1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/arch/x86_64/mm/init.c
5 * Copyright (C) 1995 Linus Torvalds
6 * Copyright (C) 2000 Pavel Machek <pavel@ucw.cz>
7 * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
10 #include <linux/signal.h>
11 #include <linux/sched.h>
12 #include <linux/kernel.h>
13 #include <linux/errno.h>
14 #include <linux/string.h>
15 #include <linux/types.h>
16 #include <linux/ptrace.h>
17 #include <linux/mman.h>
19 #include <linux/swap.h>
20 #include <linux/smp.h>
21 #include <linux/init.h>
22 #include <linux/initrd.h>
23 #include <linux/pagemap.h>
24 #include <linux/memblock.h>
25 #include <linux/proc_fs.h>
26 #include <linux/pci.h>
27 #include <linux/pfn.h>
28 #include <linux/poison.h>
29 #include <linux/dma-mapping.h>
30 #include <linux/memory.h>
31 #include <linux/memory_hotplug.h>
32 #include <linux/memremap.h>
33 #include <linux/nmi.h>
34 #include <linux/gfp.h>
35 #include <linux/kcore.h>
37 #include <asm/processor.h>
38 #include <asm/bios_ebda.h>
39 #include <linux/uaccess.h>
40 #include <asm/pgtable.h>
41 #include <asm/pgalloc.h>
43 #include <asm/fixmap.h>
44 #include <asm/e820/api.h>
47 #include <asm/mmu_context.h>
48 #include <asm/proto.h>
50 #include <asm/sections.h>
51 #include <asm/kdebug.h>
53 #include <asm/set_memory.h>
55 #include <asm/uv/uv.h>
56 #include <asm/setup.h>
58 #include "mm_internal.h"
60 #include "ident_map.c"
62 #define DEFINE_POPULATE(fname, type1, type2, init) \
63 static inline void fname##_init(struct mm_struct *mm, \
64 type1##_t *arg1, type2##_t *arg2, bool init) \
67 fname##_safe(mm, arg1, arg2); \
69 fname(mm, arg1, arg2); \
72 DEFINE_POPULATE(p4d_populate, p4d, pud, init)
73 DEFINE_POPULATE(pgd_populate, pgd, p4d, init)
74 DEFINE_POPULATE(pud_populate, pud, pmd, init)
75 DEFINE_POPULATE(pmd_populate_kernel, pmd, pte, init)
77 #define DEFINE_ENTRY(type1, type2, init) \
78 static inline void set_##type1##_init(type1##_t *arg1, \
79 type2##_t arg2, bool init) \
82 set_##type1##_safe(arg1, arg2); \
84 set_##type1(arg1, arg2); \
87 DEFINE_ENTRY(p4d, p4d, init)
88 DEFINE_ENTRY(pud, pud, init)
89 DEFINE_ENTRY(pmd, pmd, init)
90 DEFINE_ENTRY(pte, pte, init)
94 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
95 * physical space so we can cache the place of the first one and move
96 * around without checking the pgd every time.
99 /* Bits supported by the hardware: */
100 pteval_t __supported_pte_mask __read_mostly = ~0;
101 /* Bits allowed in normal kernel mappings: */
102 pteval_t __default_kernel_pte_mask __read_mostly = ~0;
103 EXPORT_SYMBOL_GPL(__supported_pte_mask);
104 /* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
105 EXPORT_SYMBOL(__default_kernel_pte_mask);
107 int force_personality32;
111 * Control non executable heap for 32bit processes.
