xref: /linux/arch/arm/kernel/head.S (revision 4f1933620f57145212cdbb1ac6ce099eeeb21c5a)
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/procinfo.h>
21#include <asm/ptrace.h>
22#include <asm/asm-offsets.h>
23#include <asm/memory.h>
24#include <asm/thread_info.h>
25#include <asm/system.h>
26
27#define PROCINFO_MMUFLAGS	8
28#define PROCINFO_INITFUNC	12
29
30#define MACHINFO_TYPE		0
31#define MACHINFO_PHYSRAM	4
32#define MACHINFO_PHYSIO		8
33#define MACHINFO_PGOFFIO	12
34#define MACHINFO_NAME		16
35
36#define KERNEL_RAM_ADDR	(PAGE_OFFSET + TEXT_OFFSET)
37
38/*
39 * swapper_pg_dir is the virtual address of the initial page table.
40 * We place the page tables 16K below KERNEL_RAM_ADDR.  Therefore, we must
41 * make sure that KERNEL_RAM_ADDR is correctly set.  Currently, we expect
42 * the least significant 16 bits to be 0x8000, but we could probably
43 * relax this restriction to KERNEL_RAM_ADDR >= PAGE_OFFSET + 0x4000.
44 */
45#if (KERNEL_RAM_ADDR & 0xffff) != 0x8000
46#error KERNEL_RAM_ADDR must start at 0xXXXX8000
47#endif
48
49	.globl	swapper_pg_dir
50	.equ	swapper_pg_dir, KERNEL_RAM_ADDR - 0x4000
51
52	.macro	pgtbl, rd
53	ldr	\rd, =(__virt_to_phys(KERNEL_RAM_ADDR - 0x4000))
54	.endm
55
56#ifdef CONFIG_XIP_KERNEL
57#define TEXTADDR  XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR)
58#else
59#define TEXTADDR  KERNEL_RAM_ADDR
60#endif
61
62/*
63 * Kernel startup entry point.
64 * ---------------------------
65 *
66 * This is normally called from the decompressor code.  The requirements
67 * are: MMU = off, D-cache = off, I-cache = dont care, r0 = 0,
68 * r1 = machine nr.
69 *
70 * This code is mostly position independent, so if you link the kernel at
71 * 0xc0008000, you call this at __pa(0xc0008000).
72 *
73 * See linux/arch/arm/tools/mach-types for the complete list of machine
74 * numbers for r1.
75 *
76 * We're trying to keep crap to a minimum; DO NOT add any machine specific
77 * crap here - that's what the boot loader (or in extreme, well justified
78 * circumstances, zImage) is for.
79 */
80	__INIT
81	.type	stext, %function
82ENTRY(stext)
83	msr	cpsr_c, #PSR_F_BIT | PSR_I_BIT | MODE_SVC @ ensure svc mode
84						@ and irqs disabled
85	bl	__lookup_processor_type		@ r5=procinfo r9=cpuid
86	movs	r10, r5				@ invalid processor (r5=0)?
87	beq	__error_p			@ yes, error 'p'
88	bl	__lookup_machine_type		@ r5=machinfo
89	movs	r8, r5				@ invalid machine (r5=0)?
90	beq	__error_a			@ yes, error 'a'
91	bl	__create_page_tables
92
93	/*
94	 * The following calls CPU specific code in a position independent
95	 * manner.  See arch/arm/mm/proc-*.S for details.  r10 = base of
96	 * xxx_proc_info structure selected by __lookup_machine_type
97	 * above.  On return, the CPU will be ready for the MMU to be
98	 * turned on, and r0 will hold the CPU control register value.
99	 */
100	ldr	r13, __switch_data		@ address to jump to after
101						@ mmu has been enabled
102	adr	lr, __enable_mmu		@ return (PIC) address
103	add	pc, r10, #PROCINFO_INITFUNC
104
105	.type	__switch_data, %object
106__switch_data:
107	.long	__mmap_switched
108	.long	__data_loc			@ r4
109	.long	__data_start			@ r5
110	.long	__bss_start			@ r6
111	.long	_end				@ r7
112	.long	processor_id			@ r4
113	.long	__machine_arch_type		@ r5
114	.long	cr_alignment			@ r6
115	.long	init_thread_union + THREAD_START_SP @ sp
116
117/*
118 * The following fragment of code is executed with the MMU on, and uses
119 * absolute addresses; this is not position independent.
