xref: /linux/arch/arm/kernel/head.S (revision 9ce7677cfd7cd871adb457c80bea3b581b839641)
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/*
37 * swapper_pg_dir is the virtual address of the initial page table.
38 * We place the page tables 16K below KERNEL_RAM_ADDR.  Therefore, we must
39 * make sure that KERNEL_RAM_ADDR is correctly set.  Currently, we expect
40 * the least significant 16 bits to be 0x8000, but we could probably
41 * relax this restriction to KERNEL_RAM_ADDR >= PAGE_OFFSET + 0x4000.
42 */
43#if (KERNEL_RAM_ADDR & 0xffff) != 0x8000
44#error KERNEL_RAM_ADDR must start at 0xXXXX8000
45#endif
46
47	.globl	swapper_pg_dir
48	.equ	swapper_pg_dir, KERNEL_RAM_ADDR - 0x4000
49
50	.macro	pgtbl, rd
51	ldr	\rd, =(__virt_to_phys(KERNEL_RAM_ADDR - 0x4000))
52	.endm
53
54#ifdef CONFIG_XIP_KERNEL
55#define TEXTADDR  XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR)
56#else
57#define TEXTADDR  KERNEL_RAM_ADDR
58#endif
59
60/*
61 * Kernel startup entry point.
62 * ---------------------------
63 *
64 * This is normally called from the decompressor code.  The requirements
65 * are: MMU = off, D-cache = off, I-cache = dont care, r0 = 0,
66 * r1 = machine nr.
67 *
68 * This code is mostly position independent, so if you link the kernel at
69 * 0xc0008000, you call this at __pa(0xc0008000).
70 *
71 * See linux/arch/arm/tools/mach-types for the complete list of machine
72 * numbers for r1.
73 *
74 * We're trying to keep crap to a minimum; DO NOT add any machine specific
75 * crap here - that's what the boot loader (or in extreme, well justified
76 * circumstances, zImage) is for.
77 */
78	__INIT
79	.type	stext, %function
80ENTRY(stext)
81	msr	cpsr_c, #PSR_F_BIT | PSR_I_BIT | MODE_SVC @ ensure svc mode
82						@ and irqs disabled
83	bl	__lookup_processor_type		@ r5=procinfo r9=cpuid
84	movs	r10, r5				@ invalid processor (r5=0)?
85	beq	__error_p			@ yes, error 'p'
86	bl	__lookup_machine_type		@ r5=machinfo
87	movs	r8, r5				@ invalid machine (r5=0)?
88	beq	__error_a			@ yes, error 'a'
89	bl	__create_page_tables
90
91	/*
92	 * The following calls CPU specific code in a position independent
93	 * manner.  See arch/arm/mm/proc-*.S for details.  r10 = base of
94	 * xxx_proc_info structure selected by __lookup_machine_type
95	 * above.  On return, the CPU will be ready for the MMU to be
96	 * turned on, and r0 will hold the CPU control register value.
97	 */
98	ldr	r13, __switch_data		@ address to jump to after
99						@ mmu has been enabled
100	adr	lr, __enable_mmu		@ return (PIC) address
101	add	pc, r10, #PROCINFO_INITFUNC
102
103	.type	__switch_data, %object
104__switch_data:
105	.long	__mmap_switched
106	.long	__data_loc			@ r4
107	.long	__data_start			@ r5
108	.long	__bss_start			@ r6
109	.long	_end				@ r7
110	.long	processor_id			@ r4
111	.long	__machine_arch_type		@ r5
112	.long	cr_alignment			@ r6
113	.long	init_thread_union + THREAD_START_SP @ sp
114
115/*
116 * The following fragment of code is executed with the MMU on, and uses
117 * absolute addresses; this is not position independent.
