xref: /openbmc/linux/arch/arm64/kernel/head.S (revision b830f94f)
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
16#include <asm/assembler.h>
17#include <asm/boot.h>
18#include <asm/ptrace.h>
19#include <asm/asm-offsets.h>
20#include <asm/cache.h>
21#include <asm/cputype.h>
22#include <asm/elf.h>
23#include <asm/image.h>
24#include <asm/kernel-pgtable.h>
25#include <asm/kvm_arm.h>
26#include <asm/memory.h>
27#include <asm/pgtable-hwdef.h>
28#include <asm/pgtable.h>
29#include <asm/page.h>
30#include <asm/smp.h>
31#include <asm/sysreg.h>
32#include <asm/thread_info.h>
33#include <asm/virt.h>
34
35#include "efi-header.S"
36
37#define __PHYS_OFFSET	(KERNEL_START - TEXT_OFFSET)
38
39#if (TEXT_OFFSET & 0xfff) != 0
40#error TEXT_OFFSET must be at least 4KB aligned
41#elif (PAGE_OFFSET & 0x1fffff) != 0
42#error PAGE_OFFSET must be at least 2MB aligned
43#elif TEXT_OFFSET > 0x1fffff
44#error TEXT_OFFSET must be less than 2MB
45#endif
46
47/*
48 * Kernel startup entry point.
49 * ---------------------------
50 *
51 * The requirements are:
52 *   MMU = off, D-cache = off, I-cache = on or off,
53 *   x0 = physical address to the FDT blob.
54 *
55 * This code is mostly position independent so you call this at
56 * __pa(PAGE_OFFSET + TEXT_OFFSET).
57 *
58 * Note that the callee-saved registers are used for storing variables
59 * that are useful before the MMU is enabled. The allocations are described
60 * in the entry routines.
61 */
62	__HEAD
63_head:
64	/*
65	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
66	 */
67#ifdef CONFIG_EFI
68	/*
69	 * This add instruction has no meaningful effect except that
70	 * its opcode forms the magic "MZ" signature required by UEFI.
71	 */
72	add	x13, x18, #0x16
73	b	stext
74#else
75	b	stext				// branch to kernel start, magic
76	.long	0				// reserved
77#endif
78	le64sym	_kernel_offset_le		// Image load offset from start of RAM, little-endian
79	le64sym	_kernel_size_le			// Effective size of kernel image, little-endian
80	le64sym	_kernel_flags_le		// Informative flags, little-endian
81	.quad	0				// reserved
82	.quad	0				// reserved
83	.quad	0				// reserved
84	.ascii	ARM64_IMAGE_MAGIC		// Magic number
85#ifdef CONFIG_EFI
86	.long	pe_header - _head		// Offset to the PE header.
87
88pe_header:
89	__EFI_PE_HEADER
90#else
91	.long	0				// reserved
92#endif
93
94	__INIT
95
96	/*
97	 * The following callee saved general purpose registers are used on the
98	 * primary lowlevel boot path:
99	 *
100	 *  Register   Scope                      Purpose
101	 *  x21        stext() .. start_kernel()  FDT pointer passed at boot in x0
102	 *  x23        stext() .. start_kernel()  physical misalignment/KASLR offset
103	 *  x28        __create_page_tables()     callee preserved temp register
104	 *  x19/x20    __primary_switch()         callee preserved temp registers
105	 */
106ENTRY(stext)
107	bl	preserve_boot_args
108	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
109	adrp	x23, __PHYS_OFFSET
110	and	x23, x23, MIN_KIMG_ALIGN - 1	// KASLR offset, defaults to 0
111	bl	set_cpu_boot_mode_flag
112	bl	__create_page_tables
113	/*
114	 * The following calls CPU setup code, see arch/arm64/mm/proc.S for
115	 * details.
116	 * On return, the CPU will be ready for the MMU to be turned on and
117	 * the TCR will have been set.
