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 )
155 * The kernel build system appends the size of the
156 * decompressed kernel at the end of the compressed data
157 * in little-endian form.
159 .macro get_inflated_image_size, res:req, tmp1:req, tmp2:req
160 adr \res, .Linflated_image_size_offset
162 add \tmp1, \tmp1, \res @ address of inflated image size
164 ldrb \res, [\tmp1] @ get_unaligned_le32
165 ldrb \tmp2, [\tmp1, #1]
166 orr \res, \res, \tmp2, lsl #8
167 ldrb \tmp2, [\tmp1, #2]
168 ldrb \tmp1, [\tmp1, #3]
169 orr \res, \res, \tmp2, lsl #16
170 orr \res, \res, \tmp1, lsl #24
173 .section ".start", "ax"
175 * sort out different calling conventions
179 * Always enter in ARM state for CPUs that support the ARM ISA.
180 * As of today (2014) that's exactly the members of the A and R
185 .type start,#function
187 * These 7 nops along with the 1 nop immediately below for
188 * !THUMB2 form 8 nops that make the compressed kernel bootable
189 * on legacy ARM systems that were assuming the kernel in a.out
190 * binary format. The boot loaders on these systems would
191 * jump 32 bytes into the image to skip the a.out header.
192 * with these 8 nops filling exactly 32 bytes, things still
193 * work as expected on these legacy systems. Thumb2 mode keeps
194 * 7 of the nops as it turns out that some boot loaders
195 * were patching the initial instructions of the kernel, i.e
196 * had started to exploit this "patch area".
201 #ifndef CONFIG_THUMB2_KERNEL
204 AR_CLASS( sub pc, pc, #3 ) @ A/R: switch to Thumb2 mode
205 M_CLASS( nop.w ) @ M: already in Thumb2 mode
210 .word _magic_sig @ Magic numbers to help the loader
211 .word _magic_start @ absolute load/run zImage address
212 .word _magic_end @ zImage end address
213 .word 0x04030201 @ endianness flag
214 .word 0x45454545 @ another magic number to indicate
215 .word _magic_table @ additional data table
219 ARM_BE8( setend be ) @ go BE8 if compiled for BE8
220 AR_CLASS( mrs r9, cpsr )
221 #ifdef CONFIG_ARM_VIRT_EXT
222 bl __hyp_stub_install @ get into SVC mode, reversibly
224 mov r7, r1 @ save architecture ID
225 mov r8, r2 @ save atags pointer
227 #ifndef CONFIG_CPU_V7M
229 * Booting from Angel - need to enter SVC mode and disable
230 * FIQs/IRQs (numeric definitions from angel arm.h source).
231 * We only do this if we were in user mode on entry.
233 mrs r2, cpsr @ get current mode
234 tst r2, #3 @ not user?
236 mov r0, #0x17 @ angel_SWIreason_EnterSVC
237 ARM( swi 0x123456 ) @ angel_SWI_ARM
238 THUMB( svc 0xab ) @ angel_SWI_THUMB
240 safe_svcmode_maskall r0
241 msr spsr_cxsf, r9 @ Save the CPU boot mode in
245 * Note that some cache flushing and other stuff may
246 * be needed here - is there an Angel SWI call for this?
250 * some architecture specific code can be inserted
251 * by the linker here, but it should preserve r7, r8, and r9.
256 #ifdef CONFIG_AUTO_ZRELADDR
258 * Find the start of physical memory. As we are executing
259 * without the MMU on, we are in the physical address space.
260 * We just need to get rid of any offset by aligning the
263 * This alignment is a balance between the requirements of
264 * different platforms - we have chosen 128MB to allow
265 * platforms which align the start of their physical memory
266 * to 128MB to use this feature, while allowing the zImage
267 * to be placed within the first 128MB of memory on other
268 * platforms. Increasing the alignment means we place
269 * stricter alignment requirements on the start of physical
270 * memory, but relaxing it means that we break people who
271 * are already placing their zImage in (eg) the top 64MB
275 and r4, r4, #0xf8000000
276 /* Determine final kernel image address. */
277 add r4, r4, #TEXT_OFFSET
283 * Set up a page table only if it won't overwrite ourself.
284 * That means r4 < pc || r4 - 16k page directory > &_end.
285 * Given that r4 > &_end is most unfrequent, we add a rough
286 * additional 1MB of room for a possible appended DTB.
