2 * linux/arch/arm/boot/compressed/head.S
4 * Copyright (C) 1996-2002 Russell King
5 * Copyright (C) 2004 Hyok S. Choi (MPU support)
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.
11 #include <linux/linkage.h>
12 #include <asm/assembler.h>
15 #include "efi-header.S"
17 AR_CLASS( .arch armv7-a )
18 M_CLASS( .arch armv7-m )
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.
29 #if defined(CONFIG_DEBUG_ICEDCC)
31 #if defined(CONFIG_CPU_V6) || defined(CONFIG_CPU_V6K) || defined(CONFIG_CPU_V7)
32 .macro loadsp, rb, tmp
35 mcr p14, 0, \ch, c0, c5, 0
37 #elif defined(CONFIG_CPU_XSCALE)
38 .macro loadsp, rb, tmp
41 mcr p14, 0, \ch, c8, c0, 0
44 .macro loadsp, rb, tmp
47 mcr p14, 0, \ch, c1, c0, 0
53 #include CONFIG_DEBUG_LL_INCLUDE
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
65 add \rb, \rb, #0x00010000 @ Ser1
69 .macro loadsp, rb, tmp
87 .macro debug_reloc_start
90 kphex r6, 8 /* processor id */
92 kphex r7, 8 /* architecture id */
93 #ifdef CONFIG_CPU_CP15
95 mrc p15, 0, r0, c1, c0
96 kphex r0, 8 /* control reg */
99 kphex r5, 8 /* decompressed kernel start */
101 kphex r9, 8 /* decompressed kernel end */
103 kphex r4, 8 /* kernel execution address */
108 .macro debug_reloc_end
110 kphex r5, 8 /* end of kernel */
113 bl memdump /* dump 256 bytes at start of kernel */
117 .section ".start", #alloc, #execinstr
119 * sort out different calling conventions
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
129 .type start,#function
135 THUMB( badr r12, 1f )
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
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
151 mov r7, r1 @ save architecture ID
152 mov r8, r2 @ save atags pointer
154 #ifndef CONFIG_CPU_V7M
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.
160 mrs r2, cpsr @ get current mode
161 tst r2, #3 @ not user?
163 mov r0, #0x17 @ angel_SWIreason_EnterSVC
164 ARM( swi 0x123456 ) @ angel_SWI_ARM
165 THUMB( svc 0xab ) @ angel_SWI_THUMB
167 safe_svcmode_maskall r0
168 msr spsr_cxsf, r9 @ Save the CPU boot mode in
172 * Note that some cache flushing and other stuff may
173 * be needed here - is there an Angel SWI call for this?
177 * some architecture specific code can be inserted
178 * by the linker here, but it should preserve r7, r8, and r9.
183 #ifdef CONFIG_AUTO_ZRELADDR
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
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
202 and r4, r4, #0xf8000000
203 /* Determine final kernel image address. */
204 add r4, r4, #TEXT_OFFSET
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.
220 orrcc r4, r4, #1 @ remember we skipped cache_on
224 ldmia r0, {r1, r2, r3, r6, r10, r11, r12}
228 * We might be running at a different address. We need
229 * to fix up various pointers.
231 sub r0, r0, r1 @ calculate the delta offset
232 add r6, r6, r0 @ _edata
233 add r10, r10, r0 @ inflated kernel size location
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.
242 orr r9, r9, lr, lsl #8
245 orr r9, r9, lr, lsl #16
246 orr r9, r9, r10, lsl #24
248 #ifndef CONFIG_ZBOOT_ROM
249 /* malloc space is above the relocated stack (64k max) */
251 add r10, sp, #0x10000
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.
261 mov r5, #0 @ init dtb size to 0
262 #ifdef CONFIG_ARM_APPENDED_DTB
267 * r4 = final kernel address (possibly with LSB set)
268 * r5 = appended dtb size (still unknown)
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
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.
284 ldr r1, =0xedfe0dd0 @ sig is 0xd00dfeed big endian
289 bne dtb_check_done @ not found
291 #ifdef CONFIG_ARM_ATAG_DTB_COMPAT
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
301 /* Get the initial DTB size */
304 /* convert to little endian */
305 eor r1, r5, r5, ror #16
306 bic r1, r1, #0x00ff0000
308 eor r5, r5, r1, lsr #8
310 /* 50% DTB growth should be good enough */
311 add r5, r5, r5, lsr #1
312 /* preserve 64-bit alignment */
315 /* clamp to 32KB min and 1MB max */
320 /* temporarily relocate the stack past the DTB work space */
323 stmfd sp!, {r0-r3, ip, lr}
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.
