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/cache.h>
21 #include <linux/export.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/init.h>
25 #include <linux/libfdt.h>
26 #include <linux/mman.h>
27 #include <linux/nodemask.h>
28 #include <linux/memblock.h>
32 #include <asm/barrier.h>
33 #include <asm/cputype.h>
34 #include <asm/fixmap.h>
35 #include <asm/kasan.h>
36 #include <asm/kernel-pgtable.h>
37 #include <asm/sections.h>
38 #include <asm/setup.h>
39 #include <asm/sizes.h>
41 #include <asm/memblock.h>
42 #include <asm/mmu_context.h>
43 #include <asm/ptdump.h>
45 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
47 u64 kimage_voffset __ro_after_init;
48 EXPORT_SYMBOL(kimage_voffset);
51 * Empty_zero_page is a special page that is used for zero-initialized data
54 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
55 EXPORT_SYMBOL(empty_zero_page);
57 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
58 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
59 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
61 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
62 unsigned long size, pgprot_t vma_prot)
65 return pgprot_noncached(vma_prot);
66 else if (file->f_flags & O_SYNC)
67 return pgprot_writecombine(vma_prot);
70 EXPORT_SYMBOL(phys_mem_access_prot);
72 static phys_addr_t __init early_pgtable_alloc(void)
77 phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
80 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
81 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
84 ptr = pte_set_fixmap(phys);
86 memset(ptr, 0, PAGE_SIZE);
89 * Implicit barriers also ensure the zeroed page is visible to the page
97 static bool pgattr_change_is_safe(u64 old, u64 new)
100 * The following mapping attributes may be updated in live
101 * kernel mappings without the need for break-before-make.
103 static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE;
105 return old == 0 || new == 0 || ((old ^ new) & ~mask) == 0;
108 static void alloc_init_pte(pmd_t *pmd, unsigned long addr,
109 unsigned long end, unsigned long pfn,
111 phys_addr_t (*pgtable_alloc)(void),
112 bool page_mappings_only)
114 pgprot_t __prot = prot;
117 BUG_ON(pmd_sect(*pmd));
118 if (pmd_none(*pmd)) {
119 phys_addr_t pte_phys;
120 BUG_ON(!pgtable_alloc);
121 pte_phys = pgtable_alloc();
122 pte = pte_set_fixmap(pte_phys);
123 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
126 BUG_ON(pmd_bad(*pmd));
128 pte = pte_set_fixmap_offset(pmd, addr);
130 pte_t old_pte = *pte;
133 * Set the contiguous bit for the subsequent group of PTEs if
134 * its size and alignment are appropriate.
136 if (((addr | PFN_PHYS(pfn)) & ~CONT_PTE_MASK) == 0) {
137 if (end - addr >= CONT_PTE_SIZE && !page_mappings_only)
138 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
143 set_pte(pte, pfn_pte(pfn, __prot));
147 * After the PTE entry has been populated once, we
148 * only allow updates to the permission attributes.
150 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte), pte_val(*pte)));
152 } while (pte++, addr += PAGE_SIZE, addr != end);
157 static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
158 phys_addr_t phys, pgprot_t prot,
159 phys_addr_t (*pgtable_alloc)(void),
160 bool page_mappings_only)
162 pgprot_t __prot = prot;
167 * Check for initial section mappings in the pgd/pud and remove them.
169 BUG_ON(pud_sect(*pud));
170 if (pud_none(*pud)) {
171 phys_addr_t pmd_phys;
172 BUG_ON(!pgtable_alloc);
173 pmd_phys = pgtable_alloc();
174 pmd = pmd_set_fixmap(pmd_phys);
175 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
178 BUG_ON(pud_bad(*pud));
180 pmd = pmd_set_fixmap_offset(pud, addr);
182 pmd_t old_pmd = *pmd;
184 next = pmd_addr_end(addr, end);
186 /* try section mapping first */
187 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
188 !page_mappings_only) {
190 * Set the contiguous bit for the subsequent group of
191 * PMDs if its size and alignment are appropriate.
193 if (((addr | phys) & ~CONT_PMD_MASK) == 0) {
194 if (end - addr >= CONT_PMD_SIZE)
195 __prot = __pgprot(pgprot_val(prot) |
200 pmd_set_huge(pmd, phys, __prot);
203 * After the PMD entry has been populated once, we
204 * only allow updates to the permission attributes.
