xref: /openbmc/linux/arch/arm64/kernel/head.S (revision 9d749629)
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/memory.h>
30#include <asm/thread_info.h>
31#include <asm/pgtable-hwdef.h>
32#include <asm/pgtable.h>
33#include <asm/page.h>
34#include <asm/virt.h>
35
36/*
37 * swapper_pg_dir is the virtual address of the initial page table. We place
38 * the page tables 3 * PAGE_SIZE below KERNEL_RAM_VADDR. The idmap_pg_dir has
39 * 2 pages and is placed below swapper_pg_dir.
40 */
41#define KERNEL_RAM_VADDR	(PAGE_OFFSET + TEXT_OFFSET)
42
43#if (KERNEL_RAM_VADDR & 0xfffff) != 0x80000
44#error KERNEL_RAM_VADDR must start at 0xXXX80000
45#endif
46
47#define SWAPPER_DIR_SIZE	(3 * PAGE_SIZE)
48#define IDMAP_DIR_SIZE		(2 * PAGE_SIZE)
49
50	.globl	swapper_pg_dir
51	.equ	swapper_pg_dir, KERNEL_RAM_VADDR - SWAPPER_DIR_SIZE
52
53	.globl	idmap_pg_dir
54	.equ	idmap_pg_dir, swapper_pg_dir - IDMAP_DIR_SIZE
55
56	.macro	pgtbl, ttb0, ttb1, phys
57	add	\ttb1, \phys, #TEXT_OFFSET - SWAPPER_DIR_SIZE
58	sub	\ttb0, \ttb1, #IDMAP_DIR_SIZE
59	.endm
60
61#ifdef CONFIG_ARM64_64K_PAGES
62#define BLOCK_SHIFT	PAGE_SHIFT
63#define BLOCK_SIZE	PAGE_SIZE
64#else
65#define BLOCK_SHIFT	SECTION_SHIFT
66#define BLOCK_SIZE	SECTION_SIZE
67#endif
68
69#define KERNEL_START	KERNEL_RAM_VADDR
70#define KERNEL_END	_end
71
72/*
73 * Initial memory map attributes.
74 */
75#ifndef CONFIG_SMP
76#define PTE_FLAGS	PTE_TYPE_PAGE | PTE_AF
77#define PMD_FLAGS	PMD_TYPE_SECT | PMD_SECT_AF
78#else
79#define PTE_FLAGS	PTE_TYPE_PAGE | PTE_AF | PTE_SHARED
80#define PMD_FLAGS	PMD_TYPE_SECT | PMD_SECT_AF | PMD_SECT_S
81#endif
82
83#ifdef CONFIG_ARM64_64K_PAGES
84#define MM_MMUFLAGS	PTE_ATTRINDX(MT_NORMAL) | PTE_FLAGS
85#else
86#define MM_MMUFLAGS	PMD_ATTRINDX(MT_NORMAL) | PMD_FLAGS
87#endif
88
89/*
90 * Kernel startup entry point.
91 * ---------------------------
92 *
93 * The requirements are:
94 *   MMU = off, D-cache = off, I-cache = on or off,
95 *   x0 = physical address to the FDT blob.
96 *
97 * This code is mostly position independent so you call this at
98 * __pa(PAGE_OFFSET + TEXT_OFFSET).
99 *
100 * Note that the callee-saved registers are used for storing variables
101 * that are useful before the MMU is enabled. The allocations are described
102 * in the entry routines.
103 */
104	__HEAD
105
106	/*
107	 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
108	 */
109	b	stext				// branch to kernel start, magic
110	.long	0				// reserved
111	.quad	TEXT_OFFSET			// Image load offset from start of RAM
112	.quad	0				// reserved
113	.quad	0				// reserved
114
115ENTRY(stext)
116	mov	x21, x0				// x21=FDT
117	bl	__calc_phys_offset		// x24=PHYS_OFFSET, x28=PHYS_OFFSET-PAGE_OFFSET
118	bl	el2_setup			// Drop to EL1
119	mrs	x22, midr_el1			// x22=cpuid
120	mov	x0, x22
121	bl	lookup_processor_type
122	mov	x23, x0				// x23=current cpu_table
123	cbz	x23, __error_p			// invalid processor (x23=0)?