293 orrcc r4, r4, #1 @ remember we skipped cache_on
297 ldmia r0, {r1, r2, r3, r6, r11, r12}
301 * We might be running at a different address. We need
302 * to fix up various pointers.
304 sub r0, r0, r1 @ calculate the delta offset
305 add r6, r6, r0 @ _edata
307 get_inflated_image_size r9, r10, lr
309 #ifndef CONFIG_ZBOOT_ROM
310 /* malloc space is above the relocated stack (64k max) */
312 add r10, sp, #0x10000
315 * With ZBOOT_ROM the bss/stack is non relocatable,
316 * but someone could still run this code from RAM,
317 * in which case our reference is _edata.
322 mov r5, #0 @ init dtb size to 0
323 #ifdef CONFIG_ARM_APPENDED_DTB
328 * r4 = final kernel address (possibly with LSB set)
329 * r5 = appended dtb size (still unknown)
331 * r7 = architecture ID
332 * r8 = atags/device tree pointer
333 * r9 = size of decompressed image
334 * r10 = end of this image, including bss/stack/malloc space if non XIP
339 * if there are device trees (dtb) appended to zImage, advance r10 so that the
340 * dtb data will get relocated along with the kernel if necessary.
345 ldr r1, =0xedfe0dd0 @ sig is 0xd00dfeed big endian
350 bne dtb_check_done @ not found
352 #ifdef CONFIG_ARM_ATAG_DTB_COMPAT
354 * OK... Let's do some funky business here.
355 * If we do have a DTB appended to zImage, and we do have
356 * an ATAG list around, we want the later to be translated
357 * and folded into the former here. No GOT fixup has occurred
358 * yet, but none of the code we're about to call uses any
362 /* Get the initial DTB size */
365 /* convert to little endian */
366 eor r1, r5, r5, ror #16
367 bic r1, r1, #0x00ff0000
369 eor r5, r5, r1, lsr #8
371 /* 50% DTB growth should be good enough */
372 add r5, r5, r5, lsr #1
373 /* preserve 64-bit alignment */
376 /* clamp to 32KB min and 1MB max */
381 /* temporarily relocate the stack past the DTB work space */
384 stmfd sp!, {r0-r3, ip, lr}
391 * If returned value is 1, there is no ATAG at the location
392 * pointed by r8. Try the typical 0x100 offset from start
393 * of RAM and hope for the best.
396 sub r0, r4, #TEXT_OFFSET
403 ldmfd sp!, {r0-r3, ip, lr}
407 mov r8, r6 @ use the appended device tree
410 * Make sure that the DTB doesn't end up in the final
411 * kernel's .bss area. To do so, we adjust the decompressed
412 * kernel size to compensate if that .bss size is larger
413 * than the relocated code.
415 ldr r5, =_kernel_bss_size
416 adr r1, wont_overwrite
421 /* Get the current DTB size */
424 /* convert r5 (dtb size) to little endian */
425 eor r1, r5, r5, ror #16
426 bic r1, r1, #0x00ff0000
428 eor r5, r5, r1, lsr #8
431 /* preserve 64-bit alignment */
435 /* relocate some pointers past the appended dtb */
443 * Check to see if we will overwrite ourselves.
444 * r4 = final kernel address (possibly with LSB set)
445 * r9 = size of decompressed image
446 * r10 = end of this image, including bss/stack/malloc space if non XIP
448 * r4 - 16k page directory >= r10 -> OK
449 * r4 + image length <= address of wont_overwrite -> OK
450 * Note: the possible LSB in r4 is harmless here.
456 adr r9, wont_overwrite
461 * Relocate ourselves past the end of the decompressed kernel.
463 * r10 = end of the decompressed kernel
464 * Because we always copy ahead, we need to do it from the end and go
465 * backward in case the source and destination overlap.
468 * Bump to the next 256-byte boundary with the size of
469 * the relocation code added. This avoids overwriting
470 * ourself when the offset is small.
472 add r10, r10, #((reloc_code_end - restart + 256) & ~255)
475 /* Get start of code we want to copy and align it down. */
479 /* Relocate the hyp vector base if necessary */
480 #ifdef CONFIG_ARM_VIRT_EXT
482 and r0, r0, #MODE_MASK
487 * Compute the address of the hyp vectors after relocation.
488 * This requires some arithmetic since we cannot directly
489 * reference __hyp_stub_vectors in a PC-relative way.
490 * Call __hyp_set_vectors with the new address so that we
491 * can HVC again after the copy.
