msr	sctlr_el1, x0
499	mov	w0, #BOOT_CPU_MODE_EL1		// This cpu booted in EL1
500	isb
501	ret
502
5031:	mov_q	x0, (SCTLR_EL2_RES1 | ENDIAN_SET_EL2)
504	msr	sctlr_el2, x0
505
506#ifdef CONFIG_ARM64_VHE
507	/*
508	 * Check for VHE being present. For the rest of the EL2 setup,
509	 * x2 being non-zero indicates that we do have VHE, and that the
510	 * kernel is intended to run at EL2.
511	 */
512	mrs	x2, id_aa64mmfr1_el1
513	ubfx	x2, x2, #ID_AA64MMFR1_VHE_SHIFT, #4
514#else
515	mov	x2, xzr
516#endif
517
518	/* Hyp configuration. */
519	mov_q	x0, HCR_HOST_NVHE_FLAGS
520	cbz	x2, set_hcr
521	mov_q	x0, HCR_HOST_VHE_FLAGS
522set_hcr:
523	msr	hcr_el2, x0
524	isb
525
526	/*
527	 * Allow Non-secure EL1 and EL0 to access physical timer and counter.
528	 * This is not necessary for VHE, since the host kernel runs in EL2,
529	 * and EL0 accesses are configured in the later stage of boot process.
530	 * Note that when HCR_EL2.E2H == 1, CNTHCTL_EL2 has the same bit layout
531	 * as CNTKCTL_EL1, and CNTKCTL_EL1 accessing instructions are redefined
532	 * to access CNTHCTL_EL2. This allows the kernel designed to run at EL1
533	 * to transparently mess with the EL0 bits via CNTKCTL_EL1 access in
534	 * EL2.
535	 */
536	cbnz	x2, 1f
537	mrs	x0, cnthctl_el2
538	orr	x0, x0, #3			// Enable EL1 physical timers
539	msr	cnthctl_el2, x0
5401:
541	msr	cntvoff_el2, xzr		// Clear virtual offset
542
543#ifdef CONFIG_ARM_GIC_V3
544	/* GICv3 system register access */
545	mrs	x0, id_aa64pfr0_el1
546	ubfx	x0, x0, #ID_AA64PFR0_GIC_SHIFT, #4
547	cbz	x0, 3f
548
549	mrs_s	x0, SYS_ICC_SRE_EL2
550	orr	x0, x0, #ICC_SRE_EL2_SRE	// Set ICC_SRE_EL2.SRE==1
551	orr	x0, x0, #ICC_SRE_EL2_ENABLE	// Set ICC_SRE_EL2.Enable==1
552	msr_s	SYS_ICC_SRE_EL2, x0
553	isb					// Make sure SRE is now set
554	mrs_s	x0, SYS_ICC_SRE_EL2		// Read SRE back,
555	tbz	x0, #0, 3f			// and check that it sticks
556	msr_s	SYS_ICH_HCR_EL2, xzr		// Reset ICC_HCR_EL2 to defaults
557
5583:
559#endif
560
561	/* Populate ID registers. */
562	mrs	x0, midr_el1
563	mrs	x1, mpidr_el1
564	msr	vpidr_el2, x0
565	msr	vmpidr_el2, x1
566
567#ifdef CONFIG_COMPAT
568	msr	hstr_el2, xzr			// Disable CP15 traps to EL2
569#endif
570
571	/* EL2 debug */
572	mrs	x1, id_aa64dfr0_el1
573	sbfx	x0, x1, #ID_AA64DFR0_PMUVER_SHIFT, #4
574	cmp	x0, #1
575	b.lt	4f				// Skip if no PMU present
576	mrs	x0, pmcr_el0			// Disable debug access traps
577	ubfx	x0, x0, #11, #5			// to EL2 and allow access to
5784:
579	csel	x3, xzr, x0, lt			// all PMU counters from EL1
580
581	/* Statistical profiling */
582	ubfx	x0, x1, #ID_AA64DFR0_PMSVER_SHIFT, #4
583	cbz	x0, 7f				// Skip if SPE not present
584	cbnz	x2, 6f				// VHE?