120 *
121 *  r0  = cp#15 control register
122 *  r1  = machine ID
123 *  r9  = processor ID
124 */
125	.type	__mmap_switched, %function
126__mmap_switched:
127	adr	r3, __switch_data + 4
128
129	ldmia	r3!, {r4, r5, r6, r7}
130	cmp	r4, r5				@ Copy data segment if needed
1311:	cmpne	r5, r6
132	ldrne	fp, [r4], #4
133	strne	fp, [r5], #4
134	bne	1b
135
136	mov	fp, #0				@ Clear BSS (and zero fp)
1371:	cmp	r6, r7
138	strcc	fp, [r6],#4
139	bcc	1b
140
141	ldmia	r3, {r4, r5, r6, sp}
142	str	r9, [r4]			@ Save processor ID
143	str	r1, [r5]			@ Save machine type
144	bic	r4, r0, #CR_A			@ Clear 'A' bit
145	stmia	r6, {r0, r4}			@ Save control register values
146	b	start_kernel
147
148#if defined(CONFIG_SMP)
149	.type   secondary_startup, #function
150ENTRY(secondary_startup)
151	/*
152	 * Common entry point for secondary CPUs.
153	 *
154	 * Ensure that we're in SVC mode, and IRQs are disabled.  Lookup
155	 * the processor type - there is no need to check the machine type
156	 * as it has already been validated by the primary processor.
157	 */
158	msr	cpsr_c, #PSR_F_BIT | PSR_I_BIT | MODE_SVC
159	bl	__lookup_processor_type
160	movs	r10, r5				@ invalid processor?
161	moveq	r0, #'p'			@ yes, error 'p'
162	beq	__error
163
164	/*
165	 * Use the page tables supplied from  __cpu_up.
166	 */
167	adr	r4, __secondary_data
168	ldmia	r4, {r5, r6, r13}		@ address to jump to after
169	sub	r4, r4, r5			@ mmu has been enabled
170	ldr	r4, [r6, r4]			@ get secondary_data.pgdir
171	adr	lr, __enable_mmu		@ return address
172	add	pc, r10, #12			@ initialise processor
173						@ (return control reg)
174
175	/*
176	 * r6  = &secondary_data
177	 */
178ENTRY(__secondary_switched)
179	ldr	sp, [r6, #4]			@ get secondary_data.stack
180	mov	fp, #0
181	b	secondary_start_kernel
182
183	.type	__secondary_data, %object
184__secondary_data:
185	.long	.
186	.long	secondary_data
187	.long	__secondary_switched
188#endif /* defined(CONFIG_SMP) */
189
190
191
192/*
193 * Setup common bits before finally enabling the MMU.  Essentially
194 * this is just loading the page table pointer and domain access
195 * registers.
196 */
197	.type	__enable_mmu, %function
198__enable_mmu:
199#ifdef CONFIG_ALIGNMENT_TRAP
200	orr	r0, r0, #CR_A
201#else
202	bic	r0, r0, #CR_A
203#endif
204#ifdef CONFIG_CPU_DCACHE_DISABLE
205	bic	r0, r0, #CR_C
206#endif
207#ifdef CONFIG_CPU_BPREDICT_DISABLE
208	bic	r0, r0, #CR_Z
209#endif
210#ifdef CONFIG_CPU_ICACHE_DISABLE
211	bic	r0, r0, #CR_I
212#endif
213	mov	r5, #(domain_val(DOMAIN_USER, DOMAIN_MANAGER) | \
214		      domain_val(DOMAIN_KERNEL, DOMAIN_MANAGER) | \
215		      domain_val(DOMAIN_TABLE, DOMAIN_MANAGER) | \
216		      domain_val(DOMAIN_IO, DOMAIN_CLIENT))
217	mcr	p15, 0, r5, c3, c0, 0		@ load domain access register
218	mcr	p15, 0, r4, c2, c0, 0		@ load page table pointer
219	b	__turn_mmu_on
220
221/*
222 * Enable the MMU.  This completely changes the structure of the visible
223 * memory space.  You will not be able to trace execution through this.