335 sub r0, r4, #TEXT_OFFSET
342 ldmfd sp!, {r0-r3, ip, lr}
346 mov r8, r6 @ use the appended device tree
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.
354 ldr r5, =_kernel_bss_size
355 adr r1, wont_overwrite
360 /* Get the current DTB size */
363 /* convert r5 (dtb size) to little endian */
364 eor r1, r5, r5, ror #16
365 bic r1, r1, #0x00ff0000
367 eor r5, r5, r1, lsr #8
370 /* preserve 64-bit alignment */
374 /* relocate some pointers past the appended dtb */
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
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.
395 adr r9, wont_overwrite
400 * Relocate ourselves past the end of the decompressed kernel.
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.
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.
411 add r10, r10, #((reloc_code_end - restart + 256) & ~255)
414 /* Get start of code we want to copy and align it down. */
418 /* Relocate the hyp vector base if necessary */
419 #ifdef CONFIG_ARM_VIRT_EXT
421 and r0, r0, #MODE_MASK
426 * Compute the address of the hyp vectors after relocation.
427 * This requires some arithmetic since we cannot directly
428 * reference __hyp_stub_vectors in a PC-relative way.
429 * Call __hyp_set_vectors with the new address so that we
430 * can HVC again after the copy.
433 movw r1, #:lower16:__hyp_stub_vectors - 0b
434 movt r1, #:upper16:__hyp_stub_vectors - 0b
442 sub r9, r6, r5 @ size to copy
443 add r9, r9, #31 @ rounded up to a multiple
444 bic r9, r9, #31 @ ... of 32 bytes
448 1: ldmdb r6!, {r0 - r3, r10 - r12, lr}
450 stmdb r9!, {r0 - r3, r10 - r12, lr}
453 /* Preserve offset to relocated code. */
456 #ifndef CONFIG_ZBOOT_ROM
457 /* cache_clean_flush may use the stack, so relocate it */
469 * If delta is zero, we are running at the address we were linked at.
473 * r4 = kernel execution address (possibly with LSB set)
474 * r5 = appended dtb size (0 if not present)
475 * r7 = architecture ID
487 #ifndef CONFIG_ZBOOT_ROM
489 * If we're running fully PIC === CONFIG_ZBOOT_ROM = n,
490 * we need to fix up pointers into the BSS region.
491 * Note that the stack pointer has already been fixed up.
497 * Relocate all entries in the GOT table.
498 * Bump bss entries to _edata + dtb size
500 1: ldr r1, [r11, #0] @ relocate entries in the GOT
501 add r1, r1, r0 @ This fixes up C references
502 cmp r1, r2 @ if entry >= bss_start &&
503 cmphs r3, r1 @ bss_end > entry
504 addhi r1, r1, r5 @ entry += dtb size
505 str r1, [r11], #4 @ next entry
509 /* bump our bss pointers too */
516 * Relocate entries in the GOT table. We only relocate
517 * the entries that are outside the (relocated) BSS region.
519 1: ldr r1, [r11, #0] @ relocate entries in the GOT
520 cmp r1, r2 @ entry < bss_start ||
521 cmphs r3, r1 @ _end < entry
522 addlo r1, r1, r0 @ table. This fixes up the
523 str r1, [r11], #4 @ C references.
528 not_relocated: mov r0, #0
529 1: str r0, [r2], #4 @ clear bss
537 * Did we skip the cache setup earlier?
538 * That is indicated by the LSB in r4.
546 * The C runtime environment should now be setup sufficiently.
547 * Set up some pointers, and start decompressing.
548 * r4 = kernel execution address
549 * r7 = architecture ID
553 mov r1, sp @ malloc space above stack
554 add r2, sp, #0x10000 @ 64k max
559 mov r1, r7 @ restore architecture number
560 mov r2, r8 @ restore atags pointer
562 #ifdef CONFIG_ARM_VIRT_EXT
563 mrs r0, spsr @ Get saved CPU boot mode
564 and r0, r0, #MODE_MASK
565 cmp r0, #HYP_MODE @ if not booted in HYP mode...
