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));
334 pudp = pud_set_fixmap_offset(p4dp, addr);
336 pud_t old_pud = READ_ONCE(*pudp);
338 next = pud_addr_end(addr, end);
341 * For 4K granule only, attempt to put down a 1GB block
343 if (pud_sect_supported() &&
344 ((addr | next | phys) & ~PUD_MASK) == 0 &&
345 (flags & NO_BLOCK_MAPPINGS) == 0) {
346 pud_set_huge(pudp, phys, prot);
349 * After the PUD entry has been populated once, we
350 * only allow updates to the permission attributes.
352 BUG_ON(!pgattr_change_is_safe(pud_val(old_pud),
353 READ_ONCE(pud_val(*pudp))));
355 alloc_init_cont_pmd(pudp, addr, next, phys, prot,
356 pgtable_alloc, flags);
358 BUG_ON(pud_val(old_pud) != 0 &&
359 pud_val(old_pud) != READ_ONCE(pud_val(*pudp)));
362 } while (pudp++, addr = next, addr != end);
367 static void __create_pgd_mapping_locked(pgd_t *pgdir, phys_addr_t phys,
368 unsigned long virt, phys_addr_t size,
370 phys_addr_t (*pgtable_alloc)(int),
373 unsigned long addr, end, next;
374 pgd_t *pgdp = pgd_offset_pgd(pgdir, virt);
377 * If the virtual and physical address don't have the same offset
378 * within a page, we cannot map the region as the caller expects.
380 if (WARN_ON((phys ^ virt) & ~PAGE_MASK))
384 addr = virt & PAGE_MASK;
385 end = PAGE_ALIGN(virt + size);
388 next = pgd_addr_end(addr, end);
389 alloc_init_pud(pgdp, addr, next, phys, prot, pgtable_alloc,
392 } while (pgdp++, addr = next, addr != end);
395 static void __create_pgd_mapping(pgd_t *pgdir, phys_addr_t phys,
396 unsigned long virt, phys_addr_t size,
398 phys_addr_t (*pgtable_alloc)(int),
401 mutex_lock(&fixmap_lock);
402 __create_pgd_mapping_locked(pgdir, phys, virt, size, prot,
403 pgtable_alloc, flags);
404 mutex_unlock(&fixmap_lock);
407 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
408 extern __alias(__create_pgd_mapping_locked)
409 void create_kpti_ng_temp_pgd(pgd_t *pgdir, phys_addr_t phys, unsigned long virt,
410 phys_addr_t size, pgprot_t prot,
411 phys_addr_t (*pgtable_alloc)(int), int flags);
414 static phys_addr_t __pgd_pgtable_alloc(int shift)
416 void *ptr = (void *)__get_free_page(GFP_PGTABLE_KERNEL);
419 /* Ensure the zeroed page is visible to the page table walker */
424 static phys_addr_t pgd_pgtable_alloc(int shift)
426 phys_addr_t pa = __pgd_pgtable_alloc(shift);
429 * Call proper page table ctor in case later we need to
430 * call core mm functions like apply_to_page_range() on
431 * this pre-allocated page table.
433 * We don't select ARCH_ENABLE_SPLIT_PMD_PTLOCK if pmd is
434 * folded, and if so pgtable_pmd_page_ctor() becomes nop.
436 if (shift == PAGE_SHIFT)
437 BUG_ON(!pgtable_pte_page_ctor(phys_to_page(pa)));
438 else if (shift == PMD_SHIFT)
439 BUG_ON(!pgtable_pmd_page_ctor(phys_to_page(pa)));
445 * This function can only be used to modify existing table entries,
446 * without allocating new levels of table. Note that this permits the
447 * creation of new section or page entries.
449 static void __init create_mapping_noalloc(phys_addr_t phys, unsigned long virt,
450 phys_addr_t size, pgprot_t prot)
452 if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
453 pr_warn("BUG: not creating mapping for %pa at 0x%016lx - outside kernel range\n",
457 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
461 void __init create_pgd_mapping(struct mm_struct *mm, phys_addr_t phys,
462 unsigned long virt, phys_addr_t size,
463 pgprot_t prot, bool page_mappings_only)
467 BUG_ON(mm == &init_mm);
469 if (page_mappings_only)
470 flags = NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
472 __create_pgd_mapping(mm->pgd, phys, virt, size, prot,
473 pgd_pgtable_alloc, flags);
476 static void update_mapping_prot(phys_addr_t phys, unsigned long virt,
477 phys_addr_t size, pgprot_t prot)
479 if ((virt >= PAGE_END) && (virt < VMALLOC_START)) {
480 pr_warn("BUG: not updating mapping for %pa at 0x%016lx - outside kernel range\n",
485 __create_pgd_mapping(init_mm.pgd, phys, virt, size, prot, NULL,
488 /* flush the TLBs after updating live kernel mappings */
489 flush_tlb_kernel_range(virt, virt + size);
492 static void __init __map_memblock(pgd_t *pgdp, phys_addr_t start,
493 phys_addr_t end, pgprot_t prot, int flags)
495 __create_pgd_mapping(pgdp, start, __phys_to_virt(start), end - start,
496 prot, early_pgtable_alloc, flags);
499 void __init mark_linear_text_alias_ro(void)
502 * Remove the write permissions from the linear alias of .text/.rodata
504 update_mapping_prot(__pa_symbol(_stext), (unsigned long)lm_alias(_stext),
505 (unsigned long)__init_begin - (unsigned long)_stext,
509 static bool crash_mem_map __initdata;
511 static int __init enable_crash_mem_map(char *arg)
514 * Proper parameter parsing is done by reserve_crashkernel(). We only
515 * need to know if the linear map has to avoid block mappings so that
516 * the crashkernel reservations can be unmapped later.