118 *
119 *  r0  = cp#15 control register
120 *  r1  = machine ID
121 *  r9  = processor ID
122 */
123	.type	__mmap_switched, %function
124__mmap_switched:
125	adr	r3, __switch_data + 4
126
127	ldmia	r3!, {r4, r5, r6, r7}
128	cmp	r4, r5				@ Copy data segment if needed
1291:	cmpne	r5, r6
130	ldrne	fp, [r4], #4
131	strne	fp, [r5], #4
132	bne	1b
133
134	mov	fp, #0				@ Clear BSS (and zero fp)
1351:	cmp	r6, r7
136	strcc	fp, [r6],#4
137	bcc	1b
138
139	ldmia	r3, {r4, r5, r6, sp}
140	str	r9, [r4]			@ Save processor ID
141	str	r1, [r5]			@ Save machine type
142	bic	r4, r0, #CR_A			@ Clear 'A' bit
143	stmia	r6, {r0, r4}			@ Save control register values
144	b	start_kernel
145
146#if defined(CONFIG_SMP)
147	.type   secondary_startup, #function
148ENTRY(secondary_startup)
149	/*
150	 * Common entry point for secondary CPUs.
151	 *
152	 * Ensure that we're in SVC mode, and IRQs are disabled.  Lookup
153	 * the processor type - there is no need to check the machine type
154	 * as it has already been validated by the primary processor.
155	 */
156	msr	cpsr_c, #PSR_F_BIT | PSR_I_BIT | MODE_SVC
157	bl	__lookup_processor_type
158	movs	r10, r5				@ invalid processor?
159	moveq	r0, #'p'			@ yes, error 'p'
160	beq	__error
161
162	/*
163	 * Use the page tables supplied from  __cpu_up.
164	 */
165	adr	r4, __secondary_data
166	ldmia	r4, {r5, r6, r13}		@ address to jump to after
167	sub	r4, r4, r5			@ mmu has been enabled
168	ldr	r4, [r6, r4]			@ get secondary_data.pgdir
169	adr	lr, __enable_mmu		@ return address
170	add	pc, r10, #12			@ initialise processor
171						@ (return control reg)
172
173	/*
174	 * r6  = &secondary_data
175	 */
176ENTRY(__secondary_switched)
177	ldr	sp, [r6, #4]			@ get secondary_data.stack
178	mov	fp, #0
179	b	secondary_start_kernel
180
181	.type	__secondary_data, %object
182__secondary_data:
183	.long	.
184	.long	secondary_data
185	.long	__secondary_switched
186#endif /* defined(CONFIG_SMP) */
187
188
189
190/*
191 * Setup common bits before finally enabling the MMU.  Essentially
192 * this is just loading the page table pointer and domain access
193 * registers.
194 */
195	.type	__enable_mmu, %function
196__enable_mmu:
197#ifdef CONFIG_ALIGNMENT_TRAP
198	orr	r0, r0, #CR_A
199#else
200	bic	r0, r0, #CR_A
201#endif
202#ifdef CONFIG_CPU_DCACHE_DISABLE
203	bic	r0, r0, #CR_C
204#endif
205#ifdef CONFIG_CPU_BPREDICT_DISABLE
206	bic	r0, r0, #CR_Z
207#endif
208#ifdef CONFIG_CPU_ICACHE_DISABLE
209	bic	r0, r0, #CR_I
210#endif
211	mov	r5, #(domain_val(DOMAIN_USER, DOMAIN_MANAGER) | \
212		      domain_val(DOMAIN_KERNEL, DOMAIN_MANAGER) | \
213		      domain_val(DOMAIN_TABLE, DOMAIN_MANAGER) | \
214		      domain_val(DOMAIN_IO, DOMAIN_CLIENT))
215	mcr	p15, 0, r5, c3, c0, 0		@ load domain access register
216	mcr	p15, 0, r4, c2, c0, 0		@ load page table pointer
217	b	__turn_mmu_on
218
219/*
220 * Enable the MMU.  This completely changes the structure of the visible
221 * memory space.  You will not be able to trace execution through this.