494 movw r1, #:lower16:__hyp_stub_vectors - 0b
495 movt r1, #:upper16:__hyp_stub_vectors - 0b
503 sub r9, r6, r5 @ size to copy
504 add r9, r9, #31 @ rounded up to a multiple
505 bic r9, r9, #31 @ ... of 32 bytes
513 * We are about to copy the kernel to a new memory area.
514 * The boundaries of the new memory area can be found in
515 * r10 and r9, whilst r5 and r6 contain the boundaries
516 * of the memory we are going to copy.
517 * Calling dbgkc will help with the printing of this
520 dbgkc r5, r6, r10, r9
523 1: ldmdb r6!, {r0 - r3, r10 - r12, lr}
525 stmdb r9!, {r0 - r3, r10 - r12, lr}
528 /* Preserve offset to relocated code. */
531 mov r0, r9 @ start of relocated zImage
532 add r1, sp, r6 @ end of relocated zImage
541 * If delta is zero, we are running at the address we were linked at.
545 * r4 = kernel execution address (possibly with LSB set)
546 * r5 = appended dtb size (0 if not present)
547 * r7 = architecture ID
559 #ifndef CONFIG_ZBOOT_ROM
561 * If we're running fully PIC === CONFIG_ZBOOT_ROM = n,
562 * we need to fix up pointers into the BSS region.
563 * Note that the stack pointer has already been fixed up.
569 * Relocate all entries in the GOT table.
570 * Bump bss entries to _edata + dtb size
572 1: ldr r1, [r11, #0] @ relocate entries in the GOT
573 add r1, r1, r0 @ This fixes up C references
574 cmp r1, r2 @ if entry >= bss_start &&
575 cmphs r3, r1 @ bss_end > entry
576 addhi r1, r1, r5 @ entry += dtb size
577 str r1, [r11], #4 @ next entry
581 /* bump our bss pointers too */
588 * Relocate entries in the GOT table. We only relocate
589 * the entries that are outside the (relocated) BSS region.
591 1: ldr r1, [r11, #0] @ relocate entries in the GOT
592 cmp r1, r2 @ entry < bss_start ||
593 cmphs r3, r1 @ _end < entry
594 addlo r1, r1, r0 @ table. This fixes up the
595 str r1, [r11], #4 @ C references.
600 not_relocated: mov r0, #0
601 1: str r0, [r2], #4 @ clear bss
609 * Did we skip the cache setup earlier?
610 * That is indicated by the LSB in r4.
618 * The C runtime environment should now be setup sufficiently.
619 * Set up some pointers, and start decompressing.
620 * r4 = kernel execution address
621 * r7 = architecture ID
625 mov r1, sp @ malloc space above stack
626 add r2, sp, #0x10000 @ 64k max
630 get_inflated_image_size r1, r2, r3
632 mov r0, r4 @ start of inflated image
633 add r1, r1, r0 @ end of inflated image
637 #ifdef CONFIG_ARM_VIRT_EXT
638 mrs r0, spsr @ Get saved CPU boot mode
639 and r0, r0, #MODE_MASK
640 cmp r0, #HYP_MODE @ if not booted in HYP mode...
641 bne __enter_kernel @ boot kernel directly
643 adr r12, .L__hyp_reentry_vectors_offset
648 __HVC(0) @ otherwise bounce to hyp mode
650 b . @ should never be reached
653 .L__hyp_reentry_vectors_offset: .long __hyp_reentry_vectors - .
661 .word __bss_start @ r2
664 .word _got_start @ r11
666 .word .L_user_stack_end @ sp
667 .word _end - restart + 16384 + 1024*1024
670 .Linflated_image_size_offset:
671 .long (input_data_end - 4) - .
673 #ifdef CONFIG_ARCH_RPC
675 params: ldr r0, =0x10000100 @ params_phys for RPC
682 * dcache_line_size - get the minimum D-cache line size from the CTR register
685 .macro dcache_line_size, reg, tmp
686 #ifdef CONFIG_CPU_V7M
687 movw \tmp, #:lower16:BASEADDR_V7M_SCB + V7M_SCB_CTR
688 movt \tmp, #:upper16:BASEADDR_V7M_SCB + V7M_SCB_CTR
691 mrc p15, 0, \tmp, c0, c0, 1 @ read ctr
694 and \tmp, \tmp, #0xf @ cache line size encoding
695 mov \reg, #4 @ bytes per word
696 mov \reg, \reg, lsl \tmp @ actual cache line size
700 * Turn on the cache. We need to setup some page tables so that we
701 * can have both the I and D caches on.
