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>
28 #include <asm/barrier.h>
29 #include <asm/cputype.h>
30 #include <asm/fixmap.h>
31 #include <asm/kasan.h>
32 #include <asm/kernel-pgtable.h>
33 #include <asm/sections.h>
34 #include <asm/setup.h>
35 #include <linux/sizes.h>
37 #include <asm/mmu_context.h>
38 #include <asm/ptdump.h>
39 #include <asm/tlbflush.h>
40 #include <asm/pgalloc.h>
42 #define NO_BLOCK_MAPPINGS BIT(0)
43 #define NO_CONT_MAPPINGS BIT(1)
44 #define NO_EXEC_MAPPINGS BIT(2) /* assumes FEAT_HPDS is not used */
46 int idmap_t0sz __ro_after_init;
49 u64 vabits_actual __ro_after_init = VA_BITS_MIN;
50 EXPORT_SYMBOL(vabits_actual);
53 u64 kimage_vaddr __ro_after_init = (u64)&_text;
54 EXPORT_SYMBOL(kimage_vaddr);
56 u64 kimage_voffset __ro_after_init;
57 EXPORT_SYMBOL(kimage_voffset);
59 u32 __boot_cpu_mode[] = { BOOT_CPU_MODE_EL2, BOOT_CPU_MODE_EL1 };
62 * The booting CPU updates the failed status @__early_cpu_boot_status,
63 * with MMU turned off.
65 long __section(".mmuoff.data.write") __early_cpu_boot_status;
68 * Empty_zero_page is a special page that is used for zero-initialized data
71 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss;
72 EXPORT_SYMBOL(empty_zero_page);
74 static pte_t bm_pte[PTRS_PER_PTE] __page_aligned_bss;
75 static pmd_t bm_pmd[PTRS_PER_PMD] __page_aligned_bss __maybe_unused;
76 static pud_t bm_pud[PTRS_PER_PUD] __page_aligned_bss __maybe_unused;
78 static DEFINE_SPINLOCK(swapper_pgdir_lock);
79 static DEFINE_MUTEX(fixmap_lock);
81 void set_swapper_pgd(pgd_t *pgdp, pgd_t pgd)
85 spin_lock(&swapper_pgdir_lock);
86 fixmap_pgdp = pgd_set_fixmap(__pa_symbol(pgdp));
87 WRITE_ONCE(*fixmap_pgdp, pgd);
89 * We need dsb(ishst) here to ensure the page-table-walker sees
90 * our new entry before set_p?d() returns. The fixmap's
91 * flush_tlb_kernel_range() via clear_fixmap() does this for us.
94 spin_unlock(&swapper_pgdir_lock);
97 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
98 unsigned long size, pgprot_t vma_prot)
100 if (!pfn_is_map_memory(pfn))
101 return pgprot_noncached(vma_prot);
102 else if (file->f_flags & O_SYNC)
103 return pgprot_writecombine(vma_prot);
106 EXPORT_SYMBOL(phys_mem_access_prot);
108 static phys_addr_t __init early_pgtable_alloc(int shift)
113 phys = memblock_phys_alloc_range(PAGE_SIZE, PAGE_SIZE, 0,
114 MEMBLOCK_ALLOC_NOLEAKTRACE);
116 panic("Failed to allocate page table page\n");
119 * The FIX_{PGD,PUD,PMD} slots may be in active use, but the FIX_PTE
120 * slot will be free, so we can (ab)use the FIX_PTE slot to initialise
121 * any level of table.
123 ptr = pte_set_fixmap(phys);
125 memset(ptr, 0, PAGE_SIZE);
128 * Implicit barriers also ensure the zeroed page is visible to the page
136 static bool pgattr_change_is_safe(u64 old, u64 new)
139 * The following mapping attributes may be updated in live
140 * kernel mappings without the need for break-before-make.
142 pteval_t mask = PTE_PXN | PTE_RDONLY | PTE_WRITE | PTE_NG;
144 /* creating or taking down mappings is always safe */
145 if (old == 0 || new == 0)
148 /* live contiguous mappings may not be manipulated at all */
149 if ((old | new) & PTE_CONT)
152 /* Transitioning from Non-Global to Global is unsafe */
153 if (old & ~new & PTE_NG)
157 * Changing the memory type between Normal and Normal-Tagged is safe
158 * since Tagged is considered a permission attribute from the
159 * mismatched attribute aliases perspective.
161 if (((old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
162 (old & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)) &&
163 ((new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL) ||
164 (new & PTE_ATTRINDX_MASK) == PTE_ATTRINDX(MT_NORMAL_TAGGED)))
165 mask |= PTE_ATTRINDX_MASK;
167 return ((old ^ new) & ~mask) == 0;
170 static void init_pte(pmd_t *pmdp, unsigned long addr, unsigned long end,
171 phys_addr_t phys, pgprot_t prot)
175 ptep = pte_set_fixmap_offset(pmdp, addr);
177 pte_t old_pte = READ_ONCE(*ptep);
179 set_pte(ptep, pfn_pte(__phys_to_pfn(phys), prot));
182 * After the PTE entry has been populated once, we
183 * only allow updates to the permission attributes.