222 * If you have an enquiry about this, *please* check the linux-arm-kernel
223 * mailing list archives BEFORE sending another post to the list.
224 *
225 *  r0  = cp#15 control register
226 *  r13 = *virtual* address to jump to upon completion
227 *
228 * other registers depend on the function called upon completion
229 */
230	.align	5
231	.type	__turn_mmu_on, %function
232__turn_mmu_on:
233	mov	r0, r0
234	mcr	p15, 0, r0, c1, c0, 0		@ write control reg
235	mrc	p15, 0, r3, c0, c0, 0		@ read id reg
236	mov	r3, r3
237	mov	r3, r3
238	mov	pc, r13
239
240
241
242/*
243 * Setup the initial page tables.  We only setup the barest
244 * amount which are required to get the kernel running, which
245 * generally means mapping in the kernel code.
246 *
247 * r8  = machinfo
248 * r9  = cpuid
249 * r10 = procinfo
250 *
251 * Returns:
252 *  r0, r3, r5, r6, r7 corrupted
253 *  r4 = physical page table address
254 */
255	.type	__create_page_tables, %function
256__create_page_tables:
257	ldr	r5, [r8, #MACHINFO_PHYSRAM]	@ physram
258	pgtbl	r4				@ page table address
259
260	/*
261	 * Clear the 16K level 1 swapper page table
262	 */
263	mov	r0, r4
264	mov	r3, #0
265	add	r6, r0, #0x4000
2661:	str	r3, [r0], #4
267	str	r3, [r0], #4
268	str	r3, [r0], #4
269	str	r3, [r0], #4
270	teq	r0, r6
271	bne	1b
272
273	ldr	r7, [r10, #PROCINFO_MMUFLAGS]	@ mmuflags
274
275	/*
276	 * Create identity mapping for first MB of kernel to
277	 * cater for the MMU enable.  This identity mapping
278	 * will be removed by paging_init().  We use our current program
279	 * counter to determine corresponding section base address.
280	 */
281	mov	r6, pc, lsr #20			@ start of kernel section
282	orr	r3, r7, r6, lsl #20		@ flags + kernel base
283	str	r3, [r4, r6, lsl #2]		@ identity mapping
284
285	/*
286	 * Now setup the pagetables for our kernel direct
287	 * mapped region.  We round TEXTADDR down to the
288	 * nearest megabyte boundary.  It is assumed that
289	 * the kernel fits within 4 contigous 1MB sections.
290	 */
291	add	r0, r4,  #(TEXTADDR & 0xff000000) >> 18	@ start of kernel
292	str	r3, [r0, #(TEXTADDR & 0x00f00000) >> 18]!
293	add	r3, r3, #1 << 20
294	str	r3, [r0, #4]!			@ KERNEL + 1MB
295	add	r3, r3, #1 << 20
296	str	r3, [r0, #4]!			@ KERNEL + 2MB
297	add	r3, r3, #1 << 20
298	str	r3, [r0, #4]			@ KERNEL + 3MB
299
300	/*
301	 * Then map first 1MB of ram in case it contains our boot params.
302	 */
303	add	r0, r4, #PAGE_OFFSET >> 18
304	orr	r6, r5, r7
305	str	r6, [r0]
306
307#ifdef CONFIG_XIP_KERNEL
308	/*
309	 * Map some ram to cover our .data and .bss areas.
310	 * Mapping 3MB should be plenty.
311	 */
312	sub	r3, r4, r5
313	mov	r3, r3, lsr #20
314	add	r0, r0, r3, lsl #2
315	add	r6, r6, r3, lsl #20
316	str	r6, [r0], #4
317	add	r6, r6, #(1 << 20)
318	str	r6, [r0], #4
319	add	r6, r6, #(1 << 20)
320	str	r6, [r0]
321#endif
322
323#ifdef CONFIG_DEBUG_LL
324	bic	r7, r7, #0x0c			@ turn off cacheable
325						@ and bufferable bits
326	/*
327	 * Map in IO space for serial debugging.