703 * We place the page tables 16k down from the kernel execution address,
704 * and we hope that nothing else is using it. If we're using it, we
708 * r4 = kernel execution address
709 * r7 = architecture number
712 * r0, r1, r2, r3, r9, r10, r12 corrupted
713 * This routine must preserve:
717 cache_on: mov r3, #8 @ cache_on function
721 * Initialize the highest priority protection region, PR7
722 * to cover all 32bit address and cacheable and bufferable.
724 __armv4_mpu_cache_on:
725 mov r0, #0x3f @ 4G, the whole
726 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
727 mcr p15, 0, r0, c6, c7, 1
730 mcr p15, 0, r0, c2, c0, 0 @ D-cache on
731 mcr p15, 0, r0, c2, c0, 1 @ I-cache on
732 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
735 mcr p15, 0, r0, c5, c0, 1 @ I-access permission
736 mcr p15, 0, r0, c5, c0, 0 @ D-access permission
739 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
740 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
741 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
742 mrc p15, 0, r0, c1, c0, 0 @ read control reg
743 @ ...I .... ..D. WC.M
744 orr r0, r0, #0x002d @ .... .... ..1. 11.1
745 orr r0, r0, #0x1000 @ ...1 .... .... ....
747 mcr p15, 0, r0, c1, c0, 0 @ write control reg
750 mcr p15, 0, r0, c7, c5, 0 @ flush(inval) I-Cache
751 mcr p15, 0, r0, c7, c6, 0 @ flush(inval) D-Cache
754 __armv3_mpu_cache_on:
755 mov r0, #0x3f @ 4G, the whole
756 mcr p15, 0, r0, c6, c7, 0 @ PR7 Area Setting
759 mcr p15, 0, r0, c2, c0, 0 @ cache on
760 mcr p15, 0, r0, c3, c0, 0 @ write-buffer on
763 mcr p15, 0, r0, c5, c0, 0 @ access permission
766 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
768 * ?? ARMv3 MMU does not allow reading the control register,
769 * does this really work on ARMv3 MPU?
771 mrc p15, 0, r0, c1, c0, 0 @ read control reg
772 @ .... .... .... WC.M
773 orr r0, r0, #0x000d @ .... .... .... 11.1
774 /* ?? this overwrites the value constructed above? */
776 mcr p15, 0, r0, c1, c0, 0 @ write control reg
778 /* ?? invalidate for the second time? */
779 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
782 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
788 __setup_mmu: sub r3, r4, #16384 @ Page directory size
789 bic r3, r3, #0xff @ Align the pointer
792 * Initialise the page tables, turning on the cacheable and bufferable
793 * bits for the RAM area only.
797 mov r9, r9, lsl #18 @ start of RAM
798 add r10, r9, #0x10000000 @ a reasonable RAM size
799 mov r1, #0x12 @ XN|U + section mapping
800 orr r1, r1, #3 << 10 @ AP=11
802 1: cmp r1, r9 @ if virt > start of RAM
803 cmphs r10, r1 @ && end of RAM > virt
804 bic r1, r1, #0x1c @ clear XN|U + C + B
805 orrlo r1, r1, #0x10 @ Set XN|U for non-RAM
806 orrhs r1, r1, r6 @ set RAM section settings
807 str r1, [r0], #4 @ 1:1 mapping
812 * If ever we are running from Flash, then we surely want the cache
813 * to be enabled also for our execution instance... We map 2MB of it
814 * so there is no map overlap problem for up to 1 MB compressed kernel.
815 * If the execution is in RAM then we would only be duplicating the above.
