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/barrier.h>
34 #include <asm/cputype.h>
35 #include <asm/fixmap.h>
36 #include <asm/kasan.h>
37 #include <asm/kernel-pgtable.h>
38 #include <asm/sections.h>
39 #include <asm/setup.h>
40 #include <asm/sizes.h>
42 #include <asm/memblock.h>
43 #include <asm/mmu_context.h>
47 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
49 u64 kimage_voffset __read_mostly;
50 EXPORT_SYMBOL(kimage_voffset);
53 * Empty_zero_page is a special page that is used for zero-initialized data
56 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
57 EXPORT_SYMBOL(empty_zero_page);
59 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
60 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
61 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
63 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
64 unsigned long size, pgprot_t vma_prot)
67 return pgprot_noncached(vma_prot);
68 else if (file->f_flags & O_SYNC)
69 return pgprot_writecombine(vma_prot);
72 EXPORT_SYMBOL(phys_mem_access_prot);
74 static phys_addr_t __init early_pgtable_alloc(void)
79 phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
83 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
84 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
87 ptr = pte_set_fixmap(phys);
89 memset(ptr, 0, PAGE_SIZE);
92 * Implicit barriers also ensure the zeroed page is visible to the page
101 * remap a PMD into pages
103 static void split_pmd(pmd_t *pmd, pte_t *pte)
105 unsigned long pfn = pmd_pfn(*pmd);
110 * Need to have the least restrictive permissions available
111 * permissions will be fixed up later
113 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
115 } while (pte++, i++, i < PTRS_PER_PTE);
118 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
119 unsigned long end, unsigned long pfn,
121 phys_addr_t (*pgtable_alloc)(void))
125 if (pmd_none(*pmd) || pmd_sect(*pmd)) {
126 phys_addr_t pte_phys;
127 BUG_ON(!pgtable_alloc);
128 pte_phys = pgtable_alloc();
129 pte = pte_set_fixmap(pte_phys);
132 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
136 BUG_ON(pmd_bad(*pmd));
138 pte = pte_set_fixmap_offset(pmd, addr);
140 set_pte(pte, pfn_pte(pfn, prot));
142 } while (pte++, addr += PAGE_SIZE, addr != end);
147 static void split_pud(pud_t *old_pud, pmd_t *pmd)
149 unsigned long addr = pud_pfn(*old_pud) << PAGE_SHIFT;
150 pgprot_t prot = __pgprot(pud_val(*old_pud) ^ addr);
154 set_pmd(pmd, __pmd(addr | pgprot_val(prot)));
156 } while (pmd++, i++, i < PTRS_PER_PMD);
159 #ifdef CONFIG_DEBUG_PAGEALLOC
160 static bool block_mappings_allowed(phys_addr_t (*pgtable_alloc)(void))
164 * If debug_page_alloc is enabled we must map the linear map
165 * using pages. However, other mappings created by
166 * create_mapping_noalloc must use sections in some cases. Allow
167 * sections to be used in those cases, where no pgtable_alloc
168 * function is provided.
170 return !pgtable_alloc || !debug_pagealloc_enabled();
173 static bool block_mappings_allowed(phys_addr_t (*pgtable_alloc)(void))
179 static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
180 phys_addr_t phys, pgprot_t prot,
181 phys_addr_t (*pgtable_alloc)(void))
187 * Check for initial section mappings in the pgd/pud and remove them.
189 if (pud_none(*pud) || pud_sect(*pud)) {
190 phys_addr_t pmd_phys;
191 BUG_ON(!pgtable_alloc);
192 pmd_phys = pgtable_alloc();
193 pmd = pmd_set_fixmap(pmd_phys);
194 if (pud_sect(*pud)) {
196 * need to have the 1G of mappings continue to be
201 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
205 BUG_ON(pud_bad(*pud));
207 pmd = pmd_set_fixmap_offset(pud, addr);
209 next = pmd_addr_end(addr, end);
210 /* try section mapping first */
211 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
212 block_mappings_allowed(pgtable_alloc)) {
214 set_pmd(pmd, __pmd(phys |
215 pgprot_val(mk_sect_prot(prot))));
217 * Check for previous table entries created during
218 * boot (__create_page_tables) and flush them.