112 * To control the stack too use noexec=off
114 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
115 * off PROT_READ implies PROT_EXEC
117 static int __init nonx32_setup(char *str)
119 if (!strcmp(str, "on"))
120 force_personality32 &= ~READ_IMPLIES_EXEC;
121 else if (!strcmp(str, "off"))
122 force_personality32 |= READ_IMPLIES_EXEC;
125 __setup("noexec32=", nonx32_setup);
127 static void sync_global_pgds_l5(unsigned long start, unsigned long end)
131 for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
132 const pgd_t *pgd_ref = pgd_offset_k(addr);
135 /* Check for overflow */
139 if (pgd_none(*pgd_ref))
142 spin_lock(&pgd_lock);
143 list_for_each_entry(page, &pgd_list, lru) {
145 spinlock_t *pgt_lock;
147 pgd = (pgd_t *)page_address(page) + pgd_index(addr);
148 /* the pgt_lock only for Xen */
149 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
152 if (!pgd_none(*pgd_ref) && !pgd_none(*pgd))
153 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
156 set_pgd(pgd, *pgd_ref);
158 spin_unlock(pgt_lock);
160 spin_unlock(&pgd_lock);
164 static void sync_global_pgds_l4(unsigned long start, unsigned long end)
168 for (addr = start; addr <= end; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
169 pgd_t *pgd_ref = pgd_offset_k(addr);
170 const p4d_t *p4d_ref;
174 * With folded p4d, pgd_none() is always false, we need to
175 * handle synchonization on p4d level.
177 MAYBE_BUILD_BUG_ON(pgd_none(*pgd_ref));
178 p4d_ref = p4d_offset(pgd_ref, addr);
180 if (p4d_none(*p4d_ref))
183 spin_lock(&pgd_lock);
184 list_for_each_entry(page, &pgd_list, lru) {
187 spinlock_t *pgt_lock;
189 pgd = (pgd_t *)page_address(page) + pgd_index(addr);
190 p4d = p4d_offset(pgd, addr);
191 /* the pgt_lock only for Xen */
192 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
195 if (!p4d_none(*p4d_ref) && !p4d_none(*p4d))
196 BUG_ON(p4d_page_vaddr(*p4d)
197 != p4d_page_vaddr(*p4d_ref));
200 set_p4d(p4d, *p4d_ref);
202 spin_unlock(pgt_lock);
204 spin_unlock(&pgd_lock);
209 * When memory was added make sure all the processes MM have
210 * suitable PGD entries in the local PGD level page.
212 void sync_global_pgds(unsigned long start, unsigned long end)
214 if (pgtable_l5_enabled())
215 sync_global_pgds_l5(start, end);
217 sync_global_pgds_l4(start, end);
221 * NOTE: This function is marked __ref because it calls __init function
222 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
224 static __ref void *spp_getpage(void)
229 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
231 ptr = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
233 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
234 panic("set_pte_phys: cannot allocate page data %s\n",
235 after_bootmem ? "after bootmem" : "");
238 pr_debug("spp_getpage %p\n", ptr);
243 static p4d_t *fill_p4d(pgd_t *pgd, unsigned long vaddr)
245 if (pgd_none(*pgd)) {
246 p4d_t *p4d = (p4d_t *)spp_getpage();
247 pgd_populate(&init_mm, pgd, p4d);
248 if (p4d != p4d_offset(pgd, 0))
249 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
250 p4d, p4d_offset(pgd, 0));
252 return p4d_offset(pgd, vaddr);
255 static pud_t *fill_pud(p4d_t *p4d, unsigned long vaddr)
257 if (p4d_none(*p4d)) {
258 pud_t *pud = (pud_t *)spp_getpage();
259 p4d_populate(&init_mm, p4d, pud);
260 if (pud != pud_offset(p4d, 0))
261 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
262 pud, pud_offset(p4d, 0));
264 return pud_offset(p4d, vaddr);
267 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
269 if (pud_none(*pud)) {
270 pmd_t *pmd = (pmd_t *) spp_getpage();
271 pud_populate(&init_mm, pud, pmd);
272 if (pmd != pmd_offset(pud, 0))
273 printk(KERN_ERR "PAGETABLE BUG #02! %p <-> %p\n",
274 pmd, pmd_offset(pud, 0));
276 return pmd_offset(pud, vaddr);
279 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
281 if (pmd_none(*pmd)) {
282 pte_t *pte = (pte_t *) spp_getpage();
283 pmd_populate_kernel(&init_mm, pmd, pte);
284 if (pte != pte_offset_kernel(pmd, 0))
285 printk(KERN_ERR "PAGETABLE BUG #03!\n");
287 return pte_offset_kernel(pmd, vaddr);
290 static void __set_pte_vaddr(pud_t *pud, unsigned long vaddr, pte_t new_pte)
292 pmd_t *pmd = fill_pmd(pud, vaddr);
293 pte_t *pte = fill_pte(pmd, vaddr);
295 set_pte(pte, new_pte);
298 * It's enough to flush this one mapping.
