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>
20 #include <linux/memremap.h>
21 #include <linux/memory.h>
25 #include <linux/vmalloc.h>
26 #include <linux/set_memory.h>
27 #include <linux/kfence.h>
29 #include <asm/barrier.h>
30 #include <asm/cputype.h>
31 #include <asm/fixmap.h>
32 #include <asm/kasan.h>
33 #include <asm/kernel-pgtable.h>
34 #include <asm/sections.h>
35 #include <asm/setup.h>
36 #include <linux/sizes.h>
38 #include <asm/mmu_context.h>
39 #include <asm/ptdump.h>
40 #include <asm/tlbflush.h>
41 #include <asm/pgalloc.h>
42 #include <asm/kfence.h>
44 #define NO_BLOCK_MAPPINGS BIT(0)
45 #define NO_CONT_MAPPINGS BIT(1)
46 #define NO_EXEC_MAPPINGS BIT(2) /* assumes FEAT_HPDS is not used */
48 u64 kimage_voffset __ro_after_init;
49 EXPORT_SYMBOL(kimage_voffset);
51 u32 __boot_cpu_mode[] = { BOOT_CPU_MODE_EL2, BOOT_CPU_MODE_EL1 };
53 static bool rodata_is_rw __ro_after_init = true;
56 * The booting CPU updates the failed status @__early_cpu_boot_status,
57 * with MMU turned off.
59 long __section(".mmuoff.data.write") __early_cpu_boot_status;
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 DEFINE_SPINLOCK(swapper_pgdir_lock);
69 static DEFINE_MUTEX(fixmap_lock);
71 void noinstr set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
76 * Don't bother with the fixmap if swapper_pg_dir is still mapped
77 * writable in the kernel mapping.
80 WRITE_ONCE(*pgdp, pgd);
86 spin_lock(&swapper_pgdir_lock);
87 fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
88 WRITE_ONCE(*fixmap_pgdp, pgd);
90 * We need dsb(ishst) here to ensure the page-table-walker sees
91 * our new entry before set_p?d() returns. The fixmap's
92 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
95 spin_unlock(&swapper_pgdir_lock);
98 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
99 unsigned long size, pgprot_t vma_prot)
101 if (!pfn_is_map_memory(pfn))
102 return pgprot_noncached(vma_prot);
103 else if (file->f_flags & O_SYNC)
104 return pgprot_writecombine(vma_prot);
107 EXPORT_SYMBOL(phys_mem_access_prot);
109 static phys_addr_t __init early_pgtable_alloc(int shift)
114 phys = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0,
115 MEMBLOCK_ALLOC_NOLEAKTRACE);
117 panic("Failed to allocate page table page\n");
120 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
121 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
122 * any level of table.
124 ptr = pte_set_fixmap(phys);
126 memset(ptr, 0, PAGE_SIZE);
129 * Implicit barriers also ensure the zeroed page is visible to the page
137 bool pgattr_change_is_safe(u64 old, u64 new)
140 * The following mapping attributes may be updated in live
141 * kernel mappings without the need for break-before-make.
143 pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
145 /* creating or taking down mappings is always safe */
146 if (!pte_valid(__pte(old)) || !pte_valid(__pte(new)))
149 /* A live entry's pfn should not change */
150 if (pte_pfn(__pte(old)) != pte_pfn(__pte(new)))
153 /* live contiguous mappings may not be manipulated at all */
154 if ((old | new) & PTE_CONT)
157 /* Transitioning from Non-Global to Global is unsafe */
158 if (old & ~new & PTE_NG)
162 * Changing the memory type between Normal and Normal-Tagged is safe
163 * since Tagged is considered a permission attribute from the
164 * mismatched attribute aliases perspective.
166 if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
167 (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) &&
168 ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
169 (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)))
170 mask |= PTE_ATTRINDX_MASK;
172 return ((old ^ new) & ~mask) == 0;
175 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
176 phys_addr_t phys, pgprot_t prot)
180 ptep = pte_set_fixmap_offset(pmdp, addr);
182 pte_t old_pte = __ptep_get(ptep);
184 __set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
187 * After the PTE entry has been populated once, we
188 * only allow updates to the permission attributes.
