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
133 #ifndef CONFIG_THUMB2_KERNEL
136 AR_CLASS( sub pc, pc, #3 ) @ A/R: switch to Thumb2 mode
137 M_CLASS( nop.w ) @ M: already in Thumb2 mode
142 .word _magic_sig @ Magic numbers to help the loader
143 .word _magic_start @ absolute load/run zImage address
144 .word _magic_end @ zImage end address
145 .word 0x04030201 @ endianness flag
146 .word 0x45454545 @ another magic number to indicate
147 .word _magic_table @ additional data table
151 ARM_BE8( setend be ) @ go BE8 if compiled for BE8
152 AR_CLASS( mrs r9, cpsr )
153 #ifdef CONFIG_ARM_VIRT_EXT
154 bl __hyp_stub_install @ get into SVC mode, reversibly
156 mov r7, r1 @ save architecture ID
157 mov r8, r2 @ save atags pointer
159 #ifndef CONFIG_CPU_V7M
161 * Booting from Angel - need to enter SVC mode and disable
162 * FIQs/IRQs (numeric definitions from angel arm.h source).
163 * We only do this if we were in user mode on entry.
165 mrs r2, cpsr @ get current mode
166 tst r2, #3 @ not user?
168 mov r0, #0x17 @ angel_SWIreason_EnterSVC
169 ARM( swi 0x123456 ) @ angel_SWI_ARM
170 THUMB( svc 0xab ) @ angel_SWI_THUMB
172 safe_svcmode_maskall r0
173 msr spsr_cxsf, r9 @ Save the CPU boot mode in
177 * Note that some cache flushing and other stuff may
178 * be needed here - is there an Angel SWI call for this?
182 * some architecture specific code can be inserted
183 * by the linker here, but it should preserve r7, r8, and r9.
188 #ifdef CONFIG_AUTO_ZRELADDR
190 * Find the start of physical memory. As we are executing
191 * without the MMU on, we are in the physical address space.
192 * We just need to get rid of any offset by aligning the
195 * This alignment is a balance between the requirements of
196 * different platforms - we have chosen 128MB to allow
197 * platforms which align the start of their physical memory
198 * to 128MB to use this feature, while allowing the zImage
199 * to be placed within the first 128MB of memory on other
200 * platforms. Increasing the alignment means we place
201 * stricter alignment requirements on the start of physical
202 * memory, but relaxing it means that we break people who
203 * are already placing their zImage in (eg) the top 64MB
207 and r4, r4, #0xf8000000
208 /* Determine final kernel image address. */
209 add r4, r4, #TEXT_OFFSET
215 * Set up a page table only if it won't overwrite ourself.
216 * That means r4 < pc || r4 - 16k page directory > &_end.
217 * Given that r4 > &_end is most unfrequent, we add a rough
218 * additional 1MB of room for a possible appended DTB.
225 orrcc r4, r4, #1 @ remember we skipped cache_on
229 ldmia r0, {r1, r2, r3, r6, r10, r11, r12}
233 * We might be running at a different address. We need
234 * to fix up various pointers.
236 sub r0, r0, r1 @ calculate the delta offset
237 add r6, r6, r0 @ _edata
238 add r10, r10, r0 @ inflated kernel size location
241 * The kernel build system appends the size of the
242 * decompressed kernel at the end of the compressed data
243 * in little-endian form.
247 orr r9, r9, lr, lsl #8
250 orr r9, r9, lr, lsl #16
251 orr r9, r9, r10, lsl #24
253 #ifndef CONFIG_ZBOOT_ROM
254 /* malloc space is above the relocated stack (64k max) */
256 add r10, sp, #0x10000
259 * With ZBOOT_ROM the bss/stack is non relocatable,
260 * but someone could still run this code from RAM,
261 * in which case our reference is _edata.
