1 /* SPDX-License-Identifier: GPL-2.0-only */
3 * linux/arch/arm/boot/compressed/head.S
5 * Copyright (C) 1996-2002 Russell King
6 * Copyright (C) 2004 Hyok S. Choi (MPU support)
8 #include <linux/linkage.h>
9 #include <asm/assembler.h>
12 #include "efi-header.S"
14 AR_CLASS( .arch armv7-a )
15 M_CLASS( .arch armv7-m )
20 * Note that these macros must not contain any code which is not
21 * 100% relocatable. Any attempt to do so will result in a crash.
22 * Please select one of the following when turning on debugging.
26 #if defined(CONFIG_DEBUG_ICEDCC)
28 #if defined(CONFIG_CPU_V6) || defined(CONFIG_CPU_V6K) || defined(CONFIG_CPU_V7)
29 .macro loadsp, rb, tmp1, tmp2
32 mcr p14, 0, \ch, c0, c5, 0
34 #elif defined(CONFIG_CPU_XSCALE)
35 .macro loadsp, rb, tmp1, tmp2
38 mcr p14, 0, \ch, c8, c0, 0
41 .macro loadsp, rb, tmp1, tmp2
44 mcr p14, 0, \ch, c1, c0, 0
50 #include CONFIG_DEBUG_LL_INCLUDE
56 #if defined(CONFIG_ARCH_SA1100)
57 .macro loadsp, rb, tmp1, tmp2
58 mov \rb, #0x80000000 @ physical base address
59 #ifdef CONFIG_DEBUG_LL_SER3
60 add \rb, \rb, #0x00050000 @ Ser3
62 add \rb, \rb, #0x00010000 @ Ser1
66 .macro loadsp, rb, tmp1, tmp2
67 addruart \rb, \tmp1, \tmp2
84 .macro debug_reloc_start
87 kphex r6, 8 /* processor id */
89 kphex r7, 8 /* architecture id */
90 #ifdef CONFIG_CPU_CP15
92 mrc p15, 0, r0, c1, c0
93 kphex r0, 8 /* control reg */
96 kphex r5, 8 /* decompressed kernel start */
98 kphex r9, 8 /* decompressed kernel end */
100 kphex r4, 8 /* kernel execution address */
105 .macro debug_reloc_end
107 kphex r5, 8 /* end of kernel */
110 bl memdump /* dump 256 bytes at start of kernel */
115 * Debug kernel copy by printing the memory addresses involved
117 .macro dbgkc, begin, end, cbegin, cend
124 kphex \begin, 8 /* Start of compressed kernel */
128 kphex \end, 8 /* End of compressed kernel */
133 kphex \cbegin, 8 /* Start of kernel copy */
137 kphex \cend, 8 /* End of kernel copy */
143 .macro enable_cp15_barriers, reg
144 mrc p15, 0, \reg, c1, c0, 0 @ read SCTLR
145 tst \reg, #(1 << 5) @ CP15BEN bit set?
147 orr \reg, \reg, #(1 << 5) @ CP15 barrier instructions
148 mcr p15, 0, \reg, c1, c0, 0 @ write SCTLR
149 ARM( .inst 0xf57ff06f @ v7+ isb )
154 .section ".start", "ax"
156 * sort out different calling conventions
160 * Always enter in ARM state for CPUs that support the ARM ISA.
161 * As of today (2014) that's exactly the members of the A and R
166 .type start,#function
168 * These 7 nops along with the 1 nop immediately below for
169 * !THUMB2 form 8 nops that make the compressed kernel bootable
170 * on legacy ARM systems that were assuming the kernel in a.out
171 * binary format. The boot loaders on these systems would
172 * jump 32 bytes into the image to skip the a.out header.
173 * with these 8 nops filling exactly 32 bytes, things still
174 * work as expected on these legacy systems. Thumb2 mode keeps
175 * 7 of the nops as it turns out that some boot loaders
176 * were patching the initial instructions of the kernel, i.e
177 * had started to exploit this "patch area".