118	 */
119	bl	__cpu_setup			// initialise processor
120	b	__primary_switch
121ENDPROC(stext)
122
123/*
124 * Preserve the arguments passed by the bootloader in x0 .. x3
125 */
126preserve_boot_args:
127	mov	x21, x0				// x21=FDT
128
129	adr_l	x0, boot_args			// record the contents of
130	stp	x21, x1, [x0]			// x0 .. x3 at kernel entry
131	stp	x2, x3, [x0, #16]
132
133	dmb	sy				// needed before dc ivac with
134						// MMU off
135
136	mov	x1, #0x20			// 4 x 8 bytes
137	b	__inval_dcache_area		// tail call
138ENDPROC(preserve_boot_args)
139
140/*
141 * Macro to create a table entry to the next page.
142 *
143 *	tbl:	page table address
144 *	virt:	virtual address
145 *	shift:	#imm page table shift
146 *	ptrs:	#imm pointers per table page
147 *
148 * Preserves:	virt
149 * Corrupts:	ptrs, tmp1, tmp2
150 * Returns:	tbl -> next level table page address
151 */
152	.macro	create_table_entry, tbl, virt, shift, ptrs, tmp1, tmp2
153	add	\tmp1, \tbl, #PAGE_SIZE
154	phys_to_pte \tmp2, \tmp1
155	orr	\tmp2, \tmp2, #PMD_TYPE_TABLE	// address of next table and entry type
156	lsr	\tmp1, \virt, #\shift
157	sub	\ptrs, \ptrs, #1
158	and	\tmp1, \tmp1, \ptrs		// table index
159	str	\tmp2, [\tbl, \tmp1, lsl #3]
160	add	\tbl, \tbl, #PAGE_SIZE		// next level table page
161	.endm
162
163/*
164 * Macro to populate page table entries, these entries can be pointers to the next level
165 * or last level entries pointing to physical memory.
166 *
167 *	tbl:	page table address
168 *	rtbl:	pointer to page table or physical memory
169 *	index:	start index to write
170 *	eindex:	end index to write - [index, eindex] written to
171 *	flags:	flags for pagetable entry to or in
172 *	inc:	increment to rtbl between each entry
173 *	tmp1:	temporary variable
174 *
175 * Preserves:	tbl, eindex, flags, inc
176 * Corrupts:	index, tmp1
177 * Returns:	rtbl
178 */
179	.macro populate_entries, tbl, rtbl, index, eindex, flags, inc, tmp1
180.Lpe\@:	phys_to_pte \tmp1, \rtbl
181	orr	\tmp1, \tmp1, \flags	// tmp1 = table entry
182	str	\tmp1, [\tbl, \index, lsl #3]
183	add	\rtbl, \rtbl, \inc	// rtbl = pa next level
184	add	\index, \index, #1
185	cmp	\index, \eindex
186	b.ls	.Lpe\@
187	.endm
188
189/*
190 * Compute indices of table entries from virtual address range. If multiple entries
191 * were needed in the previous page table level then the next page table level is assumed
192 * to be composed of multiple pages. (This effectively scales the end index).
193 *
194 *	vstart:	virtual address of start of range
195 *	vend:	virtual address of end of range
196 *	shift:	shift used to transform virtual address into index
197 *	ptrs:	number of entries in page table
198 *	istart:	index in table corresponding to vstart
199 *	iend:	index in table corresponding to vend
200 *	count:	On entry: how many extra entries were required in previous level, scales
201 *			  our end index.
202 *		On exit: returns how many extra entries required for next page table level
203 *
204 * Preserves:	vstart, vend, shift, ptrs
205 * Returns:	istart, iend, count
206 */
207	.macro compute_indices, vstart, vend, shift, ptrs, istart, iend, count
208	lsr	\iend, \vend, \shift
209	mov	\istart, \ptrs
210	sub	\istart, \istart, #1
211	and	\iend, \iend, \istart	// iend = (vend >> shift) & (ptrs - 1)
212	mov	\istart, \ptrs
213	mul	\istart, \istart, \count
214	add	\iend, \iend, \istart	// iend += (count - 1) * ptrs
215					// our entries span multiple tables
216
217	lsr	\istart, \vstart, \shift
218	mov	\count, \ptrs
219	sub	\count, \count, #1
220	and	\istart, \istart, \count
221
222	sub	\count, \iend, \istart
223	.endm
224
225/*
226 * Map memory for specified virtual address range. Each level of page table needed supports
227 * multiple entries. If a level requires n entries the next page table level is assumed to be
228 * formed from n pages.