299 * (PGE mappings get flushed as well)
301 __flush_tlb_one_kernel(vaddr);
304 void set_pte_vaddr_p4d(p4d_t *p4d_page, unsigned long vaddr, pte_t new_pte)
306 p4d_t *p4d = p4d_page + p4d_index(vaddr);
307 pud_t *pud = fill_pud(p4d, vaddr);
309 __set_pte_vaddr(pud, vaddr, new_pte);
312 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
314 pud_t *pud = pud_page + pud_index(vaddr);
316 __set_pte_vaddr(pud, vaddr, new_pte);
319 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
324 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
326 pgd = pgd_offset_k(vaddr);
327 if (pgd_none(*pgd)) {
329 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
333 p4d_page = p4d_offset(pgd, 0);
334 set_pte_vaddr_p4d(p4d_page, vaddr, pteval);
337 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
343 pgd = pgd_offset_k(vaddr);
344 p4d = fill_p4d(pgd, vaddr);
345 pud = fill_pud(p4d, vaddr);
346 return fill_pmd(pud, vaddr);
349 pte_t * __init populate_extra_pte(unsigned long vaddr)
353 pmd = populate_extra_pmd(vaddr);
354 return fill_pte(pmd, vaddr);
358 * Create large page table mappings for a range of physical addresses.
360 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
361 enum page_cache_mode cache)
369 pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
370 pgprot_val(pgprot_4k_2_large(cachemode2pgprot(cache)));
371 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
372 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
373 pgd = pgd_offset_k((unsigned long)__va(phys));
374 if (pgd_none(*pgd)) {
375 p4d = (p4d_t *) spp_getpage();
376 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE |
379 p4d = p4d_offset(pgd, (unsigned long)__va(phys));
380 if (p4d_none(*p4d)) {
381 pud = (pud_t *) spp_getpage();
382 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE |
385 pud = pud_offset(p4d, (unsigned long)__va(phys));
386 if (pud_none(*pud)) {
387 pmd = (pmd_t *) spp_getpage();
388 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
391 pmd = pmd_offset(pud, phys);
392 BUG_ON(!pmd_none(*pmd));
393 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
397 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
399 __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
402 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
404 __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
408 * The head.S code sets up the kernel high mapping:
410 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
412 * phys_base holds the negative offset to the kernel, which is added
413 * to the compile time generated pmds. This results in invalid pmds up
414 * to the point where we hit the physaddr 0 mapping.
416 * We limit the mappings to the region from _text to _brk_end. _brk_end
417 * is rounded up to the 2MB boundary. This catches the invalid pmds as
418 * well, as they are located before _text:
420 void __init cleanup_highmap(void)
422 unsigned long vaddr = __START_KERNEL_map;
423 unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
424 unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
425 pmd_t *pmd = level2_kernel_pgt;
428 * Native path, max_pfn_mapped is not set yet.
429 * Xen has valid max_pfn_mapped set in
430 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
433 vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
435 for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
438 if (vaddr < (unsigned long) _text || vaddr > end)
439 set_pmd(pmd, __pmd(0));
444 * Create PTE level page table mapping for physical addresses.
445 * It returns the last physical address mapped.
447 static unsigned long __meminit
448 phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
449 pgprot_t prot, bool init)
451 unsigned long pages = 0, paddr_next;
452 unsigned long paddr_last = paddr_end;
456 pte = pte_page + pte_index(paddr);
457 i = pte_index(paddr);
459 for (; i < PTRS_PER_PTE; i++, paddr = paddr_next, pte++) {
460 paddr_next = (paddr & PAGE_MASK) + PAGE_SIZE;
461 if (paddr >= paddr_end) {
462 if (!after_bootmem &&
463 !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
465 !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
466 E820_TYPE_RESERVED_KERN))
467 set_pte_init(pte, __pte(0), init);
472 * We will re-use the existing mapping.
