2 * linux/arch/arm/kernel/head.S
4 * Copyright (C) 1994-2002 Russell King
5 * Copyright (c) 2003 ARM Limited
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 as
10 * published by the Free Software Foundation.
12 * Kernel startup code for all 32-bit CPUs
14 #include <linux/linkage.h>
15 #include <linux/init.h>
17 #include <asm/assembler.h>
19 #include <asm/domain.h>
20 #include <asm/ptrace.h>
21 #include <asm/asm-offsets.h>
22 #include <asm/memory.h>
23 #include <asm/thread_info.h>
24 #include <asm/pgtable.h>
26 #if defined(CONFIG_DEBUG_LL) && !defined(CONFIG_DEBUG_SEMIHOSTING)
27 #include CONFIG_DEBUG_LL_INCLUDE
31 * swapper_pg_dir is the virtual address of the initial page table.
32 * We place the page tables 16K below KERNEL_RAM_VADDR. Therefore, we must
33 * make sure that KERNEL_RAM_VADDR is correctly set. Currently, we expect
34 * the least significant 16 bits to be 0x8000, but we could probably
35 * relax this restriction to KERNEL_RAM_VADDR >= PAGE_OFFSET + 0x4000.
37 #define KERNEL_RAM_VADDR (PAGE_OFFSET + TEXT_OFFSET)
38 #if (KERNEL_RAM_VADDR & 0xffff) != 0x8000
39 #error KERNEL_RAM_VADDR must start at 0xXXXX8000
42 #ifdef CONFIG_ARM_LPAE
43 /* LPAE requires an additional page for the PGD */
44 #define PG_DIR_SIZE 0x5000
47 #define PG_DIR_SIZE 0x4000
52 .equ swapper_pg_dir, KERNEL_RAM_VADDR - PG_DIR_SIZE
54 .macro pgtbl, rd, phys
55 add \rd, \phys, #TEXT_OFFSET - PG_DIR_SIZE
59 * Kernel startup entry point.
60 * ---------------------------
62 * This is normally called from the decompressor code. The requirements
63 * are: MMU = off, D-cache = off, I-cache = dont care, r0 = 0,
64 * r1 = machine nr, r2 = atags or dtb pointer.
66 * This code is mostly position independent, so if you link the kernel at
67 * 0xc0008000, you call this at __pa(0xc0008000).
69 * See linux/arch/arm/tools/mach-types for the complete list of machine
72 * We're trying to keep crap to a minimum; DO NOT add any machine specific
73 * crap here - that's what the boot loader (or in extreme, well justified
74 * circumstances, zImage) is for.
80 ARM_BE8(setend be ) @ ensure we are in BE8 mode
82 THUMB( adr r9, BSYM(1f) ) @ Kernel is always entered in ARM.
83 THUMB( bx r9 ) @ If this is a Thumb-2 kernel,
84 THUMB( .thumb ) @ switch to Thumb now.
87 #ifdef CONFIG_ARM_VIRT_EXT
90 @ ensure svc mode and all interrupts masked
91 safe_svcmode_maskall r9
93 mrc p15, 0, r9, c0, c0 @ get processor id
94 bl __lookup_processor_type @ r5=procinfo r9=cpuid
95 movs r10, r5 @ invalid processor (r5=0)?
96 THUMB( it eq ) @ force fixup-able long branch encoding
97 beq __error_p @ yes, error 'p'
99 #ifdef CONFIG_ARM_LPAE
100 mrc p15, 0, r3, c0, c1, 4 @ read ID_MMFR0
101 and r3, r3, #0xf @ extract VMSA support
102 cmp r3, #5 @ long-descriptor translation table format?
103 THUMB( it lo ) @ force fixup-able long branch encoding
104 blo __error_p @ only classic page table format
107 #ifndef CONFIG_XIP_KERNEL
110 sub r4, r3, r4 @ (PHYS_OFFSET - PAGE_OFFSET)
111 add r8, r8, r4 @ PHYS_OFFSET
113 ldr r8, =PLAT_PHYS_OFFSET @ always constant in this case
117 * r1 = machine no, r2 = atags or dtb,
118 * r8 = phys_offset, r9 = cpuid, r10 = procinfo
121 #ifdef CONFIG_SMP_ON_UP
124 #ifdef CONFIG_ARM_PATCH_PHYS_VIRT
127 bl __create_page_tables
130 * The following calls CPU specific code in a position independent
131 * manner. See arch/arm/mm/proc-*.S for details. r10 = base of
132 * xxx_proc_info structure selected by __lookup_processor_type
133 * above. On return, the CPU will be ready for the MMU to be
134 * turned on, and r0 will hold the CPU control register value.
