xref: /linux/arch/arm/kernel/head.S (revision 858259cf7d1c443c836a2022b78cb281f0a9b95e)
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/config.h>
15#include <linux/linkage.h>
16#include <linux/init.h>
17
18#include <asm/assembler.h>
19#include <asm/domain.h>
20#include <asm/mach-types.h>
21#include <asm/procinfo.h>
22#include <asm/ptrace.h>
23#include <asm/asm-offsets.h>
24#include <asm/memory.h>
25#include <asm/thread_info.h>
26#include <asm/system.h>
27
28#define PROCINFO_MMUFLAGS	8
29#define PROCINFO_INITFUNC	12
30
31#define MACHINFO_TYPE		0
32#define MACHINFO_PHYSRAM	4
33#define MACHINFO_PHYSIO		8
34#define MACHINFO_PGOFFIO	12
35#define MACHINFO_NAME		16
36
37/*
38 * swapper_pg_dir is the virtual address of the initial page table.
39 * We place the page tables 16K below KERNEL_RAM_ADDR.  Therefore, we must
40 * make sure that KERNEL_RAM_ADDR is correctly set.  Currently, we expect
41 * the least significant 16 bits to be 0x8000, but we could probably
42 * relax this restriction to KERNEL_RAM_ADDR >= PAGE_OFFSET + 0x4000.
43 */
44#if (KERNEL_RAM_ADDR & 0xffff) != 0x8000
45#error KERNEL_RAM_ADDR must start at 0xXXXX8000
46#endif
47
48	.globl	swapper_pg_dir
49	.equ	swapper_pg_dir, KERNEL_RAM_ADDR - 0x4000
50
51	.macro	pgtbl, rd
52	ldr	\rd, =(__virt_to_phys(KERNEL_RAM_ADDR - 0x4000))
53	.endm
54
55#ifdef CONFIG_XIP_KERNEL
56#define TEXTADDR  XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR)
57#else
58#define TEXTADDR  KERNEL_RAM_ADDR
59#endif
60
61/*
62 * Kernel startup entry point.
63 * ---------------------------
64 *
65 * This is normally called from the decompressor code.  The requirements
66 * are: MMU = off, D-cache = off, I-cache = dont care, r0 = 0,
67 * r1 = machine nr.
68 *
69 * This code is mostly position independent, so if you link the kernel at
70 * 0xc0008000, you call this at __pa(0xc0008000).
71 *
72 * See linux/arch/arm/tools/mach-types for the complete list of machine
73 * numbers for r1.
74 *
75 * We're trying to keep crap to a minimum; DO NOT add any machine specific
76 * crap here - that's what the boot loader (or in extreme, well justified
77 * circumstances, zImage) is for.
78 */
79	__INIT
80	.type	stext, %function
81ENTRY(stext)
82	msr	cpsr_c, #PSR_F_BIT | PSR_I_BIT | MODE_SVC @ ensure svc mode
83						@ and irqs disabled
84	bl	__lookup_processor_type		@ r5=procinfo r9=cpuid
85	movs	r10, r5				@ invalid processor (r5=0)?
86	beq	__error_p				@ yes, error 'p'
87	bl	__lookup_machine_type		@ r5=machinfo
88	movs	r8, r5				@ invalid machine (r5=0)?
89	beq	__error_a			@ yes, error 'a'
90	bl	__create_page_tables
91
92	/*
93	 * The following calls CPU specific code in a position independent
94	 * manner.  See arch/arm/mm/proc-*.S for details.  r10 = base of
95	 * xxx_proc_info structure selected by __lookup_machine_type
96	 * above.  On return, the CPU will be ready for the MMU to be
97	 * turned on, and r0 will hold the CPU control register value.
98	 */
99	ldr	r13, __switch_data		@ address to jump to after
100						@ mmu has been enabled
101	adr	lr, __enable_mmu		@ return (PIC) address
102	add	pc, r10, #PROCINFO_INITFUNC
103
104	.type	__switch_data, %object
105__switch_data:
106	.long	__mmap_switched
107	.long	__data_loc			@ r4
108	.long	__data_start			@ r5
109	.long	__bss_start			@ r6
110	.long	_end				@ r7
111	.long	processor_id			@ r4
112	.long	__machine_arch_type		@ r5
113	.long	cr_alignment			@ r6
114	.long	init_thread_union + THREAD_START_SP @ sp
115
116/*
117 * The following fragment of code is executed with the MMU on, and uses
118 * absolute addresses; this is not position independent.