190 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
191 pte_val(__ptep_get(ptep))));
194 } while (ptep++, addr += PAGE_SIZE, addr != end);
199 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
200 unsigned long end, phys_addr_t phys,
202 phys_addr_t (*pgtable_alloc)(int),
206 pmd_t pmd = READ_ONCE(*pmdp);
208 BUG_ON(pmd_sect(pmd));
210 pmdval_t pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
211 phys_addr_t pte_phys;
213 if (flags & NO_EXEC_MAPPINGS)
214 pmdval |= PMD_TABLE_PXN;
215 BUG_ON(!pgtable_alloc);
216 pte_phys = pgtable_alloc(PAGE_SHIFT);
217 __pmd_populate(pmdp, pte_phys, pmdval);
218 pmd = READ_ONCE(*pmdp);
220 BUG_ON(pmd_bad(pmd));
223 pgprot_t __prot = prot;
225 next = pte_cont_addr_end(addr, end);
227 /* use a contiguous mapping if the range is suitably aligned */
228 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
229 (flags & NO_CONT_MAPPINGS) == 0)
230 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
232 init_pte(pmdp, addr, next, phys, __prot);
235 } while (addr = next, addr != end);
238 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
239 phys_addr_t phys, pgprot_t prot,
240 phys_addr_t (*pgtable_alloc)(int), int flags)
245 pmdp = pmd_set_fixmap_offset(pudp, addr);
247 pmd_t old_pmd = READ_ONCE(*pmdp);
249 next = pmd_addr_end(addr, end);
251 /* try section mapping first */
252 if (((addr | next | phys) & ~PMD_MASK) == 0 &&
253 (flags & NO_BLOCK_MAPPINGS) == 0) {
254 pmd_set_huge(pmdp, phys, prot);
257 * After the PMD entry has been populated once, we
258 * only allow updates to the permission attributes.
260 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
261 READ_ONCE(pmd_val(*pmdp))));
263 alloc_init_cont_pte(pmdp, addr, next, phys, prot,
264 pgtable_alloc, flags);
266 BUG_ON(pmd_val(old_pmd) != 0 &&
267 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
270 } while (pmdp++, addr = next, addr != end);
275 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
276 unsigned long end, phys_addr_t phys,
278 phys_addr_t (*pgtable_alloc)(int), int flags)
281 pud_t pud = READ_ONCE(*pudp);
284 * Check for initial section mappings in the pgd/pud.
286 BUG_ON(pud_sect(pud));
288 pudval_t pudval = PUD_TYPE_TABLE | PUD_TABLE_UXN;
289 phys_addr_t pmd_phys;
291 if (flags & NO_EXEC_MAPPINGS)
292 pudval |= PUD_TABLE_PXN;
293 BUG_ON(!pgtable_alloc);
294 pmd_phys = pgtable_alloc(PMD_SHIFT);
295 __pud_populate(pudp, pmd_phys, pudval);
296 pud = READ_ONCE(*pudp);
298 BUG_ON(pud_bad(pud));
301 pgprot_t __prot = prot;
303 next = pmd_cont_addr_end(addr, end);
305 /* use a contiguous mapping if the range is suitably aligned */
306 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
307 (flags & NO_CONT_MAPPINGS) == 0)
308 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
310 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
313 } while (addr = next, addr != end);
316 static void alloc_init_pud(p4d_t *p4dp, unsigned long addr, unsigned long end,
317 phys_addr_t phys, pgprot_t prot,
318 phys_addr_t (*pgtable_alloc)(int),
322 p4d_t p4d = READ_ONCE(*p4dp);
326 p4dval_t p4dval = P4D_TYPE_TABLE | P4D_TABLE_UXN;
327 phys_addr_t pud_phys;
329 if (flags & NO_EXEC_MAPPINGS)
330 p4dval |= P4D_TABLE_PXN;
331 BUG_ON(!pgtable_alloc);
332 pud_phys = pgtable_alloc(PUD_SHIFT);
333 __p4d_populate(p4dp, pud_phys, p4dval);
334 p4d = READ_ONCE(*p4dp);
336 BUG_ON(p4d_bad(p4d));
338 pudp = pud_set_fixmap_offset(p4dp, addr);
340 pud_t old_pud = READ_ONCE(*pudp);
342 next = pud_addr_end(addr, end);
345 * For 4K granule only, attempt to put down a 1GB block
347 if (pud_sect_supported() &&
348 ((addr | next | phys) & ~PUD_MASK) == 0 &&
349 (flags & NO_BLOCK_MAPPINGS) == 0) {
350 pud_set_huge(pudp, phys, prot);
353 * After the PUD entry has been populated once, we
354 * only allow updates to the permission attributes.
