xref: /linux/arch/arm/boot/compressed/head.S (revision 63307d015b91e626c97bb82e88054af3d0b74643)
1/*
2 *  linux/arch/arm/boot/compressed/head.S
3 *
4 *  Copyright (C) 1996-2002 Russell King
5 *  Copyright (C) 2004 Hyok S. Choi (MPU support)
6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
10 */
11#include <linux/linkage.h>
12#include <asm/assembler.h>
13#include <asm/v7m.h>
14
15#include "efi-header.S"
16
17 AR_CLASS(	.arch	armv7-a	)
18 M_CLASS(	.arch	armv7-m	)
19
20/*
21 * Debugging stuff
22 *
23 * Note that these macros must not contain any code which is not
24 * 100% relocatable.  Any attempt to do so will result in a crash.
25 * Please select one of the following when turning on debugging.
26 */
27#ifdef DEBUG
28
29#if defined(CONFIG_DEBUG_ICEDCC)
30
31#if defined(CONFIG_CPU_V6) || defined(CONFIG_CPU_V6K) || defined(CONFIG_CPU_V7)
32		.macro	loadsp, rb, tmp1, tmp2
33		.endm
34		.macro	writeb, ch, rb
35		mcr	p14, 0, \ch, c0, c5, 0
36		.endm
37#elif defined(CONFIG_CPU_XSCALE)
38		.macro	loadsp, rb, tmp1, tmp2
39		.endm
40		.macro	writeb, ch, rb
41		mcr	p14, 0, \ch, c8, c0, 0
42		.endm
43#else
44		.macro	loadsp, rb, tmp1, tmp2
45		.endm
46		.macro	writeb, ch, rb
47		mcr	p14, 0, \ch, c1, c0, 0
48		.endm
49#endif
50
51#else
52
53#include CONFIG_DEBUG_LL_INCLUDE
54
55		.macro	writeb,	ch, rb
56		senduart \ch, \rb
57		.endm
58
59#if defined(CONFIG_ARCH_SA1100)
60		.macro	loadsp, rb, tmp1, tmp2
61		mov	\rb, #0x80000000	@ physical base address
62#ifdef CONFIG_DEBUG_LL_SER3
63		add	\rb, \rb, #0x00050000	@ Ser3
64#else
65		add	\rb, \rb, #0x00010000	@ Ser1
66#endif
67		.endm
68#else
69		.macro	loadsp,	rb, tmp1, tmp2
70		addruart \rb, \tmp1, \tmp2
71		.endm
72#endif
73#endif
74#endif
75
76		.macro	kputc,val
77		mov	r0, \val
78		bl	putc
79		.endm
80
81		.macro	kphex,val,len
82		mov	r0, \val
83		mov	r1, #\len
84		bl	phex
85		.endm
86
87		.macro	debug_reloc_start
88#ifdef DEBUG
89		kputc	#'\n'
90		kphex	r6, 8		/* processor id */
91		kputc	#':'
92		kphex	r7, 8		/* architecture id */
93#ifdef CONFIG_CPU_CP15
94		kputc	#':'
95		mrc	p15, 0, r0, c1, c0
96		kphex	r0, 8		/* control reg */
97#endif
98		kputc	#'\n'
99		kphex	r5, 8		/* decompressed kernel start */
100		kputc	#'-'
101		kphex	r9, 8		/* decompressed kernel end  */
102		kputc	#'>'
103		kphex	r4, 8		/* kernel execution address */
104		kputc	#'\n'
105#endif
106		.endm
107
108		.macro	debug_reloc_end
109#ifdef DEBUG
110		kphex	r5, 8		/* end of kernel */
111		kputc	#'\n'
112		mov	r0, r4
113		bl	memdump		/* dump 256 bytes at start of kernel */
114#endif
115		.endm
116
117		/*
118		 * Debug kernel copy by printing the memory addresses involved
119		 */
120		.macro dbgkc, begin, end, cbegin, cend
121#ifdef DEBUG
122		kputc   #'\n'
123		kputc   #'C'
124		kputc   #':'
125		kputc   #'0'
126		kputc   #'x'
127		kphex   \begin, 8	/* Start of compressed kernel */
128		kputc	#'-'
129		kputc	#'0'
130		kputc	#'x'
131		kphex	\end, 8		/* End of compressed kernel */
132		kputc	#'-'
133		kputc	#'>'
134		kputc   #'0'
135		kputc   #'x'
136		kphex   \cbegin, 8	/* Start of kernel copy */
137		kputc	#'-'
138		kputc	#'0'
139		kputc	#'x'
140		kphex	\cend, 8	/* End of kernel copy */
141		kputc	#'\n'
142		kputc	#'\r'
143#endif
144		.endm
145
146		.section ".start", #alloc, #execinstr
147/*
148 * sort out different calling conventions
149 */
150		.align
151		/*
152		 * Always enter in ARM state for CPUs that support the ARM ISA.
153		 * As of today (2014) that's exactly the members of the A and R
154		 * classes.
155		 */
156 AR_CLASS(	.arm	)
157start:
158		.type	start,#function
159		.rept	7
160		__nop
161		.endr
162#ifndef CONFIG_THUMB2_KERNEL
163		mov	r0, r0
164#else
165 AR_CLASS(	sub	pc, pc, #3	)	@ A/R: switch to Thumb2 mode
166  M_CLASS(	nop.w			)	@ M: already in Thumb2 mode
167		.thumb
168#endif
169		W(b)	1f
170
171		.word	_magic_sig	@ Magic numbers to help the loader
172		.word	_magic_start	@ absolute load/run zImage address
173		.word	_magic_end	@ zImage end address
174		.word	0x04030201	@ endianness flag
175		.word	0x45454545	@ another magic number to indicate
176		.word	_magic_table	@ additional data table
177
178		__EFI_HEADER
1791:
180 ARM_BE8(	setend	be		)	@ go BE8 if compiled for BE8
181 AR_CLASS(	mrs	r9, cpsr	)
182#ifdef CONFIG_ARM_VIRT_EXT
183		bl	__hyp_stub_install	@ get into SVC mode, reversibly
184#endif
185		mov	r7, r1			@ save architecture ID
186		mov	r8, r2			@ save atags pointer
187
188#ifndef CONFIG_CPU_V7M
189		/*
190		 * Booting from Angel - need to enter SVC mode and disable
191		 * FIQs/IRQs (numeric definitions from angel arm.h source).
192		 * We only do this if we were in user mode on entry.
193		 */
194		mrs	r2, cpsr		@ get current mode
195		tst	r2, #3			@ not user?
196		bne	not_angel
197		mov	r0, #0x17		@ angel_SWIreason_EnterSVC
198 ARM(		swi	0x123456	)	@ angel_SWI_ARM
199 THUMB(		svc	0xab		)	@ angel_SWI_THUMB
200not_angel:
201		safe_svcmode_maskall r0
202		msr	spsr_cxsf, r9		@ Save the CPU boot mode in
203						@ SPSR
204#endif
205		/*
206		 * Note that some cache flushing and other stuff may
207		 * be needed here - is there an Angel SWI call for this?
208		 */
209
210		/*
211		 * some architecture specific code can be inserted
212		 * by the linker here, but it should preserve r7, r8, and r9.
213		 */
214
215		.text
216
217#ifdef CONFIG_AUTO_ZRELADDR
218		/*
219		 * Find the start of physical memory.  As we are executing
220		 * without the MMU on, we are in the physical address space.