182 #ifndef CONFIG_THUMB2_KERNEL
185 AR_CLASS( sub pc, pc, #3 ) @ A/R: switch to Thumb2 mode
186 M_CLASS( nop.w ) @ M: already in Thumb2 mode
191 .word _magic_sig @ Magic numbers to help the loader
192 .word _magic_start @ absolute load/run zImage address
193 .word _magic_end @ zImage end address
194 .word 0x04030201 @ endianness flag
195 .word 0x45454545 @ another magic number to indicate
196 .word _magic_table @ additional data table
200 ARM_BE8( setend be ) @ go BE8 if compiled for BE8
201 AR_CLASS( mrs r9, cpsr )
202 #ifdef CONFIG_ARM_VIRT_EXT
203 bl __hyp_stub_install @ get into SVC mode, reversibly
205 mov r7, r1 @ save architecture ID
206 mov r8, r2 @ save atags pointer
208 #ifndef CONFIG_CPU_V7M
210 * Booting from Angel - need to enter SVC mode and disable
211 * FIQs/IRQs (numeric definitions from angel arm.h source).
212 * We only do this if we were in user mode on entry.
214 mrs r2, cpsr @ get current mode
215 tst r2, #3 @ not user?
217 mov r0, #0x17 @ angel_SWIreason_EnterSVC
218 ARM( swi 0x123456 ) @ angel_SWI_ARM
219 THUMB( svc 0xab ) @ angel_SWI_THUMB
221 safe_svcmode_maskall r0
222 msr spsr_cxsf, r9 @ Save the CPU boot mode in
226 * Note that some cache flushing and other stuff may
227 * be needed here - is there an Angel SWI call for this?
231 * some architecture specific code can be inserted
232 * by the linker here, but it should preserve r7, r8, and r9.
237 #ifdef CONFIG_AUTO_ZRELADDR
239 * Find the start of physical memory. As we are executing
240 * without the MMU on, we are in the physical address space.
241 * We just need to get rid of any offset by aligning the
244 * This alignment is a balance between the requirements of
245 * different platforms - we have chosen 128MB to allow
246 * platforms which align the start of their physical memory
247 * to 128MB to use this feature, while allowing the zImage
248 * to be placed within the first 128MB of memory on other
249 * platforms. Increasing the alignment means we place
250 * stricter alignment requirements on the start of physical
251 * memory, but relaxing it means that we break people who
252 * are already placing their zImage in (eg) the top 64MB
256 and r4, r4, #0xf8000000
257 /* Determine final kernel image address. */
258 add r4, r4, #TEXT_OFFSET
264 * Set up a page table only if it won't overwrite ourself.
265 * That means r4 < pc || r4 - 16k page directory > &_end.
266 * Given that r4 > &_end is most unfrequent, we add a rough
267 * additional 1MB of room for a possible appended DTB.
274 orrcc r4, r4, #1 @ remember we skipped cache_on
278 ldmia r0, {r1, r2, r3, r6, r10, r11, r12}
282 * We might be running at a different address. We need
283 * to fix up various pointers.
285 sub r0, r0, r1 @ calculate the delta offset
286 add r6, r6, r0 @ _edata
287 add r10, r10, r0 @ inflated kernel size location
290 * The kernel build system appends the size of the
291 * decompressed kernel at the end of the compressed data
292 * in little-endian form.
296 orr r9, r9, lr, lsl #8
299 orr r9, r9, lr, lsl #16
300 orr r9, r9, r10, lsl #24
302 #ifndef CONFIG_ZBOOT_ROM
303 /* malloc space is above the relocated stack (64k max) */
305 add r10, sp, #0x10000
308 * With ZBOOT_ROM the bss/stack is non relocatable,
309 * but someone could still run this code from RAM,
310 * in which case our reference is _edata.
315 mov r5, #0 @ init dtb size to 0
316 #ifdef CONFIG_ARM_APPENDED_DTB
321 * r4 = final kernel address (possibly with LSB set)
322 * r5 = appended dtb size (still unknown)
324 * r7 = architecture ID
325 * r8 = atags/device tree pointer
326 * r9 = size of decompressed image
327 * r10 = end of this image, including bss/stack/malloc space if non XIP
332 * if there are device trees (dtb) appended to zImage, advance r10 so that the
333 * dtb data will get relocated along with the kernel if necessary.
338 ldr r1, =0xedfe0dd0 @ sig is 0xd00dfeed big endian
343 bne dtb_check_done @ not found
345 #ifdef CONFIG_ARM_ATAG_DTB_COMPAT
347 * OK... Let's do some funky business here.
