xref: /linux/arch/arm64/kernel/head.S (revision 55f3538c4923e9dfca132e99ebec370e8094afda)
1/*
2 * Low-level CPU initialisation
3 * Based on arch/arm/kernel/head.S
4 *
5 * Copyright (C) 1994-2002 Russell King
6 * Copyright (C) 2003-2012 ARM Ltd.
7 * Authors:	Catalin Marinas <catalin.marinas@arm.com>
8 *		Will Deacon <will.deacon@arm.com>
9 *
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17 * GNU General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public License
20 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21 */
22
23#include <linux/linkage.h>
24#include <linux/init.h>
25#include <linux/irqchip/arm-gic-v3.h>
26
27#include <asm/assembler.h>
28#include <asm/boot.h>
29#include <asm/ptrace.h>
30#include <asm/asm-offsets.h>
31#include <asm/cache.h>
32#include <asm/cputype.h>
33#include <asm/elf.h>
34#include <asm/kernel-pgtable.h>
35#include <asm/kvm_arm.h>
36#include <asm/memory.h>
37#include <asm/pgtable-hwdef.h>
38#include <asm/pgtable.h>
39#include <asm/page.h>
40#include <asm/smp.h>
41#include <asm/sysreg.h>
42#include <asm/thread_info.h>
43#include <asm/virt.h>
44
45#include "efi-header.S"
46
47#define __PHYS_OFFSET	(KERNEL_START - TEXT_OFFSET)
48
49#if (TEXT_OFFSET & 0xfff) != 0
50#error TEXT_OFFSET must be at least 4KB aligned
51#elif (PAGE_OFFSET & 0x1fffff) != 0
52#error PAGE_OFFSET must be at least 2MB aligned
53#elif TEXT_OFFSET > 0x1fffff
54#error TEXT_OFFSET must be less than 2MB
55#endif
56
57/*
58 * Kernel startup entry point.
59 * ---------------------------
60 *
61 * The requirements are:
62 *   MMU = off, D-cache = off, I-cache = on or off,
63 *   x0 = physical address to the FDT blob.
64 *
65 * This code is mostly position independent so you call this at
66 * __pa(PAGE_OFFSET + TEXT_OFFSET).
67 *
68 * Note that the callee-saved registers are used for storing variables
69 * that are useful before the MMU is enabled. The allocations are described
70 * in the entry routines.
71 */
72	__HEAD
73_head:
74	/*
75	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
76	 */
77#ifdef CONFIG_EFI
78	/*
79	 * This add instruction has no meaningful effect except that
80	 * its opcode forms the magic "MZ" signature required by UEFI.
81	 */
82	add	x13, x18, #0x16
83	b	stext
84#else
85	b	stext				// branch to kernel start, magic
86	.long	0				// reserved
87#endif
88	le64sym	_kernel_offset_le		// Image load offset from start of RAM, little-endian
89	le64sym	_kernel_size_le			// Effective size of kernel image, little-endian
90	le64sym	_kernel_flags_le		// Informative flags, little-endian
91	.quad	0				// reserved
92	.quad	0				// reserved
93	.quad	0				// reserved
94	.ascii	"ARM\x64"			// Magic number
95#ifdef CONFIG_EFI
96	.long	pe_header - _head		// Offset to the PE header.
97
98pe_header:
99	__EFI_PE_HEADER
100#else
101	.long	0				// reserved
102#endif
103
104	__INIT
105
106	/*
107	 * The following callee saved general purpose registers are used on the
108	 * primary lowlevel boot path:
109	 *
110	 *  Register   Scope                      Purpose
111	 *  x21        stext() .. start_kernel()  FDT pointer passed at boot in x0
112	 *  x23        stext() .. start_kernel()  physical misalignment/KASLR offset
113	 *  x28        __create_page_tables()     callee preserved temp register
114	 *  x19/x20    __primary_switch()         callee preserved temp registers
115	 */
116ENTRY(stext)
117	bl	preserve_boot_args
118	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
119	adrp	x23, __PHYS_OFFSET
120	and	x23, x23, MIN_KIMG_ALIGN - 1	// KASLR offset, defaults to 0
121	bl	set_cpu_boot_mode_flag
122	bl	__create_page_tables
123	/*
124	 * The following calls CPU setup code, see arch/arm64/mm/proc.S for
125	 * details.
