2 * Based on arch/arm/mm/mmu.c
4 * Copyright (C) 1995-2005 Russell King
5 * Copyright (C) 2012 ARM Ltd.
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 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
20 #include <linux/export.h>
21 #include <linux/kernel.h>
22 #include <linux/errno.h>
23 #include <linux/init.h>
24 #include <linux/libfdt.h>
25 #include <linux/mman.h>
26 #include <linux/nodemask.h>
27 #include <linux/memblock.h>
30 #include <linux/slab.h>
31 #include <linux/stop_machine.h>
33 #include <asm/cputype.h>
34 #include <asm/fixmap.h>
35 #include <asm/kernel-pgtable.h>
36 #include <asm/sections.h>
37 #include <asm/setup.h>
38 #include <asm/sizes.h>
40 #include <asm/memblock.h>
41 #include <asm/mmu_context.h>
45 u64 idmap_t0sz
= TCR_T0SZ(VA_BITS
);
48 * Empty_zero_page is a special page that is used for zero-initialized data
51 struct page
*empty_zero_page
;
52 EXPORT_SYMBOL(empty_zero_page
);
54 pgprot_t
phys_mem_access_prot(struct file
*file
, unsigned long pfn
,
55 unsigned long size
, pgprot_t vma_prot
)
58 return pgprot_noncached(vma_prot
);
59 else if (file
->f_flags
& O_SYNC
)
60 return pgprot_writecombine(vma_prot
);
63 EXPORT_SYMBOL(phys_mem_access_prot
);
65 static void __init
*early_alloc(unsigned long sz
)
70 phys
= memblock_alloc(sz
, sz
);
78 * remap a PMD into pages
80 static void split_pmd(pmd_t
*pmd
, pte_t
*pte
)
82 unsigned long pfn
= pmd_pfn(*pmd
);
87 * Need to have the least restrictive permissions available
88 * permissions will be fixed up later
90 set_pte(pte
, pfn_pte(pfn
, PAGE_KERNEL_EXEC
));
92 } while (pte
++, i
++, i
< PTRS_PER_PTE
);
95 static void alloc_init_pte(pmd_t
*pmd
, unsigned long addr
,
96 unsigned long end
, unsigned long pfn
,
98 void *(*alloc
)(unsigned long size
))
102 if (pmd_none(*pmd
) || pmd_sect(*pmd
)) {
103 pte
= alloc(PTRS_PER_PTE
* sizeof(pte_t
));
106 __pmd_populate(pmd
, __pa(pte
), PMD_TYPE_TABLE
);
109 BUG_ON(pmd_bad(*pmd
));
111 pte
= pte_offset_kernel(pmd
, addr
);
113 set_pte(pte
, pfn_pte(pfn
, prot
));
115 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
118 static void split_pud(pud_t
*old_pud
, pmd_t
*pmd
)
120 unsigned long addr
= pud_pfn(*old_pud
) << PAGE_SHIFT
;
121 pgprot_t prot
= __pgprot(pud_val(*old_pud
) ^ addr
);
125 set_pmd(pmd
, __pmd(addr
| pgprot_val(prot
)));
127 } while (pmd
++, i
++, i
< PTRS_PER_PMD
);
130 static void alloc_init_pmd(struct mm_struct
*mm
, pud_t
*pud
,
131 unsigned long addr
, unsigned long end
,
132 phys_addr_t phys
, pgprot_t prot
,
133 void *(*alloc
)(unsigned long size
))
139 * Check for initial section mappings in the pgd/pud and remove them.
141 if (pud_none(*pud
) || pud_sect(*pud
)) {
142 pmd
= alloc(PTRS_PER_PMD
* sizeof(pmd_t
));
143 if (pud_sect(*pud
)) {
145 * need to have the 1G of mappings continue to be
150 pud_populate(mm
, pud
, pmd
);
153 BUG_ON(pud_bad(*pud
));
155 pmd
= pmd_offset(pud
, addr
);
157 next
= pmd_addr_end(addr
, end
);
158 /* try section mapping first */
159 if (((addr
| next
| phys
) & ~SECTION_MASK
) == 0) {
161 set_pmd(pmd
, __pmd(phys
|
162 pgprot_val(mk_sect_prot(prot
))));
164 * Check for previous table entries created during
165 * boot (__create_page_tables) and flush them.