348 * If we do have a DTB appended to zImage, and we do have
349 * an ATAG list around, we want the later to be translated
350 * and folded into the former here. No GOT fixup has occurred
351 * yet, but none of the code we're about to call uses any
355 /* Get the initial DTB size */
358 /* convert to little endian */
359 eor r1, r5, r5, ror #16
360 bic r1, r1, #0x00ff0000
362 eor r5, r5, r1, lsr #8
364 /* 50% DTB growth should be good enough */
365 add r5, r5, r5, lsr #1
366 /* preserve 64-bit alignment */
369 /* clamp to 32KB min and 1MB max */
374 /* temporarily relocate the stack past the DTB work space */
377 stmfd sp!, {r0-r3, ip, lr}
384 * If returned value is 1, there is no ATAG at the location
385 * pointed by r8. Try the typical 0x100 offset from start
386 * of RAM and hope for the best.
389 sub r0, r4, #TEXT_OFFSET
396 ldmfd sp!, {r0-r3, ip, lr}
400 mov r8, r6 @ use the appended device tree
403 * Make sure that the DTB doesn't end up in the final
404 * kernel's .bss area. To do so, we adjust the decompressed
405 * kernel size to compensate if that .bss size is larger
406 * than the relocated code.
408 ldr r5, =_kernel_bss_size
409 adr r1, wont_overwrite
414 /* Get the current DTB size */
417 /* convert r5 (dtb size) to little endian */
418 eor r1, r5, r5, ror #16
419 bic r1, r1, #0x00ff0000
421 eor r5, r5, r1, lsr #8
424 /* preserve 64-bit alignment */
428 /* relocate some pointers past the appended dtb */
436 * Check to see if we will overwrite ourselves.
437 * r4 = final kernel address (possibly with LSB set)
438 * r9 = size of decompressed image
439 * r10 = end of this image, including bss/stack/malloc space if non XIP
441 * r4 - 16k page directory >= r10 -> OK
442 * r4 + image length <= address of wont_overwrite -> OK
443 * Note: the possible LSB in r4 is harmless here.
449 adr r9, wont_overwrite
454 * Relocate ourselves past the end of the decompressed kernel.
456 * r10 = end of the decompressed kernel
457 * Because we always copy ahead, we need to do it from the end and go
458 * backward in case the source and destination overlap.
461 * Bump to the next 256-byte boundary with the size of
462 * the relocation code added. This avoids overwriting
463 * ourself when the offset is small.
465 add r10, r10, #((reloc_code_end - restart + 256) & ~255)
468 /* Get start of code we want to copy and align it down. */
472 /* Relocate the hyp vector base if necessary */
473 #ifdef CONFIG_ARM_VIRT_EXT
475 and r0, r0, #MODE_MASK
480 * Compute the address of the hyp vectors after relocation.
481 * This requires some arithmetic since we cannot directly
482 * reference __hyp_stub_vectors in a PC-relative way.
483 * Call __hyp_set_vectors with the new address so that we
484 * can HVC again after the copy.
487 movw r1, #:lower16:__hyp_stub_vectors - 0b
488 movt r1, #:upper16:__hyp_stub_vectors - 0b
496 sub r9, r6, r5 @ size to copy
497 add r9, r9, #31 @ rounded up to a multiple
498 bic r9, r9, #31 @ ... of 32 bytes
506 * We are about to copy the kernel to a new memory area.
507 * The boundaries of the new memory area can be found in
508 * r10 and r9, whilst r5 and r6 contain the boundaries
509 * of the memory we are going to copy.
510 * Calling dbgkc will help with the printing of this
513 dbgkc r5, r6, r10, r9
516 1: ldmdb r6!, {r0 - r3, r10 - r12, lr}
518 stmdb r9!, {r0 - r3, r10 - r12, lr}
521 /* Preserve offset to relocated code. */
524 #ifndef CONFIG_ZBOOT_ROM
525 /* cache_clean_flush may use the stack, so relocate it */
537 * If delta is zero, we are running at the address we were linked at.
541 * r4 = kernel execution address (possibly with LSB set)
542 * r5 = appended dtb size (0 if not present)
543 * r7 = architecture ID
555 #ifndef CONFIG_ZBOOT_ROM
557 * If we're running fully PIC === CONFIG_ZBOOT_ROM = n,
558 * we need to fix up pointers into the BSS region.
559 * Note that the stack pointer has already been fixed up.
565 * Relocate all entries in the GOT table.
566 * Bump bss entries to _edata + dtb size
568 1: ldr r1, [r11, #0] @ relocate entries in the GOT
569 add r1, r1, r0 @ This fixes up C references
570 cmp r1, r2 @ if entry >= bss_start &&
571 cmphs r3, r1 @ bss_end > entry
572 addhi r1, r1, r5 @ entry += dtb size
573 str r1, [r11], #4 @ next entry
577 /* bump our bss pointers too */
584 * Relocate entries in the GOT table. We only relocate
585 * the entries that are outside the (relocated) BSS region.