328	 * This allows debug messages to be output
329	 * via a serial console before paging_init.
330	 */
331	ldr	r3, [r8, #MACHINFO_PGOFFIO]
332	add	r0, r4, r3
333	rsb	r3, r3, #0x4000			@ PTRS_PER_PGD*sizeof(long)
334	cmp	r3, #0x0800			@ limit to 512MB
335	movhi	r3, #0x0800
336	add	r6, r0, r3
337	ldr	r3, [r8, #MACHINFO_PHYSIO]
338	orr	r3, r3, r7
3391:	str	r3, [r0], #4
340	add	r3, r3, #1 << 20
341	teq	r0, r6
342	bne	1b
343#if defined(CONFIG_ARCH_NETWINDER) || defined(CONFIG_ARCH_CATS)
344	/*
345	 * If we're using the NetWinder or CATS, we also need to map
346	 * in the 16550-type serial port for the debug messages
347	 */
348	add	r0, r4, #0xff000000 >> 18
349	orr	r3, r7, #0x7c000000
350	str	r3, [r0]
351#endif
352#ifdef CONFIG_ARCH_RPC
353	/*
354	 * Map in screen at 0x02000000 & SCREEN2_BASE
355	 * Similar reasons here - for debug.  This is
356	 * only for Acorn RiscPC architectures.
357	 */
358	add	r0, r4, #0x02000000 >> 18
359	orr	r3, r7, #0x02000000
360	str	r3, [r0]
361	add	r0, r4, #0xd8000000 >> 18
362	str	r3, [r0]
363#endif
364#endif
365	mov	pc, lr
366	.ltorg
367
368
369
370/*
371 * Exception handling.  Something went wrong and we can't proceed.  We
372 * ought to tell the user, but since we don't have any guarantee that
373 * we're even running on the right architecture, we do virtually nothing.
374 *
375 * If CONFIG_DEBUG_LL is set we try to print out something about the error
376 * and hope for the best (useful if bootloader fails to pass a proper
377 * machine ID for example).
378 */
379
380	.type	__error_p, %function
381__error_p:
382#ifdef CONFIG_DEBUG_LL
383	adr	r0, str_p1
384	bl	printascii
385	b	__error
386str_p1:	.asciz	"\nError: unrecognized/unsupported processor variant.\n"
387	.align
388#endif
389
390	.type	__error_a, %function
391__error_a:
392#ifdef CONFIG_DEBUG_LL
393	mov	r4, r1				@ preserve machine ID
394	adr	r0, str_a1
395	bl	printascii
396	mov	r0, r4
397	bl	printhex8
398	adr	r0, str_a2
399	bl	printascii
400	adr	r3, 3f
401	ldmia	r3, {r4, r5, r6}		@ get machine desc list
402	sub	r4, r3, r4			@ get offset between virt&phys
403	add	r5, r5, r4			@ convert virt addresses to
404	add	r6, r6, r4			@ physical address space
4051:	ldr	r0, [r5, #MACHINFO_TYPE]	@ get machine type
406	bl	printhex8
407	mov	r0, #'\t'
408	bl	printch
409	ldr     r0, [r5, #MACHINFO_NAME]	@ get machine name
410	add	r0, r0, r4
411	bl	printascii
412	mov	r0, #'\n'
413	bl	printch
414	add	r5, r5, #SIZEOF_MACHINE_DESC	@ next machine_desc
415	cmp	r5, r6
416	blo	1b
417	adr	r0, str_a3
418	bl	printascii
419	b	__error
420str_a1:	.asciz	"\nError: unrecognized/unsupported machine ID (r1 = 0x"
421str_a2:	.asciz	").\n\nAvailable machine support:\n\nID (hex)\tNAME\n"
422str_a3:	.asciz	"\nPlease check your kernel config and/or bootloader.\n"
423	.align
424#endif
425
426	.type	__error, %function
427__error:
428#ifdef CONFIG_ARCH_RPC
429/*
430 * Turn the screen red on a error - RiscPC only.