221		 * We just need to get rid of any offset by aligning the
222		 * address.
223		 *
224		 * This alignment is a balance between the requirements of
225		 * different platforms - we have chosen 128MB to allow
226		 * platforms which align the start of their physical memory
227		 * to 128MB to use this feature, while allowing the zImage
228		 * to be placed within the first 128MB of memory on other
229		 * platforms.  Increasing the alignment means we place
230		 * stricter alignment requirements on the start of physical
231		 * memory, but relaxing it means that we break people who
232		 * are already placing their zImage in (eg) the top 64MB
233		 * of this range.
234		 */
235		mov	r4, pc
236		and	r4, r4, #0xf8000000
237		/* Determine final kernel image address. */
238		add	r4, r4, #TEXT_OFFSET
239#else
240		ldr	r4, =zreladdr
241#endif
242
243		/*
244		 * Set up a page table only if it won't overwrite ourself.
245		 * That means r4 < pc || r4 - 16k page directory > &_end.
246		 * Given that r4 > &_end is most unfrequent, we add a rough
247		 * additional 1MB of room for a possible appended DTB.
248		 */
249		mov	r0, pc
250		cmp	r0, r4
251		ldrcc	r0, LC0+32
252		addcc	r0, r0, pc
253		cmpcc	r4, r0
254		orrcc	r4, r4, #1		@ remember we skipped cache_on
255		blcs	cache_on
256
257restart:	adr	r0, LC0
258		ldmia	r0, {r1, r2, r3, r6, r10, r11, r12}
259		ldr	sp, [r0, #28]
260
261		/*
262		 * We might be running at a different address.  We need
263		 * to fix up various pointers.
264		 */
265		sub	r0, r0, r1		@ calculate the delta offset
266		add	r6, r6, r0		@ _edata
267		add	r10, r10, r0		@ inflated kernel size location
268
269		/*
270		 * The kernel build system appends the size of the
271		 * decompressed kernel at the end of the compressed data
272		 * in little-endian form.
273		 */
274		ldrb	r9, [r10, #0]
275		ldrb	lr, [r10, #1]
276		orr	r9, r9, lr, lsl #8
277		ldrb	lr, [r10, #2]
278		ldrb	r10, [r10, #3]
279		orr	r9, r9, lr, lsl #16
280		orr	r9, r9, r10, lsl #24
281
282#ifndef CONFIG_ZBOOT_ROM
283		/* malloc space is above the relocated stack (64k max) */
284		add	sp, sp, r0
285		add	r10, sp, #0x10000
286#else
287		/*
288		 * With ZBOOT_ROM the bss/stack is non relocatable,
289		 * but someone could still run this code from RAM,
290		 * in which case our reference is _edata.
291		 */
292		mov	r10, r6
293#endif
294
295		mov	r5, #0			@ init dtb size to 0
296#ifdef CONFIG_ARM_APPENDED_DTB
297/*
298 *   r0  = delta
299 *   r2  = BSS start
300 *   r3  = BSS end
301 *   r4  = final kernel address (possibly with LSB set)
302 *   r5  = appended dtb size (still unknown)
303 *   r6  = _edata
304 *   r7  = architecture ID
305 *   r8  = atags/device tree pointer
306 *   r9  = size of decompressed image
307 *   r10 = end of this image, including  bss/stack/malloc space if non XIP
308 *   r11 = GOT start
309 *   r12 = GOT end
310 *   sp  = stack pointer
311 *
312 * if there are device trees (dtb) appended to zImage, advance r10 so that the
313 * dtb data will get relocated along with the kernel if necessary.
314 */
315
316		ldr	lr, [r6, #0]
317#ifndef __ARMEB__
318		ldr	r1, =0xedfe0dd0		@ sig is 0xd00dfeed big endian
319#else
320		ldr	r1, =0xd00dfeed
321#endif
322		cmp	lr, r1
323		bne	dtb_check_done		@ not found
324
325#ifdef CONFIG_ARM_ATAG_DTB_COMPAT
326		/*
327		 * OK... Let's do some funky business here.
328		 * If we do have a DTB appended to zImage, and we do have
329		 * an ATAG list around, we want the later to be translated
330		 * and folded into the former here. No GOT fixup has occurred
331		 * yet, but none of the code we're about to call uses any
332		 * global variable.
333		*/
334
335		/* Get the initial DTB size */
336		ldr	r5, [r6, #4]
337#ifndef __ARMEB__
338		/* convert to little endian */
339		eor	r1, r5, r5, ror #16
340		bic	r1, r1, #0x00ff0000
341		mov	r5, r5, ror #8
342		eor	r5, r5, r1, lsr #8
343#endif
344		/* 50% DTB growth should be good enough */
345		add	r5, r5, r5, lsr #1
346		/* preserve 64-bit alignment */
347		add	r5, r5, #7
348		bic	r5, r5, #7
349		/* clamp to 32KB min and 1MB max */
350		cmp	r5, #(1 << 15)
351		movlo	r5, #(1 << 15)
352		cmp	r5, #(1 << 20)
353		movhi	r5, #(1 << 20)
354		/* temporarily relocate the stack past the DTB work space */
355		add	sp, sp, r5
356
357		stmfd	sp!, {r0-r3, ip, lr}
358		mov	r0, r8
359		mov	r1, r6
360		mov	r2, r5
361		bl	atags_to_fdt
362
363		/*
364		 * If returned value is 1, there is no ATAG at the location
365		 * pointed by r8.  Try the typical 0x100 offset from start
366		 * of RAM and hope for the best.
367		 */
368		cmp	r0, #1
369		sub	r0, r4, #TEXT_OFFSET
370		bic	r0, r0, #1
371		add	r0, r0, #0x100
372		mov	r1, r6
373		mov	r2, r5
374		bleq	atags_to_fdt
375
376		ldmfd	sp!, {r0-r3, ip, lr}
377		sub	sp, sp, r5
378#endif
379
380		mov	r8, r6			@ use the appended device tree
381
382		/*
383		 * Make sure that the DTB doesn't end up in the final
384		 * kernel's .bss area. To do so, we adjust the decompressed
385		 * kernel size to compensate if that .bss size is larger
386		 * than the relocated code.
387		 */
388		ldr	r5, =_kernel_bss_size
389		adr	r1, wont_overwrite
390		sub	r1, r6, r1
391		subs	r1, r5, r1
392		addhi	r9, r9, r1
393
394		/* Get the current DTB size */
395		ldr	r5, [r6, #4]
396#ifndef __ARMEB__
397		/* convert r5 (dtb size) to little endian */
398		eor	r1, r5, r5, ror #16
399		bic	r1, r1, #0x00ff0000
400		mov	r5, r5, ror #8
401		eor	r5, r5, r1, lsr #8
402#endif
403
404		/* preserve 64-bit alignment */
405		add	r5, r5, #7
406		bic	r5, r5, #7
407
408		/* relocate some pointers past the appended dtb */
409		add	r6, r6, r5
410		add	r10, r10, r5
411		add	sp, sp, r5
412dtb_check_done:
413#endif
414
415/*
416 * Check to see if we will overwrite ourselves.
417 *   r4  = final kernel address (possibly with LSB set)
418 *   r9  = size of decompressed image
419 *   r10 = end of this image, including  bss/stack/malloc space if non XIP
420 * We basically want:
421 *   r4 - 16k page directory >= r10 -> OK
422 *   r4 + image length <= address of wont_overwrite -> OK
423 * Note: the possible LSB in r4 is harmless here.