126	 * On return, the CPU will be ready for the MMU to be turned on and
127	 * the TCR will have been set.
128	 */
129	bl	__cpu_setup			// initialise processor
130	b	__primary_switch
131ENDPROC(stext)
132
133/*
134 * Preserve the arguments passed by the bootloader in x0 .. x3
135 */
136preserve_boot_args:
137	mov	x21, x0				// x21=FDT
138
139	adr_l	x0, boot_args			// record the contents of
140	stp	x21, x1, [x0]			// x0 .. x3 at kernel entry
141	stp	x2, x3, [x0, #16]
142
143	dmb	sy				// needed before dc ivac with
144						// MMU off
145
146	mov	x1, #0x20			// 4 x 8 bytes
147	b	__inval_dcache_area		// tail call
148ENDPROC(preserve_boot_args)
149
150/*
151 * Macro to arrange a physical address in a page table entry, taking care of
152 * 52-bit addresses.
153 *
154 * Preserves:	phys
155 * Returns:	pte
156 */
157	.macro	phys_to_pte, phys, pte
158#ifdef CONFIG_ARM64_PA_BITS_52
159	/*
160	 * We assume \phys is 64K aligned and this is guaranteed by only
161	 * supporting this configuration with 64K pages.
162	 */
163	orr	\pte, \phys, \phys, lsr #36
164	and	\pte, \pte, #PTE_ADDR_MASK
165#else
166	mov	\pte, \phys
167#endif
168	.endm
169
170/*
171 * Macro to create a table entry to the next page.
172 *
173 *	tbl:	page table address
174 *	virt:	virtual address
175 *	shift:	#imm page table shift
176 *	ptrs:	#imm pointers per table page
177 *
178 * Preserves:	virt
179 * Corrupts:	ptrs, tmp1, tmp2
180 * Returns:	tbl -> next level table page address
181 */
182	.macro	create_table_entry, tbl, virt, shift, ptrs, tmp1, tmp2
183	add	\tmp1, \tbl, #PAGE_SIZE
184	phys_to_pte \tmp1, \tmp2
185	orr	\tmp2, \tmp2, #PMD_TYPE_TABLE	// address of next table and entry type
186	lsr	\tmp1, \virt, #\shift
187	sub	\ptrs, \ptrs, #1
188	and	\tmp1, \tmp1, \ptrs		// table index
189	str	\tmp2, [\tbl, \tmp1, lsl #3]
190	add	\tbl, \tbl, #PAGE_SIZE		// next level table page
191	.endm
192
193/*
194 * Macro to populate page table entries, these entries can be pointers to the next level
195 * or last level entries pointing to physical memory.
196 *
197 *	tbl:	page table address
198 *	rtbl:	pointer to page table or physical memory
199 *	index:	start index to write
200 *	eindex:	end index to write - [index, eindex] written to
201 *	flags:	flags for pagetable entry to or in
202 *	inc:	increment to rtbl between each entry
203 *	tmp1:	temporary variable
204 *
205 * Preserves:	tbl, eindex, flags, inc
206 * Corrupts:	index, tmp1
207 * Returns:	rtbl
208 */
209	.macro populate_entries, tbl, rtbl, index, eindex, flags, inc, tmp1
210.Lpe\@:	phys_to_pte \rtbl, \tmp1
211	orr	\tmp1, \tmp1, \flags	// tmp1 = table entry
212	str	\tmp1, [\tbl, \index, lsl #3]
213	add	\rtbl, \rtbl, \inc	// rtbl = pa next level
214	add	\index, \index, #1
215	cmp	\index, \eindex
216	b.ls	.Lpe\@
217	.endm
218
219/*
220 * Compute indices of table entries from virtual address range. If multiple entries
221 * were needed in the previous page table level then the next page table level is assumed
222 * to be composed of multiple pages. (This effectively scales the end index).