587 1: ldr r1, [r11, #0] @ relocate entries in the GOT
588 cmp r1, r2 @ entry < bss_start ||
589 cmphs r3, r1 @ _end < entry
590 addlo r1, r1, r0 @ table. This fixes up the
591 str r1, [r11], #4 @ C references.
596 not_relocated: mov r0, #0
597 1: str r0, [r2], #4 @ clear bss
605 * Did we skip the cache setup earlier?
606 * That is indicated by the LSB in r4.
614 * The C runtime environment should now be setup sufficiently.
615 * Set up some pointers, and start decompressing.
616 * r4 = kernel execution address
617 * r7 = architecture ID
621 mov r1, sp @ malloc space above stack
622 add r2, sp, #0x10000 @ 64k max
628 #ifdef CONFIG_ARM_VIRT_EXT
629 mrs r0, spsr @ Get saved CPU boot mode
630 and r0, r0, #MODE_MASK
631 cmp r0, #HYP_MODE @ if not booted in HYP mode...
632 bne __enter_kernel @ boot kernel directly
634 adr r12, .L__hyp_reentry_vectors_offset
639 __HVC(0) @ otherwise bounce to hyp mode
641 b . @ should never be reached
644 .L__hyp_reentry_vectors_offset: .long __hyp_reentry_vectors - .
652 .word __bss_start @ r2
655 .word input_data_end - 4 @ r10 (inflated size location)
656 .word _got_start @ r11
658 .word .L_user_stack_end @ sp
659 .word _end - restart + 16384 + 1024*1024
662 #ifdef CONFIG_ARCH_RPC
664 params: ldr r0, =0x10000100 @ params_phys for RPC
671 * Turn on the cache. We need to setup some page tables so that we
672 * can have both the I and D caches on.
674 * We place the page tables 16k down from the kernel execution address,
675 * and we hope that nothing else is using it. If we're using it, we
679 * r4 = kernel execution address
680 * r7 = architecture number
683 * r0, r1, r2, r3, r9, r10, r12 corrupted
684 * This routine must preserve:
688 cache_on: mov r3, #8 @ cache_on function
692 * Initialize the highest priority protection region, PR7
693 * to cover all 32bit address and cacheable and bufferable.
695 __armv4_mpu_cache_on:
696 mov r0, #0x3f @ 4G, the whole
697 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
698 mcr p15, 0, r0, c6, c7, 1
701 mcr p15, 0, r0, c2, c0, 0 @ D-cache on
702 mcr p15, 0, r0, c2, c0, 1 @ I-cache on
703 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
706 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
707 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
710 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
711 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
712 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
713 mrc p15, 0, r0, c1, c0, 0 @ read control reg
714 @ ...I .... ..D. WC.M
715 orr r0, r0, #0x002d @ .... .... ..1. 11.1
716 orr r0, r0, #0x1000 @ ...1 .... .... ....
718 mcr p15, 0, r0, c1, c0, 0 @ write control reg
721 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
722 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
725 __armv3_mpu_cache_on:
726 mov r0, #0x3f @ 4G, the whole
727 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
730 mcr p15, 0, r0, c2, c0, 0 @ cache on
731 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
734 mcr p15, 0, r0, c5, c0, 0 @ access permission
737 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
739 * ?? ARMv3 MMU does not allow reading the control register,
740 * does this really work on ARMv3 MPU?
742 mrc p15, 0, r0, c1, c0, 0 @ read control reg
743 @ .... .... .... WC.M
744 orr r0, r0, #0x000d @ .... .... .... 11.1
745 /* ?? this overwrites the value constructed above? */
747 mcr p15, 0, r0, c1, c0, 0 @ write control reg
749 /* ?? invalidate for the second time? */
750 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
753 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
759 __setup_mmu: sub r3, r4, #16384 @ Page directory size
760 bic r3, r3, #0xff @ Align the pointer
763 * Initialise the page tables, turning on the cacheable and bufferable
764 * bits for the RAM area only.
768 mov r9, r9, lsl #18 @ start of RAM
769 add r10, r9, #0x10000000 @ a reasonable RAM size
770 mov r1, #0x12 @ XN|U + section mapping
771 orr r1, r1, #3 << 10 @ AP=11
773 1: cmp r1, r9 @ if virt > start of RAM
774 cmphs r10, r1 @ && end of RAM > virt
775 bic r1, r1, #0x1c @ clear XN|U + C + B
776 orrlo r1, r1, #0x10 @ Set XN|U for non-RAM
777 orrhs r1, r1, r6 @ set RAM section settings
778 str r1, [r0], #4 @ 1:1 mapping
783 * If ever we are running from Flash, then we surely want the cache
784 * to be enabled also for our execution instance... We map 2MB of it
785 * so there is no map overlap problem for up to 1 MB compressed kernel.