167 if (!pmd_none(old_pmd
)) {
169 if (pmd_table(old_pmd
)) {
170 phys_addr_t table
= __pa(pte_offset_map(&old_pmd
, 0));
171 if (!WARN_ON_ONCE(slab_is_available()))
172 memblock_free(table
, PAGE_SIZE
);
176 alloc_init_pte(pmd
, addr
, next
, __phys_to_pfn(phys
),
180 } while (pmd
++, addr
= next
, addr
!= end
);
183 static inline bool use_1G_block(unsigned long addr
, unsigned long next
,
186 if (PAGE_SHIFT
!= 12)
189 if (((addr
| next
| phys
) & ~PUD_MASK
) != 0)
195 static void alloc_init_pud(struct mm_struct
*mm
, pgd_t
*pgd
,
196 unsigned long addr
, unsigned long end
,
197 phys_addr_t phys
, pgprot_t prot
,
198 void *(*alloc
)(unsigned long size
))
203 if (pgd_none(*pgd
)) {
204 pud
= alloc(PTRS_PER_PUD
* sizeof(pud_t
));
205 pgd_populate(mm
, pgd
, pud
);
207 BUG_ON(pgd_bad(*pgd
));
209 pud
= pud_offset(pgd
, addr
);
211 next
= pud_addr_end(addr
, end
);
214 * For 4K granule only, attempt to put down a 1GB block
216 if (use_1G_block(addr
, next
, phys
)) {
217 pud_t old_pud
= *pud
;
218 set_pud(pud
, __pud(phys
|
219 pgprot_val(mk_sect_prot(prot
))));
222 * If we have an old value for a pud, it will
223 * be pointing to a pmd table that we no longer
224 * need (from swapper_pg_dir).
226 * Look up the old pmd table and free it.
228 if (!pud_none(old_pud
)) {
230 if (pud_table(old_pud
)) {
231 phys_addr_t table
= __pa(pmd_offset(&old_pud
, 0));
232 if (!WARN_ON_ONCE(slab_is_available()))
233 memblock_free(table
, PAGE_SIZE
);
237 alloc_init_pmd(mm
, pud
, addr
, next
, phys
, prot
, alloc
);
240 } while (pud
++, addr
= next
, addr
!= end
);
244 * Create the page directory entries and any necessary page tables for the
245 * mapping specified by 'md'.
247 static void __create_mapping(struct mm_struct
*mm
, pgd_t
*pgd
,
248 phys_addr_t phys
, unsigned long virt
,
249 phys_addr_t size
, pgprot_t prot
,
250 void *(*alloc
)(unsigned long size
))
252 unsigned long addr
, length
, end
, next
;
254 addr
= virt
& PAGE_MASK
;
255 length
= PAGE_ALIGN(size
+ (virt
& ~PAGE_MASK
));
259 next
= pgd_addr_end(addr
, end
);
260 alloc_init_pud(mm
, pgd
, addr
, next
, phys
, prot
, alloc
);
262 } while (pgd
++, addr
= next
, addr
!= end
);
265 static void *late_alloc(unsigned long size
)
269 BUG_ON(size
> PAGE_SIZE
);
270 ptr
= (void *)__get_free_page(PGALLOC_GFP
);
275 static void __init
create_mapping(phys_addr_t phys
, unsigned long virt
,
276 phys_addr_t size
, pgprot_t prot
)
278 if (virt
< VMALLOC_START
) {
279 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
283 __create_mapping(&init_mm
, pgd_offset_k(virt
& PAGE_MASK
), phys
, virt
,
284 size
, prot
, early_alloc
);
287 void __init
create_pgd_mapping(struct mm_struct
*mm
, phys_addr_t phys
,
288 unsigned long virt
, phys_addr_t size
,
291 __create_mapping(mm
, pgd_offset(mm
, virt
), phys
, virt
, size
, prot
,
295 static void create_mapping_late(phys_addr_t phys
, unsigned long virt
,
296 phys_addr_t size
, pgprot_t prot
)
298 if (virt
< VMALLOC_START
) {
299 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
304 return __create_mapping(&init_mm
, pgd_offset_k(virt
& PAGE_MASK
),
305 phys
, virt
, size
, prot
, late_alloc
);
308 #ifdef CONFIG_DEBUG_RODATA
309 static void __init
__map_memblock(phys_addr_t start
, phys_addr_t end
)
312 * Set up the executable regions using the existing section mappings
313 * for now. This will get more fine grained later once all memory
316 unsigned long kernel_x_start
= round_down(__pa(_stext