185 BUG_ON(!pgattr_change_is_safe(pte_val(old_pte),
186 READ_ONCE(pte_val(*ptep))));
189 } while (ptep++, addr += PAGE_SIZE, addr != end);
194 static void alloc_init_cont_pte(pmd_t *pmdp, unsigned long addr,
195 unsigned long end, phys_addr_t phys,
197 phys_addr_t (*pgtable_alloc)(int),
201 pmd_t pmd = READ_ONCE(*pmdp);
203 BUG_ON(pmd_sect(pmd));
205 pmdval_t pmdval = PMD_TYPE_TABLE | PMD_TABLE_UXN;
206 phys_addr_t pte_phys;
208 if (flags & NO_EXEC_MAPPINGS)
209 pmdval |= PMD_TABLE_PXN;
210 BUG_ON(!pgtable_alloc);
211 pte_phys = pgtable_alloc(PAGE_SHIFT);
212 __pmd_populate(pmdp, pte_phys, pmdval);
213 pmd = READ_ONCE(*pmdp);
215 BUG_ON(pmd_bad(pmd));
218 pgprot_t __prot = prot;
220 next = pte_cont_addr_end(addr, end);
222 /* use a contiguous mapping if the range is suitably aligned */
223 if ((((addr | next | phys) & ~CONT_PTE_MASK) == 0) &&
224 (flags & NO_CONT_MAPPINGS) == 0)
225 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
227 init_pte(pmdp, addr, next, phys, __prot);
230 } while (addr = next, addr != end);
233 static void init_pmd(pud_t *pudp, unsigned long addr, unsigned long end,
234 phys_addr_t phys, pgprot_t prot,
235 phys_addr_t (*pgtable_alloc)(int), int flags)
240 pmdp = pmd_set_fixmap_offset(pudp, addr);
242 pmd_t old_pmd = READ_ONCE(*pmdp);
244 next = pmd_addr_end(addr, end);
246 /* try section mapping first */
247 if (((addr | next | phys) & ~PMD_MASK) == 0 &&
248 (flags & NO_BLOCK_MAPPINGS) == 0) {
249 pmd_set_huge(pmdp, phys, prot);
252 * After the PMD entry has been populated once, we
253 * only allow updates to the permission attributes.
255 BUG_ON(!pgattr_change_is_safe(pmd_val(old_pmd),
256 READ_ONCE(pmd_val(*pmdp))));
258 alloc_init_cont_pte(pmdp, addr, next, phys, prot,
259 pgtable_alloc, flags);
261 BUG_ON(pmd_val(old_pmd) != 0 &&
262 pmd_val(old_pmd) != READ_ONCE(pmd_val(*pmdp)));
265 } while (pmdp++, addr = next, addr != end);
270 static void alloc_init_cont_pmd(pud_t *pudp, unsigned long addr,
271 unsigned long end, phys_addr_t phys,
273 phys_addr_t (*pgtable_alloc)(int), int flags)
276 pud_t pud = READ_ONCE(*pudp);
279 * Check for initial section mappings in the pgd/pud.
281 BUG_ON(pud_sect(pud));
283 pudval_t pudval = PUD_TYPE_TABLE | PUD_TABLE_UXN;
284 phys_addr_t pmd_phys;
286 if (flags & NO_EXEC_MAPPINGS)
287 pudval |= PUD_TABLE_PXN;
288 BUG_ON(!pgtable_alloc);
289 pmd_phys = pgtable_alloc(PMD_SHIFT);
290 __pud_populate(pudp, pmd_phys, pudval);
291 pud = READ_ONCE(*pudp);
293 BUG_ON(pud_bad(pud));
296 pgprot_t __prot = prot;
298 next = pmd_cont_addr_end(addr, end);
300 /* use a contiguous mapping if the range is suitably aligned */
301 if ((((addr | next | phys) & ~CONT_PMD_MASK) == 0) &&
302 (flags & NO_CONT_MAPPINGS) == 0)
303 __prot = __pgprot(pgprot_val(prot) | PTE_CONT);
305 init_pmd(pudp, addr, next, phys, __prot, pgtable_alloc, flags);
308 } while (addr = next, addr != end);
311 static void alloc_init_pud(pgd_t *pgdp, unsigned long addr, unsigned long end,
312 phys_addr_t phys, pgprot_t prot,
313 phys_addr_t (*pgtable_alloc)(int),
318 p4d_t *p4dp = p4d_offset(pgdp, addr);
319 p4d_t p4d = READ_ONCE(*p4dp);
322 p4dval_t p4dval = P4D_TYPE_TABLE | P4D_TABLE_UXN;
323 phys_addr_t pud_phys;
325 if (flags & NO_EXEC_MAPPINGS)
326 p4dval |= P4D_TABLE_PXN;
327 BUG_ON(!pgtable_alloc);
328 pud_phys = pgtable_alloc(PUD_SHIFT);
329 __p4d_populate(p4dp, pud_phys, p4dval);
330 p4d = READ_ONCE(*p4dp);
332 BUG_ON(p4d_bad(p4d));
335 * No need for locking during early boot. And it doesn't work as
336 * expected with KASLR enabled.