266 mov r5, #0 @ init dtb size to 0
267 #ifdef CONFIG_ARM_APPENDED_DTB
272 * r4 = final kernel address (possibly with LSB set)
273 * r5 = appended dtb size (still unknown)
275 * r7 = architecture ID
276 * r8 = atags/device tree pointer
277 * r9 = size of decompressed image
278 * r10 = end of this image, including bss/stack/malloc space if non XIP
283 * if there are device trees (dtb) appended to zImage, advance r10 so that the
284 * dtb data will get relocated along with the kernel if necessary.
289 ldr r1, =0xedfe0dd0 @ sig is 0xd00dfeed big endian
294 bne dtb_check_done @ not found
296 #ifdef CONFIG_ARM_ATAG_DTB_COMPAT
298 * OK... Let's do some funky business here.
299 * If we do have a DTB appended to zImage, and we do have
300 * an ATAG list around, we want the later to be translated
301 * and folded into the former here. No GOT fixup has occurred
302 * yet, but none of the code we're about to call uses any
306 /* Get the initial DTB size */
309 /* convert to little endian */
310 eor r1, r5, r5, ror #16
311 bic r1, r1, #0x00ff0000
313 eor r5, r5, r1, lsr #8
315 /* 50% DTB growth should be good enough */
316 add r5, r5, r5, lsr #1
317 /* preserve 64-bit alignment */
320 /* clamp to 32KB min and 1MB max */
325 /* temporarily relocate the stack past the DTB work space */
328 stmfd sp!, {r0-r3, ip, lr}
335 * If returned value is 1, there is no ATAG at the location
336 * pointed by r8. Try the typical 0x100 offset from start
337 * of RAM and hope for the best.
340 sub r0, r4, #TEXT_OFFSET
347 ldmfd sp!, {r0-r3, ip, lr}
351 mov r8, r6 @ use the appended device tree
354 * Make sure that the DTB doesn't end up in the final
355 * kernel's .bss area. To do so, we adjust the decompressed
356 * kernel size to compensate if that .bss size is larger
357 * than the relocated code.
359 ldr r5, =_kernel_bss_size
360 adr r1, wont_overwrite
365 /* Get the current DTB size */
368 /* convert r5 (dtb size) to little endian */
369 eor r1, r5, r5, ror #16
370 bic r1, r1, #0x00ff0000
372 eor r5, r5, r1, lsr #8
375 /* preserve 64-bit alignment */
379 /* relocate some pointers past the appended dtb */
387 * Check to see if we will overwrite ourselves.
388 * r4 = final kernel address (possibly with LSB set)
389 * r9 = size of decompressed image
390 * r10 = end of this image, including bss/stack/malloc space if non XIP
392 * r4 - 16k page directory >= r10 -> OK
393 * r4 + image length <= address of wont_overwrite -> OK
394 * Note: the possible LSB in r4 is harmless here.
400 adr r9, wont_overwrite
405 * Relocate ourselves past the end of the decompressed kernel.
407 * r10 = end of the decompressed kernel
408 * Because we always copy ahead, we need to do it from the end and go
409 * backward in case the source and destination overlap.
412 * Bump to the next 256-byte boundary with the size of
413 * the relocation code added. This avoids overwriting
414 * ourself when the offset is small.
416 add r10, r10, #((reloc_code_end - restart + 256) & ~255)
419 /* Get start of code we want to copy and align it down. */
423 /* Relocate the hyp vector base if necessary */
424 #ifdef CONFIG_ARM_VIRT_EXT
426 and r0, r0, #MODE_MASK
431 * Compute the address of the hyp vectors after relocation.
432 * This requires some arithmetic since we cannot directly
433 * reference __hyp_stub_vectors in a PC-relative way.
434 * Call __hyp_set_vectors with the new address so that we
435 * can HVC again after the copy.