), SWAPPER_BLOCK_SIZE
);
317 unsigned long kernel_x_end
= round_up(__pa(__init_end
), SWAPPER_BLOCK_SIZE
);
319 if (end
< kernel_x_start
) {
320 create_mapping(start
, __phys_to_virt(start
),
321 end
- start
, PAGE_KERNEL
);
322 } else if (start
>= kernel_x_end
) {
323 create_mapping(start
, __phys_to_virt(start
),
324 end
- start
, PAGE_KERNEL
);
326 if (start
< kernel_x_start
)
327 create_mapping(start
, __phys_to_virt(start
),
328 kernel_x_start
- start
,
330 create_mapping(kernel_x_start
,
331 __phys_to_virt(kernel_x_start
),
332 kernel_x_end
- kernel_x_start
,
334 if (kernel_x_end
< end
)
335 create_mapping(kernel_x_end
,
336 __phys_to_virt(kernel_x_end
),
343 static void __init
__map_memblock(phys_addr_t start
, phys_addr_t end
)
345 create_mapping(start
, __phys_to_virt(start
), end
- start
,
350 static void __init
map_mem(void)
352 struct memblock_region
*reg
;
356 * Temporarily limit the memblock range. We need to do this as
357 * create_mapping requires puds, pmds and ptes to be allocated from
358 * memory addressable from the initial direct kernel mapping.
360 * The initial direct kernel mapping, located at swapper_pg_dir, gives
361 * us PUD_SIZE (with SECTION maps) or PMD_SIZE (without SECTION maps,
362 * memory starting from PHYS_OFFSET (which must be aligned to 2MB as
363 * per Documentation/arm64/booting.txt).
365 limit
= PHYS_OFFSET
+ SWAPPER_INIT_MAP_SIZE
;
366 memblock_set_current_limit(limit
);
368 /* map all the memory banks */
369 for_each_memblock(memory
, reg
) {
370 phys_addr_t start
= reg
->base
;
371 phys_addr_t end
= start
+ reg
->size
;
376 if (ARM64_SWAPPER_USES_SECTION_MAPS
) {
378 * For the first memory bank align the start address and
379 * current memblock limit to prevent create_mapping() from
380 * allocating pte page tables from unmapped memory. With
381 * the section maps, if the first block doesn't end on section
382 * size boundary, create_mapping() will try to allocate a pte
383 * page, which may be returned from an unmapped area.
384 * When section maps are not used, the pte page table for the
385 * current limit is already present in swapper_pg_dir.
388 start
= ALIGN(start
, SECTION_SIZE
);
390 limit
= end
& SECTION_MASK
;
391 memblock_set_current_limit(limit
);
394 __map_memblock(start
, end
);
397 /* Limit no longer required. */
398 memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE
);
401 static void __init
fixup_executable(void)
403 #ifdef CONFIG_DEBUG_RODATA
404 /* now that we are actually fully mapped, make the start/end more fine grained */
405 if (!IS_ALIGNED((unsigned long)_stext
, SWAPPER_BLOCK_SIZE
)) {
406 unsigned long aligned_start
= round_down(__pa(_stext
),
409 create_mapping(aligned_start
, __phys_to_virt(aligned_start
),
410 __pa(_stext
) - aligned_start
,
414 if (!IS_ALIGNED((unsigned long)__init_end
, SWAPPER_BLOCK_SIZE
)) {
415 unsigned long aligned_end
= round_up(__pa(__init_end
),
417 create_mapping(__pa(__init_end
), (unsigned long)__init_end
,
418 aligned_end
- __pa(__init_end
),
424 #ifdef CONFIG_DEBUG_RODATA
425 void mark_rodata_ro(void)
427 create_mapping_late(__pa(_stext
), (unsigned long)_stext
,
428 (unsigned long)_etext
- (unsigned long)_stext
,
434 void fixup_init(void)
436 create_mapping_late(__pa(__init_begin
), (unsigned long)__init_begin
,
437 (unsigned long)__init_end
- (unsigned long)__init_begin
,
442 * paging_init() sets up the page tables, initialises the zone memory
443 * maps and sets up the zero page.