223 *
224 *	vstart:	virtual address of start of range
225 *	vend:	virtual address of end of range
226 *	shift:	shift used to transform virtual address into index
227 *	ptrs:	number of entries in page table
228 *	istart:	index in table corresponding to vstart
229 *	iend:	index in table corresponding to vend
230 *	count:	On entry: how many extra entries were required in previous level, scales
231 *			  our end index.
232 *		On exit: returns how many extra entries required for next page table level
233 *
234 * Preserves:	vstart, vend, shift, ptrs
235 * Returns:	istart, iend, count
236 */
237	.macro compute_indices, vstart, vend, shift, ptrs, istart, iend, count
238	lsr	\iend, \vend, \shift
239	mov	\istart, \ptrs
240	sub	\istart, \istart, #1
241	and	\iend, \iend, \istart	// iend = (vend >> shift) & (ptrs - 1)
242	mov	\istart, \ptrs
243	mul	\istart, \istart, \count
244	add	\iend, \iend, \istart	// iend += (count - 1) * ptrs
245					// our entries span multiple tables
246
247	lsr	\istart, \vstart, \shift
248	mov	\count, \ptrs
249	sub	\count, \count, #1
250	and	\istart, \istart, \count
251
252	sub	\count, \iend, \istart
253	.endm
254
255/*
256 * Map memory for specified virtual address range. Each level of page table needed supports
257 * multiple entries. If a level requires n entries the next page table level is assumed to be
258 * formed from n pages.
259 *
260 *	tbl:	location of page table
261 *	rtbl:	address to be used for first level page table entry (typically tbl + PAGE_SIZE)
262 *	vstart:	start address to map
263 *	vend:	end address to map - we map [vstart, vend]
264 *	flags:	flags to use to map last level entries
265 *	phys:	physical address corresponding to vstart - physical memory is contiguous
266 *	pgds:	the number of pgd entries
267 *
268 * Temporaries:	istart, iend, tmp, count, sv - these need to be different registers
269 * Preserves:	vstart, vend, flags
270 * Corrupts:	tbl, rtbl, istart, iend, tmp, count, sv
271 */
272	.macro map_memory, tbl, rtbl, vstart, vend, flags, phys, pgds, istart, iend, tmp, count, sv
273	add \rtbl, \tbl, #PAGE_SIZE
274	mov \sv, \rtbl
275	mov \count, #0
276	compute_indices \vstart, \vend, #PGDIR_SHIFT, \pgds, \istart, \iend, \count
277	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
278	mov \tbl, \sv
279	mov \sv, \rtbl
280
281#if SWAPPER_PGTABLE_LEVELS > 3
282	compute_indices \vstart, \vend, #PUD_SHIFT, #PTRS_PER_PUD, \istart, \iend, \count
283	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
284	mov \tbl, \sv
285	mov \sv, \rtbl
286#endif
287
288#if SWAPPER_PGTABLE_LEVELS > 2
289	compute_indices \vstart, \vend, #SWAPPER_TABLE_SHIFT, #PTRS_PER_PMD, \istart, \iend, \count
290	populate_entries \tbl, \rtbl, \istart, \iend, #PMD_TYPE_TABLE, #PAGE_SIZE, \tmp
291	mov \tbl, \sv
292#endif
293
294	compute_indices \vstart, \vend, #SWAPPER_BLOCK_SHIFT, #PTRS_PER_PTE, \istart, \iend, \count
295	bic \count, \phys, #SWAPPER_BLOCK_SIZE - 1
296	populate_entries \tbl, \count, \istart, \iend, \flags, #SWAPPER_BLOCK_SIZE, \tmp
297	.endm
298
299/*
300 * Setup the initial page tables. We only setup the barest amount which is
301 * required to get the kernel running. The following sections are required:
302 *   - identity mapping to enable the MMU (low address, TTBR0)
303 *   - first few MB of the kernel linear mapping to jump to once the MMU has
304 *     been enabled
305 */
306__create_page_tables:
307	mov	x28, lr
308
309	/*
310	 * Invalidate the idmap and swapper page tables to avoid potential
311	 * dirty cache lines being evicted.