518 crash_mem_map = true;
522 early_param("crashkernel", enable_crash_mem_map);
524 static void __init map_mem(pgd_t *pgdp)
526 static const u64 direct_map_end = _PAGE_END(VA_BITS_MIN);
527 phys_addr_t kernel_start = __pa_symbol(_stext);
528 phys_addr_t kernel_end = __pa_symbol(__init_begin);
529 phys_addr_t start, end;
530 int flags = NO_EXEC_MAPPINGS;
534 * Setting hierarchical PXNTable attributes on table entries covering
535 * the linear region is only possible if it is guaranteed that no table
536 * entries at any level are being shared between the linear region and
537 * the vmalloc region. Check whether this is true for the PGD level, in
538 * which case it is guaranteed to be true for all other levels as well.
540 BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));
542 if (can_set_direct_map())
543 flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
546 * Take care not to create a writable alias for the
547 * read-only text and rodata sections of the kernel image.
548 * So temporarily mark them as NOMAP to skip mappings in
549 * the following for-loop
551 memblock_mark_nomap(kernel_start, kernel_end - kernel_start);
553 #ifdef CONFIG_KEXEC_CORE
555 if (defer_reserve_crashkernel())
556 flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
557 else if (crashk_res.end)
558 memblock_mark_nomap(crashk_res.start,
559 resource_size(&crashk_res));
563 /* map all the memory banks */
564 for_each_mem_range(i, &start, &end) {
568 * The linear map must allow allocation tags reading/writing
569 * if MTE is present. Otherwise, it has the same attributes as
572 __map_memblock(pgdp, start, end, pgprot_tagged(PAGE_KERNEL),
577 * Map the linear alias of the [_stext, __init_begin) interval
578 * as non-executable now, and remove the write permission in
579 * mark_linear_text_alias_ro() below (which will be called after
580 * alternative patching has completed). This makes the contents
581 * of the region accessible to subsystems such as hibernate,
582 * but protects it from inadvertent modification or execution.
583 * Note that contiguous mappings cannot be remapped in this way,
584 * so we should avoid them here.
586 __map_memblock(pgdp, kernel_start, kernel_end,
587 PAGE_KERNEL, NO_CONT_MAPPINGS);
588 memblock_clear_nomap(kernel_start, kernel_end - kernel_start);
591 * Use page-level mappings here so that we can shrink the region
592 * in page granularity and put back unused memory to buddy system
593 * through /sys/kernel/kexec_crash_size interface.
595 #ifdef CONFIG_KEXEC_CORE
596 if (crash_mem_map && !defer_reserve_crashkernel()) {
597 if (crashk_res.end) {
598 __map_memblock(pgdp, crashk_res.start,
601 NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS);
602 memblock_clear_nomap(crashk_res.start,
603 resource_size(&crashk_res));
609 void mark_rodata_ro(void)
611 unsigned long section_size;
614 * mark .rodata as read only. Use __init_begin rather than __end_rodata
615 * to cover NOTES and EXCEPTION_TABLE.
617 section_size = (unsigned long)__init_begin - (unsigned long)__start_rodata;
618 update_mapping_prot(__pa_symbol(__start_rodata), (unsigned long)__start_rodata,
619 section_size, PAGE_KERNEL_RO);
624 static void __init map_kernel_segment(pgd_t *pgdp, void *va_start, void *va_end,
625 pgprot_t prot, struct vm_struct *vma,
626 int flags, unsigned long vm_flags)
628 phys_addr_t pa_start = __pa_symbol(va_start);
629 unsigned long size = va_end - va_start;
631 BUG_ON(!PAGE_ALIGNED(pa_start));
632 BUG_ON(!PAGE_ALIGNED(size));
634 __create_pgd_mapping(pgdp, pa_start, (unsigned long)va_start, size, prot,
635 early_pgtable_alloc, flags);
637 if (!(vm_flags & VM_NO_GUARD))
640 vma->addr = va_start;
641 vma->phys_addr = pa_start;
643 vma->flags = VM_MAP | vm_flags;
644 vma->caller = __builtin_return_address(0);
646 vm_area_add_early(vma);
649 #ifdef CONFIG_UNMAP_KERNEL_AT_EL0
650 static int __init map_entry_trampoline(void)
654 pgprot_t prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
655 phys_addr_t pa_start = __pa_symbol(__entry_tramp_text_start);
657 /* The trampoline is always mapped and can therefore be global */
658 pgprot_val(prot) &= ~PTE_NG;
660 /* Map only the text into the trampoline page table */
661 memset(tramp_pg_dir, 0, PGD_SIZE);
662 __create_pgd_mapping(tramp_pg_dir, pa_start, TRAMP_VALIAS,
663 entry_tramp_text_size(), prot,
664 __pgd_pgtable_alloc, NO_BLOCK_MAPPINGS);
666 /* Map both the text and data into the kernel page table */
667 for (i = 0; i < DIV_ROUND_UP(entry_tramp_text_size(), PAGE_SIZE); i++)
668 __set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
669 pa_start + i * PAGE_SIZE, prot);
671 if (IS_ENABLED(CONFIG_RELOCATABLE))
672 __set_fixmap(FIX_ENTRY_TRAMP_TEXT1 - i,
673 pa_start + i * PAGE_SIZE, PAGE_KERNEL_RO);
677 core_initcall(map_entry_trampoline);
681 * Open coded check for BTI, only for use to determine configuration
682 * for early mappings for before the cpufeature code has run.
