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/ioport.h>
26 #include <linux/kexec.h>
27 #include <linux/libfdt.h>
28 #include <linux/mman.h>
29 #include <linux/nodemask.h>
30 #include <linux/memblock.h>
34 #include <linux/vmalloc.h>
36 #include <asm/barrier.h>
37 #include <asm/cputype.h>
38 #include <asm/fixmap.h>
39 #include <asm/kasan.h>
40 #include <asm/kernel-pgtable.h>
41 #include <asm/sections.h>
42 #include <asm/setup.h>
43 #include <asm/sizes.h>
45 #include <asm/memblock.h>
46 #include <asm/mmu_context.h>
47 #include <asm/ptdump.h>
48 #include <asm/tlbflush.h>
50 #define NO_BLOCK_MAPPINGS BIT(0)
51 #define NO_CONT_MAPPINGS BIT(1)
53 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
54 u64 idmap_ptrs_per_pgd = PTRS_PER_PGD;
56 u64 kimage_voffset __ro_after_init;
57 EXPORT_SYMBOL(kimage_voffset);
60 * Empty_zero_page is a special page that is used for zero-initialized data
63 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
64 EXPORT_SYMBOL(empty_zero_page);
66 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
67 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
68 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
70 static DEFINE_SPINLOCK(swapper_pgdir_lock);
72 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
76 spin_lock(&swapper_pgdir_lock);
77 fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
78 WRITE_ONCE(*fixmap_pgdp, pgd);
80 * We need dsb(ishst) here to ensure the page-table-walker sees
81 * our new entry before set_p?d() returns. The fixmap's
82 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
85 spin_unlock(&swapper_pgdir_lock);
88 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
89 unsigned long size, pgprot_t vma_prot)
92 return pgprot_noncached(vma_prot);
93 else if (file->f_flags & O_SYNC)
94 return pgprot_writecombine(vma_prot);
97 EXPORT_SYMBOL(phys_mem_access_prot);
99 static phys_addr_t __init early_pgtable_alloc(void)
104 phys = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
107 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
108 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
109 * any level of table.
111 ptr = pte_set_fixmap(phys);
113 memset(ptr, 0, PAGE_SIZE);
116 * Implicit barriers also ensure the zeroed page is visible to the page
124 static bool pgattr_change_is_safe(u64 old, u64 new)
127 * The following mapping attributes may be updated in live
128 * kernel mappings without the need for break-before-make.
130 static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
132 /* creating or taking down mappings is always safe */
133 if (old == 0 || new == 0)
136 /* live contiguous mappings may not be manipulated at all */
137 if ((old | new) & PTE_CONT)
140 /* Transitioning from Non-Global to Global is unsafe */
141 if (old & ~new & PTE_NG)
144 return ((old ^ new) & ~mask) == 0;
147 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
148 phys_addr_t phys, pgprot_t prot)
152 ptep = pte_set_fixmap_offset(pmdp, addr);
154 pte_t old_pte = READ_ONCE(*ptep);
156 set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
159 * After the PTE entry has been populated once, we
160 * only allow updates to the permission attributes.
162 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
163 READ_ONCE(pte_val(*ptep))));
166 } while (ptep++, addr += PAGE_SIZE, addr != end);
171 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
172 unsigned long end, phys_addr_t phys,
174 phys_addr_t (*pgtable_alloc)(void),
178 pmd_t pmd = READ_ONCE(*pmdp);
180 BUG_ON(pmd_sect(pmd));
182 phys_addr_t pte_phys;
183 BUG_ON(!pgtable_alloc);
184 pte_phys = pgtable_alloc();
185 __pmd_populate(pmdp, pte_phys, PMD_TYPE_TABLE);
186 pmd = READ_ONCE(*pmdp);
188 BUG_ON(pmd_bad(pmd));
191 pgprot_t __prot = prot;
193 next = pte_cont_addr_end(addr, end);
195 /* use a contiguous mapping if the range is suitably aligned */
196 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
197 (flags & NO_CONT_MAPPINGS) == 0)
198 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
200 init_pte(pmdp, addr, next, phys, __prot);
203 } while (addr = next, addr != end);
206 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
207 phys_addr_t phys, pgprot_t prot,
208 phys_addr_t (*pgtable_alloc)(void), int flags)
213 pmdp = pmd_set_fixmap_offset(pudp, addr);
215 pmd_t old_pmd = READ_ONCE(*pmdp);
217 next = pmd_addr_end(addr, end);
219 /* try section mapping first */
220 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
221 (flags & NO_BLOCK_MAPPINGS) == 0) {
222 pmd_set_huge(pmdp, phys, prot);
225 * After the PMD entry has been populated once, we
226 * only allow updates to the permission attributes.
