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, tmp1, tmp2
35 mcr p14, 0, \ch, c0, c5, 0
37 #elif defined(CONFIG_CPU_XSCALE)
38 .macro loadsp, rb, tmp1, tmp2
41 mcr p14, 0, \ch, c8, c0, 0
44 .macro loadsp, rb, tmp1, tmp2
47 mcr p14, 0, \ch, c1, c0, 0
53 #include CONFIG_DEBUG_LL_INCLUDE
59 #if defined(CONFIG_ARCH_SA1100)
60 .macro loadsp, rb, tmp1, tmp2
61 mov \rb, #0x80000000 @ physical base address
62 #ifdef CONFIG_DEBUG_LL_SER3
63 add \rb, \rb, #0x00050000 @ Ser3
65 add \rb, \rb, #0x00010000 @ Ser1
69 .macro loadsp, rb, tmp1, tmp2
70 addruart \rb, \tmp1, \tmp2
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
565 #ifdef CONFIG_ARM_VIRT_EXT
566 mrs r0, spsr @ Get saved CPU boot mode
567 and r0, r0, #MODE_MASK
568 cmp r0, #HYP_MODE @ if not booted in HYP mode...
569 bne __enter_kernel @ boot kernel directly
571 adr r12, .L__hyp_reentry_vectors_offset
576 __HVC(0) @ otherwise bounce to hyp mode
578 b . @ should never be reached
581 .L__hyp_reentry_vectors_offset: .long __hyp_reentry_vectors - .
589 .word __bss_start @ r2
592 .word input_data_end - 4 @ r10 (inflated size location)
593 .word _got_start @ r11
595 .word .L_user_stack_end @ sp
596 .word _end - restart + 16384 + 1024*1024
599 #ifdef CONFIG_ARCH_RPC
601 params: ldr r0, =0x10000100 @ params_phys for RPC
608 * Turn on the cache. We need to setup some page tables so that we
609 * can have both the I and D caches on.
611 * We place the page tables 16k down from the kernel execution address,
612 * and we hope that nothing else is using it. If we're using it, we
616 * r4 = kernel execution address
617 * r7 = architecture number
620 * r0, r1, r2, r3, r9, r10, r12 corrupted
621 * This routine must preserve:
625 cache_on: mov r3, #8 @ cache_on function
629 * Initialize the highest priority protection region, PR7
630 * to cover all 32bit address and cacheable and bufferable.
632 __armv4_mpu_cache_on:
633 mov r0, #0x3f @ 4G, the whole
634 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
635 mcr p15, 0, r0, c6, c7, 1
638 mcr p15, 0, r0, c2, c0, 0 @ D-cache on
639 mcr p15, 0, r0, c2, c0, 1 @ I-cache on
640 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
643 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
644 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
647 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
648 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
649 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
650 mrc p15, 0, r0, c1, c0, 0 @ read control reg
651 @ ...I .... ..D. WC.M
652 orr r0, r0, #0x002d @ .... .... ..1. 11.1
653 orr r0, r0, #0x1000 @ ...1 .... .... ....
655 mcr p15, 0, r0, c1, c0, 0 @ write control reg
658 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
659 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
662 __armv3_mpu_cache_on:
663 mov r0, #0x3f @ 4G, the whole
664 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
667 mcr p15, 0, r0, c2, c0, 0 @ cache on
668 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
671 mcr p15, 0, r0, c5, c0, 0 @ access permission
674 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
676 * ?? ARMv3 MMU does not allow reading the control register,
677 * does this really work on ARMv3 MPU?
679 mrc p15, 0, r0, c1, c0, 0 @ read control reg
680 @ .... .... .... WC.M
681 orr r0, r0, #0x000d @ .... .... .... 11.1
682 /* ?? this overwrites the value constructed above? */
684 mcr p15, 0, r0, c1, c0, 0 @ write control reg
686 /* ?? invalidate for the second time? */
687 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
690 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
696 __setup_mmu: sub r3, r4, #16384 @ Page directory size
697 bic r3, r3, #0xff @ Align the pointer
700 * Initialise the page tables, turning on the cacheable and bufferable
701 * bits for the RAM area only.
