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 = pgtable_alloc();
127 pte = pte_set_fixmap(pte_phys);
130 __pmd_populate(pmd, pte_phys, PMD_TYPE_TABLE);
134 BUG_ON(pmd_bad(*pmd));
136 pte = pte_set_fixmap_offset(pmd, addr);
138 set_pte(pte, pfn_pte(pfn, prot));
140 } while (pte++, addr += PAGE_SIZE, addr != end);
145 static void split_pud(pud_t *old_pud, pmd_t *pmd)
147 unsigned long addr = pud_pfn(*old_pud) << PAGE_SHIFT;
148 pgprot_t prot = __pgprot(pud_val(*old_pud) ^ addr);
152 set_pmd(pmd, __pmd(addr | pgprot_val(prot)));
154 } while (pmd++, i++, i < PTRS_PER_PMD);
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))
165 * Check for initial section mappings in the pgd/pud and remove them.
167 if (pud_none(*pud) || pud_sect(*pud)) {
168 phys_addr_t pmd_phys = pgtable_alloc();
169 pmd = pmd_set_fixmap(pmd_phys);
170 if (pud_sect(*pud)) {
172 * need to have the 1G of mappings continue to be
177 __pud_populate(pud, pmd_phys, PUD_TYPE_TABLE);
181 BUG_ON(pud_bad(*pud));
183 pmd = pmd_set_fixmap_offset(pud, addr);
185 next = pmd_addr_end(addr, end);
186 /* try section mapping first */
187 if (((addr | next | phys) & ~SECTION_MASK) == 0) {
189 set_pmd(pmd, __pmd(phys |
190 pgprot_val(mk_sect_prot(prot))));
192 * Check for previous table entries created during
193 * boot (__create_page_tables) and flush them.
195 if (!pmd_none(old_pmd)) {
197 if (pmd_table(old_pmd)) {
198 phys_addr_t table = pmd_page_paddr(old_pmd);
199 if (!WARN_ON_ONCE(slab_is_available()))
200 memblock_free(table, PAGE_SIZE);
204 alloc_init_pte(pmd, addr, next, __phys_to_pfn(phys),
205 prot, pgtable_alloc);
208 } while (pmd++, addr = next, addr != end);
213 static inline bool use_1G_block(unsigned long addr, unsigned long next,
216 if (PAGE_SHIFT != 12)
219 if (((addr | next | phys) & ~PUD_MASK) != 0)
225 static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
226 phys_addr_t phys, pgprot_t prot,
227 phys_addr_t (*pgtable_alloc)(void))
232 if (pgd_none(*pgd)) {
233 phys_addr_t pud_phys = pgtable_alloc();
234 __pgd_populate(pgd, pud_phys, PUD_TYPE_TABLE);
236 BUG_ON(pgd_bad(*pgd));
238 pud = pud_set_fixmap_offset(pgd, addr);
240 next = pud_addr_end(addr, end);
243 * For 4K granule only, attempt to put down a 1GB block
245 if (use_1G_block(addr, next, phys)) {
246 pud_t old_pud = *pud;
247 set_pud(pud, __pud(phys |
248 pgprot_val(mk_sect_prot(prot))));
251 * If we have an old value for a pud, it will
252 * be pointing to a pmd table that we no longer
253 * need (from swapper_pg_dir).
255 * Look up the old pmd table and free it.
257 if (!pud_none(old_pud)) {
259 if (pud_table(old_pud)) {
260 phys_addr_t table = pud_page_paddr(old_pud);
261 if (!WARN_ON_ONCE(slab_is_available()))
262 memblock_free(table, PAGE_SIZE);
266 alloc_init_pmd(pud, addr, next, phys, prot,
270 } while (pud++, addr = next, addr != end);
276 * Create the page directory entries and any necessary page tables for the
277 * mapping specified by 'md'.
279 static void init_pgd(pgd_t *pgd, phys_addr_t phys, unsigned long virt,
280 phys_addr_t size, pgprot_t prot,
281 phys_addr_t (*pgtable_alloc)(void))
283 unsigned long addr, length, end, next;
286 * If the virtual and physical address don't have the same offset
287 * within a page, we cannot map the region as the caller expects.