312	 */
313	adrp	x0, idmap_pg_dir
314	adrp	x1, swapper_pg_end
315	sub	x1, x1, x0
316	bl	__inval_dcache_area
317
318	/*
319	 * Clear the idmap and swapper page tables.
320	 */
321	adrp	x0, idmap_pg_dir
322	adrp	x1, swapper_pg_end
323	sub	x1, x1, x0
3241:	stp	xzr, xzr, [x0], #16
325	stp	xzr, xzr, [x0], #16
326	stp	xzr, xzr, [x0], #16
327	stp	xzr, xzr, [x0], #16
328	subs	x1, x1, #64
329	b.ne	1b
330
331	mov	x7, SWAPPER_MM_MMUFLAGS
332
333	/*
334	 * Create the identity mapping.
335	 */
336	adrp	x0, idmap_pg_dir
337	adrp	x3, __idmap_text_start		// __pa(__idmap_text_start)
338
339	/*
340	 * VA_BITS may be too small to allow for an ID mapping to be created
341	 * that covers system RAM if that is located sufficiently high in the
342	 * physical address space. So for the ID map, use an extended virtual
343	 * range in that case, and configure an additional translation level
344	 * if needed.
345	 *
346	 * Calculate the maximum allowed value for TCR_EL1.T0SZ so that the
347	 * entire ID map region can be mapped. As T0SZ == (64 - #bits used),
348	 * this number conveniently equals the number of leading zeroes in
349	 * the physical address of __idmap_text_end.
350	 */
351	adrp	x5, __idmap_text_end
352	clz	x5, x5
353	cmp	x5, TCR_T0SZ(VA_BITS)	// default T0SZ small enough?
354	b.ge	1f			// .. then skip VA range extension
355
356	adr_l	x6, idmap_t0sz
357	str	x5, [x6]
358	dmb	sy
359	dc	ivac, x6		// Invalidate potentially stale cache line
360
361#if (VA_BITS < 48)
362#define EXTRA_SHIFT	(PGDIR_SHIFT + PAGE_SHIFT - 3)
363#define EXTRA_PTRS	(1 << (PHYS_MASK_SHIFT - EXTRA_SHIFT))
364
365	/*
366	 * If VA_BITS < 48, we have to configure an additional table level.
367	 * First, we have to verify our assumption that the current value of
368	 * VA_BITS was chosen such that all translation levels are fully
369	 * utilised, and that lowering T0SZ will always result in an additional
370	 * translation level to be configured.
371	 */
372#if VA_BITS != EXTRA_SHIFT
373#error "Mismatch between VA_BITS and page size/number of translation levels"
374#endif
375
376	mov	x4, EXTRA_PTRS
377	create_table_entry x0, x3, EXTRA_SHIFT, x4, x5, x6
378#else
379	/*
380	 * If VA_BITS == 48, we don't have to configure an additional
381	 * translation level, but the top-level table has more entries.
382	 */
383	mov	x4, #1 << (PHYS_MASK_SHIFT - PGDIR_SHIFT)
384	str_l	x4, idmap_ptrs_per_pgd, x5
385#endif
3861:
387	ldr_l	x4, idmap_ptrs_per_pgd
388	mov	x5, x3				// __pa(__idmap_text_start)
389	adr_l	x6, __idmap_text_end		// __pa(__idmap_text_end)
390
391	map_memory x0, x1, x3, x6, x7, x3, x4, x10, x11, x12, x13, x14
392
393	/*
394	 * Map the kernel image (starting with PHYS_OFFSET).