119 *
120 *  r0  = cp#15 control register
121 *  r1  = machine ID
122 *  r9  = processor ID
123 */
124	.type	__mmap_switched, %function
125__mmap_switched:
126	adr	r3, __switch_data + 4
127
128	ldmia	r3!, {r4, r5, r6, r7}
129	cmp	r4, r5				@ Copy data segment if needed
1301:	cmpne	r5, r6
131	ldrne	fp, [r4], #4
132	strne	fp, [r5], #4
133	bne	1b
134
135	mov	fp, #0				@ Clear BSS (and zero fp)
1361:	cmp	r6, r7
137	strcc	fp, [r6],#4
138	bcc	1b
139
140	ldmia	r3, {r4, r5, r6, sp}
141	str	r9, [r4]			@ Save processor ID
142	str	r1, [r5]			@ Save machine type
143	bic	r4, r0, #CR_A			@ Clear 'A' bit
144	stmia	r6, {r0, r4}			@ Save control register values
145	b	start_kernel
146
147#if defined(CONFIG_SMP)
148	.type   secondary_startup, #function
149ENTRY(secondary_startup)
150	/*
151	 * Common entry point for secondary CPUs.
152	 *
153	 * Ensure that we're in SVC mode, and IRQs are disabled.  Lookup
154	 * the processor type - there is no need to check the machine type
155	 * as it has already been validated by the primary processor.
156	 */
157	msr	cpsr_c, #PSR_F_BIT | PSR_I_BIT | MODE_SVC
158	bl	__lookup_processor_type
159	movs	r10, r5				@ invalid processor?
160	moveq	r0, #'p'			@ yes, error 'p'
161	beq	__error
162
163	/*
164	 * Use the page tables supplied from  __cpu_up.
165	 */
166	adr	r4, __secondary_data
167	ldmia	r4, {r5, r6, r13}		@ address to jump to after
168	sub	r4, r4, r5			@ mmu has been enabled
169	ldr	r4, [r6, r4]			@ get secondary_data.pgdir
170	adr	lr, __enable_mmu		@ return address
171	add	pc, r10, #12			@ initialise processor
172						@ (return control reg)
173
174	/*
175	 * r6  = &secondary_data
176	 */
177ENTRY(__secondary_switched)
178	ldr	sp, [r6, #4]			@ get secondary_data.stack
179	mov	fp, #0
180	b	secondary_start_kernel
181
182	.type	__secondary_data, %object
183__secondary_data:
184	.long	.
185	.long	secondary_data
186	.long	__secondary_switched
187#endif /* defined(CONFIG_SMP) */
188
189
190
191/*
192 * Setup common bits before finally enabling the MMU.  Essentially
193 * this is just loading the page table pointer and domain access
194 * registers.
195 */
196	.type	__enable_mmu, %function
197__enable_mmu:
198#ifdef CONFIG_ALIGNMENT_TRAP
199	orr	r0, r0, #CR_A
200#else
201	bic	r0, r0, #CR_A
202#endif
203#ifdef CONFIG_CPU_DCACHE_DISABLE
204	bic	r0, r0, #CR_C
205#endif
206#ifdef CONFIG_CPU_BPREDICT_DISABLE
207	bic	r0, r0, #CR_Z
208#endif
209#ifdef CONFIG_CPU_ICACHE_DISABLE
210	bic	r0, r0, #CR_I
211#endif
212	mov	r5, #(domain_val(DOMAIN_USER, DOMAIN_MANAGER) | \
213		      domain_val(DOMAIN_KERNEL, DOMAIN_MANAGER) | \
214		      domain_val(DOMAIN_TABLE, DOMAIN_MANAGER) | \
215		      domain_val(DOMAIN_IO, DOMAIN_CLIENT))
216	mcr	p15, 0, r5, c3, c0, 0		@ load domain access register
217	mcr	p15, 0, r4, c2, c0, 0		@ load page table pointer
218	b	__turn_mmu_on
219
220/*
221 * Enable the MMU.  This completely changes the structure of the visible
222 * memory space.  You will not be able to trace execution through this.