817 orr r1, r6, #0x04 @ ensure B is set for this
821 orr r1, r1, r2, lsl #20
822 add r0, r3, r2, lsl #2
829 @ Enable unaligned access on v6, to allow better code generation
830 @ for the decompressor C code:
831 __armv6_mmu_cache_on:
832 mrc p15, 0, r0, c1, c0, 0 @ read SCTLR
833 bic r0, r0, #2 @ A (no unaligned access fault)
834 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
835 mcr p15, 0, r0, c1, c0, 0 @ write SCTLR
836 b __armv4_mmu_cache_on
838 __arm926ejs_mmu_cache_on:
839 #ifdef CONFIG_CPU_DCACHE_WRITETHROUGH
840 mov r0, #4 @ put dcache in WT mode
841 mcr p15, 7, r0, c15, c0, 0
844 __armv4_mmu_cache_on:
847 mov r6, #CB_BITS | 0x12 @ U
850 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
851 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
852 mrc p15, 0, r0, c1, c0, 0 @ read control reg
853 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
855 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
856 bl __common_mmu_cache_on
858 mcr p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
862 __armv7_mmu_cache_on:
863 enable_cp15_barriers r11
866 mrc p15, 0, r11, c0, c1, 4 @ read ID_MMFR0
868 movne r6, #CB_BITS | 0x02 @ !XN
871 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
873 mcrne p15, 0, r0, c8, c7, 0 @ flush I,D TLBs
875 mrc p15, 0, r0, c1, c0, 0 @ read control reg
876 bic r0, r0, #1 << 28 @ clear SCTLR.TRE
877 orr r0, r0, #0x5000 @ I-cache enable, RR cache replacement
878 orr r0, r0, #0x003c @ write buffer
879 bic r0, r0, #2 @ A (no unaligned access fault)
880 orr r0, r0, #1 << 22 @ U (v6 unaligned access model)
881 @ (needed for ARM1176)
883 ARM_BE8( orr r0, r0, #1 << 25 ) @ big-endian page tables
884 mrcne p15, 0, r6, c2, c0, 2 @ read ttb control reg
885 orrne r0, r0, #1 @ MMU enabled
886 movne r1, #0xfffffffd @ domain 0 = client
887 bic r6, r6, #1 << 31 @ 32-bit translation system
888 bic r6, r6, #(7 << 0) | (1 << 4) @ use only ttbr0
889 mcrne p15, 0, r3, c2, c0, 0 @ load page table pointer
890 mcrne p15, 0, r1, c3, c0, 0 @ load domain access control
891 mcrne p15, 0, r6, c2, c0, 2 @ load ttb control
893 mcr p15, 0, r0, c7, c5, 4 @ ISB
894 mcr p15, 0, r0, c1, c0, 0 @ load control register
895 mrc p15, 0, r0, c1, c0, 0 @ and read it back
897 mcr p15, 0, r0, c7, c5, 4 @ ISB
902 mov r6, #CB_BITS | 0x12 @ U
905 mcr p15, 0, r0, c7, c7, 0 @ Invalidate whole cache
906 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
907 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
908 mrc p15, 0, r0, c1, c0, 0 @ read control reg
909 orr r0, r0, #0x1000 @ I-cache enable
910 bl __common_mmu_cache_on
912 mcr p15, 0, r0, c8, c7, 0 @ flush UTLB
915 __common_mmu_cache_on:
916 #ifndef CONFIG_THUMB2_KERNEL
918 orr r0, r0, #0x000d @ Write buffer, mmu
921 mcr p15, 0, r3, c2, c0, 0 @ load page table pointer
922 mcr p15, 0, r1, c3, c0, 0 @ load domain access control
924 .align 5 @ cache line aligned
925 1: mcr p15, 0, r0, c1, c0, 0 @ load control register
926 mrc p15, 0, r0, c1, c0, 0 @ and read it back to
927 sub pc, lr, r0, lsr #32 @ properly flush pipeline
930 #define PROC_ENTRY_SIZE (4*5)
933 * Here follow the relocatable cache support functions for the
934 * various processors. This is a generic hook for locating an
935 * entry and jumping to an instruction at the specified offset
936 * from the start of the block. Please note this is all position
946 call_cache_fn: adr r12, proc_types
947 #ifdef CONFIG_CPU_CP15
948 mrc p15, 0, r9, c0, c0 @ get processor ID
949 #elif defined(CONFIG_CPU_V7M)
951 * On v7-M the processor id is located in the V7M_SCB_CPUID
952 * register, but as cache handling is IMPLEMENTATION DEFINED on
953 * v7-M (if existant at all) we just return early here.
954 * If V7M_SCB_CPUID were used the cpu ID functions (i.e.
955 * __armv7_mmu_cache_{on,off,flush}) would be selected which
956 * use cp15 registers that are not implemented on v7-M.