229 *
230 *	tbl:	location of page table
231 *	rtbl:	address to be used for first level page table entry (typically tbl + PAGE_SIZE)
232 *	vstart:	start address to map
233 *	vend:	end address to map - we map [vstart, vend]
234 *	flags:	flags to use to map last level entries
235 *	phys:	physical address corresponding to vstart - physical memory is contiguous
236 *	pgds:	the number of pgd entries
237 *
238 * Temporaries:	istart, iend, tmp, count, sv - these need to be different registers
239 * Preserves:	vstart, vend, flags
240 * Corrupts:	tbl, rtbl, istart, iend, tmp, count, sv
241 */
242	.macro map_memory, tbl, rtbl, vstart, vend, flags, phys, pgds, istart, iend, tmp, count, sv
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 */
276__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_USER_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
3181:
319	adr_l	x6, vabits_user
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)	// 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 + TEXT_OFFSET	// 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 the idmap and swapper page
395	 * tables again to remove any speculatively loaded cache lines.
396	 */
397	adrp	x0, idmap_pg_dir
398	adrp	x1, init_pg_end
399	sub	x1, x1, x0
400	dmb	sy
401	bl	__inval_dcache_area
402
403	ret	x28
404ENDPROC(__create_page_tables)
405	.ltorg
406
407/*
408 * The following fragment of code is executed with the MMU enabled.
409 *
410 *   x0 = __PHYS_OFFSET
411 */
412__primary_switched:
413	adrp	x4, init_thread_union
414	add	sp, x4, #THREAD_SIZE
415	adr_l	x5, init_task
416	msr	sp_el0, x5			// Save thread_info
417
418	adr_l	x8, vectors			// load VBAR_EL1 with virtual
419	msr	vbar_el1, x8			// vector table address
420	isb
421
422	stp	xzr, x30, [sp, #-16]!
423	mov	x29, sp
424
425	str_l	x21, __fdt_pointer, x5		// Save FDT pointer
426
427	ldr_l	x4, kimage_vaddr		// Save the offset between
428	sub	x4, x4, x0			// the kernel virtual and
429	str_l	x4, kimage_voffset, x5		// physical mappings
430
431	// Clear BSS
432	adr_l	x0, __bss_start
433	mov	x1, xzr
434	adr_l	x2, __bss_stop
435	sub	x2, x2, x0
436	bl	__pi_memset
437	dsb	ishst				// Make zero page visible to PTW
438
439#ifdef CONFIG_KASAN
440	bl	kasan_early_init
441#endif
442#ifdef CONFIG_RANDOMIZE_BASE
443	tst	x23, ~(MIN_KIMG_ALIGN - 1)	// already running randomized?
444	b.ne	0f
445	mov	x0, x21				// pass FDT address in x0
446	bl	kaslr_early_init		// parse FDT for KASLR options
447	cbz	x0, 0f				// KASLR disabled? just proceed
448	orr	x23, x23, x0			// record KASLR offset
449	ldp	x29, x30, [sp], #16		// we must enable KASLR, return
450	ret					// to __primary_switch()
4510:
452#endif
453	add	sp, sp, #16
454	mov	x29, #0
455	mov	x30, #0
456	b	start_kernel
457ENDPROC(__primary_switched)
458
459/*
460 * end early head section, begin head code that is also used for
461 * hotplug and needs to have the same protections as the text region
462 */
463	.section ".idmap.text","awx"
464
465ENTRY(kimage_vaddr)
466	.quad		_text - TEXT_OFFSET
467EXPORT_SYMBOL(kimage_vaddr)
468
469/*
470 * If we're fortunate enough to boot at EL2, ensure that the world is
471 * sane before dropping to EL1.
472 *
473 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in w0 if
474 * booted in EL1 or EL2 respectively.