206 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
209 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
213 BUG_ON(pmd_val(old_pmd) != 0 &&
214 pmd_val(old_pmd) != pmd_val(*pmd));
217 } while (pmd++, addr = next, addr != end);
222 static inline bool use_1G_block(unsigned long addr, unsigned long next,
225 if (PAGE_SHIFT != 12)
228 if (((addr | next | phys) & ~PUD_MASK) != 0)
234 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
235 phys_addr_t phys, pgprot_t prot,
236 phys_addr_t (*pgtable_alloc)(void),
237 bool page_mappings_only)
242 if (pgd_none(*pgd)) {
243 phys_addr_t pud_phys;
244 BUG_ON(!pgtable_alloc);
245 pud_phys = pgtable_alloc();
246 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
248 BUG_ON(pgd_bad(*pgd));
250 pud = pud_set_fixmap_offset(pgd, addr);
252 pud_t old_pud = *pud;
254 next = pud_addr_end(addr, end);
257 * For 4K granule only, attempt to put down a 1GB block
259 if (use_1G_block(addr, next, phys) && !page_mappings_only) {
260 pud_set_huge(pud, phys, prot);
263 * After the PUD entry has been populated once, we
264 * only allow updates to the permission attributes.
266 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
269 alloc_init_pmd(pud, addr, next, phys, prot,
270 pgtable_alloc, page_mappings_only);
272 BUG_ON(pud_val(old_pud) != 0 &&
273 pud_val(old_pud) != pud_val(*pud));
276 } while (pud++, addr = next, addr != end);
281 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
282 unsigned long virt, phys_addr_t size,
284 phys_addr_t (*pgtable_alloc)(void),
285 bool page_mappings_only)
287 unsigned long addr, length, end, next;
288 pgd_t *pgd = pgd_offset_raw(pgdir, virt);
291 * If the virtual and physical address don't have the same offset
292 * within a page, we cannot map the region as the caller expects.
294 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
298 addr = virt & PAGE_MASK;
299 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
303 next = pgd_addr_end(addr, end);
304 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc,
307 } while (pgd++, addr = next, addr != end);
310 static phys_addr_t pgd_pgtable_alloc(void)
312 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
313 if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
316 /* Ensure the zeroed page is visible to the page table walker */
322 * This function can only be used to modify existing table entries,
323 * without allocating new levels of table. Note that this permits the
324 * creation of new section or page entries.
326 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
327 phys_addr_t size, pgprot_t prot)
329 if (virt < VMALLOC_START) {
330 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
334 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL, false);
337 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
338 unsigned long virt, phys_addr_t size,
339 pgprot_t prot, bool page_mappings_only)
341 BUG_ON(mm == &init_mm);
343 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
344 pgd_pgtable_alloc, page_mappings_only);
347 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
348 phys_addr_t size, pgprot_t prot)
350 if (virt < VMALLOC_START) {
351 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
356 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
357 NULL, debug_pagealloc_enabled());
360 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
362 unsigned long kernel_start = __pa(_text);
363 unsigned long kernel_end = __pa(__init_begin);
366 * Take care not to create a writable alias for the
367 * read-only text and rodata sections of the kernel image.
370 /* No overlap with the kernel text/rodata */
371 if (end < kernel_start || start >= kernel_end) {
372 __create_pgd_mapping(pgd, start, __phys_to_virt(start),
373 end - start, PAGE_KERNEL,
375 debug_pagealloc_enabled());
380 * This block overlaps the kernel text/rodata mappings.
381 * Map the portion(s) which don't overlap.
383 if (start < kernel_start)
384 __create_pgd_mapping(pgd, start,
385 __phys_to_virt(start),
386 kernel_start - start, PAGE_KERNEL,
388 debug_pagealloc_enabled());
389 if (kernel_end < end)
390 __create_pgd_mapping(pgd, kernel_end,
391 __phys_to_virt(kernel_end),
392 end - kernel_end, PAGE_KERNEL,
394 debug_pagealloc_enabled());
397 * Map the linear alias of the [_text, __init_begin) interval as
398 * read-only/non-executable. This makes the contents of the
399 * region accessible to subsystems such as hibernate, but
400 * protects it from inadvertent modification or execution.