705 mov r9, r9, lsl #18 @ start of RAM
706 add r10, r9, #0x10000000 @ a reasonable RAM size
707 mov r1, #0x12 @ XN|U + section mapping
708 orr r1, r1, #3 << 10 @ AP=11
710 1: cmp r1, r9 @ if virt > start of RAM
711 cmphs r10, r1 @ && end of RAM > virt
712 bic r1, r1, #0x1c @ clear XN|U + C + B
713 orrlo r1, r1, #0x10 @ Set XN|U for non-RAM
714 orrhs r1, r1, r6 @ set RAM section settings
715 str r1, [r0], #4 @ 1:1 mapping
720 * If ever we are running from Flash, then we surely want the cache
721 * to be enabled also for our execution instance... We map 2MB of it
722 * so there is no map overlap problem for up to 1 MB compressed kernel.
723 * If the execution is in RAM then we would only be duplicating the above.
725 orr r1, r6, #0x04 @ ensure B is set for this
729 orr r1, r1, r2, lsl #20
730 add r0, r3, r2, lsl #2
737 @ Enable unaligned access on v6, to allow better code generation
738 @ for the decompressor C code:
739 __armv6_mmu_cache_on:
740 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
741 bic r0, r0, #2 @ A (no unaligned access fault)
742 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
743 mcr p15, 0, r0, c1, c0, 0 @ write SCTLR
744 b __armv4_mmu_cache_on
746 __arm926ejs_mmu_cache_on:
747 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
748 mov r0, #4 @ put dcache in WT mode
749 mcr p15, 7, r0, c15, c0, 0
752 __armv4_mmu_cache_on:
755 mov r6, #CB_BITS | 0x12 @ U
758 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
759 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
760 mrc p15, 0, r0, c1, c0, 0 @ read control reg
761 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
763 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
764 bl __common_mmu_cache_on
766 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
770 __armv7_mmu_cache_on:
773 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
775 movne r6, #CB_BITS | 0x02 @ !XN
778 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
780 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
782 mrc p15, 0, r0, c1, c0, 0 @ read control reg
783 bic r0, r0, #1 << 28 @ clear SCTLR.TRE
784 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
785 orr r0, r0, #0x003c @ write buffer
786 bic r0, r0, #2 @ A (no unaligned access fault)
787 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
788 @ (needed for ARM1176)
790 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
791 mrcne p15, 0, r6, c2, c0, 2 @ read ttb control reg
792 orrne r0, r0, #1 @ MMU enabled
793 movne r1, #0xfffffffd @ domain 0 = client
794 bic r6, r6, #1 << 31 @ 32-bit translation system
795 bic r6, r6, #(7 << 0) | (1 << 4) @ use only ttbr0
796 mcrne p15, 0, r3, c2, c0, 0 @ load page table pointer
797 mcrne p15, 0, r1, c3, c0, 0 @ load domain access control
798 mcrne p15, 0, r6, c2, c0, 2 @ load ttb control
800 mcr p15, 0, r0, c7, c5, 4 @ ISB
801 mcr p15, 0, r0, c1, c0, 0 @ load control register
802 mrc p15, 0, r0, c1, c0, 0 @ and read it back
804 mcr p15, 0, r0, c7, c5, 4 @ ISB
809 mov r6, #CB_BITS | 0x12 @ U
812 mcr p15, 0, r0, c7, c7, 0 @ Invalidate whole cache
813 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
814 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
815 mrc p15, 0, r0, c1, c0, 0 @ read control reg
816 orr r0, r0, #0x1000 @ I-cache enable
817 bl __common_mmu_cache_on
819 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
822 __common_mmu_cache_on:
823 #ifndef CONFIG_THUMB2_KERNEL
825 orr r0, r0, #0x000d @ Write buffer, mmu
828 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
829 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
831 .align 5 @ cache line aligned
832 1: mcr p15, 0, r0, c1, c0, 0 @ load control register
833 mrc p15, 0, r0, c1, c0, 0 @ and read it back to
834 sub pc, lr, r0, lsr #32 @ properly flush pipeline
837 #define PROC_ENTRY_SIZE (4*5)
840 * Here follow the relocatable cache support functions for the
841 * various processors. This is a generic hook for locating an
842 * entry and jumping to an instruction at the specified offset
843 * from the start of the block. Please note this is all position
853 call_cache_fn: adr r12, proc_types
854 #ifdef CONFIG_CPU_CP15
855 mrc p15, 0, r9, c0, c0 @ get processor ID
856 #elif defined(CONFIG_CPU_V7M)
858 * On v7-M the processor id is located in the V7M_SCB_CPUID
859 * register, but as cache handling is IMPLEMENTATION DEFINED on
860 * v7-M (if existant at all) we just return early here.