431 */
432	mov	r0, #0x02000000
433	mov	r3, #0x11
434	orr	r3, r3, r3, lsl #8
435	orr	r3, r3, r3, lsl #16
436	str	r3, [r0], #4
437	str	r3, [r0], #4
438	str	r3, [r0], #4
439	str	r3, [r0], #4
440#endif
4411:	mov	r0, r0
442	b	1b
443
444
445/*
446 * Read processor ID register (CP#15, CR0), and look up in the linker-built
447 * supported processor list.  Note that we can't use the absolute addresses
448 * for the __proc_info lists since we aren't running with the MMU on
449 * (and therefore, we are not in the correct address space).  We have to
450 * calculate the offset.
451 *
452 * Returns:
453 *	r3, r4, r6 corrupted
454 *	r5 = proc_info pointer in physical address space
455 *	r9 = cpuid
456 */
457	.type	__lookup_processor_type, %function
458__lookup_processor_type:
459	adr	r3, 3f
460	ldmda	r3, {r5, r6, r9}
461	sub	r3, r3, r9			@ get offset between virt&phys
462	add	r5, r5, r3			@ convert virt addresses to
463	add	r6, r6, r3			@ physical address space
464	mrc	p15, 0, r9, c0, c0		@ get processor id
4651:	ldmia	r5, {r3, r4}			@ value, mask
466	and	r4, r4, r9			@ mask wanted bits
467	teq	r3, r4
468	beq	2f
469	add	r5, r5, #PROC_INFO_SZ		@ sizeof(proc_info_list)
470	cmp	r5, r6
471	blo	1b
472	mov	r5, #0				@ unknown processor
4732:	mov	pc, lr
474
475/*
476 * This provides a C-API version of the above function.
477 */
478ENTRY(lookup_processor_type)
479	stmfd	sp!, {r4 - r6, r9, lr}
480	bl	__lookup_processor_type
481	mov	r0, r5
482	ldmfd	sp!, {r4 - r6, r9, pc}
483
484/*
485 * Look in include/asm-arm/procinfo.h and arch/arm/kernel/arch.[ch] for
486 * more information about the __proc_info and __arch_info structures.
487 */
488	.long	__proc_info_begin
489	.long	__proc_info_end
4903:	.long	.
491	.long	__arch_info_begin
492	.long	__arch_info_end
493
494/*
495 * Lookup machine architecture in the linker-build list of architectures.
496 * Note that we can't use the absolute addresses for the __arch_info
497 * lists since we aren't running with the MMU on (and therefore, we are
498 * not in the correct address space).  We have to calculate the offset.
499 *
500 *  r1 = machine architecture number
501 * Returns:
502 *  r3, r4, r6 corrupted
503 *  r5 = mach_info pointer in physical address space
504 */
505	.type	__lookup_machine_type, %function
506__lookup_machine_type:
507	adr	r3, 3b
508	ldmia	r3, {r4, r5, r6}
509	sub	r3, r3, r4			@ get offset between virt&phys
510	add	r5, r5, r3			@ convert virt addresses to
511	add	r6, r6, r3			@ physical address space
5121:	ldr	r3, [r5, #MACHINFO_TYPE]	@ get machine type
513	teq	r3, r1				@ matches loader number?
514	beq	2f				@ found
515	add	r5, r5, #SIZEOF_MACHINE_DESC	@ next machine_desc
516	cmp	r5, r6
517	blo	1b
518	mov	r5, #0				@ unknown machine
5192:	mov	pc, lr
520
521/*
522 * This provides a C-API version of the above function.
523 */
524ENTRY(lookup_machine_type)
525	stmfd	sp!, {r4 - r6, lr}
526	mov	r1, r0
527	bl	__lookup_machine_type
528	mov	r0, r5
529	ldmfd	sp!, {r4 - r6, pc}
530