Loading...
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, tmp
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, tmp
39 .endm
40 .macro writeb, ch, rb
41 mcr p14, 0, \ch, c8, c0, 0
42 .endm
43#else
44 .macro loadsp, rb, tmp
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, tmp
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, tmp
70 addruart \rb, \tmp
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 .section ".start", #alloc, #execinstr
118/*
119 * sort out different calling conventions
120 */
121 .align
122 /*
123 * Always enter in ARM state for CPUs that support the ARM ISA.
124 * As of today (2014) that's exactly the members of the A and R
125 * classes.
126 */
127 AR_CLASS( .arm )
128start:
129 .type start,#function
130 .rept 7
131 __nop
132 .endr
133 ARM( mov r0, r0 )
134 ARM( b 1f )
135 THUMB( badr r12, 1f )
136 THUMB( bx r12 )
137
138 .word _magic_sig @ Magic numbers to help the loader
139 .word _magic_start @ absolute load/run zImage address
140 .word _magic_end @ zImage end address
141 .word 0x04030201 @ endianness flag
142
143 THUMB( .thumb )
1441: __EFI_HEADER
145
146 ARM_BE8( setend be ) @ go BE8 if compiled for BE8
147 AR_CLASS( mrs r9, cpsr )
148#ifdef CONFIG_ARM_VIRT_EXT
149 bl __hyp_stub_install @ get into SVC mode, reversibly
150#endif
151 mov r7, r1 @ save architecture ID
152 mov r8, r2 @ save atags pointer
153
154#ifndef CONFIG_CPU_V7M
155 /*
156 * Booting from Angel - need to enter SVC mode and disable
157 * FIQs/IRQs (numeric definitions from angel arm.h source).
158 * We only do this if we were in user mode on entry.
159 */
160 mrs r2, cpsr @ get current mode
161 tst r2, #3 @ not user?
162 bne not_angel
163 mov r0, #0x17 @ angel_SWIreason_EnterSVC
164 ARM( swi 0x123456 ) @ angel_SWI_ARM
165 THUMB( svc 0xab ) @ angel_SWI_THUMB
166not_angel:
167 safe_svcmode_maskall r0
168 msr spsr_cxsf, r9 @ Save the CPU boot mode in
169 @ SPSR
170#endif
171 /*
172 * Note that some cache flushing and other stuff may
173 * be needed here - is there an Angel SWI call for this?
174 */
175
176 /*
177 * some architecture specific code can be inserted
178 * by the linker here, but it should preserve r7, r8, and r9.
179 */
180
181 .text
182
183#ifdef CONFIG_AUTO_ZRELADDR
184 /*
185 * Find the start of physical memory. As we are executing
186 * without the MMU on, we are in the physical address space.
187 * We just need to get rid of any offset by aligning the
188 * address.
189 *
190 * This alignment is a balance between the requirements of
191 * different platforms - we have chosen 128MB to allow
192 * platforms which align the start of their physical memory
193 * to 128MB to use this feature, while allowing the zImage
194 * to be placed within the first 128MB of memory on other
195 * platforms. Increasing the alignment means we place
196 * stricter alignment requirements on the start of physical
197 * memory, but relaxing it means that we break people who
198 * are already placing their zImage in (eg) the top 64MB
199 * of this range.
200 */
201 mov r4, pc
202 and r4, r4, #0xf8000000
203 /* Determine final kernel image address. */
204 add r4, r4, #TEXT_OFFSET
205#else
206 ldr r4, =zreladdr
207#endif
208
209 /*
210 * Set up a page table only if it won't overwrite ourself.
211 * That means r4 < pc || r4 - 16k page directory > &_end.
212 * Given that r4 > &_end is most unfrequent, we add a rough
213 * additional 1MB of room for a possible appended DTB.
214 */
215 mov r0, pc
216 cmp r0, r4
217 ldrcc r0, LC0+32
218 addcc r0, r0, pc
219 cmpcc r4, r0
220 orrcc r4, r4, #1 @ remember we skipped cache_on
221 blcs cache_on
222
223restart: adr r0, LC0
224 ldmia r0, {r1, r2, r3, r6, r10, r11, r12}
225 ldr sp, [r0, #28]
226
227 /*
228 * We might be running at a different address. We need
229 * to fix up various pointers.
230 */
231 sub r0, r0, r1 @ calculate the delta offset
232 add r6, r6, r0 @ _edata
233 add r10, r10, r0 @ inflated kernel size location
234
235 /*
236 * The kernel build system appends the size of the
237 * decompressed kernel at the end of the compressed data
238 * in little-endian form.
239 */
240 ldrb r9, [r10, #0]
241 ldrb lr, [r10, #1]
242 orr r9, r9, lr, lsl #8
243 ldrb lr, [r10, #2]
244 ldrb r10, [r10, #3]
245 orr r9, r9, lr, lsl #16
246 orr r9, r9, r10, lsl #24
247
248#ifndef CONFIG_ZBOOT_ROM
249 /* malloc space is above the relocated stack (64k max) */
250 add sp, sp, r0
251 add r10, sp, #0x10000
252#else
253 /*
254 * With ZBOOT_ROM the bss/stack is non relocatable,
255 * but someone could still run this code from RAM,
256 * in which case our reference is _edata.
257 */
258 mov r10, r6
259#endif
260
261 mov r5, #0 @ init dtb size to 0
262#ifdef CONFIG_ARM_APPENDED_DTB
263/*
264 * r0 = delta
265 * r2 = BSS start
266 * r3 = BSS end
267 * r4 = final kernel address (possibly with LSB set)
268 * r5 = appended dtb size (still unknown)
269 * r6 = _edata
270 * r7 = architecture ID
271 * r8 = atags/device tree pointer
272 * r9 = size of decompressed image
273 * r10 = end of this image, including bss/stack/malloc space if non XIP
274 * r11 = GOT start
275 * r12 = GOT end
276 * sp = stack pointer
277 *
278 * if there are device trees (dtb) appended to zImage, advance r10 so that the
279 * dtb data will get relocated along with the kernel if necessary.
280 */
281
282 ldr lr, [r6, #0]
283#ifndef __ARMEB__
284 ldr r1, =0xedfe0dd0 @ sig is 0xd00dfeed big endian
285#else
286 ldr r1, =0xd00dfeed
287#endif
288 cmp lr, r1
289 bne dtb_check_done @ not found
290
291#ifdef CONFIG_ARM_ATAG_DTB_COMPAT
292 /*
293 * OK... Let's do some funky business here.
294 * If we do have a DTB appended to zImage, and we do have
295 * an ATAG list around, we want the later to be translated
296 * and folded into the former here. No GOT fixup has occurred
297 * yet, but none of the code we're about to call uses any
298 * global variable.
299 */
300
301 /* Get the initial DTB size */
302 ldr r5, [r6, #4]
303#ifndef __ARMEB__
304 /* convert to little endian */
305 eor r1, r5, r5, ror #16
306 bic r1, r1, #0x00ff0000
307 mov r5, r5, ror #8
308 eor r5, r5, r1, lsr #8
309#endif
310 /* 50% DTB growth should be good enough */
311 add r5, r5, r5, lsr #1
312 /* preserve 64-bit alignment */
313 add r5, r5, #7
314 bic r5, r5, #7
315 /* clamp to 32KB min and 1MB max */
316 cmp r5, #(1 << 15)
317 movlo r5, #(1 << 15)
318 cmp r5, #(1 << 20)
319 movhi r5, #(1 << 20)
320 /* temporarily relocate the stack past the DTB work space */
321 add sp, sp, r5
322
323 stmfd sp!, {r0-r3, ip, lr}
324 mov r0, r8
325 mov r1, r6
326 mov r2, r5
327 bl atags_to_fdt
328
329 /*
330 * If returned value is 1, there is no ATAG at the location
331 * pointed by r8. Try the typical 0x100 offset from start
332 * of RAM and hope for the best.
333 */
334 cmp r0, #1
335 sub r0, r4, #TEXT_OFFSET
336 bic r0, r0, #1
337 add r0, r0, #0x100
338 mov r1, r6
339 mov r2, r5
340 bleq atags_to_fdt
341
342 ldmfd sp!, {r0-r3, ip, lr}
343 sub sp, sp, r5
344#endif
345
346 mov r8, r6 @ use the appended device tree
347
348 /*
349 * Make sure that the DTB doesn't end up in the final
350 * kernel's .bss area. To do so, we adjust the decompressed
351 * kernel size to compensate if that .bss size is larger
352 * than the relocated code.
353 */
354 ldr r5, =_kernel_bss_size
355 adr r1, wont_overwrite
356 sub r1, r6, r1
357 subs r1, r5, r1
358 addhi r9, r9, r1
359
360 /* Get the current DTB size */
361 ldr r5, [r6, #4]
362#ifndef __ARMEB__
363 /* convert r5 (dtb size) to little endian */
364 eor r1, r5, r5, ror #16
365 bic r1, r1, #0x00ff0000
366 mov r5, r5, ror #8
367 eor r5, r5, r1, lsr #8
368#endif
369
370 /* preserve 64-bit alignment */
371 add r5, r5, #7
372 bic r5, r5, #7
373
374 /* relocate some pointers past the appended dtb */
375 add r6, r6, r5
376 add r10, r10, r5
377 add sp, sp, r5
378dtb_check_done:
379#endif
380
381/*
382 * Check to see if we will overwrite ourselves.
383 * r4 = final kernel address (possibly with LSB set)
384 * r9 = size of decompressed image
385 * r10 = end of this image, including bss/stack/malloc space if non XIP
386 * We basically want:
387 * r4 - 16k page directory >= r10 -> OK
388 * r4 + image length <= address of wont_overwrite -> OK
389 * Note: the possible LSB in r4 is harmless here.
390 */
391 add r10, r10, #16384
392 cmp r4, r10
393 bhs wont_overwrite
394 add r10, r4, r9
395 adr r9, wont_overwrite
396 cmp r10, r9
397 bls wont_overwrite
398
399/*
400 * Relocate ourselves past the end of the decompressed kernel.
401 * r6 = _edata
402 * r10 = end of the decompressed kernel
403 * Because we always copy ahead, we need to do it from the end and go
404 * backward in case the source and destination overlap.
405 */
406 /*
407 * Bump to the next 256-byte boundary with the size of
408 * the relocation code added. This avoids overwriting
409 * ourself when the offset is small.
410 */
411 add r10, r10, #((reloc_code_end - restart + 256) & ~255)
412 bic r10, r10, #255
413
414 /* Get start of code we want to copy and align it down. */
415 adr r5, restart
416 bic r5, r5, #31
417
418/* Relocate the hyp vector base if necessary */
419#ifdef CONFIG_ARM_VIRT_EXT
420 mrs r0, spsr
421 and r0, r0, #MODE_MASK
422 cmp r0, #HYP_MODE
423 bne 1f
424
425 bl __hyp_get_vectors
426 sub r0, r0, r5
427 add r0, r0, r10
428 bl __hyp_set_vectors
4291:
430#endif
431
432 sub r9, r6, r5 @ size to copy
433 add r9, r9, #31 @ rounded up to a multiple
434 bic r9, r9, #31 @ ... of 32 bytes
435 add r6, r9, r5
436 add r9, r9, r10
437
4381: ldmdb r6!, {r0 - r3, r10 - r12, lr}
439 cmp r6, r5
440 stmdb r9!, {r0 - r3, r10 - r12, lr}
441 bhi 1b
442
443 /* Preserve offset to relocated code. */
444 sub r6, r9, r6
445
446#ifndef CONFIG_ZBOOT_ROM
447 /* cache_clean_flush may use the stack, so relocate it */
448 add sp, sp, r6
449#endif
450
451 bl cache_clean_flush
452
453 badr r0, restart
454 add r0, r0, r6
455 mov pc, r0
456
457wont_overwrite:
458/*
459 * If delta is zero, we are running at the address we were linked at.
460 * r0 = delta
461 * r2 = BSS start
462 * r3 = BSS end
463 * r4 = kernel execution address (possibly with LSB set)
464 * r5 = appended dtb size (0 if not present)
465 * r7 = architecture ID
466 * r8 = atags pointer
467 * r11 = GOT start
468 * r12 = GOT end
469 * sp = stack pointer
470 */
471 orrs r1, r0, r5
472 beq not_relocated
473
474 add r11, r11, r0
475 add r12, r12, r0
476
477#ifndef CONFIG_ZBOOT_ROM
478 /*
479 * If we're running fully PIC === CONFIG_ZBOOT_ROM = n,
480 * we need to fix up pointers into the BSS region.