684 static bool arm64_early_this_cpu_has_bti(void)
688 if (!IS_ENABLED(CONFIG_ARM64_BTI_KERNEL))
691 pfr1 = __read_sysreg_by_encoding(SYS_ID_AA64PFR1_EL1);
692 return cpuid_feature_extract_unsigned_field(pfr1,
693 ID_AA64PFR1_EL1_BT_SHIFT);
697 * Create fine-grained mappings for the kernel.
699 static void __init map_kernel(pgd_t *pgdp)
701 static struct vm_struct vmlinux_text, vmlinux_rodata, vmlinux_inittext,
702 vmlinux_initdata, vmlinux_data;
705 * External debuggers may need to write directly to the text
706 * mapping to install SW breakpoints. Allow this (only) when
707 * explicitly requested with rodata=off.
709 pgprot_t text_prot = rodata_enabled ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC;
712 * If we have a CPU that supports BTI and a kernel built for
713 * BTI then mark the kernel executable text as guarded pages
714 * now so we don't have to rewrite the page tables later.
716 if (arm64_early_this_cpu_has_bti())
717 text_prot = __pgprot_modify(text_prot, PTE_GP, PTE_GP);
720 * Only rodata will be remapped with different permissions later on,
721 * all other segments are allowed to use contiguous mappings.
723 map_kernel_segment(pgdp, _stext, _etext, text_prot, &vmlinux_text, 0,
725 map_kernel_segment(pgdp, __start_rodata, __inittext_begin, PAGE_KERNEL,
726 &vmlinux_rodata, NO_CONT_MAPPINGS, VM_NO_GUARD);
727 map_kernel_segment(pgdp, __inittext_begin, __inittext_end, text_prot,
728 &vmlinux_inittext, 0, VM_NO_GUARD);
729 map_kernel_segment(pgdp, __initdata_begin, __initdata_end, PAGE_KERNEL,
730 &vmlinux_initdata, 0, VM_NO_GUARD);
731 map_kernel_segment(pgdp, _data, _end, PAGE_KERNEL, &vmlinux_data, 0, 0);
733 if (!READ_ONCE(pgd_val(*pgd_offset_pgd(pgdp, FIXADDR_START)))) {
735 * The fixmap falls in a separate pgd to the kernel, and doesn't
736 * live in the carveout for the swapper_pg_dir. We can simply
737 * re-use the existing dir for the fixmap.
739 set_pgd(pgd_offset_pgd(pgdp, FIXADDR_START),
740 READ_ONCE(*pgd_offset_k(FIXADDR_START)));
741 } else if (CONFIG_PGTABLE_LEVELS > 3) {
746 * The fixmap shares its top level pgd entry with the kernel
747 * mapping. This can really only occur when we are running
748 * with 16k/4 levels, so we can simply reuse the pud level
751 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
752 bm_pgdp = pgd_offset_pgd(pgdp, FIXADDR_START);
753 bm_p4dp = p4d_offset(bm_pgdp, FIXADDR_START);
754 bm_pudp = pud_set_fixmap_offset(bm_p4dp, FIXADDR_START);
755 pud_populate(&init_mm, bm_pudp, lm_alias(bm_pmd));
761 kasan_copy_shadow(pgdp);
764 static void __init create_idmap(void)
766 u64 start = __pa_symbol(__idmap_text_start);
767 u64 size = __pa_symbol(__idmap_text_end) - start;
768 pgd_t *pgd = idmap_pg_dir;
771 /* check if we need an additional level of translation */
772 if (VA_BITS < 48 && idmap_t0sz < (64 - VA_BITS_MIN)) {
773 pgd_phys = early_pgtable_alloc(PAGE_SHIFT);
774 set_pgd(&idmap_pg_dir[start >> VA_BITS],
775 __pgd(pgd_phys | P4D_TYPE_TABLE));
776 pgd = __va(pgd_phys);
778 __create_pgd_mapping(pgd, start, start, size, PAGE_KERNEL_ROX,
779 early_pgtable_alloc, 0);
781 if (IS_ENABLED(CONFIG_UNMAP_KERNEL_AT_EL0)) {
782 extern u32 __idmap_kpti_flag;
783 u64 pa = __pa_symbol(&__idmap_kpti_flag);
786 * The KPTI G-to-nG conversion code needs a read-write mapping
787 * of its synchronization flag in the ID map.