395	 */
396	adrp	x0, swapper_pg_dir
397	mov_q	x5, KIMAGE_VADDR + TEXT_OFFSET	// compile time __va(_text)
398	add	x5, x5, x23			// add KASLR displacement
399	mov	x4, PTRS_PER_PGD
400	adrp	x6, _end			// runtime __pa(_end)
401	adrp	x3, _text			// runtime __pa(_text)
402	sub	x6, x6, x3			// _end - _text
403	add	x6, x6, x5			// runtime __va(_end)
404
405	map_memory x0, x1, x5, x6, x7, x3, x4, x10, x11, x12, x13, x14
406
407	/*
408	 * Since the page tables have been populated with non-cacheable
409	 * accesses (MMU disabled), invalidate the idmap and swapper page
410	 * tables again to remove any speculatively loaded cache lines.
411	 */
412	adrp	x0, idmap_pg_dir
413	adrp	x1, swapper_pg_end
414	sub	x1, x1, x0
415	dmb	sy
416	bl	__inval_dcache_area
417
418	ret	x28
419ENDPROC(__create_page_tables)
420	.ltorg
421
422/*
423 * The following fragment of code is executed with the MMU enabled.
424 *
425 *   x0 = __PHYS_OFFSET
426 */
427__primary_switched:
428	adrp	x4, init_thread_union
429	add	sp, x4, #THREAD_SIZE
430	adr_l	x5, init_task
431	msr	sp_el0, x5			// Save thread_info
432
433	adr_l	x8, vectors			// load VBAR_EL1 with virtual
434	msr	vbar_el1, x8			// vector table address
435	isb
436
437	stp	xzr, x30, [sp, #-16]!
438	mov	x29, sp
439
440	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
441
442	ldr_l	x4, kimage_vaddr		// Save the offset between
443	sub	x4, x4, x0			// the kernel virtual and
444	str_l	x4, kimage_voffset, x5		// physical mappings
445
446	// Clear BSS
447	adr_l	x0, __bss_start
448	mov	x1, xzr
449	adr_l	x2, __bss_stop
450	sub	x2, x2, x0
451	bl	__pi_memset
452	dsb	ishst				// Make zero page visible to PTW
453
454#ifdef CONFIG_KASAN
455	bl	kasan_early_init
456#endif
457#ifdef CONFIG_RANDOMIZE_BASE
458	tst	x23, ~(MIN_KIMG_ALIGN - 1)	// already running randomized?
459	b.ne	0f
460	mov	x0, x21				// pass FDT address in x0
461	bl	kaslr_early_init		// parse FDT for KASLR options
462	cbz	x0, 0f				// KASLR disabled? just proceed
463	orr	x23, x23, x0			// record KASLR offset
464	ldp	x29, x30, [sp], #16		// we must enable KASLR, return
465	ret					// to __primary_switch()
4660:
467#endif
468	add	sp, sp, #16
469	mov	x29, #0
470	mov	x30, #0
471	b	start_kernel
472ENDPROC(__primary_switched)
473
474/*
475 * end early head section, begin head code that is also used for
476 * hotplug and needs to have the same protections as the text region
477 */
478	.section ".idmap.text","ax"
479
480ENTRY(kimage_vaddr)
481	.quad		_text - TEXT_OFFSET
482
483/*
484 * If we're fortunate enough to boot at EL2, ensure that the world is
485 * sane before dropping to EL1.
486 *
487 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in w0 if
488 * booted in EL1 or EL2 respectively.
489 */
490ENTRY(el2_setup)
491	msr	SPsel, #1			// We want to use SP_EL{1,2}
492	mrs	x0, CurrentEL
493	cmp	x0, #CurrentEL_EL2
494	b.eq	1f
495	mov_q	x0, (SCTLR_EL1_RES1 | ENDIAN_SET_EL1)
496	msr	sctlr_el1, x0
497	mov	w0, #BOOT_CPU_MODE_EL1		// This cpu booted in EL1
498	isb
499	ret
500
5011:	mov_q	x0, (SCTLR_EL2_RES1 | ENDIAN_SET_EL2)
502	msr	sctlr_el2, x0
503
504#ifdef CONFIG_ARM64_VHE
505	/*
506	 * Check for VHE being present. For the rest of the EL2 setup,
507	 * x2 being non-zero indicates that we do have VHE, and that the
508	 * kernel is intended to run at EL2.
