xref: /linux/arch/arm/kernel/head.S (revision 95e9fd10f06cb5642028b6b851e32b8c8afb4571)
1/*
2 *  linux/arch/arm/kernel/head.S
3 *
4 *  Copyright (C) 1994-2002 Russell King
5 *  Copyright (c) 2003 ARM Limited
6 *  All Rights Reserved
7 *
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
11 *
12 *  Kernel startup code for all 32-bit CPUs
13 */
14#include <linux/linkage.h>
15#include <linux/init.h>
16
17#include <asm/assembler.h>
18#include <asm/cp15.h>
19#include <asm/domain.h>
20#include <asm/ptrace.h>
21#include <asm/asm-offsets.h>
22#include <asm/memory.h>
23#include <asm/thread_info.h>
24#include <asm/pgtable.h>
25
26#ifdef CONFIG_DEBUG_LL
27#include <mach/debug-macro.S>
28#endif
29
30/*
31 * swapper_pg_dir is the virtual address of the initial page table.
32 * We place the page tables 16K below KERNEL_RAM_VADDR.  Therefore, we must
33 * make sure that KERNEL_RAM_VADDR is correctly set.  Currently, we expect
34 * the least significant 16 bits to be 0x8000, but we could probably
35 * relax this restriction to KERNEL_RAM_VADDR >= PAGE_OFFSET + 0x4000.
36 */
37#define KERNEL_RAM_VADDR	(PAGE_OFFSET + TEXT_OFFSET)
38#if (KERNEL_RAM_VADDR & 0xffff) != 0x8000
39#error KERNEL_RAM_VADDR must start at 0xXXXX8000
40#endif
41
42#ifdef CONFIG_ARM_LPAE
43	/* LPAE requires an additional page for the PGD */
44#define PG_DIR_SIZE	0x5000
45#define PMD_ORDER	3
46#else
47#define PG_DIR_SIZE	0x4000
48#define PMD_ORDER	2
49#endif
50
51	.globl	swapper_pg_dir
52	.equ	swapper_pg_dir, KERNEL_RAM_VADDR - PG_DIR_SIZE
53
54	.macro	pgtbl, rd, phys
55	add	\rd, \phys, #TEXT_OFFSET - PG_DIR_SIZE
56	.endm
57
58/*
59 * Kernel startup entry point.
60 * ---------------------------
61 *
62 * This is normally called from the decompressor code.  The requirements
63 * are: MMU = off, D-cache = off, I-cache = dont care, r0 = 0,
64 * r1 = machine nr, r2 = atags or dtb pointer.
65 *
66 * This code is mostly position independent, so if you link the kernel at
67 * 0xc0008000, you call this at __pa(0xc0008000).
68 *
69 * See linux/arch/arm/tools/mach-types for the complete list of machine
70 * numbers for r1.
71 *
72 * We're trying to keep crap to a minimum; DO NOT add any machine specific
73 * crap here - that's what the boot loader (or in extreme, well justified
74 * circumstances, zImage) is for.
75 */
76	.arm
77
78	__HEAD
79ENTRY(stext)
80
81 THUMB(	adr	r9, BSYM(1f)	)	@ Kernel is always entered in ARM.
82 THUMB(	bx	r9		)	@ If this is a Thumb-2 kernel,
83 THUMB(	.thumb			)	@ switch to Thumb now.
84 THUMB(1:			)
85
86	setmode	PSR_F_BIT | PSR_I_BIT | SVC_MODE, r9 @ ensure svc mode
87						@ and irqs disabled
88	mrc	p15, 0, r9, c0, c0		@ get processor id
89	bl	__lookup_processor_type		@ r5=procinfo r9=cpuid
90	movs	r10, r5				@ invalid processor (r5=0)?
91 THUMB( it	eq )		@ force fixup-able long branch encoding
92	beq	__error_p			@ yes, error 'p'
93
94#ifdef CONFIG_ARM_LPAE
95	mrc	p15, 0, r3, c0, c1, 4		@ read ID_MMFR0
96	and	r3, r3, #0xf			@ extract VMSA support
97	cmp	r3, #5				@ long-descriptor translation table format?