445 void __init
paging_init(void)
452 /* allocate the zero page. */
453 zero_page
= early_alloc(PAGE_SIZE
);
457 empty_zero_page
= virt_to_page(zero_page
);
460 * TTBR0 is only used for the identity mapping at this stage. Make it
461 * point to zero page to avoid speculatively fetching new entries.
463 cpu_set_reserved_ttbr0();
464 local_flush_tlb_all();
465 cpu_set_default_tcr_t0sz();
469 * Check whether a kernel address is valid (derived from arch/x86/).
471 int kern_addr_valid(unsigned long addr
)
478 if ((((long)addr
) >> VA_BITS
) != -1UL)
481 pgd
= pgd_offset_k(addr
);
485 pud
= pud_offset(pgd
, addr
);
490 return pfn_valid(pud_pfn(*pud
));
492 pmd
= pmd_offset(pud
, addr
);
497 return pfn_valid(pmd_pfn(*pmd
));
499 pte
= pte_offset_kernel(pmd
, addr
);
503 return pfn_valid(pte_pfn(*pte
));
505 #ifdef CONFIG_SPARSEMEM_VMEMMAP
506 #if !ARM64_SWAPPER_USES_SECTION_MAPS
507 int __meminit
vmemmap_populate(unsigned long start
, unsigned long end
, int node
)
509 return vmemmap_populate_basepages(start
, end
, node
);
511 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
512 int __meminit
vmemmap_populate(unsigned long start
, unsigned long end
, int node
)
514 unsigned long addr
= start
;
521 next
= pmd_addr_end(addr
, end
);
523 pgd
= vmemmap_pgd_populate(addr
, node
);
527 pud
= vmemmap_pud_populate(pgd
, addr
, node
);
531 pmd
= pmd_offset(pud
, addr
);
532 if (pmd_none(*pmd
)) {
535 p
= vmemmap_alloc_block_buf(PMD_SIZE
, node
);
539 set_pmd(pmd
, __pmd(__pa(p
) | PROT_SECT_NORMAL
));
541 vmemmap_verify((pte_t
*)pmd
, node
, addr
, next
);
542 } while (addr
= next
, addr
!= end
);
546 #endif /* CONFIG_ARM64_64K_PAGES */
547 void vmemmap_free(unsigned long start
, unsigned long end
)
550 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
552 static pte_t bm_pte
[PTRS_PER_PTE
] __page_aligned_bss
;
553 #if CONFIG_PGTABLE_LEVELS > 2
554 static pmd_t bm_pmd
[PTRS_PER_PMD
] __page_aligned_bss
;
556 #if CONFIG_PGTABLE_LEVELS > 3
557 static pud_t bm_pud
[PTRS_PER_PUD
] __page_aligned_bss
;
560 static inline pud_t
* fixmap_pud(unsigned long addr
)
562 pgd_t
*pgd
= pgd_offset_k(addr
);
564 BUG_ON(pgd_none(*pgd
) || pgd_bad(*pgd
));
566 return pud_offset(pgd
, addr
);
569 static inline pmd_t
* fixmap_pmd(unsigned long addr
)
571 pud_t
*pud
= fixmap_pud(addr
);
573 BUG_ON(pud_none(*pud
) || pud_bad(*pud
));
575 return pmd_offset(pud
, addr
);
578 static inline pte_t
* fixmap_pte(unsigned long addr
)
580 pmd_t
*pmd
= fixmap_pmd(addr
);
582 BUG_ON(pmd_none(*pmd
) || pmd_bad(*pmd
));
584 return pte_offset_kernel(pmd
, addr
);
587 void __init
early_fixmap_init(void)
592 unsigned long addr
= FIXADDR_START
;
594 pgd
= pgd_offset_k(addr
);
595 pgd_populate(&init_mm
, pgd
, bm_pud
);
596 pud
= pud_offset(pgd
, addr
);
597 pud_populate(&init_mm
, pud
, bm_pmd
);
598 pmd
= pmd_offset(pud
, addr
);
599 pmd_populate_kernel(&init_mm
, pmd
, bm_pte
);
602 * The boot-ioremap range spans multiple pmds, for which