424 */
425		add	r10, r10, #16384
426		cmp	r4, r10
427		bhs	wont_overwrite
428		add	r10, r4, r9
429		adr	r9, wont_overwrite
430		cmp	r10, r9
431		bls	wont_overwrite
432
433/*
434 * Relocate ourselves past the end of the decompressed kernel.
435 *   r6  = _edata
436 *   r10 = end of the decompressed kernel
437 * Because we always copy ahead, we need to do it from the end and go
438 * backward in case the source and destination overlap.
439 */
440		/*
441		 * Bump to the next 256-byte boundary with the size of
442		 * the relocation code added. This avoids overwriting
443		 * ourself when the offset is small.
444		 */
445		add	r10, r10, #((reloc_code_end - restart + 256) & ~255)
446		bic	r10, r10, #255
447
448		/* Get start of code we want to copy and align it down. */
449		adr	r5, restart
450		bic	r5, r5, #31
451
452/* Relocate the hyp vector base if necessary */
453#ifdef CONFIG_ARM_VIRT_EXT
454		mrs	r0, spsr
455		and	r0, r0, #MODE_MASK
456		cmp	r0, #HYP_MODE
457		bne	1f
458
459		/*
460		 * Compute the address of the hyp vectors after relocation.
461		 * This requires some arithmetic since we cannot directly
462		 * reference __hyp_stub_vectors in a PC-relative way.
463		 * Call __hyp_set_vectors with the new address so that we
464		 * can HVC again after the copy.
465		 */
4660:		adr	r0, 0b
467		movw	r1, #:lower16:__hyp_stub_vectors - 0b
468		movt	r1, #:upper16:__hyp_stub_vectors - 0b
469		add	r0, r0, r1
470		sub	r0, r0, r5
471		add	r0, r0, r10
472		bl	__hyp_set_vectors
4731:
474#endif
475
476		sub	r9, r6, r5		@ size to copy
477		add	r9, r9, #31		@ rounded up to a multiple
478		bic	r9, r9, #31		@ ... of 32 bytes
479		add	r6, r9, r5
480		add	r9, r9, r10
481
482#ifdef DEBUG
483		sub     r10, r6, r5
484		sub     r10, r9, r10
485		/*
486		 * We are about to copy the kernel to a new memory area.
487		 * The boundaries of the new memory area can be found in
488		 * r10 and r9, whilst r5 and r6 contain the boundaries
489		 * of the memory we are going to copy.
490		 * Calling dbgkc will help with the printing of this
491		 * information.
492		 */
493		dbgkc	r5, r6, r10, r9
494#endif
495
4961:		ldmdb	r6!, {r0 - r3, r10 - r12, lr}
497		cmp	r6, r5
498		stmdb	r9!, {r0 - r3, r10 - r12, lr}
499		bhi	1b
500
501		/* Preserve offset to relocated code. */
502		sub	r6, r9, r6
503
504#ifndef CONFIG_ZBOOT_ROM
505		/* cache_clean_flush may use the stack, so relocate it */
506		add	sp, sp, r6
507#endif
508
509		bl	cache_clean_flush
510
511		badr	r0, restart
512		add	r0, r0, r6
513		mov	pc, r0
514
515wont_overwrite:
516/*
517 * If delta is zero, we are running at the address we were linked at.
518 *   r0  = delta
519 *   r2  = BSS start
520 *   r3  = BSS end
521 *   r4  = kernel execution address (possibly with LSB set)
522 *   r5  = appended dtb size (0 if not present)
523 *   r7  = architecture ID
524 *   r8  = atags pointer
525 *   r11 = GOT start
526 *   r12 = GOT end
527 *   sp  = stack pointer
528 */
529		orrs	r1, r0, r5
530		beq	not_relocated
531
532		add	r11, r11, r0
533		add	r12, r12, r0
534
535#ifndef CONFIG_ZBOOT_ROM
536		/*
537		 * If we're running fully PIC === CONFIG_ZBOOT_ROM = n,
538		 * we need to fix up pointers into the BSS region.
539		 * Note that the stack pointer has already been fixed up.
540		 */
541		add	r2, r2, r0
542		add	r3, r3, r0
543
544		/*
545		 * Relocate all entries in the GOT table.
546		 * Bump bss entries to _edata + dtb size
547		 */
5481:		ldr	r1, [r11, #0]		@ relocate entries in the GOT
549		add	r1, r1, r0		@ This fixes up C references
550		cmp	r1, r2			@ if entry >= bss_start &&
551		cmphs	r3, r1			@       bss_end > entry
552		addhi	r1, r1, r5		@    entry += dtb size
553		str	r1, [r11], #4		@ next entry
554		cmp	r11, r12
555		blo	1b
556
557		/* bump our bss pointers too */
558		add	r2, r2, r5
559		add	r3, r3, r5
560
561#else
562
563		/*
564		 * Relocate entries in the GOT table.  We only relocate
565		 * the entries that are outside the (relocated) BSS region.
566		 */
5671:		ldr	r1, [r11, #0]		@ relocate entries in the GOT
568		cmp	r1, r2			@ entry < bss_start ||
569		cmphs	r3, r1			@ _end < entry
570		addlo	r1, r1, r0		@ table.  This fixes up the
571		str	r1, [r11], #4		@ C references.
572		cmp	r11, r12
573		blo	1b
574#endif
575
576not_relocated:	mov	r0, #0
5771:		str	r0, [r2], #4		@ clear bss
578		str	r0, [r2], #4
579		str	r0, [r2], #4
580		str	r0, [r2], #4
581		cmp	r2, r3
582		blo	1b
583
584		/*
585		 * Did we skip the cache setup earlier?
586		 * That is indicated by the LSB in r4.
587		 * Do it now if so.
588		 */
589		tst	r4, #1
590		bic	r4, r4, #1
591		blne	cache_on
592
593/*
594 * The C runtime environment should now be setup sufficiently.
595 * Set up some pointers, and start decompressing.
596 *   r4  = kernel execution address
597 *   r7  = architecture ID
598 *   r8  = atags pointer
599 */
600		mov	r0, r4
601		mov	r1, sp			@ malloc space above stack
602		add	r2, sp, #0x10000	@ 64k max
603		mov	r3, r7
604		bl	decompress_kernel
605		bl	cache_clean_flush
606		bl	cache_off
607
608#ifdef CONFIG_ARM_VIRT_EXT
609		mrs	r0, spsr		@ Get saved CPU boot mode
610		and	r0, r0, #MODE_MASK
611		cmp	r0, #HYP_MODE		@ if not booted in HYP mode...
612		bne	__enter_kernel		@ boot kernel directly
613
614		adr	r12, .L__hyp_reentry_vectors_offset
615		ldr	r0, [r12]
616		add	r0, r0, r12
617
618		bl	__hyp_set_vectors
619		__HVC(0)			@ otherwise bounce to hyp mode
620
621		b	.			@ should never be reached
622
623		.align	2
624.L__hyp_reentry_vectors_offset:	.long	__hyp_reentry_vectors - .