861 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
862 * __armv7_mmu_cache_{on,off,flush}) would be selected which
863 * use cp15 registers that are not implemented on v7-M.
867 ldr r9, =CONFIG_PROCESSOR_ID
869 1: ldr r1, [r12, #0] @ get value
870 ldr r2, [r12, #4] @ get mask
871 eor r1, r1, r9 @ (real ^ match)
873 ARM( addeq pc, r12, r3 ) @ call cache function
874 THUMB( addeq r12, r3 )
875 THUMB( moveq pc, r12 ) @ call cache function
876 add r12, r12, #PROC_ENTRY_SIZE
880 * Table for cache operations. This is basically:
883 * - 'cache on' method instruction
884 * - 'cache off' method instruction
885 * - 'cache flush' method instruction
887 * We match an entry using: ((real_id ^ match) & mask) == 0
889 * Writethrough caches generally only need 'on' and 'off'
890 * methods. Writeback caches _must_ have the flush method
894 .type proc_types,#object
896 .word 0x41000000 @ old ARM ID
905 .word 0x41007000 @ ARM7/710
914 .word 0x41807200 @ ARM720T (writethrough)
916 W(b) __armv4_mmu_cache_on
917 W(b) __armv4_mmu_cache_off
921 .word 0x41007400 @ ARM74x
923 W(b) __armv3_mpu_cache_on
924 W(b) __armv3_mpu_cache_off
925 W(b) __armv3_mpu_cache_flush
927 .word 0x41009400 @ ARM94x
929 W(b) __armv4_mpu_cache_on
930 W(b) __armv4_mpu_cache_off
931 W(b) __armv4_mpu_cache_flush
933 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
935 W(b) __arm926ejs_mmu_cache_on
936 W(b) __armv4_mmu_cache_off
937 W(b) __armv5tej_mmu_cache_flush
939 .word 0x00007000 @ ARM7 IDs
948 @ Everything from here on will be the new ID system.
950 .word 0x4401a100 @ sa110 / sa1100
952 W(b) __armv4_mmu_cache_on
953 W(b) __armv4_mmu_cache_off
954 W(b) __armv4_mmu_cache_flush
956 .word 0x6901b110 @ sa1110
958 W(b) __armv4_mmu_cache_on
959 W(b) __armv4_mmu_cache_off
960 W(b) __armv4_mmu_cache_flush
963 .word 0xffffff00 @ PXA9xx
964 W(b) __armv4_mmu_cache_on
965 W(b) __armv4_mmu_cache_off
966 W(b) __armv4_mmu_cache_flush
968 .word 0x56158000 @ PXA168
970 W(b) __armv4_mmu_cache_on
971 W(b) __armv4_mmu_cache_off
972 W(b) __armv5tej_mmu_cache_flush
974 .word 0x56050000 @ Feroceon
976 W(b) __armv4_mmu_cache_on
977 W(b) __armv4_mmu_cache_off
978 W(b) __armv5tej_mmu_cache_flush
980 #ifdef CONFIG_CPU_FEROCEON_OLD_ID
981 /* this conflicts with the standard ARMv5TE entry */
982 .long 0x41009260 @ Old Feroceon
984 b __armv4_mmu_cache_on
985 b __armv4_mmu_cache_off
986 b __armv5tej_mmu_cache_flush
989 .word 0x66015261 @ FA526
991 W(b) __fa526_cache_on
992 W(b) __armv4_mmu_cache_off
993 W(b) __fa526_cache_flush
995 @ These match on the architecture ID
997 .word 0x00020000 @ ARMv4T
999 W(b) __armv4_mmu_cache_on
1000 W(b) __armv4_mmu_cache_off
1001 W(b) __armv4_mmu_cache_flush
1003 .word 0x00050000 @ ARMv5TE
1005 W(b) __armv4_mmu_cache_on
1006 W(b) __armv4_mmu_cache_off
1007 W(b) __armv4_mmu_cache_flush
1009 .word 0x00060000 @ ARMv5TEJ
1011 W(b) __armv4_mmu_cache_on
1012 W(b) __armv4_mmu_cache_off
1013 W(b) __armv5tej_mmu_cache_flush
1015 .word 0x0007b000 @ ARMv6
1017 W(b) __armv6_mmu_cache_on
1018 W(b) __armv4_mmu_cache_off
1019 W(b) __armv6_mmu_cache_flush
1021 .word 0x000f0000 @ new CPU Id
1023 W(b) __armv7_mmu_cache_on
1024 W(b) __armv7_mmu_cache_off
1025 W(b) __armv7_mmu_cache_flush
1027 .word 0 @ unrecognised type
1036 .size proc_types, . - proc_types
1039 * If you get a "non-constant expression in ".if" statement"
1040 * error from the assembler on this line, check that you have
1041 * not accidentally written a "b" instruction where you should
1042 * have written W(b).
