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/mman.h>
25 #include <linux/nodemask.h>
26 #include <linux/memblock.h>
29 #include <linux/slab.h>
30 #include <linux/stop_machine.h>
32 #include <asm/cputype.h>
33 #include <asm/fixmap.h>
34 #include <asm/sections.h>
35 #include <asm/setup.h>
36 #include <asm/sizes.h>
38 #include <asm/memblock.h>
39 #include <asm/mmu_context.h>
44 * Empty_zero_page is a special page that is used for zero-initialized data
47 struct page *empty_zero_page;
48 EXPORT_SYMBOL(empty_zero_page);
50 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
51 unsigned long size, pgprot_t vma_prot)
54 return pgprot_noncached(vma_prot);
55 else if (file->f_flags & O_SYNC)
56 return pgprot_writecombine(vma_prot);
59 EXPORT_SYMBOL(phys_mem_access_prot);
61 static void __init *early_alloc(unsigned long sz)
63 void *ptr = __va(memblock_alloc(sz, sz));
70 * remap a PMD into pages
72 static void split_pmd(pmd_t *pmd, pte_t *pte)
74 unsigned long pfn = pmd_pfn(*pmd);
79 * Need to have the least restrictive permissions available
80 * permissions will be fixed up later
82 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
84 } while (pte++, i++, i < PTRS_PER_PTE);
87 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
88 unsigned long end, unsigned long pfn,
90 void *(*alloc)(unsigned long size))
94 if (pmd_none(*pmd) || pmd_sect(*pmd)) {
95 pte = alloc(PTRS_PER_PTE * sizeof(pte_t));
98 __pmd_populate(pmd, __pa(pte), PMD_TYPE_TABLE);
101 BUG_ON(pmd_bad(*pmd));
103 pte = pte_offset_kernel(pmd, addr);
105 set_pte(pte, pfn_pte(pfn, prot));
107 } while (pte++, addr += PAGE_SIZE, addr != end);
110 void split_pud(pud_t *old_pud, pmd_t *pmd)
112 unsigned long addr = pud_pfn(*old_pud) << PAGE_SHIFT;
113 pgprot_t prot = __pgprot(pud_val(*old_pud) ^ addr);
117 set_pmd(pmd, __pmd(addr | prot));
119 } while (pmd++, i++, i < PTRS_PER_PMD);
122 static void alloc_init_pmd(struct mm_struct *mm, pud_t *pud,
123 unsigned long addr, unsigned long end,
124 phys_addr_t phys, pgprot_t prot,
125 void *(*alloc)(unsigned long size))
131 * Check for initial section mappings in the pgd/pud and remove them.
133 if (pud_none(*pud) || pud_sect(*pud)) {
134 pmd = alloc(PTRS_PER_PMD * sizeof(pmd_t));
135 if (pud_sect(*pud)) {
137 * need to have the 1G of mappings continue to be
142 pud_populate(mm, pud, pmd);
145 BUG_ON(pud_bad(*pud));
147 pmd = pmd_offset(pud, addr);
149 next = pmd_addr_end(addr, end);
150 /* try section mapping first */
151 if (((addr | next | phys) & ~SECTION_MASK) == 0) {
153 set_pmd(pmd, __pmd(phys |
154 pgprot_val(mk_sect_prot(prot))));
156 * Check for previous table entries created during
157 * boot (__create_page_tables) and flush them.
