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(int shift)
105 phys = memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
107 panic("Failed to allocate page table page\n");
110 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
111 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
112 * any level of table.
114 ptr = pte_set_fixmap(phys);
116 memset(ptr, 0, PAGE_SIZE);
119 * Implicit barriers also ensure the zeroed page is visible to the page
127 static bool pgattr_change_is_safe(u64 old, u64 new)
130 * The following mapping attributes may be updated in live
131 * kernel mappings without the need for break-before-make.
133 static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
135 /* creating or taking down mappings is always safe */
136 if (old == 0 || new == 0)
139 /* live contiguous mappings may not be manipulated at all */
140 if ((old | new) & PTE_CONT)
143 /* Transitioning from Non-Global to Global is unsafe */
144 if (old & ~new & PTE_NG)
147 return ((old ^ new) & ~mask) == 0;
150 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
151 phys_addr_t phys, pgprot_t prot)
155 ptep = pte_set_fixmap_offset(pmdp, addr);
157 pte_t old_pte = READ_ONCE(*ptep);
159 set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
162 * After the PTE entry has been populated once, we
163 * only allow updates to the permission attributes.
165 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
166 READ_ONCE(pte_val(*ptep))));
169 } while (ptep++, addr += PAGE_SIZE, addr != end);
174 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
175 unsigned long end, phys_addr_t phys,
177 phys_addr_t (*pgtable_alloc)(int),
181 pmd_t pmd = READ_ONCE(*pmdp);
183 BUG_ON(pmd_sect(pmd));
185 phys_addr_t pte_phys;
186 BUG_ON(!pgtable_alloc);
187 pte_phys = pgtable_alloc(PAGE_SHIFT);
188 __pmd_populate(pmdp, pte_phys, PMD_TYPE_TABLE);
189 pmd = READ_ONCE(*pmdp);
191 BUG_ON(pmd_bad(pmd));
194 pgprot_t __prot = prot;
196 next = pte_cont_addr_end(addr, end);
198 /* use a contiguous mapping if the range is suitably aligned */
199 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
200 (flags & NO_CONT_MAPPINGS) == 0)
201 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
203 init_pte(pmdp, addr, next, phys, __prot);
206 } while (addr = next, addr != end);
209 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
210 phys_addr_t phys, pgprot_t prot,
211 phys_addr_t (*pgtable_alloc)(int), int flags)
216 pmdp = pmd_set_fixmap_offset(pudp, addr);
218 pmd_t old_pmd = READ_ONCE(*pmdp);
220 next = pmd_addr_end(addr, end);
222 /* try section mapping first */
223 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
224 (flags & NO_BLOCK_MAPPINGS) == 0) {
225 pmd_set_huge(pmdp, phys, prot);
228 * After the PMD entry has been populated once, we
229 * only allow updates to the permission attributes.
231 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
232 READ_ONCE(pmd_val(*pmdp))));
234 alloc_init_cont_pte(pmdp, addr, next, phys, prot,
235 pgtable_alloc, flags);
237 BUG_ON(pmd_val(old_pmd) != 0 &&
238 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
241 } while (pmdp++, addr = next, addr != end);
246 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
247 unsigned long end, phys_addr_t phys,
249 phys_addr_t (*pgtable_alloc)(int), int flags)
252 pud_t pud = READ_ONCE(*pudp);
255 * Check for initial section mappings in the pgd/pud.
257 BUG_ON(pud_sect(pud));
259 phys_addr_t pmd_phys;
260 BUG_ON(!pgtable_alloc);
261 pmd_phys = pgtable_alloc(PMD_SHIFT);
262 __pud_populate(pudp, pmd_phys, PUD_TYPE_TABLE);
263 pud = READ_ONCE(*pudp);
265 BUG_ON(pud_bad(pud));
268 pgprot_t __prot = prot;
270 next = pmd_cont_addr_end(addr, end);
272 /* use a contiguous mapping if the range is suitably aligned */
273 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
274 (flags & NO_CONT_MAPPINGS) == 0)
275 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
277 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
280 } while (addr = next, addr != end);
283 static inline bool use_1G_block(unsigned long addr, unsigned long next,
286 if (PAGE_SHIFT != 12)
289 if (((addr | next | phys) & ~PUD_MASK) != 0)
295 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
296 phys_addr_t phys, pgprot_t prot,
297 phys_addr_t (*pgtable_alloc)(int),
302 pgd_t pgd = READ_ONCE(*pgdp);
305 phys_addr_t pud_phys;
306 BUG_ON(!pgtable_alloc);
307 pud_phys = pgtable_alloc(PUD_SHIFT);
308 __pgd_populate(pgdp, pud_phys, PUD_TYPE_TABLE);
309 pgd = READ_ONCE(*pgdp);
311 BUG_ON(pgd_bad(pgd));
313 pudp = pud_set_fixmap_offset(pgdp, addr);
315 pud_t old_pud = READ_ONCE(*pudp);
317 next = pud_addr_end(addr, end);
320 * For 4K granule only, attempt to put down a 1GB block
322 if (use_1G_block(addr, next, phys) &&
323 (flags & NO_BLOCK_MAPPINGS) == 0) {
324 pud_set_huge(pudp, phys, prot);
327 * After the PUD entry has been populated once, we
328 * only allow updates to the permission attributes.
330 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
331 READ_ONCE(pud_val(*pudp))));
333 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
334 pgtable_alloc, flags);
336 BUG_ON(pud_val(old_pud) != 0 &&
337 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
340 } while (pudp++, addr = next, addr != end);
345 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
346 unsigned long virt, phys_addr_t size,
348 phys_addr_t (*pgtable_alloc)(int),
351 unsigned long addr, length, end, next;
352 pgd_t *pgdp = pgd_offset_raw(pgdir, virt);
355 * If the virtual and physical address don't have the same offset
356 * within a page, we cannot map the region as the caller expects.
358 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
362 addr = virt & PAGE_MASK;
363 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
367 next = pgd_addr_end(addr, end);
368 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
371 } while (pgdp++, addr = next, addr != end);
374 static phys_addr_t __pgd_pgtable_alloc(int shift)
376 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
379 /* Ensure the zeroed page is visible to the page table walker */
384 static phys_addr_t pgd_pgtable_alloc(int shift)
386 phys_addr_t pa = __pgd_pgtable_alloc(shift);
389 * Call proper page table ctor in case later we need to
390 * call core mm functions like apply_to_page_range() on
391 * this pre-allocated page table.
393 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
394 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
396 if (shift == PAGE_SHIFT)
397 BUG_ON(!pgtable_page_ctor(phys_to_page(pa)));
398 else if (shift == PMD_SHIFT)
399 BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
405 * This function can only be used to modify existing table entries,
406 * without allocating new levels of table. Note that this permits the
407 * creation of new section or page entries.
409 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
410 phys_addr_t size, pgprot_t prot)
412 if (virt < VMALLOC_START) {
413 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
417 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
421 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
422 unsigned long virt, phys_addr_t size,
423 pgprot_t prot, bool page_mappings_only)
427 BUG_ON(mm == &init_mm);
429 if (page_mappings_only)
430 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
432 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
433 pgd_pgtable_alloc, flags);
436 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
437 phys_addr_t size, pgprot_t prot)
439 if (virt < VMALLOC_START) {
440 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
445 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
448 /* flush the TLBs after updating live kernel mappings */
449 flush_tlb_kernel_range(virt, virt + size);
452 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
453 phys_addr_t end, pgprot_t prot, int flags)
455 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
456 prot, early_pgtable_alloc, flags);
459 void __init mark_linear_text_alias_ro(void)
462 * Remove the write permissions from the linear alias of .text/.rodata
464 update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
465 (unsigned long)__init_begin - (unsigned long)_text,
469 static void __init map_mem(pgd_t *pgdp)
471 phys_addr_t kernel_start = __pa_symbol(_text);
472 phys_addr_t kernel_end = __pa_symbol(__init_begin);
473 struct memblock_region *reg;
476 if (rodata_full || debug_pagealloc_enabled())
477 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
480 * Take care not to create a writable alias for the
481 * read-only text and rodata sections of the kernel image.
482 * So temporarily mark them as NOMAP to skip mappings in
483 * the following for-loop
485 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
486 #ifdef CONFIG_KEXEC_CORE
488 memblock_mark_nomap(crashk_res.start,
489 resource_size(&crashk_res));
492 /* map all the memory banks */
493 for_each_memblock(memory, reg) {
494 phys_addr_t start = reg->base;
495 phys_addr_t end = start + reg->size;
499 if (memblock_is_nomap(reg))
502 __map_memblock(pgdp, start, end, PAGE_KERNEL, flags);
506 * Map the linear alias of the [_text, __init_begin) interval
507 * as non-executable now, and remove the write permission in
508 * mark_linear_text_alias_ro() below (which will be called after
509 * alternative patching has completed). This makes the contents
510 * of the region accessible to subsystems such as hibernate,
511 * but protects it from inadvertent modification or execution.
512 * Note that contiguous mappings cannot be remapped in this way,
513 * so we should avoid them here.
515 __map_memblock(pgdp, kernel_start, kernel_end,
516 PAGE_KERNEL, NO_CONT_MAPPINGS);
517 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
519 #ifdef CONFIG_KEXEC_CORE
521 * Use page-level mappings here so that we can shrink the region
522 * in page granularity and put back unused memory to buddy system
523 * through /sys/kernel/kexec_crash_size interface.
525 if (crashk_res.end) {
526 __map_memblock(pgdp, crashk_res.start, crashk_res.end + 1,
528 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
529 memblock_clear_nomap(crashk_res.start,
530 resource_size(&crashk_res));
535 void mark_rodata_ro(void)
537 unsigned long section_size;
540 * mark .rodata as read only. Use __init_begin rather than __end_rodata
541 * to cover NOTES and EXCEPTION_TABLE.
543 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
544 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
545 section_size, PAGE_KERNEL_RO);
550 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
551 pgprot_t prot, struct vm_struct *vma,
552 int flags, unsigned long vm_flags)
554 phys_addr_t pa_start = __pa_symbol(va_start);
555 unsigned long size = va_end - va_start;
557 BUG_ON(!PAGE_ALIGNED(pa_start));
558 BUG_ON(!PAGE_ALIGNED(size));
560 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
561 early_pgtable_alloc, flags);
563 if (!(vm_flags & VM_NO_GUARD))
566 vma->addr = va_start;
567 vma->phys_addr = pa_start;
569 vma->flags = VM_MAP | vm_flags;
570 vma->caller = __builtin_return_address(0);
572 vm_area_add_early(vma);
575 static int __init parse_rodata(char *arg)
577 int ret = strtobool(arg, &rodata_enabled);
583 /* permit 'full' in addition to boolean options */
584 if (strcmp(arg, "full"))
587 rodata_enabled = true;
591 early_param("rodata", parse_rodata);
593 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
594 static int __init map_entry_trampoline(void)
596 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
597 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
599 /* The trampoline is always mapped and can therefore be global */
600 pgprot_val(prot) &= ~PTE_NG;
602 /* Map only the text into the trampoline page table */
603 memset(tramp_pg_dir, 0, PGD_SIZE);
604 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE,
605 prot, __pgd_pgtable_alloc, 0);
607 /* Map both the text and data into the kernel page table */
608 __set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot);
609 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
610 extern char __entry_tramp_data_start[];
612 __set_fixmap(FIX_ENTRY_TRAMP_DATA,
613 __pa_symbol(__entry_tramp_data_start),
619 core_initcall(map_entry_trampoline);
623 * Create fine-grained mappings for the kernel.
625 static void __init map_kernel(pgd_t *pgdp)
627 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
628 vmlinux_initdata, vmlinux_data;
631 * External debuggers may need to write directly to the text
632 * mapping to install SW breakpoints. Allow this (only) when
633 * explicitly requested with rodata=off.
635 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
638 * Only rodata will be remapped with different permissions later on,
639 * all other segments are allowed to use contiguous mappings.
641 map_kernel_segment(pgdp, _text, _etext, text_prot, &vmlinux_text, 0,
643 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
644 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
645 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
646 &vmlinux_inittext, 0, VM_NO_GUARD);
647 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
648 &vmlinux_initdata, 0, VM_NO_GUARD);
649 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
651 if (!READ_ONCE(pgd_val(*pgd_offset_raw(pgdp, FIXADDR_START)))) {
653 * The fixmap falls in a separate pgd to the kernel, and doesn't
654 * live in the carveout for the swapper_pg_dir. We can simply
655 * re-use the existing dir for the fixmap.
657 set_pgd(pgd_offset_raw(pgdp, FIXADDR_START),
658 READ_ONCE(*pgd_offset_k(FIXADDR_START)));
659 } else if (CONFIG_PGTABLE_LEVELS > 3) {
661 * The fixmap shares its top level pgd entry with the kernel
662 * mapping. This can really only occur when we are running
663 * with 16k/4 levels, so we can simply reuse the pud level
666 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
667 pud_populate(&init_mm,
668 pud_set_fixmap_offset(pgdp, FIXADDR_START),
675 kasan_copy_shadow(pgdp);
678 void __init paging_init(void)
680 pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
687 cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
688 init_mm.pgd = swapper_pg_dir;
690 memblock_free(__pa_symbol(init_pg_dir),
691 __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
693 memblock_allow_resize();
697 * Check whether a kernel address is valid (derived from arch/x86/).
699 int kern_addr_valid(unsigned long addr)
706 if ((((long)addr) >> VA_BITS) != -1UL)
709 pgdp = pgd_offset_k(addr);
710 if (pgd_none(READ_ONCE(*pgdp)))
713 pudp = pud_offset(pgdp, addr);
714 pud = READ_ONCE(*pudp);
719 return pfn_valid(pud_pfn(pud));
721 pmdp = pmd_offset(pudp, addr);
722 pmd = READ_ONCE(*pmdp);
727 return pfn_valid(pmd_pfn(pmd));
729 ptep = pte_offset_kernel(pmdp, addr);
730 pte = READ_ONCE(*ptep);
734 return pfn_valid(pte_pfn(pte));
736 #ifdef CONFIG_SPARSEMEM_VMEMMAP
737 #if !ARM64_SWAPPER_USES_SECTION_MAPS
738 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
739 struct vmem_altmap *altmap)
741 return vmemmap_populate_basepages(start, end, node);
743 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
744 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
745 struct vmem_altmap *altmap)
747 unsigned long addr = start;
754 next = pmd_addr_end(addr, end);
756 pgdp = vmemmap_pgd_populate(addr, node);
760 pudp = vmemmap_pud_populate(pgdp, addr, node);
764 pmdp = pmd_offset(pudp, addr);
765 if (pmd_none(READ_ONCE(*pmdp))) {
768 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
772 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
774 vmemmap_verify((pte_t *)pmdp, node, addr, next);
775 } while (addr = next, addr != end);
779 #endif /* CONFIG_ARM64_64K_PAGES */
780 void vmemmap_free(unsigned long start, unsigned long end,
781 struct vmem_altmap *altmap)
784 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
786 static inline pud_t * fixmap_pud(unsigned long addr)
788 pgd_t *pgdp = pgd_offset_k(addr);
789 pgd_t pgd = READ_ONCE(*pgdp);
791 BUG_ON(pgd_none(pgd) || pgd_bad(pgd));
793 return pud_offset_kimg(pgdp, addr);
796 static inline pmd_t * fixmap_pmd(unsigned long addr)
798 pud_t *pudp = fixmap_pud(addr);
799 pud_t pud = READ_ONCE(*pudp);
801 BUG_ON(pud_none(pud) || pud_bad(pud));
803 return pmd_offset_kimg(pudp, addr);
806 static inline pte_t * fixmap_pte(unsigned long addr)
808 return &bm_pte[pte_index(addr)];
812 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
813 * directly on kernel symbols (bm_p*d). This function is called too early to use
814 * lm_alias so __p*d_populate functions must be used to populate with the
815 * physical address from __pa_symbol.
817 void __init early_fixmap_init(void)
822 unsigned long addr = FIXADDR_START;
824 pgdp = pgd_offset_k(addr);
825 pgd = READ_ONCE(*pgdp);
826 if (CONFIG_PGTABLE_LEVELS > 3 &&
827 !(pgd_none(pgd) || pgd_page_paddr(pgd) == __pa_symbol(bm_pud))) {
829 * We only end up here if the kernel mapping and the fixmap
830 * share the top level pgd entry, which should only happen on
831 * 16k/4 levels configurations.
833 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
834 pudp = pud_offset_kimg(pgdp, addr);
837 __pgd_populate(pgdp, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
838 pudp = fixmap_pud(addr);
840 if (pud_none(READ_ONCE(*pudp)))
841 __pud_populate(pudp, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
842 pmdp = fixmap_pmd(addr);
843 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
846 * The boot-ioremap range spans multiple pmds, for which
847 * we are not prepared:
849 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
850 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
852 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
853 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
855 pr_warn("pmdp %p != %p, %p\n",
856 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
857 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
858 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
859 fix_to_virt(FIX_BTMAP_BEGIN));
860 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
861 fix_to_virt(FIX_BTMAP_END));
863 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
864 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
869 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
870 * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
872 void __set_fixmap(enum fixed_addresses idx,
873 phys_addr_t phys, pgprot_t flags)
875 unsigned long addr = __fix_to_virt(idx);
878 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
880 ptep = fixmap_pte(addr);
882 if (pgprot_val(flags)) {
883 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
885 pte_clear(&init_mm, addr, ptep);
886 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
890 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
892 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
897 * Check whether the physical FDT address is set and meets the minimum
898 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
899 * at least 8 bytes so that we can always access the magic and size
900 * fields of the FDT header after mapping the first chunk, double check
901 * here if that is indeed the case.
903 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
904 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
908 * Make sure that the FDT region can be mapped without the need to
909 * allocate additional translation table pages, so that it is safe
910 * to call create_mapping_noalloc() this early.
912 * On 64k pages, the FDT will be mapped using PTEs, so we need to
913 * be in the same PMD as the rest of the fixmap.
914 * On 4k pages, we'll use section mappings for the FDT so we only
915 * have to be in the same PUD.
917 BUILD_BUG_ON(dt_virt_base % SZ_2M);
919 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
920 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
922 offset = dt_phys % SWAPPER_BLOCK_SIZE;
923 dt_virt = (void *)dt_virt_base + offset;
925 /* map the first chunk so we can read the size from the header */
926 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
927 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
929 if (fdt_magic(dt_virt) != FDT_MAGIC)
932 *size = fdt_totalsize(dt_virt);
933 if (*size > MAX_FDT_SIZE)
936 if (offset + *size > SWAPPER_BLOCK_SIZE)
937 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
938 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
943 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
948 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
952 memblock_reserve(dt_phys, size);
956 int __init arch_ioremap_pud_supported(void)
958 /* only 4k granule supports level 1 block mappings */
959 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
962 int __init arch_ioremap_pmd_supported(void)
967 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
969 pgprot_t sect_prot = __pgprot(PUD_TYPE_SECT |
970 pgprot_val(mk_sect_prot(prot)));
971 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), sect_prot);
973 /* Only allow permission changes for now */
974 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
978 BUG_ON(phys & ~PUD_MASK);
979 set_pud(pudp, new_pud);
983 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
985 pgprot_t sect_prot = __pgprot(PMD_TYPE_SECT |
986 pgprot_val(mk_sect_prot(prot)));
987 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), sect_prot);
989 /* Only allow permission changes for now */
990 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
994 BUG_ON(phys & ~PMD_MASK);
995 set_pmd(pmdp, new_pmd);
999 int pud_clear_huge(pud_t *pudp)
1001 if (!pud_sect(READ_ONCE(*pudp)))
1007 int pmd_clear_huge(pmd_t *pmdp)
1009 if (!pmd_sect(READ_ONCE(*pmdp)))
1015 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1020 pmd = READ_ONCE(*pmdp);
1022 if (!pmd_table(pmd)) {
1027 table = pte_offset_kernel(pmdp, addr);
1029 __flush_tlb_kernel_pgtable(addr);
1030 pte_free_kernel(NULL, table);
1034 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1039 unsigned long next, end;
1041 pud = READ_ONCE(*pudp);
1043 if (!pud_table(pud)) {
1048 table = pmd_offset(pudp, addr);
1051 end = addr + PUD_SIZE;
1053 pmd_free_pte_page(pmdp, next);
1054 } while (pmdp++, next += PMD_SIZE, next != end);
1057 __flush_tlb_kernel_pgtable(addr);
1058 pmd_free(NULL, table);
1062 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1064 return 0; /* Don't attempt a block mapping */
1067 #ifdef CONFIG_MEMORY_HOTPLUG
1068 int arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
1073 if (rodata_full || debug_pagealloc_enabled())
1074 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1076 __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1077 size, PAGE_KERNEL, __pgd_pgtable_alloc, flags);
1079 return __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1080 altmap, want_memblock);