960 ldr r9, =CONFIG_PROCESSOR_ID
962 1: ldr r1, [r12, #0] @ get value
963 ldr r2, [r12, #4] @ get mask
964 eor r1, r1, r9 @ (real ^ match)
966 ARM( addeq pc, r12, r3 ) @ call cache function
967 THUMB( addeq r12, r3 )
968 THUMB( moveq pc, r12 ) @ call cache function
969 add r12, r12, #PROC_ENTRY_SIZE
973 * Table for cache operations. This is basically:
976 * - 'cache on' method instruction
977 * - 'cache off' method instruction
978 * - 'cache flush' method instruction
980 * We match an entry using: ((real_id ^ match) & mask) == 0
982 * Writethrough caches generally only need 'on' and 'off'
983 * methods. Writeback caches _must_ have the flush method
987 .type proc_types,#object
989 .word 0x41000000 @ old ARM ID
998 .word 0x41007000 @ ARM7/710
1007 .word 0x41807200 @ ARM720T (writethrough)
1009 W(b) __armv4_mmu_cache_on
1010 W(b) __armv4_mmu_cache_off
1014 .word 0x41007400 @ ARM74x
1016 W(b) __armv3_mpu_cache_on
1017 W(b) __armv3_mpu_cache_off
1018 W(b) __armv3_mpu_cache_flush
1020 .word 0x41009400 @ ARM94x
1022 W(b) __armv4_mpu_cache_on
1023 W(b) __armv4_mpu_cache_off
1024 W(b) __armv4_mpu_cache_flush
1026 .word 0x41069260 @ ARM926EJ-S (v5TEJ)
1028 W(b) __arm926ejs_mmu_cache_on
1029 W(b) __armv4_mmu_cache_off
1030 W(b) __armv5tej_mmu_cache_flush
1032 .word 0x00007000 @ ARM7 IDs
1041 @ Everything from here on will be the new ID system.
1043 .word 0x4401a100 @ sa110 / sa1100
1045 W(b) __armv4_mmu_cache_on
1046 W(b) __armv4_mmu_cache_off
1047 W(b) __armv4_mmu_cache_flush
1049 .word 0x6901b110 @ sa1110
1051 W(b) __armv4_mmu_cache_on
1052 W(b) __armv4_mmu_cache_off
1053 W(b) __armv4_mmu_cache_flush
1056 .word 0xffffff00 @ PXA9xx
1057 W(b) __armv4_mmu_cache_on
1058 W(b) __armv4_mmu_cache_off
1059 W(b) __armv4_mmu_cache_flush
1061 .word 0x56158000 @ PXA168
1063 W(b) __armv4_mmu_cache_on
1064 W(b) __armv4_mmu_cache_off
1065 W(b) __armv5tej_mmu_cache_flush
1067 .word 0x56050000 @ Feroceon
1069 W(b) __armv4_mmu_cache_on
1070 W(b) __armv4_mmu_cache_off
1071 W(b) __armv5tej_mmu_cache_flush
1073 #ifdef CONFIG_CPU_FEROCEON_OLD_ID
1074 /* this conflicts with the standard ARMv5TE entry */
1075 .long 0x41009260 @ Old Feroceon
1077 b __armv4_mmu_cache_on
1078 b __armv4_mmu_cache_off
1079 b __armv5tej_mmu_cache_flush
1082 .word 0x66015261 @ FA526
1084 W(b) __fa526_cache_on
1085 W(b) __armv4_mmu_cache_off
1086 W(b) __fa526_cache_flush
1088 @ These match on the architecture ID
1090 .word 0x00020000 @ ARMv4T
1092 W(b) __armv4_mmu_cache_on
1093 W(b) __armv4_mmu_cache_off
1094 W(b) __armv4_mmu_cache_flush
1096 .word 0x00050000 @ ARMv5TE
1098 W(b) __armv4_mmu_cache_on
1099 W(b) __armv4_mmu_cache_off
1100 W(b) __armv4_mmu_cache_flush
1102 .word 0x00060000 @ ARMv5TEJ
1104 W(b) __armv4_mmu_cache_on
1105 W(b) __armv4_mmu_cache_off
1106 W(b) __armv5tej_mmu_cache_flush
1108 .word 0x0007b000 @ ARMv6
1110 W(b) __armv6_mmu_cache_on
1111 W(b) __armv4_mmu_cache_off
1112 W(b) __armv6_mmu_cache_flush
1114 .word 0x000f0000 @ new CPU Id
1116 W(b) __armv7_mmu_cache_on
1117 W(b) __armv7_mmu_cache_off
1118 W(b) __armv7_mmu_cache_flush
1120 .word 0 @ unrecognised type
1129 .size proc_types, . - proc_types
1132 * If you get a "non-constant expression in ".if" statement"
1133 * error from the assembler on this line, check that you have
1134 * not accidentally written a "b" instruction where you should
1135 * have written W(b).