789 __create_pgd_mapping(pgd, pa, pa, sizeof(u32), PAGE_KERNEL,
790 early_pgtable_alloc, 0);
794 void __init paging_init(void)
796 pgd_t *pgdp = pgd_set_fixmap(__pa_symbol(swapper_pg_dir));
797 extern pgd_t init_idmap_pg_dir[];
799 idmap_t0sz = 63UL - __fls(__pa_symbol(_end) | GENMASK(VA_BITS_MIN - 1, 0));
806 cpu_replace_ttbr1(lm_alias(swapper_pg_dir), init_idmap_pg_dir);
807 init_mm.pgd = swapper_pg_dir;
809 memblock_phys_free(__pa_symbol(init_pg_dir),
810 __pa_symbol(init_pg_end) - __pa_symbol(init_pg_dir));
812 memblock_allow_resize();
817 #ifdef CONFIG_MEMORY_HOTPLUG
818 static void free_hotplug_page_range(struct page *page, size_t size,
819 struct vmem_altmap *altmap)
822 vmem_altmap_free(altmap, size >> PAGE_SHIFT);
824 WARN_ON(PageReserved(page));
825 free_pages((unsigned long)page_address(page), get_order(size));
829 static void free_hotplug_pgtable_page(struct page *page)
831 free_hotplug_page_range(page, PAGE_SIZE, NULL);
834 static bool pgtable_range_aligned(unsigned long start, unsigned long end,
835 unsigned long floor, unsigned long ceiling,
848 if (end - 1 > ceiling - 1)
853 static void unmap_hotplug_pte_range(pmd_t *pmdp, unsigned long addr,
854 unsigned long end, bool free_mapped,
855 struct vmem_altmap *altmap)
860 ptep = pte_offset_kernel(pmdp, addr);
861 pte = READ_ONCE(*ptep);
865 WARN_ON(!pte_present(pte));
866 pte_clear(&init_mm, addr, ptep);
867 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
869 free_hotplug_page_range(pte_page(pte),
871 } while (addr += PAGE_SIZE, addr < end);
874 static void unmap_hotplug_pmd_range(pud_t *pudp, unsigned long addr,
875 unsigned long end, bool free_mapped,
876 struct vmem_altmap *altmap)
882 next = pmd_addr_end(addr, end);
883 pmdp = pmd_offset(pudp, addr);
884 pmd = READ_ONCE(*pmdp);
888 WARN_ON(!pmd_present(pmd));
893 * One TLBI should be sufficient here as the PMD_SIZE
894 * range is mapped with a single block entry.
896 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
898 free_hotplug_page_range(pmd_page(pmd),
902 WARN_ON(!pmd_table(pmd));
903 unmap_hotplug_pte_range(pmdp, addr, next, free_mapped, altmap);
904 } while (addr = next, addr < end);
907 static void unmap_hotplug_pud_range(p4d_t *p4dp, unsigned long addr,
908 unsigned long end, bool free_mapped,
909 struct vmem_altmap *altmap)
915 next = pud_addr_end(addr, end);
916 pudp = pud_offset(p4dp, addr);
917 pud = READ_ONCE(*pudp);
921 WARN_ON(!pud_present(pud));
926 * One TLBI should be sufficient here as the PUD_SIZE
927 * range is mapped with a single block entry.
929 flush_tlb_kernel_range(addr, addr + PAGE_SIZE);
931 free_hotplug_page_range(pud_page(pud),
935 WARN_ON(!pud_table(pud));
936 unmap_hotplug_pmd_range(pudp, addr, next, free_mapped, altmap);
937 } while (addr = next, addr < end);
940 static void unmap_hotplug_p4d_range(pgd_t *pgdp, unsigned long addr,
941 unsigned long end, bool free_mapped,
942 struct vmem_altmap *altmap)
948 next = p4d_addr_end(addr, end);
949 p4dp = p4d_offset(pgdp, addr);
950 p4d = READ_ONCE(*p4dp);
954 WARN_ON(!p4d_present(p4d));
955 unmap_hotplug_pud_range(p4dp, addr, next, free_mapped, altmap);
956 } while (addr = next, addr < end);
959 static void unmap_hotplug_range(unsigned long addr, unsigned long end,
960 bool free_mapped, struct vmem_altmap *altmap)
966 * altmap can only be used as vmemmap mapping backing memory.
967 * In case the backing memory itself is not being freed, then
968 * altmap is irrelevant. Warn about this inconsistency when
971 WARN_ON(!free_mapped && altmap);
974 next = pgd_addr_end(addr, end);
975 pgdp = pgd_offset_k(addr);
976 pgd = READ_ONCE(*pgdp);
980 WARN_ON(!pgd_present(pgd));
981 unmap_hotplug_p4d_range(pgdp, addr, next, free_mapped, altmap);
982 } while (addr = next, addr < end);
985 static void free_empty_pte_table(pmd_t *pmdp, unsigned long addr,
986 unsigned long end, unsigned long floor,
987 unsigned long ceiling)
990 unsigned long i, start = addr;
993 ptep = pte_offset_kernel(pmdp, addr);
994 pte = READ_ONCE(*ptep);
997 * This is just a sanity check here which verifies that
998 * pte clearing has been done by earlier unmap loops.