585	mrs_s	x4, SYS_PMBIDR_EL1		// If SPE available at EL2,
586	and	x4, x4, #(1 << SYS_PMBIDR_EL1_P_SHIFT)
587	cbnz	x4, 5f				// then permit sampling of physical
588	mov	x4, #(1 << SYS_PMSCR_EL2_PCT_SHIFT | \
589		      1 << SYS_PMSCR_EL2_PA_SHIFT)
590	msr_s	SYS_PMSCR_EL2, x4		// addresses and physical counter
5915:
592	mov	x1, #(MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT)
593	orr	x3, x3, x1			// If we don't have VHE, then
594	b	7f				// use EL1&0 translation.
5956:						// For VHE, use EL2 translation
596	orr	x3, x3, #MDCR_EL2_TPMS		// and disable access from EL1
5977:
598	msr	mdcr_el2, x3			// Configure debug traps
599
600	/* LORegions */
601	mrs	x1, id_aa64mmfr1_el1
602	ubfx	x0, x1, #ID_AA64MMFR1_LOR_SHIFT, 4
603	cbz	x0, 1f
604	msr_s	SYS_LORC_EL1, xzr
6051:
606
607	/* Stage-2 translation */
608	msr	vttbr_el2, xzr
609
610	cbz	x2, install_el2_stub
611
612	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
613	isb
614	ret
615
616SYM_INNER_LABEL(install_el2_stub, SYM_L_LOCAL)
617	/*
618	 * When VHE is not in use, early init of EL2 and EL1 needs to be
619	 * done here.
620	 * When VHE _is_ in use, EL1 will not be used in the host and
621	 * requires no configuration, and all non-hyp-specific EL2 setup
622	 * will be done via the _EL1 system register aliases in __cpu_setup.
623	 */
624	mov_q	x0, (SCTLR_EL1_RES1 | ENDIAN_SET_EL1)
625	msr	sctlr_el1, x0
626
627	/* Coprocessor traps. */
628	mov	x0, #0x33ff
629	msr	cptr_el2, x0			// Disable copro. traps to EL2
630
631	/* SVE register access */
632	mrs	x1, id_aa64pfr0_el1
633	ubfx	x1, x1, #ID_AA64PFR0_SVE_SHIFT, #4
634	cbz	x1, 7f
635
636	bic	x0, x0, #CPTR_EL2_TZ		// Also disable SVE traps
637	msr	cptr_el2, x0			// Disable copro. traps to EL2
638	isb
639	mov	x1, #ZCR_ELx_LEN_MASK		// SVE: Enable full vector
640	msr_s	SYS_ZCR_EL2, x1			// length for EL1.
641
642	/* Hypervisor stub */
6437:	adr_l	x0, __hyp_stub_vectors
644	msr	vbar_el2, x0
645
646	/* spsr */
647	mov	x0, #(PSR_F_BIT | PSR_I_BIT | PSR_A_BIT | PSR_D_BIT |\
648		      PSR_MODE_EL1h)
649	msr	spsr_el2, x0
650	msr	elr_el2, lr
651	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
652	eret
653SYM_FUNC_END(el2_setup)
654
655/*
656 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
657 * in w0. See arch/arm64/include/asm/virt.h for more info.
658 */
659SYM_FUNC_START_LOCAL(set_cpu_boot_mode_flag)
660	adr_l	x1, __boot_cpu_mode
661	cmp	w0, #BOOT_CPU_MODE_EL2
662	b.ne	1f
663	add	x1, x1, #4
6641:	str	w0, [x1]			// This CPU has booted in EL1
665	dmb	sy
666	dc	ivac, x1			// Invalidate potentially stale cache line
667	ret
668SYM_FUNC_END(set_cpu_boot_mode_flag)
669
670/*
671 * These values are written with the MMU off, but read with the MMU on.
672 * Writers will invalidate the corresponding address, discarding up to a
673 * 'Cache Writeback Granule' (CWG) worth of data. The linker script ensures
674 * sufficient alignment that the CWG doesn't overlap another section.
675 */
676	.pushsection ".mmuoff.data.write", "aw"
677/*
678 * We need to find out the CPU boot mode long after boot, so we need to
679 * store it in a writable variable.