98 THUMB( it	lo )				@ force fixup-able long branch encoding
99	blo	__error_p			@ only classic page table format
100#endif
101
102#ifndef CONFIG_XIP_KERNEL
103	adr	r3, 2f
104	ldmia	r3, {r4, r8}
105	sub	r4, r3, r4			@ (PHYS_OFFSET - PAGE_OFFSET)
106	add	r8, r8, r4			@ PHYS_OFFSET
107#else
108	ldr	r8, =PHYS_OFFSET		@ always constant in this case
109#endif
110
111	/*
112	 * r1 = machine no, r2 = atags or dtb,
113	 * r8 = phys_offset, r9 = cpuid, r10 = procinfo
114	 */
115	bl	__vet_atags
116#ifdef CONFIG_SMP_ON_UP
117	bl	__fixup_smp
118#endif
119#ifdef CONFIG_ARM_PATCH_PHYS_VIRT
120	bl	__fixup_pv_table
121#endif
122	bl	__create_page_tables
123
124	/*
125	 * The following calls CPU specific code in a position independent
126	 * manner.  See arch/arm/mm/proc-*.S for details.  r10 = base of
127	 * xxx_proc_info structure selected by __lookup_processor_type
128	 * above.  On return, the CPU will be ready for the MMU to be
129	 * turned on, and r0 will hold the CPU control register value.
130	 */
131	ldr	r13, =__mmap_switched		@ address to jump to after
132						@ mmu has been enabled
133	adr	lr, BSYM(1f)			@ return (PIC) address
134	mov	r8, r4				@ set TTBR1 to swapper_pg_dir
135 ARM(	add	pc, r10, #PROCINFO_INITFUNC	)
136 THUMB(	add	r12, r10, #PROCINFO_INITFUNC	)
137 THUMB(	mov	pc, r12				)
1381:	b	__enable_mmu
139ENDPROC(stext)
140	.ltorg
141#ifndef CONFIG_XIP_KERNEL
1422:	.long	.
143	.long	PAGE_OFFSET
144#endif
145
146/*
147 * Setup the initial page tables.  We only setup the barest
148 * amount which are required to get the kernel running, which
149 * generally means mapping in the kernel code.
150 *
151 * r8 = phys_offset, r9 = cpuid, r10 = procinfo
152 *
153 * Returns:
154 *  r0, r3, r5-r7 corrupted
155 *  r4 = physical page table address
156 */
157__create_page_tables:
158	pgtbl	r4, r8				@ page table address
159
160	/*
161	 * Clear the swapper page table
162	 */
163	mov	r0, r4
164	mov	r3, #0
165	add	r6, r0, #PG_DIR_SIZE
1661:	str	r3, [r0], #4
167	str	r3, [r0], #4
168	str	r3, [r0], #4
169	str	r3, [r0], #4
170	teq	r0, r6
171	bne	1b
172
173#ifdef CONFIG_ARM_LPAE
174	/*
175	 * Build the PGD table (first level) to point to the PMD table. A PGD
176	 * entry is 64-bit wide.
177	 */
178	mov	r0, r4
179	add	r3, r4, #0x1000			@ first PMD table address
180	orr	r3, r3, #3			@ PGD block type
181	mov	r6, #4				@ PTRS_PER_PGD
182	mov	r7, #1 << (55 - 32)		@ L_PGD_SWAPPER
1831:	str	r3, [r0], #4			@ set bottom PGD entry bits
184	str	r7, [r0], #4			@ set top PGD entry bits
185	add	r3, r3, #0x1000			@ next PMD table
186	subs	r6, r6, #1
187	bne	1b
188
189	add	r4, r4, #0x1000			@ point to the PMD tables
190#endif
191
192	ldr	r7, [r10, #PROCINFO_MM_MMUFLAGS] @ mm_mmuflags
193
194	/*
195	 * Create identity mapping to cater for __enable_mmu.
196	 * This identity mapping will be removed by paging_init().