625#else
626		b	__enter_kernel
627#endif
628
629		.align	2
630		.type	LC0, #object
631LC0:		.word	LC0			@ r1
632		.word	__bss_start		@ r2
633		.word	_end			@ r3
634		.word	_edata			@ r6
635		.word	input_data_end - 4	@ r10 (inflated size location)
636		.word	_got_start		@ r11
637		.word	_got_end		@ ip
638		.word	.L_user_stack_end	@ sp
639		.word	_end - restart + 16384 + 1024*1024
640		.size	LC0, . - LC0
641
642#ifdef CONFIG_ARCH_RPC
643		.globl	params
644params:		ldr	r0, =0x10000100		@ params_phys for RPC
645		mov	pc, lr
646		.ltorg
647		.align
648#endif
649
650/*
651 * Turn on the cache.  We need to setup some page tables so that we
652 * can have both the I and D caches on.
653 *
654 * We place the page tables 16k down from the kernel execution address,
655 * and we hope that nothing else is using it.  If we're using it, we
656 * will go pop!
657 *
658 * On entry,
659 *  r4 = kernel execution address
660 *  r7 = architecture number
661 *  r8 = atags pointer
662 * On exit,
663 *  r0, r1, r2, r3, r9, r10, r12 corrupted
664 * This routine must preserve:
665 *  r4, r7, r8
666 */
667		.align	5
668cache_on:	mov	r3, #8			@ cache_on function
669		b	call_cache_fn
670
671/*
672 * Initialize the highest priority protection region, PR7
673 * to cover all 32bit address and cacheable and bufferable.
674 */
675__armv4_mpu_cache_on:
676		mov	r0, #0x3f		@ 4G, the whole
677		mcr	p15, 0, r0, c6, c7, 0	@ PR7 Area Setting
678		mcr 	p15, 0, r0, c6, c7, 1
679
680		mov	r0, #0x80		@ PR7
681		mcr	p15, 0, r0, c2, c0, 0	@ D-cache on
682		mcr	p15, 0, r0, c2, c0, 1	@ I-cache on
683		mcr	p15, 0, r0, c3, c0, 0	@ write-buffer on
684
685		mov	r0, #0xc000
686		mcr	p15, 0, r0, c5, c0, 1	@ I-access permission
687		mcr	p15, 0, r0, c5, c0, 0	@ D-access permission
688
689		mov	r0, #0
690		mcr	p15, 0, r0, c7, c10, 4	@ drain write buffer
691		mcr	p15, 0, r0, c7, c5, 0	@ flush(inval) I-Cache
692		mcr	p15, 0, r0, c7, c6, 0	@ flush(inval) D-Cache
693		mrc	p15, 0, r0, c1, c0, 0	@ read control reg
694						@ ...I .... ..D. WC.M
695		orr	r0, r0, #0x002d		@ .... .... ..1. 11.1
696		orr	r0, r0, #0x1000		@ ...1 .... .... ....
697
698		mcr	p15, 0, r0, c1, c0, 0	@ write control reg
699
700		mov	r0, #0
701		mcr	p15, 0, r0, c7, c5, 0	@ flush(inval) I-Cache
702		mcr	p15, 0, r0, c7, c6, 0	@ flush(inval) D-Cache
703		mov	pc, lr
704
705__armv3_mpu_cache_on:
706		mov	r0, #0x3f		@ 4G, the whole
707		mcr	p15, 0, r0, c6, c7, 0	@ PR7 Area Setting
708
709		mov	r0, #0x80		@ PR7
710		mcr	p15, 0, r0, c2, c0, 0	@ cache on
711		mcr	p15, 0, r0, c3, c0, 0	@ write-buffer on
712
713		mov	r0, #0xc000
714		mcr	p15, 0, r0, c5, c0, 0	@ access permission
715
716		mov	r0, #0
717		mcr	p15, 0, r0, c7, c0, 0	@ invalidate whole cache v3
718		/*
719		 * ?? ARMv3 MMU does not allow reading the control register,
720		 * does this really work on ARMv3 MPU?
721		 */
722		mrc	p15, 0, r0, c1, c0, 0	@ read control reg
723						@ .... .... .... WC.M
724		orr	r0, r0, #0x000d		@ .... .... .... 11.1
725		/* ?? this overwrites the value constructed above? */
726		mov	r0, #0
727		mcr	p15, 0, r0, c1, c0, 0	@ write control reg
728
729		/* ?? invalidate for the second time? */
730		mcr	p15, 0, r0, c7, c0, 0	@ invalidate whole cache v3
731		mov	pc, lr
732
733#ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
734#define CB_BITS 0x08
735#else
736#define CB_BITS 0x0c
737#endif
738
739__setup_mmu:	sub	r3, r4, #16384		@ Page directory size
740		bic	r3, r3, #0xff		@ Align the pointer
741		bic	r3, r3, #0x3f00
742/*
743 * Initialise the page tables, turning on the cacheable and bufferable
744 * bits for the RAM area only.
745 */
746		mov	r0, r3
747		mov	r9, r0, lsr #18
748		mov	r9, r9, lsl #18		@ start of RAM
749		add	r10, r9, #0x10000000	@ a reasonable RAM size
750		mov	r1, #0x12		@ XN|U + section mapping
751		orr	r1, r1, #3 << 10	@ AP=11
752		add	r2, r3, #16384
7531:		cmp	r1, r9			@ if virt > start of RAM
754		cmphs	r10, r1			@   && end of RAM > virt
755		bic	r1, r1, #0x1c		@ clear XN|U + C + B
756		orrlo	r1, r1, #0x10		@ Set XN|U for non-RAM
757		orrhs	r1, r1, r6		@ set RAM section settings
758		str	r1, [r0], #4		@ 1:1 mapping
759		add	r1, r1, #1048576
760		teq	r0, r2
761		bne	1b
762/*
763 * If ever we are running from Flash, then we surely want the cache
764 * to be enabled also for our execution instance...  We map 2MB of it
765 * so there is no map overlap problem for up to 1 MB compressed kernel.
766 * If the execution is in RAM then we would only be duplicating the above.