475 */
476ENTRY(el2_setup)
477	msr	SPsel, #1			// We want to use SP_EL{1,2}
478	mrs	x0, CurrentEL
479	cmp	x0, #CurrentEL_EL2
480	b.eq	1f
481	mov_q	x0, (SCTLR_EL1_RES1 | ENDIAN_SET_EL1)
482	msr	sctlr_el1, x0
483	mov	w0, #BOOT_CPU_MODE_EL1		// This cpu booted in EL1
484	isb
485	ret
486
4871:	mov_q	x0, (SCTLR_EL2_RES1 | ENDIAN_SET_EL2)
488	msr	sctlr_el2, x0
489
490#ifdef CONFIG_ARM64_VHE
491	/*
492	 * Check for VHE being present. For the rest of the EL2 setup,
493	 * x2 being non-zero indicates that we do have VHE, and that the
494	 * kernel is intended to run at EL2.
495	 */
496	mrs	x2, id_aa64mmfr1_el1
497	ubfx	x2, x2, #ID_AA64MMFR1_VHE_SHIFT, #4
498#else
499	mov	x2, xzr
500#endif
501
502	/* Hyp configuration. */
503	mov_q	x0, HCR_HOST_NVHE_FLAGS
504	cbz	x2, set_hcr
505	mov_q	x0, HCR_HOST_VHE_FLAGS
506set_hcr:
507	msr	hcr_el2, x0
508	isb
509
510	/*
511	 * Allow Non-secure EL1 and EL0 to access physical timer and counter.
512	 * This is not necessary for VHE, since the host kernel runs in EL2,
513	 * and EL0 accesses are configured in the later stage of boot process.
514	 * Note that when HCR_EL2.E2H == 1, CNTHCTL_EL2 has the same bit layout
515	 * as CNTKCTL_EL1, and CNTKCTL_EL1 accessing instructions are redefined
516	 * to access CNTHCTL_EL2. This allows the kernel designed to run at EL1
517	 * to transparently mess with the EL0 bits via CNTKCTL_EL1 access in
518	 * EL2.
519	 */
520	cbnz	x2, 1f
521	mrs	x0, cnthctl_el2
522	orr	x0, x0, #3			// Enable EL1 physical timers
523	msr	cnthctl_el2, x0
5241:
525	msr	cntvoff_el2, xzr		// Clear virtual offset
526
527#ifdef CONFIG_ARM_GIC_V3
528	/* GICv3 system register access */
529	mrs	x0, id_aa64pfr0_el1
530	ubfx	x0, x0, #ID_AA64PFR0_GIC_SHIFT, #4
531	cbz	x0, 3f
532
533	mrs_s	x0, SYS_ICC_SRE_EL2
534	orr	x0, x0, #ICC_SRE_EL2_SRE	// Set ICC_SRE_EL2.SRE==1
535	orr	x0, x0, #ICC_SRE_EL2_ENABLE	// Set ICC_SRE_EL2.Enable==1
536	msr_s	SYS_ICC_SRE_EL2, x0
537	isb					// Make sure SRE is now set
538	mrs_s	x0, SYS_ICC_SRE_EL2		// Read SRE back,
539	tbz	x0, #0, 3f			// and check that it sticks
540	msr_s	SYS_ICH_HCR_EL2, xzr		// Reset ICC_HCR_EL2 to defaults
541
5423:
543#endif
544
545	/* Populate ID registers. */
546	mrs	x0, midr_el1
547	mrs	x1, mpidr_el1
548	msr	vpidr_el2, x0
549	msr	vmpidr_el2, x1
550
551#ifdef CONFIG_COMPAT
552	msr	hstr_el2, xzr			// Disable CP15 traps to EL2
553#endif
554
555	/* EL2 debug */
556	mrs	x1, id_aa64dfr0_el1
557	sbfx	x0, x1, #ID_AA64DFR0_PMUVER_SHIFT, #4
558	cmp	x0, #1
559	b.lt	4f				// Skip if no PMU present
560	mrs	x0, pmcr_el0			// Disable debug access traps
561	ubfx	x0, x0, #11, #5			// to EL2 and allow access to
5624:
563	csel	x3, xzr, x0, lt			// all PMU counters from EL1
564
565	/* Statistical profiling */
566	ubfx	x0, x1, #ID_AA64DFR0_PMSVER_SHIFT, #4
567	cbz	x0, 7f				// Skip if SPE not present
568	cbnz	x2, 6f				// VHE?