197	 */
198	adr	r0, __turn_mmu_on_loc
199	ldmia	r0, {r3, r5, r6}
200	sub	r0, r0, r3			@ virt->phys offset
201	add	r5, r5, r0			@ phys __turn_mmu_on
202	add	r6, r6, r0			@ phys __turn_mmu_on_end
203	mov	r5, r5, lsr #SECTION_SHIFT
204	mov	r6, r6, lsr #SECTION_SHIFT
205
2061:	orr	r3, r7, r5, lsl #SECTION_SHIFT	@ flags + kernel base
207	str	r3, [r4, r5, lsl #PMD_ORDER]	@ identity mapping
208	cmp	r5, r6
209	addlo	r5, r5, #1			@ next section
210	blo	1b
211
212	/*
213	 * Map our RAM from the start to the end of the kernel .bss section.
214	 */
215	add	r0, r4, #PAGE_OFFSET >> (SECTION_SHIFT - PMD_ORDER)
216	ldr	r6, =(_end - 1)
217	orr	r3, r8, r7
218	add	r6, r4, r6, lsr #(SECTION_SHIFT - PMD_ORDER)
2191:	str	r3, [r0], #1 << PMD_ORDER
220	add	r3, r3, #1 << SECTION_SHIFT
221	cmp	r0, r6
222	bls	1b
223
224#ifdef CONFIG_XIP_KERNEL
225	/*
226	 * Map the kernel image separately as it is not located in RAM.
227	 */
228#define XIP_START XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR)
229	mov	r3, pc
230	mov	r3, r3, lsr #SECTION_SHIFT
231	orr	r3, r7, r3, lsl #SECTION_SHIFT
232	add	r0, r4,  #(XIP_START & 0xff000000) >> (SECTION_SHIFT - PMD_ORDER)
233	str	r3, [r0, #((XIP_START & 0x00f00000) >> SECTION_SHIFT) << PMD_ORDER]!
234	ldr	r6, =(_edata_loc - 1)
235	add	r0, r0, #1 << PMD_ORDER
236	add	r6, r4, r6, lsr #(SECTION_SHIFT - PMD_ORDER)
2371:	cmp	r0, r6
238	add	r3, r3, #1 << SECTION_SHIFT
239	strls	r3, [r0], #1 << PMD_ORDER
240	bls	1b
241#endif
242
243	/*
244	 * Then map boot params address in r2 if specified.
245	 */
246	mov	r0, r2, lsr #SECTION_SHIFT
247	movs	r0, r0, lsl #SECTION_SHIFT
248	subne	r3, r0, r8
249	addne	r3, r3, #PAGE_OFFSET
250	addne	r3, r4, r3, lsr #(SECTION_SHIFT - PMD_ORDER)
251	orrne	r6, r7, r0
252	strne	r6, [r3]
253
254#ifdef CONFIG_DEBUG_LL
255#if !defined(CONFIG_DEBUG_ICEDCC) && !defined(CONFIG_DEBUG_SEMIHOSTING)
256	/*
257	 * Map in IO space for serial debugging.
258	 * This allows debug messages to be output
259	 * via a serial console before paging_init.
260	 */
261	addruart r7, r3, r0
262
263	mov	r3, r3, lsr #SECTION_SHIFT
264	mov	r3, r3, lsl #PMD_ORDER
265
266	add	r0, r4, r3
267	mov	r3, r7, lsr #SECTION_SHIFT
268	ldr	r7, [r10, #PROCINFO_IO_MMUFLAGS] @ io_mmuflags
269	orr	r3, r7, r3, lsl #SECTION_SHIFT
270#ifdef CONFIG_ARM_LPAE
271	mov	r7, #1 << (54 - 32)		@ XN
272#else
273	orr	r3, r3, #PMD_SECT_XN
274#endif
275	str	r3, [r0], #4
276#ifdef CONFIG_ARM_LPAE
277	str	r7, [r0], #4
278#endif
279
280#else /* CONFIG_DEBUG_ICEDCC || CONFIG_DEBUG_SEMIHOSTING */
281	/* we don't need any serial debugging mappings */
282	ldr	r7, [r10, #PROCINFO_IO_MMUFLAGS] @ io_mmuflags
283#endif
284
285#if defined(CONFIG_ARCH_NETWINDER) || defined(CONFIG_ARCH_CATS)
286	/*
287	 * If we're using the NetWinder or CATS, we also need to map
288	 * in the 16550-type serial port for the debug messages
289	 */
290	add	r0, r4, #0xff000000 >> (SECTION_SHIFT - PMD_ORDER)
291	orr	r3, r7, #0x7c000000
292	str	r3, [r0]
293#endif
294#ifdef CONFIG_ARCH_RPC
295	/*
296	 * Map in screen at 0x02000000 & SCREEN2_BASE
297	 * Similar reasons here - for debug.  This is
298	 * only for Acorn RiscPC architectures.