566 bne __enter_kernel @ boot kernel directly
568 adr r12, .L__hyp_reentry_vectors_offset
573 __HVC(0) @ otherwise bounce to hyp mode
575 b . @ should never be reached
578 .L__hyp_reentry_vectors_offset: .long __hyp_reentry_vectors - .
586 .word __bss_start @ r2
589 .word input_data_end - 4 @ r10 (inflated size location)
590 .word _got_start @ r11
592 .word .L_user_stack_end @ sp
593 .word _end - restart + 16384 + 1024*1024
596 #ifdef CONFIG_ARCH_RPC
598 params: ldr r0, =0x10000100 @ params_phys for RPC
605 * Turn on the cache. We need to setup some page tables so that we
606 * can have both the I and D caches on.
608 * We place the page tables 16k down from the kernel execution address,
609 * and we hope that nothing else is using it. If we're using it, we
613 * r4 = kernel execution address
614 * r7 = architecture number
617 * r0, r1, r2, r3, r9, r10, r12 corrupted
618 * This routine must preserve:
622 cache_on: mov r3, #8 @ cache_on function
626 * Initialize the highest priority protection region, PR7
627 * to cover all 32bit address and cacheable and bufferable.
629 __armv4_mpu_cache_on:
630 mov r0, #0x3f @ 4G, the whole
631 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
632 mcr p15, 0, r0, c6, c7, 1
635 mcr p15, 0, r0, c2, c0, 0 @ D-cache on
636 mcr p15, 0, r0, c2, c0, 1 @ I-cache on
637 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
640 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
641 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
644 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
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 mrc p15, 0, r0, c1, c0, 0 @ read control reg
648 @ ...I .... ..D. WC.M
649 orr r0, r0, #0x002d @ .... .... ..1. 11.1
650 orr r0, r0, #0x1000 @ ...1 .... .... ....
652 mcr p15, 0, r0, c1, c0, 0 @ write control reg
655 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
656 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
659 __armv3_mpu_cache_on:
660 mov r0, #0x3f @ 4G, the whole
661 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
664 mcr p15, 0, r0, c2, c0, 0 @ cache on
665 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
668 mcr p15, 0, r0, c5, c0, 0 @ access permission
671 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
673 * ?? ARMv3 MMU does not allow reading the control register,
674 * does this really work on ARMv3 MPU?
676 mrc p15, 0, r0, c1, c0, 0 @ read control reg
677 @ .... .... .... WC.M
678 orr r0, r0, #0x000d @ .... .... .... 11.1
679 /* ?? this overwrites the value constructed above? */
681 mcr p15, 0, r0, c1, c0, 0 @ write control reg
683 /* ?? invalidate for the second time? */
684 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
687 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
693 __setup_mmu: sub r3, r4, #16384 @ Page directory size
694 bic r3, r3, #0xff @ Align the pointer
697 * Initialise the page tables, turning on the cacheable and bufferable
698 * bits for the RAM area only.
702 mov r9, r9, lsl #18 @ start of RAM
703 add r10, r9, #0x10000000 @ a reasonable RAM size
704 mov r1, #0x12 @ XN|U + section mapping
705 orr r1, r1, #3 << 10 @ AP=11
707 1: cmp r1, r9 @ if virt > start of RAM
708 cmphs r10, r1 @ && end of RAM > virt
709 bic r1, r1, #0x1c @ clear XN|U + C + B
710 orrlo r1, r1, #0x10 @ Set XN|U for non-RAM
711 orrhs r1, r1, r6 @ set RAM section settings
712 str r1, [r0], #4 @ 1:1 mapping
717 * If ever we are running from Flash, then we surely want the cache
718 * to be enabled also for our execution instance... We map 2MB of it
719 * so there is no map overlap problem for up to 1 MB compressed kernel.
720 * If the execution is in RAM then we would only be duplicating the above.