473 * Xen for example has some special requirements, like mapping
474 * pagetable pages as RO. So assume someone who pre-setup
475 * these mappings are more intelligent.
477 if (!pte_none(*pte)) {
484 pr_info(" pte=%p addr=%lx pte=%016lx\n", pte, paddr,
485 pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL).pte);
487 set_pte_init(pte, pfn_pte(paddr >> PAGE_SHIFT, prot), init);
488 paddr_last = (paddr & PAGE_MASK) + PAGE_SIZE;
491 update_page_count(PG_LEVEL_4K, pages);
497 * Create PMD level page table mapping for physical addresses. The virtual
498 * and physical address have to be aligned at this level.
499 * It returns the last physical address mapped.
501 static unsigned long __meminit
502 phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
503 unsigned long page_size_mask, pgprot_t prot, bool init)
505 unsigned long pages = 0, paddr_next;
506 unsigned long paddr_last = paddr_end;
508 int i = pmd_index(paddr);
510 for (; i < PTRS_PER_PMD; i++, paddr = paddr_next) {
511 pmd_t *pmd = pmd_page + pmd_index(paddr);
513 pgprot_t new_prot = prot;
515 paddr_next = (paddr & PMD_MASK) + PMD_SIZE;
516 if (paddr >= paddr_end) {
517 if (!after_bootmem &&
518 !e820__mapped_any(paddr & PMD_MASK, paddr_next,
520 !e820__mapped_any(paddr & PMD_MASK, paddr_next,
521 E820_TYPE_RESERVED_KERN))
522 set_pmd_init(pmd, __pmd(0), init);
526 if (!pmd_none(*pmd)) {
527 if (!pmd_large(*pmd)) {
528 spin_lock(&init_mm.page_table_lock);
529 pte = (pte_t *)pmd_page_vaddr(*pmd);
530 paddr_last = phys_pte_init(pte, paddr,
533 spin_unlock(&init_mm.page_table_lock);
537 * If we are ok with PG_LEVEL_2M mapping, then we will
538 * use the existing mapping,
540 * Otherwise, we will split the large page mapping but
541 * use the same existing protection bits except for
542 * large page, so that we don't violate Intel's TLB
543 * Application note (317080) which says, while changing
544 * the page sizes, new and old translations should
545 * not differ with respect to page frame and
548 if (page_size_mask & (1 << PG_LEVEL_2M)) {
551 paddr_last = paddr_next;
554 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
557 if (page_size_mask & (1<<PG_LEVEL_2M)) {
559 spin_lock(&init_mm.page_table_lock);
560 set_pte_init((pte_t *)pmd,
561 pfn_pte((paddr & PMD_MASK) >> PAGE_SHIFT,
562 __pgprot(pgprot_val(prot) | _PAGE_PSE)),
564 spin_unlock(&init_mm.page_table_lock);
565 paddr_last = paddr_next;
569 pte = alloc_low_page();
570 paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot, init);
572 spin_lock(&init_mm.page_table_lock);
573 pmd_populate_kernel_init(&init_mm, pmd, pte, init);
574 spin_unlock(&init_mm.page_table_lock);
576 update_page_count(PG_LEVEL_2M, pages);
581 * Create PUD level page table mapping for physical addresses. The virtual
582 * and physical address do not have to be aligned at this level. KASLR can
583 * randomize virtual addresses up to this level.
584 * It returns the last physical address mapped.
