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
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
438 1: ldmdb r6!, {r0 - r3, r10 - r12, lr}
440 stmdb r9!, {r0 - r3, r10 - r12, lr}
443 /* Preserve offset to relocated code. */
446 #ifndef CONFIG_ZBOOT_ROM
447 /* cache_clean_flush may use the stack, so relocate it */
459 * If delta is zero, we are running at the address we were linked at.
463 * r4 = kernel execution address (possibly with LSB set)
464 * r5 = appended dtb size (0 if not present)
465 * r7 = architecture ID
477 #ifndef CONFIG_ZBOOT_ROM
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.
487 * Relocate all entries in the GOT table.
488 * Bump bss entries to _edata + dtb size
490 1: 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
499 /* bump our bss pointers too */
506 * Relocate entries in the GOT table. We only relocate
507 * the entries that are outside the (relocated) BSS region.
509 1: 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.
518 not_relocated: mov r0, #0
519 1: str r0, [r2], #4 @ clear bss
527 * Did we skip the cache setup earlier?
528 * That is indicated by the LSB in r4.
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
543 mov r1, sp @ malloc space above stack
544 add r2, sp, #0x10000 @ 64k max
549 mov r1, r7 @ restore architecture number
550 mov r2, r8 @ restore atags pointer
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
558 adr r12, .L__hyp_reentry_vectors_offset
563 __HVC(0) @ otherwise bounce to hyp mode
565 b . @ should never be reached
568 .L__hyp_reentry_vectors_offset: .long __hyp_reentry_vectors - .
576 .word __bss_start @ r2
579 .word input_data_end - 4 @ r10 (inflated size location)
580 .word _got_start @ r11
582 .word .L_user_stack_end @ sp
583 .word _end - restart + 16384 + 1024*1024
586 #ifdef CONFIG_ARCH_RPC
588 params: ldr r0, =0x10000100 @ params_phys for RPC
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.
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
603 * r4 = kernel execution address
604 * r7 = architecture number
607 * r0, r1, r2, r3, r9, r10, r12 corrupted
608 * This routine must preserve:
612 cache_on: mov r3, #8 @ cache_on function
616 * Initialize the highest priority protection region, PR7
617 * to cover all 32bit address and cacheable and bufferable.
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
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
630 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
631 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
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 .... .... ....
642 mcr p15, 0, r0, c1, c0, 0 @ write control reg
645 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
646 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
649 __armv3_mpu_cache_on:
650 mov r0, #0x3f @ 4G, the whole
651 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
654 mcr p15, 0, r0, c2, c0, 0 @ cache on
655 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
658 mcr p15, 0, r0, c5, c0, 0 @ access permission
661 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
663 * ?? ARMv3 MMU does not allow reading the control register,
664 * does this really work on ARMv3 MPU?
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? */
671 mcr p15, 0, r0, c1, c0, 0 @ write control reg
673 /* ?? invalidate for the second time? */
674 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
677 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
683 __setup_mmu: sub r3, r4, #16384 @ Page directory size
684 bic r3, r3, #0xff @ Align the pointer
687 * Initialise the page tables, turning on the cacheable and bufferable
688 * bits for the RAM area only.
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
697 1: 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
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.
712 orr r1, r6, #0x04 @ ensure B is set for this
716 orr r1, r1, r2, lsl #20
717 add r0, r3, r2, lsl #2
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
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
739 __armv4_mmu_cache_on:
742 mov r6, #CB_BITS | 0x12 @ U
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
750 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
751 bl __common_mmu_cache_on
753 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
757 __armv7_mmu_cache_on:
760 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
762 movne r6, #CB_BITS | 0x02 @ !XN
765 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
767 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
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)
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
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
791 mcr p15, 0, r0, c7, c5, 4 @ ISB
796 mov r6, #CB_BITS | 0x12 @ U
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
806 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
809 __common_mmu_cache_on:
810 #ifndef CONFIG_THUMB2_KERNEL
812 orr r0, r0, #0x000d @ Write buffer, mmu
815 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
816 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
818 .align 5 @ cache line aligned
819 1: 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
824 #define PROC_ENTRY_SIZE (4*5)
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
840 call_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)
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.