481 * Note that the stack pointer has already been fixed up.
482 */
483 add r2, r2, r0
484 add r3, r3, r0
485
486 /*
487 * Relocate all entries in the GOT table.
488 * Bump bss entries to _edata + dtb size
489 */
4901: ldr r1, [r11, #0] @ relocate entries in the GOT
491 add r1, r1, r0 @ This fixes up C references
492 cmp r1, r2 @ if entry >= bss_start &&
493 cmphs r3, r1 @ bss_end > entry
494 addhi r1, r1, r5 @ entry += dtb size
495 str r1, [r11], #4 @ next entry
496 cmp r11, r12
497 blo 1b
498
499 /* bump our bss pointers too */
500 add r2, r2, r5
501 add r3, r3, r5
502
503#else
504
505 /*
506 * Relocate entries in the GOT table. We only relocate
507 * the entries that are outside the (relocated) BSS region.
508 */
5091: ldr r1, [r11, #0] @ relocate entries in the GOT
510 cmp r1, r2 @ entry < bss_start ||
511 cmphs r3, r1 @ _end < entry
512 addlo r1, r1, r0 @ table. This fixes up the
513 str r1, [r11], #4 @ C references.
514 cmp r11, r12
515 blo 1b
516#endif
517
518not_relocated: mov r0, #0
5191: str r0, [r2], #4 @ clear bss
520 str r0, [r2], #4
521 str r0, [r2], #4
522 str r0, [r2], #4
523 cmp r2, r3
524 blo 1b
525
526 /*
527 * Did we skip the cache setup earlier?
528 * That is indicated by the LSB in r4.
529 * Do it now if so.
530 */
531 tst r4, #1
532 bic r4, r4, #1
533 blne cache_on
534
535/*
536 * The C runtime environment should now be setup sufficiently.
537 * Set up some pointers, and start decompressing.
538 * r4 = kernel execution address
539 * r7 = architecture ID
540 * r8 = atags pointer
541 */
542 mov r0, r4
543 mov r1, sp @ malloc space above stack
544 add r2, sp, #0x10000 @ 64k max
545 mov r3, r7
546 bl decompress_kernel
547 bl cache_clean_flush
548 bl cache_off
549 mov r1, r7 @ restore architecture number
550 mov r2, r8 @ restore atags pointer
551
552#ifdef CONFIG_ARM_VIRT_EXT
553 mrs r0, spsr @ Get saved CPU boot mode
554 and r0, r0, #MODE_MASK
555 cmp r0, #HYP_MODE @ if not booted in HYP mode...
556 bne __enter_kernel @ boot kernel directly
557
558 adr r12, .L__hyp_reentry_vectors_offset
559 ldr r0, [r12]
560 add r0, r0, r12
561
562 bl __hyp_set_vectors
563 __HVC(0) @ otherwise bounce to hyp mode
564
565 b . @ should never be reached
566
567 .align 2
568.L__hyp_reentry_vectors_offset: .long __hyp_reentry_vectors - .
569#else
570 b __enter_kernel
571#endif
572
573 .align 2
574 .type LC0, #object
575LC0: .word LC0 @ r1
576 .word __bss_start @ r2
577 .word _end @ r3
578 .word _edata @ r6
579 .word input_data_end - 4 @ r10 (inflated size location)
580 .word _got_start @ r11
581 .word _got_end @ ip
582 .word .L_user_stack_end @ sp
583 .word _end - restart + 16384 + 1024*1024
584 .size LC0, . - LC0
585
586#ifdef CONFIG_ARCH_RPC
587 .globl params
588params: ldr r0, =0x10000100 @ params_phys for RPC
589 mov pc, lr
590 .ltorg
591 .align
592#endif
593
594/*
595 * Turn on the cache. We need to setup some page tables so that we
596 * can have both the I and D caches on.
597 *
598 * We place the page tables 16k down from the kernel execution address,
599 * and we hope that nothing else is using it. If we're using it, we
600 * will go pop!
601 *
602 * On entry,
603 * r4 = kernel execution address
604 * r7 = architecture number
605 * r8 = atags pointer
606 * On exit,
607 * r0, r1, r2, r3, r9, r10, r12 corrupted
608 * This routine must preserve:
609 * r4, r7, r8
610 */
611 .align 5
612cache_on: mov r3, #8 @ cache_on function
613 b call_cache_fn
614
615/*
616 * Initialize the highest priority protection region, PR7
617 * to cover all 32bit address and cacheable and bufferable.
618 */
619__armv4_mpu_cache_on:
620 mov r0, #0x3f @ 4G, the whole
621 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
622 mcr p15, 0, r0, c6, c7, 1
623
624 mov r0, #0x80 @ PR7
625 mcr p15, 0, r0, c2, c0, 0 @ D-cache on
626 mcr p15, 0, r0, c2, c0, 1 @ I-cache on
627 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
628
629 mov r0, #0xc000
630 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
631 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
632
633 mov r0, #0
634 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
635 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
636 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
637 mrc p15, 0, r0, c1, c0, 0 @ read control reg
638 @ ...I .... ..D. WC.M
639 orr r0, r0, #0x002d @ .... .... ..1. 11.1
640 orr r0, r0, #0x1000 @ ...1 .... .... ....
641
642 mcr p15, 0, r0, c1, c0, 0 @ write control reg
643
644 mov r0, #0
645 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
646 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
647 mov pc, lr
648
649__armv3_mpu_cache_on:
650 mov r0, #0x3f @ 4G, the whole
651 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
652
653 mov r0, #0x80 @ PR7
654 mcr p15, 0, r0, c2, c0, 0 @ cache on
655 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
656
657 mov r0, #0xc000
658 mcr p15, 0, r0, c5, c0, 0 @ access permission
659
660 mov r0, #0
661 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
662 /*
663 * ?? ARMv3 MMU does not allow reading the control register,
664 * does this really work on ARMv3 MPU?
665 */
666 mrc p15, 0, r0, c1, c0, 0 @ read control reg
667 @ .... .... .... WC.M
668 orr r0, r0, #0x000d @ .... .... .... 11.1
669 /* ?? this overwrites the value constructed above? */
670 mov r0, #0
671 mcr p15, 0, r0, c1, c0, 0 @ write control reg
672
673 /* ?? invalidate for the second time? */
674 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
675 mov pc, lr
676
677#ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
678#define CB_BITS 0x08
679#else
680#define CB_BITS 0x0c
681#endif
682
683__setup_mmu: sub r3, r4, #16384 @ Page directory size
684 bic r3, r3, #0xff @ Align the pointer
685 bic r3, r3, #0x3f00
686/*
687 * Initialise the page tables, turning on the cacheable and bufferable
688 * bits for the RAM area only.
689 */
690 mov r0, r3
691 mov r9, r0, lsr #18
692 mov r9, r9, lsl #18 @ start of RAM
693 add r10, r9, #0x10000000 @ a reasonable RAM size
694 mov r1, #0x12 @ XN|U + section mapping
695 orr r1, r1, #3 << 10 @ AP=11
696 add r2, r3, #16384
6971: cmp r1, r9 @ if virt > start of RAM
698 cmphs r10, r1 @ && end of RAM > virt
699 bic r1, r1, #0x1c @ clear XN|U + C + B
700 orrlo r1, r1, #0x10 @ Set XN|U for non-RAM
701 orrhs r1, r1, r6 @ set RAM section settings
702 str r1, [r0], #4 @ 1:1 mapping
703 add r1, r1, #1048576
704 teq r0, r2
705 bne 1b
706/*
707 * If ever we are running from Flash, then we surely want the cache
708 * to be enabled also for our execution instance... We map 2MB of it
709 * so there is no map overlap problem for up to 1 MB compressed kernel.
710 * If the execution is in RAM then we would only be duplicating the above.
711 */
712 orr r1, r6, #0x04 @ ensure B is set for this
713 orr r1, r1, #3 << 10
714 mov r2, pc
715 mov r2, r2, lsr #20
716 orr r1, r1, r2, lsl #20
717 add r0, r3, r2, lsl #2
718 str r1, [r0], #4
719 add r1, r1, #1048576
720 str r1, [r0]
721 mov pc, lr
722ENDPROC(__setup_mmu)
723
724@ Enable unaligned access on v6, to allow better code generation
725@ for the decompressor C code:
726__armv6_mmu_cache_on:
727 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
728 bic r0, r0, #2 @ A (no unaligned access fault)
729 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
730 mcr p15, 0, r0, c1, c0, 0 @ write SCTLR
731 b __armv4_mmu_cache_on
732
733__arm926ejs_mmu_cache_on:
734#ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
735 mov r0, #4 @ put dcache in WT mode
736 mcr p15, 7, r0, c15, c0, 0
737#endif
738
739__armv4_mmu_cache_on:
740 mov r12, lr
741#ifdef CONFIG_MMU
742 mov r6, #CB_BITS | 0x12 @ U
743 bl __setup_mmu
744 mov r0, #0
745 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
746 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
747 mrc p15, 0, r0, c1, c0, 0 @ read control reg
748 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
749 orr r0, r0, #0x0030
750 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
751 bl __common_mmu_cache_on
752 mov r0, #0
753 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
754#endif
755 mov pc, r12
756
757__armv7_mmu_cache_on:
758 mov r12, lr
759#ifdef CONFIG_MMU
760 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
761 tst r11, #0xf @ VMSA
762 movne r6, #CB_BITS | 0x02 @ !XN
763 blne __setup_mmu
764 mov r0, #0
765 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
766 tst r11, #0xf @ VMSA
767 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
768#endif
769 mrc p15, 0, r0, c1, c0, 0 @ read control reg
770 bic r0, r0, #1 << 28 @ clear SCTLR.TRE
771 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
772 orr r0, r0, #0x003c @ write buffer
773 bic r0, r0, #2 @ A (no unaligned access fault)
774 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
775 @ (needed for ARM1176)
776#ifdef CONFIG_MMU
777 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
778 mrcne p15, 0, r6, c2, c0, 2 @ read ttb control reg
779 orrne r0, r0, #1 @ MMU enabled
780 movne r1, #0xfffffffd @ domain 0 = client
781 bic r6, r6, #1 << 31 @ 32-bit translation system
782 bic r6, r6, #3 << 0 @ use only ttbr0
783 mcrne p15, 0, r3, c2, c0, 0 @ load page table pointer
784 mcrne p15, 0, r1, c3, c0, 0 @ load domain access control
785 mcrne p15, 0, r6, c2, c0, 2 @ load ttb control
786#endif
787 mcr p15, 0, r0, c7, c5, 4 @ ISB
788 mcr p15, 0, r0, c1, c0, 0 @ load control register
789 mrc p15, 0, r0, c1, c0, 0 @ and read it back
790 mov r0, #0
791 mcr p15, 0, r0, c7, c5, 4 @ ISB
792 mov pc, r12
793
794__fa526_cache_on:
795 mov r12, lr
796 mov r6, #CB_BITS | 0x12 @ U
797 bl __setup_mmu
798 mov r0, #0
799 mcr p15, 0, r0, c7, c7, 0 @ Invalidate whole cache
800 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
801 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
802 mrc p15, 0, r0, c1, c0, 0 @ read control reg
803 orr r0, r0, #0x1000 @ I-cache enable
804 bl __common_mmu_cache_on
805 mov r0, #0
806 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
807 mov pc, r12
808
809__common_mmu_cache_on:
810#ifndef CONFIG_THUMB2_KERNEL
811#ifndef DEBUG
812 orr r0, r0, #0x000d @ Write buffer, mmu
813#endif
814 mov r1, #-1
815 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
816 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
817 b 1f
818 .align 5 @ cache line aligned
8191: mcr p15, 0, r0, c1, c0, 0 @ load control register
820 mrc p15, 0, r0, c1, c0, 0 @ and read it back to
821 sub pc, lr, r0, lsr #32 @ properly flush pipeline
822#endif
823
824#define PROC_ENTRY_SIZE (4*5)
825
826/*
827 * Here follow the relocatable cache support functions for the
828 * various processors. This is a generic hook for locating an
829 * entry and jumping to an instruction at the specified offset
830 * from the start of the block. Please note this is all position
831 * independent code.