438 movw r1, #:lower16:__hyp_stub_vectors - 0b
439 movt r1, #:upper16:__hyp_stub_vectors - 0b
447 sub r9, r6, r5 @ size to copy
448 add r9, r9, #31 @ rounded up to a multiple
449 bic r9, r9, #31 @ ... of 32 bytes
453 1: ldmdb r6!, {r0 - r3, r10 - r12, lr}
455 stmdb r9!, {r0 - r3, r10 - r12, lr}
458 /* Preserve offset to relocated code. */
461 #ifndef CONFIG_ZBOOT_ROM
462 /* cache_clean_flush may use the stack, so relocate it */
474 * If delta is zero, we are running at the address we were linked at.
478 * r4 = kernel execution address (possibly with LSB set)
479 * r5 = appended dtb size (0 if not present)
480 * r7 = architecture ID
492 #ifndef CONFIG_ZBOOT_ROM
494 * If we're running fully PIC === CONFIG_ZBOOT_ROM = n,
495 * we need to fix up pointers into the BSS region.
496 * Note that the stack pointer has already been fixed up.
502 * Relocate all entries in the GOT table.
503 * Bump bss entries to _edata + dtb size
505 1: ldr r1, [r11, #0] @ relocate entries in the GOT
506 add r1, r1, r0 @ This fixes up C references
507 cmp r1, r2 @ if entry >= bss_start &&
508 cmphs r3, r1 @ bss_end > entry
509 addhi r1, r1, r5 @ entry += dtb size
510 str r1, [r11], #4 @ next entry
514 /* bump our bss pointers too */
521 * Relocate entries in the GOT table. We only relocate
522 * the entries that are outside the (relocated) BSS region.
524 1: ldr r1, [r11, #0] @ relocate entries in the GOT
525 cmp r1, r2 @ entry < bss_start ||
526 cmphs r3, r1 @ _end < entry
527 addlo r1, r1, r0 @ table. This fixes up the
528 str r1, [r11], #4 @ C references.
533 not_relocated: mov r0, #0
534 1: str r0, [r2], #4 @ clear bss
542 * Did we skip the cache setup earlier?
543 * That is indicated by the LSB in r4.
551 * The C runtime environment should now be setup sufficiently.
552 * Set up some pointers, and start decompressing.
553 * r4 = kernel execution address
554 * r7 = architecture ID
558 mov r1, sp @ malloc space above stack
559 add r2, sp, #0x10000 @ 64k max
564 mov r1, r7 @ restore architecture number
565 mov r2, r8 @ restore atags pointer
567 #ifdef CONFIG_ARM_VIRT_EXT
568 mrs r0, spsr @ Get saved CPU boot mode
569 and r0, r0, #MODE_MASK
570 cmp r0, #HYP_MODE @ if not booted in HYP mode...
571 bne __enter_kernel @ boot kernel directly
573 adr r12, .L__hyp_reentry_vectors_offset
578 __HVC(0) @ otherwise bounce to hyp mode
580 b . @ should never be reached
583 .L__hyp_reentry_vectors_offset: .long __hyp_reentry_vectors - .
591 .word __bss_start @ r2
594 .word input_data_end - 4 @ r10 (inflated size location)
595 .word _got_start @ r11
597 .word .L_user_stack_end @ sp
598 .word _end - restart + 16384 + 1024*1024
601 #ifdef CONFIG_ARCH_RPC
603 params: ldr r0, =0x10000100 @ params_phys for RPC
610 * Turn on the cache. We need to setup some page tables so that we
611 * can have both the I and D caches on.
613 * We place the page tables 16k down from the kernel execution address,
614 * and we hope that nothing else is using it. If we're using it, we
618 * r4 = kernel execution address
619 * r7 = architecture number
622 * r0, r1, r2, r3, r9, r10, r12 corrupted
623 * This routine must preserve:
627 cache_on: mov r3, #8 @ cache_on function
631 * Initialize the highest priority protection region, PR7
632 * to cover all 32bit address and cacheable and bufferable.