228 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
229 READ_ONCE(pmd_val(*pmdp))));
231 alloc_init_cont_pte(pmdp, addr, next, phys, prot,
232 pgtable_alloc, flags);
234 BUG_ON(pmd_val(old_pmd) != 0 &&
235 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
238 } while (pmdp++, addr = next, addr != end);
243 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
244 unsigned long end, phys_addr_t phys,
246 phys_addr_t (*pgtable_alloc)(void), int flags)
249 pud_t pud = READ_ONCE(*pudp);
252 * Check for initial section mappings in the pgd/pud.
254 BUG_ON(pud_sect(pud));
256 phys_addr_t pmd_phys;
257 BUG_ON(!pgtable_alloc);
258 pmd_phys = pgtable_alloc();
259 __pud_populate(pudp, pmd_phys, PUD_TYPE_TABLE);
260 pud = READ_ONCE(*pudp);
262 BUG_ON(pud_bad(pud));
265 pgprot_t __prot = prot;
267 next = pmd_cont_addr_end(addr, end);
269 /* use a contiguous mapping if the range is suitably aligned */
270 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
271 (flags & NO_CONT_MAPPINGS) == 0)
272 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
274 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
277 } while (addr = next, addr != end);
280 static inline bool use_1G_block(unsigned long addr, unsigned long next,
283 if (PAGE_SHIFT != 12)
286 if (((addr | next | phys) & ~PUD_MASK) != 0)
292 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
293 phys_addr_t phys, pgprot_t prot,
294 phys_addr_t (*pgtable_alloc)(void),
299 pgd_t pgd = READ_ONCE(*pgdp);
302 phys_addr_t pud_phys;
303 BUG_ON(!pgtable_alloc);
304 pud_phys = pgtable_alloc();
305 __pgd_populate(pgdp, pud_phys, PUD_TYPE_TABLE);
306 pgd = READ_ONCE(*pgdp);
308 BUG_ON(pgd_bad(pgd));
310 pudp = pud_set_fixmap_offset(pgdp, addr);
312 pud_t old_pud = READ_ONCE(*pudp);
314 next = pud_addr_end(addr, end);
317 * For 4K granule only, attempt to put down a 1GB block
319 if (use_1G_block(addr, next, phys) &&
320 (flags & NO_BLOCK_MAPPINGS) == 0) {
321 pud_set_huge(pudp, phys, prot);
324 * After the PUD entry has been populated once, we
325 * only allow updates to the permission attributes.
327 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
328 READ_ONCE(pud_val(*pudp))));
330 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
331 pgtable_alloc, flags);
333 BUG_ON(pud_val(old_pud) != 0 &&
334 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
337 } while (pudp++, addr = next, addr != end);
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 (*pgtable_alloc)(void),
348 unsigned long addr, length, end, next;
349 pgd_t *pgdp = pgd_offset_raw(pgdir, virt);
352 * If the virtual and physical address don't have the same offset
353 * within a page, we cannot map the region as the caller expects.
355 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
359 addr = virt & PAGE_MASK;
360 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
364 next = pgd_addr_end(addr, end);
365 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
368 } while (pgdp++, addr = next, addr != end);
371 static phys_addr_t pgd_pgtable_alloc(void)
373 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
374 if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
377 /* Ensure the zeroed page is visible to the page table walker */
383 * This function can only be used to modify existing table entries,
384 * without allocating new levels of table. Note that this permits the
385 * creation of new section or page entries.
387 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
388 phys_addr_t size, pgprot_t prot)
390 if (virt < VMALLOC_START) {
391 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
395 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
399 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
400 unsigned long virt, phys_addr_t size,
401 pgprot_t prot, bool page_mappings_only)
405 BUG_ON(mm == &init_mm);
407 if (page_mappings_only)
408 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
410 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
411 pgd_pgtable_alloc, flags);
414 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
415 phys_addr_t size, pgprot_t prot)
417 if (virt < VMALLOC_START) {
418 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
423 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
426 /* flush the TLBs after updating live kernel mappings */
427 flush_tlb_kernel_range(virt, virt + size);
430 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
431 phys_addr_t end, pgprot_t prot, int flags)
433 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
434 prot, early_pgtable_alloc, flags);
437 void __init mark_linear_text_alias_ro(void)
440 * Remove the write permissions from the linear alias of .text/.rodata
442 update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
443 (unsigned long)__init_begin - (unsigned long)_text,
447 static void __init map_mem(pgd_t *pgdp)
449 phys_addr_t kernel_start = __pa_symbol(_text);
450 phys_addr_t kernel_end = __pa_symbol(__init_begin);
451 struct memblock_region *reg;
454 if (debug_pagealloc_enabled())
455 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
458 * Take care not to create a writable alias for the
459 * read-only text and rodata sections of the kernel image.