854 ldr r9, =CONFIG_PROCESSOR_ID
856 1: ldr r1, [r12, #0] @ get value
857 ldr r2, [r12, #4] @ get mask
858 eor r1, r1, r9 @ (real ^ match)
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
867 * Table for cache operations. This is basically:
870 * - 'cache on' method instruction
871 * - 'cache off' method instruction
872 * - 'cache flush' method instruction
874 * We match an entry using: ((real_id ^ match) & mask) == 0
876 * Writethrough caches generally only need 'on' and 'off'
877 * methods. Writeback caches _must_ have the flush method
881 .type proc_types,#object
883 .word 0x41000000 @ old ARM ID
892 .word 0x41007000 @ ARM7/710
901 .word 0x41807200 @ ARM720T (writethrough)
903 W(b) __armv4_mmu_cache_on
904 W(b) __armv4_mmu_cache_off
908 .word 0x41007400 @ ARM74x
910 W(b) __armv3_mpu_cache_on
911 W(b) __armv3_mpu_cache_off
912 W(b) __armv3_mpu_cache_flush
914 .word 0x41009400 @ ARM94x
916 W(b) __armv4_mpu_cache_on
917 W(b) __armv4_mpu_cache_off
918 W(b) __armv4_mpu_cache_flush
920 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
922 W(b) __arm926ejs_mmu_cache_on
923 W(b) __armv4_mmu_cache_off
924 W(b) __armv5tej_mmu_cache_flush
926 .word 0x00007000 @ ARM7 IDs
935 @ Everything from here on will be the new ID system.
937 .word 0x4401a100 @ sa110 / sa1100
939 W(b) __armv4_mmu_cache_on
940 W(b) __armv4_mmu_cache_off
941 W(b) __armv4_mmu_cache_flush
943 .word 0x6901b110 @ sa1110
945 W(b) __armv4_mmu_cache_on
946 W(b) __armv4_mmu_cache_off
947 W(b) __armv4_mmu_cache_flush
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
955 .word 0x56158000 @ PXA168
957 W(b) __armv4_mmu_cache_on
958 W(b) __armv4_mmu_cache_off
959 W(b) __armv5tej_mmu_cache_flush
961 .word 0x56050000 @ Feroceon
963 W(b) __armv4_mmu_cache_on
964 W(b) __armv4_mmu_cache_off
965 W(b) __armv5tej_mmu_cache_flush
967 #ifdef CONFIG_CPU_FEROCEON_OLD_ID
968 /* this conflicts with the standard ARMv5TE entry */
969 .long 0x41009260 @ Old Feroceon
971 b __armv4_mmu_cache_on
972 b __armv4_mmu_cache_off
973 b __armv5tej_mmu_cache_flush
976 .word 0x66015261 @ FA526
978 W(b) __fa526_cache_on
979 W(b) __armv4_mmu_cache_off
980 W(b) __fa526_cache_flush
982 @ These match on the architecture ID
984 .word 0x00020000 @ ARMv4T
986 W(b) __armv4_mmu_cache_on
987 W(b) __armv4_mmu_cache_off
988 W(b) __armv4_mmu_cache_flush
990 .word 0x00050000 @ ARMv5TE
992 W(b) __armv4_mmu_cache_on
993 W(b) __armv4_mmu_cache_off
994 W(b) __armv4_mmu_cache_flush
996 .word 0x00060000 @ ARMv5TEJ
998 W(b) __armv4_mmu_cache_on
999 W(b) __armv4_mmu_cache_off
1000 W(b) __armv5tej_mmu_cache_flush
1002 .word 0x0007b000 @ ARMv6
1004 W(b) __armv6_mmu_cache_on
1005 W(b) __armv4_mmu_cache_off
1006 W(b) __armv6_mmu_cache_flush
1008 .word 0x000f0000 @ new CPU Id
1010 W(b) __armv7_mmu_cache_on
1011 W(b) __armv7_mmu_cache_off
1012 W(b) __armv7_mmu_cache_flush
1014 .word 0 @ unrecognised type
1023 .size proc_types, . - proc_types
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).