299	 */
300	add	r0, r4, #0x02000000 >> (SECTION_SHIFT - PMD_ORDER)
301	orr	r3, r7, #0x02000000
302	str	r3, [r0]
303	add	r0, r4, #0xd8000000 >> (SECTION_SHIFT - PMD_ORDER)
304	str	r3, [r0]
305#endif
306#endif
307#ifdef CONFIG_ARM_LPAE
308	sub	r4, r4, #0x1000		@ point to the PGD table
309#endif
310	mov	pc, lr
311ENDPROC(__create_page_tables)
312	.ltorg
313	.align
314__turn_mmu_on_loc:
315	.long	.
316	.long	__turn_mmu_on
317	.long	__turn_mmu_on_end
318
319#if defined(CONFIG_SMP)
320	__CPUINIT
321ENTRY(secondary_startup)
322	/*
323	 * Common entry point for secondary CPUs.
324	 *
325	 * Ensure that we're in SVC mode, and IRQs are disabled.  Lookup
326	 * the processor type - there is no need to check the machine type
327	 * as it has already been validated by the primary processor.
328	 */
329	setmode	PSR_F_BIT | PSR_I_BIT | SVC_MODE, r9
330	mrc	p15, 0, r9, c0, c0		@ get processor id
331	bl	__lookup_processor_type
332	movs	r10, r5				@ invalid processor?
333	moveq	r0, #'p'			@ yes, error 'p'
334 THUMB( it	eq )		@ force fixup-able long branch encoding
335	beq	__error_p
336
337	/*
338	 * Use the page tables supplied from  __cpu_up.
339	 */
340	adr	r4, __secondary_data
341	ldmia	r4, {r5, r7, r12}		@ address to jump to after
342	sub	lr, r4, r5			@ mmu has been enabled
343	ldr	r4, [r7, lr]			@ get secondary_data.pgdir
344	add	r7, r7, #4
345	ldr	r8, [r7, lr]			@ get secondary_data.swapper_pg_dir
346	adr	lr, BSYM(__enable_mmu)		@ return address
347	mov	r13, r12			@ __secondary_switched address
348 ARM(	add	pc, r10, #PROCINFO_INITFUNC	) @ initialise processor
349						  @ (return control reg)
350 THUMB(	add	r12, r10, #PROCINFO_INITFUNC	)
351 THUMB(	mov	pc, r12				)
352ENDPROC(secondary_startup)
353
354	/*
355	 * r6  = &secondary_data
356	 */
357ENTRY(__secondary_switched)
358	ldr	sp, [r7, #4]			@ get secondary_data.stack
359	mov	fp, #0
360	b	secondary_start_kernel
361ENDPROC(__secondary_switched)
362
363	.align
364
365	.type	__secondary_data, %object
366__secondary_data:
367	.long	.
368	.long	secondary_data
369	.long	__secondary_switched
370#endif /* defined(CONFIG_SMP) */
371
372
373
374/*
375 * Setup common bits before finally enabling the MMU.  Essentially
376 * this is just loading the page table pointer and domain access
377 * registers.