634 __armv4_mpu_cache_on:
635 mov r0, #0x3f @ 4G, the whole
636 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
637 mcr p15, 0, r0, c6, c7, 1
640 mcr p15, 0, r0, c2, c0, 0 @ D-cache on
641 mcr p15, 0, r0, c2, c0, 1 @ I-cache on
642 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
645 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
646 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
649 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
650 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
651 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
652 mrc p15, 0, r0, c1, c0, 0 @ read control reg
653 @ ...I .... ..D. WC.M
654 orr r0, r0, #0x002d @ .... .... ..1. 11.1
655 orr r0, r0, #0x1000 @ ...1 .... .... ....
657 mcr p15, 0, r0, c1, c0, 0 @ write control reg
660 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
661 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
664 __armv3_mpu_cache_on:
665 mov r0, #0x3f @ 4G, the whole
666 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
669 mcr p15, 0, r0, c2, c0, 0 @ cache on
670 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
673 mcr p15, 0, r0, c5, c0, 0 @ access permission
676 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
678 * ?? ARMv3 MMU does not allow reading the control register,
679 * does this really work on ARMv3 MPU?
681 mrc p15, 0, r0, c1, c0, 0 @ read control reg
682 @ .... .... .... WC.M
683 orr r0, r0, #0x000d @ .... .... .... 11.1
684 /* ?? this overwrites the value constructed above? */
686 mcr p15, 0, r0, c1, c0, 0 @ write control reg
688 /* ?? invalidate for the second time? */
689 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
692 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
698 __setup_mmu: sub r3, r4, #16384 @ Page directory size
699 bic r3, r3, #0xff @ Align the pointer
702 * Initialise the page tables, turning on the cacheable and bufferable
703 * bits for the RAM area only.
707 mov r9, r9, lsl #18 @ start of RAM
708 add r10, r9, #0x10000000 @ a reasonable RAM size
709 mov r1, #0x12 @ XN|U + section mapping
710 orr r1, r1, #3 << 10 @ AP=11
712 1: cmp r1, r9 @ if virt > start of RAM
713 cmphs r10, r1 @ && end of RAM > virt
714 bic r1, r1, #0x1c @ clear XN|U + C + B
715 orrlo r1, r1, #0x10 @ Set XN|U for non-RAM
716 orrhs r1, r1, r6 @ set RAM section settings
717 str r1, [r0], #4 @ 1:1 mapping
722 * If ever we are running from Flash, then we surely want the cache
723 * to be enabled also for our execution instance... We map 2MB of it
724 * so there is no map overlap problem for up to 1 MB compressed kernel.
725 * If the execution is in RAM then we would only be duplicating the above.
727 orr r1, r6, #0x04 @ ensure B is set for this
731 orr r1, r1, r2, lsl #20
732 add r0, r3, r2, lsl #2
739 @ Enable unaligned access on v6, to allow better code generation
740 @ for the decompressor C code:
741 __armv6_mmu_cache_on:
742 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
743 bic r0, r0, #2 @ A (no unaligned access fault)
744 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
745 mcr p15, 0, r0, c1, c0, 0 @ write SCTLR
746 b __armv4_mmu_cache_on
748 __arm926ejs_mmu_cache_on:
749 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
750 mov r0, #4 @ put dcache in WT mode
751 mcr p15, 7, r0, c15, c0, 0
754 __armv4_mmu_cache_on:
757 mov r6, #CB_BITS | 0x12 @ U
760 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
761 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
762 mrc p15, 0, r0, c1, c0, 0 @ read control reg
763 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
765 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
766 bl __common_mmu_cache_on
768 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
772 __armv7_mmu_cache_on:
775 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
777 movne r6, #CB_BITS | 0x02 @ !XN
780 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
782 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
