2 * Low-level CPU initialisation
3 * Based on arch/arm/kernel/head.S
5 * Copyright (C) 1994-2002 Russell King
6 * Copyright (C) 2003-2012 ARM Ltd.
7 * Authors: Catalin Marinas <catalin.marinas@arm.com>
8 * Will Deacon <will.deacon@arm.com>
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License version 2 as
12 * published by the Free Software Foundation.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program. If not, see <http://www.gnu.org/licenses/>.
23 #include <linux/linkage.h>
24 #include <linux/init.h>
26 #include <asm/assembler.h>
27 #include <asm/ptrace.h>
28 #include <asm/asm-offsets.h>
29 #include <asm/cputype.h>
30 #include <asm/memory.h>
31 #include <asm/thread_info.h>
32 #include <asm/pgtable-hwdef.h>
33 #include <asm/pgtable.h>
38 * swapper_pg_dir is the virtual address of the initial page table. We place
39 * the page tables 3 * PAGE_SIZE below KERNEL_RAM_VADDR. The idmap_pg_dir has
40 * 2 pages and is placed below swapper_pg_dir.
42 #define KERNEL_RAM_VADDR (PAGE_OFFSET + TEXT_OFFSET)
44 #if (KERNEL_RAM_VADDR & 0xfffff) != 0x80000
45 #error KERNEL_RAM_VADDR must start at 0xXXX80000
48 #define SWAPPER_DIR_SIZE (3 * PAGE_SIZE)
49 #define IDMAP_DIR_SIZE (2 * PAGE_SIZE)
52 .equ swapper_pg_dir, KERNEL_RAM_VADDR - SWAPPER_DIR_SIZE
55 .equ idmap_pg_dir, swapper_pg_dir - IDMAP_DIR_SIZE
57 .macro pgtbl, ttb0, ttb1, phys
58 add \ttb1, \phys, #TEXT_OFFSET - SWAPPER_DIR_SIZE
59 sub \ttb0, \ttb1, #IDMAP_DIR_SIZE
62 #ifdef CONFIG_ARM64_64K_PAGES
63 #define BLOCK_SHIFT PAGE_SHIFT
64 #define BLOCK_SIZE PAGE_SIZE
66 #define BLOCK_SHIFT SECTION_SHIFT
67 #define BLOCK_SIZE SECTION_SIZE
70 #define KERNEL_START KERNEL_RAM_VADDR
71 #define KERNEL_END _end
74 * Initial memory map attributes.
77 #define PTE_FLAGS PTE_TYPE_PAGE | PTE_AF
78 #define PMD_FLAGS PMD_TYPE_SECT | PMD_SECT_AF
80 #define PTE_FLAGS PTE_TYPE_PAGE | PTE_AF | PTE_SHARED
81 #define PMD_FLAGS PMD_TYPE_SECT | PMD_SECT_AF | PMD_SECT_S
84 #ifdef CONFIG_ARM64_64K_PAGES
85 #define MM_MMUFLAGS PTE_ATTRINDX(MT_NORMAL) | PTE_FLAGS
87 #define MM_MMUFLAGS PMD_ATTRINDX(MT_NORMAL) | PMD_FLAGS
91 * Kernel startup entry point.
92 * ---------------------------
94 * The requirements are:
95 * MMU = off, D-cache = off, I-cache = on or off,
96 * x0 = physical address to the FDT blob.
98 * This code is mostly position independent so you call this at
99 * __pa(PAGE_OFFSET + TEXT_OFFSET).
101 * Note that the callee-saved registers are used for storing variables
102 * that are useful before the MMU is enabled. The allocations are described
103 * in the entry routines.
108 * DO NOT MODIFY. Image header expected by Linux boot-loaders.