378 *
379 *  r0  = cp#15 control register
380 *  r1  = machine ID
381 *  r2  = atags or dtb pointer
382 *  r4  = page table pointer
383 *  r9  = processor ID
384 *  r13 = *virtual* address to jump to upon completion
385 */
386__enable_mmu:
387#if defined(CONFIG_ALIGNMENT_TRAP) && __LINUX_ARM_ARCH__ < 6
388	orr	r0, r0, #CR_A
389#else
390	bic	r0, r0, #CR_A
391#endif
392#ifdef CONFIG_CPU_DCACHE_DISABLE
393	bic	r0, r0, #CR_C
394#endif
395#ifdef CONFIG_CPU_BPREDICT_DISABLE
396	bic	r0, r0, #CR_Z
397#endif
398#ifdef CONFIG_CPU_ICACHE_DISABLE
399	bic	r0, r0, #CR_I
400#endif
401#ifdef CONFIG_ARM_LPAE
402	mov	r5, #0
403	mcrr	p15, 0, r4, r5, c2		@ load TTBR0
404#else
405	mov	r5, #(domain_val(DOMAIN_USER, DOMAIN_MANAGER) | \
406		      domain_val(DOMAIN_KERNEL, DOMAIN_MANAGER) | \
407		      domain_val(DOMAIN_TABLE, DOMAIN_MANAGER) | \
408		      domain_val(DOMAIN_IO, DOMAIN_CLIENT))
409	mcr	p15, 0, r5, c3, c0, 0		@ load domain access register
410	mcr	p15, 0, r4, c2, c0, 0		@ load page table pointer
411#endif
412	b	__turn_mmu_on
413ENDPROC(__enable_mmu)
414
415/*
416 * Enable the MMU.  This completely changes the structure of the visible
417 * memory space.  You will not be able to trace execution through this.
418 * If you have an enquiry about this, *please* check the linux-arm-kernel
419 * mailing list archives BEFORE sending another post to the list.
420 *
421 *  r0  = cp#15 control register
422 *  r1  = machine ID
423 *  r2  = atags or dtb pointer
424 *  r9  = processor ID
425 *  r13 = *virtual* address to jump to upon completion
426 *
427 * other registers depend on the function called upon completion
428 */
429	.align	5
430	.pushsection	.idmap.text, "ax"
431ENTRY(__turn_mmu_on)
432	mov	r0, r0
433	instr_sync
434	mcr	p15, 0, r0, c1, c0, 0		@ write control reg
435	mrc	p15, 0, r3, c0, c0, 0		@ read id reg
436	instr_sync
437	mov	r3, r3
438	mov	r3, r13
439	mov	pc, r3
440__turn_mmu_on_end:
441ENDPROC(__turn_mmu_on)
442	.popsection
443
444
445#ifdef CONFIG_SMP_ON_UP
446	__INIT
447__fixup_smp:
448	and	r3, r9, #0x000f0000	@ architecture version
449	teq	r3, #0x000f0000		@ CPU ID supported?
450	bne	__fixup_smp_on_up	@ no, assume UP
451
452	bic	r3, r9, #0x00ff0000
453	bic	r3, r3, #0x0000000f	@ mask 0xff00fff0
454	mov	r4, #0x41000000
455	orr	r4, r4, #0x0000b000
456	orr	r4, r4, #0x00000020	@ val 0x4100b020
457	teq	r3, r4			@ ARM 11MPCore?
458	moveq	pc, lr			@ yes, assume SMP
459
460	mrc	p15, 0, r0, c0, c0, 5	@ read MPIDR
461	and	r0, r0, #0xc0000000	@ multiprocessing extensions and
462	teq	r0, #0x80000000		@ not part of a uniprocessor system?
463	moveq	pc, lr			@ yes, assume SMP
464
465__fixup_smp_on_up:
466	adr	r0, 1f
467	ldmia	r0, {r3 - r5}
468	sub	r3, r0, r3
469	add	r4, r4, r3
470	add	r5, r5, r3
471	b	__do_fixup_smp_on_up
472ENDPROC(__fixup_smp)
473
474	.align
4751:	.word	.
476	.word	__smpalt_begin
477	.word	__smpalt_end
478
479	.pushsection .data
480	.globl	smp_on_up
481smp_on_up:
482	ALT_SMP(.long	1)
483	ALT_UP(.long	0)
484	.popsection
485#endif
486
487	.text
488__do_fixup_smp_on_up:
489	cmp	r4, r5
490	movhs	pc, lr
491	ldmia	r4!, {r0, r6}
492 ARM(	str	r6, [r0, r3]	)
493 THUMB(	add	r0, r0, r3	)
494#ifdef __ARMEB__
495 THUMB(	mov	r6, r6, ror #16	)	@ Convert word order for big-endian.
496#endif
497 THUMB(	strh	r6, [r0], #2	)	@ For Thumb-2, store as two halfwords
498 THUMB(	mov	r6, r6, lsr #16	)	@ to be robust against misaligned r3.
499 THUMB(	strh	r6, [r0]	)
500	b	__do_fixup_smp_on_up
501ENDPROC(__do_fixup_smp_on_up)
502
503ENTRY(fixup_smp)
504	stmfd	sp!, {r4 - r6, lr}
505	mov	r4, r0
506	add	r5, r0, r1
507	mov	r3, #0
508	bl	__do_fixup_smp_on_up
509	ldmfd	sp!, {r4 - r6, pc}
510ENDPROC(fixup_smp)
511
512#ifdef CONFIG_ARM_PATCH_PHYS_VIRT
513
514/* __fixup_pv_table - patch the stub instructions with the delta between
515 * PHYS_OFFSET and PAGE_OFFSET, which is assumed to be 16MiB aligned and
516 * can be expressed by an immediate shifter operand. The stub instruction
517 * has a form of '(add|sub) rd, rn, #imm'.
518 */
519	__HEAD
520__fixup_pv_table:
521	adr	r0, 1f
522	ldmia	r0, {r3-r5, r7}
523	sub	r3, r0, r3	@ PHYS_OFFSET - PAGE_OFFSET
524	add	r4, r4, r3	@ adjust table start address
525	add	r5, r5, r3	@ adjust table end address
526	add	r7, r7, r3	@ adjust __pv_phys_offset address
527	str	r8, [r7]	@ save computed PHYS_OFFSET to __pv_phys_offset
528	mov	r6, r3, lsr #24	@ constant for add/sub instructions
529	teq	r3, r6, lsl #24 @ must be 16MiB aligned
530THUMB(	it	ne		@ cross section branch )
531	bne	__error
532	str	r6, [r7, #4]	@ save to __pv_offset
533	b	__fixup_a_pv_table
534ENDPROC(__fixup_pv_table)
535
536	.align
5371:	.long	.
538	.long	__pv_table_begin
539	.long	__pv_table_end
5402:	.long	__pv_phys_offset
541
542	.text
543__fixup_a_pv_table:
544#ifdef CONFIG_THUMB2_KERNEL
545	lsls	r6, #24
546	beq	2f
547	clz	r7, r6
548	lsr	r6, #24
549	lsl	r6, r7
550	bic	r6, #0x0080
551	lsrs	r7, #1
552	orrcs	r6, #0x0080
553	orr	r6, r6, r7, lsl #12
554	orr	r6, #0x4000
555	b	2f
5561:	add     r7, r3
557	ldrh	ip, [r7, #2]
558	and	ip, 0x8f00
559	orr	ip, r6	@ mask in offset bits 31-24
560	strh	ip, [r7, #2]
5612:	cmp	r4, r5
562	ldrcc	r7, [r4], #4	@ use branch for delay slot
563	bcc	1b
564	bx	lr
565#else
566	b	2f
5671:	ldr	ip, [r7, r3]
568	bic	ip, ip, #0x000000ff
569	orr	ip, ip, r6	@ mask in offset bits 31-24
570	str	ip, [r7, r3]
5712:	cmp	r4, r5
572	ldrcc	r7, [r4], #4	@ use branch for delay slot
573	bcc	1b
574	mov	pc, lr
575#endif
576ENDPROC(__fixup_a_pv_table)
577
578ENTRY(fixup_pv_table)
579	stmfd	sp!, {r4 - r7, lr}
580	ldr	r2, 2f			@ get address of __pv_phys_offset
581	mov	r3, #0			@ no offset
582	mov	r4, r0			@ r0 = table start
583	add	r5, r0, r1		@ r1 = table size
584	ldr	r6, [r2, #4]		@ get __pv_offset
585	bl	__fixup_a_pv_table
586	ldmfd	sp!, {r4 - r7, pc}
587ENDPROC(fixup_pv_table)
588
589	.align
5902:	.long	__pv_phys_offset
591
592	.data
593	.globl	__pv_phys_offset
594	.type	__pv_phys_offset, %object
595__pv_phys_offset:
596	.long	0
597	.size	__pv_phys_offset, . - __pv_phys_offset
598__pv_offset:
599	.long	0
600#endif
601
602#include "head-common.S"
603