680 *
681 * This is not in .bss, because we set it sufficiently early that the boot-time
682 * zeroing of .bss would clobber it.
683 */
684SYM_DATA_START(__boot_cpu_mode)
685	.long	BOOT_CPU_MODE_EL2
686	.long	BOOT_CPU_MODE_EL1
687SYM_DATA_END(__boot_cpu_mode)
688/*
689 * The booting CPU updates the failed status @__early_cpu_boot_status,
690 * with MMU turned off.
691 */
692SYM_DATA_START(__early_cpu_boot_status)
693	.quad 	0
694SYM_DATA_END(__early_cpu_boot_status)
695
696	.popsection
697
698	/*
699	 * This provides a "holding pen" for platforms to hold all secondary
700	 * cores are held until we're ready for them to initialise.
701	 */
702SYM_FUNC_START(secondary_holding_pen)
703	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
704	bl	set_cpu_boot_mode_flag
705	mrs	x0, mpidr_el1
706	mov_q	x1, MPIDR_HWID_BITMASK
707	and	x0, x0, x1
708	adr_l	x3, secondary_holding_pen_release
709pen:	ldr	x4, [x3]
710	cmp	x4, x0
711	b.eq	secondary_startup
712	wfe
713	b	pen
714SYM_FUNC_END(secondary_holding_pen)
715
716	/*
717	 * Secondary entry point that jumps straight into the kernel. Only to
718	 * be used where CPUs are brought online dynamically by the kernel.
719	 */
720SYM_FUNC_START(secondary_entry)
721	bl	el2_setup			// Drop to EL1
722	bl	set_cpu_boot_mode_flag
723	b	secondary_startup
724SYM_FUNC_END(secondary_entry)
725
726SYM_FUNC_START_LOCAL(secondary_startup)
727	/*
728	 * Common entry point for secondary CPUs.
729	 */
730	bl	__cpu_secondary_check52bitva
731	bl	__cpu_setup			// initialise processor
732	adrp	x1, swapper_pg_dir
733	bl	__enable_mmu
734	ldr	x8, =__secondary_switched
735	br	x8
736SYM_FUNC_END(secondary_startup)
737
738SYM_FUNC_START_LOCAL(__secondary_switched)
739	adr_l	x5, vectors
740	msr	vbar_el1, x5
741	isb
742
743	adr_l	x0, secondary_data
744	ldr	x1, [x0, #CPU_BOOT_STACK]	// get secondary_data.stack
745	cbz	x1, __secondary_too_slow
746	mov	sp, x1
747	ldr	x2, [x0, #CPU_BOOT_TASK]
748	cbz	x2, __secondary_too_slow
749	msr	sp_el0, x2
750	scs_load_current
751	mov	x29, #0
752	mov	x30, #0
753
754#ifdef CONFIG_ARM64_PTR_AUTH
755	ptrauth_keys_init_cpu x2, x3, x4, x5
756#endif
757
758	b	secondary_start_kernel
759SYM_FUNC_END(__secondary_switched)
760
761SYM_FUNC_START_LOCAL(__secondary_too_slow)
762	wfe
763	wfi
764	b	__secondary_too_slow
765SYM_FUNC_END(__secondary_too_slow)
766
767/*
768 * The booting CPU updates the failed status @__early_cpu_boot_status,
769 * with MMU turned off.
770 *
771 * update_early_cpu_boot_status tmp, status
772 *  - Corrupts tmp1, tmp2
773 *  - Writes 'status' to __early_cpu_boot_status and makes sure
774 *    it is committed to memory.
775 */
776
777	.macro	update_early_cpu_boot_status status, tmp1, tmp2
778	mov	\tmp2, #\status
779	adr_l	\tmp1, __early_cpu_boot_status
780	str	\tmp2, [\tmp1]
781	dmb	sy
782	dc	ivac, \tmp1			// Invalidate potentially stale cache line
783	.endm
784
785/*
786 * Enable the MMU.
787 *
788 *  x0  = SCTLR_EL1 value for turning on the MMU.
789 *  x1  = TTBR1_EL1 value
790 *
791 * Returns to the caller via x30/lr. This requires the caller to be covered
792 * by the .idmap.text section.