110 b stext // branch to kernel start, magic
112 .quad TEXT_OFFSET // Image load offset from start of RAM
118 .byte 0x41 // Magic number, "ARM\x64"
125 mov x21, x0 // x21=FDT
126 bl __calc_phys_offset // x24=PHYS_OFFSET, x28=PHYS_OFFSET-PAGE_OFFSET
127 bl el2_setup // Drop to EL1
128 mrs x22, midr_el1 // x22=cpuid
130 bl lookup_processor_type
131 mov x23, x0 // x23=current cpu_table
132 cbz x23, __error_p // invalid processor (x23=0)?
134 bl __create_page_tables // x25=TTBR0, x26=TTBR1
136 * The following calls CPU specific code in a position independent
137 * manner. See arch/arm64/mm/proc.S for details. x23 = base of
138 * cpu_info structure selected by lookup_processor_type above.
139 * On return, the CPU will be ready for the MMU to be turned on and
140 * the TCR will have been set.
142 ldr x27, __switch_data // address to jump to after
143 // MMU has been enabled
144 adr lr, __enable_mmu // return (PIC) address
145 ldr x12, [x23, #CPU_INFO_SETUP]
146 add x12, x12, x28 // __virt_to_phys
147 br x12 // initialise processor
151 * If we're fortunate enough to boot at EL2, ensure that the world is
152 * sane before dropping to EL1.
156 cmp x0, #PSR_MODE_EL2t
157 ccmp x0, #PSR_MODE_EL2h, #0x4, ne
158 ldr x0, =__boot_cpu_mode // Compute __boot_cpu_mode
161 str wzr, [x0] // Remember we don't have EL2...
164 /* Hyp configuration. */
165 1: ldr w1, =BOOT_CPU_MODE_EL2
166 str w1, [x0, #4] // This CPU has EL2
167 mov x0, #(1 << 31) // 64-bit EL1
170 /* Generic timers. */
172 orr x0, x0, #3 // Enable EL1 physical timers
174 msr cntvoff_el2, xzr // Clear virtual offset
176 /* Populate ID registers. */
183 mov x0, #0x0800 // Set/clear RES{1,0} bits
184 movk x0, #0x30d0, lsl #16
187 /* Coprocessor traps. */
189 msr cptr_el2, x0 // Disable copro. traps to EL2
192 msr hstr_el2, xzr // Disable CP15 traps to EL2
195 /* Stage-2 translation */
198 /* Hypervisor stub */
199 adr x0, __hyp_stub_vectors
203 mov x0, #(PSR_F_BIT | PSR_I_BIT | PSR_A_BIT | PSR_D_BIT |\
211 * We need to find out the CPU boot mode long after boot, so we need to
212 * store it in a writable variable.
214 * This is not in .bss, because we set it sufficiently early that the boot-time
215 * zeroing of .bss would clobber it.
218 ENTRY(__boot_cpu_mode)
219 .long BOOT_CPU_MODE_EL2
230 .quad secondary_holding_pen_release
233 * This provides a "holding pen" for platforms to hold all secondary
234 * cores are held until we're ready for them to initialise.
236 ENTRY(secondary_holding_pen)
237 bl __calc_phys_offset // x24=phys offset
238 bl el2_setup // Drop to EL1
240 ldr x1, =MPIDR_HWID_BITMASK
248 b.eq secondary_startup
251 ENDPROC(secondary_holding_pen)
254 * Secondary entry point that jumps straight into the kernel. Only to
255 * be used where CPUs are brought online dynamically by the kernel.
257 ENTRY(secondary_entry)
258 bl __calc_phys_offset // x2=phys offset
259 bl el2_setup // Drop to EL1
261 ENDPROC(secondary_entry)
263 ENTRY(secondary_startup)
265 * Common entry point for secondary CPUs.
267 mrs x22, midr_el1 // x22=cpuid
269 bl lookup_processor_type
270 mov x23, x0 // x23=current cpu_table
271 cbz x23, __error_p // invalid processor (x23=0)?