289 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
293 addr = virt & PAGE_MASK;
294 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
298 next = pgd_addr_end(addr, end);
299 alloc_init_pud(pgd, addr, next, phys, prot, pgtable_alloc);
301 } while (pgd++, addr = next, addr != end);
304 static phys_addr_t late_pgtable_alloc(void)
306 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
309 /* Ensure the zeroed page is visible to the page table walker */
314 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
315 unsigned long virt, phys_addr_t size,
317 phys_addr_t (*alloc)(void))
319 init_pgd(pgd_offset_raw(pgdir, virt), phys, virt, size, prot, alloc);
322 static void __init create_mapping(phys_addr_t phys, unsigned long virt,
323 phys_addr_t size, pgprot_t prot)
325 if (virt < VMALLOC_START) {
326 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
330 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
331 early_pgtable_alloc);
334 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
335 unsigned long virt, phys_addr_t size,
338 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
342 static void create_mapping_late(phys_addr_t phys, unsigned long virt,
343 phys_addr_t size, pgprot_t prot)
345 if (virt < VMALLOC_START) {
346 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
351 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot,
355 static void __init __map_memblock(pgd_t *pgd, phys_addr_t start, phys_addr_t end)
358 unsigned long kernel_start = __pa(_stext);
359 unsigned long kernel_end = __pa(_etext);
362 * Take care not to create a writable alias for the
363 * read-only text and rodata sections of the kernel image.
366 /* No overlap with the kernel text */
367 if (end < kernel_start || start >= kernel_end) {
368 __create_pgd_mapping(pgd, start, __phys_to_virt(start),
369 end - start, PAGE_KERNEL,
370 early_pgtable_alloc);
375 * This block overlaps the kernel text mapping. Map the portion(s) which
378 if (start < kernel_start)
379 __create_pgd_mapping(pgd, start,
380 __phys_to_virt(start),
381 kernel_start - start, PAGE_KERNEL,
382 early_pgtable_alloc);
383 if (kernel_end < end)
384 __create_pgd_mapping(pgd, kernel_end,
385 __phys_to_virt(kernel_end),
386 end - kernel_end, PAGE_KERNEL,
387 early_pgtable_alloc);
390 static void __init map_mem(pgd_t *pgd)
392 struct memblock_region *reg;
394 /* map all the memory banks */
395 for_each_memblock(memory, reg) {
396 phys_addr_t start = reg->base;
397 phys_addr_t end = start + reg->size;
401 if (memblock_is_nomap(reg))
404 __map_memblock(pgd, start, end);
408 void mark_rodata_ro(void)
410 if (!IS_ENABLED(CONFIG_DEBUG_RODATA))
413 create_mapping_late(__pa(_stext), (unsigned long)_stext,
414 (unsigned long)_etext - (unsigned long)_stext,
418 void fixup_init(void)
421 * Unmap the __init region but leave the VM area in place. This
422 * prevents the region from being reused for kernel modules, which
423 * is not supported by kallsyms.
425 unmap_kernel_range((u64)__init_begin, (u64)(__init_end - __init_begin));
428 static void __init map_kernel_chunk(pgd_t *pgd, void *va_start, void *va_end,
429 pgprot_t prot, struct vm_struct *vma)
431 phys_addr_t pa_start = __pa(va_start);
432 unsigned long size = va_end - va_start;
434 BUG_ON(!PAGE_ALIGNED(pa_start));
435 BUG_ON(!PAGE_ALIGNED(size));
437 __create_pgd_mapping(pgd, pa_start, (unsigned long)va_start, size, prot,
438 early_pgtable_alloc);
440 vma->addr = va_start;
441 vma->phys_addr = pa_start;
444 vma->caller = map_kernel_chunk;
446 vm_area_add_early(vma);
450 * Create fine-grained mappings for the kernel.