832 *
833 * r1 = corrupted
834 * r2 = corrupted
835 * r3 = block offset
836 * r9 = corrupted
837 * r12 = corrupted
838 */
839
840call_cache_fn: adr r12, proc_types
841#ifdef CONFIG_CPU_CP15
842 mrc p15, 0, r9, c0, c0 @ get processor ID
843#elif defined(CONFIG_CPU_V7M)
844 /*
845 * On v7-M the processor id is located in the V7M_SCB_CPUID
846 * register, but as cache handling is IMPLEMENTATION DEFINED on
847 * v7-M (if existant at all) we just return early here.
848 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
849 * __armv7_mmu_cache_{on,off,flush}) would be selected which
850 * use cp15 registers that are not implemented on v7-M.
851 */
852 bx lr
853#else
854 ldr r9, =CONFIG_PROCESSOR_ID
855#endif
8561: ldr r1, [r12, #0] @ get value
857 ldr r2, [r12, #4] @ get mask
858 eor r1, r1, r9 @ (real ^ match)
859 tst r1, r2 @ & mask
860 ARM( addeq pc, r12, r3 ) @ call cache function
861 THUMB( addeq r12, r3 )
862 THUMB( moveq pc, r12 ) @ call cache function
863 add r12, r12, #PROC_ENTRY_SIZE
864 b 1b
865
866/*
867 * Table for cache operations. This is basically:
868 * - CPU ID match
869 * - CPU ID mask
870 * - 'cache on' method instruction
871 * - 'cache off' method instruction
872 * - 'cache flush' method instruction
873 *
874 * We match an entry using: ((real_id ^ match) & mask) == 0
875 *
876 * Writethrough caches generally only need 'on' and 'off'
877 * methods. Writeback caches _must_ have the flush method
878 * defined.
879 */
880 .align 2
881 .type proc_types,#object
882proc_types:
883 .word 0x41000000 @ old ARM ID
884 .word 0xff00f000
885 mov pc, lr
886 THUMB( nop )
887 mov pc, lr
888 THUMB( nop )
889 mov pc, lr
890 THUMB( nop )
891
892 .word 0x41007000 @ ARM7/710
893 .word 0xfff8fe00
894 mov pc, lr
895 THUMB( nop )
896 mov pc, lr
897 THUMB( nop )
898 mov pc, lr
899 THUMB( nop )
900
901 .word 0x41807200 @ ARM720T (writethrough)
902 .word 0xffffff00
903 W(b) __armv4_mmu_cache_on
904 W(b) __armv4_mmu_cache_off
905 mov pc, lr
906 THUMB( nop )
907
908 .word 0x41007400 @ ARM74x
909 .word 0xff00ff00
910 W(b) __armv3_mpu_cache_on
911 W(b) __armv3_mpu_cache_off
912 W(b) __armv3_mpu_cache_flush
913
914 .word 0x41009400 @ ARM94x
915 .word 0xff00ff00
916 W(b) __armv4_mpu_cache_on
917 W(b) __armv4_mpu_cache_off
918 W(b) __armv4_mpu_cache_flush
919
920 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
921 .word 0xff0ffff0
922 W(b) __arm926ejs_mmu_cache_on
923 W(b) __armv4_mmu_cache_off
924 W(b) __armv5tej_mmu_cache_flush
925
926 .word 0x00007000 @ ARM7 IDs
927 .word 0x0000f000
928 mov pc, lr
929 THUMB( nop )
930 mov pc, lr
931 THUMB( nop )
932 mov pc, lr
933 THUMB( nop )
934
935 @ Everything from here on will be the new ID system.
936
937 .word 0x4401a100 @ sa110 / sa1100
938 .word 0xffffffe0
939 W(b) __armv4_mmu_cache_on
940 W(b) __armv4_mmu_cache_off
941 W(b) __armv4_mmu_cache_flush
942
943 .word 0x6901b110 @ sa1110
944 .word 0xfffffff0
945 W(b) __armv4_mmu_cache_on
946 W(b) __armv4_mmu_cache_off
947 W(b) __armv4_mmu_cache_flush
948
949 .word 0x56056900
950 .word 0xffffff00 @ PXA9xx
951 W(b) __armv4_mmu_cache_on
952 W(b) __armv4_mmu_cache_off
953 W(b) __armv4_mmu_cache_flush
954
955 .word 0x56158000 @ PXA168
956 .word 0xfffff000
957 W(b) __armv4_mmu_cache_on
958 W(b) __armv4_mmu_cache_off
959 W(b) __armv5tej_mmu_cache_flush
960
961 .word 0x56050000 @ Feroceon
962 .word 0xff0f0000
963 W(b) __armv4_mmu_cache_on
964 W(b) __armv4_mmu_cache_off
965 W(b) __armv5tej_mmu_cache_flush
966
967#ifdef CONFIG_CPU_FEROCEON_OLD_ID
968 /* this conflicts with the standard ARMv5TE entry */
969 .long 0x41009260 @ Old Feroceon
970 .long 0xff00fff0
971 b __armv4_mmu_cache_on
972 b __armv4_mmu_cache_off
973 b __armv5tej_mmu_cache_flush
974#endif
975
976 .word 0x66015261 @ FA526
977 .word 0xff01fff1
978 W(b) __fa526_cache_on
979 W(b) __armv4_mmu_cache_off
980 W(b) __fa526_cache_flush
981
982 @ These match on the architecture ID
983
984 .word 0x00020000 @ ARMv4T
985 .word 0x000f0000
986 W(b) __armv4_mmu_cache_on
987 W(b) __armv4_mmu_cache_off
988 W(b) __armv4_mmu_cache_flush
989
990 .word 0x00050000 @ ARMv5TE
991 .word 0x000f0000
992 W(b) __armv4_mmu_cache_on
993 W(b) __armv4_mmu_cache_off
994 W(b) __armv4_mmu_cache_flush
995
996 .word 0x00060000 @ ARMv5TEJ
997 .word 0x000f0000
998 W(b) __armv4_mmu_cache_on
999 W(b) __armv4_mmu_cache_off
1000 W(b) __armv5tej_mmu_cache_flush
1001
1002 .word 0x0007b000 @ ARMv6
1003 .word 0x000ff000
1004 W(b) __armv6_mmu_cache_on
1005 W(b) __armv4_mmu_cache_off
1006 W(b) __armv6_mmu_cache_flush
1007
1008 .word 0x000f0000 @ new CPU Id
1009 .word 0x000f0000
1010 W(b) __armv7_mmu_cache_on
1011 W(b) __armv7_mmu_cache_off
1012 W(b) __armv7_mmu_cache_flush
1013
1014 .word 0 @ unrecognised type
1015 .word 0
1016 mov pc, lr
1017 THUMB( nop )
1018 mov pc, lr
1019 THUMB( nop )
1020 mov pc, lr
1021 THUMB( nop )
1022
1023 .size proc_types, . - proc_types
1024
1025 /*
1026 * If you get a "non-constant expression in ".if" statement"
1027 * error from the assembler on this line, check that you have
1028 * not accidentally written a "b" instruction where you should
1029 * have written W(b).
1030 */
1031 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1032 .error "The size of one or more proc_types entries is wrong."
1033 .endif
1034
1035/*
1036 * Turn off the Cache and MMU. ARMv3 does not support
1037 * reading the control register, but ARMv4 does.
1038 *
1039 * On exit,
1040 * r0, r1, r2, r3, r9, r12 corrupted
1041 * This routine must preserve:
1042 * r4, r7, r8
1043 */
1044 .align 5
1045cache_off: mov r3, #12 @ cache_off function
1046 b call_cache_fn
1047
1048__armv4_mpu_cache_off:
1049 mrc p15, 0, r0, c1, c0
1050 bic r0, r0, #0x000d
1051 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
1052 mov r0, #0
1053 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
1054 mcr p15, 0, r0, c7, c6, 0 @ flush D-Cache
1055 mcr p15, 0, r0, c7, c5, 0 @ flush I-Cache
1056 mov pc, lr
1057
1058__armv3_mpu_cache_off:
1059 mrc p15, 0, r0, c1, c0
1060 bic r0, r0, #0x000d
1061 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1062 mov r0, #0
1063 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1064 mov pc, lr
1065
1066__armv4_mmu_cache_off:
1067#ifdef CONFIG_MMU
1068 mrc p15, 0, r0, c1, c0
1069 bic r0, r0, #0x000d
1070 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1071 mov r0, #0
1072 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1073 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1074#endif
1075 mov pc, lr
1076
1077__armv7_mmu_cache_off:
1078 mrc p15, 0, r0, c1, c0
1079#ifdef CONFIG_MMU
1080 bic r0, r0, #0x000d
1081#else
1082 bic r0, r0, #0x000c
1083#endif
1084 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1085 mov r12, lr
1086 bl __armv7_mmu_cache_flush
1087 mov r0, #0
1088#ifdef CONFIG_MMU
1089 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
1090#endif
1091 mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC
1092 mcr p15, 0, r0, c7, c10, 4 @ DSB
1093 mcr p15, 0, r0, c7, c5, 4 @ ISB
1094 mov pc, r12
1095
1096/*
1097 * Clean and flush the cache to maintain consistency.
1098 *
1099 * On exit,
1100 * r1, r2, r3, r9, r10, r11, r12 corrupted
1101 * This routine must preserve:
1102 * r4, r6, r7, r8
1103 */
1104 .align 5
1105cache_clean_flush:
1106 mov r3, #16
1107 b call_cache_fn
1108
1109__armv4_mpu_cache_flush:
1110 tst r4, #1
1111 movne pc, lr
1112 mov r2, #1
1113 mov r3, #0
1114 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1115 mov r1, #7 << 5 @ 8 segments
11161: orr r3, r1, #63 << 26 @ 64 entries
11172: mcr p15, 0, r3, c7, c14, 2 @ clean & invalidate D index
1118 subs r3, r3, #1 << 26
1119 bcs 2b @ entries 63 to 0
1120 subs r1, r1, #1 << 5
1121 bcs 1b @ segments 7 to 0
1122
1123 teq r2, #0
1124 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1125 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1126 mov pc, lr
1127
1128__fa526_cache_flush:
1129 tst r4, #1
1130 movne pc, lr
1131 mov r1, #0
1132 mcr p15, 0, r1, c7, c14, 0 @ clean and invalidate D cache
1133 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1134 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1135 mov pc, lr
1136
1137__armv6_mmu_cache_flush:
1138 mov r1, #0
1139 tst r4, #1
1140 mcreq p15, 0, r1, c7, c14, 0 @ clean+invalidate D
1141 mcr p15, 0, r1, c7, c5, 0 @ invalidate I+BTB
1142 mcreq p15, 0, r1, c7, c15, 0 @ clean+invalidate unified
1143 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1144 mov pc, lr
1145
1146__armv7_mmu_cache_flush:
1147 tst r4, #1
1148 bne iflush
1149 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1150 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1151 mov r10, #0
1152 beq hierarchical
1153 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
1154 b iflush
1155hierarchical:
1156 mcr p15, 0, r10, c7, c10, 5 @ DMB
1157 stmfd sp!, {r0-r7, r9-r11}
1158 mrc p15, 1, r0, c0, c0, 1 @ read clidr
1159 ands r3, r0, #0x7000000 @ extract loc from clidr
1160 mov r3, r3, lsr #23 @ left align loc bit field
1161 beq finished @ if loc is 0, then no need to clean
1162 mov r10, #0 @ start clean at cache level 0
1163loop1:
1164 add r2, r10, r10, lsr #1 @ work out 3x current cache level
1165 mov r1, r0, lsr r2 @ extract cache type bits from clidr
1166 and r1, r1, #7 @ mask of the bits for current cache only
1167 cmp r1, #2 @ see what cache we have at this level
1168 blt skip @ skip if no cache, or just i-cache
1169 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1170 mcr p15, 0, r10, c7, c5, 4 @ isb to sych the new cssr&csidr
1171 mrc p15, 1, r1, c0, c0, 0 @ read the new csidr
1172 and r2, r1, #7 @ extract the length of the cache lines
1173 add r2, r2, #4 @ add 4 (line length offset)
1174 ldr r4, =0x3ff
1175 ands r4, r4, r1, lsr #3 @ find maximum number on the way size
1176 clz r5, r4 @ find bit position of way size increment
1177 ldr r7, =0x7fff
1178 ands r7, r7, r1, lsr #13 @ extract max number of the index size
1179loop2:
1180 mov r9, r4 @ create working copy of max way size
1181loop3:
1182 ARM( orr r11, r10, r9, lsl r5 ) @ factor way and cache number into r11
1183 ARM( orr r11, r11, r7, lsl r2 ) @ factor index number into r11
1184 THUMB( lsl r6, r9, r5 )
1185 THUMB( orr r11, r10, r6 ) @ factor way and cache number into r11
1186 THUMB( lsl r6, r7, r2 )
1187 THUMB( orr r11, r11, r6 ) @ factor index number into r11
1188 mcr p15, 0, r11, c7, c14, 2 @ clean & invalidate by set/way
1189 subs r9, r9, #1 @ decrement the way
1190 bge loop3
1191 subs r7, r7, #1 @ decrement the index
1192 bge loop2
1193skip:
1194 add r10, r10, #2 @ increment cache number
1195 cmp r3, r10
1196 bgt loop1
1197finished:
1198 ldmfd sp!, {r0-r7, r9-r11}
1199 mov r10, #0 @ swith back to cache level 0
1200 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1201iflush:
1202 mcr p15, 0, r10, c7, c10, 4 @ DSB
1203 mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB
1204 mcr p15, 0, r10, c7, c10, 4 @ DSB
1205 mcr p15, 0, r10, c7, c5, 4 @ ISB
1206 mov pc, lr
1207
1208__armv5tej_mmu_cache_flush:
1209 tst r4, #1
1210 movne pc, lr
12111: mrc p15, 0, r15, c7, c14, 3 @ test,clean,invalidate D cache
1212 bne 1b
1213 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1214 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1215 mov pc, lr
1216
1217__armv4_mmu_cache_flush:
1218 tst r4, #1
1219 movne pc, lr
1220 mov r2, #64*1024 @ default: 32K dcache size (*2)
1221 mov r11, #32 @ default: 32 byte line size
1222 mrc p15, 0, r3, c0, c0, 1 @ read cache type
1223 teq r3, r9 @ cache ID register present?