356 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
357 READ_ONCE(pud_val(*pudp))));
359 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
360 pgtable_alloc, flags);
362 BUG_ON(pud_val(old_pud) != 0 &&
363 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
366 } while (pudp++, addr = next, addr != end);
371 static void alloc_init_p4d(pgd_t *pgdp, unsigned long addr, unsigned long end,
372 phys_addr_t phys, pgprot_t prot,
373 phys_addr_t (*pgtable_alloc)(int),
377 pgd_t pgd = READ_ONCE(*pgdp);
381 pgdval_t pgdval = PGD_TYPE_TABLE | PGD_TABLE_UXN;
382 phys_addr_t p4d_phys;
384 if (flags & NO_EXEC_MAPPINGS)
385 pgdval |= PGD_TABLE_PXN;
386 BUG_ON(!pgtable_alloc);
387 p4d_phys = pgtable_alloc(P4D_SHIFT);
388 __pgd_populate(pgdp, p4d_phys, pgdval);
389 pgd = READ_ONCE(*pgdp);
391 BUG_ON(pgd_bad(pgd));
393 p4dp = p4d_set_fixmap_offset(pgdp, addr);
395 p4d_t old_p4d = READ_ONCE(*p4dp);
397 next = p4d_addr_end(addr, end);
399 alloc_init_pud(p4dp, addr, next, phys, prot,
400 pgtable_alloc, flags);
402 BUG_ON(p4d_val(old_p4d) != 0 &&
403 p4d_val(old_p4d) != READ_ONCE(p4d_val(*p4dp)));
406 } while (p4dp++, addr = next, addr != end);
411 static void __create_pgd_mapping_locked(pgd_t *pgdir, phys_addr_t phys,
412 unsigned long virt, phys_addr_t size,
414 phys_addr_t (*pgtable_alloc)(int),
417 unsigned long addr, end, next;
418 pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
421 * If the virtual and physical address don't have the same offset
422 * within a page, we cannot map the region as the caller expects.
424 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
428 addr = virt & PAGE_MASK;
429 end = PAGE_ALIGN(virt + size);
432 next = pgd_addr_end(addr, end);
433 alloc_init_p4d(pgdp, addr, next, phys, prot, pgtable_alloc,
436 } while (pgdp++, addr = next, addr != end);
439 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
440 unsigned long virt, phys_addr_t size,
442 phys_addr_t (*pgtable_alloc)(int),
445 mutex_lock(&fixmap_lock);
446 __create_pgd_mapping_locked(pgdir, phys, virt, size, prot,
447 pgtable_alloc, flags);
448 mutex_unlock(&fixmap_lock);
451 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
452 extern __alias(__create_pgd_mapping_locked)
453 void create_kpti_ng_temp_pgd(pgd_t *pgdir, phys_addr_t phys, unsigned long virt,
454 phys_addr_t size, pgprot_t prot,
455 phys_addr_t (*pgtable_alloc)(int), int flags);
458 static phys_addr_t __pgd_pgtable_alloc(int shift)
460 void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
463 /* Ensure the zeroed page is visible to the page table walker */
468 static phys_addr_t pgd_pgtable_alloc(int shift)
470 phys_addr_t pa = __pgd_pgtable_alloc(shift);
471 struct ptdesc *ptdesc = page_ptdesc(phys_to_page(pa));
474 * Call proper page table ctor in case later we need to
475 * call core mm functions like apply_to_page_range() on
476 * this pre-allocated page table.
478 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
479 * folded, and if so pagetable_pte_ctor() becomes nop.
481 if (shift == PAGE_SHIFT)
482 BUG_ON(!pagetable_pte_ctor(ptdesc));
483 else if (shift == PMD_SHIFT)
484 BUG_ON(!pagetable_pmd_ctor(ptdesc));
490 * This function can only be used to modify existing table entries,
491 * without allocating new levels of table. Note that this permits the
492 * creation of new section or page entries.
494 void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
495 phys_addr_t size, pgprot_t prot)
497 if (virt < PAGE_OFFSET) {
498 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
502 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
506 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
507 unsigned long virt, phys_addr_t size,
508 pgprot_t prot, bool page_mappings_only)
512 BUG_ON(mm == &init_mm);
514 if (page_mappings_only)
515 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
517 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
518 pgd_pgtable_alloc, flags);
521 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
522 phys_addr_t size, pgprot_t prot)
524 if (virt < PAGE_OFFSET) {
525 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
530 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
533 /* flush the TLBs after updating live kernel mappings */
534 flush_tlb_kernel_range(virt, virt + size);
537 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
538 phys_addr_t end, pgprot_t prot, int flags)
540 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
541 prot, early_pgtable_alloc, flags);
544 void __init mark_linear_text_alias_ro(void)
547 * Remove the write permissions from the linear alias of .text/.rodata
549 update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