136 ldr r13, =__mmap_switched @ address to jump to after
137 @ mmu has been enabled
138 adr lr, BSYM(1f) @ return (PIC) address
139 mov r8, r4 @ set TTBR1 to swapper_pg_dir
140 ARM( add pc, r10, #PROCINFO_INITFUNC )
141 THUMB( add r12, r10, #PROCINFO_INITFUNC )
146 #ifndef CONFIG_XIP_KERNEL
152 * Setup the initial page tables. We only setup the barest
153 * amount which are required to get the kernel running, which
154 * generally means mapping in the kernel code.
156 * r8 = phys_offset, r9 = cpuid, r10 = procinfo
159 * r0, r3, r5-r7 corrupted
160 * r4 = page table (see ARCH_PGD_SHIFT in asm/memory.h)
162 __create_page_tables:
163 pgtbl r4, r8 @ page table address
166 * Clear the swapper page table
170 add r6, r0, #PG_DIR_SIZE
178 #ifdef CONFIG_ARM_LPAE
180 * Build the PGD table (first level) to point to the PMD table. A PGD
181 * entry is 64-bit wide.
184 add r3, r4, #0x1000 @ first PMD table address
185 orr r3, r3, #3 @ PGD block type
186 mov r6, #4 @ PTRS_PER_PGD
187 mov r7, #1 << (55 - 32) @ L_PGD_SWAPPER
189 #ifdef CONFIG_CPU_ENDIAN_BE8
190 str r7, [r0], #4 @ set top PGD entry bits
191 str r3, [r0], #4 @ set bottom PGD entry bits
193 str r3, [r0], #4 @ set bottom PGD entry bits
194 str r7, [r0], #4 @ set top PGD entry bits
196 add r3, r3, #0x1000 @ next PMD table
200 add r4, r4, #0x1000 @ point to the PMD tables
201 #ifdef CONFIG_CPU_ENDIAN_BE8
202 add r4, r4, #4 @ we only write the bottom word
206 ldr r7, [r10, #PROCINFO_MM_MMUFLAGS] @ mm_mmuflags
209 * Create identity mapping to cater for __enable_mmu.
210 * This identity mapping will be removed by paging_init().
212 adr r0, __turn_mmu_on_loc
213 ldmia r0, {r3, r5, r6}
214 sub r0, r0, r3 @ virt->phys offset
215 add r5, r5, r0 @ phys __turn_mmu_on
216 add r6, r6, r0 @ phys __turn_mmu_on_end
217 mov r5, r5, lsr #SECTION_SHIFT
218 mov r6, r6, lsr #SECTION_SHIFT
220 1: orr r3, r7, r5, lsl #SECTION_SHIFT @ flags + kernel base
221 str r3, [r4, r5, lsl #PMD_ORDER] @ identity mapping
223 addlo r5, r5, #1 @ next section
227 * Map our RAM from the start to the end of the kernel .bss section.
229 add r0, r4, #PAGE_OFFSET >> (SECTION_SHIFT - PMD_ORDER)
232 add r6, r4, r6, lsr #(SECTION_SHIFT - PMD_ORDER)
233 1: str r3, [r0], #1 << PMD_ORDER
234 add r3, r3, #1 << SECTION_SHIFT
238 #ifdef CONFIG_XIP_KERNEL
240 * Map the kernel image separately as it is not located in RAM.
242 #define XIP_START XIP_VIRT_ADDR(CONFIG_XIP_PHYS_ADDR)
244 mov r3, r3, lsr #SECTION_SHIFT
245 orr r3, r7, r3, lsl #SECTION_SHIFT
246 add r0, r4, #(XIP_START & 0xff000000) >> (SECTION_SHIFT - PMD_ORDER)
247 str r3, [r0, #((XIP_START & 0x00f00000) >> SECTION_SHIFT) << PMD_ORDER]!
248 ldr r6, =(_edata_loc - 1)
249 add r0, r0, #1 << PMD_ORDER
250 add r6, r4, r6, lsr #(SECTION_SHIFT - PMD_ORDER)
252 add r3, r3, #1 << SECTION_SHIFT
253 strls r3, [r0], #1 << PMD_ORDER
258 * Then map boot params address in r2 if specified.
259 * We map 2 sections in case the ATAGs/DTB crosses a section boundary.