223 * If you have an enquiry about this, *please* check the linux-arm-kernel
224 * mailing list archives BEFORE sending another post to the list.
225 *
226 *  r0  = cp#15 control register
227 *  r13 = *virtual* address to jump to upon completion
228 *
229 * other registers depend on the function called upon completion
230 */
231	.align	5
232	.type	__turn_mmu_on, %function
233__turn_mmu_on:
234	mov	r0, r0
235	mcr	p15, 0, r0, c1, c0, 0		@ write control reg
236	mrc	p15, 0, r3, c0, c0, 0		@ read id reg
237	mov	r3, r3
238	mov	r3, r3
239	mov	pc, r13
240
241
242
243/*
244 * Setup the initial page tables.  We only setup the barest
245 * amount which are required to get the kernel running, which
246 * generally means mapping in the kernel code.
247 *
248 * r8  = machinfo
249 * r9  = cpuid
250 * r10 = procinfo
251 *
252 * Returns:
253 *  r0, r3, r5, r6, r7 corrupted
254 *  r4 = physical page table address
255 */
256	.type	__create_page_tables, %function
257__create_page_tables:
258	ldr	r5, [r8, #MACHINFO_PHYSRAM]	@ physram
259	pgtbl	r4				@ page table address
260
261	/*
262	 * Clear the 16K level 1 swapper page table
263	 */
264	mov	r0, r4
265	mov	r3, #0
266	add	r6, r0, #0x4000
2671:	str	r3, [r0], #4
268	str	r3, [r0], #4
269	str	r3, [r0], #4
270	str	r3, [r0], #4
271	teq	r0, r6
272	bne	1b
273
274	ldr	r7, [r10, #PROCINFO_MMUFLAGS]	@ mmuflags
275
276	/*
277	 * Create identity mapping for first MB of kernel to
278	 * cater for the MMU enable.  This identity mapping
279	 * will be removed by paging_init().  We use our current program
280	 * counter to determine corresponding section base address.
281	 */
282	mov	r6, pc, lsr #20			@ start of kernel section
283	orr	r3, r7, r6, lsl #20		@ flags + kernel base
284	str	r3, [r4, r6, lsl #2]		@ identity mapping
285
286	/*
287	 * Now setup the pagetables for our kernel direct
288	 * mapped region.  We round TEXTADDR down to the
289	 * nearest megabyte boundary.  It is assumed that
290	 * the kernel fits within 4 contigous 1MB sections.
291	 */
292	add	r0, r4,  #(TEXTADDR & 0xff000000) >> 18	@ start of kernel
293	str	r3, [r0, #(TEXTADDR & 0x00f00000) >> 18]!
294	add	r3, r3, #1 << 20
295	str	r3, [r0, #4]!			@ KERNEL + 1MB
296	add	r3, r3, #1 << 20
297	str	r3, [r0, #4]!			@ KERNEL + 2MB
298	add	r3, r3, #1 << 20
299	str	r3, [r0, #4]			@ KERNEL + 3MB
300
301	/*
302	 * Then map first 1MB of ram in case it contains our boot params.
303	 */
304	add	r0, r4, #PAGE_OFFSET >> 18
305	orr	r6, r5, r7
306	str	r6, [r0]
307
308#ifdef CONFIG_XIP_KERNEL
309	/*
310	 * Map some ram to cover our .data and .bss areas.
311	 * Mapping 3MB should be plenty.
312	 */
313	sub	r3, r4, r5
314	mov	r3, r3, lsr #20
315	add	r0, r0, r3, lsl #2
316	add	r6, r6, r3, lsl #20
317	str	r6, [r0], #4
318	add	r6, r6, #(1 << 20)
319	str	r6, [r0], #4
320	add	r6, r6, #(1 << 20)
321	str	r6, [r0]
322#endif
323
324#ifdef CONFIG_DEBUG_LL
325	bic	r7, r7, #0x0c			@ turn off cacheable
326						@ and bufferable bits
327	/*
328	 * Map in IO space for serial debugging.