586 static unsigned long __meminit
587 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
588 unsigned long page_size_mask, bool init)
590 unsigned long pages = 0, paddr_next;
591 unsigned long paddr_last = paddr_end;
592 unsigned long vaddr = (unsigned long)__va(paddr);
593 int i = pud_index(vaddr);
595 for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
598 pgprot_t prot = PAGE_KERNEL;
600 vaddr = (unsigned long)__va(paddr);
601 pud = pud_page + pud_index(vaddr);
602 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
604 if (paddr >= paddr_end) {
605 if (!after_bootmem &&
606 !e820__mapped_any(paddr & PUD_MASK, paddr_next,
608 !e820__mapped_any(paddr & PUD_MASK, paddr_next,
609 E820_TYPE_RESERVED_KERN))
610 set_pud_init(pud, __pud(0), init);
614 if (!pud_none(*pud)) {
615 if (!pud_large(*pud)) {
616 pmd = pmd_offset(pud, 0);
617 paddr_last = phys_pmd_init(pmd, paddr,
624 * If we are ok with PG_LEVEL_1G mapping, then we will
625 * use the existing mapping.
627 * Otherwise, we will split the gbpage mapping but use
628 * the same existing protection bits except for large
629 * page, so that we don't violate Intel's TLB
630 * Application note (317080) which says, while changing
631 * the page sizes, new and old translations should
632 * not differ with respect to page frame and
635 if (page_size_mask & (1 << PG_LEVEL_1G)) {
638 paddr_last = paddr_next;
641 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
644 if (page_size_mask & (1<<PG_LEVEL_1G)) {
646 spin_lock(&init_mm.page_table_lock);
647 set_pte_init((pte_t *)pud,
648 pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
651 spin_unlock(&init_mm.page_table_lock);
652 paddr_last = paddr_next;
656 pmd = alloc_low_page();
657 paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
658 page_size_mask, prot, init);
660 spin_lock(&init_mm.page_table_lock);
661 pud_populate_init(&init_mm, pud, pmd, init);
662 spin_unlock(&init_mm.page_table_lock);
665 update_page_count(PG_LEVEL_1G, pages);
670 static unsigned long __meminit
671 phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
672 unsigned long page_size_mask, bool init)
674 unsigned long vaddr, vaddr_end, vaddr_next, paddr_next, paddr_last;
676 paddr_last = paddr_end;
677 vaddr = (unsigned long)__va(paddr);
678 vaddr_end = (unsigned long)__va(paddr_end);
680 if (!pgtable_l5_enabled())
681 return phys_pud_init((pud_t *) p4d_page, paddr, paddr_end,
682 page_size_mask, init);
684 for (; vaddr < vaddr_end; vaddr = vaddr_next) {
685 p4d_t *p4d = p4d_page + p4d_index(vaddr);
688 vaddr_next = (vaddr & P4D_MASK) + P4D_SIZE;
691 if (paddr >= paddr_end) {
692 paddr_next = __pa(vaddr_next);
693 if (!after_bootmem &&
694 !e820__mapped_any(paddr & P4D_MASK, paddr_next,
696 !e820__mapped_any(paddr & P4D_MASK, paddr_next,
697 E820_TYPE_RESERVED_KERN))
698 set_p4d_init(p4d, __p4d(0), init);
702 if (!p4d_none(*p4d)) {
703 pud = pud_offset(p4d, 0);
704 paddr_last = phys_pud_init(pud, paddr, __pa(vaddr_end),
705 page_size_mask, init);
709 pud = alloc_low_page();
710 paddr_last = phys_pud_init(pud, paddr, __pa(vaddr_end),
711 page_size_mask, init);
713 spin_lock(&init_mm.page_table_lock);
714 p4d_populate_init(&init_mm, p4d, pud, init);
715 spin_unlock(&init_mm.page_table_lock);
721 static unsigned long __meminit
722 __kernel_physical_mapping_init(unsigned long paddr_start,
723 unsigned long paddr_end,
724 unsigned long page_size_mask,
727 bool pgd_changed = false;
728 unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
730 paddr_last = paddr_end;
731 vaddr = (unsigned long)__va(paddr_start);
732 vaddr_end = (unsigned long)__va(paddr_end);
735 for (; vaddr < vaddr_end; vaddr = vaddr_next) {
736 pgd_t *pgd = pgd_offset_k(vaddr);
739 vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
742 p4d = (p4d_t *)pgd_page_vaddr(*pgd);
743 paddr_last = phys_p4d_init(p4d, __pa(vaddr),
750 p4d = alloc_low_page();
751 paddr_last = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
752 page_size_mask, init);
754 spin_lock(&init_mm.page_table_lock);
755 if (pgtable_l5_enabled())
756 pgd_populate_init(&init_mm, pgd, p4d, init);
758 p4d_populate_init(&init_mm, p4d_offset(pgd, vaddr),
759 (pud_t *) p4d, init);
761 spin_unlock(&init_mm.page_table_lock);
766 sync_global_pgds(vaddr_start, vaddr_end - 1);
773 * Create page table mapping for the physical memory for specific physical
774 * addresses. Note that it can only be used to populate non-present entries.