1000 WARN_ON(!pte_none(pte));
1001 } while (addr += PAGE_SIZE, addr < end);
1003 if (!pgtable_range_aligned(start, end, floor, ceiling, PMD_MASK))
1007 * Check whether we can free the pte page if the rest of the
1008 * entries are empty. Overlap with other regions have been
1009 * handled by the floor/ceiling check.
1011 ptep = pte_offset_kernel(pmdp, 0UL);
1012 for (i = 0; i < PTRS_PER_PTE; i++) {
1013 if (!pte_none(READ_ONCE(ptep[i])))
1018 __flush_tlb_kernel_pgtable(start);
1019 free_hotplug_pgtable_page(virt_to_page(ptep));
1022 static void free_empty_pmd_table(pud_t *pudp, unsigned long addr,
1023 unsigned long end, unsigned long floor,
1024 unsigned long ceiling)
1027 unsigned long i, next, start = addr;
1030 next = pmd_addr_end(addr, end);
1031 pmdp = pmd_offset(pudp, addr);
1032 pmd = READ_ONCE(*pmdp);
1036 WARN_ON(!pmd_present(pmd) || !pmd_table(pmd) || pmd_sect(pmd));
1037 free_empty_pte_table(pmdp, addr, next, floor, ceiling);
1038 } while (addr = next, addr < end);
1040 if (CONFIG_PGTABLE_LEVELS <= 2)
1043 if (!pgtable_range_aligned(start, end, floor, ceiling, PUD_MASK))
1047 * Check whether we can free the pmd page if the rest of the
1048 * entries are empty. Overlap with other regions have been
1049 * handled by the floor/ceiling check.
1051 pmdp = pmd_offset(pudp, 0UL);
1052 for (i = 0; i < PTRS_PER_PMD; i++) {
1053 if (!pmd_none(READ_ONCE(pmdp[i])))
1058 __flush_tlb_kernel_pgtable(start);
1059 free_hotplug_pgtable_page(virt_to_page(pmdp));
1062 static void free_empty_pud_table(p4d_t *p4dp, unsigned long addr,
1063 unsigned long end, unsigned long floor,
1064 unsigned long ceiling)
1067 unsigned long i, next, start = addr;
1070 next = pud_addr_end(addr, end);
1071 pudp = pud_offset(p4dp, addr);
1072 pud = READ_ONCE(*pudp);
1076 WARN_ON(!pud_present(pud) || !pud_table(pud) || pud_sect(pud));
1077 free_empty_pmd_table(pudp, addr, next, floor, ceiling);
1078 } while (addr = next, addr < end);
1080 if (CONFIG_PGTABLE_LEVELS <= 3)
1083 if (!pgtable_range_aligned(start, end, floor, ceiling, PGDIR_MASK))
1087 * Check whether we can free the pud page if the rest of the
1088 * entries are empty. Overlap with other regions have been
1089 * handled by the floor/ceiling check.
1091 pudp = pud_offset(p4dp, 0UL);
1092 for (i = 0; i < PTRS_PER_PUD; i++) {
1093 if (!pud_none(READ_ONCE(pudp[i])))
1098 __flush_tlb_kernel_pgtable(start);
1099 free_hotplug_pgtable_page(virt_to_page(pudp));
1102 static void free_empty_p4d_table(pgd_t *pgdp, unsigned long addr,
1103 unsigned long end, unsigned long floor,
1104 unsigned long ceiling)
1110 next = p4d_addr_end(addr, end);
1111 p4dp = p4d_offset(pgdp, addr);
1112 p4d = READ_ONCE(*p4dp);
1116 WARN_ON(!p4d_present(p4d));
1117 free_empty_pud_table(p4dp, addr, next, floor, ceiling);
1118 } while (addr = next, addr < end);
1121 static void free_empty_tables(unsigned long addr, unsigned long end,
1122 unsigned long floor, unsigned long ceiling)
1128 next = pgd_addr_end(addr, end);
1129 pgdp = pgd_offset_k(addr);
1130 pgd = READ_ONCE(*pgdp);
1134 WARN_ON(!pgd_present(pgd));
1135 free_empty_p4d_table(pgdp, addr, next, floor, ceiling);
1136 } while (addr = next, addr < end);
1140 void __meminit vmemmap_set_pmd(pmd_t *pmdp, void *p, int node,
1141 unsigned long addr, unsigned long next)
1143 pmd_set_huge(pmdp, __pa(p), __pgprot(PROT_SECT_NORMAL));
1146 int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node,
1147 unsigned long addr, unsigned long next)
1149 vmemmap_verify((pte_t *)pmdp, node, addr, next);
1153 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1154 struct vmem_altmap *altmap)
1156 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1158 if (!IS_ENABLED(CONFIG_ARM64_4K_PAGES))
1159 return vmemmap_populate_basepages(start, end, node, altmap);
1161 return vmemmap_populate_hugepages(start, end, node, altmap);
1164 #ifdef CONFIG_MEMORY_HOTPLUG
1165 void vmemmap_free(unsigned long start, unsigned long end,
1166 struct vmem_altmap *altmap)
1168 WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
1170 unmap_hotplug_range(start, end, true, altmap);
1171 free_empty_tables(start, end, VMEMMAP_START, VMEMMAP_END);
1173 #endif /* CONFIG_MEMORY_HOTPLUG */
1175 static inline pud_t *fixmap_pud(unsigned long addr)
1177 pgd_t *pgdp = pgd_offset_k(addr);
1178 p4d_t *p4dp = p4d_offset(pgdp, addr);
1179 p4d_t p4d = READ_ONCE(*p4dp);
1181 BUG_ON(p4d_none(p4d) || p4d_bad(p4d));
1183 return pud_offset_kimg(p4dp, addr);
1186 static inline pmd_t *fixmap_pmd(unsigned long addr)
1188 pud_t *pudp = fixmap_pud(addr);
1189 pud_t pud = READ_ONCE(*pudp);
1191 BUG_ON(pud_none(pud) || pud_bad(pud));
1193 return pmd_offset_kimg(pudp, addr);
1196 static inline pte_t *fixmap_pte(unsigned long addr)
1198 return &bm_pte[pte_index(addr)];
1202 * The p*d_populate functions call virt_to_phys implicitly so they can't be used
1203 * directly on kernel symbols (bm_p*d). This function is called too early to use
1204 * lm_alias so __p*d_populate functions must be used to populate with the
1205 * physical address from __pa_symbol.