329	 * This allows debug messages to be output
330	 * via a serial console before paging_init.
331	 */
332	ldr	r3, [r8, #MACHINFO_PGOFFIO]
333	add	r0, r4, r3
334	rsb	r3, r3, #0x4000			@ PTRS_PER_PGD*sizeof(long)
335	cmp	r3, #0x0800			@ limit to 512MB
336	movhi	r3, #0x0800
337	add	r6, r0, r3
338	ldr	r3, [r8, #MACHINFO_PHYSIO]
339	orr	r3, r3, r7
3401:	str	r3, [r0], #4
341	add	r3, r3, #1 << 20
342	teq	r0, r6
343	bne	1b
344#if defined(CONFIG_ARCH_NETWINDER) || defined(CONFIG_ARCH_CATS)
345	/*
346	 * If we're using the NetWinder, we need to map in
347	 * the 16550-type serial port for the debug messages
348	 */
349	teq	r1, #MACH_TYPE_NETWINDER
350	teqne	r1, #MACH_TYPE_CATS
351	bne	1f
352	add	r0, r4, #0xff000000 >> 18
353	orr	r3, r7, #0x7c000000
354	str	r3, [r0]
3551:
356#endif
357#ifdef CONFIG_ARCH_RPC
358	/*
359	 * Map in screen at 0x02000000 & SCREEN2_BASE
360	 * Similar reasons here - for debug.  This is
361	 * only for Acorn RiscPC architectures.
362	 */
363	add	r0, r4, #0x02000000 >> 18
364	orr	r3, r7, #0x02000000
365	str	r3, [r0]
366	add	r0, r4, #0xd8000000 >> 18
367	str	r3, [r0]
368#endif
369#endif
370	mov	pc, lr
371	.ltorg
372
373
374
375/*
376 * Exception handling.  Something went wrong and we can't proceed.  We
377 * ought to tell the user, but since we don't have any guarantee that
378 * we're even running on the right architecture, we do virtually nothing.
379 *
380 * If CONFIG_DEBUG_LL is set we try to print out something about the error
381 * and hope for the best (useful if bootloader fails to pass a proper
382 * machine ID for example).
383 */
384
385	.type	__error_p, %function
386__error_p:
387#ifdef CONFIG_DEBUG_LL
388	adr	r0, str_p1
389	bl	printascii
390	b	__error
391str_p1:	.asciz	"\nError: unrecognized/unsupported processor variant.\n"
392	.align
393#endif
394
395	.type	__error_a, %function
396__error_a:
397#ifdef CONFIG_DEBUG_LL
398	mov	r4, r1				@ preserve machine ID
399	adr	r0, str_a1
400	bl	printascii
401	mov	r0, r4
402	bl	printhex8
403	adr	r0, str_a2
404	bl	printascii
405	adr	r3, 3f
406	ldmia	r3, {r4, r5, r6}		@ get machine desc list
407	sub	r4, r3, r4			@ get offset between virt&phys
408	add	r5, r5, r4			@ convert virt addresses to
409	add	r6, r6, r4			@ physical address space
4101:	ldr	r0, [r5, #MACHINFO_TYPE]	@ get machine type
411	bl	printhex8
412	mov	r0, #'\t'
413	bl	printch
414	ldr     r0, [r5, #MACHINFO_NAME]	@ get machine name
415	add	r0, r0, r4
416	bl	printascii
417	mov	r0, #'\n'
418	bl	printch
419	add	r5, r5, #SIZEOF_MACHINE_DESC	@ next machine_desc
420	cmp	r5, r6
421	blo	1b
422	adr	r0, str_a3
423	bl	printascii
424	b	__error
425str_a1:	.asciz	"\nError: unrecognized/unsupported machine ID (r1 = 0x"
426str_a2:	.asciz	").\n\nAvailable machine support:\n\nID (hex)\tNAME\n"
427str_a3:	.asciz	"\nPlease check your kernel config and/or bootloader.\n"
428	.align
429#endif
430
431	.type	__error, %function
432__error:
433#ifdef CONFIG_ARCH_RPC
434/*
435 * Turn the screen red on a error - RiscPC only.