509	 */
510	mrs	x2, id_aa64mmfr1_el1
511	ubfx	x2, x2, #8, #4
512#else
513	mov	x2, xzr
514#endif
515
516	/* Hyp configuration. */
517	mov	x0, #HCR_RW			// 64-bit EL1
518	cbz	x2, set_hcr
519	orr	x0, x0, #HCR_TGE		// Enable Host Extensions
520	orr	x0, x0, #HCR_E2H
521set_hcr:
522	msr	hcr_el2, x0
523	isb
524
525	/*
526	 * Allow Non-secure EL1 and EL0 to access physical timer and counter.
527	 * This is not necessary for VHE, since the host kernel runs in EL2,
528	 * and EL0 accesses are configured in the later stage of boot process.
529	 * Note that when HCR_EL2.E2H == 1, CNTHCTL_EL2 has the same bit layout
530	 * as CNTKCTL_EL1, and CNTKCTL_EL1 accessing instructions are redefined
531	 * to access CNTHCTL_EL2. This allows the kernel designed to run at EL1
532	 * to transparently mess with the EL0 bits via CNTKCTL_EL1 access in
533	 * EL2.
534	 */
535	cbnz	x2, 1f
536	mrs	x0, cnthctl_el2
537	orr	x0, x0, #3			// Enable EL1 physical timers
538	msr	cnthctl_el2, x0
5391:
540	msr	cntvoff_el2, xzr		// Clear virtual offset
541
542#ifdef CONFIG_ARM_GIC_V3
543	/* GICv3 system register access */
544	mrs	x0, id_aa64pfr0_el1
545	ubfx	x0, x0, #24, #4
546	cmp	x0, #1
547	b.ne	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		// Check ID_AA64DFR0_EL1 PMUVer
573	sbfx	x0, x1, #8, #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, #32, #4			// Check ID_AA64DFR0_EL1 PMSVer
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	/* Stage-2 translation */
601	msr	vttbr_el2, xzr
602
603	cbz	x2, install_el2_stub
604
605	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
606	isb
607	ret
608
609install_el2_stub:
610	/*
611	 * When VHE is not in use, early init of EL2 and EL1 needs to be
612	 * done here.
613	 * When VHE _is_ in use, EL1 will not be used in the host and
614	 * requires no configuration, and all non-hyp-specific EL2 setup
615	 * will be done via the _EL1 system register aliases in __cpu_setup.
616	 */
617	mov_q	x0, (SCTLR_EL1_RES1 | ENDIAN_SET_EL1)
618	msr	sctlr_el1, x0
619
620	/* Coprocessor traps. */
621	mov	x0, #0x33ff
622	msr	cptr_el2, x0			// Disable copro. traps to EL2
623
624	/* SVE register access */
625	mrs	x1, id_aa64pfr0_el1
626	ubfx	x1, x1, #ID_AA64PFR0_SVE_SHIFT, #4
627	cbz	x1, 7f
628
629	bic	x0, x0, #CPTR_EL2_TZ		// Also disable SVE traps
630	msr	cptr_el2, x0			// Disable copro. traps to EL2
631	isb
632	mov	x1, #ZCR_ELx_LEN_MASK		// SVE: Enable full vector
633	msr_s	SYS_ZCR_EL2, x1			// length for EL1.
634
635	/* Hypervisor stub */
6367:	adr_l	x0, __hyp_stub_vectors
637	msr	vbar_el2, x0
638
639	/* spsr */
640	mov	x0, #(PSR_F_BIT | PSR_I_BIT | PSR_A_BIT | PSR_D_BIT |\
641		      PSR_MODE_EL1h)
642	msr	spsr_el2, x0
643	msr	elr_el2, lr
644	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
645	eret
646ENDPROC(el2_setup)
647
648/*
649 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
650 * in w0. See arch/arm64/include/asm/virt.h for more info.
