xref: /openbmc/linux/arch/arm64/kernel/head.S (revision 12eb4683)
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
26#include <asm/assembler.h>
27#include <asm/ptrace.h>
28#include <asm/asm-offsets.h>
29#include <asm/cputype.h>
30#include <asm/memory.h>
31#include <asm/thread_info.h>
32#include <asm/pgtable-hwdef.h>
33#include <asm/pgtable.h>
34#include <asm/page.h>
35#include <asm/virt.h>
36
37/*
38 * swapper_pg_dir is the virtual address of the initial page table. We place
39 * the page tables 3 * PAGE_SIZE below KERNEL_RAM_VADDR. The idmap_pg_dir has
40 * 2 pages and is placed below swapper_pg_dir.
41 */
42#define KERNEL_RAM_VADDR	(PAGE_OFFSET + TEXT_OFFSET)
43
44#if (KERNEL_RAM_VADDR & 0xfffff) != 0x80000
45#error KERNEL_RAM_VADDR must start at 0xXXX80000
46#endif
47
48#define SWAPPER_DIR_SIZE	(3 * PAGE_SIZE)
49#define IDMAP_DIR_SIZE		(2 * PAGE_SIZE)
50
51	.globl	swapper_pg_dir
52	.equ	swapper_pg_dir, KERNEL_RAM_VADDR - SWAPPER_DIR_SIZE
53
54	.globl	idmap_pg_dir
55	.equ	idmap_pg_dir, swapper_pg_dir - IDMAP_DIR_SIZE
56
57	.macro	pgtbl, ttb0, ttb1, phys
58	add	\ttb1, \phys, #TEXT_OFFSET - SWAPPER_DIR_SIZE
59	sub	\ttb0, \ttb1, #IDMAP_DIR_SIZE
60	.endm
61
62#ifdef CONFIG_ARM64_64K_PAGES
63#define BLOCK_SHIFT	PAGE_SHIFT
64#define BLOCK_SIZE	PAGE_SIZE
65#else
66#define BLOCK_SHIFT	SECTION_SHIFT
67#define BLOCK_SIZE	SECTION_SIZE
68#endif
69
70#define KERNEL_START	KERNEL_RAM_VADDR
71#define KERNEL_END	_end
72
73/*
74 * Initial memory map attributes.
75 */
76#ifndef CONFIG_SMP
77#define PTE_FLAGS	PTE_TYPE_PAGE | PTE_AF
78#define PMD_FLAGS	PMD_TYPE_SECT | PMD_SECT_AF
79#else
80#define PTE_FLAGS	PTE_TYPE_PAGE | PTE_AF | PTE_SHARED
81#define PMD_FLAGS	PMD_TYPE_SECT | PMD_SECT_AF | PMD_SECT_S
82#endif
83
84#ifdef CONFIG_ARM64_64K_PAGES
85#define MM_MMUFLAGS	PTE_ATTRINDX(MT_NORMAL) | PTE_FLAGS
86#else
87#define MM_MMUFLAGS	PMD_ATTRINDX(MT_NORMAL) | PMD_FLAGS
88#endif
89
90/*
91 * Kernel startup entry point.
92 * ---------------------------
93 *
94 * The requirements are:
95 *   MMU = off, D-cache = off, I-cache = on or off,
96 *   x0 = physical address to the FDT blob.
97 *
98 * This code is mostly position independent so you call this at
99 * __pa(PAGE_OFFSET + TEXT_OFFSET).
100 *
101 * Note that the callee-saved registers are used for storing variables
102 * that are useful before the MMU is enabled. The allocations are described
103 * in the entry routines.
104 */
105	__HEAD
106
107	/*
108	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
109	 */
110	b	stext				// branch to kernel start, magic
111	.long	0				// reserved
112	.quad	TEXT_OFFSET			// Image load offset from start of RAM
113	.quad	0				// reserved
114	.quad	0				// reserved
115	.quad	0				// reserved
116	.quad	0				// reserved
117	.quad	0				// reserved
118	.byte	0x41				// Magic number, "ARM\x64"
119	.byte	0x52
120	.byte	0x4d
121	.byte	0x64
122	.word	0				// reserved
123
124ENTRY(stext)
125	mov	x21, x0				// x21=FDT
126	bl	el2_setup			// Drop to EL1, w20=cpu_boot_mode
127	bl	__calc_phys_offset		// x24=PHYS_OFFSET, x28=PHYS_OFFSET-PAGE_OFFSET
128	bl	set_cpu_boot_mode_flag
129	mrs	x22, midr_el1			// x22=cpuid
130	mov	x0, x22
131	bl	lookup_processor_type
132	mov	x23, x0				// x23=current cpu_table
133	cbz	x23, __error_p			// invalid processor (x23=0)?