1224 beq no_cache_id
1225 mov r1, r3, lsr #18
1226 and r1, r1, #7
1227 mov r2, #1024
1228 mov r2, r2, lsl r1 @ base dcache size *2
1229 tst r3, #1 << 14 @ test M bit
1230 addne r2, r2, r2, lsr #1 @ +1/2 size if M == 1
1231 mov r3, r3, lsr #12
1232 and r3, r3, #3
1233 mov r11, #8
1234 mov r11, r11, lsl r3 @ cache line size in bytes
1235no_cache_id:
1236 mov r1, pc
1237 bic r1, r1, #63 @ align to longest cache line
1238 add r2, r1, r2
12391:
1240 ARM( ldr r3, [r1], r11 ) @ s/w flush D cache
1241 THUMB( ldr r3, [r1] ) @ s/w flush D cache
1242 THUMB( add r1, r1, r11 )
1243 teq r1, r2
1244 bne 1b
1245
1246 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1247 mcr p15, 0, r1, c7, c6, 0 @ flush D cache
1248 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1249 mov pc, lr
1250
1251__armv3_mmu_cache_flush:
1252__armv3_mpu_cache_flush:
1253 tst r4, #1
1254 movne pc, lr
1255 mov r1, #0
1256 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1257 mov pc, lr
1258
1259/*
1260 * Various debugging routines for printing hex characters and
1261 * memory, which again must be relocatable.
1262 */
1263#ifdef DEBUG
1264 .align 2
1265 .type phexbuf,#object
1266phexbuf: .space 12
1267 .size phexbuf, . - phexbuf
1268
1269@ phex corrupts {r0, r1, r2, r3}
1270phex: adr r3, phexbuf
1271 mov r2, #0
1272 strb r2, [r3, r1]
12731: subs r1, r1, #1
1274 movmi r0, r3
1275 bmi puts
1276 and r2, r0, #15
1277 mov r0, r0, lsr #4
1278 cmp r2, #10
1279 addge r2, r2, #7
1280 add r2, r2, #'0'
1281 strb r2, [r3, r1]
1282 b 1b
1283
1284@ puts corrupts {r0, r1, r2, r3}
1285puts: loadsp r3, r1
12861: ldrb r2, [r0], #1
1287 teq r2, #0
1288 moveq pc, lr
12892: writeb r2, r3
1290 mov r1, #0x00020000
12913: subs r1, r1, #1
1292 bne 3b
1293 teq r2, #'\n'
1294 moveq r2, #'\r'
1295 beq 2b
1296 teq r0, #0
1297 bne 1b
1298 mov pc, lr
1299@ putc corrupts {r0, r1, r2, r3}
1300putc:
1301 mov r2, r0
1302 mov r0, #0
1303 loadsp r3, r1
1304 b 2b
1305
1306@ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1307memdump: mov r12, r0
1308 mov r10, lr
1309 mov r11, #0
13102: mov r0, r11, lsl #2
1311 add r0, r0, r12
1312 mov r1, #8
1313 bl phex
1314 mov r0, #':'
1315 bl putc
13161: mov r0, #' '
1317 bl putc
1318 ldr r0, [r12, r11, lsl #2]
1319 mov r1, #8
1320 bl phex
1321 and r0, r11, #7
1322 teq r0, #3
1323 moveq r0, #' '
1324 bleq putc
1325 and r0, r11, #7
1326 add r11, r11, #1
1327 teq r0, #7
1328 bne 1b
1329 mov r0, #'\n'
1330 bl putc
1331 cmp r11, #64
1332 blt 2b
1333 mov pc, r10
1334#endif
1335
1336 .ltorg
1337
1338#ifdef CONFIG_ARM_VIRT_EXT
1339.align 5
1340__hyp_reentry_vectors:
1341 W(b) . @ reset
1342 W(b) . @ undef
1343 W(b) . @ svc
1344 W(b) . @ pabort
1345 W(b) . @ dabort
1346 W(b) __enter_kernel @ hyp
1347 W(b) . @ irq
1348 W(b) . @ fiq
1349#endif /* CONFIG_ARM_VIRT_EXT */
1350
1351__enter_kernel:
1352 mov r0, #0 @ must be 0
1353 ARM( mov pc, r4 ) @ call kernel
1354 M_CLASS( add r4, r4, #1 ) @ enter in Thumb mode for M class
1355 THUMB( bx r4 ) @ entry point is always ARM for A/R classes
1356
1357reloc_code_end:
1358
1359#ifdef CONFIG_EFI_STUB
1360 .align 2
1361_start: .long start - .
1362
1363ENTRY(efi_stub_entry)
1364 @ allocate space on stack for passing current zImage address
1365 @ and for the EFI stub to return of new entry point of
1366 @ zImage, as EFI stub may copy the kernel. Pointer address
1367 @ is passed in r2. r0 and r1 are passed through from the
1368 @ EFI firmware to efi_entry
1369 adr ip, _start
1370 ldr r3, [ip]
1371 add r3, r3, ip
1372 stmfd sp!, {r3, lr}
1373 mov r2, sp @ pass zImage address in r2
1374 bl efi_entry
1375
1376 @ Check for error return from EFI stub. r0 has FDT address
1377 @ or error code.
1378 cmn r0, #1
1379 beq efi_load_fail
1380
1381 @ Preserve return value of efi_entry() in r4
1382 mov r4, r0
1383 bl cache_clean_flush
1384 bl cache_off
1385
1386 @ Set parameters for booting zImage according to boot protocol
1387 @ put FDT address in r2, it was returned by efi_entry()
1388 @ r1 is the machine type, and r0 needs to be 0
1389 mov r0, #0
1390 mov r1, #0xFFFFFFFF
1391 mov r2, r4
1392
1393 @ Branch to (possibly) relocated zImage that is in [sp]
1394 ldr lr, [sp]
1395 ldr ip, =start_offset
1396 add lr, lr, ip
1397 mov pc, lr @ no mode switch
1398
1399efi_load_fail:
1400 @ Return EFI_LOAD_ERROR to EFI firmware on error.
1401 ldr r0, =0x80000001
1402 ldmfd sp!, {ip, pc}
1403ENDPROC(efi_stub_entry)
1404#endif
1405
1406 .align
1407 .section ".stack", "aw", %nobits
1408.L_user_stack: .space 4096
1409.L_user_stack_end:
1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * linux/arch/arm/boot/compressed/head.S
4 *
5 * Copyright (C) 1996-2002 Russell King
6 * Copyright (C) 2004 Hyok S. Choi (MPU support)
7 */
8#include <linux/linkage.h>
9#include <asm/assembler.h>
10#include <asm/v7m.h>
11
12#include "efi-header.S"
13
14#ifdef __ARMEB__
15#define OF_DT_MAGIC 0xd00dfeed
16#else
17#define OF_DT_MAGIC 0xedfe0dd0
18#endif
19
20 AR_CLASS( .arch armv7-a )
21 M_CLASS( .arch armv7-m )
22
23/*
24 * Debugging stuff
25 *
26 * Note that these macros must not contain any code which is not
27 * 100% relocatable. Any attempt to do so will result in a crash.
28 * Please select one of the following when turning on debugging.
29 */
30#ifdef DEBUG
31
32#if defined(CONFIG_DEBUG_ICEDCC)
33
34#if defined(CONFIG_CPU_V6) || defined(CONFIG_CPU_V6K) || defined(CONFIG_CPU_V7)
35 .macro loadsp, rb, tmp1, tmp2
36 .endm
37 .macro writeb, ch, rb, tmp
38 mcr p14, 0, \ch, c0, c5, 0
39 .endm
40#elif defined(CONFIG_CPU_XSCALE)
41 .macro loadsp, rb, tmp1, tmp2
42 .endm
43 .macro writeb, ch, rb, tmp
44 mcr p14, 0, \ch, c8, c0, 0
45 .endm
46#else
47 .macro loadsp, rb, tmp1, tmp2
48 .endm
49 .macro writeb, ch, rb, tmp
50 mcr p14, 0, \ch, c1, c0, 0
51 .endm
52#endif
53
54#else
55
56#include CONFIG_DEBUG_LL_INCLUDE
57
58 .macro writeb, ch, rb, tmp
59#ifdef CONFIG_DEBUG_UART_FLOW_CONTROL
60 waituartcts \tmp, \rb
61#endif
62 waituarttxrdy \tmp, \rb
63 senduart \ch, \rb
64 busyuart \tmp, \rb
65 .endm
66
67#if defined(CONFIG_ARCH_SA1100)
68 .macro loadsp, rb, tmp1, tmp2
69 mov \rb, #0x80000000 @ physical base address
70 add \rb, \rb, #0x00010000 @ Ser1
71 .endm
72#else
73 .macro loadsp, rb, tmp1, tmp2
74 addruart \rb, \tmp1, \tmp2
75 .endm
76#endif
77#endif
78#endif
79
80 .macro kputc,val
81 mov r0, \val
82 bl putc
83 .endm
84
85 .macro kphex,val,len
86 mov r0, \val
87 mov r1, #\len
88 bl phex
89 .endm
90
91 /*
92 * Debug kernel copy by printing the memory addresses involved
93 */
94 .macro dbgkc, begin, end, cbegin, cend
95#ifdef DEBUG
96 kputc #'C'
97 kputc #':'
98 kputc #'0'
99 kputc #'x'
100 kphex \begin, 8 /* Start of compressed kernel */
101 kputc #'-'
102 kputc #'0'
103 kputc #'x'
104 kphex \end, 8 /* End of compressed kernel */
105 kputc #'-'
106 kputc #'>'
107 kputc #'0'
108 kputc #'x'
109 kphex \cbegin, 8 /* Start of kernel copy */
110 kputc #'-'
111 kputc #'0'
112 kputc #'x'
113 kphex \cend, 8 /* End of kernel copy */
114 kputc #'\n'
115#endif
116 .endm
117
118 /*
119 * Debug print of the final appended DTB location
120 */
121 .macro dbgadtb, begin, size
122#ifdef DEBUG
123 kputc #'D'
124 kputc #'T'
125 kputc #'B'
126 kputc #':'
127 kputc #'0'
128 kputc #'x'
129 kphex \begin, 8 /* Start of appended DTB */
130 kputc #' '
131 kputc #'('
132 kputc #'0'
133 kputc #'x'
134 kphex \size, 8 /* Size of appended DTB */
135 kputc #')'
136 kputc #'\n'
137#endif
138 .endm
139
140 .macro enable_cp15_barriers, reg
141 mrc p15, 0, \reg, c1, c0, 0 @ read SCTLR
142 tst \reg, #(1 << 5) @ CP15BEN bit set?