550 (unsigned long)__init_begin - (unsigned long)_stext,
556 bool __ro_after_init kfence_early_init = !!CONFIG_KFENCE_SAMPLE_INTERVAL;
558 /* early_param() will be parsed before map_mem() below. */
559 static int __init parse_kfence_early_init(char *arg)
563 if (get_option(&arg, &val))
564 kfence_early_init = !!val;
567 early_param("kfence.sample_interval", parse_kfence_early_init);
569 static phys_addr_t __init arm64_kfence_alloc_pool(void)
571 phys_addr_t kfence_pool;
573 if (!kfence_early_init)
576 kfence_pool = memblock_phys_alloc(KFENCE_POOL_SIZE, PAGE_SIZE);
578 pr_err("failed to allocate kfence pool\n");
579 kfence_early_init = false;
583 /* Temporarily mark as NOMAP. */
584 memblock_mark_nomap(kfence_pool, KFENCE_POOL_SIZE);
589 static void __init arm64_kfence_map_pool(phys_addr_t kfence_pool, pgd_t *pgdp)
594 /* KFENCE pool needs page-level mapping. */
595 __map_memblock(pgdp, kfence_pool, kfence_pool + KFENCE_POOL_SIZE,
596 pgprot_tagged(PAGE_KERNEL),
597 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
598 memblock_clear_nomap(kfence_pool, KFENCE_POOL_SIZE);
599 __kfence_pool = phys_to_virt(kfence_pool);
601 #else /* CONFIG_KFENCE */
603 static inline phys_addr_t arm64_kfence_alloc_pool(void) { return 0; }
604 static inline void arm64_kfence_map_pool(phys_addr_t kfence_pool, pgd_t *pgdp) { }
606 #endif /* CONFIG_KFENCE */
608 static void __init map_mem(pgd_t *pgdp)
610 static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
611 phys_addr_t kernel_start = __pa_symbol(_stext);
612 phys_addr_t kernel_end = __pa_symbol(__init_begin);
613 phys_addr_t start, end;
614 phys_addr_t early_kfence_pool;
615 int flags = NO_EXEC_MAPPINGS;
619 * Setting hierarchical PXNTable attributes on table entries covering
620 * the linear region is only possible if it is guaranteed that no table
621 * entries at any level are being shared between the linear region and
622 * the vmalloc region. Check whether this is true for the PGD level, in
623 * which case it is guaranteed to be true for all other levels as well.
624 * (Unless we are running with support for LPA2, in which case the
625 * entire reduced VA space is covered by a single pgd_t which will have
626 * been populated without the PXNTable attribute by the time we get here.)
628 BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end) &&
629 pgd_index(_PAGE_OFFSET(VA_BITS_MIN)) != PTRS_PER_PGD - 1);
631 early_kfence_pool = arm64_kfence_alloc_pool();
633 if (can_set_direct_map())
634 flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
637 * Take care not to create a writable alias for the
638 * read-only text and rodata sections of the kernel image.
639 * So temporarily mark them as NOMAP to skip mappings in
640 * the following for-loop
642 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
644 /* map all the memory banks */
645 for_each_mem_range(i, &start, &end) {
649 * The linear map must allow allocation tags reading/writing
650 * if MTE is present. Otherwise, it has the same attributes as
653 __map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
658 * Map the linear alias of the [_stext, __init_begin) interval
659 * as non-executable now, and remove the write permission in
660 * mark_linear_text_alias_ro() below (which will be called after
661 * alternative patching has completed). This makes the contents
662 * of the region accessible to subsystems such as hibernate,
663 * but protects it from inadvertent modification or execution.
664 * Note that contiguous mappings cannot be remapped in this way,
665 * so we should avoid them here.
667 __map_memblock(pgdp, kernel_start, kernel_end,
668 PAGE_KERNEL, NO_CONT_MAPPINGS);
669 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
670 arm64_kfence_map_pool(early_kfence_pool, pgdp);
673 void mark_rodata_ro(void)
675 unsigned long section_size;
678 * mark .rodata as read only. Use __init_begin rather than __end_rodata
679 * to cover NOTES and EXCEPTION_TABLE.
681 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
682 WRITE_ONCE(rodata_is_rw, false);
683 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
684 section_size, PAGE_KERNEL_RO);
687 static void __init declare_vma(struct vm_struct *vma,
688 void *va_start, void *va_end,
689 unsigned long vm_flags)
691 phys_addr_t pa_start = __pa_symbol(va_start);
692 unsigned long size = va_end - va_start;
694 BUG_ON(!PAGE_ALIGNED(pa_start));
695 BUG_ON(!PAGE_ALIGNED(size));
697 if (!(vm_flags & VM_NO_GUARD))
700 vma->addr = va_start;
701 vma->phys_addr = pa_start;
703 vma->flags = VM_MAP | vm_flags;
704 vma->caller = __builtin_return_address(0);
706 vm_area_add_early(vma);
709 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
710 static pgprot_t kernel_exec_prot(void)