651 */
652set_cpu_boot_mode_flag:
653	adr_l	x1, __boot_cpu_mode
654	cmp	w0, #BOOT_CPU_MODE_EL2
655	b.ne	1f
656	add	x1, x1, #4
6571:	str	w0, [x1]			// This CPU has booted in EL1
658	dmb	sy
659	dc	ivac, x1			// Invalidate potentially stale cache line
660	ret
661ENDPROC(set_cpu_boot_mode_flag)
662
663/*
664 * These values are written with the MMU off, but read with the MMU on.
665 * Writers will invalidate the corresponding address, discarding up to a
666 * 'Cache Writeback Granule' (CWG) worth of data. The linker script ensures
667 * sufficient alignment that the CWG doesn't overlap another section.
668 */
669	.pushsection ".mmuoff.data.write", "aw"
670/*
671 * We need to find out the CPU boot mode long after boot, so we need to
672 * store it in a writable variable.
673 *
674 * This is not in .bss, because we set it sufficiently early that the boot-time
675 * zeroing of .bss would clobber it.
676 */
677ENTRY(__boot_cpu_mode)
678	.long	BOOT_CPU_MODE_EL2
679	.long	BOOT_CPU_MODE_EL1
680/*
681 * The booting CPU updates the failed status @__early_cpu_boot_status,
682 * with MMU turned off.
683 */
684ENTRY(__early_cpu_boot_status)
685	.long 	0
686
687	.popsection
688
689	/*
690	 * This provides a "holding pen" for platforms to hold all secondary
691	 * cores are held until we're ready for them to initialise.
692	 */
693ENTRY(secondary_holding_pen)
694	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
695	bl	set_cpu_boot_mode_flag
696	mrs	x0, mpidr_el1
697	mov_q	x1, MPIDR_HWID_BITMASK
698	and	x0, x0, x1
699	adr_l	x3, secondary_holding_pen_release
700pen:	ldr	x4, [x3]
701	cmp	x4, x0
702	b.eq	secondary_startup
703	wfe
704	b	pen
705ENDPROC(secondary_holding_pen)
706
707	/*
708	 * Secondary entry point that jumps straight into the kernel. Only to
709	 * be used where CPUs are brought online dynamically by the kernel.
710	 */
711ENTRY(secondary_entry)
712	bl	el2_setup			// Drop to EL1
713	bl	set_cpu_boot_mode_flag
714	b	secondary_startup
715ENDPROC(secondary_entry)
716
717secondary_startup:
718	/*
719	 * Common entry point for secondary CPUs.
720	 */
721	bl	__cpu_setup			// initialise processor
722	bl	__enable_mmu
723	ldr	x8, =__secondary_switched
724	br	x8
725ENDPROC(secondary_startup)
726
727__secondary_switched:
728	adr_l	x5, vectors
729	msr	vbar_el1, x5
730	isb
731
732	adr_l	x0, secondary_data
733	ldr	x1, [x0, #CPU_BOOT_STACK]	// get secondary_data.stack
734	mov	sp, x1
735	ldr	x2, [x0, #CPU_BOOT_TASK]
736	msr	sp_el0, x2
737	mov	x29, #0
738	mov	x30, #0
739	b	secondary_start_kernel
740ENDPROC(__secondary_switched)
741
742/*
743 * The booting CPU updates the failed status @__early_cpu_boot_status,
744 * with MMU turned off.
745 *
746 * update_early_cpu_boot_status tmp, status
747 *  - Corrupts tmp1, tmp2
748 *  - Writes 'status' to __early_cpu_boot_status and makes sure
749 *    it is committed to memory.
750 */
751
752	.macro	update_early_cpu_boot_status status, tmp1, tmp2
753	mov	\tmp2, #\status
754	adr_l	\tmp1, __early_cpu_boot_status
755	str	\tmp2, [\tmp1]
756	dmb	sy
757	dc	ivac, \tmp1			// Invalidate potentially stale cache line
758	.endm
759
760/*
761 * Enable the MMU.
762 *
763 *  x0  = SCTLR_EL1 value for turning on the MMU.