767 */
768		orr	r1, r6, #0x04		@ ensure B is set for this
769		orr	r1, r1, #3 << 10
770		mov	r2, pc
771		mov	r2, r2, lsr #20
772		orr	r1, r1, r2, lsl #20
773		add	r0, r3, r2, lsl #2
774		str	r1, [r0], #4
775		add	r1, r1, #1048576
776		str	r1, [r0]
777		mov	pc, lr
778ENDPROC(__setup_mmu)
779
780@ Enable unaligned access on v6, to allow better code generation
781@ for the decompressor C code:
782__armv6_mmu_cache_on:
783		mrc	p15, 0, r0, c1, c0, 0	@ read SCTLR
784		bic	r0, r0, #2		@ A (no unaligned access fault)
785		orr	r0, r0, #1 << 22	@ U (v6 unaligned access model)
786		mcr	p15, 0, r0, c1, c0, 0	@ write SCTLR
787		b	__armv4_mmu_cache_on
788
789__arm926ejs_mmu_cache_on:
790#ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
791		mov	r0, #4			@ put dcache in WT mode
792		mcr	p15, 7, r0, c15, c0, 0
793#endif
794
795__armv4_mmu_cache_on:
796		mov	r12, lr
797#ifdef CONFIG_MMU
798		mov	r6, #CB_BITS | 0x12	@ U
799		bl	__setup_mmu
800		mov	r0, #0
801		mcr	p15, 0, r0, c7, c10, 4	@ drain write buffer
802		mcr	p15, 0, r0, c8, c7, 0	@ flush I,D TLBs
803		mrc	p15, 0, r0, c1, c0, 0	@ read control reg
804		orr	r0, r0, #0x5000		@ I-cache enable, RR cache replacement
805		orr	r0, r0, #0x0030
806 ARM_BE8(	orr	r0, r0, #1 << 25 )	@ big-endian page tables
807		bl	__common_mmu_cache_on
808		mov	r0, #0
809		mcr	p15, 0, r0, c8, c7, 0	@ flush I,D TLBs
810#endif
811		mov	pc, r12
812
813__armv7_mmu_cache_on:
814		mov	r12, lr
815#ifdef CONFIG_MMU
816		mrc	p15, 0, r11, c0, c1, 4	@ read ID_MMFR0
817		tst	r11, #0xf		@ VMSA
818		movne	r6, #CB_BITS | 0x02	@ !XN
819		blne	__setup_mmu
820		mov	r0, #0
821		mcr	p15, 0, r0, c7, c10, 4	@ drain write buffer
822		tst	r11, #0xf		@ VMSA
823		mcrne	p15, 0, r0, c8, c7, 0	@ flush I,D TLBs
824#endif
825		mrc	p15, 0, r0, c1, c0, 0	@ read control reg
826		bic	r0, r0, #1 << 28	@ clear SCTLR.TRE
827		orr	r0, r0, #0x5000		@ I-cache enable, RR cache replacement
828		orr	r0, r0, #0x003c		@ write buffer
829		bic	r0, r0, #2		@ A (no unaligned access fault)
830		orr	r0, r0, #1 << 22	@ U (v6 unaligned access model)
831						@ (needed for ARM1176)
832#ifdef CONFIG_MMU
833 ARM_BE8(	orr	r0, r0, #1 << 25 )	@ big-endian page tables
834		mrcne   p15, 0, r6, c2, c0, 2   @ read ttb control reg
835		orrne	r0, r0, #1		@ MMU enabled
836		movne	r1, #0xfffffffd		@ domain 0 = client
837		bic     r6, r6, #1 << 31        @ 32-bit translation system
838		bic     r6, r6, #(7 << 0) | (1 << 4)	@ use only ttbr0
839		mcrne	p15, 0, r3, c2, c0, 0	@ load page table pointer
840		mcrne	p15, 0, r1, c3, c0, 0	@ load domain access control
841		mcrne   p15, 0, r6, c2, c0, 2   @ load ttb control
842#endif
843		mcr	p15, 0, r0, c7, c5, 4	@ ISB
844		mcr	p15, 0, r0, c1, c0, 0	@ load control register
845		mrc	p15, 0, r0, c1, c0, 0	@ and read it back
846		mov	r0, #0
847		mcr	p15, 0, r0, c7, c5, 4	@ ISB
848		mov	pc, r12
849
850__fa526_cache_on:
851		mov	r12, lr
852		mov	r6, #CB_BITS | 0x12	@ U
853		bl	__setup_mmu
854		mov	r0, #0
855		mcr	p15, 0, r0, c7, c7, 0	@ Invalidate whole cache
856		mcr	p15, 0, r0, c7, c10, 4	@ drain write buffer
857		mcr	p15, 0, r0, c8, c7, 0	@ flush UTLB
858		mrc	p15, 0, r0, c1, c0, 0	@ read control reg
859		orr	r0, r0, #0x1000		@ I-cache enable
860		bl	__common_mmu_cache_on
861		mov	r0, #0
862		mcr	p15, 0, r0, c8, c7, 0	@ flush UTLB
863		mov	pc, r12
864
865__common_mmu_cache_on:
866#ifndef CONFIG_THUMB2_KERNEL
867#ifndef DEBUG
868		orr	r0, r0, #0x000d		@ Write buffer, mmu
869#endif
870		mov	r1, #-1
871		mcr	p15, 0, r3, c2, c0, 0	@ load page table pointer
872		mcr	p15, 0, r1, c3, c0, 0	@ load domain access control
873		b	1f
874		.align	5			@ cache line aligned
8751:		mcr	p15, 0, r0, c1, c0, 0	@ load control register
876		mrc	p15, 0, r0, c1, c0, 0	@ and read it back to
877		sub	pc, lr, r0, lsr #32	@ properly flush pipeline
878#endif
879
880#define PROC_ENTRY_SIZE (4*5)
881
882/*
883 * Here follow the relocatable cache support functions for the
884 * various processors.  This is a generic hook for locating an
885 * entry and jumping to an instruction at the specified offset
886 * from the start of the block.  Please note this is all position
887 * independent code.
888 *
889 *  r1  = corrupted
890 *  r2  = corrupted
891 *  r3  = block offset
892 *  r9  = corrupted
893 *  r12 = corrupted
894 */
895
896call_cache_fn:	adr	r12, proc_types
897#ifdef CONFIG_CPU_CP15
898		mrc	p15, 0, r9, c0, c0	@ get processor ID
899#elif defined(CONFIG_CPU_V7M)
900		/*
901		 * On v7-M the processor id is located in the V7M_SCB_CPUID
902		 * register, but as cache handling is IMPLEMENTATION DEFINED on
903		 * v7-M (if existant at all) we just return early here.
904		 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
905		 * __armv7_mmu_cache_{on,off,flush}) would be selected which
906		 * use cp15 registers that are not implemented on v7-M.
907		 */
908		bx	lr
909#else
910		ldr	r9, =CONFIG_PROCESSOR_ID
911#endif
9121:		ldr	r1, [r12, #0]		@ get value
913		ldr	r2, [r12, #4]		@ get mask
914		eor	r1, r1, r9		@ (real ^ match)
915		tst	r1, r2			@       & mask
916 ARM(		addeq	pc, r12, r3		) @ call cache function
917 THUMB(		addeq	r12, r3			)
918 THUMB(		moveq	pc, r12			) @ call cache function
919		add	r12, r12, #PROC_ENTRY_SIZE
920		b	1b
921
922/*
923 * Table for cache operations.  This is basically:
924 *   - CPU ID match
925 *   - CPU ID mask
926 *   - 'cache on' method instruction
927 *   - 'cache off' method instruction
928 *   - 'cache flush' method instruction
929 *
930 * We match an entry using: ((real_id ^ match) & mask) == 0
931 *
932 * Writethrough caches generally only need 'on' and 'off'
933 * methods.  Writeback caches _must_ have the flush method
934 * defined.
935 */
936		.align	2
937		.type	proc_types,#object
938proc_types:
939		.word	0x41000000		@ old ARM ID
940		.word	0xff00f000
941		mov	pc, lr
942 THUMB(		nop				)
943		mov	pc, lr
944 THUMB(		nop				)
945		mov	pc, lr
946 THUMB(		nop				)
947
948		.word	0x41007000		@ ARM7/710
949		.word	0xfff8fe00
950		mov	pc, lr
951 THUMB(		nop				)
952		mov	pc, lr
953 THUMB(		nop				)
954		mov	pc, lr
955 THUMB(		nop				)
956
957		.word	0x41807200		@ ARM720T (writethrough)
958		.word	0xffffff00
959		W(b)	__armv4_mmu_cache_on
960		W(b)	__armv4_mmu_cache_off
961		mov	pc, lr
962 THUMB(		nop				)
963
964		.word	0x41007400		@ ARM74x
965		.word	0xff00ff00
966		W(b)	__armv3_mpu_cache_on
967		W(b)	__armv3_mpu_cache_off
968		W(b)	__armv3_mpu_cache_flush
969
970		.word	0x41009400		@ ARM94x
971		.word	0xff00ff00
972		W(b)	__armv4_mpu_cache_on
973		W(b)	__armv4_mpu_cache_off
974		W(b)	__armv4_mpu_cache_flush
975
976		.word	0x41069260		@ ARM926EJ-S (v5TEJ)
977		.word	0xff0ffff0
978		W(b)	__arm926ejs_mmu_cache_on
979		W(b)	__armv4_mmu_cache_off
980		W(b)	__armv5tej_mmu_cache_flush
981
982		.word	0x00007000		@ ARM7 IDs
983		.word	0x0000f000
984		mov	pc, lr
985 THUMB(		nop				)
986		mov	pc, lr
987 THUMB(		nop				)
988		mov	pc, lr
989 THUMB(		nop				)
990
991		@ Everything from here on will be the new ID system.