124	bl	__vet_fdt
125	bl	__create_page_tables		// x25=TTBR0, x26=TTBR1
126	/*
127	 * The following calls CPU specific code in a position independent
128	 * manner. See arch/arm64/mm/proc.S for details. x23 = base of
129	 * cpu_info structure selected by lookup_processor_type above.
130	 * On return, the CPU will be ready for the MMU to be turned on and
131	 * the TCR will have been set.
132	 */
133	ldr	x27, __switch_data		// address to jump to after
134						// MMU has been enabled
135	adr	lr, __enable_mmu		// return (PIC) address
136	ldr	x12, [x23, #CPU_INFO_SETUP]
137	add	x12, x12, x28			// __virt_to_phys
138	br	x12				// initialise processor
139ENDPROC(stext)
140
141/*
142 * If we're fortunate enough to boot at EL2, ensure that the world is
143 * sane before dropping to EL1.
144 */
145ENTRY(el2_setup)
146	mrs	x0, CurrentEL
147	cmp	x0, #PSR_MODE_EL2t
148	ccmp	x0, #PSR_MODE_EL2h, #0x4, ne
149	ldr	x0, =__boot_cpu_mode		// Compute __boot_cpu_mode
150	add	x0, x0, x28
151	b.eq	1f
152	str	wzr, [x0]			// Remember we don't have EL2...
153	ret
154
155	/* Hyp configuration. */
1561:	ldr	w1, =BOOT_CPU_MODE_EL2
157	str	w1, [x0, #4]			// This CPU has EL2
158	mov	x0, #(1 << 31)			// 64-bit EL1
159	msr	hcr_el2, x0
160
161	/* Generic timers. */
162	mrs	x0, cnthctl_el2
163	orr	x0, x0, #3			// Enable EL1 physical timers
164	msr	cnthctl_el2, x0
165	msr	cntvoff_el2, xzr		// Clear virtual offset
166
167	/* Populate ID registers. */
168	mrs	x0, midr_el1
169	mrs	x1, mpidr_el1
170	msr	vpidr_el2, x0
171	msr	vmpidr_el2, x1
172
173	/* sctlr_el1 */
174	mov	x0, #0x0800			// Set/clear RES{1,0} bits
175	movk	x0, #0x30d0, lsl #16
176	msr	sctlr_el1, x0
177
178	/* Coprocessor traps. */
179	mov	x0, #0x33ff
180	msr	cptr_el2, x0			// Disable copro. traps to EL2
181
182#ifdef CONFIG_COMPAT
183	msr	hstr_el2, xzr			// Disable CP15 traps to EL2
184#endif
185
186	/* Stage-2 translation */
187	msr	vttbr_el2, xzr
188
189	/* Hypervisor stub */
190	adr	x0, __hyp_stub_vectors
191	msr	vbar_el2, x0
192
193	/* spsr */
194	mov	x0, #(PSR_F_BIT | PSR_I_BIT | PSR_A_BIT | PSR_D_BIT |\
195		      PSR_MODE_EL1h)
196	msr	spsr_el2, x0
197	msr	elr_el2, lr
198	eret
199ENDPROC(el2_setup)
200
201/*
202 * We need to find out the CPU boot mode long after boot, so we need to
203 * store it in a writable variable.
204 *
205 * This is not in .bss, because we set it sufficiently early that the boot-time
206 * zeroing of .bss would clobber it.
207 */
208	.pushsection	.data
209ENTRY(__boot_cpu_mode)
210	.long	BOOT_CPU_MODE_EL2
211	.long	0
212	.popsection
213
214	.align	3
2152:	.quad	.
216	.quad	PAGE_OFFSET
217
218#ifdef CONFIG_SMP
219	.pushsection    .smp.pen.text, "ax"
220	.align	3
2211:	.quad	.