793 *
794 * Checks if the selected granule size is supported by the CPU.
795 * If it isn't, park the CPU
796 */
797SYM_FUNC_START(__enable_mmu)
798	mrs	x2, ID_AA64MMFR0_EL1
799	ubfx	x2, x2, #ID_AA64MMFR0_TGRAN_SHIFT, 4
800	cmp     x2, #ID_AA64MMFR0_TGRAN_SUPPORTED_MIN
801	b.lt    __no_granule_support
802	cmp     x2, #ID_AA64MMFR0_TGRAN_SUPPORTED_MAX
803	b.gt    __no_granule_support
804	update_early_cpu_boot_status 0, x2, x3
805	adrp	x2, idmap_pg_dir
806	phys_to_ttbr x1, x1
807	phys_to_ttbr x2, x2
808	msr	ttbr0_el1, x2			// load TTBR0
809	offset_ttbr1 x1, x3
810	msr	ttbr1_el1, x1			// load TTBR1
811	isb
812	msr	sctlr_el1, x0
813	isb
814	/*
815	 * Invalidate the local I-cache so that any instructions fetched
816	 * speculatively from the PoC are discarded, since they may have
817	 * been dynamically patched at the PoU.
818	 */
819	ic	iallu
820	dsb	nsh
821	isb
822	ret
823SYM_FUNC_END(__enable_mmu)
824
825SYM_FUNC_START(__cpu_secondary_check52bitva)
826#ifdef CONFIG_ARM64_VA_BITS_52
827	ldr_l	x0, vabits_actual
828	cmp	x0, #52
829	b.ne	2f
830
831	mrs_s	x0, SYS_ID_AA64MMFR2_EL1
832	and	x0, x0, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
833	cbnz	x0, 2f
834
835	update_early_cpu_boot_status \
836		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_52_BIT_VA, x0, x1
8371:	wfe
838	wfi
839	b	1b
840
841#endif
8422:	ret
843SYM_FUNC_END(__cpu_secondary_check52bitva)
844
845SYM_FUNC_START_LOCAL(__no_granule_support)
846	/* Indicate that this CPU can't boot and is stuck in the kernel */
847	update_early_cpu_boot_status \
848		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_NO_GRAN, x1, x2
8491:
850	wfe
851	wfi
852	b	1b
853SYM_FUNC_END(__no_granule_support)
854
855#ifdef CONFIG_RELOCATABLE
856SYM_FUNC_START_LOCAL(__relocate_kernel)
857	/*
858	 * Iterate over each entry in the relocation table, and apply the
859	 * relocations in place.
860	 */
861	ldr	w9, =__rela_offset		// offset to reloc table
862	ldr	w10, =__rela_size		// size of reloc table
863
864	mov_q	x11, KIMAGE_VADDR		// default virtual offset
865	add	x11, x11, x23			// actual virtual offset
866	add	x9, x9, x11			// __va(.rela)
867	add	x10, x9, x10			// __va(.rela) + sizeof(.rela)
868
8690:	cmp	x9, x10
870	b.hs	1f
871	ldp	x12, x13, [x9], #24
872	ldr	x14, [x9, #-8]
873	cmp	w13, #R_AARCH64_RELATIVE
874	b.ne	0b
875	add	x14, x14, x23			// relocate
876	str	x14, [x12, x23]
877	b	0b
878
8791:
880#ifdef CONFIG_RELR
881	/*
882	 * Apply RELR relocations.
883	 *
884	 * RELR is a compressed format for storing relative relocations. The
885	 * encoded sequence of entries looks like:
886	 * [ AAAAAAAA BBBBBBB1 BBBBBBB1 ... AAAAAAAA BBBBBB1 ... ]
887	 *
888	 * i.e. start with an address, followed by any number of bitmaps. The
889	 * address entry encodes 1 relocation. The subsequent bitmap entries
890	 * encode up to 63 relocations each, at subsequent offsets following
891	 * the last address entry.
892	 *
893	 * The bitmap entries must have 1 in the least significant bit. The
894	 * assumption here is that an address cannot have 1 in lsb. Odd
895	 * addresses are not supported. Any odd addresses are stored in the RELA
896	 * section, which is handled above.