134	bl	__vet_fdt
135	bl	__create_page_tables		// x25=TTBR0, x26=TTBR1
136	/*
137	 * The following calls CPU specific code in a position independent
138	 * manner. See arch/arm64/mm/proc.S for details. x23 = base of
139	 * cpu_info structure selected by lookup_processor_type above.
140	 * On return, the CPU will be ready for the MMU to be turned on and
141	 * the TCR will have been set.
142	 */
143	ldr	x27, __switch_data		// address to jump to after
144						// MMU has been enabled
145	adr	lr, __enable_mmu		// return (PIC) address
146	ldr	x12, [x23, #CPU_INFO_SETUP]
147	add	x12, x12, x28			// __virt_to_phys
148	br	x12				// initialise processor
149ENDPROC(stext)
150
151/*
152 * If we're fortunate enough to boot at EL2, ensure that the world is
153 * sane before dropping to EL1.
154 *
155 * Returns either BOOT_CPU_MODE_EL1 or BOOT_CPU_MODE_EL2 in x20 if
156 * booted in EL1 or EL2 respectively.
157 */
158ENTRY(el2_setup)
159	mrs	x0, CurrentEL
160	cmp	x0, #PSR_MODE_EL2t
161	ccmp	x0, #PSR_MODE_EL2h, #0x4, ne
162	b.ne	1f
163	mrs	x0, sctlr_el2
164CPU_BE(	orr	x0, x0, #(1 << 25)	)	// Set the EE bit for EL2
165CPU_LE(	bic	x0, x0, #(1 << 25)	)	// Clear the EE bit for EL2
166	msr	sctlr_el2, x0
167	b	2f
1681:	mrs	x0, sctlr_el1
169CPU_BE(	orr	x0, x0, #(3 << 24)	)	// Set the EE and E0E bits for EL1
170CPU_LE(	bic	x0, x0, #(3 << 24)	)	// Clear the EE and E0E bits for EL1
171	msr	sctlr_el1, x0
172	mov	w20, #BOOT_CPU_MODE_EL1		// This cpu booted in EL1
173	isb
174	ret
175
176	/* Hyp configuration. */
1772:	mov	x0, #(1 << 31)			// 64-bit EL1
178	msr	hcr_el2, x0
179
180	/* Generic timers. */
181	mrs	x0, cnthctl_el2
182	orr	x0, x0, #3			// Enable EL1 physical timers
183	msr	cnthctl_el2, x0
184	msr	cntvoff_el2, xzr		// Clear virtual offset
185
186	/* Populate ID registers. */
187	mrs	x0, midr_el1
188	mrs	x1, mpidr_el1
189	msr	vpidr_el2, x0
190	msr	vmpidr_el2, x1
191
192	/* sctlr_el1 */
193	mov	x0, #0x0800			// Set/clear RES{1,0} bits
194CPU_BE(	movk	x0, #0x33d0, lsl #16	)	// Set EE and E0E on BE systems
195CPU_LE(	movk	x0, #0x30d0, lsl #16	)	// Clear EE and E0E on LE systems
196	msr	sctlr_el1, x0
197
198	/* Coprocessor traps. */
199	mov	x0, #0x33ff
200	msr	cptr_el2, x0			// Disable copro. traps to EL2
201
202#ifdef CONFIG_COMPAT
203	msr	hstr_el2, xzr			// Disable CP15 traps to EL2
204#endif
205
206	/* Stage-2 translation */
207	msr	vttbr_el2, xzr
208
209	/* Hypervisor stub */
210	adr	x0, __hyp_stub_vectors
211	msr	vbar_el2, x0
212
213	/* spsr */
214	mov	x0, #(PSR_F_BIT | PSR_I_BIT | PSR_A_BIT | PSR_D_BIT |\
215		      PSR_MODE_EL1h)
216	msr	spsr_el2, x0
217	msr	elr_el2, lr
218	mov	w20, #BOOT_CPU_MODE_EL2		// This CPU booted in EL2
219	eret
220ENDPROC(el2_setup)
221
222/*
223 * Sets the __boot_cpu_mode flag depending on the CPU boot mode passed
224 * in x20. See arch/arm64/include/asm/virt.h for more info.