220 if (!pmd_none(old_pmd)) {
222 if (pmd_table(old_pmd)) {
223 phys_addr_t table = pmd_page_paddr(old_pmd);
224 if (!WARN_ON_ONCE(slab_is_available()))
225 memblock_free(table, PAGE_SIZE);
229 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
230 prot, pgtable_alloc);
233 } while (pmd++, addr = next, addr != end);
238 static inline bool use_1G_block(unsigned long addr, unsigned long next,
241 if (PAGE_SHIFT != 12)
244 if (((addr | next | phys) & ~PUD_MASK) != 0)
250 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
251 phys_addr_t phys, pgprot_t prot,
252 phys_addr_t (*pgtable_alloc)(void))
257 if (pgd_none(*pgd)) {
258 phys_addr_t pud_phys;
259 BUG_ON(!pgtable_alloc);
260 pud_phys = pgtable_alloc();
261 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
263 BUG_ON(pgd_bad(*pgd));
265 pud = pud_set_fixmap_offset(pgd, addr);
267 next = pud_addr_end(addr, end);
270 * For 4K granule only, attempt to put down a 1GB block
272 if (use_1G_block(addr, next, phys) &&
273 block_mappings_allowed(pgtable_alloc)) {
274 pud_t old_pud = *pud;
275 set_pud(pud, __pud(phys |
276 pgprot_val(mk_sect_prot(prot))));
279 * If we have an old value for a pud, it will
280 * be pointing to a pmd table that we no longer
281 * need (from swapper_pg_dir).
283 * Look up the old pmd table and free it.
285 if (!pud_none(old_pud)) {
287 if (pud_table(old_pud)) {
288 phys_addr_t table = pud_page_paddr(old_pud);
289 if (!WARN_ON_ONCE(slab_is_available()))
290 memblock_free(table, PAGE_SIZE);
294 alloc_init_pmd(pud, addr, next, phys, prot,
298 } while (pud++, addr = next, addr != end);
304 * Create the page directory entries and any necessary page tables for the
305 * mapping specified by 'md'.
307 static void init_pgd(pgd_t *pgd, phys_addr_t phys, unsigned long virt,
308 phys_addr_t size, pgprot_t prot,
309 phys_addr_t (*pgtable_alloc)(void))
311 unsigned long addr, length, end, next;
314 * If the virtual and physical address don't have the same offset
315 * within a page, we cannot map the region as the caller expects.
317 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
321 addr = virt & PAGE_MASK;
322 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
326 next = pgd_addr_end(addr, end);
327 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc);
329 } while (pgd++, addr = next, addr != end);
332 static phys_addr_t late_pgtable_alloc(void)
334 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
337 /* Ensure the zeroed page is visible to the page table walker */
342 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
343 unsigned long virt, phys_addr_t size,
345 phys_addr_t (*alloc)(void))
347 init_pgd(pgd_offset_raw(pgdir, virt), phys, virt, size, prot, alloc);
351 * This function can only be used to modify existing table entries,
352 * without allocating new levels of table. Note that this permits the
353 * creation of new section or page entries.
355 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
356 phys_addr_t size, pgprot_t prot)
358 if (virt < VMALLOC_START) {
359 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
363 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
367 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
368 unsigned long virt, phys_addr_t size,
371 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
375 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
376 phys_addr_t size, pgprot_t prot)
378 if (virt < VMALLOC_START) {
379 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
384 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
388 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
390 unsigned long kernel_start = __pa(_stext);
391 unsigned long kernel_end = __pa(_etext);
394 * Take care not to create a writable alias for the
395 * read-only text and rodata sections of the kernel image.
398 /* No overlap with the kernel text */
399 if (end < kernel_start || start >= kernel_end) {
400 __create_pgd_mapping(pgd, start, __phys_to_virt(start),
401 end - start, PAGE_KERNEL,
402 early_pgtable_alloc);
407 * This block overlaps the kernel text mapping.
408 * Map the portion(s) which don't overlap.
410 if (start < kernel_start)
411 __create_pgd_mapping(pgd, start,
412 __phys_to_virt(start),
413 kernel_start - start, PAGE_KERNEL,
414 early_pgtable_alloc);
415 if (kernel_end < end)
416 __create_pgd_mapping(pgd, kernel_end,
417 __phys_to_virt(kernel_end),
418 end - kernel_end, PAGE_KERNEL,
419 early_pgtable_alloc);
422 * Map the linear alias of the [_stext, _etext) interval as
423 * read-only/non-executable. This makes the contents of the
424 * region accessible to subsystems such as hibernate, but
425 * protects it from inadvertent modification or execution.