261 mov r0, r2, lsr #SECTION_SHIFT
262 movs r0, r0, lsl #SECTION_SHIFT
264 addne r3, r3, #PAGE_OFFSET
265 addne r3, r4, r3, lsr #(SECTION_SHIFT - PMD_ORDER)
267 strne r6, [r3], #1 << PMD_ORDER
268 addne r6, r6, #1 << SECTION_SHIFT
271 #if defined(CONFIG_ARM_LPAE) && defined(CONFIG_CPU_ENDIAN_BE8)
272 sub r4, r4, #4 @ Fixup page table pointer
273 @ for 64-bit descriptors
276 #ifdef CONFIG_DEBUG_LL
277 #if !defined(CONFIG_DEBUG_ICEDCC) && !defined(CONFIG_DEBUG_SEMIHOSTING)
279 * Map in IO space for serial debugging.
280 * This allows debug messages to be output
281 * via a serial console before paging_init.
285 mov r3, r3, lsr #SECTION_SHIFT
286 mov r3, r3, lsl #PMD_ORDER
289 mov r3, r7, lsr #SECTION_SHIFT
290 ldr r7, [r10, #PROCINFO_IO_MMUFLAGS] @ io_mmuflags
291 orr r3, r7, r3, lsl #SECTION_SHIFT
292 #ifdef CONFIG_ARM_LPAE
293 mov r7, #1 << (54 - 32) @ XN
294 #ifdef CONFIG_CPU_ENDIAN_BE8
302 orr r3, r3, #PMD_SECT_XN
306 #else /* CONFIG_DEBUG_ICEDCC || CONFIG_DEBUG_SEMIHOSTING */
307 /* we don't need any serial debugging mappings */
308 ldr r7, [r10, #PROCINFO_IO_MMUFLAGS] @ io_mmuflags
311 #if defined(CONFIG_ARCH_NETWINDER) || defined(CONFIG_ARCH_CATS)
313 * If we're using the NetWinder or CATS, we also need to map
314 * in the 16550-type serial port for the debug messages
316 add r0, r4, #0xff000000 >> (SECTION_SHIFT - PMD_ORDER)
317 orr r3, r7, #0x7c000000
320 #ifdef CONFIG_ARCH_RPC
322 * Map in screen at 0x02000000 & SCREEN2_BASE
323 * Similar reasons here - for debug. This is
324 * only for Acorn RiscPC architectures.
326 add r0, r4, #0x02000000 >> (SECTION_SHIFT - PMD_ORDER)
327 orr r3, r7, #0x02000000
329 add r0, r4, #0xd8000000 >> (SECTION_SHIFT - PMD_ORDER)
333 #ifdef CONFIG_ARM_LPAE
334 sub r4, r4, #0x1000 @ point to the PGD table
335 mov r4, r4, lsr #ARCH_PGD_SHIFT
338 ENDPROC(__create_page_tables)
344 .long __turn_mmu_on_end
346 #if defined(CONFIG_SMP)
348 ENTRY(secondary_startup)
350 * Common entry point for secondary CPUs.
352 * Ensure that we're in SVC mode, and IRQs are disabled. Lookup
353 * the processor type - there is no need to check the machine type
354 * as it has already been validated by the primary processor.
357 ARM_BE8(setend be) @ ensure we are in BE8 mode
359 #ifdef CONFIG_ARM_VIRT_EXT
360 bl __hyp_stub_install_secondary
362 safe_svcmode_maskall r9
364 mrc p15, 0, r9, c0, c0 @ get processor id
365 bl __lookup_processor_type
366 movs r10, r5 @ invalid processor?
367 moveq r0, #'p' @ yes, error 'p'
368 THUMB( it eq ) @ force fixup-able long branch encoding
372 * Use the page tables supplied from __cpu_up.