460 * So temporarily mark them as NOMAP to skip mappings in
461 * the following for-loop
463 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
464 #ifdef CONFIG_KEXEC_CORE
466 memblock_mark_nomap(crashk_res.start,
467 resource_size(&crashk_res));
470 /* map all the memory banks */
471 for_each_memblock(memory, reg) {
472 phys_addr_t start = reg->base;
473 phys_addr_t end = start + reg->size;
477 if (memblock_is_nomap(reg))
480 __map_memblock(pgdp, start, end, PAGE_KERNEL, flags);
484 * Map the linear alias of the [_text, __init_begin) interval
485 * as non-executable now, and remove the write permission in
486 * mark_linear_text_alias_ro() below (which will be called after
487 * alternative patching has completed). This makes the contents
488 * of the region accessible to subsystems such as hibernate,
489 * but protects it from inadvertent modification or execution.
490 * Note that contiguous mappings cannot be remapped in this way,
491 * so we should avoid them here.
493 __map_memblock(pgdp, kernel_start, kernel_end,
494 PAGE_KERNEL, NO_CONT_MAPPINGS);
495 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
497 #ifdef CONFIG_KEXEC_CORE
499 * Use page-level mappings here so that we can shrink the region
500 * in page granularity and put back unused memory to buddy system
501 * through /sys/kernel/kexec_crash_size interface.
503 if (crashk_res.end) {
504 __map_memblock(pgdp, crashk_res.start, crashk_res.end + 1,
506 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
507 memblock_clear_nomap(crashk_res.start,
508 resource_size(&crashk_res));
513 void mark_rodata_ro(void)
515 unsigned long section_size;
518 * mark .rodata as read only. Use __init_begin rather than __end_rodata
519 * to cover NOTES and EXCEPTION_TABLE.
521 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
522 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
523 section_size, PAGE_KERNEL_RO);
528 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
529 pgprot_t prot, struct vm_struct *vma,
530 int flags, unsigned long vm_flags)
532 phys_addr_t pa_start = __pa_symbol(va_start);
533 unsigned long size = va_end - va_start;
535 BUG_ON(!PAGE_ALIGNED(pa_start));
536 BUG_ON(!PAGE_ALIGNED(size));
538 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
539 early_pgtable_alloc, flags);
541 if (!(vm_flags & VM_NO_GUARD))
544 vma->addr = va_start;
545 vma->phys_addr = pa_start;
547 vma->flags = VM_MAP | vm_flags;
548 vma->caller = __builtin_return_address(0);
550 vm_area_add_early(vma);
553 static int __init parse_rodata(char *arg)
555 return strtobool(arg, &rodata_enabled);
557 early_param("rodata", parse_rodata);
559 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
560 static int __init map_entry_trampoline(void)
562 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
563 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
565 /* The trampoline is always mapped and can therefore be global */
566 pgprot_val(prot) &= ~PTE_NG;
568 /* Map only the text into the trampoline page table */
569 memset(tramp_pg_dir, 0, PGD_SIZE);
570 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE,
571 prot, pgd_pgtable_alloc, 0);
573 /* Map both the text and data into the kernel page table */
574 __set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot);
575 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
576 extern char __entry_tramp_data_start[];
578 __set_fixmap(FIX_ENTRY_TRAMP_DATA,
579 __pa_symbol(__entry_tramp_data_start),
585 core_initcall(map_entry_trampoline);
589 * Create fine-grained mappings for the kernel.
591 static void __init map_kernel(pgd_t *pgdp)
593 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
594 vmlinux_initdata, vmlinux_data;
597 * External debuggers may need to write directly to the text
598 * mapping to install SW breakpoints. Allow this (only) when
599 * explicitly requested with rodata=off.
601 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
604 * Only rodata will be remapped with different permissions later on,
605 * all other segments are allowed to use contiguous mappings.
