xref: /linux/arch/arm/kernel/head.S (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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#if defined(CONFIG_DEBUG_LL) && !defined(CONFIG_DEBUG_SEMIHOSTING)
27#include CONFIG_DEBUG_LL_INCLUDE
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#ifdef CONFIG_ARM_VIRT_EXT
87	bl	__hyp_stub_install
88#endif
89	@ ensure svc mode and all interrupts masked
90	safe_svcmode_maskall r9
91
92	mrc	p15, 0, r9, c0, c0		@ get processor id
93	bl	__lookup_processor_type		@ r5=procinfo r9=cpuid
94	movs	r10, r5				@ invalid processor (r5=0)?
95 THUMB( it	eq )		@ force fixup-able long branch encoding
96	beq	__error_p			@ yes, error 'p'
97
98#ifdef CONFIG_ARM_LPAE
99	mrc	p15, 0, r3, c0, c1, 4		@ read ID_MMFR0
100	and	r3, r3, #0xf			@ extract VMSA support
101	cmp	r3, #5				@ long-descriptor translation table format?
102 THUMB( it	lo )				@ force fixup-able long branch encoding
103	blo	__error_p			@ only classic page table format
104#endif
105
106#ifndef CONFIG_XIP_KERNEL
107	adr	r3, 2f
108	ldmia	r3, {r4, r8}
109	sub	r4, r3, r4			@ (PHYS_OFFSET - PAGE_OFFSET)
110	add	r8, r8, r4			@ PHYS_OFFSET
111#else
112	ldr	r8, =PHYS_OFFSET		@ always constant in this case
113#endif
114
115	/*
116	 * r1 = machine no, r2 = atags or dtb,
117	 * r8 = phys_offset, r9 = cpuid, r10 = procinfo
118	 */
119	bl	__vet_atags
120#ifdef CONFIG_SMP_ON_UP
121	bl	__fixup_smp
122#endif
123#ifdef CONFIG_ARM_PATCH_PHYS_VIRT
124	bl	__fixup_pv_table
125#endif
126	bl	__create_page_tables
127
128	/*
129	 * The following calls CPU specific code in a position independent
130	 * manner.  See arch/arm/mm/proc-*.S for details.  r10 = base of
131	 * xxx_proc_info structure selected by __lookup_processor_type
132	 * above.  On return, the CPU will be ready for the MMU to be
133	 * turned on, and r0 will hold the CPU control register value.
134	 */
135	ldr	r13, =__mmap_switched		@ address to jump to after
136						@ mmu has been enabled
137	adr	lr, BSYM(1f)			@ return (PIC) address
138	mov	r8, r4				@ set TTBR1 to swapper_pg_dir
139 ARM(	add	pc, r10, #PROCINFO_INITFUNC	)
140 THUMB(	add	r12, r10, #PROCINFO_INITFUNC	)
141 THUMB(	mov	pc, r12				)
1421:	b	__enable_mmu
143ENDPROC(stext)
144	.ltorg
145#ifndef CONFIG_XIP_KERNEL
1462:	.long	.
147	.long	PAGE_OFFSET
148#endif
149
150/*
151 * Setup the initial page tables.  We only setup the barest
152 * amount which are required to get the kernel running, which
153 * generally means mapping in the kernel code.
154 *
155 * r8 = phys_offset, r9 = cpuid, r10 = procinfo
156 *
157 * Returns:
158 *  r0, r3, r5-r7 corrupted
159 *  r4 = physical page table address
160 */
161__create_page_tables:
162	pgtbl	r4, r8				@ page table address
163
164	/*
165	 * Clear the swapper page table
166	 */
167	mov	r0, r4
168	mov	r3, #0
169	add	r6, r0, #PG_DIR_SIZE
1701:	str	r3, [r0], #4
171	str	r3, [r0], #4
172	str	r3, [r0], #4
173	str	r3, [r0], #4
174	teq	r0, r6
175	bne	1b
176
177#ifdef CONFIG_ARM_LPAE
178	/*
179	 * Build the PGD table (first level) to point to the PMD table. A PGD
180	 * entry is 64-bit wide.