786 * If the execution is in RAM then we would only be duplicating the above.
788 orr r1, r6, #0x04 @ ensure B is set for this
792 orr r1, r1, r2, lsl #20
793 add r0, r3, r2, lsl #2
800 @ Enable unaligned access on v6, to allow better code generation
801 @ for the decompressor C code:
802 __armv6_mmu_cache_on:
803 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
804 bic r0, r0, #2 @ A (no unaligned access fault)
805 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
806 mcr p15, 0, r0, c1, c0, 0 @ write SCTLR
807 b __armv4_mmu_cache_on
809 __arm926ejs_mmu_cache_on:
810 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
811 mov r0, #4 @ put dcache in WT mode
812 mcr p15, 7, r0, c15, c0, 0
815 __armv4_mmu_cache_on:
818 mov r6, #CB_BITS | 0x12 @ U
821 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
822 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
823 mrc p15, 0, r0, c1, c0, 0 @ read control reg
824 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
826 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
827 bl __common_mmu_cache_on
829 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
833 __armv7_mmu_cache_on:
834 enable_cp15_barriers r11
837 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
839 movne r6, #CB_BITS | 0x02 @ !XN
842 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
844 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
846 mrc p15, 0, r0, c1, c0, 0 @ read control reg
847 bic r0, r0, #1 << 28 @ clear SCTLR.TRE
848 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
849 orr r0, r0, #0x003c @ write buffer
850 bic r0, r0, #2 @ A (no unaligned access fault)
851 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
852 @ (needed for ARM1176)
854 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
855 mrcne p15, 0, r6, c2, c0, 2 @ read ttb control reg
856 orrne r0, r0, #1 @ MMU enabled
857 movne r1, #0xfffffffd @ domain 0 = client
858 bic r6, r6, #1 << 31 @ 32-bit translation system
859 bic r6, r6, #(7 << 0) | (1 << 4) @ use only ttbr0
860 mcrne p15, 0, r3, c2, c0, 0 @ load page table pointer
861 mcrne p15, 0, r1, c3, c0, 0 @ load domain access control
862 mcrne p15, 0, r6, c2, c0, 2 @ load ttb control
864 mcr p15, 0, r0, c7, c5, 4 @ ISB
865 mcr p15, 0, r0, c1, c0, 0 @ load control register
866 mrc p15, 0, r0, c1, c0, 0 @ and read it back
868 mcr p15, 0, r0, c7, c5, 4 @ ISB
873 mov r6, #CB_BITS | 0x12 @ U
876 mcr p15, 0, r0, c7, c7, 0 @ Invalidate whole cache
877 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
878 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
879 mrc p15, 0, r0, c1, c0, 0 @ read control reg
880 orr r0, r0, #0x1000 @ I-cache enable
881 bl __common_mmu_cache_on
883 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
886 __common_mmu_cache_on:
887 #ifndef CONFIG_THUMB2_KERNEL
889 orr r0, r0, #0x000d @ Write buffer, mmu
892 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
893 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
895 .align 5 @ cache line aligned
896 1: mcr p15, 0, r0, c1, c0, 0 @ load control register
897 mrc p15, 0, r0, c1, c0, 0 @ and read it back to
898 sub pc, lr, r0, lsr #32 @ properly flush pipeline
901 #define PROC_ENTRY_SIZE (4*5)
904 * Here follow the relocatable cache support functions for the
905 * various processors. This is a generic hook for locating an
906 * entry and jumping to an instruction at the specified offset
907 * from the start of the block. Please note this is all position
917 call_cache_fn: adr r12, proc_types
918 #ifdef CONFIG_CPU_CP15
919 mrc p15, 0, r9, c0, c0 @ get processor ID
920 #elif defined(CONFIG_CPU_V7M)
922 * On v7-M the processor id is located in the V7M_SCB_CPUID
923 * register, but as cache handling is IMPLEMENTATION DEFINED on
924 * v7-M (if existant at all) we just return early here.
925 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
926 * __armv7_mmu_cache_{on,off,flush}) would be selected which
927 * use cp15 registers that are not implemented on v7-M.