1207 void __init early_fixmap_init(void)
1213 unsigned long addr = FIXADDR_START;
1215 pgdp = pgd_offset_k(addr);
1216 p4dp = p4d_offset(pgdp, addr);
1217 p4d = READ_ONCE(*p4dp);
1218 if (CONFIG_PGTABLE_LEVELS > 3 &&
1219 !(p4d_none(p4d) || p4d_page_paddr(p4d) == __pa_symbol(bm_pud))) {
1221 * We only end up here if the kernel mapping and the fixmap
1222 * share the top level pgd entry, which should only happen on
1223 * 16k/4 levels configurations.
1225 BUG_ON(!IS_ENABLED(CONFIG_ARM64_16K_PAGES));
1226 pudp = pud_offset_kimg(p4dp, addr);
1229 __p4d_populate(p4dp, __pa_symbol(bm_pud), P4D_TYPE_TABLE);
1230 pudp = fixmap_pud(addr);
1232 if (pud_none(READ_ONCE(*pudp)))
1233 __pud_populate(pudp, __pa_symbol(bm_pmd), PUD_TYPE_TABLE);
1234 pmdp = fixmap_pmd(addr);
1235 __pmd_populate(pmdp, __pa_symbol(bm_pte), PMD_TYPE_TABLE);
1238 * The boot-ioremap range spans multiple pmds, for which
1239 * we are not prepared:
1241 BUILD_BUG_ON((__fix_to_virt(FIX_BTMAP_BEGIN) >> PMD_SHIFT)
1242 != (__fix_to_virt(FIX_BTMAP_END) >> PMD_SHIFT));
1244 if ((pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)))
1245 || pmdp != fixmap_pmd(fix_to_virt(FIX_BTMAP_END))) {
1247 pr_warn("pmdp %p != %p, %p\n",
1248 pmdp, fixmap_pmd(fix_to_virt(FIX_BTMAP_BEGIN)),
1249 fixmap_pmd(fix_to_virt(FIX_BTMAP_END)));
1250 pr_warn("fix_to_virt(FIX_BTMAP_BEGIN): %08lx\n",
1251 fix_to_virt(FIX_BTMAP_BEGIN));
1252 pr_warn("fix_to_virt(FIX_BTMAP_END): %08lx\n",
1253 fix_to_virt(FIX_BTMAP_END));
1255 pr_warn("FIX_BTMAP_END: %d\n", FIX_BTMAP_END);
1256 pr_warn("FIX_BTMAP_BEGIN: %d\n", FIX_BTMAP_BEGIN);
1261 * Unusually, this is also called in IRQ context (ghes_iounmap_irq) so if we
1262 * ever need to use IPIs for TLB broadcasting, then we're in trouble here.
1264 void __set_fixmap(enum fixed_addresses idx,
1265 phys_addr_t phys, pgprot_t flags)
1267 unsigned long addr = __fix_to_virt(idx);
1270 BUG_ON(idx <= FIX_HOLE || idx >= __end_of_fixed_addresses);
1272 ptep = fixmap_pte(addr);
1274 if (pgprot_val(flags)) {
1275 set_pte(ptep, pfn_pte(phys >> PAGE_SHIFT, flags));
1277 pte_clear(&init_mm, addr, ptep);
1278 flush_tlb_kernel_range(addr, addr+PAGE_SIZE);
1282 void *__init fixmap_remap_fdt(phys_addr_t dt_phys, int *size, pgprot_t prot)
1284 const u64 dt_virt_base = __fix_to_virt(FIX_FDT);
1289 * Check whether the physical FDT address is set and meets the minimum
1290 * alignment requirement. Since we are relying on MIN_FDT_ALIGN to be
1291 * at least 8 bytes so that we can always access the magic and size
1292 * fields of the FDT header after mapping the first chunk, double check
1293 * here if that is indeed the case.