603 * we are not preparted:
605 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN
) >> PMD_SHIFT
)
606 != (__fix_to_virt(FIX_BTMAP_END
) >> PMD_SHIFT
));
608 if ((pmd
!= fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN
)))
609 || pmd
!= fixmap_pmd(fix_to_virt(FIX_BTMAP_END
))) {
611 pr_warn("pmd %p != %p, %p\n",
612 pmd
, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN
)),
613 fixmap_pmd(fix_to_virt(FIX_BTMAP_END
)));
614 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
615 fix_to_virt(FIX_BTMAP_BEGIN
));
616 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
617 fix_to_virt(FIX_BTMAP_END
));
619 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END
);
620 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN
);
624 void __set_fixmap(enum fixed_addresses idx
,
625 phys_addr_t phys
, pgprot_t flags
)
627 unsigned long addr
= __fix_to_virt(idx
);
630 BUG_ON(idx
<= FIX_HOLE
|| idx
>= __end_of_fixed_addresses
);
632 pte
= fixmap_pte(addr
);
634 if (pgprot_val(flags
)) {
635 set_pte(pte
, pfn_pte(phys
>> PAGE_SHIFT
, flags
));
637 pte_clear(&init_mm
, addr
, pte
);
638 flush_tlb_kernel_range(addr
, addr
+PAGE_SIZE
);
642 void *__init
fixmap_remap_fdt(phys_addr_t dt_phys
)
644 const u64 dt_virt_base
= __fix_to_virt(FIX_FDT
);
645 pgprot_t prot
= PAGE_KERNEL_RO
;
650 * Check whether the physical FDT address is set and meets the minimum
651 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
652 * at least 8 bytes so that we can always access the size field of the
653 * FDT header after mapping the first chunk, double check here if that
654 * is indeed the case.
656 BUILD_BUG_ON(MIN_FDT_ALIGN
< 8);
657 if (!dt_phys
|| dt_phys
% MIN_FDT_ALIGN
)
661 * Make sure that the FDT region can be mapped without the need to
662 * allocate additional translation table pages, so that it is safe
663 * to call create_mapping() this early.
665 * On 64k pages, the FDT will be mapped using PTEs, so we need to
666 * be in the same PMD as the rest of the fixmap.
667 * On 4k pages, we'll use section mappings for the FDT so we only
668 * have to be in the same PUD.
670 BUILD_BUG_ON(dt_virt_base
% SZ_2M
);
672 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END
) >> SWAPPER_TABLE_SHIFT
!=
673 __fix_to_virt(FIX_BTMAP_BEGIN
) >> SWAPPER_TABLE_SHIFT
);
675 offset
= dt_phys
% SWAPPER_BLOCK_SIZE
;
676 dt_virt
= (void *)dt_virt_base
+ offset
;
678 /* map the first chunk so we can read the size from the header */
679 create_mapping(round_down(dt_phys
, SWAPPER_BLOCK_SIZE
), dt_virt_base
,
680 SWAPPER_BLOCK_SIZE
, prot
);
682 if (fdt_check_header(dt_virt
) != 0)
685 size
= fdt_totalsize(dt_virt
);
686 if (size
> MAX_FDT_SIZE
)
689 if (offset
+ size
> SWAPPER_BLOCK_SIZE
)
690 create_mapping(round_down(dt_phys
, SWAPPER_BLOCK_SIZE
), dt_virt_base
,
691 round_up(offset
+ size
, SWAPPER_BLOCK_SIZE
), prot
);
693 memblock_reserve(dt_phys
, size
);