181	 */
182	mov	r0, r4
183	add	r3, r4, #0x1000			@ first PMD table address
184	orr	r3, r3, #3			@ PGD block type
185	mov	r6, #4				@ PTRS_PER_PGD
186	mov	r7, #1 << (55 - 32)		@ L_PGD_SWAPPER
1871:	str	r3, [r0], #4			@ set bottom PGD entry bits
188	str	r7, [r0], #4			@ set top PGD entry bits
189	add	r3, r3, #0x1000			@ next PMD table
190	subs	r6, r6, #1
191	bne	1b
192
193	add	r4, r4, #0x1000			@ point to the PMD tables
194#endif
195
196	ldr	r7, [r10, #PROCINFO_MM_MMUFLAGS] @ mm_mmuflags
197
198	/*
199	 * Create identity mapping to cater for __enable_mmu.
200	 * This identity mapping will be removed by paging_init().
201	 */
202	adr	r0, __turn_mmu_on_loc
203	ldmia	r0, {r3, r5, r6}
204	sub	r0, r0, r3			@ virt->phys offset
205	add	r5, r5, r0			@ phys __turn_mmu_on
206	add	r6, r6, r0			@ phys __turn_mmu_on_end
207	mov	r5, r5, lsr #SECTION_SHIFT
208	mov	r6, r6, lsr #SECTION_SHIFT
209
2101:	orr	r3, r7, r5, lsl #SECTION_SHIFT	@ flags + kernel base
211	str	r3, [r4, r5, lsl #PMD_ORDER]	@ identity mapping
212	cmp	r5, r6
213	addlo	r5, r5, #1			@ next section
214	blo	1b
215
216	/*
217	 * Map our RAM from the start to the end of the kernel .bss section.
218	 */
219	add	r0, r4, #PAGE_OFFSET >> (SECTION_SHIFT - PMD_ORDER)
220	ldr	r6, =(_end - 1)
221	orr	r3, r8, r7
222	add	r6, r4, r6, lsr #(SECTION_SHIFT - PMD_ORDER)
2231:	str	r3, [r0], #1 << PMD_ORDER
224	add	r3, r3, #1 << SECTION_SHIFT
225	cmp	r0, r6
226	bls	1b
227
228#ifdef CONFIG_XIP_KERNEL
229	/*
230	 * Map the kernel image separately as it is not located in RAM.
231	 */
232#define XIP_START XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR)
233	mov	r3, pc
234	mov	r3, r3, lsr #SECTION_SHIFT
235	orr	r3, r7, r3, lsl #SECTION_SHIFT
236	add	r0, r4,  #(XIP_START & 0xff000000) >> (SECTION_SHIFT - PMD_ORDER)
237	str	r3, [r0, #((XIP_START & 0x00f00000) >> SECTION_SHIFT) << PMD_ORDER]!
238	ldr	r6, =(_edata_loc - 1)
239	add	r0, r0, #1 << PMD_ORDER
240	add	r6, r4, r6, lsr #(SECTION_SHIFT - PMD_ORDER)
2411:	cmp	r0, r6
242	add	r3, r3, #1 << SECTION_SHIFT
243	strls	r3, [r0], #1 << PMD_ORDER
244	bls	1b
245#endif
246
247	/*
248	 * Then map boot params address in r2 if specified.
249	 */
250	mov	r0, r2, lsr #SECTION_SHIFT
251	movs	r0, r0, lsl #SECTION_SHIFT
252	subne	r3, r0, r8
253	addne	r3, r3, #PAGE_OFFSET
254	addne	r3, r4, r3, lsr #(SECTION_SHIFT - PMD_ORDER)
255	orrne	r6, r7, r0
256	strne	r6, [r3]
257
258#ifdef CONFIG_DEBUG_LL
259#if !defined(CONFIG_DEBUG_ICEDCC) && !defined(CONFIG_DEBUG_SEMIHOSTING)
260	/*
261	 * Map in IO space for serial debugging.
262	 * This allows debug messages to be output
263	 * via a serial console before paging_init.
