1 // SPDX-License-Identifier: GPL-2.0-only
3 * Based on arch/arm/mm/mmu.c
5 * Copyright (C) 1995-2005 Russell King
6 * Copyright (C) 2012 ARM Ltd.
9 #include <linux/cache.h>
10 #include <linux/export.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/init.h>
14 #include <linux/ioport.h>
15 #include <linux/kexec.h>
16 #include <linux/libfdt.h>
17 #include <linux/mman.h>
18 #include <linux/nodemask.h>
19 #include <linux/memblock.h>
23 #include <linux/vmalloc.h>
25 #include <asm/barrier.h>
26 #include <asm/cputype.h>
27 #include <asm/fixmap.h>
28 #include <asm/kasan.h>
29 #include <asm/kernel-pgtable.h>
30 #include <asm/sections.h>
31 #include <asm/setup.h>
32 #include <linux/sizes.h>
34 #include <asm/mmu_context.h>
35 #include <asm/ptdump.h>
36 #include <asm/tlbflush.h>
38 #define NO_BLOCK_MAPPINGS BIT(0)
39 #define NO_CONT_MAPPINGS BIT(1)
41 u64 idmap_t0sz = TCR_T0SZ(VA_BITS_MIN);
42 u64 idmap_ptrs_per_pgd = PTRS_PER_PGD;
44 u64 __section(".mmuoff.data.write") vabits_actual;
45 EXPORT_SYMBOL(vabits_actual);
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 static DEFINE_SPINLOCK(swapper_pgdir_lock);
63 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
67 spin_lock(&swapper_pgdir_lock);
68 fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
69 WRITE_ONCE(*fixmap_pgdp, pgd);
71 * We need dsb(ishst) here to ensure the page-table-walker sees
72 * our new entry before set_p?d() returns. The fixmap's
73 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
76 spin_unlock(&swapper_pgdir_lock);
79 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
80 unsigned long size, pgprot_t vma_prot)
83 return pgprot_noncached(vma_prot);
84 else if (file->f_flags & O_SYNC)
85 return pgprot_writecombine(vma_prot);
88 EXPORT_SYMBOL(phys_mem_access_prot);
90 static phys_addr_t __init early_pgtable_alloc(int shift)
95 phys = memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
97 panic("Failed to allocate page table page\n");
100 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
101 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
102 * any level of table.
104 ptr = pte_set_fixmap(phys);
106 memset(ptr, 0, PAGE_SIZE);
109 * Implicit barriers also ensure the zeroed page is visible to the page
117 static bool pgattr_change_is_safe(u64 old, u64 new)
120 * The following mapping attributes may be updated in live
121 * kernel mappings without the need for break-before-make.
123 static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
125 /* creating or taking down mappings is always safe */
126 if (old == 0 || new == 0)
129 /* live contiguous mappings may not be manipulated at all */
130 if ((old | new) & PTE_CONT)
133 /* Transitioning from Non-Global to Global is unsafe */
134 if (old & ~new & PTE_NG)
137 return ((old ^ new) & ~mask) == 0;
140 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
141 phys_addr_t phys, pgprot_t prot)
145 ptep = pte_set_fixmap_offset(pmdp, addr);
147 pte_t old_pte = READ_ONCE(*ptep);
149 set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
152 * After the PTE entry has been populated once, we
153 * only allow updates to the permission attributes.