224 * If you have an enquiry about this, *please* check the linux-arm-kernel
225 * mailing list archives BEFORE sending another post to the list.
226 *
227 *  r0  = cp#15 control register
228 *  r13 = *virtual* address to jump to upon completion
229 *
230 * other registers depend on the function called upon completion
231 */
232	.align	5
233	.type	__turn_mmu_on, %function
234__turn_mmu_on:
235	mov	r0, r0
236	mcr	p15, 0, r0, c1, c0, 0		@ write control reg
237	mrc	p15, 0, r3, c0, c0, 0		@ read id reg
238	mov	r3, r3
239	mov	r3, r3
240	mov	pc, r13
241
242
243
244/*
245 * Setup the initial page tables.  We only setup the barest
246 * amount which are required to get the kernel running, which
247 * generally means mapping in the kernel code.
248 *
249 * r8  = machinfo
250 * r9  = cpuid
251 * r10 = procinfo
252 *
253 * Returns:
254 *  r0, r3, r6, r7 corrupted
255 *  r4 = physical page table address
256 */
257	.type	__create_page_tables, %function
258__create_page_tables:
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, r7, #PHYS_OFFSET
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, #PHYS_OFFSET
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 or CATS, we also need to map
347	 * in the 16550-type serial port for the debug messages
348	 */
349	add	r0, r4, #0xff000000 >> 18
350	orr	r3, r7, #0x7c000000
351	str	r3, [r0]
352#endif
353#ifdef CONFIG_ARCH_RPC
354	/*
355	 * Map in screen at 0x02000000 & SCREEN2_BASE
356	 * Similar reasons here - for debug.  This is
357	 * only for Acorn RiscPC architectures.
358	 */
359	add	r0, r4, #0x02000000 >> 18
360	orr	r3, r7, #0x02000000
361	str	r3, [r0]
362	add	r0, r4, #0xd8000000 >> 18
363	str	r3, [r0]
364#endif
365#endif
366	mov	pc, lr
367	.ltorg
368
369
370
371/*
372 * Exception handling.  Something went wrong and we can't proceed.  We
373 * ought to tell the user, but since we don't have any guarantee that
374 * we're even running on the right architecture, we do virtually nothing.
375 *
376 * If CONFIG_DEBUG_LL is set we try to print out something about the error
377 * and hope for the best (useful if bootloader fails to pass a proper
378 * machine ID for example).
379 */
380
381	.type	__error_p, %function
382__error_p:
383#ifdef CONFIG_DEBUG_LL
384	adr	r0, str_p1
385	bl	printascii
386	b	__error
387str_p1:	.asciz	"\nError: unrecognized/unsupported processor variant.\n"
388	.align
389#endif
390
391	.type	__error_a, %function
392__error_a:
393#ifdef CONFIG_DEBUG_LL
394	mov	r4, r1				@ preserve machine ID
395	adr	r0, str_a1
396	bl	printascii
397	mov	r0, r4
398	bl	printhex8
399	adr	r0, str_a2
400	bl	printascii
401	adr	r3, 3f
402	ldmia	r3, {r4, r5, r6}		@ get machine desc list
403	sub	r4, r3, r4			@ get offset between virt&phys
404	add	r5, r5, r4			@ convert virt addresses to
405	add	r6, r6, r4			@ physical address space
4061:	ldr	r0, [r5, #MACHINFO_TYPE]	@ get machine type
407	bl	printhex8
408	mov	r0, #'\t'
409	bl	printch
410	ldr     r0, [r5, #MACHINFO_NAME]	@ get machine name
411	add	r0, r0, r4
412	bl	printascii
413	mov	r0, #'\n'
414	bl	printch
415	add	r5, r5, #SIZEOF_MACHINE_DESC	@ next machine_desc
416	cmp	r5, r6
417	blo	1b
418	adr	r0, str_a3
419	bl	printascii
420	b	__error
421str_a1:	.asciz	"\nError: unrecognized/unsupported machine ID (r1 = 0x"
422str_a2:	.asciz	").\n\nAvailable machine support:\n\nID (hex)\tNAME\n"
423str_a3:	.asciz	"\nPlease check your kernel config and/or bootloader.\n"
424	.align
425#endif
426
427	.type	__error, %function
428__error:
429#ifdef CONFIG_ARCH_RPC
430/*
431 * Turn the screen red on a error - RiscPC only.