264	 */
265	addruart r7, r3, r0
266
267	mov	r3, r3, lsr #SECTION_SHIFT
268	mov	r3, r3, lsl #PMD_ORDER
269
270	add	r0, r4, r3
271	mov	r3, r7, lsr #SECTION_SHIFT
272	ldr	r7, [r10, #PROCINFO_IO_MMUFLAGS] @ io_mmuflags
273	orr	r3, r7, r3, lsl #SECTION_SHIFT
274#ifdef CONFIG_ARM_LPAE
275	mov	r7, #1 << (54 - 32)		@ XN
276#else
277	orr	r3, r3, #PMD_SECT_XN
278#endif
279	str	r3, [r0], #4
280#ifdef CONFIG_ARM_LPAE
281	str	r7, [r0], #4
282#endif
283
284#else /* CONFIG_DEBUG_ICEDCC || CONFIG_DEBUG_SEMIHOSTING */
285	/* we don't need any serial debugging mappings */
286	ldr	r7, [r10, #PROCINFO_IO_MMUFLAGS] @ io_mmuflags
287#endif
288
289#if defined(CONFIG_ARCH_NETWINDER) || defined(CONFIG_ARCH_CATS)
290	/*
291	 * If we're using the NetWinder or CATS, we also need to map
292	 * in the 16550-type serial port for the debug messages
293	 */
294	add	r0, r4, #0xff000000 >> (SECTION_SHIFT - PMD_ORDER)
295	orr	r3, r7, #0x7c000000
296	str	r3, [r0]
297#endif
298#ifdef CONFIG_ARCH_RPC
299	/*
300	 * Map in screen at 0x02000000 & SCREEN2_BASE
301	 * Similar reasons here - for debug.  This is
302	 * only for Acorn RiscPC architectures.
303	 */
304	add	r0, r4, #0x02000000 >> (SECTION_SHIFT - PMD_ORDER)
305	orr	r3, r7, #0x02000000
306	str	r3, [r0]
307	add	r0, r4, #0xd8000000 >> (SECTION_SHIFT - PMD_ORDER)
308	str	r3, [r0]
309#endif
310#endif
311#ifdef CONFIG_ARM_LPAE
312	sub	r4, r4, #0x1000		@ point to the PGD table
313#endif
314	mov	pc, lr
315ENDPROC(__create_page_tables)
316	.ltorg
317	.align
318__turn_mmu_on_loc:
319	.long	.
320	.long	__turn_mmu_on
321	.long	__turn_mmu_on_end
322
323#if defined(CONFIG_SMP)
324	__CPUINIT
325ENTRY(secondary_startup)
326	/*
327	 * Common entry point for secondary CPUs.
328	 *
329	 * Ensure that we're in SVC mode, and IRQs are disabled.  Lookup
330	 * the processor type - there is no need to check the machine type
331	 * as it has already been validated by the primary processor.
332	 */
333#ifdef CONFIG_ARM_VIRT_EXT
334	bl	__hyp_stub_install
335#endif
336	safe_svcmode_maskall r9
337
338	mrc	p15, 0, r9, c0, c0		@ get processor id
339	bl	__lookup_processor_type
340	movs	r10, r5				@ invalid processor?
341	moveq	r0, #'p'			@ yes, error 'p'
342 THUMB( it	eq )		@ force fixup-able long branch encoding
343	beq	__error_p
344
345	/*
346	 * Use the page tables supplied from  __cpu_up.
347	 */
348	adr	r4, __secondary_data
349	ldmia	r4, {r5, r7, r12}		@ address to jump to after
350	sub	lr, r4, r5			@ mmu has been enabled
351	ldr	r4, [r7, lr]			@ get secondary_data.pgdir
352	add	r7, r7, #4
353	ldr	r8, [r7, lr]			@ get secondary_data.swapper_pg_dir
354	adr	lr, BSYM(__enable_mmu)		@ return address
355	mov	r13, r12			@ __secondary_switched address
356 ARM(	add	pc, r10, #PROCINFO_INITFUNC	) @ initialise processor
357						  @ (return control reg)
358 THUMB(	add	r12, r10, #PROCINFO_INITFUNC	)
359 THUMB(	mov	pc, r12				)
360ENDPROC(secondary_startup)
361
362	/*
363	 * r6  = &secondary_data
364	 */
365ENTRY(__secondary_switched)
366	ldr	sp, [r7, #4]			@ get secondary_data.stack
367	mov	fp, #0
368	b	secondary_start_kernel
369ENDPROC(__secondary_switched)
370
371	.align
372
373	.type	__secondary_data, %object
374__secondary_data:
375	.long	.
376	.long	secondary_data
377	.long	__secondary_switched
378#endif /* defined(CONFIG_SMP) */
379
380
381
382/*
383 * Setup common bits before finally enabling the MMU.  Essentially
384 * this is just loading the page table pointer and domain access
385 * registers.