143 bne .L_\@
144 orr \reg, \reg, #(1 << 5) @ CP15 barrier instructions
145 mcr p15, 0, \reg, c1, c0, 0 @ write SCTLR
146 ARM( .inst 0xf57ff06f @ v7+ isb )
147 THUMB( isb )
148.L_\@:
149 .endm
150
151 /*
152 * The kernel build system appends the size of the
153 * decompressed kernel at the end of the compressed data
154 * in little-endian form.
155 */
156 .macro get_inflated_image_size, res:req, tmp1:req, tmp2:req
157 adr \res, .Linflated_image_size_offset
158 ldr \tmp1, [\res]
159 add \tmp1, \tmp1, \res @ address of inflated image size
160
161 ldrb \res, [\tmp1] @ get_unaligned_le32
162 ldrb \tmp2, [\tmp1, #1]
163 orr \res, \res, \tmp2, lsl #8
164 ldrb \tmp2, [\tmp1, #2]
165 ldrb \tmp1, [\tmp1, #3]
166 orr \res, \res, \tmp2, lsl #16
167 orr \res, \res, \tmp1, lsl #24
168 .endm
169
170 .macro be32tocpu, val, tmp
171#ifndef __ARMEB__
172 /* convert to little endian */
173 rev_l \val, \tmp
174#endif
175 .endm
176
177 .section ".start", "ax"
178/*
179 * sort out different calling conventions
180 */
181 .align
182 /*
183 * Always enter in ARM state for CPUs that support the ARM ISA.
184 * As of today (2014) that's exactly the members of the A and R
185 * classes.
186 */
187 AR_CLASS( .arm )
188start:
189 .type start,#function
190 /*
191 * These 7 nops along with the 1 nop immediately below for
192 * !THUMB2 form 8 nops that make the compressed kernel bootable
193 * on legacy ARM systems that were assuming the kernel in a.out
194 * binary format. The boot loaders on these systems would
195 * jump 32 bytes into the image to skip the a.out header.
196 * with these 8 nops filling exactly 32 bytes, things still
197 * work as expected on these legacy systems. Thumb2 mode keeps
198 * 7 of the nops as it turns out that some boot loaders
199 * were patching the initial instructions of the kernel, i.e
200 * had started to exploit this "patch area".
201 */
202 __initial_nops
203 .rept 5
204 __nop
205 .endr
206#ifndef CONFIG_THUMB2_KERNEL
207 __nop
208#else
209 AR_CLASS( sub pc, pc, #3 ) @ A/R: switch to Thumb2 mode
210 M_CLASS( nop.w ) @ M: already in Thumb2 mode
211 .thumb
212#endif
213 W(b) 1f
214
215 .word _magic_sig @ Magic numbers to help the loader
216 .word _magic_start @ absolute load/run zImage address
217 .word _magic_end @ zImage end address
218 .word 0x04030201 @ endianness flag
219 .word 0x45454545 @ another magic number to indicate
220 .word _magic_table @ additional data table
221
222 __EFI_HEADER
2231:
224 ARM_BE8( setend be ) @ go BE8 if compiled for BE8
225 AR_CLASS( mrs r9, cpsr )
226#ifdef CONFIG_ARM_VIRT_EXT
227 bl __hyp_stub_install @ get into SVC mode, reversibly
228#endif
229 mov r7, r1 @ save architecture ID
230 mov r8, r2 @ save atags pointer
231
232#ifndef CONFIG_CPU_V7M
233 /*
234 * Booting from Angel - need to enter SVC mode and disable
235 * FIQs/IRQs (numeric definitions from angel arm.h source).
236 * We only do this if we were in user mode on entry.
237 */
238 mrs r2, cpsr @ get current mode
239 tst r2, #3 @ not user?
240 bne not_angel
241 mov r0, #0x17 @ angel_SWIreason_EnterSVC
242 ARM( swi 0x123456 ) @ angel_SWI_ARM
243 THUMB( svc 0xab ) @ angel_SWI_THUMB
244not_angel:
245 safe_svcmode_maskall r0
246 msr spsr_cxsf, r9 @ Save the CPU boot mode in
247 @ SPSR
248#endif
249 /*
250 * Note that some cache flushing and other stuff may
251 * be needed here - is there an Angel SWI call for this?
252 */
253
254 /*
255 * some architecture specific code can be inserted
256 * by the linker here, but it should preserve r7, r8, and r9.
257 */
258
259 .text
260
261#ifdef CONFIG_AUTO_ZRELADDR
262 /*
263 * Find the start of physical memory. As we are executing
264 * without the MMU on, we are in the physical address space.
265 * We just need to get rid of any offset by aligning the
266 * address.
267 *
268 * This alignment is a balance between the requirements of
269 * different platforms - we have chosen 128MB to allow
270 * platforms which align the start of their physical memory
271 * to 128MB to use this feature, while allowing the zImage
272 * to be placed within the first 128MB of memory on other
273 * platforms. Increasing the alignment means we place
274 * stricter alignment requirements on the start of physical
275 * memory, but relaxing it means that we break people who
276 * are already placing their zImage in (eg) the top 64MB
277 * of this range.
278 */
279 mov r0, pc
280 and r0, r0, #0xf8000000
281#ifdef CONFIG_USE_OF
282 adr r1, LC1
283#ifdef CONFIG_ARM_APPENDED_DTB
284 /*
285 * Look for an appended DTB. If found, we cannot use it to
286 * validate the calculated start of physical memory, as its
287 * memory nodes may need to be augmented by ATAGS stored at
288 * an offset from the same start of physical memory.
289 */
290 ldr r2, [r1, #4] @ get &_edata
291 add r2, r2, r1 @ relocate it
292 ldr r2, [r2] @ get DTB signature
293 ldr r3, =OF_DT_MAGIC
294 cmp r2, r3 @ do we have a DTB there?
295 beq 1f @ if yes, skip validation
296#endif /* CONFIG_ARM_APPENDED_DTB */
297
298 /*
299 * Make sure we have some stack before calling C code.
300 * No GOT fixup has occurred yet, but none of the code we're
301 * about to call uses any global variables.
302 */
303 ldr sp, [r1] @ get stack location
304 add sp, sp, r1 @ apply relocation
305
306 /* Validate calculated start against passed DTB */
307 mov r1, r8
308 bl fdt_check_mem_start
3091:
310#endif /* CONFIG_USE_OF */
311 /* Determine final kernel image address. */
312 add r4, r0, #TEXT_OFFSET
313#else
314 ldr r4, =zreladdr
315#endif
316
317 /*
318 * Set up a page table only if it won't overwrite ourself.
319 * That means r4 < pc || r4 - 16k page directory > &_end.
320 * Given that r4 > &_end is most unfrequent, we add a rough
321 * additional 1MB of room for a possible appended DTB.
322 */
323 mov r0, pc
324 cmp r0, r4
325 ldrcc r0, .Lheadroom
326 addcc r0, r0, pc
327 cmpcc r4, r0
328 orrcc r4, r4, #1 @ remember we skipped cache_on
329 blcs cache_on
330
331restart: adr r0, LC1
332 ldr sp, [r0]
333 ldr r6, [r0, #4]
334 add sp, sp, r0
335 add r6, r6, r0
336
337 get_inflated_image_size r9, r10, lr
338
339#ifndef CONFIG_ZBOOT_ROM
340 /* malloc space is above the relocated stack (64k max) */
341 add r10, sp, #MALLOC_SIZE
342#else
343 /*
344 * With ZBOOT_ROM the bss/stack is non relocatable,
345 * but someone could still run this code from RAM,
346 * in which case our reference is _edata.
347 */
348 mov r10, r6
349#endif
350
351 mov r5, #0 @ init dtb size to 0
352#ifdef CONFIG_ARM_APPENDED_DTB
353/*
354 * r4 = final kernel address (possibly with LSB set)
355 * r5 = appended dtb size (still unknown)
356 * r6 = _edata
357 * r7 = architecture ID
358 * r8 = atags/device tree pointer
359 * r9 = size of decompressed image
360 * r10 = end of this image, including bss/stack/malloc space if non XIP
361 * sp = stack pointer
362 *
363 * if there are device trees (dtb) appended to zImage, advance r10 so that the
364 * dtb data will get relocated along with the kernel if necessary.
365 */
366
367 ldr lr, [r6, #0]
368 ldr r1, =OF_DT_MAGIC
369 cmp lr, r1
370 bne dtb_check_done @ not found
371
372#ifdef CONFIG_ARM_ATAG_DTB_COMPAT
373 /*
374 * OK... Let's do some funky business here.
375 * If we do have a DTB appended to zImage, and we do have
376 * an ATAG list around, we want the later to be translated
377 * and folded into the former here. No GOT fixup has occurred
378 * yet, but none of the code we're about to call uses any
379 * global variable.
380 */
381
382 /* Get the initial DTB size */
383 ldr r5, [r6, #4]
384 be32tocpu r5, r1
385 dbgadtb r6, r5
386 /* 50% DTB growth should be good enough */
387 add r5, r5, r5, lsr #1
388 /* preserve 64-bit alignment */
389 add r5, r5, #7
390 bic r5, r5, #7
391 /* clamp to 32KB min and 1MB max */
392 cmp r5, #(1 << 15)
393 movlo r5, #(1 << 15)
394 cmp r5, #(1 << 20)
395 movhi r5, #(1 << 20)
396 /* temporarily relocate the stack past the DTB work space */
397 add sp, sp, r5
398
399 mov r0, r8
400 mov r1, r6
401 mov r2, r5
402 bl atags_to_fdt
403
404 /*
405 * If returned value is 1, there is no ATAG at the location
406 * pointed by r8. Try the typical 0x100 offset from start
407 * of RAM and hope for the best.
408 */
409 cmp r0, #1
410 sub r0, r4, #TEXT_OFFSET
411 bic r0, r0, #1
412 add r0, r0, #0x100
413 mov r1, r6
414 mov r2, r5
415 bleq atags_to_fdt
416
417 sub sp, sp, r5
418#endif
419
420 mov r8, r6 @ use the appended device tree
421
422 /*
423 * Make sure that the DTB doesn't end up in the final
424 * kernel's .bss area. To do so, we adjust the decompressed
425 * kernel size to compensate if that .bss size is larger
426 * than the relocated code.
427 */
428 ldr r5, =_kernel_bss_size
429 adr r1, wont_overwrite
430 sub r1, r6, r1
431 subs r1, r5, r1
432 addhi r9, r9, r1
433
434 /* Get the current DTB size */
435 ldr r5, [r6, #4]
436 be32tocpu r5, r1
437
438 /* preserve 64-bit alignment */
439 add r5, r5, #7
440 bic r5, r5, #7
441
442 /* relocate some pointers past the appended dtb */
443 add r6, r6, r5
444 add r10, r10, r5
445 add sp, sp, r5
446dtb_check_done:
447#endif
448
449/*
450 * Check to see if we will overwrite ourselves.
451 * r4 = final kernel address (possibly with LSB set)
452 * r9 = size of decompressed image
453 * r10 = end of this image, including bss/stack/malloc space if non XIP
454 * We basically want:
455 * r4 - 16k page directory >= r10 -> OK
456 * r4 + image length <= address of wont_overwrite -> OK
457 * Note: the possible LSB in r4 is harmless here.
458 */
459 add r10, r10, #16384
460 cmp r4, r10
461 bhs wont_overwrite
462 add r10, r4, r9
463 adr r9, wont_overwrite
464 cmp r10, r9
465 bls wont_overwrite
466
467/*
468 * Relocate ourselves past the end of the decompressed kernel.
469 * r6 = _edata
470 * r10 = end of the decompressed kernel
471 * Because we always copy ahead, we need to do it from the end and go
472 * backward in case the source and destination overlap.
473 */
474 /*
475 * Bump to the next 256-byte boundary with the size of
476 * the relocation code added. This avoids overwriting
477 * ourself when the offset is small.