374 adr r4, __secondary_data
375 ldmia r4, {r5, r7, r12} @ address to jump to after
376 sub lr, r4, r5 @ mmu has been enabled
377 ldr r4, [r7, lr] @ get secondary_data.pgdir
379 ldr r8, [r7, lr] @ get secondary_data.swapper_pg_dir
380 adr lr, BSYM(__enable_mmu) @ return address
381 mov r13, r12 @ __secondary_switched address
382 ARM( add pc, r10, #PROCINFO_INITFUNC ) @ initialise processor
383 @ (return control reg)
384 THUMB( add r12, r10, #PROCINFO_INITFUNC )
386 ENDPROC(secondary_startup)
389 * r6 = &secondary_data
391 ENTRY(__secondary_switched)
392 ldr sp, [r7, #4] @ get secondary_data.stack
394 b secondary_start_kernel
395 ENDPROC(__secondary_switched)
399 .type __secondary_data, %object
403 .long __secondary_switched
404 #endif /* defined(CONFIG_SMP) */
409 * Setup common bits before finally enabling the MMU. Essentially
410 * this is just loading the page table pointer and domain access
413 * r0 = cp#15 control register
415 * r2 = atags or dtb pointer
416 * r4 = page table (see ARCH_PGD_SHIFT in asm/memory.h)
418 * r13 = *virtual* address to jump to upon completion
421 #if defined(CONFIG_ALIGNMENT_TRAP) && __LINUX_ARM_ARCH__ < 6
426 #ifdef CONFIG_CPU_DCACHE_DISABLE
429 #ifdef CONFIG_CPU_BPREDICT_DISABLE
432 #ifdef CONFIG_CPU_ICACHE_DISABLE
435 #ifndef CONFIG_ARM_LPAE
436 mov r5, #(domain_val(DOMAIN_USER, DOMAIN_MANAGER) | \
437 domain_val(DOMAIN_KERNEL, DOMAIN_MANAGER) | \
438 domain_val(DOMAIN_TABLE, DOMAIN_MANAGER) | \
439 domain_val(DOMAIN_IO, DOMAIN_CLIENT))
440 mcr p15, 0, r5, c3, c0, 0 @ load domain access register
441 mcr p15, 0, r4, c2, c0, 0 @ load page table pointer
444 ENDPROC(__enable_mmu)
447 * Enable the MMU. This completely changes the structure of the visible
448 * memory space. You will not be able to trace execution through this.
449 * If you have an enquiry about this, *please* check the linux-arm-kernel
450 * mailing list archives BEFORE sending another post to the list.
452 * r0 = cp#15 control register
454 * r2 = atags or dtb pointer
456 * r13 = *virtual* address to jump to upon completion
458 * other registers depend on the function called upon completion
461 .pushsection .idmap.text, "ax"
465 mcr p15, 0, r0, c1, c0, 0 @ write control reg
466 mrc p15, 0, r3, c0, c0, 0 @ read id reg
472 ENDPROC(__turn_mmu_on)
476 #ifdef CONFIG_SMP_ON_UP
479 and r3, r9, #0x000f0000 @ architecture version
480 teq r3, #0x000f0000 @ CPU ID supported?
481 bne __fixup_smp_on_up @ no, assume UP
483 bic r3, r9, #0x00ff0000
484 bic r3, r3, #0x0000000f @ mask 0xff00fff0
486 orr r4, r4, #0x0000b000
487 orr r4, r4, #0x00000020 @ val 0x4100b020
488 teq r3, r4 @ ARM 11MPCore?
489 moveq pc, lr @ yes, assume SMP
491 mrc p15, 0, r0, c0, c0, 5 @ read MPIDR
492 and r0, r0, #0xc0000000 @ multiprocessing extensions and
493 teq r0, #0x80000000 @ not part of a uniprocessor system?
494 bne __fixup_smp_on_up @ no, assume UP
496 @ Core indicates it is SMP. Check for Aegis SOC where a single
497 @ Cortex-A9 CPU is present but SMP operations fault.
499 orr r4, r4, #0x0000c000
500 orr r4, r4, #0x00000090
501 teq r3, r4 @ Check for ARM Cortex-A9
502 movne pc, lr @ Not ARM Cortex-A9,
504 @ If a future SoC *does* use 0x0 as the PERIPH_BASE, then the
505 @ below address check will need to be #ifdef'd or equivalent
506 @ for the Aegis platform.
507 mrc p15, 4, r0, c15, c0 @ get SCU base address
508 teq r0, #0x0 @ '0' on actual UP A9 hardware
509 beq __fixup_smp_on_up @ So its an A9 UP
510 ldr r0, [r0, #4] @ read SCU Config
511 ARM_BE8(rev r0, r0) @ byteswap if big endian
512 and r0, r0, #0x3 @ number of CPUs
522 b __do_fixup_smp_on_up
539 __do_fixup_smp_on_up:
543 ARM( str r6, [r0, r3] )
544 THUMB( add r0, r0, r3 )
546 THUMB( mov r6, r6, ror #16 ) @ Convert word order for big-endian.
548 THUMB( strh r6, [r0], #2 ) @ For Thumb-2, store as two halfwords
549 THUMB( mov r6, r6, lsr #16 ) @ to be robust against misaligned r3.
550 THUMB( strh r6, [r0] )
551 b __do_fixup_smp_on_up
552 ENDPROC(__do_fixup_smp_on_up)
555 stmfd sp!, {r4 - r6, lr}
559 bl __do_fixup_smp_on_up
560 ldmfd sp!, {r4 - r6, pc}
564 #define LOW_OFFSET 0x4
565 #define HIGH_OFFSET 0x0
567 #define LOW_OFFSET 0x0
568 #define HIGH_OFFSET 0x4
571 #ifdef CONFIG_ARM_PATCH_PHYS_VIRT
573 /* __fixup_pv_table - patch the stub instructions with the delta between
574 * PHYS_OFFSET and PAGE_OFFSET, which is assumed to be 16MiB aligned and
575 * can be expressed by an immediate shifter operand. The stub instruction
576 * has a form of '(add|sub) rd, rn, #imm'.
583 subs r3, r0, r3 @ PHYS_OFFSET - PAGE_OFFSET
584 add r4, r4, r3 @ adjust table start address
585 add r5, r5, r3 @ adjust table end address
586 add r6, r6, r3 @ adjust __pv_phys_offset address
587 add r7, r7, r3 @ adjust __pv_offset address
588 str r8, [r6, #LOW_OFFSET] @ save computed PHYS_OFFSET to __pv_phys_offset
589 strcc ip, [r7, #HIGH_OFFSET] @ save to __pv_offset high bits
590 mov r6, r3, lsr #24 @ constant for add/sub instructions
591 teq r3, r6, lsl #24 @ must be 16MiB aligned
592 THUMB( it ne @ cross section branch )
594 str r3, [r7, #LOW_OFFSET] @ save to __pv_offset low bits
596 ENDPROC(__fixup_pv_table)
600 .long __pv_table_begin
602 2: .long __pv_phys_offset
610 ldr r0, [r6, #HIGH_OFFSET] @ pv_offset high word
611 ldr r6, [r6, #LOW_OFFSET] @ pv_offset low word
614 #ifdef CONFIG_THUMB2_KERNEL
615 moveq r0, #0x200000 @ set bit 21, mov to mvn instruction
624 orr r6, r6, r7, lsl #12
629 ARM_BE8(rev16 ip, ip)
632 orrne ip, r6 @ mask in offset bits 31-24
633 orreq ip, r0 @ mask in offset bits 7-0
634 ARM_BE8(rev16 ip, ip)
638 ARM_BE8(rev16 ip, ip)
640 orr ip, ip, r0, lsr #16
641 ARM_BE8(rev16 ip, ip)
644 ldrcc r7, [r4], #4 @ use branch for delay slot
648 #ifdef CONFIG_CPU_ENDIAN_BE8
649 moveq r0, #0x00004000 @ set bit 22, mov to mvn instruction
651 moveq r0, #0x400000 @ set bit 22, mov to mvn instruction
655 #ifdef CONFIG_CPU_ENDIAN_BE8
656 @ in BE8, we load data in BE, but instructions still in LE
657 bic ip, ip, #0xff000000
658 tst ip, #0x000f0000 @ check the rotation field
659 orrne ip, ip, r6, lsl #24 @ mask in offset bits 31-24
660 biceq ip, ip, #0x00004000 @ clear bit 22
661 orreq ip, ip, r0 @ mask in offset bits 7-0
663 bic ip, ip, #0x000000ff
664 tst ip, #0xf00 @ check the rotation field
665 orrne ip, ip, r6 @ mask in offset bits 31-24
666 biceq ip, ip, #0x400000 @ clear bit 22
667 orreq ip, ip, r0 @ mask in offset bits 7-0
671 ldrcc r7, [r4], #4 @ use branch for delay slot
675 ENDPROC(__fixup_a_pv_table)
680 ENTRY(fixup_pv_table)
681 stmfd sp!, {r4 - r7, lr}
682 mov r3, #0 @ no offset
683 mov r4, r0 @ r0 = table start
684 add r5, r0, r1 @ r1 = table size
685 bl __fixup_a_pv_table
686 ldmfd sp!, {r4 - r7, pc}
687 ENDPROC(fixup_pv_table)
690 .globl __pv_phys_offset
691 .type __pv_phys_offset, %object
694 .size __pv_phys_offset, . -__pv_phys_offset
697 .type __pv_offset, %object
700 .size __pv_offset, . -__pv_offset
703 #include "head-common.S"