569	mrs_s	x4, SYS_PMBIDR_EL1		// If SPE available at EL2,
570	and	x4, x4, #(1 << SYS_PMBIDR_EL1_P_SHIFT)
571	cbnz	x4, 5f				// then permit sampling of physical
572	mov	x4, #(1 << SYS_PMSCR_EL2_PCT_SHIFT | \
573		      1 << SYS_PMSCR_EL2_PA_SHIFT)
574	msr_s	SYS_PMSCR_EL2, x4		// addresses and physical counter
5755:
576	mov	x1, #(MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT)
577	orr	x3, x3, x1			// If we don't have VHE, then
578	b	7f				// use EL1&0 translation.
5796:						// For VHE, use EL2 translation
580	orr	x3, x3, #MDCR_EL2_TPMS		// and disable access from EL1
5817:
582	msr	mdcr_el2, x3			// Configure debug traps
583
584	/* LORegions */
585	mrs	x1, id_aa64mmfr1_el1
586	ubfx	x0, x1, #ID_AA64MMFR1_LOR_SHIFT, 4
587	cbz	x0, 1f
588	msr_s	SYS_LORC_EL1, xzr
5891:
590
591	/* Stage-2 translation */
592	msr	vttbr_el2, xzr
593
594	cbz	x2, install_el2_stub
595
596	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
597	isb
598	ret
599
600install_el2_stub:
601	/*
602	 * When VHE is not in use, early init of EL2 and EL1 needs to be
603	 * done here.
604	 * When VHE _is_ in use, EL1 will not be used in the host and
605	 * requires no configuration, and all non-hyp-specific EL2 setup
606	 * will be done via the _EL1 system register aliases in __cpu_setup.
607	 */
608	mov_q	x0, (SCTLR_EL1_RES1 | ENDIAN_SET_EL1)
609	msr	sctlr_el1, x0
610
611	/* Coprocessor traps. */
612	mov	x0, #0x33ff
613	msr	cptr_el2, x0			// Disable copro. traps to EL2
614
615	/* SVE register access */
616	mrs	x1, id_aa64pfr0_el1
617	ubfx	x1, x1, #ID_AA64PFR0_SVE_SHIFT, #4
618	cbz	x1, 7f
619
620	bic	x0, x0, #CPTR_EL2_TZ		// Also disable SVE traps
621	msr	cptr_el2, x0			// Disable copro. traps to EL2
622	isb
623	mov	x1, #ZCR_ELx_LEN_MASK		// SVE: Enable full vector
624	msr_s	SYS_ZCR_EL2, x1			// length for EL1.
625
626	/* Hypervisor stub */
6277:	adr_l	x0, __hyp_stub_vectors
628	msr	vbar_el2, x0
629
630	/* spsr */
631	mov	x0, #(PSR_F_BIT | PSR_I_BIT | PSR_A_BIT | PSR_D_BIT |\
632		      PSR_MODE_EL1h)
633	msr	spsr_el2, x0
634	msr	elr_el2, lr
635	mov	w0, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
636	eret
637ENDPROC(el2_setup)
638
639/*
640 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
641 * in w0. See arch/arm64/include/asm/virt.h for more info.
642 */
643set_cpu_boot_mode_flag:
644	adr_l	x1, __boot_cpu_mode
645	cmp	w0, #BOOT_CPU_MODE_EL2
646	b.ne	1f
647	add	x1, x1, #4
6481:	str	w0, [x1]			// This CPU has booted in EL1
649	dmb	sy
650	dc	ivac, x1			// Invalidate potentially stale cache line
651	ret
652ENDPROC(set_cpu_boot_mode_flag)
653
654/*
655 * These values are written with the MMU off, but read with the MMU on.
656 * Writers will invalidate the corresponding address, discarding up to a
657 * 'Cache Writeback Granule' (CWG) worth of data. The linker script ensures
658 * sufficient alignment that the CWG doesn't overlap another section.
659 */
660	.pushsection ".mmuoff.data.write", "aw"
661/*
662 * We need to find out the CPU boot mode long after boot, so we need to
663 * store it in a writable variable.