225 */
226ENTRY(set_cpu_boot_mode_flag)
227	ldr	x1, =__boot_cpu_mode		// Compute __boot_cpu_mode
228	add	x1, x1, x28
229	cmp	w20, #BOOT_CPU_MODE_EL2
230	b.ne	1f
231	add	x1, x1, #4
2321:	str	w20, [x1]			// This CPU has booted in EL1
233	ret
234ENDPROC(set_cpu_boot_mode_flag)
235
236/*
237 * We need to find out the CPU boot mode long after boot, so we need to
238 * store it in a writable variable.
239 *
240 * This is not in .bss, because we set it sufficiently early that the boot-time
241 * zeroing of .bss would clobber it.
242 */
243	.pushsection	.data
244ENTRY(__boot_cpu_mode)
245	.long	BOOT_CPU_MODE_EL2
246	.long	0
247	.popsection
248
249	.align	3
2502:	.quad	.
251	.quad	PAGE_OFFSET
252
253#ifdef CONFIG_SMP
254	.align	3
2551:	.quad	.
256	.quad	secondary_holding_pen_release
257
258	/*
259	 * This provides a "holding pen" for platforms to hold all secondary
260	 * cores are held until we're ready for them to initialise.
261	 */
262ENTRY(secondary_holding_pen)
263	bl	el2_setup			// Drop to EL1, w20=cpu_boot_mode
264	bl	__calc_phys_offset		// x24=PHYS_OFFSET, x28=PHYS_OFFSET-PAGE_OFFSET
265	bl	set_cpu_boot_mode_flag
266	mrs	x0, mpidr_el1
267	ldr     x1, =MPIDR_HWID_BITMASK
268	and	x0, x0, x1
269	adr	x1, 1b
270	ldp	x2, x3, [x1]
271	sub	x1, x1, x2
272	add	x3, x3, x1
273pen:	ldr	x4, [x3]
274	cmp	x4, x0
275	b.eq	secondary_startup
276	wfe
277	b	pen
278ENDPROC(secondary_holding_pen)
279
280	/*
281	 * Secondary entry point that jumps straight into the kernel. Only to
282	 * be used where CPUs are brought online dynamically by the kernel.
283	 */
284ENTRY(secondary_entry)
285	bl	el2_setup			// Drop to EL1
286	bl	__calc_phys_offset		// x24=PHYS_OFFSET, x28=PHYS_OFFSET-PAGE_OFFSET
287	bl	set_cpu_boot_mode_flag
288	b	secondary_startup
289ENDPROC(secondary_entry)
290
291ENTRY(secondary_startup)
292	/*
293	 * Common entry point for secondary CPUs.
294	 */
295	mrs	x22, midr_el1			// x22=cpuid
296	mov	x0, x22
297	bl	lookup_processor_type
298	mov	x23, x0				// x23=current cpu_table
299	cbz	x23, __error_p			// invalid processor (x23=0)?
300
301	pgtbl	x25, x26, x24			// x25=TTBR0, x26=TTBR1
302	ldr	x12, [x23, #CPU_INFO_SETUP]
303	add	x12, x12, x28			// __virt_to_phys
304	blr	x12				// initialise processor
305
306	ldr	x21, =secondary_data
307	ldr	x27, =__secondary_switched	// address to jump to after enabling the MMU
308	b	__enable_mmu
309ENDPROC(secondary_startup)
310
311ENTRY(__secondary_switched)
312	ldr	x0, [x21]			// get secondary_data.stack
313	mov	sp, x0
314	mov	x29, #0
315	b	secondary_start_kernel
316ENDPROC(__secondary_switched)
317#endif	/* CONFIG_SMP */
318
319/*
320 * Setup common bits before finally enabling the MMU. Essentially this is just
321 * loading the page table pointer and vector base registers.
322 *
323 * On entry to this code, x0 must contain the SCTLR_EL1 value for turning on
324 * the MMU.
325 */
326__enable_mmu:
327	ldr	x5, =vectors
328	msr	vbar_el1, x5
329	msr	ttbr0_el1, x25			// load TTBR0
330	msr	ttbr1_el1, x26			// load TTBR1
331	isb
332	b	__turn_mmu_on
333ENDPROC(__enable_mmu)
334
335/*
336 * Enable the MMU. This completely changes the structure of the visible memory
337 * space. You will not be able to trace execution through this.
338 *
339 *  x0  = system control register
340 *  x27 = *virtual* address to jump to upon completion
341 *
342 * other registers depend on the function called upon completion
343 */
344	.align	6
345__turn_mmu_on:
346	msr	sctlr_el1, x0
347	isb
348	br	x27
349ENDPROC(__turn_mmu_on)
350
351/*
352 * Calculate the start of physical memory.