607 map_kernel_segment(pgdp, _text, _etext, text_prot, &vmlinux_text, 0,
609 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
610 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
611 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
612 &vmlinux_inittext, 0, VM_NO_GUARD);
613 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
614 &vmlinux_initdata, 0, VM_NO_GUARD);
615 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
617 if (!READ_ONCE(pgd_val(*pgd_offset_raw(pgdp, FIXADDR_START)))) {
619 * The fixmap falls in a separate pgd to the kernel, and doesn't
620 * live in the carveout for the swapper_pg_dir. We can simply
621 * re-use the existing dir for the fixmap.
623 set_pgd(pgd_offset_raw(pgdp, FIXADDR_START),
624 READ_ONCE(*pgd_offset_k(FIXADDR_START)));
625 } else if (CONFIG_PGTABLE_LEVELS > 3) {
627 * The fixmap shares its top level pgd entry with the kernel
628 * mapping. This can really only occur when we are running
629 * with 16k/4 levels, so we can simply reuse the pud level
632 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
633 pud_populate(&init_mm,
634 pud_set_fixmap_offset(pgdp, FIXADDR_START),
641 kasan_copy_shadow(pgdp);
645 * paging_init() sets up the page tables, initialises the zone memory
646 * maps and sets up the zero page.
648 void __init paging_init(void)
650 pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
657 cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
658 init_mm.pgd = swapper_pg_dir;
660 memblock_free(__pa_symbol(init_pg_dir),
661 __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
665 * Check whether a kernel address is valid (derived from arch/x86/).
667 int kern_addr_valid(unsigned long addr)
674 if ((((long)addr) >> VA_BITS) != -1UL)
677 pgdp = pgd_offset_k(addr);
678 if (pgd_none(READ_ONCE(*pgdp)))
681 pudp = pud_offset(pgdp, addr);
682 pud = READ_ONCE(*pudp);
687 return pfn_valid(pud_pfn(pud));
689 pmdp = pmd_offset(pudp, addr);
690 pmd = READ_ONCE(*pmdp);
695 return pfn_valid(pmd_pfn(pmd));
697 ptep = pte_offset_kernel(pmdp, addr);
698 pte = READ_ONCE(*ptep);
702 return pfn_valid(pte_pfn(pte));
704 #ifdef CONFIG_SPARSEMEM_VMEMMAP
705 #if !ARM64_SWAPPER_USES_SECTION_MAPS
706 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
707 struct vmem_altmap *altmap)
709 return vmemmap_populate_basepages(start, end, node);
711 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
712 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
713 struct vmem_altmap *altmap)
715 unsigned long addr = start;
722 next = pmd_addr_end(addr, end);
724 pgdp = vmemmap_pgd_populate(addr, node);
728 pudp = vmemmap_pud_populate(pgdp, addr, node);
732 pmdp = pmd_offset(pudp, addr);
733 if (pmd_none(READ_ONCE(*pmdp))) {
736 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
740 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
742 vmemmap_verify((pte_t *)pmdp, node, addr, next);
743 } while (addr = next, addr != end);
747 #endif /* CONFIG_ARM64_64K_PAGES */
748 void vmemmap_free(unsigned long start, unsigned long end,
749 struct vmem_altmap *altmap)
752 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
754 static inline pud_t * fixmap_pud(unsigned long addr)
756 pgd_t *pgdp = pgd_offset_k(addr);
757 pgd_t pgd = READ_ONCE(*pgdp);
759 BUG_ON(pgd_none(pgd) || pgd_bad(pgd));
761 return pud_offset_kimg(pgdp, addr);
764 static inline pmd_t * fixmap_pmd(unsigned long addr)
766 pud_t *pudp = fixmap_pud(addr);
767 pud_t pud = READ_ONCE(*pudp);
769 BUG_ON(pud_none(pud) || pud_bad(pud));
771 return pmd_offset_kimg(pudp, addr);
774 static inline pte_t * fixmap_pte(unsigned long addr)
776 return &bm_pte[pte_index(addr)];
780 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
781 * directly on kernel symbols (bm_p*d). This function is called too early to use
782 * lm_alias so __p*d_populate functions must be used to populate with the
783 * physical address from __pa_symbol.
785 void __init early_fixmap_init(void)
790 unsigned long addr = FIXADDR_START;
792 pgdp = pgd_offset_k(addr);
793 pgd = READ_ONCE(*pgdp);
794 if (CONFIG_PGTABLE_LEVELS > 3 &&
795 !(pgd_none(pgd) || pgd_page_paddr(pgd) == __pa_symbol(bm_pud))) {
797 * We only end up here if the kernel mapping and the fixmap
798 * share the top level pgd entry, which should only happen on
799 * 16k/4 levels configurations.