478 */
479 add r10, r10, #((reloc_code_end - restart + 256) & ~255)
480 bic r10, r10, #255
481
482 /* Get start of code we want to copy and align it down. */
483 adr r5, restart
484 bic r5, r5, #31
485
486/* Relocate the hyp vector base if necessary */
487#ifdef CONFIG_ARM_VIRT_EXT
488 mrs r0, spsr
489 and r0, r0, #MODE_MASK
490 cmp r0, #HYP_MODE
491 bne 1f
492
493 /*
494 * Compute the address of the hyp vectors after relocation.
495 * Call __hyp_set_vectors with the new address so that we
496 * can HVC again after the copy.
497 */
498 adr_l r0, __hyp_stub_vectors
499 sub r0, r0, r5
500 add r0, r0, r10
501 bl __hyp_set_vectors
5021:
503#endif
504
505 sub r9, r6, r5 @ size to copy
506 add r9, r9, #31 @ rounded up to a multiple
507 bic r9, r9, #31 @ ... of 32 bytes
508 add r6, r9, r5
509 add r9, r9, r10
510
511#ifdef DEBUG
512 sub r10, r6, r5
513 sub r10, r9, r10
514 /*
515 * We are about to copy the kernel to a new memory area.
516 * The boundaries of the new memory area can be found in
517 * r10 and r9, whilst r5 and r6 contain the boundaries
518 * of the memory we are going to copy.
519 * Calling dbgkc will help with the printing of this
520 * information.
521 */
522 dbgkc r5, r6, r10, r9
523#endif
524
5251: ldmdb r6!, {r0 - r3, r10 - r12, lr}
526 cmp r6, r5
527 stmdb r9!, {r0 - r3, r10 - r12, lr}
528 bhi 1b
529
530 /* Preserve offset to relocated code. */
531 sub r6, r9, r6
532
533 mov r0, r9 @ start of relocated zImage
534 add r1, sp, r6 @ end of relocated zImage
535 bl cache_clean_flush
536
537 badr r0, restart
538 add r0, r0, r6
539 mov pc, r0
540
541wont_overwrite:
542 adr r0, LC0
543 ldmia r0, {r1, r2, r3, r11, r12}
544 sub r0, r0, r1 @ calculate the delta offset
545
546/*
547 * If delta is zero, we are running at the address we were linked at.
548 * r0 = delta
549 * r2 = BSS start
550 * r3 = BSS end
551 * r4 = kernel execution address (possibly with LSB set)
552 * r5 = appended dtb size (0 if not present)
553 * r7 = architecture ID
554 * r8 = atags pointer
555 * r11 = GOT start
556 * r12 = GOT end
557 * sp = stack pointer
558 */
559 orrs r1, r0, r5
560 beq not_relocated
561
562 add r11, r11, r0
563 add r12, r12, r0
564
565#ifndef CONFIG_ZBOOT_ROM
566 /*
567 * If we're running fully PIC === CONFIG_ZBOOT_ROM = n,
568 * we need to fix up pointers into the BSS region.
569 * Note that the stack pointer has already been fixed up.
570 */
571 add r2, r2, r0
572 add r3, r3, r0
573
574 /*
575 * Relocate all entries in the GOT table.
576 * Bump bss entries to _edata + dtb size
577 */
5781: ldr r1, [r11, #0] @ relocate entries in the GOT
579 add r1, r1, r0 @ This fixes up C references
580 cmp r1, r2 @ if entry >= bss_start &&
581 cmphs r3, r1 @ bss_end > entry
582 addhi r1, r1, r5 @ entry += dtb size
583 str r1, [r11], #4 @ next entry
584 cmp r11, r12
585 blo 1b
586
587 /* bump our bss pointers too */
588 add r2, r2, r5
589 add r3, r3, r5
590
591#else
592
593 /*
594 * Relocate entries in the GOT table. We only relocate
595 * the entries that are outside the (relocated) BSS region.
596 */
5971: ldr r1, [r11, #0] @ relocate entries in the GOT
598 cmp r1, r2 @ entry < bss_start ||
599 cmphs r3, r1 @ _end < entry
600 addlo r1, r1, r0 @ table. This fixes up the
601 str r1, [r11], #4 @ C references.
602 cmp r11, r12
603 blo 1b
604#endif
605
606not_relocated: mov r0, #0
6071: str r0, [r2], #4 @ clear bss
608 str r0, [r2], #4
609 str r0, [r2], #4
610 str r0, [r2], #4
611 cmp r2, r3
612 blo 1b
613
614 /*
615 * Did we skip the cache setup earlier?
616 * That is indicated by the LSB in r4.
617 * Do it now if so.
618 */
619 tst r4, #1
620 bic r4, r4, #1
621 blne cache_on
622
623/*
624 * The C runtime environment should now be setup sufficiently.
625 * Set up some pointers, and start decompressing.
626 * r4 = kernel execution address
627 * r7 = architecture ID
628 * r8 = atags pointer
629 */
630 mov r0, r4
631 mov r1, sp @ malloc space above stack
632 add r2, sp, #MALLOC_SIZE @ 64k max
633 mov r3, r7
634 bl decompress_kernel
635
636 get_inflated_image_size r1, r2, r3
637
638 mov r0, r4 @ start of inflated image
639 add r1, r1, r0 @ end of inflated image
640 bl cache_clean_flush
641 bl cache_off
642
643#ifdef CONFIG_ARM_VIRT_EXT
644 mrs r0, spsr @ Get saved CPU boot mode
645 and r0, r0, #MODE_MASK
646 cmp r0, #HYP_MODE @ if not booted in HYP mode...
647 bne __enter_kernel @ boot kernel directly
648
649 adr_l r0, __hyp_reentry_vectors
650 bl __hyp_set_vectors
651 __HVC(0) @ otherwise bounce to hyp mode
652
653 b . @ should never be reached
654#else
655 b __enter_kernel
656#endif
657
658 .align 2
659 .type LC0, #object
660LC0: .word LC0 @ r1
661 .word __bss_start @ r2
662 .word _end @ r3
663 .word _got_start @ r11
664 .word _got_end @ ip
665 .size LC0, . - LC0
666
667 .type LC1, #object
668LC1: .word .L_user_stack_end - LC1 @ sp
669 .word _edata - LC1 @ r6
670 .size LC1, . - LC1
671
672.Lheadroom:
673 .word _end - restart + 16384 + 1024*1024
674
675.Linflated_image_size_offset:
676 .long (input_data_end - 4) - .
677
678#ifdef CONFIG_ARCH_RPC
679 .globl params
680params: ldr r0, =0x10000100 @ params_phys for RPC
681 mov pc, lr
682 .ltorg
683 .align
684#endif
685
686/*
687 * dcache_line_size - get the minimum D-cache line size from the CTR register
688 * on ARMv7.
689 */
690 .macro dcache_line_size, reg, tmp
691#ifdef CONFIG_CPU_V7M
692 movw \tmp, #:lower16:BASEADDR_V7M_SCB + V7M_SCB_CTR
693 movt \tmp, #:upper16:BASEADDR_V7M_SCB + V7M_SCB_CTR
694 ldr \tmp, [\tmp]
695#else
696 mrc p15, 0, \tmp, c0, c0, 1 @ read ctr
697#endif
698 lsr \tmp, \tmp, #16
699 and \tmp, \tmp, #0xf @ cache line size encoding
700 mov \reg, #4 @ bytes per word
701 mov \reg, \reg, lsl \tmp @ actual cache line size
702 .endm
703
704/*
705 * Turn on the cache. We need to setup some page tables so that we
706 * can have both the I and D caches on.
707 *
708 * We place the page tables 16k down from the kernel execution address,
709 * and we hope that nothing else is using it. If we're using it, we
710 * will go pop!
711 *
712 * On entry,
713 * r4 = kernel execution address
714 * r7 = architecture number
715 * r8 = atags pointer
716 * On exit,
717 * r0, r1, r2, r3, r9, r10, r12 corrupted
718 * This routine must preserve:
719 * r4, r7, r8
720 */
721 .align 5
722cache_on: mov r3, #8 @ cache_on function
723 b call_cache_fn
724
725/*
726 * Initialize the highest priority protection region, PR7
727 * to cover all 32bit address and cacheable and bufferable.
728 */
729__armv4_mpu_cache_on:
730 mov r0, #0x3f @ 4G, the whole
731 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
732 mcr p15, 0, r0, c6, c7, 1
733
734 mov r0, #0x80 @ PR7
735 mcr p15, 0, r0, c2, c0, 0 @ D-cache on
736 mcr p15, 0, r0, c2, c0, 1 @ I-cache on
737 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
738
739 mov r0, #0xc000
740 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
741 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
742
743 mov r0, #0
744 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
745 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
746 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
747 mrc p15, 0, r0, c1, c0, 0 @ read control reg
748 @ ...I .... ..D. WC.M
749 orr r0, r0, #0x002d @ .... .... ..1. 11.1
750 orr r0, r0, #0x1000 @ ...1 .... .... ....
751
752 mcr p15, 0, r0, c1, c0, 0 @ write control reg
753
754 mov r0, #0
755 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
756 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
757 mov pc, lr
758
759__armv3_mpu_cache_on:
760 mov r0, #0x3f @ 4G, the whole
761 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
762
763 mov r0, #0x80 @ PR7
764 mcr p15, 0, r0, c2, c0, 0 @ cache on
765 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
766
767 mov r0, #0xc000
768 mcr p15, 0, r0, c5, c0, 0 @ access permission
769
770 mov r0, #0
771 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
772 /*
773 * ?? ARMv3 MMU does not allow reading the control register,
774 * does this really work on ARMv3 MPU?
775 */
776 mrc p15, 0, r0, c1, c0, 0 @ read control reg
777 @ .... .... .... WC.M
778 orr r0, r0, #0x000d @ .... .... .... 11.1
779 /* ?? this overwrites the value constructed above? */
780 mov r0, #0
781 mcr p15, 0, r0, c1, c0, 0 @ write control reg
782
783 /* ?? invalidate for the second time? */
784 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
785 mov pc, lr
786
787#ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
788#define CB_BITS 0x08
789#else
790#define CB_BITS 0x0c
791#endif
792
793__setup_mmu: sub r3, r4, #16384 @ Page directory size
794 bic r3, r3, #0xff @ Align the pointer
795 bic r3, r3, #0x3f00
796/*
797 * Initialise the page tables, turning on the cacheable and bufferable
798 * bits for the RAM area only.
799 */
800 mov r0, r3
801 mov r9, r0, lsr #18
802 mov r9, r9, lsl #18 @ start of RAM
803 add r10, r9, #0x10000000 @ a reasonable RAM size
804 mov r1, #0x12 @ XN|U + section mapping
805 orr r1, r1, #3 << 10 @ AP=11
806 add r2, r3, #16384
8071: cmp r1, r9 @ if virt > start of RAM
808 cmphs r10, r1 @ && end of RAM > virt
809 bic r1, r1, #0x1c @ clear XN|U + C + B
810 orrlo r1, r1, #0x10 @ Set XN|U for non-RAM
811 orrhs r1, r1, r6 @ set RAM section settings
812 str r1, [r0], #4 @ 1:1 mapping
813 add r1, r1, #1048576
814 teq r0, r2
815 bne 1b
816/*
817 * If ever we are running from Flash, then we surely want the cache
818 * to be enabled also for our execution instance... We map 2MB of it
819 * so there is no map overlap problem for up to 1 MB compressed kernel.
820 * If the execution is in RAM then we would only be duplicating the above.