1137 .if (. - proc_types) % PROC_ENTRY_SIZE != 0
1138 .error "The size of one or more proc_types entries is wrong."
1142 * Turn off the Cache and MMU. ARMv3 does not support
1143 * reading the control register, but ARMv4 does.
1146 * r0, r1, r2, r3, r9, r12 corrupted
1147 * This routine must preserve:
1151 cache_off: mov r3, #12 @ cache_off function
1154 __armv4_mpu_cache_off:
1155 mrc p15, 0, r0, c1, c0
1157 mcr p15, 0, r0, c1, c0 @ turn MPU and cache off
1159 mcr p15, 0, r0, c7, c10, 4 @ drain write buffer
1160 mcr p15, 0, r0, c7, c6, 0 @ flush D-Cache
1161 mcr p15, 0, r0, c7, c5, 0 @ flush I-Cache
1164 __armv3_mpu_cache_off:
1165 mrc p15, 0, r0, c1, c0
1167 mcr p15, 0, r0, c1, c0, 0 @ turn MPU and cache off
1169 mcr p15, 0, r0, c7, c0, 0 @ invalidate whole cache v3
1172 __armv4_mmu_cache_off:
1174 mrc p15, 0, r0, c1, c0
1176 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1178 mcr p15, 0, r0, c7, c7 @ invalidate whole cache v4
1179 mcr p15, 0, r0, c8, c7 @ invalidate whole TLB v4
1183 __armv7_mmu_cache_off:
1184 mrc p15, 0, r0, c1, c0
1190 mcr p15, 0, r0, c1, c0 @ turn MMU and cache off
1193 mcr p15, 0, r0, c8, c7, 0 @ invalidate whole TLB
1195 mcr p15, 0, r0, c7, c5, 6 @ invalidate BTC
1196 mcr p15, 0, r0, c7, c10, 4 @ DSB
1197 mcr p15, 0, r0, c7, c5, 4 @ ISB
1201 * Clean and flush the cache to maintain consistency.
1204 * r0 = start address
1205 * r1 = end address (exclusive)
1207 * r1, r2, r3, r9, r10, r11, r12 corrupted
1208 * This routine must preserve:
1217 __armv4_mpu_cache_flush:
1222 mcr p15, 0, ip, c7, c6, 0 @ invalidate D cache
1223 mov r1, #7 << 5 @ 8 segments
1224 1: orr r3, r1, #63 << 26 @ 64 entries
1225 2: mcr p15, 0, r3, c7, c14, 2 @ clean & invalidate D index
1226 subs r3, r3, #1 << 26
1227 bcs 2b @ entries 63 to 0
1228 subs r1, r1, #1 << 5
1229 bcs 1b @ segments 7 to 0
1232 mcrne p15, 0, ip, c7, c5, 0 @ invalidate I cache
1233 mcr p15, 0, ip, c7, c10, 4 @ drain WB
1236 __fa526_cache_flush:
1240 mcr p15, 0, r1, c7, c14, 0 @ clean and invalidate D cache
1241 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1242 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1245 __armv6_mmu_cache_flush:
1248 mcreq p15, 0, r1, c7, c14, 0 @ clean+invalidate D
1249 mcr p15, 0, r1, c7, c5, 0 @ invalidate I+BTB
1250 mcreq p15, 0, r1, c7, c15, 0 @ clean+invalidate unified
1251 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1254 __armv7_mmu_cache_flush:
1255 enable_cp15_barriers r10
1258 mrc p15, 0, r10, c0, c1, 5 @ read ID_MMFR1
1259 tst r10, #0xf << 16 @ hierarchical cache (ARMv7)
1262 mcr p15, 0, r10, c7, c14, 0 @ clean+invalidate D
1265 dcache_line_size r1, r2 @ r1 := dcache min line size
1266 sub r2, r1, #1 @ r2 := line size mask
1267 bic r0, r0, r2 @ round down start to line size
1268 sub r11, r11, #1 @ end address is exclusive
1269 bic r11, r11, r2 @ round down end to line size
1270 0: cmp r0, r11 @ finished?