353 */
354__calc_phys_offset:
355	adr	x0, 1f
356	ldp	x1, x2, [x0]
357	sub	x28, x0, x1			// x28 = PHYS_OFFSET - PAGE_OFFSET
358	add	x24, x2, x28			// x24 = PHYS_OFFSET
359	ret
360ENDPROC(__calc_phys_offset)
361
362	.align 3
3631:	.quad	.
364	.quad	PAGE_OFFSET
365
366/*
367 * Macro to populate the PGD for the corresponding block entry in the next
368 * level (tbl) for the given virtual address.
369 *
370 * Preserves:	pgd, tbl, virt
371 * Corrupts:	tmp1, tmp2
372 */
373	.macro	create_pgd_entry, pgd, tbl, virt, tmp1, tmp2
374	lsr	\tmp1, \virt, #PGDIR_SHIFT
375	and	\tmp1, \tmp1, #PTRS_PER_PGD - 1	// PGD index
376	orr	\tmp2, \tbl, #3			// PGD entry table type
377	str	\tmp2, [\pgd, \tmp1, lsl #3]
378	.endm
379
380/*
381 * Macro to populate block entries in the page table for the start..end
382 * virtual range (inclusive).
383 *
384 * Preserves:	tbl, flags
385 * Corrupts:	phys, start, end, pstate
386 */
387	.macro	create_block_map, tbl, flags, phys, start, end, idmap=0
388	lsr	\phys, \phys, #BLOCK_SHIFT
389	.if	\idmap
390	and	\start, \phys, #PTRS_PER_PTE - 1	// table index
391	.else
392	lsr	\start, \start, #BLOCK_SHIFT
393	and	\start, \start, #PTRS_PER_PTE - 1	// table index
394	.endif
395	orr	\phys, \flags, \phys, lsl #BLOCK_SHIFT	// table entry
396	.ifnc	\start,\end
397	lsr	\end, \end, #BLOCK_SHIFT
398	and	\end, \end, #PTRS_PER_PTE - 1		// table end index
399	.endif
4009999:	str	\phys, [\tbl, \start, lsl #3]		// store the entry
401	.ifnc	\start,\end
402	add	\start, \start, #1			// next entry
403	add	\phys, \phys, #BLOCK_SIZE		// next block
404	cmp	\start, \end
405	b.ls	9999b
406	.endif
407	.endm
408
409/*
410 * Setup the initial page tables. We only setup the barest amount which is
411 * required to get the kernel running. The following sections are required:
412 *   - identity mapping to enable the MMU (low address, TTBR0)
413 *   - first few MB of the kernel linear mapping to jump to once the MMU has
414 *     been enabled, including the FDT blob (TTBR1)
415 *   - UART mapping if CONFIG_EARLY_PRINTK is enabled (TTBR1)
416 */
417__create_page_tables:
418	pgtbl	x25, x26, x24			// idmap_pg_dir and swapper_pg_dir addresses
419
420	/*
421	 * Clear the idmap and swapper page tables.
422	 */
423	mov	x0, x25
424	add	x6, x26, #SWAPPER_DIR_SIZE
4251:	stp	xzr, xzr, [x0], #16
426	stp	xzr, xzr, [x0], #16
427	stp	xzr, xzr, [x0], #16
428	stp	xzr, xzr, [x0], #16
429	cmp	x0, x6
430	b.lo	1b
431
432	ldr	x7, =MM_MMUFLAGS
433
434	/*
435	 * Create the identity mapping.
436	 */
437	add	x0, x25, #PAGE_SIZE		// section table address
438	adr	x3, __turn_mmu_on		// virtual/physical address
439	create_pgd_entry x25, x0, x3, x5, x6
440	create_block_map x0, x7, x3, x5, x5, idmap=1
441
442	/*
443	 * Map the kernel image (starting with PHYS_OFFSET).
444	 */
445	add	x0, x26, #PAGE_SIZE		// section table address
446	mov	x5, #PAGE_OFFSET
447	create_pgd_entry x26, x0, x5, x3, x6
448	ldr	x6, =KERNEL_END - 1
449	mov	x3, x24				// phys offset
450	create_block_map x0, x7, x3, x5, x6
451
452	/*
453	 * Map the FDT blob (maximum 2MB; must be within 512MB of
454	 * PHYS_OFFSET).
455	 */
456	mov	x3, x21				// FDT phys address
457	and	x3, x3, #~((1 << 21) - 1)	// 2MB aligned
458	mov	x6, #PAGE_OFFSET
459	sub	x5, x3, x24			// subtract PHYS_OFFSET
460	tst	x5, #~((1 << 29) - 1)		// within 512MB?