1044 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1045 .error "The size of one or more proc_types entries is wrong."
1049 * Turn off the Cache and MMU. ARMv3 does not support
1050 * reading the control register, but ARMv4 does.
1053 * r0, r1, r2, r3, r9, r12 corrupted
1054 * This routine must preserve:
1058 cache_off: mov r3, #12 @ cache_off function
1061 __armv4_mpu_cache_off:
1062 mrc p15, 0, r0, c1, c0
1064 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
1066 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
1067 mcr p15, 0, r0, c7, c6, 0 @ flush D-Cache
1068 mcr p15, 0, r0, c7, c5, 0 @ flush I-Cache
1071 __armv3_mpu_cache_off:
1072 mrc p15, 0, r0, c1, c0
1074 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1076 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1079 __armv4_mmu_cache_off:
1081 mrc p15, 0, r0, c1, c0
1083 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1085 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1086 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1090 __armv7_mmu_cache_off:
1091 mrc p15, 0, r0, c1, c0
1097 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1099 bl __armv7_mmu_cache_flush
1102 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
1104 mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC
1105 mcr p15, 0, r0, c7, c10, 4 @ DSB
1106 mcr p15, 0, r0, c7, c5, 4 @ ISB
1110 * Clean and flush the cache to maintain consistency.
1113 * r1, r2, r3, r9, r10, r11, r12 corrupted
1114 * This routine must preserve:
1122 __armv4_mpu_cache_flush:
1127 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1128 mov r1, #7 << 5 @ 8 segments
1129 1: orr r3, r1, #63 << 26 @ 64 entries
1130 2: mcr p15, 0, r3, c7, c14, 2 @ clean & invalidate D index
1131 subs r3, r3, #1 << 26
1132 bcs 2b @ entries 63 to 0
1133 subs r1, r1, #1 << 5
1134 bcs 1b @ segments 7 to 0
1137 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1138 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1141 __fa526_cache_flush:
1145 mcr p15, 0, r1, c7, c14, 0 @ clean and invalidate D cache
1146 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1147 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1150 __armv6_mmu_cache_flush:
1153 mcreq p15, 0, r1, c7, c14, 0 @ clean+invalidate D
1154 mcr p15, 0, r1, c7, c5, 0 @ invalidate I+BTB
1155 mcreq p15, 0, r1, c7, c15, 0 @ clean+invalidate unified
1156 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1159 __armv7_mmu_cache_flush:
1162 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1163 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1166 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
1169 mcr p15, 0, r10, c7, c10, 5 @ DMB
1170 stmfd sp!, {r0-r7, r9-r11}
1171 mrc p15, 1, r0, c0, c0, 1 @ read clidr
1172 ands r3, r0, #0x7000000 @ extract loc from clidr
1173 mov r3, r3, lsr #23 @ left align loc bit field
1174 beq finished @ if loc is 0, then no need to clean
1175 mov r10, #0 @ start clean at cache level 0
1177 add r2, r10, r10, lsr #1 @ work out 3x current cache level
1178 mov r1, r0, lsr r2 @ extract cache type bits from clidr
1179 and r1, r1, #7 @ mask of the bits for current cache only
1180 cmp r1, #2 @ see what cache we have at this level
1181 blt skip @ skip if no cache, or just i-cache