452 static void __init map_kernel(pgd_t *pgd)
454 static struct vm_struct vmlinux_text, vmlinux_init, vmlinux_data;
456 map_kernel_chunk(pgd, _stext, _etext, PAGE_KERNEL_EXEC, &vmlinux_text);
457 map_kernel_chunk(pgd, __init_begin, __init_end, PAGE_KERNEL_EXEC,
459 map_kernel_chunk(pgd, _data, _end, PAGE_KERNEL, &vmlinux_data);
461 if (!pgd_val(*pgd_offset_raw(pgd, FIXADDR_START))) {
463 * The fixmap falls in a separate pgd to the kernel, and doesn't
464 * live in the carveout for the swapper_pg_dir. We can simply
465 * re-use the existing dir for the fixmap.
467 set_pgd(pgd_offset_raw(pgd, FIXADDR_START),
468 *pgd_offset_k(FIXADDR_START));
469 } else if (CONFIG_PGTABLE_LEVELS > 3) {
471 * The fixmap shares its top level pgd entry with the kernel
472 * mapping. This can really only occur when we are running
473 * with 16k/4 levels, so we can simply reuse the pud level
476 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
477 set_pud(pud_set_fixmap_offset(pgd, FIXADDR_START),
478 __pud(__pa(bm_pmd) | PUD_TYPE_TABLE));
484 kasan_copy_shadow(pgd);
488 * paging_init() sets up the page tables, initialises the zone memory
489 * maps and sets up the zero page.
491 void __init paging_init(void)
493 phys_addr_t pgd_phys = early_pgtable_alloc();
494 pgd_t *pgd = pgd_set_fixmap(pgd_phys);
500 * We want to reuse the original swapper_pg_dir so we don't have to
501 * communicate the new address to non-coherent secondaries in
502 * secondary_entry, and so cpu_switch_mm can generate the address with
503 * adrp+add rather than a load from some global variable.
505 * To do this we need to go via a temporary pgd.
507 cpu_replace_ttbr1(__va(pgd_phys));
508 memcpy(swapper_pg_dir, pgd, PAGE_SIZE);
509 cpu_replace_ttbr1(swapper_pg_dir);
512 memblock_free(pgd_phys, PAGE_SIZE);
515 * We only reuse the PGD from the swapper_pg_dir, not the pud + pmd
518 memblock_free(__pa(swapper_pg_dir) + PAGE_SIZE,
519 SWAPPER_DIR_SIZE - PAGE_SIZE);
525 * Check whether a kernel address is valid (derived from arch/x86/).
527 int kern_addr_valid(unsigned long addr)
534 if ((((long)addr) >> VA_BITS) != -1UL)
537 pgd = pgd_offset_k(addr);
541 pud = pud_offset(pgd, addr);
546 return pfn_valid(pud_pfn(*pud));
548 pmd = pmd_offset(pud, addr);
553 return pfn_valid(pmd_pfn(*pmd));
555 pte = pte_offset_kernel(pmd, addr);
559 return pfn_valid(pte_pfn(*pte));
561 #ifdef CONFIG_SPARSEMEM_VMEMMAP
562 #if !ARM64_SWAPPER_USES_SECTION_MAPS
563 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
565 return vmemmap_populate_basepages(start, end, node);
567 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
568 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node)
570 unsigned long addr = start;
577 next = pmd_addr_end(addr, end);
579 pgd = vmemmap_pgd_populate(addr, node);
583 pud = vmemmap_pud_populate(pgd, addr, node);
587 pmd = pmd_offset(pud, addr);
588 if (pmd_none(*pmd)) {
591 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
595 set_pmd(pmd, __pmd(__pa(p) | PROT_SECT_NORMAL));
597 vmemmap_verify((pte_t *)pmd, node, addr, next);
598 } while (addr = next, addr != end);
602 #endif /* CONFIG_ARM64_64K_PAGES */
603 void vmemmap_free(unsigned long start, unsigned long end)
606 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
608 static inline pud_t * fixmap_pud(unsigned long addr)
610 pgd_t *pgd = pgd_offset_k(addr);
612 BUG_ON(pgd_none(*pgd) || pgd_bad(*pgd));
614 return pud_offset_kimg(pgd, addr);
617 static inline pmd_t * fixmap_pmd(unsigned long addr)
619 pud_t *pud = fixmap_pud(addr);
621 BUG_ON(pud_none(*pud) || pud_bad(*pud));
623 return pmd_offset_kimg(pud, addr);
626 static inline pte_t * fixmap_pte(unsigned long addr)
628 return &bm_pte[pte_index(addr)];
631 void __init early_fixmap_init(void)
636 unsigned long addr = FIXADDR_START;
638 pgd = pgd_offset_k(addr);
639 if (CONFIG_PGTABLE_LEVELS > 3 && !pgd_none(*pgd)) {
641 * We only end up here if the kernel mapping and the fixmap
642 * share the top level pgd entry, which should only happen on
643 * 16k/4 levels configurations.