722 orr r1, r6, #0x04 @ ensure B is set for this
726 orr r1, r1, r2, lsl #20
727 add r0, r3, r2, lsl #2
734 @ Enable unaligned access on v6, to allow better code generation
735 @ for the decompressor C code:
736 __armv6_mmu_cache_on:
737 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
738 bic r0, r0, #2 @ A (no unaligned access fault)
739 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
740 mcr p15, 0, r0, c1, c0, 0 @ write SCTLR
741 b __armv4_mmu_cache_on
743 __arm926ejs_mmu_cache_on:
744 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
745 mov r0, #4 @ put dcache in WT mode
746 mcr p15, 7, r0, c15, c0, 0
749 __armv4_mmu_cache_on:
752 mov r6, #CB_BITS | 0x12 @ U
755 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
756 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
757 mrc p15, 0, r0, c1, c0, 0 @ read control reg
758 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
760 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
761 bl __common_mmu_cache_on
763 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
767 __armv7_mmu_cache_on:
770 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
772 movne r6, #CB_BITS | 0x02 @ !XN
775 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
777 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
779 mrc p15, 0, r0, c1, c0, 0 @ read control reg
780 bic r0, r0, #1 << 28 @ clear SCTLR.TRE
781 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
782 orr r0, r0, #0x003c @ write buffer
783 bic r0, r0, #2 @ A (no unaligned access fault)
784 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
785 @ (needed for ARM1176)
787 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
788 mrcne p15, 0, r6, c2, c0, 2 @ read ttb control reg
789 orrne r0, r0, #1 @ MMU enabled
790 movne r1, #0xfffffffd @ domain 0 = client
791 bic r6, r6, #1 << 31 @ 32-bit translation system
792 bic r6, r6, #(7 << 0) | (1 << 4) @ use only ttbr0
793 mcrne p15, 0, r3, c2, c0, 0 @ load page table pointer
794 mcrne p15, 0, r1, c3, c0, 0 @ load domain access control
795 mcrne p15, 0, r6, c2, c0, 2 @ load ttb control
797 mcr p15, 0, r0, c7, c5, 4 @ ISB
798 mcr p15, 0, r0, c1, c0, 0 @ load control register
799 mrc p15, 0, r0, c1, c0, 0 @ and read it back
801 mcr p15, 0, r0, c7, c5, 4 @ ISB
806 mov r6, #CB_BITS | 0x12 @ U
809 mcr p15, 0, r0, c7, c7, 0 @ Invalidate whole cache
810 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
811 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
812 mrc p15, 0, r0, c1, c0, 0 @ read control reg
813 orr r0, r0, #0x1000 @ I-cache enable
814 bl __common_mmu_cache_on
816 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
819 __common_mmu_cache_on:
820 #ifndef CONFIG_THUMB2_KERNEL
822 orr r0, r0, #0x000d @ Write buffer, mmu
825 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
826 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
828 .align 5 @ cache line aligned
829 1: mcr p15, 0, r0, c1, c0, 0 @ load control register
830 mrc p15, 0, r0, c1, c0, 0 @ and read it back to
831 sub pc, lr, r0, lsr #32 @ properly flush pipeline
834 #define PROC_ENTRY_SIZE (4*5)
837 * Here follow the relocatable cache support functions for the
838 * various processors. This is a generic hook for locating an
839 * entry and jumping to an instruction at the specified offset
840 * from the start of the block. Please note this is all position
850 call_cache_fn: adr r12, proc_types
851 #ifdef CONFIG_CPU_CP15
852 mrc p15, 0, r9, c0, c0 @ get processor ID
853 #elif defined(CONFIG_CPU_V7M)
855 * On v7-M the processor id is located in the V7M_SCB_CPUID
856 * register, but as cache handling is IMPLEMENTATION DEFINED on
857 * v7-M (if existant at all) we just return early here.
858 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
859 * __armv7_mmu_cache_{on,off,flush}) would be selected which
860 * use cp15 registers that are not implemented on v7-M.