402 __create_pgd_mapping(pgd, kernel_start, __phys_to_virt(kernel_start),
403 kernel_end - kernel_start, PAGE_KERNEL_RO,
404 early_pgtable_alloc, debug_pagealloc_enabled());
407 static void __init map_mem(pgd_t *pgd)
409 struct memblock_region *reg;
411 /* map all the memory banks */
412 for_each_memblock(memory, reg) {
413 phys_addr_t start = reg->base;
414 phys_addr_t end = start + reg->size;
418 if (memblock_is_nomap(reg))
421 __map_memblock(pgd, start, end);
425 void mark_rodata_ro(void)
427 unsigned long section_size;
429 section_size = (unsigned long)_etext - (unsigned long)_text;
430 create_mapping_late(__pa(_text), (unsigned long)_text,
431 section_size, PAGE_KERNEL_ROX);
433 * mark .rodata as read only. Use __init_begin rather than __end_rodata
434 * to cover NOTES and EXCEPTION_TABLE.
436 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
437 create_mapping_late(__pa(__start_rodata), (unsigned long)__start_rodata,
438 section_size, PAGE_KERNEL_RO);
440 /* flush the TLBs after updating live kernel mappings */
446 static void __init map_kernel_segment(pgd_t *pgd, void *va_start, void *va_end,
447 pgprot_t prot, struct vm_struct *vma)
449 phys_addr_t pa_start = __pa(va_start);
450 unsigned long size = va_end - va_start;
452 BUG_ON(!PAGE_ALIGNED(pa_start));
453 BUG_ON(!PAGE_ALIGNED(size));
455 __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
456 early_pgtable_alloc, debug_pagealloc_enabled());
458 vma->addr = va_start;
459 vma->phys_addr = pa_start;
462 vma->caller = __builtin_return_address(0);
464 vm_area_add_early(vma);
468 * Create fine-grained mappings for the kernel.
470 static void __init map_kernel(pgd_t *pgd)
472 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_init, vmlinux_data;
474 map_kernel_segment(pgd, _text, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
475 map_kernel_segment(pgd, __start_rodata, __init_begin, PAGE_KERNEL, &vmlinux_rodata);
476 map_kernel_segment(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
478 map_kernel_segment(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
480 if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
482 * The fixmap falls in a separate pgd to the kernel, and doesn't
483 * live in the carveout for the swapper_pg_dir. We can simply
484 * re-use the existing dir for the fixmap.
486 set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
487 *pgd_offset_k(FIXADDR_START));
488 } else if (CONFIG_PGTABLE_LEVELS > 3) {
490 * The fixmap shares its top level pgd entry with the kernel
491 * mapping. This can really only occur when we are running
492 * with 16k/4 levels, so we can simply reuse the pud level
495 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
496 set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
497 __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
503 kasan_copy_shadow(pgd);
507 * paging_init() sets up the page tables, initialises the zone memory
508 * maps and sets up the zero page.
510 void __init paging_init(void)
512 phys_addr_t pgd_phys = early_pgtable_alloc();
513 pgd_t *pgd = pgd_set_fixmap(pgd_phys);
519 * We want to reuse the original swapper_pg_dir so we don't have to
520 * communicate the new address to non-coherent secondaries in
521 * secondary_entry, and so cpu_switch_mm can generate the address with
522 * adrp+add rather than a load from some global variable.
524 * To do this we need to go via a temporary pgd.
526 cpu_replace_ttbr1(__va(pgd_phys));
527 memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
528 cpu_replace_ttbr1(swapper_pg_dir);
531 memblock_free(pgd_phys, PAGE_SIZE);
534 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
537 memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
538 SWAPPER_DIR_SIZE - PAGE_SIZE);
542 * Check whether a kernel address is valid (derived from arch/x86/).