931 ldr r9, =CONFIG_PROCESSOR_ID
933 1: ldr r1, [r12, #0] @ get value
934 ldr r2, [r12, #4] @ get mask
935 eor r1, r1, r9 @ (real ^ match)
937 ARM( addeq pc, r12, r3 ) @ call cache function
938 THUMB( addeq r12, r3 )
939 THUMB( moveq pc, r12 ) @ call cache function
940 add r12, r12, #PROC_ENTRY_SIZE
944 * Table for cache operations. This is basically:
947 * - 'cache on' method instruction
948 * - 'cache off' method instruction
949 * - 'cache flush' method instruction
951 * We match an entry using: ((real_id ^ match) & mask) == 0
953 * Writethrough caches generally only need 'on' and 'off'
954 * methods. Writeback caches _must_ have the flush method
958 .type proc_types,#object
960 .word 0x41000000 @ old ARM ID
969 .word 0x41007000 @ ARM7/710
978 .word 0x41807200 @ ARM720T (writethrough)
980 W(b) __armv4_mmu_cache_on
981 W(b) __armv4_mmu_cache_off
985 .word 0x41007400 @ ARM74x
987 W(b) __armv3_mpu_cache_on
988 W(b) __armv3_mpu_cache_off
989 W(b) __armv3_mpu_cache_flush
991 .word 0x41009400 @ ARM94x
993 W(b) __armv4_mpu_cache_on
994 W(b) __armv4_mpu_cache_off
995 W(b) __armv4_mpu_cache_flush
997 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
999 W(b) __arm926ejs_mmu_cache_on
1000 W(b) __armv4_mmu_cache_off
1001 W(b) __armv5tej_mmu_cache_flush
1003 .word 0x00007000 @ ARM7 IDs
1012 @ Everything from here on will be the new ID system.
1014 .word 0x4401a100 @ sa110 / sa1100
1016 W(b) __armv4_mmu_cache_on
1017 W(b) __armv4_mmu_cache_off
1018 W(b) __armv4_mmu_cache_flush
1020 .word 0x6901b110 @ sa1110
1022 W(b) __armv4_mmu_cache_on
1023 W(b) __armv4_mmu_cache_off
1024 W(b) __armv4_mmu_cache_flush
1027 .word 0xffffff00 @ PXA9xx
1028 W(b) __armv4_mmu_cache_on
1029 W(b) __armv4_mmu_cache_off
1030 W(b) __armv4_mmu_cache_flush
1032 .word 0x56158000 @ PXA168
1034 W(b) __armv4_mmu_cache_on
1035 W(b) __armv4_mmu_cache_off
1036 W(b) __armv5tej_mmu_cache_flush
1038 .word 0x56050000 @ Feroceon
1040 W(b) __armv4_mmu_cache_on
1041 W(b) __armv4_mmu_cache_off
1042 W(b) __armv5tej_mmu_cache_flush
1044 #ifdef CONFIG_CPU_FEROCEON_OLD_ID
1045 /* this conflicts with the standard ARMv5TE entry */
1046 .long 0x41009260 @ Old Feroceon
1048 b __armv4_mmu_cache_on
1049 b __armv4_mmu_cache_off
1050 b __armv5tej_mmu_cache_flush
1053 .word 0x66015261 @ FA526
1055 W(b) __fa526_cache_on
1056 W(b) __armv4_mmu_cache_off
1057 W(b) __fa526_cache_flush
1059 @ These match on the architecture ID
1061 .word 0x00020000 @ ARMv4T
1063 W(b) __armv4_mmu_cache_on
1064 W(b) __armv4_mmu_cache_off
1065 W(b) __armv4_mmu_cache_flush
1067 .word 0x00050000 @ ARMv5TE
1069 W(b) __armv4_mmu_cache_on
1070 W(b) __armv4_mmu_cache_off
1071 W(b) __armv4_mmu_cache_flush
1073 .word 0x00060000 @ ARMv5TEJ
1075 W(b) __armv4_mmu_cache_on
1076 W(b) __armv4_mmu_cache_off
1077 W(b) __armv5tej_mmu_cache_flush
1079 .word 0x0007b000 @ ARMv6
1081 W(b) __armv6_mmu_cache_on
1082 W(b) __armv4_mmu_cache_off
1083 W(b) __armv6_mmu_cache_flush
1085 .word 0x000f0000 @ new CPU Id
1087 W(b) __armv7_mmu_cache_on
1088 W(b) __armv7_mmu_cache_off
1089 W(b) __armv7_mmu_cache_flush
1091 .word 0 @ unrecognised type
1100 .size proc_types, . - proc_types
1103 * If you get a "non-constant expression in ".if" statement"
1104 * error from the assembler on this line, check that you have
1105 * not accidentally written a "b" instruction where you should
1106 * have written W(b).