338 if (system_state != SYSTEM_BOOTING)
339 mutex_lock(&fixmap_lock);
340 pudp = pud_set_fixmap_offset(p4dp, addr);
342 pud_t old_pud = READ_ONCE(*pudp);
344 next = pud_addr_end(addr, end);
347 * For 4K granule only, attempt to put down a 1GB block
349 if (pud_sect_supported() &&
350 ((addr | next | phys) & ~PUD_MASK) == 0 &&
351 (flags & NO_BLOCK_MAPPINGS) == 0) {
352 pud_set_huge(pudp, phys, prot);
355 * After the PUD entry has been populated once, we
356 * only allow updates to the permission attributes.
358 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
359 READ_ONCE(pud_val(*pudp))));
361 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
362 pgtable_alloc, flags);
364 BUG_ON(pud_val(old_pud) != 0 &&
365 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
368 } while (pudp++, addr = next, addr != end);
371 if (system_state != SYSTEM_BOOTING)
372 mutex_unlock(&fixmap_lock);
375 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
376 unsigned long virt, phys_addr_t size,
378 phys_addr_t (*pgtable_alloc)(int),
381 unsigned long addr, end, next;
382 pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
385 * If the virtual and physical address don't have the same offset
386 * within a page, we cannot map the region as the caller expects.
388 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
392 addr = virt & PAGE_MASK;
393 end = PAGE_ALIGN(virt + size);
396 next = pgd_addr_end(addr, end);
397 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
400 } while (pgdp++, addr = next, addr != end);
403 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
404 extern __alias(__create_pgd_mapping)
405 void create_kpti_ng_temp_pgd(pgd_t *pgdir, phys_addr_t phys, unsigned long virt,
406 phys_addr_t size, pgprot_t prot,
407 phys_addr_t (*pgtable_alloc)(int), int flags);
410 static phys_addr_t __pgd_pgtable_alloc(int shift)
412 void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
415 /* Ensure the zeroed page is visible to the page table walker */
420 static phys_addr_t pgd_pgtable_alloc(int shift)
422 phys_addr_t pa = __pgd_pgtable_alloc(shift);
425 * Call proper page table ctor in case later we need to
426 * call core mm functions like apply_to_page_range() on
427 * this pre-allocated page table.
429 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
430 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
432 if (shift == PAGE_SHIFT)
433 BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa)));
434 else if (shift == PMD_SHIFT)
435 BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
441 * This function can only be used to modify existing table entries,
442 * without allocating new levels of table. Note that this permits the
443 * creation of new section or page entries.
445 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
446 phys_addr_t size, pgprot_t prot)
448 if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
449 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
453 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
457 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
458 unsigned long virt, phys_addr_t size,
459 pgprot_t prot, bool page_mappings_only)
463 BUG_ON(mm == &init_mm);
465 if (page_mappings_only)
466 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
468 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
469 pgd_pgtable_alloc, flags);
472 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
473 phys_addr_t size, pgprot_t prot)
475 if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
476 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
481 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
484 /* flush the TLBs after updating live kernel mappings */
485 flush_tlb_kernel_range(virt, virt + size);
488 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
489 phys_addr_t end, pgprot_t prot, int flags)
491 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
492 prot, early_pgtable_alloc, flags);
495 void __init mark_linear_text_alias_ro(void)
498 * Remove the write permissions from the linear alias of .text/.rodata
500 update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
501 (unsigned long)__init_begin - (unsigned long)_stext,
505 static bool crash_mem_map __initdata;
507 static int __init enable_crash_mem_map(char *arg)
510 * Proper parameter parsing is done by reserve_crashkernel(). We only
511 * need to know if the linear map has to avoid block mappings so that
512 * the crashkernel reservations can be unmapped later.
514 crash_mem_map = true;
518 early_param("crashkernel", enable_crash_mem_map);
520 static void __init map_mem(pgd_t *pgdp)
522 static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
523 phys_addr_t kernel_start = __pa_symbol(_stext);
524 phys_addr_t kernel_end = __pa_symbol(__init_begin);
525 phys_addr_t start, end;
526 int flags = NO_EXEC_MAPPINGS;
530 * Setting hierarchical PXNTable attributes on table entries covering
531 * the linear region is only possible if it is guaranteed that no table
532 * entries at any level are being shared between the linear region and
533 * the vmalloc region. Check whether this is true for the PGD level, in
534 * which case it is guaranteed to be true for all other levels as well.
536 BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));
538 if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
539 flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
542 * Take care not to create a writable alias for the
543 * read-only text and rodata sections of the kernel image.
544 * So temporarily mark them as NOMAP to skip mappings in
545 * the following for-loop
547 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
549 #ifdef CONFIG_KEXEC_CORE
551 if (defer_reserve_crashkernel())
552 flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
553 else if (crashk_res.end)
554 memblock_mark_nomap(crashk_res.start,
555 resource_size(&crashk_res));
559 /* map all the memory banks */
560 for_each_mem_range(i, &start, &end) {
564 * The linear map must allow allocation tags reading/writing
565 * if MTE is present. Otherwise, it has the same attributes as
568 __map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
573 * Map the linear alias of the [_stext, __init_begin) interval
574 * as non-executable now, and remove the write permission in
575 * mark_linear_text_alias_ro() below (which will be called after
576 * alternative patching has completed). This makes the contents
577 * of the region accessible to subsystems such as hibernate,
578 * but protects it from inadvertent modification or execution.
579 * Note that contiguous mappings cannot be remapped in this way,
580 * so we should avoid them here.