764 *
765 * Returns to the caller via x30/lr. This requires the caller to be covered
766 * by the .idmap.text section.
767 *
768 * Checks if the selected granule size is supported by the CPU.
769 * If it isn't, park the CPU
770 */
771ENTRY(__enable_mmu)
772	mrs	x1, ID_AA64MMFR0_EL1
773	ubfx	x2, x1, #ID_AA64MMFR0_TGRAN_SHIFT, 4
774	cmp	x2, #ID_AA64MMFR0_TGRAN_SUPPORTED
775	b.ne	__no_granule_support
776	update_early_cpu_boot_status 0, x1, x2
777	adrp	x1, idmap_pg_dir
778	adrp	x2, swapper_pg_dir
779	phys_to_ttbr x1, x3
780	phys_to_ttbr x2, x4
781	msr	ttbr0_el1, x3			// load TTBR0
782	msr	ttbr1_el1, x4			// load TTBR1
783	isb
784	msr	sctlr_el1, x0
785	isb
786	/*
787	 * Invalidate the local I-cache so that any instructions fetched
788	 * speculatively from the PoC are discarded, since they may have
789	 * been dynamically patched at the PoU.
790	 */
791	ic	iallu
792	dsb	nsh
793	isb
794	ret
795ENDPROC(__enable_mmu)
796
797__no_granule_support:
798	/* Indicate that this CPU can't boot and is stuck in the kernel */
799	update_early_cpu_boot_status CPU_STUCK_IN_KERNEL, x1, x2
8001:
801	wfe
802	wfi
803	b	1b
804ENDPROC(__no_granule_support)
805
806#ifdef CONFIG_RELOCATABLE
807__relocate_kernel:
808	/*
809	 * Iterate over each entry in the relocation table, and apply the
810	 * relocations in place.
811	 */
812	ldr	w9, =__rela_offset		// offset to reloc table
813	ldr	w10, =__rela_size		// size of reloc table
814
815	mov_q	x11, KIMAGE_VADDR		// default virtual offset
816	add	x11, x11, x23			// actual virtual offset
817	add	x9, x9, x11			// __va(.rela)
818	add	x10, x9, x10			// __va(.rela) + sizeof(.rela)
819
8200:	cmp	x9, x10
821	b.hs	1f
822	ldp	x11, x12, [x9], #24
823	ldr	x13, [x9, #-8]
824	cmp	w12, #R_AARCH64_RELATIVE
825	b.ne	0b
826	add	x13, x13, x23			// relocate
827	str	x13, [x11, x23]
828	b	0b
8291:	ret
830ENDPROC(__relocate_kernel)
831#endif
832
833__primary_switch:
834#ifdef CONFIG_RANDOMIZE_BASE
835	mov	x19, x0				// preserve new SCTLR_EL1 value
836	mrs	x20, sctlr_el1			// preserve old SCTLR_EL1 value
837#endif
838
839	bl	__enable_mmu
840#ifdef CONFIG_RELOCATABLE
841	bl	__relocate_kernel
842#ifdef CONFIG_RANDOMIZE_BASE
843	ldr	x8, =__primary_switched
844	adrp	x0, __PHYS_OFFSET
845	blr	x8
846
847	/*
848	 * If we return here, we have a KASLR displacement in x23 which we need
849	 * to take into account by discarding the current kernel mapping and
850	 * creating a new one.
851	 */
852	pre_disable_mmu_workaround
853	msr	sctlr_el1, x20			// disable the MMU
854	isb
855	bl	__create_page_tables		// recreate kernel mapping
856
857	tlbi	vmalle1				// Remove any stale TLB entries
858	dsb	nsh
859
860	msr	sctlr_el1, x19			// re-enable the MMU
861	isb
862	ic	iallu				// flush instructions fetched
863	dsb	nsh				// via old mapping
864	isb
865
866	bl	__relocate_kernel
867#endif
868#endif
869	ldr	x8, =__primary_switched
870	adrp	x0, __PHYS_OFFSET
871	br	x8
872ENDPROC(__primary_switch)
873