664 *
665 * This is not in .bss, because we set it sufficiently early that the boot-time
666 * zeroing of .bss would clobber it.
667 */
668ENTRY(__boot_cpu_mode)
669	.long	BOOT_CPU_MODE_EL2
670	.long	BOOT_CPU_MODE_EL1
671/*
672 * The booting CPU updates the failed status @__early_cpu_boot_status,
673 * with MMU turned off.
674 */
675ENTRY(__early_cpu_boot_status)
676	.quad 	0
677
678	.popsection
679
680	/*
681	 * This provides a "holding pen" for platforms to hold all secondary
682	 * cores are held until we're ready for them to initialise.
683	 */
684ENTRY(secondary_holding_pen)
685	bl	el2_setup			// Drop to EL1, w0=cpu_boot_mode
686	bl	set_cpu_boot_mode_flag
687	mrs	x0, mpidr_el1
688	mov_q	x1, MPIDR_HWID_BITMASK
689	and	x0, x0, x1
690	adr_l	x3, secondary_holding_pen_release
691pen:	ldr	x4, [x3]
692	cmp	x4, x0
693	b.eq	secondary_startup
694	wfe
695	b	pen
696ENDPROC(secondary_holding_pen)
697
698	/*
699	 * Secondary entry point that jumps straight into the kernel. Only to
700	 * be used where CPUs are brought online dynamically by the kernel.
701	 */
702ENTRY(secondary_entry)
703	bl	el2_setup			// Drop to EL1
704	bl	set_cpu_boot_mode_flag
705	b	secondary_startup
706ENDPROC(secondary_entry)
707
708secondary_startup:
709	/*
710	 * Common entry point for secondary CPUs.
711	 */
712	bl	__cpu_secondary_check52bitva
713	bl	__cpu_setup			// initialise processor
714	adrp	x1, swapper_pg_dir
715	bl	__enable_mmu
716	ldr	x8, =__secondary_switched
717	br	x8
718ENDPROC(secondary_startup)
719
720__secondary_switched:
721	adr_l	x5, vectors
722	msr	vbar_el1, x5
723	isb
724
725	adr_l	x0, secondary_data
726	ldr	x1, [x0, #CPU_BOOT_STACK]	// get secondary_data.stack
727	mov	sp, x1
728	ldr	x2, [x0, #CPU_BOOT_TASK]
729	msr	sp_el0, x2
730	mov	x29, #0
731	mov	x30, #0
732	b	secondary_start_kernel
733ENDPROC(__secondary_switched)
734
735/*
736 * The booting CPU updates the failed status @__early_cpu_boot_status,
737 * with MMU turned off.
738 *
739 * update_early_cpu_boot_status tmp, status
740 *  - Corrupts tmp1, tmp2
741 *  - Writes 'status' to __early_cpu_boot_status and makes sure
742 *    it is committed to memory.
743 */
744
745	.macro	update_early_cpu_boot_status status, tmp1, tmp2
746	mov	\tmp2, #\status
747	adr_l	\tmp1, __early_cpu_boot_status
748	str	\tmp2, [\tmp1]
749	dmb	sy
750	dc	ivac, \tmp1			// Invalidate potentially stale cache line
751	.endm
752
753/*
754 * Enable the MMU.
755 *
756 *  x0  = SCTLR_EL1 value for turning on the MMU.
757 *  x1  = TTBR1_EL1 value
758 *
759 * Returns to the caller via x30/lr. This requires the caller to be covered
760 * by the .idmap.text section.
761 *
762 * Checks if the selected granule size is supported by the CPU.
763 * If it isn't, park the CPU
764 */
765ENTRY(__enable_mmu)
766	mrs	x2, ID_AA64MMFR0_EL1
767	ubfx	x2, x2, #ID_AA64MMFR0_TGRAN_SHIFT, 4
768	cmp	x2, #ID_AA64MMFR0_TGRAN_SUPPORTED
769	b.ne	__no_granule_support
770	update_early_cpu_boot_status 0, x2, x3
771	adrp	x2, idmap_pg_dir
772	phys_to_ttbr x1, x1
773	phys_to_ttbr x2, x2
774	msr	ttbr0_el1, x2			// load TTBR0
775	offset_ttbr1 x1
776	msr	ttbr1_el1, x1			// load TTBR1
777	isb
778	msr	sctlr_el1, x0
779	isb
780	/*
781	 * Invalidate the local I-cache so that any instructions fetched
782	 * speculatively from the PoC are discarded, since they may have
783	 * been dynamically patched at the PoU.