1108 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1109 .error "The size of one or more proc_types entries is wrong."
1113 * Turn off the Cache and MMU. ARMv3 does not support
1114 * reading the control register, but ARMv4 does.
1117 * r0, r1, r2, r3, r9, r12 corrupted
1118 * This routine must preserve:
1122 cache_off: mov r3, #12 @ cache_off function
1125 __armv4_mpu_cache_off:
1126 mrc p15, 0, r0, c1, c0
1128 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
1130 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
1131 mcr p15, 0, r0, c7, c6, 0 @ flush D-Cache
1132 mcr p15, 0, r0, c7, c5, 0 @ flush I-Cache
1135 __armv3_mpu_cache_off:
1136 mrc p15, 0, r0, c1, c0
1138 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1140 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1143 __armv4_mmu_cache_off:
1145 mrc p15, 0, r0, c1, c0
1147 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1149 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1150 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1154 __armv7_mmu_cache_off:
1155 mrc p15, 0, r0, c1, c0
1161 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1163 bl __armv7_mmu_cache_flush
1166 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
1168 mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC
1169 mcr p15, 0, r0, c7, c10, 4 @ DSB
1170 mcr p15, 0, r0, c7, c5, 4 @ ISB
1174 * Clean and flush the cache to maintain consistency.
1177 * r1, r2, r3, r9, r10, r11, r12 corrupted
1178 * This routine must preserve:
1186 __armv4_mpu_cache_flush:
1191 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1192 mov r1, #7 << 5 @ 8 segments
1193 1: orr r3, r1, #63 << 26 @ 64 entries
1194 2: mcr p15, 0, r3, c7, c14, 2 @ clean & invalidate D index
1195 subs r3, r3, #1 << 26
1196 bcs 2b @ entries 63 to 0
1197 subs r1, r1, #1 << 5
1198 bcs 1b @ segments 7 to 0
1201 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1202 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1205 __fa526_cache_flush:
1209 mcr p15, 0, r1, c7, c14, 0 @ clean and invalidate D cache
1210 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1211 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1214 __armv6_mmu_cache_flush:
1217 mcreq p15, 0, r1, c7, c14, 0 @ clean+invalidate D
1218 mcr p15, 0, r1, c7, c5, 0 @ invalidate I+BTB
1219 mcreq p15, 0, r1, c7, c15, 0 @ clean+invalidate unified
1220 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1223 __armv7_mmu_cache_flush:
1224 enable_cp15_barriers r10
1227 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1228 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1231 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
1234 mcr p15, 0, r10, c7, c10, 5 @ DMB
1235 stmfd sp!, {r0-r7, r9-r11}
1236 mrc p15, 1, r0, c0, c0, 1 @ read clidr
1237 ands r3, r0, #0x7000000 @ extract loc from clidr
1238 mov r3, r3, lsr #23 @ left align loc bit field
1239 beq finished @ if loc is 0, then no need to clean
1240 mov r10, #0 @ start clean at cache level 0
1242 add r2, r10, r10, lsr #1 @ work out 3x current cache level
1243 mov r1, r0, lsr r2 @ extract cache type bits from clidr
1244 and r1, r1, #7 @ mask of the bits for current cache only
1245 cmp r1, #2 @ see what cache we have at this level
1246 blt skip @ skip if no cache, or just i-cache
1247 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1248 mcr p15, 0, r10, c7, c5, 4 @ isb to sych the new cssr&csidr
1249 mrc p15, 1, r1, c0, c0, 0 @ read the new csidr
1250 and r2, r1, #7 @ extract the length of the cache lines
1251 add r2, r2, #4 @ add 4 (line length offset)
1253 ands r4, r4, r1, lsr #3 @ find maximum number on the way size
1254 clz r5, r4 @ find bit position of way size increment
1256 ands r7, r7, r1, lsr #13 @ extract max number of the index size
1258 mov r9, r4 @ create working copy of max way size
1260 ARM( orr r11, r10, r9, lsl r5 ) @ factor way and cache number into r11
1261 ARM( orr r11, r11, r7, lsl r2 ) @ factor index number into r11
1262 THUMB( lsl r6, r9, r5 )
1263 THUMB( orr r11, r10, r6 ) @ factor way and cache number into r11
1264 THUMB( lsl r6, r7, r2 )
1265 THUMB( orr r11, r11, r6 ) @ factor index number into r11
1266 mcr p15, 0, r11, c7, c14, 2 @ clean & invalidate by set/way
1267 subs r9, r9, #1 @ decrement the way
1269 subs r7, r7, #1 @ decrement the index
1272 add r10, r10, #2 @ increment cache number
1276 ldmfd sp!, {r0-r7, r9-r11}
1277 mov r10, #0 @ switch back to cache level 0
1278 mcr p15, 2, r10, c0, c0, 0 @ select current cache level in cssr
1280 mcr p15, 0, r10, c7, c10, 4 @ DSB
1281 mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB
1282 mcr p15, 0, r10, c7, c10, 4 @ DSB
1283 mcr p15, 0, r10, c7, c5, 4 @ ISB
1286 __armv5tej_mmu_cache_flush:
1289 1: mrc p15, 0, APSR_nzcv, c7, c14, 3 @ test,clean,invalidate D cache
1291 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1292 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1295 __armv4_mmu_cache_flush:
1298 mov r2, #64*1024 @ default: 32K dcache size (*2)
1299 mov r11, #32 @ default: 32 byte line size
1300 mrc p15, 0, r3, c0, c0, 1 @ read cache type
1301 teq r3, r9 @ cache ID register present?