712 return rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
715 static int __init map_entry_trampoline(void)
719 if (!arm64_kernel_unmapped_at_el0())
722 pgprot_t prot = kernel_exec_prot();
723 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
725 /* The trampoline is always mapped and can therefore be global */
726 pgprot_val(prot) &= ~PTE_NG;
728 /* Map only the text into the trampoline page table */
729 memset(tramp_pg_dir, 0, PGD_SIZE);
730 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
731 entry_tramp_text_size(), prot,
732 __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
734 /* Map both the text and data into the kernel page table */
735 for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
736 __set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
737 pa_start + i * PAGE_SIZE, prot);
739 if (IS_ENABLED(CONFIG_RELOCATABLE))
740 __set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
741 pa_start + i * PAGE_SIZE, PAGE_KERNEL_RO);
745 core_initcall(map_entry_trampoline);
749 * Declare the VMA areas for the kernel
751 static void __init declare_kernel_vmas(void)
753 static struct vm_struct vmlinux_seg[KERNEL_SEGMENT_COUNT];
755 declare_vma(&vmlinux_seg[0], _stext, _etext, VM_NO_GUARD);
756 declare_vma(&vmlinux_seg[1], __start_rodata, __inittext_begin, VM_NO_GUARD);
757 declare_vma(&vmlinux_seg[2], __inittext_begin, __inittext_end, VM_NO_GUARD);
758 declare_vma(&vmlinux_seg[3], __initdata_begin, __initdata_end, VM_NO_GUARD);
759 declare_vma(&vmlinux_seg[4], _data, _end, 0);
762 void __pi_map_range(u64 *pgd, u64 start, u64 end, u64 pa, pgprot_t prot,
763 int level, pte_t *tbl, bool may_use_cont, u64 va_offset);
765 static u8 idmap_ptes[IDMAP_LEVELS - 1][PAGE_SIZE] __aligned(PAGE_SIZE) __ro_after_init,
766 kpti_ptes[IDMAP_LEVELS - 1][PAGE_SIZE] __aligned(PAGE_SIZE) __ro_after_init;
768 static void __init create_idmap(void)
770 u64 start = __pa_symbol(__idmap_text_start);
771 u64 end = __pa_symbol(__idmap_text_end);
772 u64 ptep = __pa_symbol(idmap_ptes);
774 __pi_map_range(&ptep, start, end, start, PAGE_KERNEL_ROX,
775 IDMAP_ROOT_LEVEL, (pte_t *)idmap_pg_dir, false,
776 __phys_to_virt(ptep) - ptep);
778 if (IS_ENABLED(CONFIG_UNMAP_KERNEL_AT_EL0) && !arm64_use_ng_mappings) {
779 extern u32 __idmap_kpti_flag;
780 u64 pa = __pa_symbol(&__idmap_kpti_flag);
783 * The KPTI G-to-nG conversion code needs a read-write mapping
784 * of its synchronization flag in the ID map.
786 ptep = __pa_symbol(kpti_ptes);
787 __pi_map_range(&ptep, pa, pa + sizeof(u32), pa, PAGE_KERNEL,
788 IDMAP_ROOT_LEVEL, (pte_t *)idmap_pg_dir, false,
789 __phys_to_virt(ptep) - ptep);
793 void __init paging_init(void)
795 map_mem(swapper_pg_dir);
797 memblock_allow_resize();
800 declare_kernel_vmas();
803 #ifdef CONFIG_MEMORY_HOTPLUG
804 static void free_hotplug_page_range(struct page *page, size_t size,
805 struct vmem_altmap *altmap)
808 vmem_altmap_free(altmap, size >> PAGE_SHIFT);
810 WARN_ON(PageReserved(page));
811 free_pages((unsigned long)page_address(page), get_order(size));
815 static void free_hotplug_pgtable_page(struct page *page)
817 free_hotplug_page_range(page, PAGE_SIZE, NULL);
820 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
821 unsigned long floor, unsigned long ceiling,
834 if (end - 1 > ceiling - 1)
839 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
840 unsigned long end, bool free_mapped,
841 struct vmem_altmap *altmap)
846 ptep = pte_offset_kernel(pmdp, addr);
847 pte = __ptep_get(ptep);
851 WARN_ON(!pte_present(pte));
852 __pte_clear(&init_mm, addr, ptep);
853 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
855 free_hotplug_page_range(pte_page(pte),
857 } while (addr += PAGE_SIZE, addr < end);
860 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
861 unsigned long end, bool free_mapped,
862 struct vmem_altmap *altmap)
868 next = pmd_addr_end(addr, end);
869 pmdp = pmd_offset(pudp, addr);
870 pmd = READ_ONCE(*pmdp);
874 WARN_ON(!pmd_present(pmd));
879 * One TLBI should be sufficient here as the PMD_SIZE
880 * range is mapped with a single block entry.
882 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
884 free_hotplug_page_range(pmd_page(pmd),
888 WARN_ON(!pmd_table(pmd));
889 unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap);
890 } while (addr = next, addr < end);
893 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
894 unsigned long end, bool free_mapped,
895 struct vmem_altmap *altmap)
901 next = pud_addr_end(addr, end);
902 pudp = pud_offset(p4dp, addr);
903 pud = READ_ONCE(*pudp);
907 WARN_ON(!pud_present(pud));
912 * One TLBI should be sufficient here as the PUD_SIZE
913 * range is mapped with a single block entry.