1272 mcr p15, 0, r0, c7, c14, 1 @ Dcache clean/invalidate by VA
1276 mcr p15, 0, r10, c7, c10, 4 @ DSB
1277 mcr p15, 0, r10, c7, c5, 0 @ invalidate I+BTB
1278 mcr p15, 0, r10, c7, c10, 4 @ DSB
1279 mcr p15, 0, r10, c7, c5, 4 @ ISB
1282 __armv5tej_mmu_cache_flush:
1285 1: mrc p15, 0, APSR_nzcv, c7, c14, 3 @ test,clean,invalidate D cache
1287 mcr p15, 0, r0, c7, c5, 0 @ flush I cache
1288 mcr p15, 0, r0, c7, c10, 4 @ drain WB
1291 __armv4_mmu_cache_flush:
1294 mov r2, #64*1024 @ default: 32K dcache size (*2)
1295 mov r11, #32 @ default: 32 byte line size
1296 mrc p15, 0, r3, c0, c0, 1 @ read cache type
1297 teq r3, r9 @ cache ID register present?
1302 mov r2, r2, lsl r1 @ base dcache size *2
1303 tst r3, #1 << 14 @ test M bit
1304 addne r2, r2, r2, lsr #1 @ +1/2 size if M == 1
1308 mov r11, r11, lsl r3 @ cache line size in bytes
1311 bic r1, r1, #63 @ align to longest cache line
1314 ARM( ldr r3, [r1], r11 ) @ s/w flush D cache
1315 THUMB( ldr r3, [r1] ) @ s/w flush D cache
1316 THUMB( add r1, r1, r11 )
1320 mcr p15, 0, r1, c7, c5, 0 @ flush I cache
1321 mcr p15, 0, r1, c7, c6, 0 @ flush D cache
1322 mcr p15, 0, r1, c7, c10, 4 @ drain WB
1325 __armv3_mmu_cache_flush:
1326 __armv3_mpu_cache_flush:
1330 mcr p15, 0, r1, c7, c0, 0 @ invalidate whole cache v3
1334 * Various debugging routines for printing hex characters and
1335 * memory, which again must be relocatable.
1339 .type phexbuf,#object
1341 .size phexbuf, . - phexbuf
1343 @ phex corrupts {r0, r1, r2, r3}
1344 phex: adr r3, phexbuf
1358 @ puts corrupts {r0, r1, r2, r3}
1359 puts: loadsp r3, r2, r1
1360 1: ldrb r2, [r0], #1
1373 @ putc corrupts {r0, r1, r2, r3}
1380 @ memdump corrupts {r0, r1, r2, r3, r10, r11, r12, lr}
1381 memdump: mov r12, r0
1384 2: mov r0, r11, lsl #2
1392 ldr r0, [r12, r11, lsl #2]
1412 #ifdef CONFIG_ARM_VIRT_EXT
1414 __hyp_reentry_vectors:
1420 W(b) __enter_kernel @ hyp
1423 #endif /* CONFIG_ARM_VIRT_EXT */
1426 mov r0, #0 @ must be 0
1427 mov r1, r7 @ restore architecture number
1428 mov r2, r8 @ restore atags pointer
1429 ARM( mov pc, r4 ) @ call kernel
1430 M_CLASS( add r4, r4, #1 ) @ enter in Thumb mode for M class
1431 THUMB( bx r4 ) @ entry point is always ARM for A/R classes
1435 #ifdef CONFIG_EFI_STUB
1436 ENTRY(efi_enter_kernel)
1437 mov r7, r0 @ preserve image base
1438 mov r4, r1 @ preserve DT pointer
1440 mov r0, r4 @ DT start
1441 add r1, r4, r2 @ DT end
1442 bl cache_clean_flush
1444 mov r0, r7 @ relocated zImage
1445 ldr r1, =_edata @ size of zImage
1446 add r1, r1, r0 @ end of zImage
1447 bl cache_clean_flush
1449 @ The PE/COFF loader might not have cleaned the code we are
1450 @ running beyond the PoU, and so calling cache_off below from
1451 @ inside the PE/COFF loader allocated region is unsafe unless
1452 @ we explicitly clean it to the PoC.
1453 ARM( adrl r0, call_cache_fn )
1454 THUMB( adr r0, call_cache_fn ) @ region of code we will
1455 adr r1, 0f @ run with MMU off
1456 bl cache_clean_flush
1459 @ Set parameters for booting zImage according to boot protocol
1460 @ put FDT address in r2, it was returned by efi_entry()
1461 @ r1 is the machine type, and r0 needs to be 0
1465 add r7, r7, #(__efi_start - start)
1466 mov pc, r7 @ no mode switch
1467 ENDPROC(efi_enter_kernel)
1472 .section ".stack", "aw", %nobits
1473 .L_user_stack: .space 4096