436 */
437	mov	r0, #0x02000000
438	mov	r3, #0x11
439	orr	r3, r3, r3, lsl #8
440	orr	r3, r3, r3, lsl #16
441	str	r3, [r0], #4
442	str	r3, [r0], #4
443	str	r3, [r0], #4
444	str	r3, [r0], #4
445#endif
4461:	mov	r0, r0
447	b	1b
448
449
450/*
451 * Read processor ID register (CP#15, CR0), and look up in the linker-built
452 * supported processor list.  Note that we can't use the absolute addresses
453 * for the __proc_info lists since we aren't running with the MMU on
454 * (and therefore, we are not in the correct address space).  We have to
455 * calculate the offset.
456 *
457 * Returns:
458 *	r3, r4, r6 corrupted
459 *	r5 = proc_info pointer in physical address space
460 *	r9 = cpuid
461 */
462	.type	__lookup_processor_type, %function
463__lookup_processor_type:
464	adr	r3, 3f
465	ldmda	r3, {r5, r6, r9}
466	sub	r3, r3, r9			@ get offset between virt&phys
467	add	r5, r5, r3			@ convert virt addresses to
468	add	r6, r6, r3			@ physical address space
469	mrc	p15, 0, r9, c0, c0		@ get processor id
4701:	ldmia	r5, {r3, r4}			@ value, mask
471	and	r4, r4, r9			@ mask wanted bits
472	teq	r3, r4
473	beq	2f
474	add	r5, r5, #PROC_INFO_SZ		@ sizeof(proc_info_list)
475	cmp	r5, r6
476	blo	1b
477	mov	r5, #0				@ unknown processor
4782:	mov	pc, lr
479
480/*
481 * This provides a C-API version of the above function.
482 */
483ENTRY(lookup_processor_type)
484	stmfd	sp!, {r4 - r6, r9, lr}
485	bl	__lookup_processor_type
486	mov	r0, r5
487	ldmfd	sp!, {r4 - r6, r9, pc}
488
489/*
490 * Look in include/asm-arm/procinfo.h and arch/arm/kernel/arch.[ch] for
491 * more information about the __proc_info and __arch_info structures.
492 */
493	.long	__proc_info_begin
494	.long	__proc_info_end
4953:	.long	.
496	.long	__arch_info_begin
497	.long	__arch_info_end
498
499/*
500 * Lookup machine architecture in the linker-build list of architectures.
501 * Note that we can't use the absolute addresses for the __arch_info
502 * lists since we aren't running with the MMU on (and therefore, we are
503 * not in the correct address space).  We have to calculate the offset.
504 *
505 *  r1 = machine architecture number
506 * Returns:
507 *  r3, r4, r6 corrupted
508 *  r5 = mach_info pointer in physical address space
509 */
510	.type	__lookup_machine_type, %function
511__lookup_machine_type:
512	adr	r3, 3b
513	ldmia	r3, {r4, r5, r6}
514	sub	r3, r3, r4			@ get offset between virt&phys
515	add	r5, r5, r3			@ convert virt addresses to
516	add	r6, r6, r3			@ physical address space
5171:	ldr	r3, [r5, #MACHINFO_TYPE]	@ get machine type
518	teq	r3, r1				@ matches loader number?
519	beq	2f				@ found
520	add	r5, r5, #SIZEOF_MACHINE_DESC	@ next machine_desc
521	cmp	r5, r6
522	blo	1b
523	mov	r5, #0				@ unknown machine
5242:	mov	pc, lr
525
526/*
527 * This provides a C-API version of the above function.
528 */
529ENTRY(lookup_machine_type)
530	stmfd	sp!, {r4 - r6, lr}
531	mov	r1, r0
532	bl	__lookup_machine_type
533	mov	r0, r5
534	ldmfd	sp!, {r4 - r6, pc}
535