582 __map_memblock(pgdp, kernel_start, kernel_end,
583 PAGE_KERNEL, NO_CONT_MAPPINGS);
584 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
587 * Use page-level mappings here so that we can shrink the region
588 * in page granularity and put back unused memory to buddy system
589 * through /sys/kernel/kexec_crash_size interface.
591 #ifdef CONFIG_KEXEC_CORE
592 if (crash_mem_map && !defer_reserve_crashkernel()) {
593 if (crashk_res.end) {
594 __map_memblock(pgdp, crashk_res.start,
597 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
598 memblock_clear_nomap(crashk_res.start,
599 resource_size(&crashk_res));
605 void mark_rodata_ro(void)
607 unsigned long section_size;
610 * mark .rodata as read only. Use __init_begin rather than __end_rodata
611 * to cover NOTES and EXCEPTION_TABLE.
613 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
614 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
615 section_size, PAGE_KERNEL_RO);
620 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
621 pgprot_t prot, struct vm_struct *vma,
622 int flags, unsigned long vm_flags)
624 phys_addr_t pa_start = __pa_symbol(va_start);
625 unsigned long size = va_end - va_start;
627 BUG_ON(!PAGE_ALIGNED(pa_start));
628 BUG_ON(!PAGE_ALIGNED(size));
630 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
631 early_pgtable_alloc, flags);
633 if (!(vm_flags & VM_NO_GUARD))
636 vma->addr = va_start;
637 vma->phys_addr = pa_start;
639 vma->flags = VM_MAP | vm_flags;
640 vma->caller = __builtin_return_address(0);
642 vm_area_add_early(vma);
645 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
646 static int __init map_entry_trampoline(void)
650 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
651 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
653 /* The trampoline is always mapped and can therefore be global */
654 pgprot_val(prot) &= ~PTE_NG;
656 /* Map only the text into the trampoline page table */
657 memset(tramp_pg_dir, 0, PGD_SIZE);
658 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
659 entry_tramp_text_size(), prot,
660 __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
662 /* Map both the text and data into the kernel page table */
663 for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
664 __set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
665 pa_start + i * PAGE_SIZE, prot);
667 if (IS_ENABLED(CONFIG_RELOCATABLE))
668 __set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
669 pa_start + i * PAGE_SIZE, PAGE_KERNEL_RO);
673 core_initcall(map_entry_trampoline);
677 * Open coded check for BTI, only for use to determine configuration
678 * for early mappings for before the cpufeature code has run.
680 static bool arm64_early_this_cpu_has_bti(void)
684 if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
687 pfr1 = __read_sysreg_by_encoding(SYS_ID_AA64PFR1_EL1);
688 return cpuid_feature_extract_unsigned_field(pfr1,
689 ID_AA64PFR1_EL1_BT_SHIFT);
693 * Create fine-grained mappings for the kernel.
695 static void __init map_kernel(pgd_t *pgdp)
697 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
698 vmlinux_initdata, vmlinux_data;
701 * External debuggers may need to write directly to the text
702 * mapping to install SW breakpoints. Allow this (only) when
703 * explicitly requested with rodata=off.
705 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
708 * If we have a CPU that supports BTI and a kernel built for
709 * BTI then mark the kernel executable text as guarded pages
710 * now so we don't have to rewrite the page tables later.
712 if (arm64_early_this_cpu_has_bti())
713 text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP);
716 * Only rodata will be remapped with different permissions later on,
717 * all other segments are allowed to use contiguous mappings.
719 map_kernel_segment(pgdp, _stext, _etext, text_prot, &vmlinux_text, 0,
721 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
722 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
723 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
724 &vmlinux_inittext, 0, VM_NO_GUARD);
725 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
726 &vmlinux_initdata, 0, VM_NO_GUARD);
727 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
729 if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdp, FIXADDR_START)))) {
731 * The fixmap falls in a separate pgd to the kernel, and doesn't
732 * live in the carveout for the swapper_pg_dir. We can simply
733 * re-use the existing dir for the fixmap.
735 set_pgd(pgd_offset_pgd(pgdp, FIXADDR_START),
736 READ_ONCE(*pgd_offset_k(FIXADDR_START)));
737 } else if (CONFIG_PGTABLE_LEVELS > 3) {
742 * The fixmap shares its top level pgd entry with the kernel
743 * mapping. This can really only occur when we are running
744 * with 16k/4 levels, so we can simply reuse the pud level
747 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
748 bm_pgdp = pgd_offset_pgd(pgdp, FIXADDR_START);
749 bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_START);
750 bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_START);
751 pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
757 kasan_copy_shadow(pgdp);
760 static void __init create_idmap(void)
762 u64 start = __pa_symbol(__idmap_text_start);
763 u64 size = __pa_symbol(__idmap_text_end) - start;
764 pgd_t *pgd = idmap_pg_dir;
767 /* check if we need an additional level of translation */
768 if (VA_BITS < 48 && idmap_t0sz < (64 - VA_BITS_MIN)) {
769 pgd_phys = early_pgtable_alloc(PAGE_SHIFT);
770 set_pgd(&idmap_pg_dir[start >> VA_BITS],
771 __pgd(pgd_phys | P4D_TYPE_TABLE));
772 pgd = __va(pgd_phys);
774 __create_pgd_mapping(pgd, start, start, size, PAGE_KERNEL_ROX,
775 early_pgtable_alloc, 0);
777 if (IS_ENABLED(CONFIG_UNMAP_KERNEL_AT_EL0)) {
778 extern u32 __idmap_kpti_flag;
779 u64 pa = __pa_symbol(&__idmap_kpti_flag);
782 * The KPTI G-to-nG conversion code needs a read-write mapping
783 * of its synchronization flag in the ID map.