1295 BUILD_BUG_ON(MIN_FDT_ALIGN < 8);
1296 if (!dt_phys || dt_phys % MIN_FDT_ALIGN)
1300 * Make sure that the FDT region can be mapped without the need to
1301 * allocate additional translation table pages, so that it is safe
1302 * to call create_mapping_noalloc() this early.
1304 * On 64k pages, the FDT will be mapped using PTEs, so we need to
1305 * be in the same PMD as the rest of the fixmap.
1306 * On 4k pages, we'll use section mappings for the FDT so we only
1307 * have to be in the same PUD.
1309 BUILD_BUG_ON(dt_virt_base % SZ_2M);
1311 BUILD_BUG_ON(__fix_to_virt(FIX_FDT_END) >> SWAPPER_TABLE_SHIFT !=
1312 __fix_to_virt(FIX_BTMAP_BEGIN) >> SWAPPER_TABLE_SHIFT);
1314 offset = dt_phys % SWAPPER_BLOCK_SIZE;
1315 dt_virt = (void *)dt_virt_base + offset;
1317 /* map the first chunk so we can read the size from the header */
1318 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE),
1319 dt_virt_base, SWAPPER_BLOCK_SIZE, prot);
1321 if (fdt_magic(dt_virt) != FDT_MAGIC)
1324 *size = fdt_totalsize(dt_virt);
1325 if (*size > MAX_FDT_SIZE)
1328 if (offset + *size > SWAPPER_BLOCK_SIZE)
1329 create_mapping_noalloc(round_down(dt_phys, SWAPPER_BLOCK_SIZE), dt_virt_base,
1330 round_up(offset + *size, SWAPPER_BLOCK_SIZE), prot);
1335 int pud_set_huge(pud_t *pudp, phys_addr_t phys, pgprot_t prot)
1337 pud_t new_pud = pfn_pud(__phys_to_pfn(phys), mk_pud_sect_prot(prot));
1339 /* Only allow permission changes for now */
1340 if (!pgattr_change_is_safe(READ_ONCE(pud_val(*pudp)),
1344 VM_BUG_ON(phys & ~PUD_MASK);
1345 set_pud(pudp, new_pud);
1349 int pmd_set_huge(pmd_t *pmdp, phys_addr_t phys, pgprot_t prot)
1351 pmd_t new_pmd = pfn_pmd(__phys_to_pfn(phys), mk_pmd_sect_prot(prot));
1353 /* Only allow permission changes for now */
1354 if (!pgattr_change_is_safe(READ_ONCE(pmd_val(*pmdp)),
1358 VM_BUG_ON(phys & ~PMD_MASK);
1359 set_pmd(pmdp, new_pmd);
1363 int pud_clear_huge(pud_t *pudp)
1365 if (!pud_sect(READ_ONCE(*pudp)))
1371 int pmd_clear_huge(pmd_t *pmdp)
1373 if (!pmd_sect(READ_ONCE(*pmdp)))
1379 int pmd_free_pte_page(pmd_t *pmdp, unsigned long addr)
1384 pmd = READ_ONCE(*pmdp);
1386 if (!pmd_table(pmd)) {
1391 table = pte_offset_kernel(pmdp, addr);
1393 __flush_tlb_kernel_pgtable(addr);
1394 pte_free_kernel(NULL, table);
1398 int pud_free_pmd_page(pud_t *pudp, unsigned long addr)
1403 unsigned long next, end;
1405 pud = READ_ONCE(*pudp);
1407 if (!pud_table(pud)) {
1412 table = pmd_offset(pudp, addr);
1415 end = addr + PUD_SIZE;
1417 pmd_free_pte_page(pmdp, next);
1418 } while (pmdp++, next += PMD_SIZE, next != end);
1421 __flush_tlb_kernel_pgtable(addr);
1422 pmd_free(NULL, table);
1426 #ifdef CONFIG_MEMORY_HOTPLUG
1427 static void __remove_pgd_mapping(pgd_t *pgdir, unsigned long start, u64 size)
1429 unsigned long end = start + size;
1431 WARN_ON(pgdir != init_mm.pgd);
1432 WARN_ON((start < PAGE_OFFSET) || (end > PAGE_END));
1434 unmap_hotplug_range(start, end, false, NULL);
1435 free_empty_tables(start, end, PAGE_OFFSET, PAGE_END);
1438 struct range arch_get_mappable_range(void)
1440 struct range mhp_range;
1441 u64 start_linear_pa = __pa(_PAGE_OFFSET(vabits_actual));
1442 u64 end_linear_pa = __pa(PAGE_END - 1);
1444 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE)) {
1446 * Check for a wrap, it is possible because of randomized linear
1447 * mapping the start physical address is actually bigger than
1448 * the end physical address. In this case set start to zero
1449 * because [0, end_linear_pa] range must still be able to cover
1450 * all addressable physical addresses.
1452 if (start_linear_pa > end_linear_pa)
1453 start_linear_pa = 0;
1456 WARN_ON(start_linear_pa > end_linear_pa);
1459 * Linear mapping region is the range [PAGE_OFFSET..(PAGE_END - 1)]
1460 * accommodating both its ends but excluding PAGE_END. Max physical
1461 * range which can be mapped inside this linear mapping range, must
1462 * also be derived from its end points.