801 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
802 pudp = pud_offset_kimg(pgdp, addr);
805 __pgd_populate(pgdp, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
806 pudp = fixmap_pud(addr);
808 if (pud_none(READ_ONCE(*pudp)))
809 __pud_populate(pudp, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
810 pmdp = fixmap_pmd(addr);
811 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
814 * The boot-ioremap range spans multiple pmds, for which
815 * we are not prepared:
817 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
818 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
820 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
821 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
823 pr_warn("pmdp %p != %p, %p\n",
824 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
825 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
826 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
827 fix_to_virt(FIX_BTMAP_BEGIN));
828 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
829 fix_to_virt(FIX_BTMAP_END));
831 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
832 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
837 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
838 * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
840 void __set_fixmap(enum fixed_addresses idx,
841 phys_addr_t phys, pgprot_t flags)
843 unsigned long addr = __fix_to_virt(idx);
846 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
848 ptep = fixmap_pte(addr);
850 if (pgprot_val(flags)) {
851 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
853 pte_clear(&init_mm, addr, ptep);
854 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
858 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
860 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
865 * Check whether the physical FDT address is set and meets the minimum
866 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
867 * at least 8 bytes so that we can always access the magic and size
868 * fields of the FDT header after mapping the first chunk, double check
869 * here if that is indeed the case.
871 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
872 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
876 * Make sure that the FDT region can be mapped without the need to
877 * allocate additional translation table pages, so that it is safe
878 * to call create_mapping_noalloc() this early.
880 * On 64k pages, the FDT will be mapped using PTEs, so we need to
881 * be in the same PMD as the rest of the fixmap.
882 * On 4k pages, we'll use section mappings for the FDT so we only
883 * have to be in the same PUD.
885 BUILD_BUG_ON(dt_virt_base % SZ_2M);
887 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
888 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
890 offset = dt_phys % SWAPPER_BLOCK_SIZE;
891 dt_virt = (void *)dt_virt_base + offset;
893 /* map the first chunk so we can read the size from the header */
894 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
895 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
897 if (fdt_magic(dt_virt) != FDT_MAGIC)
900 *size = fdt_totalsize(dt_virt);
901 if (*size > MAX_FDT_SIZE)
904 if (offset + *size > SWAPPER_BLOCK_SIZE)
905 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
906 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
911 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
916 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
920 memblock_reserve(dt_phys, size);
924 int __init arch_ioremap_pud_supported(void)
926 /* only 4k granule supports level 1 block mappings */
927 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
930 int __init arch_ioremap_pmd_supported(void)
935 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
937 pgprot_t sect_prot = __pgprot(PUD_TYPE_SECT |
938 pgprot_val(mk_sect_prot(prot)));
939 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), sect_prot);
941 /* Only allow permission changes for now */
942 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
946 BUG_ON(phys & ~PUD_MASK);
947 set_pud(pudp, new_pud);
951 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
953 pgprot_t sect_prot = __pgprot(PMD_TYPE_SECT |
954 pgprot_val(mk_sect_prot(prot)));
955 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), sect_prot);
957 /* Only allow permission changes for now */
958 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
962 BUG_ON(phys & ~PMD_MASK);
963 set_pmd(pmdp, new_pmd);
967 int pud_clear_huge(pud_t *pudp)
969 if (!pud_sect(READ_ONCE(*pudp)))
975 int pmd_clear_huge(pmd_t *pmdp)
977 if (!pmd_sect(READ_ONCE(*pmdp)))
983 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
988 pmd = READ_ONCE(*pmdp);
990 if (!pmd_present(pmd))
992 if (!pmd_table(pmd)) {
993 VM_WARN_ON(!pmd_table(pmd));
997 table = pte_offset_kernel(pmdp, addr);
999 __flush_tlb_kernel_pgtable(addr);
1000 pte_free_kernel(NULL, table);
1004 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1009 unsigned long next, end;
1011 pud = READ_ONCE(*pudp);
1013 if (!pud_present(pud))
1015 if (!pud_table(pud)) {
1016 VM_WARN_ON(!pud_table(pud));
1020 table = pmd_offset(pudp, addr);
1023 end = addr + PUD_SIZE;
1025 pmd_free_pte_page(pmdp, next);
1026 } while (pmdp++, next += PMD_SIZE, next != end);
1029 __flush_tlb_kernel_pgtable(addr);
1030 pmd_free(NULL, table);