427 __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
428 kernel_end - kernel_start, PAGE_KERNEL_RO,
429 early_pgtable_alloc);
432 static void __init map_mem(pgd_t *pgd)
434 struct memblock_region *reg;
436 /* map all the memory banks */
437 for_each_memblock(memory, reg) {
438 phys_addr_t start = reg->base;
439 phys_addr_t end = start + reg->size;
443 if (memblock_is_nomap(reg))
446 __map_memblock(pgd, start, end);
450 void mark_rodata_ro(void)
452 if (!IS_ENABLED(CONFIG_DEBUG_RODATA))
455 create_mapping_late(__pa(_stext), (unsigned long)_stext,
456 (unsigned long)_etext - (unsigned long)_stext,
460 void fixup_init(void)
463 * Unmap the __init region but leave the VM area in place. This
464 * prevents the region from being reused for kernel modules, which
465 * is not supported by kallsyms.
467 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
470 static void __init map_kernel_chunk(pgd_t *pgd, void *va_start, void *va_end,
471 pgprot_t prot, struct vm_struct *vma)
473 phys_addr_t pa_start = __pa(va_start);
474 unsigned long size = va_end - va_start;
476 BUG_ON(!PAGE_ALIGNED(pa_start));
477 BUG_ON(!PAGE_ALIGNED(size));
479 __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
480 early_pgtable_alloc);
482 vma->addr = va_start;
483 vma->phys_addr = pa_start;
486 vma->caller = __builtin_return_address(0);
488 vm_area_add_early(vma);
492 * Create fine-grained mappings for the kernel.
494 static void __init map_kernel(pgd_t *pgd)
496 static struct vm_struct vmlinux_text, vmlinux_init, vmlinux_data;
498 map_kernel_chunk(pgd, _stext, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
499 map_kernel_chunk(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
501 map_kernel_chunk(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
503 if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
505 * The fixmap falls in a separate pgd to the kernel, and doesn't
506 * live in the carveout for the swapper_pg_dir. We can simply
507 * re-use the existing dir for the fixmap.
509 set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
510 *pgd_offset_k(FIXADDR_START));
511 } else if (CONFIG_PGTABLE_LEVELS > 3) {
513 * The fixmap shares its top level pgd entry with the kernel
514 * mapping. This can really only occur when we are running
515 * with 16k/4 levels, so we can simply reuse the pud level
518 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
519 set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
520 __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
526 kasan_copy_shadow(pgd);
530 * paging_init() sets up the page tables, initialises the zone memory
531 * maps and sets up the zero page.
533 void __init paging_init(void)
535 phys_addr_t pgd_phys = early_pgtable_alloc();
536 pgd_t *pgd = pgd_set_fixmap(pgd_phys);
542 * We want to reuse the original swapper_pg_dir so we don't have to
543 * communicate the new address to non-coherent secondaries in
544 * secondary_entry, and so cpu_switch_mm can generate the address with
545 * adrp+add rather than a load from some global variable.
547 * To do this we need to go via a temporary pgd.
549 cpu_replace_ttbr1(__va(pgd_phys));
550 memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
551 cpu_replace_ttbr1(swapper_pg_dir);
554 memblock_free(pgd_phys, PAGE_SIZE);
557 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
560 memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
561 SWAPPER_DIR_SIZE - PAGE_SIZE);
567 * Check whether a kernel address is valid (derived from arch/x86/).