915 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
917 free_hotplug_page_range(pud_page(pud),
921 WARN_ON(!pud_table(pud));
922 unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap);
923 } while (addr = next, addr < end);
926 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
927 unsigned long end, bool free_mapped,
928 struct vmem_altmap *altmap)
934 next = p4d_addr_end(addr, end);
935 p4dp = p4d_offset(pgdp, addr);
936 p4d = READ_ONCE(*p4dp);
940 WARN_ON(!p4d_present(p4d));
941 unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap);
942 } while (addr = next, addr < end);
945 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
946 bool free_mapped, struct vmem_altmap *altmap)
952 * altmap can only be used as vmemmap mapping backing memory.
953 * In case the backing memory itself is not being freed, then
954 * altmap is irrelevant. Warn about this inconsistency when
957 WARN_ON(!free_mapped && altmap);
960 next = pgd_addr_end(addr, end);
961 pgdp = pgd_offset_k(addr);
962 pgd = READ_ONCE(*pgdp);
966 WARN_ON(!pgd_present(pgd));
967 unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap);
968 } while (addr = next, addr < end);
971 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
972 unsigned long end, unsigned long floor,
973 unsigned long ceiling)
976 unsigned long i, start = addr;
979 ptep = pte_offset_kernel(pmdp, addr);
980 pte = __ptep_get(ptep);
983 * This is just a sanity check here which verifies that
984 * pte clearing has been done by earlier unmap loops.
986 WARN_ON(!pte_none(pte));
987 } while (addr += PAGE_SIZE, addr < end);
989 if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
993 * Check whether we can free the pte page if the rest of the
994 * entries are empty. Overlap with other regions have been
995 * handled by the floor/ceiling check.
997 ptep = pte_offset_kernel(pmdp, 0UL);
998 for (i = 0; i < PTRS_PER_PTE; i++) {
999 if (!pte_none(__ptep_get(&ptep[i])))
1004 __flush_tlb_kernel_pgtable(start);
1005 free_hotplug_pgtable_page(virt_to_page(ptep));
1008 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
1009 unsigned long end, unsigned long floor,
1010 unsigned long ceiling)
1013 unsigned long i, next, start = addr;
1016 next = pmd_addr_end(addr, end);
1017 pmdp = pmd_offset(pudp, addr);
1018 pmd = READ_ONCE(*pmdp);
1022 WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
1023 free_empty_pte_table(pmdp, addr, next, floor, ceiling);
1024 } while (addr = next, addr < end);
1026 if (CONFIG_PGTABLE_LEVELS <= 2)
1029 if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
1033 * Check whether we can free the pmd page if the rest of the
1034 * entries are empty. Overlap with other regions have been
1035 * handled by the floor/ceiling check.
1037 pmdp = pmd_offset(pudp, 0UL);
1038 for (i = 0; i < PTRS_PER_PMD; i++) {
1039 if (!pmd_none(READ_ONCE(pmdp[i])))
1044 __flush_tlb_kernel_pgtable(start);
1045 free_hotplug_pgtable_page(virt_to_page(pmdp));
1048 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
1049 unsigned long end, unsigned long floor,
1050 unsigned long ceiling)
1053 unsigned long i, next, start = addr;
1056 next = pud_addr_end(addr, end);
1057 pudp = pud_offset(p4dp, addr);
1058 pud = READ_ONCE(*pudp);
1062 WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1063 free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1064 } while (addr = next, addr < end);
1066 if (!pgtable_l4_enabled())
1069 if (!pgtable_range_aligned(start, end, floor, ceiling, P4D_MASK))
1073 * Check whether we can free the pud page if the rest of the
1074 * entries are empty. Overlap with other regions have been
1075 * handled by the floor/ceiling check.
1077 pudp = pud_offset(p4dp, 0UL);
1078 for (i = 0; i < PTRS_PER_PUD; i++) {
1079 if (!pud_none(READ_ONCE(pudp[i])))
1084 __flush_tlb_kernel_pgtable(start);
1085 free_hotplug_pgtable_page(virt_to_page(pudp));
1088 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1089 unsigned long end, unsigned long floor,
1090 unsigned long ceiling)
1093 unsigned long i, next, start = addr;
1096 next = p4d_addr_end(addr, end);
1097 p4dp = p4d_offset(pgdp, addr);
1098 p4d = READ_ONCE(*p4dp);
1102 WARN_ON(!p4d_present(p4d));
1103 free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1104 } while (addr = next, addr < end);
1106 if (!pgtable_l5_enabled())
1109 if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1113 * Check whether we can free the p4d page if the rest of the
1114 * entries are empty. Overlap with other regions have been
1115 * handled by the floor/ceiling check.