159 if (!pmd_none(old_pmd)) {
161 if (pmd_table(old_pmd)) {
162 phys_addr_t table = __pa(pte_offset_map(&old_pmd, 0));
163 if (!WARN_ON_ONCE(slab_is_available()))
164 memblock_free(table, PAGE_SIZE);
168 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
172 } while (pmd++, addr = next, addr != end);
175 static inline bool use_1G_block(unsigned long addr, unsigned long next,
178 if (PAGE_SHIFT != 12)
181 if (((addr | next | phys) & ~PUD_MASK) != 0)
187 static void alloc_init_pud(struct mm_struct *mm, pgd_t *pgd,
188 unsigned long addr, unsigned long end,
189 phys_addr_t phys, pgprot_t prot,
190 void *(*alloc)(unsigned long size))
195 if (pgd_none(*pgd)) {
196 pud = alloc(PTRS_PER_PUD * sizeof(pud_t));
197 pgd_populate(mm, pgd, pud);
199 BUG_ON(pgd_bad(*pgd));
201 pud = pud_offset(pgd, addr);
203 next = pud_addr_end(addr, end);
206 * For 4K granule only, attempt to put down a 1GB block
208 if (use_1G_block(addr, next, phys)) {
209 pud_t old_pud = *pud;
210 set_pud(pud, __pud(phys |
211 pgprot_val(mk_sect_prot(prot))));
214 * If we have an old value for a pud, it will
215 * be pointing to a pmd table that we no longer
216 * need (from swapper_pg_dir).
218 * Look up the old pmd table and free it.
220 if (!pud_none(old_pud)) {
222 if (pud_table(old_pud)) {
223 phys_addr_t table = __pa(pmd_offset(&old_pud, 0));
224 if (!WARN_ON_ONCE(slab_is_available()))
225 memblock_free(table, PAGE_SIZE);
229 alloc_init_pmd(mm, pud, addr, next, phys, prot, alloc);
232 } while (pud++, addr = next, addr != end);
236 * Create the page directory entries and any necessary page tables for the
237 * mapping specified by 'md'.
239 static void __create_mapping(struct mm_struct *mm, pgd_t *pgd,
240 phys_addr_t phys, unsigned long virt,
241 phys_addr_t size, pgprot_t prot,
242 void *(*alloc)(unsigned long size))
244 unsigned long addr, length, end, next;
246 addr = virt & PAGE_MASK;
247 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
251 next = pgd_addr_end(addr, end);
252 alloc_init_pud(mm, pgd, addr, next, phys, prot, alloc);
254 } while (pgd++, addr = next, addr != end);
257 static void *late_alloc(unsigned long size)
261 BUG_ON(size > PAGE_SIZE);
262 ptr = (void *)__get_free_page(PGALLOC_GFP);
267 static void __ref create_mapping(phys_addr_t phys, unsigned long virt,
268 phys_addr_t size, pgprot_t prot)
270 if (virt < VMALLOC_START) {
271 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
275 __create_mapping(&init_mm, pgd_offset_k(virt & PAGE_MASK), phys, virt,
276 size, prot, early_alloc);
279 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
280 unsigned long virt, phys_addr_t size,
283 __create_mapping(mm, pgd_offset(mm, virt), phys, virt, size, prot,
287 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
288 phys_addr_t size, pgprot_t prot)
290 if (virt < VMALLOC_START) {
291 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
296 return __create_mapping(&init_mm, pgd_offset_k(virt & PAGE_MASK),
297 phys, virt, size, prot, late_alloc);
300 #ifdef CONFIG_DEBUG_RODATA
301 static void __init __map_memblock(phys_addr_t start, phys_addr_t end)
304 * Set up the executable regions using the existing section mappings
305 * for now. This will get more fine grained later once all memory
308 unsigned long kernel_x_start = round_down(__pa(_stext), SECTION_SIZE);
309 unsigned long kernel_x_end = round_up(__pa(__init_end), SECTION_SIZE);
311 if (end < kernel_x_start) {
312 create_mapping(start, __phys_to_virt(start),
313 end - start, PAGE_KERNEL);
314 } else if (start >= kernel_x_end) {
315 create_mapping(start, __phys_to_virt(start),
316 end - start, PAGE_KERNEL);
318 if (start < kernel_x_start)
319 create_mapping(start, __phys_to_virt(start),
320 kernel_x_start - start,
322 create_mapping(kernel_x_start,
323 __phys_to_virt(kernel_x_start),
324 kernel_x_end - kernel_x_start,
326 if (kernel_x_end < end)
327 create_mapping(kernel_x_end,
328 __phys_to_virt(kernel_x_end),
335 static void __init __map_memblock(phys_addr_t start, phys_addr_t end)
337 create_mapping(start, __phys_to_virt(start), end - start,
342 static void __init map_mem(void)
344 struct memblock_region *reg;
348 * Temporarily limit the memblock range. We need to do this as
349 * create_mapping requires puds, pmds and ptes to be allocated from
350 * memory addressable from the initial direct kernel mapping.