897	 *
898	 * Excluding the least significant bit in the bitmap, each non-zero
899	 * bit in the bitmap represents a relocation to be applied to
900	 * a corresponding machine word that follows the base address
901	 * word. The second least significant bit represents the machine
902	 * word immediately following the initial address, and each bit
903	 * that follows represents the next word, in linear order. As such,
904	 * a single bitmap can encode up to 63 relocations in a 64-bit object.
905	 *
906	 * In this implementation we store the address of the next RELR table
907	 * entry in x9, the address being relocated by the current address or
908	 * bitmap entry in x13 and the address being relocated by the current
909	 * bit in x14.
910	 *
911	 * Because addends are stored in place in the binary, RELR relocations
912	 * cannot be applied idempotently. We use x24 to keep track of the
913	 * currently applied displacement so that we can correctly relocate if
914	 * __relocate_kernel is called twice with non-zero displacements (i.e.
915	 * if there is both a physical misalignment and a KASLR displacement).
916	 */
917	ldr	w9, =__relr_offset		// offset to reloc table
918	ldr	w10, =__relr_size		// size of reloc table
919	add	x9, x9, x11			// __va(.relr)
920	add	x10, x9, x10			// __va(.relr) + sizeof(.relr)
921
922	sub	x15, x23, x24			// delta from previous offset
923	cbz	x15, 7f				// nothing to do if unchanged
924	mov	x24, x23			// save new offset
925
9262:	cmp	x9, x10
927	b.hs	7f
928	ldr	x11, [x9], #8
929	tbnz	x11, #0, 3f			// branch to handle bitmaps
930	add	x13, x11, x23
931	ldr	x12, [x13]			// relocate address entry
932	add	x12, x12, x15
933	str	x12, [x13], #8			// adjust to start of bitmap
934	b	2b
935
9363:	mov	x14, x13
9374:	lsr	x11, x11, #1
938	cbz	x11, 6f
939	tbz	x11, #0, 5f			// skip bit if not set
940	ldr	x12, [x14]			// relocate bit
941	add	x12, x12, x15
942	str	x12, [x14]
943
9445:	add	x14, x14, #8			// move to next bit's address
945	b	4b
946
9476:	/*
948	 * Move to the next bitmap's address. 8 is the word size, and 63 is the
949	 * number of significant bits in a bitmap entry.
950	 */
951	add	x13, x13, #(8 * 63)
952	b	2b
953
9547:
955#endif
956	ret
957
958SYM_FUNC_END(__relocate_kernel)
959#endif
960
961SYM_FUNC_START_LOCAL(__primary_switch)
962#ifdef CONFIG_RANDOMIZE_BASE
963	mov	x19, x0				// preserve new SCTLR_EL1 value
964	mrs	x20, sctlr_el1			// preserve old SCTLR_EL1 value
965#endif
966
967	adrp	x1, init_pg_dir
968	bl	__enable_mmu
969#ifdef CONFIG_RELOCATABLE
970#ifdef CONFIG_RELR
971	mov	x24, #0				// no RELR displacement yet
972#endif
973	bl	__relocate_kernel
974#ifdef CONFIG_RANDOMIZE_BASE
975	ldr	x8, =__primary_switched
976	adrp	x0, __PHYS_OFFSET
977	blr	x8
978
979	/*
980	 * If we return here, we have a KASLR displacement in x23 which we need
981	 * to take into account by discarding the current kernel mapping and
982	 * creating a new one.
983	 */
984	pre_disable_mmu_workaround
985	msr	sctlr_el1, x20			// disable the MMU
986	isb
987	bl	__create_page_tables		// recreate kernel mapping
988
989	tlbi	vmalle1				// Remove any stale TLB entries
990	dsb	nsh
991	isb
992
993	msr	sctlr_el1, x19			// re-enable the MMU
994	isb
995	ic	iallu				// flush instructions fetched
996	dsb	nsh				// via old mapping
997	isb
998
999	bl	__relocate_kernel
1000#endif
1001#endif
1002	ldr	x8, =__primary_switched
1003	adrp	x0, __PHYS_OFFSET
1004	br	x8
1005SYM_FUNC_END(__primary_switch)
1006