992
993		.word	0x4401a100		@ sa110 / sa1100
994		.word	0xffffffe0
995		W(b)	__armv4_mmu_cache_on
996		W(b)	__armv4_mmu_cache_off
997		W(b)	__armv4_mmu_cache_flush
998
999		.word	0x6901b110		@ sa1110
1000		.word	0xfffffff0
1001		W(b)	__armv4_mmu_cache_on
1002		W(b)	__armv4_mmu_cache_off
1003		W(b)	__armv4_mmu_cache_flush
1004
1005		.word	0x56056900
1006		.word	0xffffff00		@ PXA9xx
1007		W(b)	__armv4_mmu_cache_on
1008		W(b)	__armv4_mmu_cache_off
1009		W(b)	__armv4_mmu_cache_flush
1010
1011		.word	0x56158000		@ PXA168
1012		.word	0xfffff000
1013		W(b)	__armv4_mmu_cache_on
1014		W(b)	__armv4_mmu_cache_off
1015		W(b)	__armv5tej_mmu_cache_flush
1016
1017		.word	0x56050000		@ Feroceon
1018		.word	0xff0f0000
1019		W(b)	__armv4_mmu_cache_on
1020		W(b)	__armv4_mmu_cache_off
1021		W(b)	__armv5tej_mmu_cache_flush
1022
1023#ifdef CONFIG_CPU_FEROCEON_OLD_ID
1024		/* this conflicts with the standard ARMv5TE entry */
1025		.long	0x41009260		@ Old Feroceon
1026		.long	0xff00fff0
1027		b	__armv4_mmu_cache_on
1028		b	__armv4_mmu_cache_off
1029		b	__armv5tej_mmu_cache_flush
1030#endif
1031
1032		.word	0x66015261		@ FA526
1033		.word	0xff01fff1
1034		W(b)	__fa526_cache_on
1035		W(b)	__armv4_mmu_cache_off
1036		W(b)	__fa526_cache_flush
1037
1038		@ These match on the architecture ID
1039
1040		.word	0x00020000		@ ARMv4T
1041		.word	0x000f0000
1042		W(b)	__armv4_mmu_cache_on
1043		W(b)	__armv4_mmu_cache_off
1044		W(b)	__armv4_mmu_cache_flush
1045
1046		.word	0x00050000		@ ARMv5TE
1047		.word	0x000f0000
1048		W(b)	__armv4_mmu_cache_on
1049		W(b)	__armv4_mmu_cache_off
1050		W(b)	__armv4_mmu_cache_flush
1051
1052		.word	0x00060000		@ ARMv5TEJ
1053		.word	0x000f0000
1054		W(b)	__armv4_mmu_cache_on
1055		W(b)	__armv4_mmu_cache_off
1056		W(b)	__armv5tej_mmu_cache_flush
1057
1058		.word	0x0007b000		@ ARMv6
1059		.word	0x000ff000
1060		W(b)	__armv6_mmu_cache_on
1061		W(b)	__armv4_mmu_cache_off
1062		W(b)	__armv6_mmu_cache_flush
1063
1064		.word	0x000f0000		@ new CPU Id
1065		.word	0x000f0000
1066		W(b)	__armv7_mmu_cache_on
1067		W(b)	__armv7_mmu_cache_off
1068		W(b)	__armv7_mmu_cache_flush
1069
1070		.word	0			@ unrecognised type
1071		.word	0
1072		mov	pc, lr
1073 THUMB(		nop				)
1074		mov	pc, lr
1075 THUMB(		nop				)
1076		mov	pc, lr
1077 THUMB(		nop				)
1078
1079		.size	proc_types, . - proc_types
1080
1081		/*
1082		 * If you get a "non-constant expression in ".if" statement"
1083		 * error from the assembler on this line, check that you have
1084		 * not accidentally written a "b" instruction where you should
1085		 * have written W(b).
1086		 */
1087		.if (. - proc_types) % PROC_ENTRY_SIZE != 0
1088		.error "The size of one or more proc_types entries is wrong."
1089		.endif
1090
1091/*
1092 * Turn off the Cache and MMU.  ARMv3 does not support
1093 * reading the control register, but ARMv4 does.
1094 *
1095 * On exit,
1096 *  r0, r1, r2, r3, r9, r12 corrupted
1097 * This routine must preserve:
1098 *  r4, r7, r8
1099 */
1100		.align	5
1101cache_off:	mov	r3, #12			@ cache_off function
1102		b	call_cache_fn
1103
1104__armv4_mpu_cache_off:
1105		mrc	p15, 0, r0, c1, c0
1106		bic	r0, r0, #0x000d
1107		mcr	p15, 0, r0, c1, c0	@ turn MPU and cache off
1108		mov	r0, #0
1109		mcr	p15, 0, r0, c7, c10, 4	@ drain write buffer
1110		mcr	p15, 0, r0, c7, c6, 0	@ flush D-Cache
1111		mcr	p15, 0, r0, c7, c5, 0	@ flush I-Cache
1112		mov	pc, lr
1113
1114__armv3_mpu_cache_off:
1115		mrc	p15, 0, r0, c1, c0
1116		bic	r0, r0, #0x000d
1117		mcr	p15, 0, r0, c1, c0, 0	@ turn MPU and cache off
1118		mov	r0, #0
1119		mcr	p15, 0, r0, c7, c0, 0	@ invalidate whole cache v3
1120		mov	pc, lr
1121
1122__armv4_mmu_cache_off:
1123#ifdef CONFIG_MMU
1124		mrc	p15, 0, r0, c1, c0
1125		bic	r0, r0, #0x000d
1126		mcr	p15, 0, r0, c1, c0	@ turn MMU and cache off
1127		mov	r0, #0
1128		mcr	p15, 0, r0, c7, c7	@ invalidate whole cache v4
1129		mcr	p15, 0, r0, c8, c7	@ invalidate whole TLB v4
1130#endif
1131		mov	pc, lr
1132
1133__armv7_mmu_cache_off:
1134		mrc	p15, 0, r0, c1, c0
1135#ifdef CONFIG_MMU
1136		bic	r0, r0, #0x000d
1137#else
1138		bic	r0, r0, #0x000c
1139#endif
1140		mcr	p15, 0, r0, c1, c0	@ turn MMU and cache off
1141		mov	r12, lr
1142		bl	__armv7_mmu_cache_flush
1143		mov	r0, #0
1144#ifdef CONFIG_MMU
1145		mcr	p15, 0, r0, c8, c7, 0	@ invalidate whole TLB
1146#endif
1147		mcr	p15, 0, r0, c7, c5, 6	@ invalidate BTC
1148		mcr	p15, 0, r0, c7, c10, 4	@ DSB
1149		mcr	p15, 0, r0, c7, c5, 4	@ ISB
1150		mov	pc, r12
1151
1152/*
1153 * Clean and flush the cache to maintain consistency.