785 __create_pgd_mapping(pgd, pa, pa, sizeof(u32), PAGE_KERNEL,
786 early_pgtable_alloc, 0);
790 void __init paging_init(void)
792 pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
793 extern pgd_t init_idmap_pg_dir[];
795 idmap_t0sz = 63UL - __fls(__pa_symbol(_end) | GENMASK(VA_BITS_MIN - 1, 0));
802 cpu_replace_ttbr1(lm_alias(swapper_pg_dir), init_idmap_pg_dir);
803 init_mm.pgd = swapper_pg_dir;
805 memblock_phys_free(__pa_symbol(init_pg_dir),
806 __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
808 memblock_allow_resize();
814 * Check whether a kernel address is valid (derived from arch/x86/).
816 int kern_addr_valid(unsigned long addr)
824 addr = arch_kasan_reset_tag(addr);
825 if ((((long)addr) >> VA_BITS) != -1UL)
828 pgdp = pgd_offset_k(addr);
829 if (pgd_none(READ_ONCE(*pgdp)))
832 p4dp = p4d_offset(pgdp, addr);
833 if (p4d_none(READ_ONCE(*p4dp)))
836 pudp = pud_offset(p4dp, addr);
837 pud = READ_ONCE(*pudp);
842 return pfn_valid(pud_pfn(pud));
844 pmdp = pmd_offset(pudp, addr);
845 pmd = READ_ONCE(*pmdp);
850 return pfn_valid(pmd_pfn(pmd));
852 ptep = pte_offset_kernel(pmdp, addr);
853 pte = READ_ONCE(*ptep);
857 return pfn_valid(pte_pfn(pte));
860 #ifdef CONFIG_MEMORY_HOTPLUG
861 static void free_hotplug_page_range(struct page *page, size_t size,
862 struct vmem_altmap *altmap)
865 vmem_altmap_free(altmap, size >> PAGE_SHIFT);
867 WARN_ON(PageReserved(page));
868 free_pages((unsigned long)page_address(page), get_order(size));
872 static void free_hotplug_pgtable_page(struct page *page)
874 free_hotplug_page_range(page, PAGE_SIZE, NULL);
877 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
878 unsigned long floor, unsigned long ceiling,
891 if (end - 1 > ceiling - 1)
896 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
897 unsigned long end, bool free_mapped,
898 struct vmem_altmap *altmap)
903 ptep = pte_offset_kernel(pmdp, addr);
904 pte = READ_ONCE(*ptep);
908 WARN_ON(!pte_present(pte));
909 pte_clear(&init_mm, addr, ptep);
910 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
912 free_hotplug_page_range(pte_page(pte),
914 } while (addr += PAGE_SIZE, addr < end);
917 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
918 unsigned long end, bool free_mapped,
919 struct vmem_altmap *altmap)
925 next = pmd_addr_end(addr, end);
926 pmdp = pmd_offset(pudp, addr);
927 pmd = READ_ONCE(*pmdp);
931 WARN_ON(!pmd_present(pmd));
936 * One TLBI should be sufficient here as the PMD_SIZE
937 * range is mapped with a single block entry.
939 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
941 free_hotplug_page_range(pmd_page(pmd),
945 WARN_ON(!pmd_table(pmd));
946 unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap);
947 } while (addr = next, addr < end);
950 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
951 unsigned long end, bool free_mapped,
952 struct vmem_altmap *altmap)
958 next = pud_addr_end(addr, end);
959 pudp = pud_offset(p4dp, addr);
960 pud = READ_ONCE(*pudp);
964 WARN_ON(!pud_present(pud));
969 * One TLBI should be sufficient here as the PUD_SIZE
970 * range is mapped with a single block entry.
972 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
974 free_hotplug_page_range(pud_page(pud),
978 WARN_ON(!pud_table(pud));
979 unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap);
980 } while (addr = next, addr < end);
983 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
984 unsigned long end, bool free_mapped,
985 struct vmem_altmap *altmap)
991 next = p4d_addr_end(addr, end);
992 p4dp = p4d_offset(pgdp, addr);
993 p4d = READ_ONCE(*p4dp);
997 WARN_ON(!p4d_present(p4d));
998 unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap);
999 } while (addr = next, addr < end);
1002 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
1003 bool free_mapped, struct vmem_altmap *altmap)
1009 * altmap can only be used as vmemmap mapping backing memory.
1010 * In case the backing memory itself is not being freed, then
1011 * altmap is irrelevant. Warn about this inconsistency when
1014 WARN_ON(!free_mapped && altmap);
1017 next = pgd_addr_end(addr, end);
1018 pgdp = pgd_offset_k(addr);
1019 pgd = READ_ONCE(*pgdp);
1023 WARN_ON(!pgd_present(pgd));
1024 unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap);
1025 } while (addr = next, addr < end);
1028 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
1029 unsigned long end, unsigned long floor,
1030 unsigned long ceiling)
1033 unsigned long i, start = addr;
1036 ptep = pte_offset_kernel(pmdp, addr);
1037 pte = READ_ONCE(*ptep);
1040 * This is just a sanity check here which verifies that
1041 * pte clearing has been done by earlier unmap loops.