1117 p4dp = p4d_offset(pgdp, 0UL);
1118 for (i = 0; i < PTRS_PER_P4D; i++) {
1119 if (!p4d_none(READ_ONCE(p4dp[i])))
1124 __flush_tlb_kernel_pgtable(start);
1125 free_hotplug_pgtable_page(virt_to_page(p4dp));
1128 static void free_empty_tables(unsigned long addr, unsigned long end,
1129 unsigned long floor, unsigned long ceiling)
1135 next = pgd_addr_end(addr, end);
1136 pgdp = pgd_offset_k(addr);
1137 pgd = READ_ONCE(*pgdp);
1141 WARN_ON(!pgd_present(pgd));
1142 free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1143 } while (addr = next, addr < end);
1147 void __meminit vmemmap_set_pmd(pmd_t *pmdp, void *p, int node,
1148 unsigned long addr, unsigned long next)
1150 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1153 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1154 unsigned long addr, unsigned long next)
1156 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1160 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1161 struct vmem_altmap *altmap)
1163 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1165 if (!IS_ENABLED(CONFIG_ARM64_4K_PAGES))
1166 return vmemmap_populate_basepages(start, end, node, altmap);
1168 return vmemmap_populate_hugepages(start, end, node, altmap);
1171 #ifdef CONFIG_MEMORY_HOTPLUG
1172 void vmemmap_free(unsigned long start, unsigned long end,
1173 struct vmem_altmap *altmap)
1175 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1177 unmap_hotplug_range(start, end, true, altmap);
1178 free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1180 #endif /* CONFIG_MEMORY_HOTPLUG */
1182 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1184 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1186 /* Only allow permission changes for now */
1187 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1191 VM_BUG_ON(phys & ~PUD_MASK);
1192 set_pud(pudp, new_pud);
1196 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1198 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1200 /* Only allow permission changes for now */
1201 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1205 VM_BUG_ON(phys & ~PMD_MASK);
1206 set_pmd(pmdp, new_pmd);
1210 #ifndef __PAGETABLE_P4D_FOLDED
1211 void p4d_clear_huge(p4d_t *p4dp)
1216 int pud_clear_huge(pud_t *pudp)
1218 if (!pud_sect(READ_ONCE(*pudp)))
1224 int pmd_clear_huge(pmd_t *pmdp)
1226 if (!pmd_sect(READ_ONCE(*pmdp)))
1232 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1237 pmd = READ_ONCE(*pmdp);
1239 if (!pmd_table(pmd)) {
1244 table = pte_offset_kernel(pmdp, addr);
1246 __flush_tlb_kernel_pgtable(addr);
1247 pte_free_kernel(NULL, table);
1251 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1256 unsigned long next, end;
1258 pud = READ_ONCE(*pudp);
1260 if (!pud_table(pud)) {
1265 table = pmd_offset(pudp, addr);
1268 end = addr + PUD_SIZE;
1270 pmd_free_pte_page(pmdp, next);
1271 } while (pmdp++, next += PMD_SIZE, next != end);
1274 __flush_tlb_kernel_pgtable(addr);
1275 pmd_free(NULL, table);
1279 #ifdef CONFIG_MEMORY_HOTPLUG
1280 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1282 unsigned long end = start + size;
1284 WARN_ON(pgdir != init_mm.pgd);
1285 WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1287 unmap_hotplug_range(start, end, false, NULL);
1288 free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1291 struct range arch_get_mappable_range(void)
1293 struct range mhp_range;
1294 u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1295 u64 end_linear_pa = __pa(PAGE_END - 1);
1297 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1299 * Check for a wrap, it is possible because of randomized linear
1300 * mapping the start physical address is actually bigger than
1301 * the end physical address. In this case set start to zero
1302 * because [0, end_linear_pa] range must still be able to cover
1303 * all addressable physical addresses.
1305 if (start_linear_pa > end_linear_pa)
1306 start_linear_pa = 0;
1309 WARN_ON(start_linear_pa > end_linear_pa);
1312 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1313 * accommodating both its ends but excluding PAGE_END. Max physical
1314 * range which can be mapped inside this linear mapping range, must
1315 * also be derived from its end points.
1317 mhp_range.start = start_linear_pa;
1318 mhp_range.end = end_linear_pa;
1323 int arch_add_memory(int nid, u64 start, u64 size,
1324 struct mhp_params *params)
1326 int ret, flags = NO_EXEC_MAPPINGS;
1328 VM_BUG_ON(!mhp_range_allowed(start, size, true));
1330 if (can_set_direct_map())
1331 flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1333 __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1334 size, params->pgprot, __pgd_pgtable_alloc,
1337 memblock_clear_nomap(start, size);
1339 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1342 __remove_pgd_mapping(swapper_pg_dir,
1343 __phys_to_virt(start), size);
1345 max_pfn = PFN_UP(start + size);
1346 max_low_pfn = max_pfn;
1352 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1354 unsigned long start_pfn = start >> PAGE_SHIFT;
1355 unsigned long nr_pages = size >> PAGE_SHIFT;
1357 __remove_pages(start_pfn, nr_pages, altmap);
1358 __remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1362 * This memory hotplug notifier helps prevent boot memory from being
1363 * inadvertently removed as it blocks pfn range offlining process in
1364 * __offline_pages(). Hence this prevents both offlining as well as
1365 * removal process for boot memory which is initially always online.