864 ldr r9, =CONFIG_PROCESSOR_ID
866 1: ldr r1, [r12, #0] @ get value
867 ldr r2, [r12, #4] @ get mask
868 eor r1, r1, r9 @ (real ^ match)
870 ARM( addeq pc, r12, r3 ) @ call cache function
871 THUMB( addeq r12, r3 )
872 THUMB( moveq pc, r12 ) @ call cache function
873 add r12, r12, #PROC_ENTRY_SIZE
877 * Table for cache operations. This is basically:
880 * - 'cache on' method instruction
881 * - 'cache off' method instruction
882 * - 'cache flush' method instruction
884 * We match an entry using: ((real_id ^ match) & mask) == 0
886 * Writethrough caches generally only need 'on' and 'off'
887 * methods. Writeback caches _must_ have the flush method
891 .type proc_types,#object
893 .word 0x41000000 @ old ARM ID
902 .word 0x41007000 @ ARM7/710
911 .word 0x41807200 @ ARM720T (writethrough)
913 W(b) __armv4_mmu_cache_on
914 W(b) __armv4_mmu_cache_off
918 .word 0x41007400 @ ARM74x
920 W(b) __armv3_mpu_cache_on
921 W(b) __armv3_mpu_cache_off
922 W(b) __armv3_mpu_cache_flush
924 .word 0x41009400 @ ARM94x
926 W(b) __armv4_mpu_cache_on
927 W(b) __armv4_mpu_cache_off
928 W(b) __armv4_mpu_cache_flush
930 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
932 W(b) __arm926ejs_mmu_cache_on
933 W(b) __armv4_mmu_cache_off
934 W(b) __armv5tej_mmu_cache_flush
936 .word 0x00007000 @ ARM7 IDs
945 @ Everything from here on will be the new ID system.
947 .word 0x4401a100 @ sa110 / sa1100
949 W(b) __armv4_mmu_cache_on
950 W(b) __armv4_mmu_cache_off
951 W(b) __armv4_mmu_cache_flush
953 .word 0x6901b110 @ sa1110
955 W(b) __armv4_mmu_cache_on
956 W(b) __armv4_mmu_cache_off
957 W(b) __armv4_mmu_cache_flush
960 .word 0xffffff00 @ PXA9xx
961 W(b) __armv4_mmu_cache_on
962 W(b) __armv4_mmu_cache_off
963 W(b) __armv4_mmu_cache_flush
965 .word 0x56158000 @ PXA168
967 W(b) __armv4_mmu_cache_on
968 W(b) __armv4_mmu_cache_off
969 W(b) __armv5tej_mmu_cache_flush
971 .word 0x56050000 @ Feroceon
973 W(b) __armv4_mmu_cache_on
974 W(b) __armv4_mmu_cache_off
975 W(b) __armv5tej_mmu_cache_flush
977 #ifdef CONFIG_CPU_FEROCEON_OLD_ID
978 /* this conflicts with the standard ARMv5TE entry */
979 .long 0x41009260 @ Old Feroceon
981 b __armv4_mmu_cache_on
982 b __armv4_mmu_cache_off
983 b __armv5tej_mmu_cache_flush
986 .word 0x66015261 @ FA526
988 W(b) __fa526_cache_on
989 W(b) __armv4_mmu_cache_off
990 W(b) __fa526_cache_flush
992 @ These match on the architecture ID
994 .word 0x00020000 @ ARMv4T
996 W(b) __armv4_mmu_cache_on
997 W(b) __armv4_mmu_cache_off
998 W(b) __armv4_mmu_cache_flush
1000 .word 0x00050000 @ ARMv5TE
1002 W(b) __armv4_mmu_cache_on
1003 W(b) __armv4_mmu_cache_off
1004 W(b) __armv4_mmu_cache_flush
1006 .word 0x00060000 @ ARMv5TEJ
1008 W(b) __armv4_mmu_cache_on
1009 W(b) __armv4_mmu_cache_off
1010 W(b) __armv5tej_mmu_cache_flush
1012 .word 0x0007b000 @ ARMv6
1014 W(b) __armv6_mmu_cache_on
1015 W(b) __armv4_mmu_cache_off
1016 W(b) __armv6_mmu_cache_flush
1018 .word 0x000f0000 @ new CPU Id
1020 W(b) __armv7_mmu_cache_on
1021 W(b) __armv7_mmu_cache_off
1022 W(b) __armv7_mmu_cache_flush
1024 .word 0 @ unrecognised type
1033 .size proc_types, . - proc_types
1036 * If you get a "non-constant expression in ".if" statement"
1037 * error from the assembler on this line, check that you have
1038 * not accidentally written a "b" instruction where you should
1039 * have written W(b).