775 * The virtual and physical addresses have to be aligned on PMD level
776 * down. It returns the last physical address mapped.
778 unsigned long __meminit
779 kernel_physical_mapping_init(unsigned long paddr_start,
780 unsigned long paddr_end,
781 unsigned long page_size_mask)
783 return __kernel_physical_mapping_init(paddr_start, paddr_end,
784 page_size_mask, true);
788 * This function is similar to kernel_physical_mapping_init() above with the
789 * exception that it uses set_{pud,pmd}() instead of the set_{pud,pte}_safe()
790 * when updating the mapping. The caller is responsible to flush the TLBs after
791 * the function returns.
793 unsigned long __meminit
794 kernel_physical_mapping_change(unsigned long paddr_start,
795 unsigned long paddr_end,
796 unsigned long page_size_mask)
798 return __kernel_physical_mapping_init(paddr_start, paddr_end,
799 page_size_mask, false);
803 void __init initmem_init(void)
805 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
809 void __init paging_init(void)
811 sparse_memory_present_with_active_regions(MAX_NUMNODES);
815 * clear the default setting with node 0
816 * note: don't use nodes_clear here, that is really clearing when
817 * numa support is not compiled in, and later node_set_state
818 * will not set it back.
820 node_clear_state(0, N_MEMORY);
821 node_clear_state(0, N_NORMAL_MEMORY);
827 * Memory hotplug specific functions
829 #ifdef CONFIG_MEMORY_HOTPLUG
831 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
834 static void update_end_of_memory_vars(u64 start, u64 size)
836 unsigned long end_pfn = PFN_UP(start + size);
838 if (end_pfn > max_pfn) {
840 max_low_pfn = end_pfn;
841 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
845 int add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
846 struct mhp_restrictions *restrictions)
850 ret = __add_pages(nid, start_pfn, nr_pages, restrictions);
853 /* update max_pfn, max_low_pfn and high_memory */
854 update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
855 nr_pages << PAGE_SHIFT);
860 int arch_add_memory(int nid, u64 start, u64 size,
861 struct mhp_restrictions *restrictions)
863 unsigned long start_pfn = start >> PAGE_SHIFT;
864 unsigned long nr_pages = size >> PAGE_SHIFT;
866 init_memory_mapping(start, start + size);
868 return add_pages(nid, start_pfn, nr_pages, restrictions);
871 #define PAGE_INUSE 0xFD
873 static void __meminit free_pagetable(struct page *page, int order)
876 unsigned int nr_pages = 1 << order;
878 /* bootmem page has reserved flag */
879 if (PageReserved(page)) {
880 __ClearPageReserved(page);
882 magic = (unsigned long)page->freelist;
883 if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
885 put_page_bootmem(page++);
888 free_reserved_page(page++);
890 free_pages((unsigned long)page_address(page), order);
893 static void __meminit free_hugepage_table(struct page *page,
894 struct vmem_altmap *altmap)
897 vmem_altmap_free(altmap, PMD_SIZE / PAGE_SIZE);
899 free_pagetable(page, get_order(PMD_SIZE));
902 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
907 for (i = 0; i < PTRS_PER_PTE; i++) {
913 /* free a pte talbe */
914 free_pagetable(pmd_page(*pmd), 0);
915 spin_lock(&init_mm.page_table_lock);
917 spin_unlock(&init_mm.page_table_lock);
920 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
925 for (i = 0; i < PTRS_PER_PMD; i++) {
931 /* free a pmd talbe */
932 free_pagetable(pud_page(*pud), 0);
933 spin_lock(&init_mm.page_table_lock);
935 spin_unlock(&init_mm.page_table_lock);
938 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
943 for (i = 0; i < PTRS_PER_PUD; i++) {
949 /* free a pud talbe */
950 free_pagetable(p4d_page(*p4d), 0);
951 spin_lock(&init_mm.page_table_lock);
953 spin_unlock(&init_mm.page_table_lock);
956 static void __meminit
957 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
960 unsigned long next, pages = 0;
963 phys_addr_t phys_addr;
965 pte = pte_start + pte_index(addr);
966 for (; addr < end; addr = next, pte++) {
967 next = (addr + PAGE_SIZE) & PAGE_MASK;
971 if (!pte_present(*pte))
975 * We mapped [0,1G) memory as identity mapping when
976 * initializing, in arch/x86/kernel/head_64.S. These
977 * pagetables cannot be removed.