821 */
822 orr r1, r6, #0x04 @ ensure B is set for this
823 orr r1, r1, #3 << 10
824 mov r2, pc
825 mov r2, r2, lsr #20
826 orr r1, r1, r2, lsl #20
827 add r0, r3, r2, lsl #2
828 str r1, [r0], #4
829 add r1, r1, #1048576
830 str r1, [r0]
831 mov pc, lr
832ENDPROC(__setup_mmu)
833
834@ Enable unaligned access on v6, to allow better code generation
835@ for the decompressor C code:
836__armv6_mmu_cache_on:
837 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
838 bic r0, r0, #2 @ A (no unaligned access fault)
839 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
840 mcr p15, 0, r0, c1, c0, 0 @ write SCTLR
841 b __armv4_mmu_cache_on
842
843__arm926ejs_mmu_cache_on:
844#ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
845 mov r0, #4 @ put dcache in WT mode
846 mcr p15, 7, r0, c15, c0, 0
847#endif
848
849__armv4_mmu_cache_on:
850 mov r12, lr
851#ifdef CONFIG_MMU
852 mov r6, #CB_BITS | 0x12 @ U
853 bl __setup_mmu
854 mov r0, #0
855 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
856 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
857 mrc p15, 0, r0, c1, c0, 0 @ read control reg
858 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
859 orr r0, r0, #0x0030
860 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
861 bl __common_mmu_cache_on
862 mov r0, #0
863 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
864#endif
865 mov pc, r12
866
867__armv7_mmu_cache_on:
868 enable_cp15_barriers r11
869 mov r12, lr
870#ifdef CONFIG_MMU
871 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
872 tst r11, #0xf @ VMSA
873 movne r6, #CB_BITS | 0x02 @ !XN
874 blne __setup_mmu
875 mov r0, #0
876 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
877 tst r11, #0xf @ VMSA
878 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
879#endif
880 mrc p15, 0, r0, c1, c0, 0 @ read control reg
881 bic r0, r0, #1 << 28 @ clear SCTLR.TRE
882 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
883 orr r0, r0, #0x003c @ write buffer
884 bic r0, r0, #2 @ A (no unaligned access fault)
885 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
886 @ (needed for ARM1176)
887#ifdef CONFIG_MMU
888 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
889 mrcne p15, 0, r6, c2, c0, 2 @ read ttb control reg
890 orrne r0, r0, #1 @ MMU enabled
891 movne r1, #0xfffffffd @ domain 0 = client
892 bic r6, r6, #1 << 31 @ 32-bit translation system
893 bic r6, r6, #(7 << 0) | (1 << 4) @ use only ttbr0
894 mcrne p15, 0, r3, c2, c0, 0 @ load page table pointer
895 mcrne p15, 0, r1, c3, c0, 0 @ load domain access control
896 mcrne p15, 0, r6, c2, c0, 2 @ load ttb control
897#endif
898 mcr p15, 0, r0, c7, c5, 4 @ ISB
899 mcr p15, 0, r0, c1, c0, 0 @ load control register
900 mrc p15, 0, r0, c1, c0, 0 @ and read it back
901 mov r0, #0
902 mcr p15, 0, r0, c7, c5, 4 @ ISB
903 mov pc, r12
904
905__fa526_cache_on:
906 mov r12, lr
907 mov r6, #CB_BITS | 0x12 @ U
908 bl __setup_mmu
909 mov r0, #0
910 mcr p15, 0, r0, c7, c7, 0 @ Invalidate whole cache
911 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
912 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
913 mrc p15, 0, r0, c1, c0, 0 @ read control reg
914 orr r0, r0, #0x1000 @ I-cache enable
915 bl __common_mmu_cache_on
916 mov r0, #0
917 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
918 mov pc, r12
919
920__common_mmu_cache_on:
921#ifndef CONFIG_THUMB2_KERNEL
922#ifndef DEBUG
923 orr r0, r0, #0x000d @ Write buffer, mmu
924#endif
925 mov r1, #-1
926 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
927 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
928 b 1f
929 .align 5 @ cache line aligned
9301: mcr p15, 0, r0, c1, c0, 0 @ load control register
931 mrc p15, 0, r0, c1, c0, 0 @ and read it back to
932 sub pc, lr, r0, lsr #32 @ properly flush pipeline
933#endif
934
935#define PROC_ENTRY_SIZE (4*5)
936
937/*
938 * Here follow the relocatable cache support functions for the
939 * various processors. This is a generic hook for locating an
940 * entry and jumping to an instruction at the specified offset
941 * from the start of the block. Please note this is all position
942 * independent code.
943 *
944 * r1 = corrupted
945 * r2 = corrupted
946 * r3 = block offset
947 * r9 = corrupted
948 * r12 = corrupted
949 */
950
951call_cache_fn: adr r12, proc_types
952#ifdef CONFIG_CPU_CP15
953 mrc p15, 0, r9, c0, c0 @ get processor ID
954#elif defined(CONFIG_CPU_V7M)
955 /*
956 * On v7-M the processor id is located in the V7M_SCB_CPUID
957 * register, but as cache handling is IMPLEMENTATION DEFINED on
958 * v7-M (if existant at all) we just return early here.
959 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
960 * __armv7_mmu_cache_{on,off,flush}) would be selected which
961 * use cp15 registers that are not implemented on v7-M.
962 */
963 bx lr
964#else
965 ldr r9, =CONFIG_PROCESSOR_ID
966#endif
9671: ldr r1, [r12, #0] @ get value
968 ldr r2, [r12, #4] @ get mask
969 eor r1, r1, r9 @ (real ^ match)
970 tst r1, r2 @ & mask
971 ARM( addeq pc, r12, r3 ) @ call cache function
972 THUMB( addeq r12, r3 )
973 THUMB( moveq pc, r12 ) @ call cache function
974 add r12, r12, #PROC_ENTRY_SIZE
975 b 1b
976
977/*
978 * Table for cache operations. This is basically:
979 * - CPU ID match
980 * - CPU ID mask
981 * - 'cache on' method instruction
982 * - 'cache off' method instruction
983 * - 'cache flush' method instruction
984 *
985 * We match an entry using: ((real_id ^ match) & mask) == 0
986 *
987 * Writethrough caches generally only need 'on' and 'off'
988 * methods. Writeback caches _must_ have the flush method
989 * defined.
990 */
991 .align 2
992 .type proc_types,#object
993proc_types:
994 .word 0x41000000 @ old ARM ID
995 .word 0xff00f000
996 mov pc, lr
997 THUMB( nop )
998 mov pc, lr
999 THUMB( nop )
1000 mov pc, lr
1001 THUMB( nop )
1002
1003 .word 0x41007000 @ ARM7/710
1004 .word 0xfff8fe00
1005 mov pc, lr
1006 THUMB( nop )
1007 mov pc, lr
1008 THUMB( nop )
1009 mov pc, lr
1010 THUMB( nop )
1011
1012 .word 0x41807200 @ ARM720T (writethrough)
1013 .word 0xffffff00
1014 W(b) __armv4_mmu_cache_on
1015 W(b) __armv4_mmu_cache_off
1016 mov pc, lr
1017 THUMB( nop )
1018
1019 .word 0x41007400 @ ARM74x
1020 .word 0xff00ff00
1021 W(b) __armv3_mpu_cache_on
1022 W(b) __armv3_mpu_cache_off
1023 W(b) __armv3_mpu_cache_flush
1024
1025 .word 0x41009400 @ ARM94x
1026 .word 0xff00ff00
1027 W(b) __armv4_mpu_cache_on
1028 W(b) __armv4_mpu_cache_off
1029 W(b) __armv4_mpu_cache_flush
1030
1031 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
1032 .word 0xff0ffff0
1033 W(b) __arm926ejs_mmu_cache_on
1034 W(b) __armv4_mmu_cache_off
1035 W(b) __armv5tej_mmu_cache_flush
1036
1037 .word 0x00007000 @ ARM7 IDs
1038 .word 0x0000f000
1039 mov pc, lr
1040 THUMB( nop )
1041 mov pc, lr
1042 THUMB( nop )
1043 mov pc, lr
1044 THUMB( nop )
1045
1046 @ Everything from here on will be the new ID system.
1047
1048 .word 0x4401a100 @ sa110 / sa1100
1049 .word 0xffffffe0
1050 W(b) __armv4_mmu_cache_on
1051 W(b) __armv4_mmu_cache_off
1052 W(b) __armv4_mmu_cache_flush
1053
1054 .word 0x6901b110 @ sa1110
1055 .word 0xfffffff0
1056 W(b) __armv4_mmu_cache_on
1057 W(b) __armv4_mmu_cache_off
1058 W(b) __armv4_mmu_cache_flush
1059
1060 .word 0x56056900
1061 .word 0xffffff00 @ PXA9xx
1062 W(b) __armv4_mmu_cache_on
1063 W(b) __armv4_mmu_cache_off
1064 W(b) __armv4_mmu_cache_flush
1065
1066 .word 0x56158000 @ PXA168
1067 .word 0xfffff000
1068 W(b) __armv4_mmu_cache_on
1069 W(b) __armv4_mmu_cache_off
1070 W(b) __armv5tej_mmu_cache_flush
1071
1072 .word 0x56050000 @ Feroceon
1073 .word 0xff0f0000
1074 W(b) __armv4_mmu_cache_on
1075 W(b) __armv4_mmu_cache_off
1076 W(b) __armv5tej_mmu_cache_flush
1077
1078#ifdef CONFIG_CPU_FEROCEON_OLD_ID
1079 /* this conflicts with the standard ARMv5TE entry */
1080 .long 0x41009260 @ Old Feroceon
1081 .long 0xff00fff0
1082 b __armv4_mmu_cache_on
1083 b __armv4_mmu_cache_off
1084 b __armv5tej_mmu_cache_flush
1085#endif
1086
1087 .word 0x66015261 @ FA526
1088 .word 0xff01fff1
1089 W(b) __fa526_cache_on
1090 W(b) __armv4_mmu_cache_off
1091 W(b) __fa526_cache_flush
1092
1093 @ These match on the architecture ID
1094
1095 .word 0x00020000 @ ARMv4T
1096 .word 0x000f0000
1097 W(b) __armv4_mmu_cache_on
1098 W(b) __armv4_mmu_cache_off
1099 W(b) __armv4_mmu_cache_flush
1100
1101 .word 0x00050000 @ ARMv5TE
1102 .word 0x000f0000
1103 W(b) __armv4_mmu_cache_on
1104 W(b) __armv4_mmu_cache_off
1105 W(b) __armv4_mmu_cache_flush
1106
1107 .word 0x00060000 @ ARMv5TEJ
1108 .word 0x000f0000
1109 W(b) __armv4_mmu_cache_on
1110 W(b) __armv4_mmu_cache_off
1111 W(b) __armv5tej_mmu_cache_flush
1112
1113 .word 0x0007b000 @ ARMv6
1114 .word 0x000ff000
1115 W(b) __armv6_mmu_cache_on
1116 W(b) __armv4_mmu_cache_off
1117 W(b) __armv6_mmu_cache_flush
1118
1119 .word 0x000f0000 @ new CPU Id
1120 .word 0x000f0000
1121 W(b) __armv7_mmu_cache_on
1122 W(b) __armv7_mmu_cache_off
1123 W(b) __armv7_mmu_cache_flush
1124
1125 .word 0 @ unrecognised type
1126 .word 0
1127 mov pc, lr
1128 THUMB( nop )
1129 mov pc, lr
1130 THUMB( nop )
1131 mov pc, lr
1132 THUMB( nop )
1133
1134 .size proc_types, . - proc_types
1135
1136 /*
1137 * If you get a "non-constant expression in ".if" statement"
1138 * error from the assembler on this line, check that you have
1139 * not accidentally written a "b" instruction where you should
1140 * have written W(b).
1141 */
1142 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1143 .error "The size of one or more proc_types entries is wrong."
1144 .endif
1145
1146/*
1147 * Turn off the Cache and MMU. ARMv3 does not support
1148 * reading the control register, but ARMv4 does.
1149 *
1150 * On exit,
1151 * r0, r1, r2, r3, r9, r12 corrupted
1152 * This routine must preserve:
1153 * r4, r7, r8
1154 */
1155 .align 5
1156cache_off: mov r3, #12 @ cache_off function
1157 b call_cache_fn
1158
1159__armv4_mpu_cache_off:
1160 mrc p15, 0, r0, c1, c0
1161 bic r0, r0, #0x000d
1162 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
1163 mov r0, #0
1164 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
1165 mcr p15, 0, r0, c7, c6, 0 @ flush D-Cache
1166 mcr p15, 0, r0, c7, c5, 0 @ flush I-Cache
1167 mov pc, lr
1168
1169__armv3_mpu_cache_off:
1170 mrc p15, 0, r0, c1, c0
1171 bic r0, r0, #0x000d
1172 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1173 mov r0, #0
1174 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1175 mov pc, lr
1176
1177__armv4_mmu_cache_off:
1178#ifdef CONFIG_MMU
1179 mrc p15, 0, r0, c1, c0
1180 bic r0, r0, #0x000d
1181 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1182 mov r0, #0
1183 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1184 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1185#endif
1186 mov pc, lr
1187
1188__armv7_mmu_cache_off:
1189 mrc p15, 0, r0, c1, c0
1190#ifdef CONFIG_MMU
1191 bic r0, r0, #0x0005
1192#else
1193 bic r0, r0, #0x0004
1194#endif
1195 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1196 mov r0, #0
1197#ifdef CONFIG_MMU
1198 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
1199#endif
1200 mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC
1201 mcr p15, 0, r0, c7, c10, 4 @ DSB
1202 mcr p15, 0, r0, c7, c5, 4 @ ISB
1203 mov pc, lr
1204
1205/*
1206 * Clean and flush the cache to maintain consistency.