1041 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1042 .error "The size of one or more proc_types entries is wrong."
1046 * Turn off the Cache and MMU. ARMv3 does not support
1047 * reading the control register, but ARMv4 does.
1050 * r0, r1, r2, r3, r9, r12 corrupted
1051 * This routine must preserve:
1055 cache_off: mov r3, #12 @ cache_off function
1058 __armv4_mpu_cache_off:
1059 mrc p15, 0, r0, c1, c0
1061 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
1063 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
1064 mcr p15, 0, r0, c7, c6, 0 @ flush D-Cache
1065 mcr p15, 0, r0, c7, c5, 0 @ flush I-Cache
1068 __armv3_mpu_cache_off:
1069 mrc p15, 0, r0, c1, c0
1071 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1073 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1076 __armv4_mmu_cache_off:
1078 mrc p15, 0, r0, c1, c0
1080 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1082 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1083 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1087 __armv7_mmu_cache_off:
1088 mrc p15, 0, r0, c1, c0
1094 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1096 bl __armv7_mmu_cache_flush
1099 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
1101 mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC
1102 mcr p15, 0, r0, c7, c10, 4 @ DSB
1103 mcr p15, 0, r0, c7, c5, 4 @ ISB
1107 * Clean and flush the cache to maintain consistency.
1110 * r1, r2, r3, r9, r10, r11, r12 corrupted
1111 * This routine must preserve:
1119 __armv4_mpu_cache_flush:
1124 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1125 mov r1, #7 << 5 @ 8 segments
1126 1: orr r3, r1, #63 << 26 @ 64 entries
1127 2: mcr p15, 0, r3, c7, c14, 2 @ clean & invalidate D index
1128 subs r3, r3, #1 << 26
1129 bcs 2b @ entries 63 to 0
1130 subs r1, r1, #1 << 5
1131 bcs 1b @ segments 7 to 0
1134 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1135 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1138 __fa526_cache_flush:
1142 mcr p15, 0, r1, c7, c14, 0 @ clean and invalidate D cache
1143 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1144 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1147 __armv6_mmu_cache_flush:
1150 mcreq p15, 0, r1, c7, c14, 0 @ clean+invalidate D
1151 mcr p15, 0, r1, c7, c5, 0 @ invalidate I+BTB
1152 mcreq p15, 0, r1, c7, c15, 0 @ clean+invalidate unified
1153 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1156 __armv7_mmu_cache_flush:
1159 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1160 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1163 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
1166 mcr p15, 0, r10, c7, c10, 5 @ DMB
1167 stmfd sp!, {r0-r7, r9-r11}
1168 mrc p15, 1, r0, c0, c0, 1 @ read clidr
1169 ands r3, r0, #0x7000000 @ extract loc from clidr
1170 mov r3, r3, lsr #23 @ left align loc bit field
1171 beq finished @ if loc is 0, then no need to clean
1172 mov r10, #0 @ start clean at cache level 0
1174 add r2, r10, r10, lsr #1 @ work out 3x current cache level
1175 mov r1, r0, lsr r2 @ extract cache type bits from clidr
1176 and r1, r1, #7 @ mask of the bits for current cache only
1177 cmp r1, #2 @ see what cache we have at this level
1178 blt skip @ skip if no cache, or just i-cache
1179 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1180 mcr p15, 0, r10, c7, c5, 4 @ isb to sych the new cssr&csidr
1181 mrc p15, 1, r1, c0, c0, 0 @ read the new csidr
1182 and r2, r1, #7 @ extract the length of the cache lines
1183 add r2, r2, #4 @ add 4 (line length offset)
1185 ands r4, r4, r1, lsr #3 @ find maximum number on the way size
1186 clz r5, r4 @ find bit position of way size increment
1188 ands r7, r7, r1, lsr #13 @ extract max number of the index size
1190 mov r9, r4 @ create working copy of max way size
1192 ARM( orr r11, r10, r9, lsl r5 ) @ factor way and cache number into r11
1193 ARM( orr r11, r11, r7, lsl r2 ) @ factor index number into r11
1194 THUMB( lsl r6, r9, r5 )
1195 THUMB( orr r11, r10, r6 ) @ factor way and cache number into r11
1196 THUMB( lsl r6, r7, r2 )
1197 THUMB( orr r11, r11, r6 ) @ factor index number into r11
1198 mcr p15, 0, r11, c7, c14, 2 @ clean & invalidate by set/way
1199 subs r9, r9, #1 @ decrement the way
1201 subs r7, r7, #1 @ decrement the index
1204 add r10, r10, #2 @ increment cache number
1208 ldmfd sp!, {r0-r7, r9-r11}
1209 mov r10, #0 @ switch back to cache level 0
1210 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1212 mcr p15, 0, r10, c7, c10, 4 @ DSB
1213 mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB
1214 mcr p15, 0, r10, c7, c10, 4 @ DSB
1215 mcr p15, 0, r10, c7, c5, 4 @ ISB
1218 __armv5tej_mmu_cache_flush:
1221 1: mrc p15, 0, r15, c7, c14, 3 @ test,clean,invalidate D cache
1223 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1224 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1227 __armv4_mmu_cache_flush:
1230 mov r2, #64*1024 @ default: 32K dcache size (*2)
1231 mov r11, #32 @ default: 32 byte line size
1232 mrc p15, 0, r3, c0, c0, 1 @ read cache type
1233 teq r3, r9 @ cache ID register present?