1031 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1032 .error "The size of one or more proc_types entries is wrong."
1036 * Turn off the Cache and MMU. ARMv3 does not support
1037 * reading the control register, but ARMv4 does.
1040 * r0, r1, r2, r3, r9, r12 corrupted
1041 * This routine must preserve:
1045 cache_off: mov r3, #12 @ cache_off function
1048 __armv4_mpu_cache_off:
1049 mrc p15, 0, r0, c1, c0
1051 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
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
1058 __armv3_mpu_cache_off:
1059 mrc p15, 0, r0, c1, c0
1061 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1063 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1066 __armv4_mmu_cache_off:
1068 mrc p15, 0, r0, c1, c0
1070 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1072 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1073 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1077 __armv7_mmu_cache_off:
1078 mrc p15, 0, r0, c1, c0
1084 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1086 bl __armv7_mmu_cache_flush
1089 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
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
1097 * Clean and flush the cache to maintain consistency.
1100 * r1, r2, r3, r9, r10, r11, r12 corrupted
1101 * This routine must preserve:
1109 __armv4_mpu_cache_flush:
1114 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1115 mov r1, #7 << 5 @ 8 segments
1116 1: orr r3, r1, #63 << 26 @ 64 entries
1117 2: 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
1124 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1125 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1128 __fa526_cache_flush:
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
1137 __armv6_mmu_cache_flush:
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
1146 __armv7_mmu_cache_flush:
1149 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1150 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1153 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
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
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)
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
1178 ands r7, r7, r1, lsr #13 @ extract max number of the index size
1180 mov r9, r4 @ create working copy of max way size
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
1191 subs r7, r7, #1 @ decrement the index
1194 add r10, r10, #2 @ increment cache number
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
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
1208 __armv5tej_mmu_cache_flush:
1211 1: mrc p15, 0, r15, c7, c14, 3 @ test,clean,invalidate D cache
1213 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1214 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1217 __armv4_mmu_cache_flush:
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?
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
1234 mov r11, r11, lsl r3 @ cache line size in bytes
1237 bic r1, r1, #63 @ align to longest cache line
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 )
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
1251 __armv3_mmu_cache_flush:
1252 __armv3_mpu_cache_flush:
1256 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1260 * Various debugging routines for printing hex characters and
1261 * memory, which again must be relocatable.
1265 .type phexbuf,#object
1267 .size phexbuf, . - phexbuf
1269 @ phex corrupts {r0, r1, r2, r3}
1270 phex: adr r3, phexbuf
1284 @ puts corrupts {r0, r1, r2, r3}
1286 1: ldrb r2, [r0], #1
1299 @ putc corrupts {r0, r1, r2, r3}
1306 @ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1307 memdump: mov r12, r0
1310 2: mov r0, r11, lsl #2
1318 ldr r0, [r12, r11, lsl #2]
1338 #ifdef CONFIG_ARM_VIRT_EXT
1340 __hyp_reentry_vectors:
1346 W(b) __enter_kernel @ hyp
1349 #endif /* CONFIG_ARM_VIRT_EXT */
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
1359 #ifdef CONFIG_EFI_STUB
1361 _start: .long start - .
1363 ENTRY(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
1373 mov r2, sp @ pass zImage address in r2
1376 @ Check for error return from EFI stub. r0 has FDT address
1381 @ Preserve return value of efi_entry() in r4
1383 bl cache_clean_flush
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
1393 @ Branch to (possibly) relocated zImage that is in [sp]
1395 ldr ip, =start_offset
1397 mov pc, lr @ no mode switch
1400 @ Return EFI_LOAD_ERROR to EFI firmware on error.
1403 ENDPROC(efi_stub_entry)
1407 .section ".stack", "aw", %nobits
1408 .L_user_stack: .space 4096