784 mrc p15, 0, r0, c1, c0, 0 @ read control reg
785 bic r0, r0, #1 << 28 @ clear SCTLR.TRE
786 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
787 orr r0, r0, #0x003c @ write buffer
788 bic r0, r0, #2 @ A (no unaligned access fault)
789 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
790 @ (needed for ARM1176)
792 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
793 mrcne p15, 0, r6, c2, c0, 2 @ read ttb control reg
794 orrne r0, r0, #1 @ MMU enabled
795 movne r1, #0xfffffffd @ domain 0 = client
796 bic r6, r6, #1 << 31 @ 32-bit translation system
797 bic r6, r6, #(7 << 0) | (1 << 4) @ use only ttbr0
798 mcrne p15, 0, r3, c2, c0, 0 @ load page table pointer
799 mcrne p15, 0, r1, c3, c0, 0 @ load domain access control
800 mcrne p15, 0, r6, c2, c0, 2 @ load ttb control
802 mcr p15, 0, r0, c7, c5, 4 @ ISB
803 mcr p15, 0, r0, c1, c0, 0 @ load control register
804 mrc p15, 0, r0, c1, c0, 0 @ and read it back
806 mcr p15, 0, r0, c7, c5, 4 @ ISB
811 mov r6, #CB_BITS | 0x12 @ U
814 mcr p15, 0, r0, c7, c7, 0 @ Invalidate whole cache
815 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
816 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
817 mrc p15, 0, r0, c1, c0, 0 @ read control reg
818 orr r0, r0, #0x1000 @ I-cache enable
819 bl __common_mmu_cache_on
821 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
824 __common_mmu_cache_on:
825 #ifndef CONFIG_THUMB2_KERNEL
827 orr r0, r0, #0x000d @ Write buffer, mmu
830 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
831 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
833 .align 5 @ cache line aligned
834 1: mcr p15, 0, r0, c1, c0, 0 @ load control register
835 mrc p15, 0, r0, c1, c0, 0 @ and read it back to
836 sub pc, lr, r0, lsr #32 @ properly flush pipeline
839 #define PROC_ENTRY_SIZE (4*5)
842 * Here follow the relocatable cache support functions for the
843 * various processors. This is a generic hook for locating an
844 * entry and jumping to an instruction at the specified offset
845 * from the start of the block. Please note this is all position
855 call_cache_fn: adr r12, proc_types
856 #ifdef CONFIG_CPU_CP15
857 mrc p15, 0, r9, c0, c0 @ get processor ID
858 #elif defined(CONFIG_CPU_V7M)
860 * On v7-M the processor id is located in the V7M_SCB_CPUID
861 * register, but as cache handling is IMPLEMENTATION DEFINED on
862 * v7-M (if existant at all) we just return early here.
863 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
864 * __armv7_mmu_cache_{on,off,flush}) would be selected which
865 * use cp15 registers that are not implemented on v7-M.
869 ldr r9, =CONFIG_PROCESSOR_ID
871 1: ldr r1, [r12, #0] @ get value
872 ldr r2, [r12, #4] @ get mask
873 eor r1, r1, r9 @ (real ^ match)
875 ARM( addeq pc, r12, r3 ) @ call cache function
876 THUMB( addeq r12, r3 )
877 THUMB( moveq pc, r12 ) @ call cache function
878 add r12, r12, #PROC_ENTRY_SIZE
882 * Table for cache operations. This is basically:
885 * - 'cache on' method instruction
886 * - 'cache off' method instruction
887 * - 'cache flush' method instruction
889 * We match an entry using: ((real_id ^ match) & mask) == 0
891 * Writethrough caches generally only need 'on' and 'off'
892 * methods. Writeback caches _must_ have the flush method
896 .type proc_types,#object
898 .word 0x41000000 @ old ARM ID
907 .word 0x41007000 @ ARM7/710
916 .word 0x41807200 @ ARM720T (writethrough)
918 W(b) __armv4_mmu_cache_on
919 W(b) __armv4_mmu_cache_off
923 .word 0x41007400 @ ARM74x
925 W(b) __armv3_mpu_cache_on
926 W(b) __armv3_mpu_cache_off
927 W(b) __armv3_mpu_cache_flush
929 .word 0x41009400 @ ARM94x
931 W(b) __armv4_mpu_cache_on
932 W(b) __armv4_mpu_cache_off
933 W(b) __armv4_mpu_cache_flush
935 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
937 W(b) __arm926ejs_mmu_cache_on
938 W(b) __armv4_mmu_cache_off
939 W(b) __armv5tej_mmu_cache_flush
941 .word 0x00007000 @ ARM7 IDs
950 @ Everything from here on will be the new ID system.