386 *
387 *  r0  = cp#15 control register
388 *  r1  = machine ID
389 *  r2  = atags or dtb pointer
390 *  r4  = page table pointer
391 *  r9  = processor ID
392 *  r13 = *virtual* address to jump to upon completion
393 */
394__enable_mmu:
395#if defined(CONFIG_ALIGNMENT_TRAP) && __LINUX_ARM_ARCH__ < 6
396	orr	r0, r0, #CR_A
397#else
398	bic	r0, r0, #CR_A
399#endif
400#ifdef CONFIG_CPU_DCACHE_DISABLE
401	bic	r0, r0, #CR_C
402#endif
403#ifdef CONFIG_CPU_BPREDICT_DISABLE
404	bic	r0, r0, #CR_Z
405#endif
406#ifdef CONFIG_CPU_ICACHE_DISABLE
407	bic	r0, r0, #CR_I
408#endif
409#ifdef CONFIG_ARM_LPAE
410	mov	r5, #0
411	mcrr	p15, 0, r4, r5, c2		@ load TTBR0
412#else
413	mov	r5, #(domain_val(DOMAIN_USER, DOMAIN_MANAGER) | \
414		      domain_val(DOMAIN_KERNEL, DOMAIN_MANAGER) | \
415		      domain_val(DOMAIN_TABLE, DOMAIN_MANAGER) | \
416		      domain_val(DOMAIN_IO, DOMAIN_CLIENT))
417	mcr	p15, 0, r5, c3, c0, 0		@ load domain access register
418	mcr	p15, 0, r4, c2, c0, 0		@ load page table pointer
419#endif
420	b	__turn_mmu_on
421ENDPROC(__enable_mmu)
422
423/*
424 * Enable the MMU.  This completely changes the structure of the visible
425 * memory space.  You will not be able to trace execution through this.
426 * If you have an enquiry about this, *please* check the linux-arm-kernel
427 * mailing list archives BEFORE sending another post to the list.
428 *
429 *  r0  = cp#15 control register
430 *  r1  = machine ID
431 *  r2  = atags or dtb pointer
432 *  r9  = processor ID
433 *  r13 = *virtual* address to jump to upon completion
434 *
435 * other registers depend on the function called upon completion
436 */
437	.align	5
438	.pushsection	.idmap.text, "ax"
439ENTRY(__turn_mmu_on)
440	mov	r0, r0
441	instr_sync
442	mcr	p15, 0, r0, c1, c0, 0		@ write control reg
443	mrc	p15, 0, r3, c0, c0, 0		@ read id reg
444	instr_sync
445	mov	r3, r3
446	mov	r3, r13
447	mov	pc, r3
448__turn_mmu_on_end:
449ENDPROC(__turn_mmu_on)
450	.popsection
451
452
453#ifdef CONFIG_SMP_ON_UP
454	__INIT
455__fixup_smp:
456	and	r3, r9, #0x000f0000	@ architecture version
457	teq	r3, #0x000f0000		@ CPU ID supported?
458	bne	__fixup_smp_on_up	@ no, assume UP
459
460	bic	r3, r9, #0x00ff0000
461	bic	r3, r3, #0x0000000f	@ mask 0xff00fff0
462	mov	r4, #0x41000000
463	orr	r4, r4, #0x0000b000
464	orr	r4, r4, #0x00000020	@ val 0x4100b020
465	teq	r3, r4			@ ARM 11MPCore?
466	moveq	pc, lr			@ yes, assume SMP
467
468	mrc	p15, 0, r0, c0, c0, 5	@ read MPIDR
469	and	r0, r0, #0xc0000000	@ multiprocessing extensions and
470	teq	r0, #0x80000000		@ not part of a uniprocessor system?
471	moveq	pc, lr			@ yes, assume SMP
472
473__fixup_smp_on_up:
474	adr	r0, 1f
475	ldmia	r0, {r3 - r5}
476	sub	r3, r0, r3
477	add	r4, r4, r3
478	add	r5, r5, r3
479	b	__do_fixup_smp_on_up
480ENDPROC(__fixup_smp)
481
482	.align
4831:	.word	.