1154 *
1155 * On exit,
1156 *  r1, r2, r3, r9, r10, r11, r12 corrupted
1157 * This routine must preserve:
1158 *  r4, r6, r7, r8
1159 */
1160		.align	5
1161cache_clean_flush:
1162		mov	r3, #16
1163		b	call_cache_fn
1164
1165__armv4_mpu_cache_flush:
1166		tst	r4, #1
1167		movne	pc, lr
1168		mov	r2, #1
1169		mov	r3, #0
1170		mcr	p15, 0, ip, c7, c6, 0	@ invalidate D cache
1171		mov	r1, #7 << 5		@ 8 segments
11721:		orr	r3, r1, #63 << 26	@ 64 entries
11732:		mcr	p15, 0, r3, c7, c14, 2	@ clean & invalidate D index
1174		subs	r3, r3, #1 << 26
1175		bcs	2b			@ entries 63 to 0
1176		subs 	r1, r1, #1 << 5
1177		bcs	1b			@ segments 7 to 0
1178
1179		teq	r2, #0
1180		mcrne	p15, 0, ip, c7, c5, 0	@ invalidate I cache
1181		mcr	p15, 0, ip, c7, c10, 4	@ drain WB
1182		mov	pc, lr
1183
1184__fa526_cache_flush:
1185		tst	r4, #1
1186		movne	pc, lr
1187		mov	r1, #0
1188		mcr	p15, 0, r1, c7, c14, 0	@ clean and invalidate D cache
1189		mcr	p15, 0, r1, c7, c5, 0	@ flush I cache
1190		mcr	p15, 0, r1, c7, c10, 4	@ drain WB
1191		mov	pc, lr
1192
1193__armv6_mmu_cache_flush:
1194		mov	r1, #0
1195		tst	r4, #1
1196		mcreq	p15, 0, r1, c7, c14, 0	@ clean+invalidate D
1197		mcr	p15, 0, r1, c7, c5, 0	@ invalidate I+BTB
1198		mcreq	p15, 0, r1, c7, c15, 0	@ clean+invalidate unified
1199		mcr	p15, 0, r1, c7, c10, 4	@ drain WB
1200		mov	pc, lr
1201
1202__armv7_mmu_cache_flush:
1203		tst	r4, #1
1204		bne	iflush
1205		mrc	p15, 0, r10, c0, c1, 5	@ read ID_MMFR1
1206		tst	r10, #0xf << 16		@ hierarchical cache (ARMv7)
1207		mov	r10, #0
1208		beq	hierarchical
1209		mcr	p15, 0, r10, c7, c14, 0	@ clean+invalidate D
1210		b	iflush
1211hierarchical:
1212		mcr	p15, 0, r10, c7, c10, 5	@ DMB
1213		stmfd	sp!, {r0-r7, r9-r11}
1214		mrc	p15, 1, r0, c0, c0, 1	@ read clidr
1215		ands	r3, r0, #0x7000000	@ extract loc from clidr
1216		mov	r3, r3, lsr #23		@ left align loc bit field
1217		beq	finished		@ if loc is 0, then no need to clean
1218		mov	r10, #0			@ start clean at cache level 0
1219loop1:
1220		add	r2, r10, r10, lsr #1	@ work out 3x current cache level
1221		mov	r1, r0, lsr r2		@ extract cache type bits from clidr
1222		and	r1, r1, #7		@ mask of the bits for current cache only
1223		cmp	r1, #2			@ see what cache we have at this level
1224		blt	skip			@ skip if no cache, or just i-cache
1225		mcr	p15, 2, r10, c0, c0, 0	@ select current cache level in cssr
1226		mcr	p15, 0, r10, c7, c5, 4	@ isb to sych the new cssr&csidr
1227		mrc	p15, 1, r1, c0, c0, 0	@ read the new csidr
1228		and	r2, r1, #7		@ extract the length of the cache lines
1229		add	r2, r2, #4		@ add 4 (line length offset)
1230		ldr	r4, =0x3ff
1231		ands	r4, r4, r1, lsr #3	@ find maximum number on the way size
1232		clz	r5, r4			@ find bit position of way size increment
1233		ldr	r7, =0x7fff
1234		ands	r7, r7, r1, lsr #13	@ extract max number of the index size
1235loop2:
1236		mov	r9, r4			@ create working copy of max way size
1237loop3:
1238 ARM(		orr	r11, r10, r9, lsl r5	) @ factor way and cache number into r11
1239 ARM(		orr	r11, r11, r7, lsl r2	) @ factor index number into r11
1240 THUMB(		lsl	r6, r9, r5		)
1241 THUMB(		orr	r11, r10, r6		) @ factor way and cache number into r11
1242 THUMB(		lsl	r6, r7, r2		)
1243 THUMB(		orr	r11, r11, r6		) @ factor index number into r11
1244		mcr	p15, 0, r11, c7, c14, 2	@ clean & invalidate by set/way
1245		subs	r9, r9, #1		@ decrement the way
1246		bge	loop3
1247		subs	r7, r7, #1		@ decrement the index
1248		bge	loop2
1249skip:
1250		add	r10, r10, #2		@ increment cache number
1251		cmp	r3, r10
1252		bgt	loop1
1253finished:
1254		ldmfd	sp!, {r0-r7, r9-r11}
1255		mov	r10, #0			@ switch back to cache level 0
1256		mcr	p15, 2, r10, c0, c0, 0	@ select current cache level in cssr
1257iflush:
1258		mcr	p15, 0, r10, c7, c10, 4	@ DSB
1259		mcr	p15, 0, r10, c7, c5, 0	@ invalidate I+BTB
1260		mcr	p15, 0, r10, c7, c10, 4	@ DSB
1261		mcr	p15, 0, r10, c7, c5, 4	@ ISB
1262		mov	pc, lr
1263
1264__armv5tej_mmu_cache_flush:
1265		tst	r4, #1
1266		movne	pc, lr
12671:		mrc	p15, 0, r15, c7, c14, 3	@ test,clean,invalidate D cache
1268		bne	1b
1269		mcr	p15, 0, r0, c7, c5, 0	@ flush I cache
1270		mcr	p15, 0, r0, c7, c10, 4	@ drain WB
1271		mov	pc, lr
1272
1273__armv4_mmu_cache_flush:
1274		tst	r4, #1
1275		movne	pc, lr
1276		mov	r2, #64*1024		@ default: 32K dcache size (*2)
1277		mov	r11, #32		@ default: 32 byte line size
1278		mrc	p15, 0, r3, c0, c0, 1	@ read cache type
1279		teq	r3, r9			@ cache ID register present?
