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/mmu_context.h>
46 #include <asm/ptdump.h>
47 #include <asm/tlbflush.h>
49 #define NO_BLOCK_MAPPINGS BIT(0)
50 #define NO_CONT_MAPPINGS BIT(1)
52 u64 idmap_t0sz = TCR_T0SZ(VA_BITS);
53 u64 idmap_ptrs_per_pgd = PTRS_PER_PGD;
54 u64 vabits_user __ro_after_init;
55 EXPORT_SYMBOL(vabits_user);
57 u64 kimage_voffset __ro_after_init;
58 EXPORT_SYMBOL(kimage_voffset);
61 * Empty_zero_page is a special page that is used for zero-initialized data
64 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
65 EXPORT_SYMBOL(empty_zero_page);
67 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
68 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
69 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
71 static DEFINE_SPINLOCK(swapper_pgdir_lock);
73 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
77 spin_lock(&swapper_pgdir_lock);
78 fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
79 WRITE_ONCE(*fixmap_pgdp, pgd);
81 * We need dsb(ishst) here to ensure the page-table-walker sees
82 * our new entry before set_p?d() returns. The fixmap's
83 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
86 spin_unlock(&swapper_pgdir_lock);
89 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
90 unsigned long size, pgprot_t vma_prot)
93 return pgprot_noncached(vma_prot);
94 else if (file->f_flags & O_SYNC)
95 return pgprot_writecombine(vma_prot);
98 EXPORT_SYMBOL(phys_mem_access_prot);
100 static phys_addr_t __init early_pgtable_alloc(void)
105 phys = memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
108 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
109 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
110 * any level of table.
112 ptr = pte_set_fixmap(phys);
114 memset(ptr, 0, PAGE_SIZE);
117 * Implicit barriers also ensure the zeroed page is visible to the page
125 static bool pgattr_change_is_safe(u64 old, u64 new)
128 * The following mapping attributes may be updated in live
129 * kernel mappings without the need for break-before-make.
131 static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
133 /* creating or taking down mappings is always safe */
134 if (old == 0 || new == 0)
137 /* live contiguous mappings may not be manipulated at all */
138 if ((old | new) & PTE_CONT)
141 /* Transitioning from Non-Global to Global is unsafe */
142 if (old & ~new & PTE_NG)
145 return ((old ^ new) & ~mask) == 0;
148 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
149 phys_addr_t phys, pgprot_t prot)
153 ptep = pte_set_fixmap_offset(pmdp, addr);
155 pte_t old_pte = READ_ONCE(*ptep);
157 set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
160 * After the PTE entry has been populated once, we
161 * only allow updates to the permission attributes.
163 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
164 READ_ONCE(pte_val(*ptep))));
167 } while (ptep++, addr += PAGE_SIZE, addr != end);
172 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
173 unsigned long end, phys_addr_t phys,
175 phys_addr_t (*pgtable_alloc)(void),
179 pmd_t pmd = READ_ONCE(*pmdp);
181 BUG_ON(pmd_sect(pmd));
183 phys_addr_t pte_phys;
184 BUG_ON(!pgtable_alloc);
185 pte_phys = pgtable_alloc();
186 __pmd_populate(pmdp, pte_phys, PMD_TYPE_TABLE);
187 pmd = READ_ONCE(*pmdp);
189 BUG_ON(pmd_bad(pmd));
192 pgprot_t __prot = prot;
194 next = pte_cont_addr_end(addr, end);
196 /* use a contiguous mapping if the range is suitably aligned */
197 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
198 (flags & NO_CONT_MAPPINGS) == 0)
199 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
201 init_pte(pmdp, addr, next, phys, __prot);
204 } while (addr = next, addr != end);
207 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
208 phys_addr_t phys, pgprot_t prot,
209 phys_addr_t (*pgtable_alloc)(void), int flags)
214 pmdp = pmd_set_fixmap_offset(pudp, addr);
216 pmd_t old_pmd = READ_ONCE(*pmdp);
218 next = pmd_addr_end(addr, end);
220 /* try section mapping first */
221 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
222 (flags & NO_BLOCK_MAPPINGS) == 0) {
223 pmd_set_huge(pmdp, phys, prot);
226 * After the PMD entry has been populated once, we
227 * only allow updates to the permission attributes.
229 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
230 READ_ONCE(pmd_val(*pmdp))));
232 alloc_init_cont_pte(pmdp, addr, next, phys, prot,
233 pgtable_alloc, flags);
235 BUG_ON(pmd_val(old_pmd) != 0 &&
236 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
239 } while (pmdp++, addr = next, addr != end);
244 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
245 unsigned long end, phys_addr_t phys,
247 phys_addr_t (*pgtable_alloc)(void), int flags)
250 pud_t pud = READ_ONCE(*pudp);
253 * Check for initial section mappings in the pgd/pud.
255 BUG_ON(pud_sect(pud));
257 phys_addr_t pmd_phys;
258 BUG_ON(!pgtable_alloc);
259 pmd_phys = pgtable_alloc();
260 __pud_populate(pudp, pmd_phys, PUD_TYPE_TABLE);
261 pud = READ_ONCE(*pudp);
263 BUG_ON(pud_bad(pud));
266 pgprot_t __prot = prot;
268 next = pmd_cont_addr_end(addr, end);
270 /* use a contiguous mapping if the range is suitably aligned */
271 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
272 (flags & NO_CONT_MAPPINGS) == 0)
273 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
275 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
278 } while (addr = next, addr != end);
281 static inline bool use_1G_block(unsigned long addr, unsigned long next,
284 if (PAGE_SHIFT != 12)
287 if (((addr | next | phys) & ~PUD_MASK) != 0)
293 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
294 phys_addr_t phys, pgprot_t prot,
295 phys_addr_t (*pgtable_alloc)(void),
300 pgd_t pgd = READ_ONCE(*pgdp);
303 phys_addr_t pud_phys;
304 BUG_ON(!pgtable_alloc);
305 pud_phys = pgtable_alloc();
306 __pgd_populate(pgdp, pud_phys, PUD_TYPE_TABLE);
307 pgd = READ_ONCE(*pgdp);
309 BUG_ON(pgd_bad(pgd));
311 pudp = pud_set_fixmap_offset(pgdp, addr);
313 pud_t old_pud = READ_ONCE(*pudp);
315 next = pud_addr_end(addr, end);
318 * For 4K granule only, attempt to put down a 1GB block
320 if (use_1G_block(addr, next, phys) &&
321 (flags & NO_BLOCK_MAPPINGS) == 0) {
322 pud_set_huge(pudp, phys, prot);
325 * After the PUD entry has been populated once, we
326 * only allow updates to the permission attributes.
328 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
329 READ_ONCE(pud_val(*pudp))));
331 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
332 pgtable_alloc, flags);
334 BUG_ON(pud_val(old_pud) != 0 &&
335 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
338 } while (pudp++, addr = next, addr != end);
343 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
344 unsigned long virt, phys_addr_t size,
346 phys_addr_t (*pgtable_alloc)(void),
349 unsigned long addr, length, end, next;
350 pgd_t *pgdp = pgd_offset_raw(pgdir, virt);
353 * If the virtual and physical address don't have the same offset
354 * within a page, we cannot map the region as the caller expects.
356 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
360 addr = virt & PAGE_MASK;
361 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
365 next = pgd_addr_end(addr, end);
366 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
369 } while (pgdp++, addr = next, addr != end);
372 static phys_addr_t pgd_pgtable_alloc(void)
374 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
375 if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
378 /* Ensure the zeroed page is visible to the page table walker */
384 * This function can only be used to modify existing table entries,
385 * without allocating new levels of table. Note that this permits the
386 * creation of new section or page entries.
388 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
389 phys_addr_t size, pgprot_t prot)
391 if (virt < VMALLOC_START) {
392 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
396 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
400 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
401 unsigned long virt, phys_addr_t size,
402 pgprot_t prot, bool page_mappings_only)
406 BUG_ON(mm == &init_mm);
408 if (page_mappings_only)
409 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
411 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
412 pgd_pgtable_alloc, flags);
415 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
416 phys_addr_t size, pgprot_t prot)
418 if (virt < VMALLOC_START) {
419 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
424 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
427 /* flush the TLBs after updating live kernel mappings */
428 flush_tlb_kernel_range(virt, virt + size);
431 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
432 phys_addr_t end, pgprot_t prot, int flags)
434 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
435 prot, early_pgtable_alloc, flags);
438 void __init mark_linear_text_alias_ro(void)
441 * Remove the write permissions from the linear alias of .text/.rodata
443 update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
444 (unsigned long)__init_begin - (unsigned long)_text,
448 static void __init map_mem(pgd_t *pgdp)
450 phys_addr_t kernel_start = __pa_symbol(_text);
451 phys_addr_t kernel_end = __pa_symbol(__init_begin);
452 struct memblock_region *reg;
455 if (rodata_full || debug_pagealloc_enabled())
456 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
459 * Take care not to create a writable alias for the
460 * read-only text and rodata sections of the kernel image.
461 * So temporarily mark them as NOMAP to skip mappings in
462 * the following for-loop
464 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
465 #ifdef CONFIG_KEXEC_CORE
467 memblock_mark_nomap(crashk_res.start,
468 resource_size(&crashk_res));
471 /* map all the memory banks */
472 for_each_memblock(memory, reg) {
473 phys_addr_t start = reg->base;
474 phys_addr_t end = start + reg->size;
478 if (memblock_is_nomap(reg))
481 __map_memblock(pgdp, start, end, PAGE_KERNEL, flags);
485 * Map the linear alias of the [_text, __init_begin) interval
486 * as non-executable now, and remove the write permission in
487 * mark_linear_text_alias_ro() below (which will be called after
488 * alternative patching has completed). This makes the contents
489 * of the region accessible to subsystems such as hibernate,
490 * but protects it from inadvertent modification or execution.
491 * Note that contiguous mappings cannot be remapped in this way,
492 * so we should avoid them here.
494 __map_memblock(pgdp, kernel_start, kernel_end,
495 PAGE_KERNEL, NO_CONT_MAPPINGS);
496 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
498 #ifdef CONFIG_KEXEC_CORE
500 * Use page-level mappings here so that we can shrink the region
501 * in page granularity and put back unused memory to buddy system
502 * through /sys/kernel/kexec_crash_size interface.
504 if (crashk_res.end) {
505 __map_memblock(pgdp, crashk_res.start, crashk_res.end + 1,
507 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
508 memblock_clear_nomap(crashk_res.start,
509 resource_size(&crashk_res));
514 void mark_rodata_ro(void)
516 unsigned long section_size;
519 * mark .rodata as read only. Use __init_begin rather than __end_rodata
520 * to cover NOTES and EXCEPTION_TABLE.
522 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
523 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
524 section_size, PAGE_KERNEL_RO);
529 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
530 pgprot_t prot, struct vm_struct *vma,
531 int flags, unsigned long vm_flags)
533 phys_addr_t pa_start = __pa_symbol(va_start);
534 unsigned long size = va_end - va_start;
536 BUG_ON(!PAGE_ALIGNED(pa_start));
537 BUG_ON(!PAGE_ALIGNED(size));
539 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
540 early_pgtable_alloc, flags);
542 if (!(vm_flags & VM_NO_GUARD))
545 vma->addr = va_start;
546 vma->phys_addr = pa_start;
548 vma->flags = VM_MAP | vm_flags;
549 vma->caller = __builtin_return_address(0);
551 vm_area_add_early(vma);
554 static int __init parse_rodata(char *arg)
556 int ret = strtobool(arg, &rodata_enabled);
562 /* permit 'full' in addition to boolean options */
563 if (strcmp(arg, "full"))
566 rodata_enabled = true;
570 early_param("rodata", parse_rodata);
572 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
573 static int __init map_entry_trampoline(void)
575 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
576 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
578 /* The trampoline is always mapped and can therefore be global */
579 pgprot_val(prot) &= ~PTE_NG;
581 /* Map only the text into the trampoline page table */
582 memset(tramp_pg_dir, 0, PGD_SIZE);
583 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE,
584 prot, pgd_pgtable_alloc, 0);
586 /* Map both the text and data into the kernel page table */
587 __set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot);
588 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
589 extern char __entry_tramp_data_start[];
591 __set_fixmap(FIX_ENTRY_TRAMP_DATA,
592 __pa_symbol(__entry_tramp_data_start),
598 core_initcall(map_entry_trampoline);
602 * Create fine-grained mappings for the kernel.
604 static void __init map_kernel(pgd_t *pgdp)
606 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
607 vmlinux_initdata, vmlinux_data;
610 * External debuggers may need to write directly to the text
611 * mapping to install SW breakpoints. Allow this (only) when
612 * explicitly requested with rodata=off.
614 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
617 * Only rodata will be remapped with different permissions later on,
618 * all other segments are allowed to use contiguous mappings.
620 map_kernel_segment(pgdp, _text, _etext, text_prot, &vmlinux_text, 0,
622 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
623 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
624 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
625 &vmlinux_inittext, 0, VM_NO_GUARD);
626 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
627 &vmlinux_initdata, 0, VM_NO_GUARD);
628 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
630 if (!READ_ONCE(pgd_val(*pgd_offset_raw(pgdp, FIXADDR_START)))) {
632 * The fixmap falls in a separate pgd to the kernel, and doesn't
633 * live in the carveout for the swapper_pg_dir. We can simply
634 * re-use the existing dir for the fixmap.
636 set_pgd(pgd_offset_raw(pgdp, FIXADDR_START),
637 READ_ONCE(*pgd_offset_k(FIXADDR_START)));
638 } else if (CONFIG_PGTABLE_LEVELS > 3) {
640 * The fixmap shares its top level pgd entry with the kernel
641 * mapping. This can really only occur when we are running
642 * with 16k/4 levels, so we can simply reuse the pud level
645 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
646 pud_populate(&init_mm,
647 pud_set_fixmap_offset(pgdp, FIXADDR_START),
654 kasan_copy_shadow(pgdp);
657 void __init paging_init(void)
659 pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
666 cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
667 init_mm.pgd = swapper_pg_dir;
669 memblock_free(__pa_symbol(init_pg_dir),
670 __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
672 memblock_allow_resize();
676 * Check whether a kernel address is valid (derived from arch/x86/).
678 int kern_addr_valid(unsigned long addr)
685 if ((((long)addr) >> VA_BITS) != -1UL)
688 pgdp = pgd_offset_k(addr);
689 if (pgd_none(READ_ONCE(*pgdp)))
692 pudp = pud_offset(pgdp, addr);
693 pud = READ_ONCE(*pudp);
698 return pfn_valid(pud_pfn(pud));
700 pmdp = pmd_offset(pudp, addr);
701 pmd = READ_ONCE(*pmdp);
706 return pfn_valid(pmd_pfn(pmd));
708 ptep = pte_offset_kernel(pmdp, addr);
709 pte = READ_ONCE(*ptep);
713 return pfn_valid(pte_pfn(pte));
715 #ifdef CONFIG_SPARSEMEM_VMEMMAP
716 #if !ARM64_SWAPPER_USES_SECTION_MAPS
717 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
718 struct vmem_altmap *altmap)
720 return vmemmap_populate_basepages(start, end, node);
722 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
723 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
724 struct vmem_altmap *altmap)
726 unsigned long addr = start;
733 next = pmd_addr_end(addr, end);
735 pgdp = vmemmap_pgd_populate(addr, node);
739 pudp = vmemmap_pud_populate(pgdp, addr, node);
743 pmdp = pmd_offset(pudp, addr);
744 if (pmd_none(READ_ONCE(*pmdp))) {
747 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
751 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
753 vmemmap_verify((pte_t *)pmdp, node, addr, next);
754 } while (addr = next, addr != end);
758 #endif /* CONFIG_ARM64_64K_PAGES */
759 void vmemmap_free(unsigned long start, unsigned long end,
760 struct vmem_altmap *altmap)
763 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
765 static inline pud_t * fixmap_pud(unsigned long addr)
767 pgd_t *pgdp = pgd_offset_k(addr);
768 pgd_t pgd = READ_ONCE(*pgdp);
770 BUG_ON(pgd_none(pgd) || pgd_bad(pgd));
772 return pud_offset_kimg(pgdp, addr);
775 static inline pmd_t * fixmap_pmd(unsigned long addr)
777 pud_t *pudp = fixmap_pud(addr);
778 pud_t pud = READ_ONCE(*pudp);
780 BUG_ON(pud_none(pud) || pud_bad(pud));
782 return pmd_offset_kimg(pudp, addr);
785 static inline pte_t * fixmap_pte(unsigned long addr)
787 return &bm_pte[pte_index(addr)];
791 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
792 * directly on kernel symbols (bm_p*d). This function is called too early to use
793 * lm_alias so __p*d_populate functions must be used to populate with the
794 * physical address from __pa_symbol.
796 void __init early_fixmap_init(void)
801 unsigned long addr = FIXADDR_START;
803 pgdp = pgd_offset_k(addr);
804 pgd = READ_ONCE(*pgdp);
805 if (CONFIG_PGTABLE_LEVELS > 3 &&
806 !(pgd_none(pgd) || pgd_page_paddr(pgd) == __pa_symbol(bm_pud))) {
808 * We only end up here if the kernel mapping and the fixmap
809 * share the top level pgd entry, which should only happen on
810 * 16k/4 levels configurations.
812 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
813 pudp = pud_offset_kimg(pgdp, addr);
816 __pgd_populate(pgdp, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
817 pudp = fixmap_pud(addr);
819 if (pud_none(READ_ONCE(*pudp)))
820 __pud_populate(pudp, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
821 pmdp = fixmap_pmd(addr);
822 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
825 * The boot-ioremap range spans multiple pmds, for which
826 * we are not prepared:
828 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
829 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
831 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
832 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
834 pr_warn("pmdp %p != %p, %p\n",
835 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
836 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
837 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
838 fix_to_virt(FIX_BTMAP_BEGIN));
839 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
840 fix_to_virt(FIX_BTMAP_END));
842 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
843 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
848 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
849 * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
851 void __set_fixmap(enum fixed_addresses idx,
852 phys_addr_t phys, pgprot_t flags)
854 unsigned long addr = __fix_to_virt(idx);
857 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
859 ptep = fixmap_pte(addr);
861 if (pgprot_val(flags)) {
862 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
864 pte_clear(&init_mm, addr, ptep);
865 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
869 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
871 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
876 * Check whether the physical FDT address is set and meets the minimum
877 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
878 * at least 8 bytes so that we can always access the magic and size
879 * fields of the FDT header after mapping the first chunk, double check
880 * here if that is indeed the case.
882 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
883 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
887 * Make sure that the FDT region can be mapped without the need to
888 * allocate additional translation table pages, so that it is safe
889 * to call create_mapping_noalloc() this early.
891 * On 64k pages, the FDT will be mapped using PTEs, so we need to
892 * be in the same PMD as the rest of the fixmap.
893 * On 4k pages, we'll use section mappings for the FDT so we only
894 * have to be in the same PUD.
896 BUILD_BUG_ON(dt_virt_base % SZ_2M);
898 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
899 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
901 offset = dt_phys % SWAPPER_BLOCK_SIZE;
902 dt_virt = (void *)dt_virt_base + offset;
904 /* map the first chunk so we can read the size from the header */
905 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
906 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
908 if (fdt_magic(dt_virt) != FDT_MAGIC)
911 *size = fdt_totalsize(dt_virt);
912 if (*size > MAX_FDT_SIZE)
915 if (offset + *size > SWAPPER_BLOCK_SIZE)
916 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
917 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
922 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
927 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
931 memblock_reserve(dt_phys, size);
935 int __init arch_ioremap_pud_supported(void)
937 /* only 4k granule supports level 1 block mappings */
938 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
941 int __init arch_ioremap_pmd_supported(void)
946 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
948 pgprot_t sect_prot = __pgprot(PUD_TYPE_SECT |
949 pgprot_val(mk_sect_prot(prot)));
950 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), sect_prot);
952 /* Only allow permission changes for now */
953 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
957 BUG_ON(phys & ~PUD_MASK);
958 set_pud(pudp, new_pud);
962 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
964 pgprot_t sect_prot = __pgprot(PMD_TYPE_SECT |
965 pgprot_val(mk_sect_prot(prot)));
966 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), sect_prot);
968 /* Only allow permission changes for now */
969 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
973 BUG_ON(phys & ~PMD_MASK);
974 set_pmd(pmdp, new_pmd);
978 int pud_clear_huge(pud_t *pudp)
980 if (!pud_sect(READ_ONCE(*pudp)))
986 int pmd_clear_huge(pmd_t *pmdp)
988 if (!pmd_sect(READ_ONCE(*pmdp)))
994 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
999 pmd = READ_ONCE(*pmdp);
1001 if (!pmd_table(pmd)) {
1006 table = pte_offset_kernel(pmdp, addr);
1008 __flush_tlb_kernel_pgtable(addr);
1009 pte_free_kernel(NULL, table);
1013 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1018 unsigned long next, end;
1020 pud = READ_ONCE(*pudp);
1022 if (!pud_table(pud)) {
1027 table = pmd_offset(pudp, addr);
1030 end = addr + PUD_SIZE;
1032 pmd_free_pte_page(pmdp, next);
1033 } while (pmdp++, next += PMD_SIZE, next != end);
1036 __flush_tlb_kernel_pgtable(addr);
1037 pmd_free(NULL, table);
1041 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1043 return 0; /* Don't attempt a block mapping */
1046 #ifdef CONFIG_MEMORY_HOTPLUG
1047 int arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
1052 if (rodata_full || debug_pagealloc_enabled())
1053 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1055 __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1056 size, PAGE_KERNEL, pgd_pgtable_alloc, flags);
1058 return __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1059 altmap, want_memblock);