1043 WARN_ON(!pte_none(pte));
1044 } while (addr += PAGE_SIZE, addr < end);
1046 if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
1050 * Check whether we can free the pte page if the rest of the
1051 * entries are empty. Overlap with other regions have been
1052 * handled by the floor/ceiling check.
1054 ptep = pte_offset_kernel(pmdp, 0UL);
1055 for (i = 0; i < PTRS_PER_PTE; i++) {
1056 if (!pte_none(READ_ONCE(ptep[i])))
1061 __flush_tlb_kernel_pgtable(start);
1062 free_hotplug_pgtable_page(virt_to_page(ptep));
1065 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
1066 unsigned long end, unsigned long floor,
1067 unsigned long ceiling)
1070 unsigned long i, next, start = addr;
1073 next = pmd_addr_end(addr, end);
1074 pmdp = pmd_offset(pudp, addr);
1075 pmd = READ_ONCE(*pmdp);
1079 WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
1080 free_empty_pte_table(pmdp, addr, next, floor, ceiling);
1081 } while (addr = next, addr < end);
1083 if (CONFIG_PGTABLE_LEVELS <= 2)
1086 if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
1090 * Check whether we can free the pmd page if the rest of the
1091 * entries are empty. Overlap with other regions have been
1092 * handled by the floor/ceiling check.
1094 pmdp = pmd_offset(pudp, 0UL);
1095 for (i = 0; i < PTRS_PER_PMD; i++) {
1096 if (!pmd_none(READ_ONCE(pmdp[i])))
1101 __flush_tlb_kernel_pgtable(start);
1102 free_hotplug_pgtable_page(virt_to_page(pmdp));
1105 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
1106 unsigned long end, unsigned long floor,
1107 unsigned long ceiling)
1110 unsigned long i, next, start = addr;
1113 next = pud_addr_end(addr, end);
1114 pudp = pud_offset(p4dp, addr);
1115 pud = READ_ONCE(*pudp);
1119 WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1120 free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1121 } while (addr = next, addr < end);
1123 if (CONFIG_PGTABLE_LEVELS <= 3)
1126 if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1130 * Check whether we can free the pud page if the rest of the
1131 * entries are empty. Overlap with other regions have been
1132 * handled by the floor/ceiling check.
1134 pudp = pud_offset(p4dp, 0UL);
1135 for (i = 0; i < PTRS_PER_PUD; i++) {
1136 if (!pud_none(READ_ONCE(pudp[i])))
1141 __flush_tlb_kernel_pgtable(start);
1142 free_hotplug_pgtable_page(virt_to_page(pudp));
1145 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1146 unsigned long end, unsigned long floor,
1147 unsigned long ceiling)
1153 next = p4d_addr_end(addr, end);
1154 p4dp = p4d_offset(pgdp, addr);
1155 p4d = READ_ONCE(*p4dp);
1159 WARN_ON(!p4d_present(p4d));
1160 free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1161 } while (addr = next, addr < end);
1164 static void free_empty_tables(unsigned long addr, unsigned long end,
1165 unsigned long floor, unsigned long ceiling)
1171 next = pgd_addr_end(addr, end);
1172 pgdp = pgd_offset_k(addr);
1173 pgd = READ_ONCE(*pgdp);
1177 WARN_ON(!pgd_present(pgd));
1178 free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1179 } while (addr = next, addr < end);
1183 #if !ARM64_KERNEL_USES_PMD_MAPS
1184 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1185 struct vmem_altmap *altmap)
1187 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1188 return vmemmap_populate_basepages(start, end, node, altmap);
1190 #else /* !ARM64_KERNEL_USES_PMD_MAPS */
1191 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1192 struct vmem_altmap *altmap)
1194 unsigned long addr = start;
1201 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1203 next = pmd_addr_end(addr, end);
1205 pgdp = vmemmap_pgd_populate(addr, node);
1209 p4dp = vmemmap_p4d_populate(pgdp, addr, node);
1213 pudp = vmemmap_pud_populate(p4dp, addr, node);
1217 pmdp = pmd_offset(pudp, addr);
1218 if (pmd_none(READ_ONCE(*pmdp))) {
1221 p = vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1223 if (vmemmap_populate_basepages(addr, next, node, altmap))
1228 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1230 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1231 } while (addr = next, addr != end);
1235 #endif /* !ARM64_KERNEL_USES_PMD_MAPS */
1237 #ifdef CONFIG_MEMORY_HOTPLUG
1238 void vmemmap_free(unsigned long start, unsigned long end,
1239 struct vmem_altmap *altmap)
1241 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1243 unmap_hotplug_range(start, end, true, altmap);
1244 free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1246 #endif /* CONFIG_MEMORY_HOTPLUG */
1248 static inline pud_t *fixmap_pud(unsigned long addr)
1250 pgd_t *pgdp = pgd_offset_k(addr);
1251 p4d_t *p4dp = p4d_offset(pgdp, addr);
1252 p4d_t p4d = READ_ONCE(*p4dp);
1254 BUG_ON(p4d_none(p4d) || p4d_bad(p4d));
1256 return pud_offset_kimg(p4dp, addr);
1259 static inline pmd_t *fixmap_pmd(unsigned long addr)
1261 pud_t *pudp = fixmap_pud(addr);
1262 pud_t pud = READ_ONCE(*pudp);
1264 BUG_ON(pud_none(pud) || pud_bad(pud));
1266 return pmd_offset_kimg(pudp, addr);
1269 static inline pte_t *fixmap_pte(unsigned long addr)
1271 return &bm_pte[pte_index(addr)];
1275 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
1276 * directly on kernel symbols (bm_p*d). This function is called too early to use
1277 * lm_alias so __p*d_populate functions must be used to populate with the
1278 * physical address from __pa_symbol.