222	.quad	secondary_holding_pen_release
223
224	/*
225	 * This provides a "holding pen" for platforms to hold all secondary
226	 * cores are held until we're ready for them to initialise.
227	 */
228ENTRY(secondary_holding_pen)
229	bl	__calc_phys_offset		// x24=phys offset
230	bl	el2_setup			// Drop to EL1
231	mrs	x0, mpidr_el1
232	and	x0, x0, #15			// CPU number
233	adr	x1, 1b
234	ldp	x2, x3, [x1]
235	sub	x1, x1, x2
236	add	x3, x3, x1
237pen:	ldr	x4, [x3]
238	cmp	x4, x0
239	b.eq	secondary_startup
240	wfe
241	b	pen
242ENDPROC(secondary_holding_pen)
243	.popsection
244
245ENTRY(secondary_startup)
246	/*
247	 * Common entry point for secondary CPUs.
248	 */
249	mrs	x22, midr_el1			// x22=cpuid
250	mov	x0, x22
251	bl	lookup_processor_type
252	mov	x23, x0				// x23=current cpu_table
253	cbz	x23, __error_p			// invalid processor (x23=0)?
254
255	pgtbl	x25, x26, x24			// x25=TTBR0, x26=TTBR1
256	ldr	x12, [x23, #CPU_INFO_SETUP]
257	add	x12, x12, x28			// __virt_to_phys
258	blr	x12				// initialise processor
259
260	ldr	x21, =secondary_data
261	ldr	x27, =__secondary_switched	// address to jump to after enabling the MMU
262	b	__enable_mmu
263ENDPROC(secondary_startup)
264
265ENTRY(__secondary_switched)
266	ldr	x0, [x21]			// get secondary_data.stack
267	mov	sp, x0
268	mov	x29, #0
269	b	secondary_start_kernel
270ENDPROC(__secondary_switched)
271#endif	/* CONFIG_SMP */
272
273/*
274 * Setup common bits before finally enabling the MMU. Essentially this is just
275 * loading the page table pointer and vector base registers.
276 *
277 * On entry to this code, x0 must contain the SCTLR_EL1 value for turning on
278 * the MMU.
279 */
280__enable_mmu:
281	ldr	x5, =vectors
282	msr	vbar_el1, x5
283	msr	ttbr0_el1, x25			// load TTBR0
284	msr	ttbr1_el1, x26			// load TTBR1
285	isb
286	b	__turn_mmu_on
287ENDPROC(__enable_mmu)
288
289/*
290 * Enable the MMU. This completely changes the structure of the visible memory
291 * space. You will not be able to trace execution through this.
292 *
293 *  x0  = system control register
294 *  x27 = *virtual* address to jump to upon completion
295 *
296 * other registers depend on the function called upon completion
297 */
298	.align	6
299__turn_mmu_on:
300	msr	sctlr_el1, x0
301	isb
302	br	x27
303ENDPROC(__turn_mmu_on)
304
305/*
306 * Calculate the start of physical memory.
307 */
308__calc_phys_offset:
309	adr	x0, 1f
310	ldp	x1, x2, [x0]
311	sub	x28, x0, x1			// x28 = PHYS_OFFSET - PAGE_OFFSET
312	add	x24, x2, x28			// x24 = PHYS_OFFSET
313	ret
314ENDPROC(__calc_phys_offset)
315
316	.align 3
3171:	.quad	.
318	.quad	PAGE_OFFSET
319
320/*
321 * Macro to populate the PGD for the corresponding block entry in the next
322 * level (tbl) for the given virtual address.
323 *
324 * Preserves:	pgd, tbl, virt
325 * Corrupts:	tmp1, tmp2
326 */
327	.macro	create_pgd_entry, pgd, tbl, virt, tmp1, tmp2
328	lsr	\tmp1, \virt, #PGDIR_SHIFT
329	and	\tmp1, \tmp1, #PTRS_PER_PGD - 1	// PGD index
330	orr	\tmp2, \tbl, #3			// PGD entry table type
331	str	\tmp2, [\pgd, \tmp1, lsl #3]
332	.endm
333
334/*
335 * Macro to populate block entries in the page table for the start..end
336 * virtual range (inclusive).