484	.word	__smpalt_begin
485	.word	__smpalt_end
486
487	.pushsection .data
488	.globl	smp_on_up
489smp_on_up:
490	ALT_SMP(.long	1)
491	ALT_UP(.long	0)
492	.popsection
493#endif
494
495	.text
496__do_fixup_smp_on_up:
497	cmp	r4, r5
498	movhs	pc, lr
499	ldmia	r4!, {r0, r6}
500 ARM(	str	r6, [r0, r3]	)
501 THUMB(	add	r0, r0, r3	)
502#ifdef __ARMEB__
503 THUMB(	mov	r6, r6, ror #16	)	@ Convert word order for big-endian.
504#endif
505 THUMB(	strh	r6, [r0], #2	)	@ For Thumb-2, store as two halfwords
506 THUMB(	mov	r6, r6, lsr #16	)	@ to be robust against misaligned r3.
507 THUMB(	strh	r6, [r0]	)
508	b	__do_fixup_smp_on_up
509ENDPROC(__do_fixup_smp_on_up)
510
511ENTRY(fixup_smp)
512	stmfd	sp!, {r4 - r6, lr}
513	mov	r4, r0
514	add	r5, r0, r1
515	mov	r3, #0
516	bl	__do_fixup_smp_on_up
517	ldmfd	sp!, {r4 - r6, pc}
518ENDPROC(fixup_smp)
519
520#ifdef CONFIG_ARM_PATCH_PHYS_VIRT
521
522/* __fixup_pv_table - patch the stub instructions with the delta between
523 * PHYS_OFFSET and PAGE_OFFSET, which is assumed to be 16MiB aligned and
524 * can be expressed by an immediate shifter operand. The stub instruction
525 * has a form of '(add|sub) rd, rn, #imm'.
526 */
527	__HEAD
528__fixup_pv_table:
529	adr	r0, 1f
530	ldmia	r0, {r3-r5, r7}
531	sub	r3, r0, r3	@ PHYS_OFFSET - PAGE_OFFSET
532	add	r4, r4, r3	@ adjust table start address
533	add	r5, r5, r3	@ adjust table end address
534	add	r7, r7, r3	@ adjust __pv_phys_offset address
535	str	r8, [r7]	@ save computed PHYS_OFFSET to __pv_phys_offset
536	mov	r6, r3, lsr #24	@ constant for add/sub instructions
537	teq	r3, r6, lsl #24 @ must be 16MiB aligned
538THUMB(	it	ne		@ cross section branch )
539	bne	__error
540	str	r6, [r7, #4]	@ save to __pv_offset
541	b	__fixup_a_pv_table
542ENDPROC(__fixup_pv_table)
543
544	.align
5451:	.long	.
546	.long	__pv_table_begin
547	.long	__pv_table_end
5482:	.long	__pv_phys_offset
549
550	.text
551__fixup_a_pv_table:
552#ifdef CONFIG_THUMB2_KERNEL
553	lsls	r6, #24
554	beq	2f
555	clz	r7, r6
556	lsr	r6, #24
557	lsl	r6, r7
558	bic	r6, #0x0080
559	lsrs	r7, #1
560	orrcs	r6, #0x0080
561	orr	r6, r6, r7, lsl #12
562	orr	r6, #0x4000
563	b	2f
5641:	add     r7, r3
565	ldrh	ip, [r7, #2]
566	and	ip, 0x8f00
567	orr	ip, r6	@ mask in offset bits 31-24
568	strh	ip, [r7, #2]
5692:	cmp	r4, r5
570	ldrcc	r7, [r4], #4	@ use branch for delay slot
571	bcc	1b
572	bx	lr
573#else
574	b	2f
5751:	ldr	ip, [r7, r3]
576	bic	ip, ip, #0x000000ff
577	orr	ip, ip, r6	@ mask in offset bits 31-24
578	str	ip, [r7, r3]
5792:	cmp	r4, r5
580	ldrcc	r7, [r4], #4	@ use branch for delay slot
581	bcc	1b
582	mov	pc, lr
583#endif
584ENDPROC(__fixup_a_pv_table)
585
586ENTRY(fixup_pv_table)
587	stmfd	sp!, {r4 - r7, lr}
588	ldr	r2, 2f			@ get address of __pv_phys_offset
589	mov	r3, #0			@ no offset
590	mov	r4, r0			@ r0 = table start
591	add	r5, r0, r1		@ r1 = table size
592	ldr	r6, [r2, #4]		@ get __pv_offset
593	bl	__fixup_a_pv_table
594	ldmfd	sp!, {r4 - r7, pc}
595ENDPROC(fixup_pv_table)
596
597	.align
5982:	.long	__pv_phys_offset
599
600	.data
601	.globl	__pv_phys_offset
602	.type	__pv_phys_offset, %object
603__pv_phys_offset:
604	.long	0
605	.size	__pv_phys_offset, . - __pv_phys_offset
606__pv_offset:
607	.long	0
608#endif
609
610#include "head-common.S"
611