352 * The initial direct kernel mapping, located at swapper_pg_dir, gives
353 * us PUD_SIZE (4K pages) or PMD_SIZE (64K pages) memory starting from
354 * PHYS_OFFSET (which must be aligned to 2MB as per
355 * Documentation/arm64/booting.txt).
357 if (IS_ENABLED(CONFIG_ARM64_64K_PAGES))
358 limit = PHYS_OFFSET + PMD_SIZE;
360 limit = PHYS_OFFSET + PUD_SIZE;
361 memblock_set_current_limit(limit);
363 /* map all the memory banks */
364 for_each_memblock(memory, reg) {
365 phys_addr_t start = reg->base;
366 phys_addr_t end = start + reg->size;
371 #ifndef CONFIG_ARM64_64K_PAGES
373 * For the first memory bank align the start address and
374 * current memblock limit to prevent create_mapping() from
375 * allocating pte page tables from unmapped memory.
376 * When 64K pages are enabled, the pte page table for the
377 * first PGDIR_SIZE is already present in swapper_pg_dir.
380 start = ALIGN(start, PMD_SIZE);
382 limit = end & PMD_MASK;
383 memblock_set_current_limit(limit);
386 __map_memblock(start, end);
389 /* Limit no longer required. */
390 memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
393 void __init fixup_executable(void)
395 #ifdef CONFIG_DEBUG_RODATA
396 /* now that we are actually fully mapped, make the start/end more fine grained */
397 if (!IS_ALIGNED((unsigned long)_stext, SECTION_SIZE)) {
398 unsigned long aligned_start = round_down(__pa(_stext),
401 create_mapping(aligned_start, __phys_to_virt(aligned_start),
402 __pa(_stext) - aligned_start,
406 if (!IS_ALIGNED((unsigned long)__init_end, SECTION_SIZE)) {
407 unsigned long aligned_end = round_up(__pa(__init_end),
409 create_mapping(__pa(__init_end), (unsigned long)__init_end,
410 aligned_end - __pa(__init_end),
416 #ifdef CONFIG_DEBUG_RODATA
417 void mark_rodata_ro(void)
419 create_mapping_late(__pa(_stext), (unsigned long)_stext,
420 (unsigned long)_etext - (unsigned long)_stext,
421 PAGE_KERNEL_EXEC | PTE_RDONLY);
426 void fixup_init(void)
428 create_mapping_late(__pa(__init_begin), (unsigned long)__init_begin,
429 (unsigned long)__init_end - (unsigned long)__init_begin,
434 * paging_init() sets up the page tables, initialises the zone memory
435 * maps and sets up the zero page.
437 void __init paging_init(void)
444 /* allocate the zero page. */
445 zero_page = early_alloc(PAGE_SIZE);
449 empty_zero_page = virt_to_page(zero_page);
452 * TTBR0 is only used for the identity mapping at this stage. Make it
453 * point to zero page to avoid speculatively fetching new entries.
455 cpu_set_reserved_ttbr0();
460 * Enable the identity mapping to allow the MMU disabling.
462 void setup_mm_for_reboot(void)
464 cpu_switch_mm(idmap_pg_dir, &init_mm);
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 #ifdef CONFIG_ARM64_64K_PAGES
507 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
509 return vmemmap_populate_basepages(start, end, node);
511 #else /* !CONFIG_ARM64_64K_PAGES */
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 if (idx >= __end_of_fixed_addresses) {
635 pte = fixmap_pte(addr);
637 if (pgprot_val(flags)) {
638 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
640 pte_clear(&init_mm, addr, pte);
641 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);