461	csel	x21, xzr, x21, ne		// zero the FDT pointer
462	b.ne	1f
463	add	x5, x5, x6			// __va(FDT blob)
464	add	x6, x5, #1 << 21		// 2MB for the FDT blob
465	sub	x6, x6, #1			// inclusive range
466	create_block_map x0, x7, x3, x5, x6
4671:
468#ifdef CONFIG_EARLY_PRINTK
469	/*
470	 * Create the pgd entry for the UART mapping. The full mapping is done
471	 * later based earlyprintk kernel parameter.
472	 */
473	ldr	x5, =EARLYCON_IOBASE		// UART virtual address
474	add	x0, x26, #2 * PAGE_SIZE		// section table address
475	create_pgd_entry x26, x0, x5, x6, x7
476#endif
477	ret
478ENDPROC(__create_page_tables)
479	.ltorg
480
481	.align	3
482	.type	__switch_data, %object
483__switch_data:
484	.quad	__mmap_switched
485	.quad	__data_loc			// x4
486	.quad	_data				// x5
487	.quad	__bss_start			// x6
488	.quad	_end				// x7
489	.quad	processor_id			// x4
490	.quad	__fdt_pointer			// x5
491	.quad	memstart_addr			// x6
492	.quad	init_thread_union + THREAD_START_SP // sp
493
494/*
495 * The following fragment of code is executed with the MMU on in MMU mode, and
496 * uses absolute addresses; this is not position independent.
497 */
498__mmap_switched:
499	adr	x3, __switch_data + 8
500
501	ldp	x4, x5, [x3], #16
502	ldp	x6, x7, [x3], #16
503	cmp	x4, x5				// Copy data segment if needed
5041:	ccmp	x5, x6, #4, ne
505	b.eq	2f
506	ldr	x16, [x4], #8
507	str	x16, [x5], #8
508	b	1b
5092:
5101:	cmp	x6, x7
511	b.hs	2f
512	str	xzr, [x6], #8			// Clear BSS
513	b	1b
5142:
515	ldp	x4, x5, [x3], #16
516	ldr	x6, [x3], #8
517	ldr	x16, [x3]
518	mov	sp, x16
519	str	x22, [x4]			// Save processor ID
520	str	x21, [x5]			// Save FDT pointer
521	str	x24, [x6]			// Save PHYS_OFFSET
522	mov	x29, #0
523	b	start_kernel
524ENDPROC(__mmap_switched)
525
526/*
527 * Exception handling. Something went wrong and we can't proceed. We ought to
528 * tell the user, but since we don't have any guarantee that we're even
529 * running on the right architecture, we do virtually nothing.
530 */
531__error_p:
532ENDPROC(__error_p)
533
534__error:
5351:	nop
536	b	1b
537ENDPROC(__error)
538
539/*
540 * This function gets the processor ID in w0 and searches the cpu_table[] for
541 * a match. It returns a pointer to the struct cpu_info it found. The
542 * cpu_table[] must end with an empty (all zeros) structure.
543 *
544 * This routine can be called via C code and it needs to work with the MMU
545 * both disabled and enabled (the offset is calculated automatically).
546 */
547ENTRY(lookup_processor_type)
548	adr	x1, __lookup_processor_type_data
549	ldp	x2, x3, [x1]
550	sub	x1, x1, x2			// get offset between VA and PA
551	add	x3, x3, x1			// convert VA to PA
5521:
553	ldp	w5, w6, [x3]			// load cpu_id_val and cpu_id_mask
554	cbz	w5, 2f				// end of list?
555	and	w6, w6, w0
556	cmp	w5, w6
557	b.eq	3f
558	add	x3, x3, #CPU_INFO_SZ
559	b	1b
5602:
561	mov	x3, #0				// unknown processor
5623:
563	mov	x0, x3
564	ret
565ENDPROC(lookup_processor_type)
566
567	.align	3
568	.type	__lookup_processor_type_data, %object
569__lookup_processor_type_data:
570	.quad	.
571	.quad	cpu_table
572	.size	__lookup_processor_type_data, . - __lookup_processor_type_data
573
574/*
575 * Determine validity of the x21 FDT pointer.
576 * The dtb must be 8-byte aligned and live in the first 512M of memory.
577 */
578__vet_fdt:
579	tst	x21, #0x7
580	b.ne	1f
581	cmp	x21, x24
582	b.lt	1f
583	mov	x0, #(1 << 29)
584	add	x0, x0, x24
585	cmp	x21, x0
586	b.ge	1f
587	ret
5881:
589	mov	x21, #0
590	ret
591ENDPROC(__vet_fdt)
592