952 .word 0x4401a100 @ sa110 / sa1100
954 W(b) __armv4_mmu_cache_on
955 W(b) __armv4_mmu_cache_off
956 W(b) __armv4_mmu_cache_flush
958 .word 0x6901b110 @ sa1110
960 W(b) __armv4_mmu_cache_on
961 W(b) __armv4_mmu_cache_off
962 W(b) __armv4_mmu_cache_flush
965 .word 0xffffff00 @ PXA9xx
966 W(b) __armv4_mmu_cache_on
967 W(b) __armv4_mmu_cache_off
968 W(b) __armv4_mmu_cache_flush
970 .word 0x56158000 @ PXA168
972 W(b) __armv4_mmu_cache_on
973 W(b) __armv4_mmu_cache_off
974 W(b) __armv5tej_mmu_cache_flush
976 .word 0x56050000 @ Feroceon
978 W(b) __armv4_mmu_cache_on
979 W(b) __armv4_mmu_cache_off
980 W(b) __armv5tej_mmu_cache_flush
982 #ifdef CONFIG_CPU_FEROCEON_OLD_ID
983 /* this conflicts with the standard ARMv5TE entry */
984 .long 0x41009260 @ Old Feroceon
986 b __armv4_mmu_cache_on
987 b __armv4_mmu_cache_off
988 b __armv5tej_mmu_cache_flush
991 .word 0x66015261 @ FA526
993 W(b) __fa526_cache_on
994 W(b) __armv4_mmu_cache_off
995 W(b) __fa526_cache_flush
997 @ These match on the architecture ID
999 .word 0x00020000 @ ARMv4T
1001 W(b) __armv4_mmu_cache_on
1002 W(b) __armv4_mmu_cache_off
1003 W(b) __armv4_mmu_cache_flush
1005 .word 0x00050000 @ ARMv5TE
1007 W(b) __armv4_mmu_cache_on
1008 W(b) __armv4_mmu_cache_off
1009 W(b) __armv4_mmu_cache_flush
1011 .word 0x00060000 @ ARMv5TEJ
1013 W(b) __armv4_mmu_cache_on
1014 W(b) __armv4_mmu_cache_off
1015 W(b) __armv5tej_mmu_cache_flush
1017 .word 0x0007b000 @ ARMv6
1019 W(b) __armv6_mmu_cache_on
1020 W(b) __armv4_mmu_cache_off
1021 W(b) __armv6_mmu_cache_flush
1023 .word 0x000f0000 @ new CPU Id
1025 W(b) __armv7_mmu_cache_on
1026 W(b) __armv7_mmu_cache_off
1027 W(b) __armv7_mmu_cache_flush
1029 .word 0 @ unrecognised type
1038 .size proc_types, . - proc_types
1041 * If you get a "non-constant expression in ".if" statement"
1042 * error from the assembler on this line, check that you have
1043 * not accidentally written a "b" instruction where you should
1044 * have written W(b).
1046 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1047 .error "The size of one or more proc_types entries is wrong."
1051 * Turn off the Cache and MMU. ARMv3 does not support
1052 * reading the control register, but ARMv4 does.