1238 mov r2, r2, lsl r1 @ base dcache size *2
1239 tst r3, #1 << 14 @ test M bit
1240 addne r2, r2, r2, lsr #1 @ +1/2 size if M == 1
1244 mov r11, r11, lsl r3 @ cache line size in bytes
1247 bic r1, r1, #63 @ align to longest cache line
1250 ARM( ldr r3, [r1], r11 ) @ s/w flush D cache
1251 THUMB( ldr r3, [r1] ) @ s/w flush D cache
1252 THUMB( add r1, r1, r11 )
1256 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1257 mcr p15, 0, r1, c7, c6, 0 @ flush D cache
1258 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1261 __armv3_mmu_cache_flush:
1262 __armv3_mpu_cache_flush:
1266 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1270 * Various debugging routines for printing hex characters and
1271 * memory, which again must be relocatable.
1275 .type phexbuf,#object
1277 .size phexbuf, . - phexbuf
1279 @ phex corrupts {r0, r1, r2, r3}
1280 phex: adr r3, phexbuf
1294 @ puts corrupts {r0, r1, r2, r3}
1296 1: ldrb r2, [r0], #1
1309 @ putc corrupts {r0, r1, r2, r3}
1316 @ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1317 memdump: mov r12, r0
1320 2: mov r0, r11, lsl #2
1328 ldr r0, [r12, r11, lsl #2]
1348 #ifdef CONFIG_ARM_VIRT_EXT
1350 __hyp_reentry_vectors:
1356 W(b) __enter_kernel @ hyp
1359 #endif /* CONFIG_ARM_VIRT_EXT */
1362 mov r0, #0 @ must be 0
1363 ARM( mov pc, r4 ) @ call kernel
1364 M_CLASS( add r4, r4, #1 ) @ enter in Thumb mode for M class
1365 THUMB( bx r4 ) @ entry point is always ARM for A/R classes
1369 #ifdef CONFIG_EFI_STUB
1371 _start: .long start - .
1373 ENTRY(efi_stub_entry)
1374 @ allocate space on stack for passing current zImage address
1375 @ and for the EFI stub to return of new entry point of
1376 @ zImage, as EFI stub may copy the kernel. Pointer address
1377 @ is passed in r2. r0 and r1 are passed through from the
1378 @ EFI firmware to efi_entry
1383 mov r2, sp @ pass zImage address in r2
1386 @ Check for error return from EFI stub. r0 has FDT address
1391 @ Preserve return value of efi_entry() in r4
1393 bl cache_clean_flush
1396 @ Set parameters for booting zImage according to boot protocol
1397 @ put FDT address in r2, it was returned by efi_entry()
1398 @ r1 is the machine type, and r0 needs to be 0
1403 @ Branch to (possibly) relocated zImage that is in [sp]
1405 ldr ip, =start_offset
1407 mov pc, lr @ no mode switch
1410 @ Return EFI_LOAD_ERROR to EFI firmware on error.
1413 ENDPROC(efi_stub_entry)
1417 .section ".stack", "aw", %nobits
1418 .L_user_stack: .space 4096