337 *
338 * Preserves:	tbl, flags
339 * Corrupts:	phys, start, end, pstate
340 */
341	.macro	create_block_map, tbl, flags, phys, start, end, idmap=0
342	lsr	\phys, \phys, #BLOCK_SHIFT
343	.if	\idmap
344	and	\start, \phys, #PTRS_PER_PTE - 1	// table index
345	.else
346	lsr	\start, \start, #BLOCK_SHIFT
347	and	\start, \start, #PTRS_PER_PTE - 1	// table index
348	.endif
349	orr	\phys, \flags, \phys, lsl #BLOCK_SHIFT	// table entry
350	.ifnc	\start,\end
351	lsr	\end, \end, #BLOCK_SHIFT
352	and	\end, \end, #PTRS_PER_PTE - 1		// table end index
353	.endif
3549999:	str	\phys, [\tbl, \start, lsl #3]		// store the entry
355	.ifnc	\start,\end
356	add	\start, \start, #1			// next entry
357	add	\phys, \phys, #BLOCK_SIZE		// next block
358	cmp	\start, \end
359	b.ls	9999b
360	.endif
361	.endm
362
363/*
364 * Setup the initial page tables. We only setup the barest amount which is
365 * required to get the kernel running. The following sections are required:
366 *   - identity mapping to enable the MMU (low address, TTBR0)
367 *   - first few MB of the kernel linear mapping to jump to once the MMU has
368 *     been enabled, including the FDT blob (TTBR1)
369 *   - UART mapping if CONFIG_EARLY_PRINTK is enabled (TTBR1)
370 */
371__create_page_tables:
372	pgtbl	x25, x26, x24			// idmap_pg_dir and swapper_pg_dir addresses
373
374	/*
375	 * Clear the idmap and swapper page tables.
376	 */
377	mov	x0, x25
378	add	x6, x26, #SWAPPER_DIR_SIZE
3791:	stp	xzr, xzr, [x0], #16
380	stp	xzr, xzr, [x0], #16
381	stp	xzr, xzr, [x0], #16
382	stp	xzr, xzr, [x0], #16
383	cmp	x0, x6
384	b.lo	1b
385
386	ldr	x7, =MM_MMUFLAGS
387
388	/*
389	 * Create the identity mapping.
390	 */
391	add	x0, x25, #PAGE_SIZE		// section table address
392	adr	x3, __turn_mmu_on		// virtual/physical address
393	create_pgd_entry x25, x0, x3, x5, x6
394	create_block_map x0, x7, x3, x5, x5, idmap=1
395
396	/*
397	 * Map the kernel image (starting with PHYS_OFFSET).
398	 */
399	add	x0, x26, #PAGE_SIZE		// section table address
400	mov	x5, #PAGE_OFFSET
401	create_pgd_entry x26, x0, x5, x3, x6
402	ldr	x6, =KERNEL_END - 1
403	mov	x3, x24				// phys offset
404	create_block_map x0, x7, x3, x5, x6
405
406	/*
407	 * Map the FDT blob (maximum 2MB; must be within 512MB of
408	 * PHYS_OFFSET).
409	 */
410	mov	x3, x21				// FDT phys address
411	and	x3, x3, #~((1 << 21) - 1)	// 2MB aligned
412	mov	x6, #PAGE_OFFSET
413	sub	x5, x3, x24			// subtract PHYS_OFFSET
414	tst	x5, #~((1 << 29) - 1)		// within 512MB?
415	csel	x21, xzr, x21, ne		// zero the FDT pointer
416	b.ne	1f
417	add	x5, x5, x6			// __va(FDT blob)
418	add	x6, x5, #1 << 21		// 2MB for the FDT blob
419	sub	x6, x6, #1			// inclusive range
420	create_block_map x0, x7, x3, x5, x6
4211:
422#ifdef CONFIG_EARLY_PRINTK
423	/*
424	 * Create the pgd entry for the UART mapping. The full mapping is done
425	 * later based earlyprintk kernel parameter.