544 int kern_addr_valid(unsigned long addr)
551 if ((((long)addr) >> VA_BITS) != -1UL)
554 pgd = pgd_offset_k(addr);
558 pud = pud_offset(pgd, addr);
563 return pfn_valid(pud_pfn(*pud));
565 pmd = pmd_offset(pud, addr);
570 return pfn_valid(pmd_pfn(*pmd));
572 pte = pte_offset_kernel(pmd, addr);
576 return pfn_valid(pte_pfn(*pte));
578 #ifdef CONFIG_SPARSEMEM_VMEMMAP
579 #if !ARM64_SWAPPER_USES_SECTION_MAPS
580 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
582 return vmemmap_populate_basepages(start, end, node);
584 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
585 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
587 unsigned long addr = start;
594 next = pmd_addr_end(addr, end);
596 pgd = vmemmap_pgd_populate(addr, node);
600 pud = vmemmap_pud_populate(pgd, addr, node);
604 pmd = pmd_offset(pud, addr);
605 if (pmd_none(*pmd)) {
608 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
612 set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
614 vmemmap_verify((pte_t *)pmd, node, addr, next);
615 } while (addr = next, addr != end);
619 #endif /* CONFIG_ARM64_64K_PAGES */
620 void vmemmap_free(unsigned long start, unsigned long end)
623 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
625 static inline pud_t * fixmap_pud(unsigned long addr)
627 pgd_t *pgd = pgd_offset_k(addr);
629 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
631 return pud_offset_kimg(pgd, addr);
634 static inline pmd_t * fixmap_pmd(unsigned long addr)
636 pud_t *pud = fixmap_pud(addr);
638 BUG_ON(pud_none(*pud) || pud_bad(*pud));
640 return pmd_offset_kimg(pud, addr);
643 static inline pte_t * fixmap_pte(unsigned long addr)
645 return &bm_pte[pte_index(addr)];
648 void __init early_fixmap_init(void)
653 unsigned long addr = FIXADDR_START;
655 pgd = pgd_offset_k(addr);
656 if (CONFIG_PGTABLE_LEVELS > 3 &&
657 !(pgd_none(*pgd) || pgd_page_paddr(*pgd) == __pa(bm_pud))) {
659 * We only end up here if the kernel mapping and the fixmap
660 * share the top level pgd entry, which should only happen on
661 * 16k/4 levels configurations.
663 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
664 pud = pud_offset_kimg(pgd, addr);
666 pgd_populate(&init_mm, pgd, bm_pud);
667 pud = fixmap_pud(addr);
669 pud_populate(&init_mm, pud, bm_pmd);
670 pmd = fixmap_pmd(addr);
671 pmd_populate_kernel(&init_mm, pmd, bm_pte);
674 * The boot-ioremap range spans multiple pmds, for which
675 * we are not prepared:
677 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
678 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
680 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
681 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
683 pr_warn("pmd %p != %p, %p\n",
684 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
685 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
686 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
687 fix_to_virt(FIX_BTMAP_BEGIN));
688 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
689 fix_to_virt(FIX_BTMAP_END));
691 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
692 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
696 void __set_fixmap(enum fixed_addresses idx,
697 phys_addr_t phys, pgprot_t flags)
699 unsigned long addr = __fix_to_virt(idx);
702 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
704 pte = fixmap_pte(addr);
706 if (pgprot_val(flags)) {
707 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
709 pte_clear(&init_mm, addr, pte);
710 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
714 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
716 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
721 * Check whether the physical FDT address is set and meets the minimum
722 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
723 * at least 8 bytes so that we can always access the magic and size
724 * fields of the FDT header after mapping the first chunk, double check
725 * here if that is indeed the case.
727 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
728 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
732 * Make sure that the FDT region can be mapped without the need to
733 * allocate additional translation table pages, so that it is safe
734 * to call create_mapping_noalloc() this early.
736 * On 64k pages, the FDT will be mapped using PTEs, so we need to
737 * be in the same PMD as the rest of the fixmap.
738 * On 4k pages, we'll use section mappings for the FDT so we only
739 * have to be in the same PUD.
741 BUILD_BUG_ON(dt_virt_base % SZ_2M);
743 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
744 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
746 offset = dt_phys % SWAPPER_BLOCK_SIZE;
747 dt_virt = (void *)dt_virt_base + offset;
749 /* map the first chunk so we can read the size from the header */
750 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
751 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
753 if (fdt_magic(dt_virt) != FDT_MAGIC)
756 *size = fdt_totalsize(dt_virt);
757 if (*size > MAX_FDT_SIZE)
760 if (offset + *size > SWAPPER_BLOCK_SIZE)
761 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
762 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
767 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
772 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
776 memblock_reserve(dt_phys, size);
780 int __init arch_ioremap_pud_supported(void)
782 /* only 4k granule supports level 1 block mappings */
783 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
786 int __init arch_ioremap_pmd_supported(void)
791 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
793 BUG_ON(phys & ~PUD_MASK);
794 set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
798 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
800 BUG_ON(phys & ~PMD_MASK);
801 set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
805 int pud_clear_huge(pud_t *pud)
813 int pmd_clear_huge(pmd_t *pmd)