569 int kern_addr_valid(unsigned long addr)
576 if ((((long)addr) >> VA_BITS) != -1UL)
579 pgd = pgd_offset_k(addr);
583 pud = pud_offset(pgd, addr);
588 return pfn_valid(pud_pfn(*pud));
590 pmd = pmd_offset(pud, addr);
595 return pfn_valid(pmd_pfn(*pmd));
597 pte = pte_offset_kernel(pmd, addr);
601 return pfn_valid(pte_pfn(*pte));
603 #ifdef CONFIG_SPARSEMEM_VMEMMAP
604 #if !ARM64_SWAPPER_USES_SECTION_MAPS
605 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
607 return vmemmap_populate_basepages(start, end, node);
609 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
610 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
612 unsigned long addr = start;
619 next = pmd_addr_end(addr, end);
621 pgd = vmemmap_pgd_populate(addr, node);
625 pud = vmemmap_pud_populate(pgd, addr, node);
629 pmd = pmd_offset(pud, addr);
630 if (pmd_none(*pmd)) {
633 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
637 set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
639 vmemmap_verify((pte_t *)pmd, node, addr, next);
640 } while (addr = next, addr != end);
644 #endif /* CONFIG_ARM64_64K_PAGES */
645 void vmemmap_free(unsigned long start, unsigned long end)
648 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
650 static inline pud_t * fixmap_pud(unsigned long addr)
652 pgd_t *pgd = pgd_offset_k(addr);
654 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
656 return pud_offset_kimg(pgd, addr);
659 static inline pmd_t * fixmap_pmd(unsigned long addr)
661 pud_t *pud = fixmap_pud(addr);
663 BUG_ON(pud_none(*pud) || pud_bad(*pud));
665 return pmd_offset_kimg(pud, addr);
668 static inline pte_t * fixmap_pte(unsigned long addr)
670 return &bm_pte[pte_index(addr)];
673 void __init early_fixmap_init(void)
678 unsigned long addr = FIXADDR_START;
680 pgd = pgd_offset_k(addr);
681 if (CONFIG_PGTABLE_LEVELS > 3 && !pgd_none(*pgd)) {
683 * We only end up here if the kernel mapping and the fixmap
684 * share the top level pgd entry, which should only happen on
685 * 16k/4 levels configurations.
687 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
688 pud = pud_offset_kimg(pgd, addr);
690 pgd_populate(&init_mm, pgd, bm_pud);
691 pud = fixmap_pud(addr);
693 pud_populate(&init_mm, pud, bm_pmd);
694 pmd = fixmap_pmd(addr);
695 pmd_populate_kernel(&init_mm, pmd, bm_pte);
698 * The boot-ioremap range spans multiple pmds, for which
699 * we are not prepared:
701 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
702 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
704 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
705 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
707 pr_warn("pmd %p != %p, %p\n",
708 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
709 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
710 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
711 fix_to_virt(FIX_BTMAP_BEGIN));
712 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
713 fix_to_virt(FIX_BTMAP_END));
715 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
716 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
720 void __set_fixmap(enum fixed_addresses idx,
721 phys_addr_t phys, pgprot_t flags)
723 unsigned long addr = __fix_to_virt(idx);
726 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
728 pte = fixmap_pte(addr);
730 if (pgprot_val(flags)) {
731 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
733 pte_clear(&init_mm, addr, pte);
734 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
738 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
740 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
741 pgprot_t prot = PAGE_KERNEL_RO;
746 * Check whether the physical FDT address is set and meets the minimum
747 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
748 * at least 8 bytes so that we can always access the size field of the
749 * FDT header after mapping the first chunk, double check here if that
750 * is indeed the case.
752 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
753 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
757 * Make sure that the FDT region can be mapped without the need to
758 * allocate additional translation table pages, so that it is safe
759 * to call create_mapping_noalloc() this early.
761 * On 64k pages, the FDT will be mapped using PTEs, so we need to
762 * be in the same PMD as the rest of the fixmap.
763 * On 4k pages, we'll use section mappings for the FDT so we only
764 * have to be in the same PUD.
766 BUILD_BUG_ON(dt_virt_base % SZ_2M);
768 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
769 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
771 offset = dt_phys % SWAPPER_BLOCK_SIZE;
772 dt_virt = (void *)dt_virt_base + offset;
774 /* map the first chunk so we can read the size from the header */
775 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
776 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
778 if (fdt_check_header(dt_virt) != 0)
781 size = fdt_totalsize(dt_virt);
782 if (size > MAX_FDT_SIZE)
785 if (offset + size > SWAPPER_BLOCK_SIZE)
786 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
787 round_up(offset + size, SWAPPER_BLOCK_SIZE), prot);
789 memblock_reserve(dt_phys, size);
794 int __init arch_ioremap_pud_supported(void)
796 /* only 4k granule supports level 1 block mappings */
797 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
800 int __init arch_ioremap_pmd_supported(void)
805 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
807 BUG_ON(phys & ~PUD_MASK);
808 set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
812 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
814 BUG_ON(phys & ~PMD_MASK);
815 set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
819 int pud_clear_huge(pud_t *pud)
827 int pmd_clear_huge(pmd_t *pmd)