1366 * In future if and when boot memory could be removed, this notifier
1367 * should be dropped and free_hotplug_page_range() should handle any
1368 * reserved pages allocated during boot.
1370 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1371 unsigned long action, void *data)
1373 struct mem_section *ms;
1374 struct memory_notify *arg = data;
1375 unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1376 unsigned long pfn = arg->start_pfn;
1378 if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1381 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1382 unsigned long start = PFN_PHYS(pfn);
1383 unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1385 ms = __pfn_to_section(pfn);
1386 if (!early_section(ms))
1389 if (action == MEM_GOING_OFFLINE) {
1391 * Boot memory removal is not supported. Prevent
1392 * it via blocking any attempted offline request
1393 * for the boot memory and just report it.
1395 pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1397 } else if (action == MEM_OFFLINE) {
1399 * This should have never happened. Boot memory
1400 * offlining should have been prevented by this
1401 * very notifier. Probably some memory removal
1402 * procedure might have changed which would then
1403 * require further debug.
1405 pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1408 * Core memory hotplug does not process a return
1409 * code from the notifier for MEM_OFFLINE events.
1410 * The error condition has been reported. Return
1411 * from here as if ignored.
1419 static struct notifier_block prevent_bootmem_remove_nb = {
1420 .notifier_call = prevent_bootmem_remove_notifier,
1424 * This ensures that boot memory sections on the platform are online
1425 * from early boot. Memory sections could not be prevented from being
1426 * offlined, unless for some reason they are not online to begin with.
1427 * This helps validate the basic assumption on which the above memory
1428 * event notifier works to prevent boot memory section offlining and
1429 * its possible removal.
1431 static void validate_bootmem_online(void)
1433 phys_addr_t start, end, addr;
1434 struct mem_section *ms;
1438 * Scanning across all memblock might be expensive
1439 * on some big memory systems. Hence enable this
1440 * validation only with DEBUG_VM.
1442 if (!IS_ENABLED(CONFIG_DEBUG_VM))
1445 for_each_mem_range(i, &start, &end) {
1446 for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1447 ms = __pfn_to_section(PHYS_PFN(addr));
1450 * All memory ranges in the system at this point
1451 * should have been marked as early sections.
1453 WARN_ON(!early_section(ms));
1456 * Memory notifier mechanism here to prevent boot
1457 * memory offlining depends on the fact that each
1458 * early section memory on the system is initially
1459 * online. Otherwise a given memory section which
1460 * is already offline will be overlooked and can
1461 * be removed completely. Call out such sections.
1463 if (!online_section(ms))
1464 pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1465 addr, addr + (1UL << PA_SECTION_SHIFT));
1470 static int __init prevent_bootmem_remove_init(void)
1474 if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1477 validate_bootmem_online();
1478 ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1480 pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1484 early_initcall(prevent_bootmem_remove_init);
1487 pte_t ptep_modify_prot_start(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
1489 if (alternative_has_cap_unlikely(ARM64_WORKAROUND_2645198)) {
1491 * Break-before-make (BBM) is required for all user space mappings
1492 * when the permission changes from executable to non-executable
1493 * in cases where cpu is affected with errata #2645198.
1495 if (pte_user_exec(ptep_get(ptep)))
1496 return ptep_clear_flush(vma, addr, ptep);
1498 return ptep_get_and_clear(vma->vm_mm, addr, ptep);
1501 void ptep_modify_prot_commit(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep,
1502 pte_t old_pte, pte_t pte)
1504 set_pte_at(vma->vm_mm, addr, ptep, pte);
1508 * Atomically replaces the active TTBR1_EL1 PGD with a new VA-compatible PGD,
1509 * avoiding the possibility of conflicting TLB entries being allocated.
1511 void __cpu_replace_ttbr1(pgd_t *pgdp, bool cnp)
1513 typedef void (ttbr_replace_func)(phys_addr_t);
1514 extern ttbr_replace_func idmap_cpu_replace_ttbr1;
1515 ttbr_replace_func *replace_phys;
1518 /* phys_to_ttbr() zeros lower 2 bits of ttbr with 52-bit PA */
1519 phys_addr_t ttbr1 = phys_to_ttbr(virt_to_phys(pgdp));
1522 ttbr1 |= TTBR_CNP_BIT;
1524 replace_phys = (void *)__pa_symbol(idmap_cpu_replace_ttbr1);
1526 cpu_install_idmap();
1529 * We really don't want to take *any* exceptions while TTBR1 is
1530 * in the process of being replaced so mask everything.
1532 daif = local_daif_save();
1533 replace_phys(ttbr1);
1534 local_daif_restore(daif);
1536 cpu_uninstall_idmap();