1182 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1183 mcr p15, 0, r10, c7, c5, 4 @ isb to sych the new cssr&csidr
1184 mrc p15, 1, r1, c0, c0, 0 @ read the new csidr
1185 and r2, r1, #7 @ extract the length of the cache lines
1186 add r2, r2, #4 @ add 4 (line length offset)
1188 ands r4, r4, r1, lsr #3 @ find maximum number on the way size
1189 clz r5, r4 @ find bit position of way size increment
1191 ands r7, r7, r1, lsr #13 @ extract max number of the index size
1193 mov r9, r4 @ create working copy of max way size
1195 ARM( orr r11, r10, r9, lsl r5 ) @ factor way and cache number into r11
1196 ARM( orr r11, r11, r7, lsl r2 ) @ factor index number into r11
1197 THUMB( lsl r6, r9, r5 )
1198 THUMB( orr r11, r10, r6 ) @ factor way and cache number into r11
1199 THUMB( lsl r6, r7, r2 )
1200 THUMB( orr r11, r11, r6 ) @ factor index number into r11
1201 mcr p15, 0, r11, c7, c14, 2 @ clean & invalidate by set/way
1202 subs r9, r9, #1 @ decrement the way
1204 subs r7, r7, #1 @ decrement the index
1207 add r10, r10, #2 @ increment cache number
1211 ldmfd sp!, {r0-r7, r9-r11}
1212 mov r10, #0 @ switch back to cache level 0
1213 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1215 mcr p15, 0, r10, c7, c10, 4 @ DSB
1216 mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB
1217 mcr p15, 0, r10, c7, c10, 4 @ DSB
1218 mcr p15, 0, r10, c7, c5, 4 @ ISB
1221 __armv5tej_mmu_cache_flush:
1224 1: mrc p15, 0, r15, c7, c14, 3 @ test,clean,invalidate D cache
1226 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1227 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1230 __armv4_mmu_cache_flush:
1233 mov r2, #64*1024 @ default: 32K dcache size (*2)
1234 mov r11, #32 @ default: 32 byte line size
1235 mrc p15, 0, r3, c0, c0, 1 @ read cache type
1236 teq r3, r9 @ cache ID register present?
1241 mov r2, r2, lsl r1 @ base dcache size *2
1242 tst r3, #1 << 14 @ test M bit
1243 addne r2, r2, r2, lsr #1 @ +1/2 size if M == 1
1247 mov r11, r11, lsl r3 @ cache line size in bytes
1250 bic r1, r1, #63 @ align to longest cache line
1253 ARM( ldr r3, [r1], r11 ) @ s/w flush D cache
1254 THUMB( ldr r3, [r1] ) @ s/w flush D cache
1255 THUMB( add r1, r1, r11 )
1259 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1260 mcr p15, 0, r1, c7, c6, 0 @ flush D cache
1261 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1264 __armv3_mmu_cache_flush:
1265 __armv3_mpu_cache_flush:
1269 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1273 * Various debugging routines for printing hex characters and
1274 * memory, which again must be relocatable.
1278 .type phexbuf,#object
1280 .size phexbuf, . - phexbuf
1282 @ phex corrupts {r0, r1, r2, r3}
1283 phex: adr r3, phexbuf
1297 @ puts corrupts {r0, r1, r2, r3}
1298 puts: loadsp r3, r2, r1
1299 1: ldrb r2, [r0], #1
1312 @ putc corrupts {r0, r1, r2, r3}
1319 @ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1320 memdump: mov r12, r0
1323 2: mov r0, r11, lsl #2
1331 ldr r0, [r12, r11, lsl #2]
1351 #ifdef CONFIG_ARM_VIRT_EXT
1353 __hyp_reentry_vectors:
1359 W(b) __enter_kernel @ hyp
1362 #endif /* CONFIG_ARM_VIRT_EXT */
1365 mov r0, #0 @ must be 0
1366 mov r1, r7 @ restore architecture number
1367 mov r2, r8 @ restore atags pointer
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