1207 *
1208 * On entry,
1209 * r0 = start address
1210 * r1 = end address (exclusive)
1211 * On exit,
1212 * r1, r2, r3, r9, r10, r11, r12 corrupted
1213 * This routine must preserve:
1214 * r4, r6, r7, r8
1215 */
1216 .align 5
1217cache_clean_flush:
1218 mov r3, #16
1219 mov r11, r1
1220 b call_cache_fn
1221
1222__armv4_mpu_cache_flush:
1223 tst r4, #1
1224 movne pc, lr
1225 mov r2, #1
1226 mov r3, #0
1227 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1228 mov r1, #7 << 5 @ 8 segments
12291: orr r3, r1, #63 << 26 @ 64 entries
12302: mcr p15, 0, r3, c7, c14, 2 @ clean & invalidate D index
1231 subs r3, r3, #1 << 26
1232 bcs 2b @ entries 63 to 0
1233 subs r1, r1, #1 << 5
1234 bcs 1b @ segments 7 to 0
1235
1236 teq r2, #0
1237 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1238 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1239 mov pc, lr
1240
1241__fa526_cache_flush:
1242 tst r4, #1
1243 movne pc, lr
1244 mov r1, #0
1245 mcr p15, 0, r1, c7, c14, 0 @ clean and invalidate D cache
1246 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1247 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1248 mov pc, lr
1249
1250__armv6_mmu_cache_flush:
1251 mov r1, #0
1252 tst r4, #1
1253 mcreq p15, 0, r1, c7, c14, 0 @ clean+invalidate D
1254 mcr p15, 0, r1, c7, c5, 0 @ invalidate I+BTB
1255 mcreq p15, 0, r1, c7, c15, 0 @ clean+invalidate unified
1256 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1257 mov pc, lr
1258
1259__armv7_mmu_cache_flush:
1260 enable_cp15_barriers r10
1261 tst r4, #1
1262 bne iflush
1263 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1264 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1265 mov r10, #0
1266 beq hierarchical
1267 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
1268 b iflush
1269hierarchical:
1270 dcache_line_size r1, r2 @ r1 := dcache min line size
1271 sub r2, r1, #1 @ r2 := line size mask
1272 bic r0, r0, r2 @ round down start to line size
1273 sub r11, r11, #1 @ end address is exclusive
1274 bic r11, r11, r2 @ round down end to line size
12750: cmp r0, r11 @ finished?
1276 bgt iflush
1277 mcr p15, 0, r0, c7, c14, 1 @ Dcache clean/invalidate by VA
1278 add r0, r0, r1
1279 b 0b
1280iflush:
1281 mcr p15, 0, r10, c7, c10, 4 @ DSB
1282 mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB
1283 mcr p15, 0, r10, c7, c10, 4 @ DSB
1284 mcr p15, 0, r10, c7, c5, 4 @ ISB
1285 mov pc, lr
1286
1287__armv5tej_mmu_cache_flush:
1288 tst r4, #1
1289 movne pc, lr
12901: mrc p15, 0, APSR_nzcv, c7, c14, 3 @ test,clean,invalidate D cache
1291 bne 1b
1292 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1293 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1294 mov pc, lr
1295
1296__armv4_mmu_cache_flush:
1297 tst r4, #1
1298 movne pc, lr
1299 mov r2, #64*1024 @ default: 32K dcache size (*2)
1300 mov r11, #32 @ default: 32 byte line size
1301 mrc p15, 0, r3, c0, c0, 1 @ read cache type
1302 teq r3, r9 @ cache ID register present?
1303 beq no_cache_id
1304 mov r1, r3, lsr #18
1305 and r1, r1, #7
1306 mov r2, #1024
1307 mov r2, r2, lsl r1 @ base dcache size *2
1308 tst r3, #1 << 14 @ test M bit
1309 addne r2, r2, r2, lsr #1 @ +1/2 size if M == 1
1310 mov r3, r3, lsr #12
1311 and r3, r3, #3
1312 mov r11, #8
1313 mov r11, r11, lsl r3 @ cache line size in bytes
1314no_cache_id:
1315 mov r1, pc
1316 bic r1, r1, #63 @ align to longest cache line
1317 add r2, r1, r2
13181:
1319 ARM( ldr r3, [r1], r11 ) @ s/w flush D cache
1320 THUMB( ldr r3, [r1] ) @ s/w flush D cache
1321 THUMB( add r1, r1, r11 )
1322 teq r1, r2
1323 bne 1b
1324
1325 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1326 mcr p15, 0, r1, c7, c6, 0 @ flush D cache
1327 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1328 mov pc, lr
1329
1330__armv3_mmu_cache_flush:
1331__armv3_mpu_cache_flush:
1332 tst r4, #1
1333 movne pc, lr
1334 mov r1, #0
1335 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1336 mov pc, lr
1337
1338/*
1339 * Various debugging routines for printing hex characters and
1340 * memory, which again must be relocatable.
1341 */
1342#ifdef DEBUG
1343 .align 2
1344 .type phexbuf,#object
1345phexbuf: .space 12
1346 .size phexbuf, . - phexbuf
1347
1348@ phex corrupts {r0, r1, r2, r3}
1349phex: adr r3, phexbuf
1350 mov r2, #0
1351 strb r2, [r3, r1]
13521: subs r1, r1, #1
1353 movmi r0, r3
1354 bmi puts
1355 and r2, r0, #15
1356 mov r0, r0, lsr #4
1357 cmp r2, #10
1358 addge r2, r2, #7
1359 add r2, r2, #'0'
1360 strb r2, [r3, r1]
1361 b 1b
1362
1363@ puts corrupts {r0, r1, r2, r3}
1364puts: loadsp r3, r2, r1
13651: ldrb r2, [r0], #1
1366 teq r2, #0
1367 moveq pc, lr
13682: writeb r2, r3, r1
1369 mov r1, #0x00020000
13703: subs r1, r1, #1
1371 bne 3b
1372 teq r2, #'\n'
1373 moveq r2, #'\r'
1374 beq 2b
1375 teq r0, #0
1376 bne 1b
1377 mov pc, lr
1378@ putc corrupts {r0, r1, r2, r3}
1379putc:
1380 mov r2, r0
1381 loadsp r3, r1, r0
1382 mov r0, #0
1383 b 2b
1384
1385@ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1386memdump: mov r12, r0
1387 mov r10, lr
1388 mov r11, #0
13892: mov r0, r11, lsl #2
1390 add r0, r0, r12
1391 mov r1, #8
1392 bl phex
1393 mov r0, #':'
1394 bl putc
13951: mov r0, #' '
1396 bl putc
1397 ldr r0, [r12, r11, lsl #2]
1398 mov r1, #8
1399 bl phex
1400 and r0, r11, #7
1401 teq r0, #3
1402 moveq r0, #' '
1403 bleq putc
1404 and r0, r11, #7
1405 add r11, r11, #1
1406 teq r0, #7
1407 bne 1b
1408 mov r0, #'\n'
1409 bl putc
1410 cmp r11, #64
1411 blt 2b
1412 mov pc, r10
1413#endif
1414
1415 .ltorg
1416
1417#ifdef CONFIG_ARM_VIRT_EXT
1418.align 5
1419__hyp_reentry_vectors:
1420 W(b) . @ reset
1421 W(b) . @ undef
1422#ifdef CONFIG_EFI_STUB
1423 W(b) __enter_kernel_from_hyp @ hvc from HYP
1424#else
1425 W(b) . @ svc
1426#endif
1427 W(b) . @ pabort
1428 W(b) . @ dabort
1429 W(b) __enter_kernel @ hyp
1430 W(b) . @ irq
1431 W(b) . @ fiq
1432#endif /* CONFIG_ARM_VIRT_EXT */
1433
1434__enter_kernel:
1435 mov r0, #0 @ must be 0
1436 mov r1, r7 @ restore architecture number
1437 mov r2, r8 @ restore atags pointer
1438 ARM( mov pc, r4 ) @ call kernel
1439 M_CLASS( add r4, r4, #1 ) @ enter in Thumb mode for M class
1440 THUMB( bx r4 ) @ entry point is always ARM for A/R classes
1441
1442reloc_code_end:
1443
1444#ifdef CONFIG_EFI_STUB
1445__enter_kernel_from_hyp:
1446 mrc p15, 4, r0, c1, c0, 0 @ read HSCTLR
1447 bic r0, r0, #0x5 @ disable MMU and caches
1448 mcr p15, 4, r0, c1, c0, 0 @ write HSCTLR
1449 isb
1450 b __enter_kernel
1451
1452ENTRY(efi_enter_kernel)
1453 mov r4, r0 @ preserve image base
1454 mov r8, r1 @ preserve DT pointer
1455
1456 adr_l r0, call_cache_fn
1457 adr r1, 0f @ clean the region of code we
1458 bl cache_clean_flush @ may run with the MMU off
1459
1460#ifdef CONFIG_ARM_VIRT_EXT
1461 @
1462 @ The EFI spec does not support booting on ARM in HYP mode,
1463 @ since it mandates that the MMU and caches are on, with all
1464 @ 32-bit addressable DRAM mapped 1:1 using short descriptors.
1465 @
1466 @ While the EDK2 reference implementation adheres to this,
1467 @ U-Boot might decide to enter the EFI stub in HYP mode
1468 @ anyway, with the MMU and caches either on or off.
1469 @
1470 mrs r0, cpsr @ get the current mode
1471 msr spsr_cxsf, r0 @ record boot mode
1472 and r0, r0, #MODE_MASK @ are we running in HYP mode?
1473 cmp r0, #HYP_MODE
1474 bne .Lefi_svc
1475
1476 mrc p15, 4, r1, c1, c0, 0 @ read HSCTLR
1477 tst r1, #0x1 @ MMU enabled at HYP?
1478 beq 1f
1479
1480 @
1481 @ When running in HYP mode with the caches on, we're better
1482 @ off just carrying on using the cached 1:1 mapping that the
1483 @ firmware provided. Set up the HYP vectors so HVC instructions
1484 @ issued from HYP mode take us to the correct handler code. We
1485 @ will disable the MMU before jumping to the kernel proper.
1486 @
1487 ARM( bic r1, r1, #(1 << 30) ) @ clear HSCTLR.TE
1488 THUMB( orr r1, r1, #(1 << 30) ) @ set HSCTLR.TE
1489 mcr p15, 4, r1, c1, c0, 0
1490 adr r0, __hyp_reentry_vectors
1491 mcr p15, 4, r0, c12, c0, 0 @ set HYP vector base (HVBAR)
1492 isb
1493 b .Lefi_hyp
1494
1495 @
1496 @ When running in HYP mode with the caches off, we need to drop
1497 @ into SVC mode now, and let the decompressor set up its cached
1498 @ 1:1 mapping as usual.
1499 @
15001: mov r9, r4 @ preserve image base
1501 bl __hyp_stub_install @ install HYP stub vectors
1502 safe_svcmode_maskall r1 @ drop to SVC mode
1503 msr spsr_cxsf, r0 @ record boot mode
1504 orr r4, r9, #1 @ restore image base and set LSB
1505 b .Lefi_hyp
1506.Lefi_svc:
1507#endif
1508 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
1509 tst r0, #0x1 @ MMU enabled?
1510 orreq r4, r4, #1 @ set LSB if not
1511
1512.Lefi_hyp:
1513 mov r0, r8 @ DT start
1514 add r1, r8, r2 @ DT end
1515 bl cache_clean_flush
1516
1517 adr r0, 0f @ switch to our stack
1518 ldr sp, [r0]
1519 add sp, sp, r0
1520
1521 mov r5, #0 @ appended DTB size
1522 mov r7, #0xFFFFFFFF @ machine ID
1523 b wont_overwrite
1524ENDPROC(efi_enter_kernel)
15250: .long .L_user_stack_end - .
1526#endif
1527
1528 .align
1529 .section ".stack", "aw", %nobits
1530.L_user_stack: .space 4096
1531.L_user_stack_end: