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;
55 u64 vabits_user __ro_after_init;
56 EXPORT_SYMBOL(vabits_user);
58 u64 kimage_voffset __ro_after_init;
59 EXPORT_SYMBOL(kimage_voffset);
62 * Empty_zero_page is a special page that is used for zero-initialized data
65 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
66 EXPORT_SYMBOL(empty_zero_page);
68 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
69 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
70 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
72 static DEFINE_SPINLOCK(swapper_pgdir_lock);
74 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
78 spin_lock(&swapper_pgdir_lock);
79 fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
80 WRITE_ONCE(*fixmap_pgdp, pgd);
82 * We need dsb(ishst) here to ensure the page-table-walker sees
83 * our new entry before set_p?d() returns. The fixmap's
84 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
87 spin_unlock(&swapper_pgdir_lock);
90 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
91 unsigned long size, pgprot_t vma_prot)
94 return pgprot_noncached(vma_prot);
95 else if (file->f_flags & O_SYNC)
96 return pgprot_writecombine(vma_prot);
99 EXPORT_SYMBOL(phys_mem_access_prot);
101 static phys_addr_t __init early_pgtable_alloc(void)
106 phys = memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
109 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
110 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
111 * any level of table.
113 ptr = pte_set_fixmap(phys);
115 memset(ptr, 0, PAGE_SIZE);
118 * Implicit barriers also ensure the zeroed page is visible to the page
126 static bool pgattr_change_is_safe(u64 old, u64 new)
129 * The following mapping attributes may be updated in live
130 * kernel mappings without the need for break-before-make.
132 static const pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
134 /* creating or taking down mappings is always safe */
135 if (old == 0 || new == 0)
138 /* live contiguous mappings may not be manipulated at all */
139 if ((old | new) & PTE_CONT)
142 /* Transitioning from Non-Global to Global is unsafe */
143 if (old & ~new & PTE_NG)
146 return ((old ^ new) & ~mask) == 0;
149 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
150 phys_addr_t phys, pgprot_t prot)
154 ptep = pte_set_fixmap_offset(pmdp, addr);
156 pte_t old_pte = READ_ONCE(*ptep);
158 set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
161 * After the PTE entry has been populated once, we
162 * only allow updates to the permission attributes.
164 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
165 READ_ONCE(pte_val(*ptep))));
168 } while (ptep++, addr += PAGE_SIZE, addr != end);
173 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
174 unsigned long end, phys_addr_t phys,
176 phys_addr_t (*pgtable_alloc)(void),
180 pmd_t pmd = READ_ONCE(*pmdp);
182 BUG_ON(pmd_sect(pmd));
184 phys_addr_t pte_phys;
185 BUG_ON(!pgtable_alloc);
186 pte_phys = pgtable_alloc();
187 __pmd_populate(pmdp, pte_phys, PMD_TYPE_TABLE);
188 pmd = READ_ONCE(*pmdp);
190 BUG_ON(pmd_bad(pmd));
193 pgprot_t __prot = prot;
195 next = pte_cont_addr_end(addr, end);
197 /* use a contiguous mapping if the range is suitably aligned */
198 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
199 (flags & NO_CONT_MAPPINGS) == 0)
200 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
202 init_pte(pmdp, addr, next, phys, __prot);
205 } while (addr = next, addr != end);
208 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
209 phys_addr_t phys, pgprot_t prot,
210 phys_addr_t (*pgtable_alloc)(void), int flags)
215 pmdp = pmd_set_fixmap_offset(pudp, addr);
217 pmd_t old_pmd = READ_ONCE(*pmdp);
219 next = pmd_addr_end(addr, end);
221 /* try section mapping first */
222 if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
223 (flags & NO_BLOCK_MAPPINGS) == 0) {
224 pmd_set_huge(pmdp, phys, prot);
227 * After the PMD entry has been populated once, we
228 * only allow updates to the permission attributes.
230 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
231 READ_ONCE(pmd_val(*pmdp))));
233 alloc_init_cont_pte(pmdp, addr, next, phys, prot,
234 pgtable_alloc, flags);
236 BUG_ON(pmd_val(old_pmd) != 0 &&
237 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
240 } while (pmdp++, addr = next, addr != end);
245 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
246 unsigned long end, phys_addr_t phys,
248 phys_addr_t (*pgtable_alloc)(void), int flags)
251 pud_t pud = READ_ONCE(*pudp);
254 * Check for initial section mappings in the pgd/pud.
256 BUG_ON(pud_sect(pud));
258 phys_addr_t pmd_phys;
259 BUG_ON(!pgtable_alloc);
260 pmd_phys = pgtable_alloc();
261 __pud_populate(pudp, pmd_phys, PUD_TYPE_TABLE);
262 pud = READ_ONCE(*pudp);
264 BUG_ON(pud_bad(pud));
267 pgprot_t __prot = prot;
269 next = pmd_cont_addr_end(addr, end);
271 /* use a contiguous mapping if the range is suitably aligned */
272 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
273 (flags & NO_CONT_MAPPINGS) == 0)
274 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
276 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
279 } while (addr = next, addr != end);
282 static inline bool use_1G_block(unsigned long addr, unsigned long next,
285 if (PAGE_SHIFT != 12)
288 if (((addr | next | phys) & ~PUD_MASK) != 0)
294 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
295 phys_addr_t phys, pgprot_t prot,
296 phys_addr_t (*pgtable_alloc)(void),
301 pgd_t pgd = READ_ONCE(*pgdp);
304 phys_addr_t pud_phys;
305 BUG_ON(!pgtable_alloc);
306 pud_phys = pgtable_alloc();
307 __pgd_populate(pgdp, pud_phys, PUD_TYPE_TABLE);
308 pgd = READ_ONCE(*pgdp);
310 BUG_ON(pgd_bad(pgd));
312 pudp = pud_set_fixmap_offset(pgdp, addr);
314 pud_t old_pud = READ_ONCE(*pudp);
316 next = pud_addr_end(addr, end);
319 * For 4K granule only, attempt to put down a 1GB block
321 if (use_1G_block(addr, next, phys) &&
322 (flags & NO_BLOCK_MAPPINGS) == 0) {
323 pud_set_huge(pudp, phys, prot);
326 * After the PUD entry has been populated once, we
327 * only allow updates to the permission attributes.
329 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
330 READ_ONCE(pud_val(*pudp))));
332 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
333 pgtable_alloc, flags);
335 BUG_ON(pud_val(old_pud) != 0 &&
336 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
339 } while (pudp++, addr = next, addr != end);
344 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
345 unsigned long virt, phys_addr_t size,
347 phys_addr_t (*pgtable_alloc)(void),
350 unsigned long addr, length, end, next;
351 pgd_t *pgdp = pgd_offset_raw(pgdir, virt);
354 * If the virtual and physical address don't have the same offset
355 * within a page, we cannot map the region as the caller expects.
357 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
361 addr = virt & PAGE_MASK;
362 length = PAGE_ALIGN(size + (virt & ~PAGE_MASK));
366 next = pgd_addr_end(addr, end);
367 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
370 } while (pgdp++, addr = next, addr != end);
373 static phys_addr_t pgd_pgtable_alloc(void)
375 void *ptr = (void *)__get_free_page(PGALLOC_GFP);
376 if (!ptr || !pgtable_page_ctor(virt_to_page(ptr)))
379 /* Ensure the zeroed page is visible to the page table walker */
385 * This function can only be used to modify existing table entries,
386 * without allocating new levels of table. Note that this permits the
387 * creation of new section or page entries.
389 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
390 phys_addr_t size, pgprot_t prot)
392 if (virt < VMALLOC_START) {
393 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
397 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
401 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
402 unsigned long virt, phys_addr_t size,
403 pgprot_t prot, bool page_mappings_only)
407 BUG_ON(mm == &init_mm);
409 if (page_mappings_only)
410 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
412 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
413 pgd_pgtable_alloc, flags);
416 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
417 phys_addr_t size, pgprot_t prot)
419 if (virt < VMALLOC_START) {
420 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
425 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
428 /* flush the TLBs after updating live kernel mappings */
429 flush_tlb_kernel_range(virt, virt + size);
432 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
433 phys_addr_t end, pgprot_t prot, int flags)
435 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
436 prot, early_pgtable_alloc, flags);
439 void __init mark_linear_text_alias_ro(void)
442 * Remove the write permissions from the linear alias of .text/.rodata
444 update_mapping_prot(__pa_symbol(_text), (unsigned long)lm_alias(_text),
445 (unsigned long)__init_begin - (unsigned long)_text,
449 static void __init map_mem(pgd_t *pgdp)
451 phys_addr_t kernel_start = __pa_symbol(_text);
452 phys_addr_t kernel_end = __pa_symbol(__init_begin);
453 struct memblock_region *reg;
456 if (rodata_full || debug_pagealloc_enabled())
457 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
460 * Take care not to create a writable alias for the
461 * read-only text and rodata sections of the kernel image.
462 * So temporarily mark them as NOMAP to skip mappings in
463 * the following for-loop
465 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
466 #ifdef CONFIG_KEXEC_CORE
468 memblock_mark_nomap(crashk_res.start,
469 resource_size(&crashk_res));
472 /* map all the memory banks */
473 for_each_memblock(memory, reg) {
474 phys_addr_t start = reg->base;
475 phys_addr_t end = start + reg->size;
479 if (memblock_is_nomap(reg))
482 __map_memblock(pgdp, start, end, PAGE_KERNEL, flags);
486 * Map the linear alias of the [_text, __init_begin) interval
487 * as non-executable now, and remove the write permission in
488 * mark_linear_text_alias_ro() below (which will be called after
489 * alternative patching has completed). This makes the contents
490 * of the region accessible to subsystems such as hibernate,
491 * but protects it from inadvertent modification or execution.
492 * Note that contiguous mappings cannot be remapped in this way,
493 * so we should avoid them here.
495 __map_memblock(pgdp, kernel_start, kernel_end,
496 PAGE_KERNEL, NO_CONT_MAPPINGS);
497 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
499 #ifdef CONFIG_KEXEC_CORE
501 * Use page-level mappings here so that we can shrink the region
502 * in page granularity and put back unused memory to buddy system
503 * through /sys/kernel/kexec_crash_size interface.
505 if (crashk_res.end) {
506 __map_memblock(pgdp, crashk_res.start, crashk_res.end + 1,
508 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
509 memblock_clear_nomap(crashk_res.start,
510 resource_size(&crashk_res));
515 void mark_rodata_ro(void)
517 unsigned long section_size;
520 * mark .rodata as read only. Use __init_begin rather than __end_rodata
521 * to cover NOTES and EXCEPTION_TABLE.
523 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
524 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
525 section_size, PAGE_KERNEL_RO);
530 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
531 pgprot_t prot, struct vm_struct *vma,
532 int flags, unsigned long vm_flags)
534 phys_addr_t pa_start = __pa_symbol(va_start);
535 unsigned long size = va_end - va_start;
537 BUG_ON(!PAGE_ALIGNED(pa_start));
538 BUG_ON(!PAGE_ALIGNED(size));
540 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
541 early_pgtable_alloc, flags);
543 if (!(vm_flags & VM_NO_GUARD))
546 vma->addr = va_start;
547 vma->phys_addr = pa_start;
549 vma->flags = VM_MAP | vm_flags;
550 vma->caller = __builtin_return_address(0);
552 vm_area_add_early(vma);
555 static int __init parse_rodata(char *arg)
557 int ret = strtobool(arg, &rodata_enabled);
563 /* permit 'full' in addition to boolean options */
564 if (strcmp(arg, "full"))
567 rodata_enabled = true;
571 early_param("rodata", parse_rodata);
573 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
574 static int __init map_entry_trampoline(void)
576 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
577 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
579 /* The trampoline is always mapped and can therefore be global */
580 pgprot_val(prot) &= ~PTE_NG;
582 /* Map only the text into the trampoline page table */
583 memset(tramp_pg_dir, 0, PGD_SIZE);
584 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS, PAGE_SIZE,
585 prot, pgd_pgtable_alloc, 0);
587 /* Map both the text and data into the kernel page table */
588 __set_fixmap(FIX_ENTRY_TRAMP_TEXT, pa_start, prot);
589 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
590 extern char __entry_tramp_data_start[];
592 __set_fixmap(FIX_ENTRY_TRAMP_DATA,
593 __pa_symbol(__entry_tramp_data_start),
599 core_initcall(map_entry_trampoline);
603 * Create fine-grained mappings for the kernel.
605 static void __init map_kernel(pgd_t *pgdp)
607 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
608 vmlinux_initdata, vmlinux_data;
611 * External debuggers may need to write directly to the text
612 * mapping to install SW breakpoints. Allow this (only) when
613 * explicitly requested with rodata=off.
615 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
618 * Only rodata will be remapped with different permissions later on,
619 * all other segments are allowed to use contiguous mappings.
621 map_kernel_segment(pgdp, _text, _etext, text_prot, &vmlinux_text, 0,
623 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
624 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
625 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
626 &vmlinux_inittext, 0, VM_NO_GUARD);
627 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
628 &vmlinux_initdata, 0, VM_NO_GUARD);
629 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
631 if (!READ_ONCE(pgd_val(*pgd_offset_raw(pgdp, FIXADDR_START)))) {
633 * The fixmap falls in a separate pgd to the kernel, and doesn't
634 * live in the carveout for the swapper_pg_dir. We can simply
635 * re-use the existing dir for the fixmap.
637 set_pgd(pgd_offset_raw(pgdp, FIXADDR_START),
638 READ_ONCE(*pgd_offset_k(FIXADDR_START)));
639 } else if (CONFIG_PGTABLE_LEVELS > 3) {
641 * The fixmap shares its top level pgd entry with the kernel
642 * mapping. This can really only occur when we are running
643 * with 16k/4 levels, so we can simply reuse the pud level
646 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
647 pud_populate(&init_mm,
648 pud_set_fixmap_offset(pgdp, FIXADDR_START),
655 kasan_copy_shadow(pgdp);
659 * paging_init() sets up the page tables, initialises the zone memory
660 * maps and sets up the zero page.
662 void __init paging_init(void)
664 pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
671 cpu_replace_ttbr1(lm_alias(swapper_pg_dir));
672 init_mm.pgd = swapper_pg_dir;
674 memblock_free(__pa_symbol(init_pg_dir),
675 __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
677 memblock_allow_resize();
681 * Check whether a kernel address is valid (derived from arch/x86/).
683 int kern_addr_valid(unsigned long addr)
690 if ((((long)addr) >> VA_BITS) != -1UL)
693 pgdp = pgd_offset_k(addr);
694 if (pgd_none(READ_ONCE(*pgdp)))
697 pudp = pud_offset(pgdp, addr);
698 pud = READ_ONCE(*pudp);
703 return pfn_valid(pud_pfn(pud));
705 pmdp = pmd_offset(pudp, addr);
706 pmd = READ_ONCE(*pmdp);
711 return pfn_valid(pmd_pfn(pmd));
713 ptep = pte_offset_kernel(pmdp, addr);
714 pte = READ_ONCE(*ptep);
718 return pfn_valid(pte_pfn(pte));
720 #ifdef CONFIG_SPARSEMEM_VMEMMAP
721 #if !ARM64_SWAPPER_USES_SECTION_MAPS
722 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
723 struct vmem_altmap *altmap)
725 return vmemmap_populate_basepages(start, end, node);
727 #else /* !ARM64_SWAPPER_USES_SECTION_MAPS */
728 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
729 struct vmem_altmap *altmap)
731 unsigned long addr = start;
738 next = pmd_addr_end(addr, end);
740 pgdp = vmemmap_pgd_populate(addr, node);
744 pudp = vmemmap_pud_populate(pgdp, addr, node);
748 pmdp = pmd_offset(pudp, addr);
749 if (pmd_none(READ_ONCE(*pmdp))) {
752 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
756 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
758 vmemmap_verify((pte_t *)pmdp, node, addr, next);
759 } while (addr = next, addr != end);
763 #endif /* CONFIG_ARM64_64K_PAGES */
764 void vmemmap_free(unsigned long start, unsigned long end,
765 struct vmem_altmap *altmap)
768 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
770 static inline pud_t * fixmap_pud(unsigned long addr)
772 pgd_t *pgdp = pgd_offset_k(addr);
773 pgd_t pgd = READ_ONCE(*pgdp);
775 BUG_ON(pgd_none(pgd) || pgd_bad(pgd));
777 return pud_offset_kimg(pgdp, addr);
780 static inline pmd_t * fixmap_pmd(unsigned long addr)
782 pud_t *pudp = fixmap_pud(addr);
783 pud_t pud = READ_ONCE(*pudp);
785 BUG_ON(pud_none(pud) || pud_bad(pud));
787 return pmd_offset_kimg(pudp, addr);
790 static inline pte_t * fixmap_pte(unsigned long addr)
792 return &bm_pte[pte_index(addr)];
796 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
797 * directly on kernel symbols (bm_p*d). This function is called too early to use
798 * lm_alias so __p*d_populate functions must be used to populate with the
799 * physical address from __pa_symbol.
801 void __init early_fixmap_init(void)
806 unsigned long addr = FIXADDR_START;
808 pgdp = pgd_offset_k(addr);
809 pgd = READ_ONCE(*pgdp);
810 if (CONFIG_PGTABLE_LEVELS > 3 &&
811 !(pgd_none(pgd) || pgd_page_paddr(pgd) == __pa_symbol(bm_pud))) {
813 * We only end up here if the kernel mapping and the fixmap
814 * share the top level pgd entry, which should only happen on
815 * 16k/4 levels configurations.
817 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
818 pudp = pud_offset_kimg(pgdp, addr);
821 __pgd_populate(pgdp, __pa_symbol(bm_pud), PUD_TYPE_TABLE);
822 pudp = fixmap_pud(addr);
824 if (pud_none(READ_ONCE(*pudp)))
825 __pud_populate(pudp, __pa_symbol(bm_pmd), PMD_TYPE_TABLE);
826 pmdp = fixmap_pmd(addr);
827 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
830 * The boot-ioremap range spans multiple pmds, for which
831 * we are not prepared:
833 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
834 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
836 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
837 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
839 pr_warn("pmdp %p != %p, %p\n",
840 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
841 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
842 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
843 fix_to_virt(FIX_BTMAP_BEGIN));
844 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
845 fix_to_virt(FIX_BTMAP_END));
847 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
848 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
853 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
854 * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
856 void __set_fixmap(enum fixed_addresses idx,
857 phys_addr_t phys, pgprot_t flags)
859 unsigned long addr = __fix_to_virt(idx);
862 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
864 ptep = fixmap_pte(addr);
866 if (pgprot_val(flags)) {
867 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
869 pte_clear(&init_mm, addr, ptep);
870 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
874 void *__init __fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
876 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
881 * Check whether the physical FDT address is set and meets the minimum
882 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
883 * at least 8 bytes so that we can always access the magic and size
884 * fields of the FDT header after mapping the first chunk, double check
885 * here if that is indeed the case.
887 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
888 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
892 * Make sure that the FDT region can be mapped without the need to
893 * allocate additional translation table pages, so that it is safe
894 * to call create_mapping_noalloc() this early.
896 * On 64k pages, the FDT will be mapped using PTEs, so we need to
897 * be in the same PMD as the rest of the fixmap.
898 * On 4k pages, we'll use section mappings for the FDT so we only
899 * have to be in the same PUD.
901 BUILD_BUG_ON(dt_virt_base % SZ_2M);
903 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
904 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
906 offset = dt_phys % SWAPPER_BLOCK_SIZE;
907 dt_virt = (void *)dt_virt_base + offset;
909 /* map the first chunk so we can read the size from the header */
910 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
911 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
913 if (fdt_magic(dt_virt) != FDT_MAGIC)
916 *size = fdt_totalsize(dt_virt);
917 if (*size > MAX_FDT_SIZE)
920 if (offset + *size > SWAPPER_BLOCK_SIZE)
921 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
922 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
927 void *__init fixmap_remap_fdt(phys_addr_t dt_phys)
932 dt_virt = __fixmap_remap_fdt(dt_phys, &size, PAGE_KERNEL_RO);
936 memblock_reserve(dt_phys, size);
940 int __init arch_ioremap_pud_supported(void)
942 /* only 4k granule supports level 1 block mappings */
943 return IS_ENABLED(CONFIG_ARM64_4K_PAGES);
946 int __init arch_ioremap_pmd_supported(void)
951 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
953 pgprot_t sect_prot = __pgprot(PUD_TYPE_SECT |
954 pgprot_val(mk_sect_prot(prot)));
955 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), sect_prot);
957 /* Only allow permission changes for now */
958 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
962 BUG_ON(phys & ~PUD_MASK);
963 set_pud(pudp, new_pud);
967 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
969 pgprot_t sect_prot = __pgprot(PMD_TYPE_SECT |
970 pgprot_val(mk_sect_prot(prot)));
971 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), sect_prot);
973 /* Only allow permission changes for now */
974 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
978 BUG_ON(phys & ~PMD_MASK);
979 set_pmd(pmdp, new_pmd);
983 int pud_clear_huge(pud_t *pudp)
985 if (!pud_sect(READ_ONCE(*pudp)))
991 int pmd_clear_huge(pmd_t *pmdp)
993 if (!pmd_sect(READ_ONCE(*pmdp)))
999 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1004 pmd = READ_ONCE(*pmdp);
1006 if (!pmd_table(pmd)) {
1011 table = pte_offset_kernel(pmdp, addr);
1013 __flush_tlb_kernel_pgtable(addr);
1014 pte_free_kernel(NULL, table);
1018 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1023 unsigned long next, end;
1025 pud = READ_ONCE(*pudp);
1027 if (!pud_table(pud)) {
1032 table = pmd_offset(pudp, addr);
1035 end = addr + PUD_SIZE;
1037 pmd_free_pte_page(pmdp, next);
1038 } while (pmdp++, next += PMD_SIZE, next != end);
1041 __flush_tlb_kernel_pgtable(addr);
1042 pmd_free(NULL, table);
1046 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1048 return 0; /* Don't attempt a block mapping */
1051 #ifdef CONFIG_MEMORY_HOTPLUG
1052 int arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
1057 if (rodata_full || debug_pagealloc_enabled())
1058 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1060 __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1061 size, PAGE_KERNEL, pgd_pgtable_alloc, flags);
1063 return __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1064 altmap, want_memblock);