1280 void __init early_fixmap_init(void)
1286 unsigned long addr = FIXADDR_START;
1288 pgdp = pgd_offset_k(addr);
1289 p4dp = p4d_offset(pgdp, addr);
1290 p4d = READ_ONCE(*p4dp);
1291 if (CONFIG_PGTABLE_LEVELS > 3 &&
1292 !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) {
1294 * We only end up here if the kernel mapping and the fixmap
1295 * share the top level pgd entry, which should only happen on
1296 * 16k/4 levels configurations.
1298 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
1299 pudp = pud_offset_kimg(p4dp, addr);
1302 __p4d_populate(p4dp, __pa_symbol(bm_pud), P4D_TYPE_TABLE);
1303 pudp = fixmap_pud(addr);
1305 if (pud_none(READ_ONCE(*pudp)))
1306 __pud_populate(pudp, __pa_symbol(bm_pmd), PUD_TYPE_TABLE);
1307 pmdp = fixmap_pmd(addr);
1308 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
1311 * The boot-ioremap range spans multiple pmds, for which
1312 * we are not prepared:
1314 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1315 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1317 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
1318 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
1320 pr_warn("pmdp %p != %p, %p\n",
1321 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
1322 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
1323 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1324 fix_to_virt(FIX_BTMAP_BEGIN));
1325 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1326 fix_to_virt(FIX_BTMAP_END));
1328 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1329 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1334 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
1335 * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
1337 void __set_fixmap(enum fixed_addresses idx,
1338 phys_addr_t phys, pgprot_t flags)
1340 unsigned long addr = __fix_to_virt(idx);
1343 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
1345 ptep = fixmap_pte(addr);
1347 if (pgprot_val(flags)) {
1348 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
1350 pte_clear(&init_mm, addr, ptep);
1351 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
1355 void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
1357 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
1362 * Check whether the physical FDT address is set and meets the minimum
1363 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
1364 * at least 8 bytes so that we can always access the magic and size
1365 * fields of the FDT header after mapping the first chunk, double check
1366 * here if that is indeed the case.
1368 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
1369 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
1373 * Make sure that the FDT region can be mapped without the need to
1374 * allocate additional translation table pages, so that it is safe
1375 * to call create_mapping_noalloc() this early.
1377 * On 64k pages, the FDT will be mapped using PTEs, so we need to
1378 * be in the same PMD as the rest of the fixmap.
1379 * On 4k pages, we'll use section mappings for the FDT so we only
1380 * have to be in the same PUD.
1382 BUILD_BUG_ON(dt_virt_base % SZ_2M);
1384 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
1385 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
1387 offset = dt_phys % SWAPPER_BLOCK_SIZE;
1388 dt_virt = (void *)dt_virt_base + offset;
1390 /* map the first chunk so we can read the size from the header */
1391 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
1392 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
1394 if (fdt_magic(dt_virt) != FDT_MAGIC)
1397 *size = fdt_totalsize(dt_virt);
1398 if (*size > MAX_FDT_SIZE)
1401 if (offset + *size > SWAPPER_BLOCK_SIZE)
1402 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
1403 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
1408 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1410 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1412 /* Only allow permission changes for now */
1413 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1417 VM_BUG_ON(phys & ~PUD_MASK);
1418 set_pud(pudp, new_pud);
1422 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1424 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1426 /* Only allow permission changes for now */
1427 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1431 VM_BUG_ON(phys & ~PMD_MASK);
1432 set_pmd(pmdp, new_pmd);
1436 int pud_clear_huge(pud_t *pudp)
1438 if (!pud_sect(READ_ONCE(*pudp)))
1444 int pmd_clear_huge(pmd_t *pmdp)
1446 if (!pmd_sect(READ_ONCE(*pmdp)))
1452 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1457 pmd = READ_ONCE(*pmdp);
1459 if (!pmd_table(pmd)) {
1464 table = pte_offset_kernel(pmdp, addr);
1466 __flush_tlb_kernel_pgtable(addr);
1467 pte_free_kernel(NULL, table);
1471 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1476 unsigned long next, end;
1478 pud = READ_ONCE(*pudp);
1480 if (!pud_table(pud)) {
1485 table = pmd_offset(pudp, addr);
1488 end = addr + PUD_SIZE;
1490 pmd_free_pte_page(pmdp, next);
1491 } while (pmdp++, next += PMD_SIZE, next != end);
1494 __flush_tlb_kernel_pgtable(addr);
1495 pmd_free(NULL, table);
1499 #ifdef CONFIG_MEMORY_HOTPLUG
1500 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1502 unsigned long end = start + size;
1504 WARN_ON(pgdir != init_mm.pgd);
1505 WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1507 unmap_hotplug_range(start, end, false, NULL);
1508 free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1511 struct range arch_get_mappable_range(void)
1513 struct range mhp_range;
1514 u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1515 u64 end_linear_pa = __pa(PAGE_END - 1);
1517 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1519 * Check for a wrap, it is possible because of randomized linear
1520 * mapping the start physical address is actually bigger than
1521 * the end physical address. In this case set start to zero
1522 * because [0, end_linear_pa] range must still be able to cover
1523 * all addressable physical addresses.