979 phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
980 if (phys_addr < (phys_addr_t)0x40000000)
983 if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
985 * Do not free direct mapping pages since they were
986 * freed when offlining, or simplely not in use.
989 free_pagetable(pte_page(*pte), 0);
991 spin_lock(&init_mm.page_table_lock);
992 pte_clear(&init_mm, addr, pte);
993 spin_unlock(&init_mm.page_table_lock);
995 /* For non-direct mapping, pages means nothing. */
999 * If we are here, we are freeing vmemmap pages since
1000 * direct mapped memory ranges to be freed are aligned.
1002 * If we are not removing the whole page, it means
1003 * other page structs in this page are being used and
1004 * we canot remove them. So fill the unused page_structs
1005 * with 0xFD, and remove the page when it is wholly
1008 memset((void *)addr, PAGE_INUSE, next - addr);
1010 page_addr = page_address(pte_page(*pte));
1011 if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
1012 free_pagetable(pte_page(*pte), 0);
1014 spin_lock(&init_mm.page_table_lock);
1015 pte_clear(&init_mm, addr, pte);
1016 spin_unlock(&init_mm.page_table_lock);
1021 /* Call free_pte_table() in remove_pmd_table(). */
1024 update_page_count(PG_LEVEL_4K, -pages);
1027 static void __meminit
1028 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
1029 bool direct, struct vmem_altmap *altmap)
1031 unsigned long next, pages = 0;
1036 pmd = pmd_start + pmd_index(addr);
1037 for (; addr < end; addr = next, pmd++) {
1038 next = pmd_addr_end(addr, end);
1040 if (!pmd_present(*pmd))
1043 if (pmd_large(*pmd)) {
1044 if (IS_ALIGNED(addr, PMD_SIZE) &&
1045 IS_ALIGNED(next, PMD_SIZE)) {
1047 free_hugepage_table(pmd_page(*pmd),
1050 spin_lock(&init_mm.page_table_lock);
1052 spin_unlock(&init_mm.page_table_lock);
1055 /* If here, we are freeing vmemmap pages. */
1056 memset((void *)addr, PAGE_INUSE, next - addr);
1058 page_addr = page_address(pmd_page(*pmd));
1059 if (!memchr_inv(page_addr, PAGE_INUSE,
1061 free_hugepage_table(pmd_page(*pmd),
1064 spin_lock(&init_mm.page_table_lock);
1066 spin_unlock(&init_mm.page_table_lock);
1073 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
1074 remove_pte_table(pte_base, addr, next, direct);
1075 free_pte_table(pte_base, pmd);
1078 /* Call free_pmd_table() in remove_pud_table(). */
1080 update_page_count(PG_LEVEL_2M, -pages);
1083 static void __meminit
1084 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
1085 struct vmem_altmap *altmap, bool direct)
1087 unsigned long next, pages = 0;
1092 pud = pud_start + pud_index(addr);
1093 for (; addr < end; addr = next, pud++) {
1094 next = pud_addr_end(addr, end);
1096 if (!pud_present(*pud))
1099 if (pud_large(*pud)) {
1100 if (IS_ALIGNED(addr, PUD_SIZE) &&
1101 IS_ALIGNED(next, PUD_SIZE)) {
1103 free_pagetable(pud_page(*pud),
1104 get_order(PUD_SIZE));
1106 spin_lock(&init_mm.page_table_lock);
1108 spin_unlock(&init_mm.page_table_lock);
1111 /* If here, we are freeing vmemmap pages. */
1112 memset((void *)addr, PAGE_INUSE, next - addr);
1114 page_addr = page_address(pud_page(*pud));