273 pgtbl x25, x26, x24 // x25=TTBR0, x26=TTBR1
274 ldr x12, [x23, #CPU_INFO_SETUP]
275 add x12, x12, x28 // __virt_to_phys
276 blr x12 // initialise processor
278 ldr x21, =secondary_data
279 ldr x27, =__secondary_switched // address to jump to after enabling the MMU
281 ENDPROC(secondary_startup)
283 ENTRY(__secondary_switched)
284 ldr x0, [x21] // get secondary_data.stack
287 b secondary_start_kernel
288 ENDPROC(__secondary_switched)
289 #endif /* CONFIG_SMP */
292 * Setup common bits before finally enabling the MMU. Essentially this is just
293 * loading the page table pointer and vector base registers.
295 * On entry to this code, x0 must contain the SCTLR_EL1 value for turning on
301 msr ttbr0_el1, x25 // load TTBR0
302 msr ttbr1_el1, x26 // load TTBR1
305 ENDPROC(__enable_mmu)
308 * Enable the MMU. This completely changes the structure of the visible memory
309 * space. You will not be able to trace execution through this.
311 * x0 = system control register
312 * x27 = *virtual* address to jump to upon completion
314 * other registers depend on the function called upon completion
316 * We align the entire function to the smallest power of two larger than it to
317 * ensure it fits within a single block map entry. Otherwise were PHYS_OFFSET
318 * close to the end of a 512MB or 1GB block we might require an additional
319 * table to map the entire function.
326 ENDPROC(__turn_mmu_on)
329 * Calculate the start of physical memory.
334 sub x28, x0, x1 // x28 = PHYS_OFFSET - PAGE_OFFSET
335 add x24, x2, x28 // x24 = PHYS_OFFSET
337 ENDPROC(__calc_phys_offset)
344 * Macro to populate the PGD for the corresponding block entry in the next
345 * level (tbl) for the given virtual address.
347 * Preserves: pgd, tbl, virt
348 * Corrupts: tmp1, tmp2
350 .macro create_pgd_entry, pgd, tbl, virt, tmp1, tmp2
351 lsr \tmp1, \virt, #PGDIR_SHIFT
352 and \tmp1, \tmp1, #PTRS_PER_PGD - 1 // PGD index
353 orr \tmp2, \tbl, #3 // PGD entry table type
354 str \tmp2, [\pgd, \tmp1, lsl #3]
358 * Macro to populate block entries in the page table for the start..end
359 * virtual range (inclusive).
361 * Preserves: tbl, flags
362 * Corrupts: phys, start, end, pstate
364 .macro create_block_map, tbl, flags, phys, start, end
365 lsr \phys, \phys, #BLOCK_SHIFT
366 lsr \start, \start, #BLOCK_SHIFT
367 and \start, \start, #PTRS_PER_PTE - 1 // table index
368 orr \phys, \flags, \phys, lsl #BLOCK_SHIFT // table entry
369 lsr \end, \end, #BLOCK_SHIFT
370 and \end, \end, #PTRS_PER_PTE - 1 // table end index
371 9999: str \phys, [\tbl, \start, lsl #3] // store the entry
372 add \start, \start, #1 // next entry
373 add \phys, \phys, #BLOCK_SIZE // next block
379 * Setup the initial page tables. We only setup the barest amount which is
380 * required to get the kernel running. The following sections are required:
381 * - identity mapping to enable the MMU (low address, TTBR0)
382 * - first few MB of the kernel linear mapping to jump to once the MMU has
383 * been enabled, including the FDT blob (TTBR1)
384 * - pgd entry for fixed mappings (TTBR1)
386 __create_page_tables:
387 pgtbl x25, x26, x24 // idmap_pg_dir and swapper_pg_dir addresses
390 * Clear the idmap and swapper page tables.