1055 * r0, r1, r2, r3, r9, r12 corrupted
1056 * This routine must preserve:
1060 cache_off: mov r3, #12 @ cache_off function
1063 __armv4_mpu_cache_off:
1064 mrc p15, 0, r0, c1, c0
1066 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
1068 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
1069 mcr p15, 0, r0, c7, c6, 0 @ flush D-Cache
1070 mcr p15, 0, r0, c7, c5, 0 @ flush I-Cache
1073 __armv3_mpu_cache_off:
1074 mrc p15, 0, r0, c1, c0
1076 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1078 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1081 __armv4_mmu_cache_off:
1083 mrc p15, 0, r0, c1, c0
1085 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1087 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1088 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1092 __armv7_mmu_cache_off:
1093 mrc p15, 0, r0, c1, c0
1099 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1101 bl __armv7_mmu_cache_flush
1104 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
1106 mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC
1107 mcr p15, 0, r0, c7, c10, 4 @ DSB
1108 mcr p15, 0, r0, c7, c5, 4 @ ISB
1112 * Clean and flush the cache to maintain consistency.
1115 * r1, r2, r3, r9, r10, r11, r12 corrupted
1116 * This routine must preserve:
1124 __armv4_mpu_cache_flush:
1129 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1130 mov r1, #7 << 5 @ 8 segments
1131 1: orr r3, r1, #63 << 26 @ 64 entries
1132 2: mcr p15, 0, r3, c7, c14, 2 @ clean & invalidate D index
1133 subs r3, r3, #1 << 26
1134 bcs 2b @ entries 63 to 0
1135 subs r1, r1, #1 << 5
1136 bcs 1b @ segments 7 to 0
1139 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1140 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1143 __fa526_cache_flush:
1147 mcr p15, 0, r1, c7, c14, 0 @ clean and invalidate D cache
1148 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1149 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1152 __armv6_mmu_cache_flush:
1155 mcreq p15, 0, r1, c7, c14, 0 @ clean+invalidate D
1156 mcr p15, 0, r1, c7, c5, 0 @ invalidate I+BTB
1157 mcreq p15, 0, r1, c7, c15, 0 @ clean+invalidate unified
1158 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1161 __armv7_mmu_cache_flush:
1164 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1165 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1168 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
1171 mcr p15, 0, r10, c7, c10, 5 @ DMB
1172 stmfd sp!, {r0-r7, r9-r11}
1173 mrc p15, 1, r0, c0, c0, 1 @ read clidr
1174 ands r3, r0, #0x7000000 @ extract loc from clidr
1175 mov r3, r3, lsr #23 @ left align loc bit field
1176 beq finished @ if loc is 0, then no need to clean
1177 mov r10, #0 @ start clean at cache level 0
1179 add r2, r10, r10, lsr #1 @ work out 3x current cache level
1180 mov r1, r0, lsr r2 @ extract cache type bits from clidr
1181 and r1, r1, #7 @ mask of the bits for current cache only
1182 cmp r1, #2 @ see what cache we have at this level
1183 blt skip @ skip if no cache, or just i-cache
1184 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1185 mcr p15, 0, r10, c7, c5, 4 @ isb to sych the new cssr&csidr
1186 mrc p15, 1, r1, c0, c0, 0 @ read the new csidr
1187 and r2, r1, #7 @ extract the length of the cache lines
1188 add r2, r2, #4 @ add 4 (line length offset)
1190 ands r4, r4, r1, lsr #3 @ find maximum number on the way size
1191 clz r5, r4 @ find bit position of way size increment
1193 ands r7, r7, r1, lsr #13 @ extract max number of the index size
1195 mov r9, r4 @ create working copy of max way size
1197 ARM( orr r11, r10, r9, lsl r5 ) @ factor way and cache number into r11
1198 ARM( orr r11, r11, r7, lsl r2 ) @ factor index number into r11
1199 THUMB( lsl r6, r9, r5 )
1200 THUMB( orr r11, r10, r6 ) @ factor way and cache number into r11
1201 THUMB( lsl r6, r7, r2 )
1202 THUMB( orr r11, r11, r6 ) @ factor index number into r11
1203 mcr p15, 0, r11, c7, c14, 2 @ clean & invalidate by set/way
1204 subs r9, r9, #1 @ decrement the way
1206 subs r7, r7, #1 @ decrement the index
1209 add r10, r10, #2 @ increment cache number
1213 ldmfd sp!, {r0-r7, r9-r11}
1214 mov r10, #0 @ switch back to cache level 0
1215 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1217 mcr p15, 0, r10, c7, c10, 4 @ DSB
1218 mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB
1219 mcr p15, 0, r10, c7, c10, 4 @ DSB
1220 mcr p15, 0, r10, c7, c5, 4 @ ISB
1223 __armv5tej_mmu_cache_flush:
1226 1: mrc p15, 0, r15, c7, c14, 3 @ test,clean,invalidate D cache
1228 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1229 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1232 __armv4_mmu_cache_flush:
1235 mov r2, #64*1024 @ default: 32K dcache size (*2)
1236 mov r11, #32 @ default: 32 byte line size
1237 mrc p15, 0, r3, c0, c0, 1 @ read cache type
1238 teq r3, r9 @ cache ID register present?