1306 mov r2, r2, lsl r1 @ base dcache size *2
1307 tst r3, #1 << 14 @ test M bit
1308 addne r2, r2, r2, lsr #1 @ +1/2 size if M == 1
1312 mov r11, r11, lsl r3 @ cache line size in bytes
1315 bic r1, r1, #63 @ align to longest cache line
1318 ARM( ldr r3, [r1], r11 ) @ s/w flush D cache
1319 THUMB( ldr r3, [r1] ) @ s/w flush D cache
1320 THUMB( add r1, r1, r11 )
1324 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1325 mcr p15, 0, r1, c7, c6, 0 @ flush D cache
1326 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1329 __armv3_mmu_cache_flush:
1330 __armv3_mpu_cache_flush:
1334 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1338 * Various debugging routines for printing hex characters and
1339 * memory, which again must be relocatable.
1343 .type phexbuf,#object
1345 .size phexbuf, . - phexbuf
1347 @ phex corrupts {r0, r1, r2, r3}
1348 phex: adr r3, phexbuf
1362 @ puts corrupts {r0, r1, r2, r3}
1363 puts: loadsp r3, r2, r1
1364 1: ldrb r2, [r0], #1
1377 @ putc corrupts {r0, r1, r2, r3}
1384 @ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1385 memdump: mov r12, r0
1388 2: mov r0, r11, lsl #2
1396 ldr r0, [r12, r11, lsl #2]
1416 #ifdef CONFIG_ARM_VIRT_EXT
1418 __hyp_reentry_vectors:
1424 W(b) __enter_kernel @ hyp
1427 #endif /* CONFIG_ARM_VIRT_EXT */
1430 mov r0, #0 @ must be 0
1431 mov r1, r7 @ restore architecture number
1432 mov r2, r8 @ restore atags pointer
1433 ARM( mov pc, r4 ) @ call kernel
1434 M_CLASS( add r4, r4, #1 ) @ enter in Thumb mode for M class
1435 THUMB( bx r4 ) @ entry point is always ARM for A/R classes
1439 #ifdef CONFIG_EFI_STUB
1441 _start: .long start - .
1443 ENTRY(efi_stub_entry)
1444 @ allocate space on stack for passing current zImage address
1445 @ and for the EFI stub to return of new entry point of
1446 @ zImage, as EFI stub may copy the kernel. Pointer address
1447 @ is passed in r2. r0 and r1 are passed through from the
1448 @ EFI firmware to efi_entry
1453 mov r2, sp @ pass zImage address in r2
1456 @ Check for error return from EFI stub. r0 has FDT address
1461 @ Preserve return value of efi_entry() in r4
1463 bl cache_clean_flush
1466 @ Set parameters for booting zImage according to boot protocol
1467 @ put FDT address in r2, it was returned by efi_entry()
1468 @ r1 is the machine type, and r0 needs to be 0
1473 @ Branch to (possibly) relocated zImage that is in [sp]
1475 ldr ip, =start_offset
1477 mov pc, lr @ no mode switch
1480 @ Return EFI_LOAD_ERROR to EFI firmware on error.
1483 ENDPROC(efi_stub_entry)
1487 .section ".stack", "aw", %nobits
1488 .L_user_stack: .space 4096