393 add x6, x26, #SWAPPER_DIR_SIZE
394 1: stp xzr, xzr, [x0], #16
395 stp xzr, xzr, [x0], #16
396 stp xzr, xzr, [x0], #16
397 stp xzr, xzr, [x0], #16
404 * Create the identity mapping.
406 add x0, x25, #PAGE_SIZE // section table address
407 ldr x3, =KERNEL_START
408 add x3, x3, x28 // __pa(KERNEL_START)
409 create_pgd_entry x25, x0, x3, x5, x6
411 mov x5, x3 // __pa(KERNEL_START)
412 add x6, x6, x28 // __pa(KERNEL_END)
413 create_block_map x0, x7, x3, x5, x6
416 * Map the kernel image (starting with PHYS_OFFSET).
418 add x0, x26, #PAGE_SIZE // section table address
420 create_pgd_entry x26, x0, x5, x3, x6
422 mov x3, x24 // phys offset
423 create_block_map x0, x7, x3, x5, x6
426 * Map the FDT blob (maximum 2MB; must be within 512MB of
429 mov x3, x21 // FDT phys address
430 and x3, x3, #~((1 << 21) - 1) // 2MB aligned
432 sub x5, x3, x24 // subtract PHYS_OFFSET
433 tst x5, #~((1 << 29) - 1) // within 512MB?
434 csel x21, xzr, x21, ne // zero the FDT pointer
436 add x5, x5, x6 // __va(FDT blob)
437 add x6, x5, #1 << 21 // 2MB for the FDT blob
438 sub x6, x6, #1 // inclusive range
439 create_block_map x0, x7, x3, x5, x6
442 * Create the pgd entry for the fixed mappings.
444 ldr x5, =FIXADDR_TOP // Fixed mapping virtual address
445 add x0, x26, #2 * PAGE_SIZE // section table address
446 create_pgd_entry x26, x0, x5, x6, x7
448 ENDPROC(__create_page_tables)
452 .type __switch_data, %object
454 .quad __mmap_switched
455 .quad __bss_start // x6
457 .quad processor_id // x4
458 .quad __fdt_pointer // x5
459 .quad memstart_addr // x6
460 .quad init_thread_union + THREAD_START_SP // sp
463 * The following fragment of code is executed with the MMU on in MMU mode, and
464 * uses absolute addresses; this is not position independent.
467 adr x3, __switch_data + 8
469 ldp x6, x7, [x3], #16
472 str xzr, [x6], #8 // Clear BSS
475 ldp x4, x5, [x3], #16
479 str x22, [x4] // Save processor ID
480 str x21, [x5] // Save FDT pointer
481 str x24, [x6] // Save PHYS_OFFSET
484 ENDPROC(__mmap_switched)
487 * Exception handling. Something went wrong and we can't proceed. We ought to
488 * tell the user, but since we don't have any guarantee that we're even
489 * running on the right architecture, we do virtually nothing.
500 * This function gets the processor ID in w0 and searches the cpu_table[] for
501 * a match. It returns a pointer to the struct cpu_info it found. The
502 * cpu_table[] must end with an empty (all zeros) structure.
504 * This routine can be called via C code and it needs to work with the MMU
505 * both disabled and enabled (the offset is calculated automatically).
507 ENTRY(lookup_processor_type)
508 adr x1, __lookup_processor_type_data
510 sub x1, x1, x2 // get offset between VA and PA
511 add x3, x3, x1 // convert VA to PA
513 ldp w5, w6, [x3] // load cpu_id_val and cpu_id_mask
514 cbz w5, 2f // end of list?
518 add x3, x3, #CPU_INFO_SZ
521 mov x3, #0 // unknown processor
525 ENDPROC(lookup_processor_type)
528 .type __lookup_processor_type_data, %object
529 __lookup_processor_type_data:
532 .size __lookup_processor_type_data, . - __lookup_processor_type_data
535 * Determine validity of the x21 FDT pointer.
536 * The dtb must be 8-byte aligned and live in the first 512M of memory.