432 */
433	mov	r0, #0x02000000
434	mov	r3, #0x11
435	orr	r3, r3, r3, lsl #8
436	orr	r3, r3, r3, lsl #16
437	str	r3, [r0], #4
438	str	r3, [r0], #4
439	str	r3, [r0], #4
440	str	r3, [r0], #4
441#endif
4421:	mov	r0, r0
443	b	1b
444
445
446/*
447 * Read processor ID register (CP#15, CR0), and look up in the linker-built
448 * supported processor list.  Note that we can't use the absolute addresses
449 * for the __proc_info lists since we aren't running with the MMU on
450 * (and therefore, we are not in the correct address space).  We have to
451 * calculate the offset.
452 *
453 * Returns:
454 *	r3, r4, r6 corrupted
455 *	r5 = proc_info pointer in physical address space
456 *	r9 = cpuid
457 */
458	.type	__lookup_processor_type, %function
459__lookup_processor_type:
460	adr	r3, 3f
461	ldmda	r3, {r5, r6, r9}
462	sub	r3, r3, r9			@ get offset between virt&phys
463	add	r5, r5, r3			@ convert virt addresses to
464	add	r6, r6, r3			@ physical address space
465	mrc	p15, 0, r9, c0, c0		@ get processor id
4661:	ldmia	r5, {r3, r4}			@ value, mask
467	and	r4, r4, r9			@ mask wanted bits
468	teq	r3, r4
469	beq	2f
470	add	r5, r5, #PROC_INFO_SZ		@ sizeof(proc_info_list)
471	cmp	r5, r6
472	blo	1b
473	mov	r5, #0				@ unknown processor
4742:	mov	pc, lr
475
476/*
477 * This provides a C-API version of the above function.
478 */
479ENTRY(lookup_processor_type)
480	stmfd	sp!, {r4 - r6, r9, lr}
481	bl	__lookup_processor_type
482	mov	r0, r5
483	ldmfd	sp!, {r4 - r6, r9, pc}
484
485/*
486 * Look in include/asm-arm/procinfo.h and arch/arm/kernel/arch.[ch] for
487 * more information about the __proc_info and __arch_info structures.
488 */
489	.long	__proc_info_begin
490	.long	__proc_info_end
4913:	.long	.
492	.long	__arch_info_begin
493	.long	__arch_info_end
494
495/*
496 * Lookup machine architecture in the linker-build list of architectures.
497 * Note that we can't use the absolute addresses for the __arch_info
498 * lists since we aren't running with the MMU on (and therefore, we are
499 * not in the correct address space).  We have to calculate the offset.
500 *
501 *  r1 = machine architecture number
502 * Returns:
503 *  r3, r4, r6 corrupted
504 *  r5 = mach_info pointer in physical address space
505 */
506	.type	__lookup_machine_type, %function
507__lookup_machine_type:
508	adr	r3, 3b
509	ldmia	r3, {r4, r5, r6}
510	sub	r3, r3, r4			@ get offset between virt&phys
511	add	r5, r5, r3			@ convert virt addresses to
512	add	r6, r6, r3			@ physical address space
5131:	ldr	r3, [r5, #MACHINFO_TYPE]	@ get machine type
514	teq	r3, r1				@ matches loader number?
515	beq	2f				@ found
516	add	r5, r5, #SIZEOF_MACHINE_DESC	@ next machine_desc
517	cmp	r5, r6
518	blo	1b
519	mov	r5, #0				@ unknown machine
5202:	mov	pc, lr
521
522/*
523 * This provides a C-API version of the above function.
524 */
525ENTRY(lookup_machine_type)
526	stmfd	sp!, {r4 - r6, lr}
527	mov	r1, r0
528	bl	__lookup_machine_type
529	mov	r0, r5
530	ldmfd	sp!, {r4 - r6, pc}
531