1525 if (start_linear_pa > end_linear_pa)
1526 start_linear_pa = 0;
1529 WARN_ON(start_linear_pa > end_linear_pa);
1532 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1533 * accommodating both its ends but excluding PAGE_END. Max physical
1534 * range which can be mapped inside this linear mapping range, must
1535 * also be derived from its end points.
1537 mhp_range.start = start_linear_pa;
1538 mhp_range.end = end_linear_pa;
1543 int arch_add_memory(int nid, u64 start, u64 size,
1544 struct mhp_params *params)
1546 int ret, flags = NO_EXEC_MAPPINGS;
1548 VM_BUG_ON(!mhp_range_allowed(start, size, true));
1551 * KFENCE requires linear map to be mapped at page granularity, so that
1552 * it is possible to protect/unprotect single pages in the KFENCE pool.
1554 if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
1555 flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1557 __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1558 size, params->pgprot, __pgd_pgtable_alloc,
1561 memblock_clear_nomap(start, size);
1563 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1566 __remove_pgd_mapping(swapper_pg_dir,
1567 __phys_to_virt(start), size);
1569 max_pfn = PFN_UP(start + size);
1570 max_low_pfn = max_pfn;
1576 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1578 unsigned long start_pfn = start >> PAGE_SHIFT;
1579 unsigned long nr_pages = size >> PAGE_SHIFT;
1581 __remove_pages(start_pfn, nr_pages, altmap);
1582 __remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1586 * This memory hotplug notifier helps prevent boot memory from being
1587 * inadvertently removed as it blocks pfn range offlining process in
1588 * __offline_pages(). Hence this prevents both offlining as well as
1589 * removal process for boot memory which is initially always online.
1590 * In future if and when boot memory could be removed, this notifier
1591 * should be dropped and free_hotplug_page_range() should handle any
1592 * reserved pages allocated during boot.
1594 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1595 unsigned long action, void *data)
1597 struct mem_section *ms;
1598 struct memory_notify *arg = data;
1599 unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1600 unsigned long pfn = arg->start_pfn;
1602 if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1605 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1606 unsigned long start = PFN_PHYS(pfn);
1607 unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1609 ms = __pfn_to_section(pfn);
1610 if (!early_section(ms))
1613 if (action == MEM_GOING_OFFLINE) {
1615 * Boot memory removal is not supported. Prevent
1616 * it via blocking any attempted offline request
1617 * for the boot memory and just report it.
1619 pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1621 } else if (action == MEM_OFFLINE) {
1623 * This should have never happened. Boot memory
1624 * offlining should have been prevented by this
1625 * very notifier. Probably some memory removal
1626 * procedure might have changed which would then
1627 * require further debug.
1629 pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1632 * Core memory hotplug does not process a return
1633 * code from the notifier for MEM_OFFLINE events.
1634 * The error condition has been reported. Return
1635 * from here as if ignored.
1643 static struct notifier_block prevent_bootmem_remove_nb = {
1644 .notifier_call = prevent_bootmem_remove_notifier,
1648 * This ensures that boot memory sections on the platform are online
1649 * from early boot. Memory sections could not be prevented from being
1650 * offlined, unless for some reason they are not online to begin with.
1651 * This helps validate the basic assumption on which the above memory
1652 * event notifier works to prevent boot memory section offlining and
1653 * its possible removal.
1655 static void validate_bootmem_online(void)
1657 phys_addr_t start, end, addr;
1658 struct mem_section *ms;
1662 * Scanning across all memblock might be expensive
1663 * on some big memory systems. Hence enable this
1664 * validation only with DEBUG_VM.
1666 if (!IS_ENABLED(CONFIG_DEBUG_VM))
1669 for_each_mem_range(i, &start, &end) {
1670 for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1671 ms = __pfn_to_section(PHYS_PFN(addr));
1674 * All memory ranges in the system at this point
1675 * should have been marked as early sections.
1677 WARN_ON(!early_section(ms));
1680 * Memory notifier mechanism here to prevent boot
1681 * memory offlining depends on the fact that each
1682 * early section memory on the system is initially
1683 * online. Otherwise a given memory section which
1684 * is already offline will be overlooked and can
1685 * be removed completely. Call out such sections.
1687 if (!online_section(ms))
1688 pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1689 addr, addr + (1UL << PA_SECTION_SHIFT));
1694 static int __init prevent_bootmem_remove_init(void)
1698 if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1701 validate_bootmem_online();
1702 ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1704 pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1708 early_initcall(prevent_bootmem_remove_init);