1243 mov r2, r2, lsl r1 @ base dcache size *2
1244 tst r3, #1 << 14 @ test M bit
1245 addne r2, r2, r2, lsr #1 @ +1/2 size if M == 1
1249 mov r11, r11, lsl r3 @ cache line size in bytes
1252 bic r1, r1, #63 @ align to longest cache line
1255 ARM( ldr r3, [r1], r11 ) @ s/w flush D cache
1256 THUMB( ldr r3, [r1] ) @ s/w flush D cache
1257 THUMB( add r1, r1, r11 )
1261 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1262 mcr p15, 0, r1, c7, c6, 0 @ flush D cache
1263 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1266 __armv3_mmu_cache_flush:
1267 __armv3_mpu_cache_flush:
1271 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1275 * Various debugging routines for printing hex characters and
1276 * memory, which again must be relocatable.
1280 .type phexbuf,#object
1282 .size phexbuf, . - phexbuf
1284 @ phex corrupts {r0, r1, r2, r3}
1285 phex: adr r3, phexbuf
1299 @ puts corrupts {r0, r1, r2, r3}
1301 1: ldrb r2, [r0], #1
1314 @ putc corrupts {r0, r1, r2, r3}
1321 @ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1322 memdump: mov r12, r0
1325 2: mov r0, r11, lsl #2
1333 ldr r0, [r12, r11, lsl #2]
1353 #ifdef CONFIG_ARM_VIRT_EXT
1355 __hyp_reentry_vectors:
1361 W(b) __enter_kernel @ hyp
1364 #endif /* CONFIG_ARM_VIRT_EXT */
1367 mov r0, #0 @ must be 0
1368 ARM( mov pc, r4 ) @ call kernel
1369 M_CLASS( add r4, r4, #1 ) @ enter in Thumb mode for M class
1370 THUMB( bx r4 ) @ entry point is always ARM for A/R classes
1374 #ifdef CONFIG_EFI_STUB
1376 _start: .long start - .
1378 ENTRY(efi_stub_entry)
1379 @ allocate space on stack for passing current zImage address
1380 @ and for the EFI stub to return of new entry point of
1381 @ zImage, as EFI stub may copy the kernel. Pointer address
1382 @ is passed in r2. r0 and r1 are passed through from the
1383 @ EFI firmware to efi_entry
1388 mov r2, sp @ pass zImage address in r2
1391 @ Check for error return from EFI stub. r0 has FDT address
1396 @ Preserve return value of efi_entry() in r4
1398 bl cache_clean_flush
1401 @ Set parameters for booting zImage according to boot protocol
1402 @ put FDT address in r2, it was returned by efi_entry()
1403 @ r1 is the machine type, and r0 needs to be 0
1408 @ Branch to (possibly) relocated zImage that is in [sp]
1410 ldr ip, =start_offset
1412 mov pc, lr @ no mode switch
1415 @ Return EFI_LOAD_ERROR to EFI firmware on error.
1418 ENDPROC(efi_stub_entry)
1422 .section ".stack", "aw", %nobits
1423 .L_user_stack: .space 4096