1464 mhp_range.start = start_linear_pa;
1465 mhp_range.end = end_linear_pa;
1470 int arch_add_memory(int nid, u64 start, u64 size,
1471 struct mhp_params *params)
1473 int ret, flags = NO_EXEC_MAPPINGS;
1475 VM_BUG_ON(!mhp_range_allowed(start, size, true));
1477 if (can_set_direct_map())
1478 flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;
1480 __create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
1481 size, params->pgprot, __pgd_pgtable_alloc,
1484 memblock_clear_nomap(start, size);
1486 ret = __add_pages(nid, start >> PAGE_SHIFT, size >> PAGE_SHIFT,
1489 __remove_pgd_mapping(swapper_pg_dir,
1490 __phys_to_virt(start), size);
1492 max_pfn = PFN_UP(start + size);
1493 max_low_pfn = max_pfn;
1499 void arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
1501 unsigned long start_pfn = start >> PAGE_SHIFT;
1502 unsigned long nr_pages = size >> PAGE_SHIFT;
1504 __remove_pages(start_pfn, nr_pages, altmap);
1505 __remove_pgd_mapping(swapper_pg_dir, __phys_to_virt(start), size);
1509 * This memory hotplug notifier helps prevent boot memory from being
1510 * inadvertently removed as it blocks pfn range offlining process in
1511 * __offline_pages(). Hence this prevents both offlining as well as
1512 * removal process for boot memory which is initially always online.
1513 * In future if and when boot memory could be removed, this notifier
1514 * should be dropped and free_hotplug_page_range() should handle any
1515 * reserved pages allocated during boot.
1517 static int prevent_bootmem_remove_notifier(struct notifier_block *nb,
1518 unsigned long action, void *data)
1520 struct mem_section *ms;
1521 struct memory_notify *arg = data;
1522 unsigned long end_pfn = arg->start_pfn + arg->nr_pages;
1523 unsigned long pfn = arg->start_pfn;
1525 if ((action != MEM_GOING_OFFLINE) && (action != MEM_OFFLINE))
1528 for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1529 unsigned long start = PFN_PHYS(pfn);
1530 unsigned long end = start + (1UL << PA_SECTION_SHIFT);
1532 ms = __pfn_to_section(pfn);
1533 if (!early_section(ms))
1536 if (action == MEM_GOING_OFFLINE) {
1538 * Boot memory removal is not supported. Prevent
1539 * it via blocking any attempted offline request
1540 * for the boot memory and just report it.
1542 pr_warn("Boot memory [%lx %lx] offlining attempted\n", start, end);
1544 } else if (action == MEM_OFFLINE) {
1546 * This should have never happened. Boot memory
1547 * offlining should have been prevented by this
1548 * very notifier. Probably some memory removal
1549 * procedure might have changed which would then
1550 * require further debug.
1552 pr_err("Boot memory [%lx %lx] offlined\n", start, end);
1555 * Core memory hotplug does not process a return
1556 * code from the notifier for MEM_OFFLINE events.
1557 * The error condition has been reported. Return
1558 * from here as if ignored.
1566 static struct notifier_block prevent_bootmem_remove_nb = {
1567 .notifier_call = prevent_bootmem_remove_notifier,
1571 * This ensures that boot memory sections on the platform are online
1572 * from early boot. Memory sections could not be prevented from being
1573 * offlined, unless for some reason they are not online to begin with.
1574 * This helps validate the basic assumption on which the above memory
1575 * event notifier works to prevent boot memory section offlining and
1576 * its possible removal.
1578 static void validate_bootmem_online(void)
1580 phys_addr_t start, end, addr;
1581 struct mem_section *ms;
1585 * Scanning across all memblock might be expensive
1586 * on some big memory systems. Hence enable this
1587 * validation only with DEBUG_VM.
1589 if (!IS_ENABLED(CONFIG_DEBUG_VM))
1592 for_each_mem_range(i, &start, &end) {
1593 for (addr = start; addr < end; addr += (1UL << PA_SECTION_SHIFT)) {
1594 ms = __pfn_to_section(PHYS_PFN(addr));
1597 * All memory ranges in the system at this point
1598 * should have been marked as early sections.
1600 WARN_ON(!early_section(ms));
1603 * Memory notifier mechanism here to prevent boot
1604 * memory offlining depends on the fact that each
1605 * early section memory on the system is initially
1606 * online. Otherwise a given memory section which
1607 * is already offline will be overlooked and can
1608 * be removed completely. Call out such sections.
1610 if (!online_section(ms))
1611 pr_err("Boot memory [%llx %llx] is offline, can be removed\n",
1612 addr, addr + (1UL << PA_SECTION_SHIFT));
1617 static int __init prevent_bootmem_remove_init(void)
1621 if (!IS_ENABLED(CONFIG_MEMORY_HOTREMOVE))
1624 validate_bootmem_online();
1625 ret = register_memory_notifier(&prevent_bootmem_remove_nb);
1627 pr_err("%s: Notifier registration failed %d\n", __func__, ret);
1631 early_initcall(prevent_bootmem_remove_init);