426	 */
427	ldr	x5, =EARLYCON_IOBASE		// UART virtual address
428	add	x0, x26, #2 * PAGE_SIZE		// section table address
429	create_pgd_entry x26, x0, x5, x6, x7
430#endif
431	ret
432ENDPROC(__create_page_tables)
433	.ltorg
434
435	.align	3
436	.type	__switch_data, %object
437__switch_data:
438	.quad	__mmap_switched
439	.quad	__data_loc			// x4
440	.quad	_data				// x5
441	.quad	__bss_start			// x6
442	.quad	_end				// x7
443	.quad	processor_id			// x4
444	.quad	__fdt_pointer			// x5
445	.quad	memstart_addr			// x6
446	.quad	init_thread_union + THREAD_START_SP // sp
447
448/*
449 * The following fragment of code is executed with the MMU on in MMU mode, and
450 * uses absolute addresses; this is not position independent.
451 */
452__mmap_switched:
453	adr	x3, __switch_data + 8
454
455	ldp	x4, x5, [x3], #16
456	ldp	x6, x7, [x3], #16
457	cmp	x4, x5				// Copy data segment if needed
4581:	ccmp	x5, x6, #4, ne
459	b.eq	2f
460	ldr	x16, [x4], #8
461	str	x16, [x5], #8
462	b	1b
4632:
4641:	cmp	x6, x7
465	b.hs	2f
466	str	xzr, [x6], #8			// Clear BSS
467	b	1b
4682:
469	ldp	x4, x5, [x3], #16
470	ldr	x6, [x3], #8
471	ldr	x16, [x3]
472	mov	sp, x16
473	str	x22, [x4]			// Save processor ID
474	str	x21, [x5]			// Save FDT pointer
475	str	x24, [x6]			// Save PHYS_OFFSET
476	mov	x29, #0
477	b	start_kernel
478ENDPROC(__mmap_switched)
479
480/*
481 * Exception handling. Something went wrong and we can't proceed. We ought to
482 * tell the user, but since we don't have any guarantee that we're even
483 * running on the right architecture, we do virtually nothing.
484 */
485__error_p:
486ENDPROC(__error_p)
487
488__error:
4891:	nop
490	b	1b
491ENDPROC(__error)
492
493/*
494 * This function gets the processor ID in w0 and searches the cpu_table[] for
495 * a match. It returns a pointer to the struct cpu_info it found. The
496 * cpu_table[] must end with an empty (all zeros) structure.
497 *
498 * This routine can be called via C code and it needs to work with the MMU
499 * both disabled and enabled (the offset is calculated automatically).
500 */
501ENTRY(lookup_processor_type)
502	adr	x1, __lookup_processor_type_data
503	ldp	x2, x3, [x1]
504	sub	x1, x1, x2			// get offset between VA and PA
505	add	x3, x3, x1			// convert VA to PA
5061:
507	ldp	w5, w6, [x3]			// load cpu_id_val and cpu_id_mask
508	cbz	w5, 2f				// end of list?
509	and	w6, w6, w0
510	cmp	w5, w6
511	b.eq	3f
512	add	x3, x3, #CPU_INFO_SZ
513	b	1b
5142:
515	mov	x3, #0				// unknown processor
5163:
517	mov	x0, x3
518	ret
519ENDPROC(lookup_processor_type)
520
521	.align	3
522	.type	__lookup_processor_type_data, %object
523__lookup_processor_type_data:
524	.quad	.
525	.quad	cpu_table
526	.size	__lookup_processor_type_data, . - __lookup_processor_type_data
527
528/*
529 * Determine validity of the x21 FDT pointer.
530 * The dtb must be 8-byte aligned and live in the first 512M of memory.
531 */
532__vet_fdt:
533	tst	x21, #0x7
534	b.ne	1f
535	cmp	x21, x24
536	b.lt	1f
537	mov	x0, #(1 << 29)
538	add	x0, x0, x24
539	cmp	x21, x0
540	b.ge	1f
541	ret
5421:
543	mov	x21, #0
544	ret
545ENDPROC(__vet_fdt)
546