155 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
156 READ_ONCE(pte_val(*ptep))));
159 } while (ptep++, addr += PAGE_SIZE, addr != end);
164 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
165 unsigned long end, phys_addr_t phys,
167 phys_addr_t (*pgtable_alloc)(int),
171 pmd_t pmd = READ_ONCE(*pmdp);
173 BUG_ON(pmd_sect(pmd));
175 phys_addr_t pte_phys;
176 BUG_ON(!pgtable_alloc);
177 pte_phys = pgtable_alloc(PAGE_SHIFT);
178 __pmd_populate(pmdp, pte_phys, PMD_TYPE_TABLE);
179 pmd = READ_ONCE(*pmdp);
181 BUG_ON(pmd_bad(pmd));
184 pgprot_t __prot = prot;
186 next = pte_cont_addr_end(addr, end);
188 /* use a contiguous mapping if the range is suitably aligned */
189 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
190 (flags & NO_CONT_MAPPINGS) == 0)
191 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
193 init_pte(pmdp, addr, next, phys, __prot);
196 } while (addr = next, addr != end);
199 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
200 phys_addr_t phys, pgprot_t prot,
201 phys_addr_t (*pgtable_alloc)(int), int flags)
206 pmdp = pmd_set_fixmap_offset(pudp, addr);
208 pmd_t old_pmd = READ_ONCE(*pmdp);
210 next = pmd_addr_end(addr, end);
212 /* try section mapping first */
213 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
214 (flags & NO_BLOCK_MAPPINGS) == 0) {
215 pmd_set_huge(pmdp, phys, prot);
218 * After the PMD entry has been populated once, we
219 * only allow updates to the permission attributes.
221 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
222 READ_ONCE(pmd_val(*pmdp))));
224 alloc_init_cont_pte(pmdp, addr, next, phys, prot,
225 pgtable_alloc, flags);
227 BUG_ON(pmd_val(old_pmd) != 0 &&
228 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
231 } while (pmdp++, addr = next, addr != end);
236 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
237 unsigned long end, phys_addr_t phys,
239 phys_addr_t (*pgtable_alloc)(int), int flags)
242 pud_t pud = READ_ONCE(*pudp);
245 * Check for initial section mappings in the pgd/pud.
247 BUG_ON(pud_sect(pud));
249 phys_addr_t pmd_phys;
250 BUG_ON(!pgtable_alloc);
251 pmd_phys = pgtable_alloc(PMD_SHIFT);
252 __pud_populate(pudp, pmd_phys, PUD_TYPE_TABLE);
253 pud = READ_ONCE(*pudp);
255 BUG_ON(pud_bad(pud));
258 pgprot_t __prot = prot;
260 next = pmd_cont_addr_end(addr, end);
262 /* use a contiguous mapping if the range is suitably aligned */
263 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
264 (flags & NO_CONT_MAPPINGS) == 0)
265 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
267 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
270 } while (addr = next, addr != end);
273 static inline bool use_1G_block(unsigned long addr, unsigned long next,
276 if (PAGE_SHIFT != 12)
279 if (((addr | next | phys) & ~PUD_MASK) != 0)
285 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
286 phys_addr_t phys, pgprot_t prot,
287 phys_addr_t (*pgtable_alloc)(int),
292 pgd_t pgd = READ_ONCE(*pgdp);
295 phys_addr_t pud_phys;
296 BUG_ON(!pgtable_alloc);
297 pud_phys = pgtable_alloc(PUD_SHIFT);
298 __pgd_populate(pgdp, pud_phys, PUD_TYPE_TABLE);
299 pgd = READ_ONCE(*pgdp);
301 BUG_ON(pgd_bad(pgd));
303 pudp = pud_set_fixmap_offset(pgdp, addr);
305 pud_t old_pud = READ_ONCE(*pudp);
307 next = pud_addr_end(addr, end);
310 * For 4K granule only, attempt to put down a 1GB block
312 if (use_1G_block(addr, next, phys) &&
313 (flags & NO_BLOCK_MAPPINGS) == 0) {
314 pud_set_huge(pudp, phys, prot);
317 * After the PUD entry has been populated once, we
318 * only allow updates to the permission attributes.
320 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
321 READ_ONCE(pud_val(*pudp))));
323 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
324 pgtable_alloc, flags);
326 BUG_ON(pud_val(old_pud) != 0 &&
327 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
330 } while (pudp++, addr = next, addr != end);
335 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
336 unsigned long virt, phys_addr_t size,
338 phys_addr_t (*pgtable_alloc)(int),
341 unsigned long addr, length, end, next;
342 pgd_t *pgdp = pgd_offset_raw(pgdir, virt);
345 * If the virtual and physical address don't have the same offset
346 * within a page, we cannot map the region as the caller expects.
348 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
352 addr = virt & PAGE_MASK;
353 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
357 next = pgd_addr_end(addr, end);
358 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
361 } while (pgdp++, addr = next, addr != end);
364 static phys_addr_t __pgd_pgtable_alloc(int shift)
366 void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
369 /* Ensure the zeroed page is visible to the page table walker */
374 static phys_addr_t pgd_pgtable_alloc(int shift)
376 phys_addr_t pa = __pgd_pgtable_alloc(shift);
379 * Call proper page table ctor in case later we need to
380 * call core mm functions like apply_to_page_range() on
381 * this pre-allocated page table.
383 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
384 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
386 if (shift == PAGE_SHIFT)
387 BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa)));
388 else if (shift == PMD_SHIFT)
389 BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
395 * This function can only be used to modify existing table entries,
396 * without allocating new levels of table. Note that this permits the
397 * creation of new section or page entries.
399 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
400 phys_addr_t size, pgprot_t prot)
402 if (virt < PAGE_OFFSET) {
403 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
407 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
411 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
412 unsigned long virt, phys_addr_t size,
413 pgprot_t prot, bool page_mappings_only)
417 BUG_ON(mm == &init_mm);
419 if (page_mappings_only)
420 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
422 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
423 pgd_pgtable_alloc, flags);
426 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
427 phys_addr_t size, pgprot_t prot)
429 if (virt < PAGE_OFFSET) {
430 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
435 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
438 /* flush the TLBs after updating live kernel mappings */
439 flush_tlb_kernel_range(virt, virt + size);
442 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
443 phys_addr_t end, pgprot_t prot, int flags)
445 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
446 prot, early_pgtable_alloc, flags);
449 void __init mark_linear_text_alias_ro(void)
452 * Remove the write permissions from the linear alias of .text/.rodata
454 update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
455 (unsigned long)__init_begin - (unsigned long)_text,
459 static void __init map_mem(pgd_t *pgdp)
461 phys_addr_t kernel_start = __pa_symbol(_text);
462 phys_addr_t kernel_end = __pa_symbol(__init_begin);
463 struct memblock_region *reg;
466 if (rodata_full || debug_pagealloc_enabled())
467 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
470 * Take care not to create a writable alias for the
471 * read-only text and rodata sections of the kernel image.
472 * So temporarily mark them as NOMAP to skip mappings in
473 * the following for-loop
475 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
476 #ifdef CONFIG_KEXEC_CORE
478 memblock_mark_nomap(crashk_res.start,
479 resource_size(&crashk_res));
482 /* map all the memory banks */
483 for_each_memblock(memory, reg) {
484 phys_addr_t start = reg->base;
485 phys_addr_t end = start + reg->size;
489 if (memblock_is_nomap(reg))
492 __map_memblock(pgdp, start, end, PAGE_KERNEL, flags);
496 * Map the linear alias of the [_text, __init_begin) interval
497 * as non-executable now, and remove the write permission in
498 * mark_linear_text_alias_ro() below (which will be called after
499 * alternative patching has completed). This makes the contents
500 * of the region accessible to subsystems such as hibernate,
501 * but protects it from inadvertent modification or execution.
502 * Note that contiguous mappings cannot be remapped in this way,
503 * so we should avoid them here.
505 __map_memblock(pgdp, kernel_start, kernel_end,
506 PAGE_KERNEL, NO_CONT_MAPPINGS);
507 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
509 #ifdef CONFIG_KEXEC_CORE
511 * Use page-level mappings here so that we can shrink the region
512 * in page granularity and put back unused memory to buddy system
513 * through /sys/kernel/kexec_crash_size interface.
515 if (crashk_res.end) {
516 __map_memblock(pgdp, crashk_res.start, crashk_res.end + 1,
518 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
519 memblock_clear_nomap(crashk_res.start,
520 resource_size(&crashk_res));
525 void mark_rodata_ro(void)
527 unsigned long section_size;
530 * mark .rodata as read only. Use __init_begin rather than __end_rodata
531 * to cover NOTES and EXCEPTION_TABLE.
533 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
534 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
535 section_size, PAGE_KERNEL_RO);
540 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
541 pgprot_t prot, struct vm_struct *vma,
542 int flags, unsigned long vm_flags)
544 phys_addr_t pa_start = __pa_symbol(va_start);
545 unsigned long size = va_end - va_start;
547 BUG_ON(!PAGE_ALIGNED(pa_start));
548 BUG_ON(!PAGE_ALIGNED(size));
550 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
551 early_pgtable_alloc, flags);
553 if (!(vm_flags & VM_NO_GUARD))
556 vma->addr = va_start;
557 vma->phys_addr = pa_start;
559 vma->flags = VM_MAP | vm_flags;
560 vma->caller = __builtin_return_address(0);
562 vm_area_add_early(vma);
565 static int __init parse_rodata(char *arg)
567 int ret = strtobool(arg, &rodata_enabled);
573 /* permit 'full' in addition to boolean options */
574 if (strcmp(arg, "full"))
577 rodata_enabled = true;
581 early_param("rodata", parse_rodata);
583 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
584 static int __init map_entry_trampoline(void)
588 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
589 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
591 /* The trampoline is always mapped and can therefore be global */
592 pgprot_val(prot) &= ~PTE_NG;
594 /* Map only the text into the trampoline page table */
595 memset(tramp_pg_dir, 0, PGD_SIZE);
596 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
597 entry_tramp_text_size(), prot,
598 __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
600 /* Map both the text and data into the kernel page table */
601 for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
602 __set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
603 pa_start + i * PAGE_SIZE, prot);
605 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
606 extern char __entry_tramp_data_start[];
608 __set_fixmap(FIX_ENTRY_TRAMP_DATA,
609 __pa_symbol(__entry_tramp_data_start),
615 core_initcall(map_entry_trampoline);
619 * Create fine-grained mappings for the kernel.
621 static void __init map_kernel(pgd_t *pgdp)
623 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
624 vmlinux_initdata, vmlinux_data;
627 * External debuggers may need to write directly to the text
628 * mapping to install SW breakpoints. Allow this (only) when
629 * explicitly requested with rodata=off.
631 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
634 * Only rodata will be remapped with different permissions later on,
635 * all other segments are allowed to use contiguous mappings.
637 map_kernel_segment(pgdp, _text, _etext, text_prot, &vmlinux_text, 0,
639 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
640 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
641 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
642 &vmlinux_inittext, 0, VM_NO_GUARD);
643 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
644 &vmlinux_initdata, 0, VM_NO_GUARD);
645 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
647 if (!READ_ONCE(pgd_val(*pgd_offset_raw(pgdp, FIXADDR_START)))) {
649 * The fixmap falls in a separate pgd to the kernel, and doesn't
650 * live in the carveout for the swapper_pg_dir. We can simply
651 * re-use the existing dir for the fixmap.
653 set_pgd(pgd_offset_raw(pgdp, FIXADDR_START),
654 READ_ONCE(*pgd_offset_k(FIXADDR_START)));
655 } else if (CONFIG_PGTABLE_LEVELS > 3) {
659 * The fixmap shares its top level pgd entry with the kernel
660 * mapping. This can really only occur when we are running
661 * with 16k/4 levels, so we can simply reuse the pud level
664 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
665 bm_pgdp = pgd_offset_raw(pgdp, FIXADDR_START);
666 bm_pudp = pud_set_fixmap_offset(bm_pgdp, FIXADDR_START);
667 pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
673 kasan_copy_shadow(pgdp);
676 void __init paging_init(void)
678 pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
685 cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
686 init_mm.pgd = swapper_pg_dir;
688 memblock_free(__pa_symbol(init_pg_dir),
689 __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
691 memblock_allow_resize();
695 * Check whether a kernel address is valid (derived from arch/x86/).
697 int kern_addr_valid(unsigned long addr)
704 if ((((long)addr) >> VA_BITS) != -1UL)
707 pgdp = pgd_offset_k(addr);
708 if (pgd_none(READ_ONCE(*pgdp)))
711 pudp = pud_offset(pgdp, addr);
712 pud = READ_ONCE(*pudp);
717 return pfn_valid(pud_pfn(pud));
719 pmdp = pmd_offset(pudp, addr);
720 pmd = READ_ONCE(*pmdp);
725 return pfn_valid(pmd_pfn(pmd));
727 ptep = pte_offset_kernel(pmdp, addr);
728 pte = READ_ONCE(*ptep);
732 return pfn_valid(pte_pfn(pte));
734 #ifdef CONFIG_SPARSEMEM_VMEMMAP
735 #if !ARM64_SWAPPER_USES_SECTION_MAPS
736 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
737 struct vmem_altmap *altmap)
739 return vmemmap_populate_basepages(start, end, node);
741 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
742 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
743 struct vmem_altmap *altmap)
745 unsigned long addr = start;
752 next = pmd_addr_end(addr, end);
754 pgdp = vmemmap_pgd_populate(addr, node);
758 pudp = vmemmap_pud_populate(pgdp, addr, node);
762 pmdp = pmd_offset(pudp, addr);
763 if (pmd_none(READ_ONCE(*pmdp))) {
766 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
770 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
772 vmemmap_verify((pte_t *)pmdp, node, addr, next);
773 } while (addr = next, addr != end);
777 #endif /* !ARM64_SWAPPER_USES_SECTION_MAPS */
778 void vmemmap_free(unsigned long start, unsigned long end,
779 struct vmem_altmap *altmap)
782 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
784 static inline pud_t * fixmap_pud(unsigned long addr)
786 pgd_t *pgdp = pgd_offset_k(addr);
787 pgd_t pgd = READ_ONCE(*pgdp);
789 BUG_ON(pgd_none(pgd) || pgd_bad(pgd));
791 return pud_offset_kimg(pgdp, addr);
794 static inline pmd_t * fixmap_pmd(unsigned long addr)
796 pud_t *pudp = fixmap_pud(addr);
797 pud_t pud = READ_ONCE(*pudp);
799 BUG_ON(pud_none(pud) || pud_bad(pud));
801 return pmd_offset_kimg(pudp, addr);
804 static inline pte_t * fixmap_pte(unsigned long addr)
806 return &bm_pte[pte_index(addr)];
810 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
811 * directly on kernel symbols (bm_p*d). This function is called too early to use
812 * lm_alias so __p*d_populate functions must be used to populate with the
813 * physical address from __pa_symbol.
815 void __init early_fixmap_init(void)
820 unsigned long addr = FIXADDR_START;
822 pgdp = pgd_offset_k(addr);
823 pgd = READ_ONCE(*pgdp);
824 if (CONFIG_PGTABLE_LEVELS > 3 &&
825 !(pgd_none(pgd) || pgd_page_paddr(pgd) == __pa_symbol(bm_pud))) {
827 * We only end up here if the kernel mapping and the fixmap
828 * share the top level pgd entry, which should only happen on
829 * 16k/4 levels configurations.
831 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
832 pudp = pud_offset_kimg(pgdp, addr);
835 __pgd_populate(pgdp, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
836 pudp = fixmap_pud(addr);
838 if (pud_none(READ_ONCE(*pudp)))
839 __pud_populate(pudp, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
840 pmdp = fixmap_pmd(addr);
841 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
844 * The boot-ioremap range spans multiple pmds, for which
845 * we are not prepared:
847 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
848 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
850 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
851 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
853 pr_warn("pmdp %p != %p, %p\n",
854 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
855 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
856 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
857 fix_to_virt(FIX_BTMAP_BEGIN));
858 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
859 fix_to_virt(FIX_BTMAP_END));
861 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
862 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
867 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
868 * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
870 void __set_fixmap(enum fixed_addresses idx,
871 phys_addr_t phys, pgprot_t flags)
873 unsigned long addr = __fix_to_virt(idx);
876 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
878 ptep = fixmap_pte(addr);
880 if (pgprot_val(flags)) {
881 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
883 pte_clear(&init_mm, addr, ptep);
884 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
888 void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
890 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
895 * Check whether the physical FDT address is set and meets the minimum
896 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
897 * at least 8 bytes so that we can always access the magic and size
898 * fields of the FDT header after mapping the first chunk, double check
899 * here if that is indeed the case.