645 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
646 pud = pud_offset_kimg(pgd, addr);
648 pgd_populate(&init_mm, pgd, bm_pud);
649 pud = fixmap_pud(addr);
651 pud_populate(&init_mm, pud, bm_pmd);
652 pmd = fixmap_pmd(addr);
653 pmd_populate_kernel(&init_mm, pmd, bm_pte);
656 * The boot-ioremap range spans multiple pmds, for which
657 * we are not prepared:
659 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
660 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
662 if ((pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
663 || pmd != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
665 pr_warn("pmd %p != %p, %p\n",
666 pmd, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
667 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
668 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
669 fix_to_virt(FIX_BTMAP_BEGIN));
670 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
671 fix_to_virt(FIX_BTMAP_END));
673 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
674 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
678 void __set_fixmap(enum fixed_addresses idx,
679 phys_addr_t phys, pgprot_t flags)
681 unsigned long addr = __fix_to_virt(idx);
684 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
686 pte = fixmap_pte(addr);
688 if (pgprot_val(flags)) {
689 set_pte(pte, pfn_pte(phys >> PAGE_SHIFT, flags));
691 pte_clear(&init_mm, addr, pte);
692 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
696 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
698 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
699 pgprot_t prot = PAGE_KERNEL_RO;
704 * Check whether the physical FDT address is set and meets the minimum
705 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
706 * at least 8 bytes so that we can always access the size field of the
707 * FDT header after mapping the first chunk, double check here if that
708 * is indeed the case.
710 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
711 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
715 * Make sure that the FDT region can be mapped without the need to
716 * allocate additional translation table pages, so that it is safe
717 * to call create_mapping() this early.
719 * On 64k pages, the FDT will be mapped using PTEs, so we need to
720 * be in the same PMD as the rest of the fixmap.
721 * On 4k pages, we'll use section mappings for the FDT so we only
722 * have to be in the same PUD.
724 BUILD_BUG_ON(dt_virt_base % SZ_2M);
726 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
727 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
729 offset = dt_phys % SWAPPER_BLOCK_SIZE;
730 dt_virt = (void *)dt_virt_base + offset;
732 /* map the first chunk so we can read the size from the header */
733 create_mapping(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
734 SWAPPER_BLOCK_SIZE, prot);
736 if (fdt_check_header(dt_virt) != 0)
739 size = fdt_totalsize(dt_virt);
740 if (size > MAX_FDT_SIZE)
743 if (offset + size > SWAPPER_BLOCK_SIZE)
744 create_mapping(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
745 round_up(offset + size, SWAPPER_BLOCK_SIZE), prot);
747 memblock_reserve(dt_phys, size);
752 int __init arch_ioremap_pud_supported(void)
754 /* only 4k granule supports level 1 block mappings */
755 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
758 int __init arch_ioremap_pmd_supported(void)
763 int pud_set_huge(pud_t *pud, phys_addr_t phys, pgprot_t prot)
765 BUG_ON(phys & ~PUD_MASK);
766 set_pud(pud, __pud(phys | PUD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
770 int pmd_set_huge(pmd_t *pmd, phys_addr_t phys, pgprot_t prot)
772 BUG_ON(phys & ~PMD_MASK);
773 set_pmd(pmd, __pmd(phys | PMD_TYPE_SECT | pgprot_val(mk_sect_prot(prot))));
777 int pud_clear_huge(pud_t *pud)
785 int pmd_clear_huge(pmd_t *pmd)