784	 */
785	ic	iallu
786	dsb	nsh
787	isb
788	ret
789ENDPROC(__enable_mmu)
790
791ENTRY(__cpu_secondary_check52bitva)
792#ifdef CONFIG_ARM64_USER_VA_BITS_52
793	ldr_l	x0, vabits_user
794	cmp	x0, #52
795	b.ne	2f
796
797	mrs_s	x0, SYS_ID_AA64MMFR2_EL1
798	and	x0, x0, #(0xf << ID_AA64MMFR2_LVA_SHIFT)
799	cbnz	x0, 2f
800
801	update_early_cpu_boot_status \
802		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_52_BIT_VA, x0, x1
8031:	wfe
804	wfi
805	b	1b
806
807#endif
8082:	ret
809ENDPROC(__cpu_secondary_check52bitva)
810
811__no_granule_support:
812	/* Indicate that this CPU can't boot and is stuck in the kernel */
813	update_early_cpu_boot_status \
814		CPU_STUCK_IN_KERNEL | CPU_STUCK_REASON_NO_GRAN, x1, x2
8151:
816	wfe
817	wfi
818	b	1b
819ENDPROC(__no_granule_support)
820
821#ifdef CONFIG_RELOCATABLE
822__relocate_kernel:
823	/*
824	 * Iterate over each entry in the relocation table, and apply the
825	 * relocations in place.
826	 */
827	ldr	w9, =__rela_offset		// offset to reloc table
828	ldr	w10, =__rela_size		// size of reloc table
829
830	mov_q	x11, KIMAGE_VADDR		// default virtual offset
831	add	x11, x11, x23			// actual virtual offset
832	add	x9, x9, x11			// __va(.rela)
833	add	x10, x9, x10			// __va(.rela) + sizeof(.rela)
834
8350:	cmp	x9, x10
836	b.hs	1f
837	ldp	x11, x12, [x9], #24
838	ldr	x13, [x9, #-8]
839	cmp	w12, #R_AARCH64_RELATIVE
840	b.ne	0b
841	add	x13, x13, x23			// relocate
842	str	x13, [x11, x23]
843	b	0b
8441:	ret
845ENDPROC(__relocate_kernel)
846#endif
847
848__primary_switch:
849#ifdef CONFIG_RANDOMIZE_BASE
850	mov	x19, x0				// preserve new SCTLR_EL1 value
851	mrs	x20, sctlr_el1			// preserve old SCTLR_EL1 value
852#endif
853
854	adrp	x1, init_pg_dir
855	bl	__enable_mmu
856#ifdef CONFIG_RELOCATABLE
857	bl	__relocate_kernel
858#ifdef CONFIG_RANDOMIZE_BASE
859	ldr	x8, =__primary_switched
860	adrp	x0, __PHYS_OFFSET
861	blr	x8
862
863	/*
864	 * If we return here, we have a KASLR displacement in x23 which we need
865	 * to take into account by discarding the current kernel mapping and
866	 * creating a new one.
867	 */
868	pre_disable_mmu_workaround
869	msr	sctlr_el1, x20			// disable the MMU
870	isb
871	bl	__create_page_tables		// recreate kernel mapping
872
873	tlbi	vmalle1				// Remove any stale TLB entries
874	dsb	nsh
875
876	msr	sctlr_el1, x19			// re-enable the MMU
877	isb
878	ic	iallu				// flush instructions fetched
879	dsb	nsh				// via old mapping
880	isb
881
882	bl	__relocate_kernel
883#endif
884#endif
885	ldr	x8, =__primary_switched
886	adrp	x0, __PHYS_OFFSET
887	br	x8
888ENDPROC(__primary_switch)
889