901 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
902 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
906 * Make sure that the FDT region can be mapped without the need to
907 * allocate additional translation table pages, so that it is safe
908 * to call create_mapping_noalloc() this early.
910 * On 64k pages, the FDT will be mapped using PTEs, so we need to
911 * be in the same PMD as the rest of the fixmap.
912 * On 4k pages, we'll use section mappings for the FDT so we only
913 * have to be in the same PUD.
915 BUILD_BUG_ON(dt_virt_base % SZ_2M);
917 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
918 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
920 offset = dt_phys % SWAPPER_BLOCK_SIZE;
921 dt_virt = (void *)dt_virt_base + offset;
923 /* map the first chunk so we can read the size from the header */
924 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
925 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
927 if (fdt_magic(dt_virt) != FDT_MAGIC)
930 *size = fdt_totalsize(dt_virt);
931 if (*size > MAX_FDT_SIZE)
934 if (offset + *size > SWAPPER_BLOCK_SIZE)
935 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
936 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
941 int __init arch_ioremap_p4d_supported(void)
946 int __init arch_ioremap_pud_supported(void)
949 * Only 4k granule supports level 1 block mappings.
950 * SW table walks can't handle removal of intermediate entries.
952 return IS_ENABLED(CONFIG_ARM64_4K_PAGES) &&
953 !IS_ENABLED(CONFIG_ARM64_PTDUMP_DEBUGFS);
956 int __init arch_ioremap_pmd_supported(void)
958 /* See arch_ioremap_pud_supported() */
959 return !IS_ENABLED(CONFIG_ARM64_PTDUMP_DEBUGFS);
962 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
964 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
966 /* Only allow permission changes for now */
967 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
971 VM_BUG_ON(phys & ~PUD_MASK);
972 set_pud(pudp, new_pud);
976 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
978 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
980 /* Only allow permission changes for now */
981 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
985 VM_BUG_ON(phys & ~PMD_MASK);
986 set_pmd(pmdp, new_pmd);
990 int pud_clear_huge(pud_t *pudp)
992 if (!pud_sect(READ_ONCE(*pudp)))
998 int pmd_clear_huge(pmd_t *pmdp)
1000 if (!pmd_sect(READ_ONCE(*pmdp)))
1006 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1011 pmd = READ_ONCE(*pmdp);
1013 if (!pmd_table(pmd)) {
1018 table = pte_offset_kernel(pmdp, addr);
1020 __flush_tlb_kernel_pgtable(addr);
1021 pte_free_kernel(NULL, table);
1025 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1030 unsigned long next, end;
1032 pud = READ_ONCE(*pudp);
1034 if (!pud_table(pud)) {
1039 table = pmd_offset(pudp, addr);
1042 end = addr + PUD_SIZE;
1044 pmd_free_pte_page(pmdp, next);
1045 } while (pmdp++, next += PMD_SIZE, next != end);
1048 __flush_tlb_kernel_pgtable(addr);
1049 pmd_free(NULL, table);
1053 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1055 return 0; /* Don't attempt a block mapping */
1058 #ifdef CONFIG_MEMORY_HOTPLUG
1059 int arch_add_memory(int nid, u64 start, u64 size,
1060 struct mhp_restrictions *restrictions)
1064 if (rodata_full || debug_pagealloc_enabled())
1065 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1067 __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1068 size, PAGE_KERNEL, __pgd_pgtable_alloc, flags);
1070 return __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1073 void arch_remove_memory(int nid, u64 start, u64 size,
1074 struct vmem_altmap *altmap)
1076 unsigned long start_pfn = start >> PAGE_SHIFT;
1077 unsigned long nr_pages = size >> PAGE_SHIFT;
1080 * FIXME: Cleanup page tables (also in arch_add_memory() in case
1081 * adding fails). Until then, this function should only be used
1082 * during memory hotplug (adding memory), not for memory
1083 * unplug. ARCH_ENABLE_MEMORY_HOTREMOVE must not be
1086 __remove_pages(start_pfn, nr_pages, altmap);