1280		beq	no_cache_id
1281		mov	r1, r3, lsr #18
1282		and	r1, r1, #7
1283		mov	r2, #1024
1284		mov	r2, r2, lsl r1		@ base dcache size *2
1285		tst	r3, #1 << 14		@ test M bit
1286		addne	r2, r2, r2, lsr #1	@ +1/2 size if M == 1
1287		mov	r3, r3, lsr #12
1288		and	r3, r3, #3
1289		mov	r11, #8
1290		mov	r11, r11, lsl r3	@ cache line size in bytes
1291no_cache_id:
1292		mov	r1, pc
1293		bic	r1, r1, #63		@ align to longest cache line
1294		add	r2, r1, r2
12951:
1296 ARM(		ldr	r3, [r1], r11		) @ s/w flush D cache
1297 THUMB(		ldr     r3, [r1]		) @ s/w flush D cache
1298 THUMB(		add     r1, r1, r11		)
1299		teq	r1, r2
1300		bne	1b
1301
1302		mcr	p15, 0, r1, c7, c5, 0	@ flush I cache
1303		mcr	p15, 0, r1, c7, c6, 0	@ flush D cache
1304		mcr	p15, 0, r1, c7, c10, 4	@ drain WB
1305		mov	pc, lr
1306
1307__armv3_mmu_cache_flush:
1308__armv3_mpu_cache_flush:
1309		tst	r4, #1
1310		movne	pc, lr
1311		mov	r1, #0
1312		mcr	p15, 0, r1, c7, c0, 0	@ invalidate whole cache v3
1313		mov	pc, lr
1314
1315/*
1316 * Various debugging routines for printing hex characters and
1317 * memory, which again must be relocatable.
1318 */
1319#ifdef DEBUG
1320		.align	2
1321		.type	phexbuf,#object
1322phexbuf:	.space	12
1323		.size	phexbuf, . - phexbuf
1324
1325@ phex corrupts {r0, r1, r2, r3}
1326phex:		adr	r3, phexbuf
1327		mov	r2, #0
1328		strb	r2, [r3, r1]
13291:		subs	r1, r1, #1
1330		movmi	r0, r3
1331		bmi	puts
1332		and	r2, r0, #15
1333		mov	r0, r0, lsr #4
1334		cmp	r2, #10
1335		addge	r2, r2, #7
1336		add	r2, r2, #'0'
1337		strb	r2, [r3, r1]
1338		b	1b
1339
1340@ puts corrupts {r0, r1, r2, r3}
1341puts:		loadsp	r3, r2, r1
13421:		ldrb	r2, [r0], #1
1343		teq	r2, #0
1344		moveq	pc, lr
13452:		writeb	r2, r3
1346		mov	r1, #0x00020000
13473:		subs	r1, r1, #1
1348		bne	3b
1349		teq	r2, #'\n'
1350		moveq	r2, #'\r'
1351		beq	2b
1352		teq	r0, #0
1353		bne	1b
1354		mov	pc, lr
1355@ putc corrupts {r0, r1, r2, r3}
1356putc:
1357		mov	r2, r0
1358		loadsp	r3, r1, r0
1359		mov	r0, #0
1360		b	2b
1361
1362@ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1363memdump:	mov	r12, r0
1364		mov	r10, lr
1365		mov	r11, #0
13662:		mov	r0, r11, lsl #2
1367		add	r0, r0, r12
1368		mov	r1, #8
1369		bl	phex
1370		mov	r0, #':'
1371		bl	putc
13721:		mov	r0, #' '
1373		bl	putc
1374		ldr	r0, [r12, r11, lsl #2]
1375		mov	r1, #8
1376		bl	phex
1377		and	r0, r11, #7
1378		teq	r0, #3
1379		moveq	r0, #' '
1380		bleq	putc
1381		and	r0, r11, #7
1382		add	r11, r11, #1
1383		teq	r0, #7
1384		bne	1b
1385		mov	r0, #'\n'
1386		bl	putc
1387		cmp	r11, #64
1388		blt	2b
1389		mov	pc, r10
1390#endif
1391
1392		.ltorg
1393
1394#ifdef CONFIG_ARM_VIRT_EXT
1395.align 5
1396__hyp_reentry_vectors:
1397		W(b)	.			@ reset
1398		W(b)	.			@ undef
1399		W(b)	.			@ svc
1400		W(b)	.			@ pabort
1401		W(b)	.			@ dabort
1402		W(b)	__enter_kernel		@ hyp
1403		W(b)	.			@ irq
1404		W(b)	.			@ fiq
1405#endif /* CONFIG_ARM_VIRT_EXT */
1406
1407__enter_kernel:
1408		mov	r0, #0			@ must be 0
1409		mov	r1, r7			@ restore architecture number
1410		mov	r2, r8			@ restore atags pointer
1411 ARM(		mov	pc, r4		)	@ call kernel
1412 M_CLASS(	add	r4, r4, #1	)	@ enter in Thumb mode for M class
1413 THUMB(		bx	r4		)	@ entry point is always ARM for A/R classes
1414
1415reloc_code_end:
1416
1417#ifdef CONFIG_EFI_STUB
1418		.align	2
1419_start:		.long	start - .
1420
1421ENTRY(efi_stub_entry)
1422		@ allocate space on stack for passing current zImage address
1423		@ and for the EFI stub to return of new entry point of
1424		@ zImage, as EFI stub may copy the kernel. Pointer address
1425		@ is passed in r2. r0 and r1 are passed through from the
1426		@ EFI firmware to efi_entry
1427		adr	ip, _start
1428		ldr	r3, [ip]
1429		add	r3, r3, ip
1430		stmfd	sp!, {r3, lr}
1431		mov	r2, sp			@ pass zImage address in r2
1432		bl	efi_entry
1433
1434		@ Check for error return from EFI stub. r0 has FDT address
1435		@ or error code.
1436		cmn	r0, #1
1437		beq	efi_load_fail
1438
1439		@ Preserve return value of efi_entry() in r4
1440		mov	r4, r0
1441
1442		@ our cache maintenance code relies on CP15 barrier instructions
1443		@ but since we arrived here with the MMU and caches configured
1444		@ by UEFI, we must check that the CP15BEN bit is set in SCTLR.
1445		@ Note that this bit is RAO/WI on v6 and earlier, so the ISB in
1446		@ the enable path will be executed on v7+ only.
1447		mrc	p15, 0, r1, c1, c0, 0	@ read SCTLR
1448		tst	r1, #(1 << 5)		@ CP15BEN bit set?
1449		bne	0f
1450		orr	r1, r1, #(1 << 5)	@ CP15 barrier instructions
1451		mcr	p15, 0, r1, c1, c0, 0	@ write SCTLR
1452 ARM(		.inst	0xf57ff06f		@ v7+ isb	)
1453 THUMB(		isb						)
1454
14550:		bl	cache_clean_flush
1456		bl	cache_off
1457
1458		@ Set parameters for booting zImage according to boot protocol
1459		@ put FDT address in r2, it was returned by efi_entry()
1460		@ r1 is the machine type, and r0 needs to be 0
1461		mov	r0, #0
1462		mov	r1, #0xFFFFFFFF
1463		mov	r2, r4
1464
1465		@ Branch to (possibly) relocated zImage that is in [sp]
1466		ldr	lr, [sp]
1467		ldr	ip, =start_offset
1468		add	lr, lr, ip
1469		mov	pc, lr				@ no mode switch
1470
1471efi_load_fail:
1472		@ Return EFI_LOAD_ERROR to EFI firmware on error.
1473		ldr	r0, =0x80000001
1474		ldmfd	sp!, {ip, pc}
1475ENDPROC(efi_stub_entry)
1476#endif
1477
1478		.align
1479		.section ".stack", "aw", %nobits
1480.L_user_stack:	.space	4096
1481.L_user_stack_end:
1482