1115 if (!memchr_inv(page_addr, PAGE_INUSE,
1117 free_pagetable(pud_page(*pud),
1118 get_order(PUD_SIZE));
1120 spin_lock(&init_mm.page_table_lock);
1122 spin_unlock(&init_mm.page_table_lock);
1129 pmd_base = pmd_offset(pud, 0);
1130 remove_pmd_table(pmd_base, addr, next, direct, altmap);
1131 free_pmd_table(pmd_base, pud);
1135 update_page_count(PG_LEVEL_1G, -pages);
1138 static void __meminit
1139 remove_p4d_table(p4d_t *p4d_start, unsigned long addr, unsigned long end,
1140 struct vmem_altmap *altmap, bool direct)
1142 unsigned long next, pages = 0;
1146 p4d = p4d_start + p4d_index(addr);
1147 for (; addr < end; addr = next, p4d++) {
1148 next = p4d_addr_end(addr, end);
1150 if (!p4d_present(*p4d))
1153 BUILD_BUG_ON(p4d_large(*p4d));
1155 pud_base = pud_offset(p4d, 0);
1156 remove_pud_table(pud_base, addr, next, altmap, direct);
1158 * For 4-level page tables we do not want to free PUDs, but in the
1159 * 5-level case we should free them. This code will have to change
1160 * to adapt for boot-time switching between 4 and 5 level page tables.
1162 if (pgtable_l5_enabled())
1163 free_pud_table(pud_base, p4d);
1167 update_page_count(PG_LEVEL_512G, -pages);
1170 /* start and end are both virtual address. */
1171 static void __meminit
1172 remove_pagetable(unsigned long start, unsigned long end, bool direct,
1173 struct vmem_altmap *altmap)
1180 for (addr = start; addr < end; addr = next) {
1181 next = pgd_addr_end(addr, end);
1183 pgd = pgd_offset_k(addr);
1184 if (!pgd_present(*pgd))
1187 p4d = p4d_offset(pgd, 0);
1188 remove_p4d_table(p4d, addr, next, altmap, direct);
1194 void __ref vmemmap_free(unsigned long start, unsigned long end,
1195 struct vmem_altmap *altmap)
1197 remove_pagetable(start, end, false, altmap);
1200 static void __meminit
1201 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
1203 start = (unsigned long)__va(start);
1204 end = (unsigned long)__va(end);
1206 remove_pagetable(start, end, true, NULL);
1209 void __ref arch_remove_memory(int nid, u64 start, u64 size,
1210 struct vmem_altmap *altmap)
1212 unsigned long start_pfn = start >> PAGE_SHIFT;
1213 unsigned long nr_pages = size >> PAGE_SHIFT;
1215 __remove_pages(start_pfn, nr_pages, altmap);
1216 kernel_physical_mapping_remove(start, start + size);
1218 #endif /* CONFIG_MEMORY_HOTPLUG */
1220 static struct kcore_list kcore_vsyscall;
1222 static void __init register_page_bootmem_info(void)
1227 for_each_online_node(i)
1228 register_page_bootmem_info_node(NODE_DATA(i));
1232 void __init mem_init(void)
1236 /* clear_bss() already clear the empty_zero_page */
1238 /* this will put all memory onto the freelists */
1239 memblock_free_all();
1241 x86_init.hyper.init_after_bootmem();
1244 * Must be done after boot memory is put on freelist, because here we
1245 * might set fields in deferred struct pages that have not yet been
1246 * initialized, and memblock_free_all() initializes all the reserved
1247 * deferred pages for us.
1249 register_page_bootmem_info();
1251 /* Register memory areas for /proc/kcore */
1252 if (get_gate_vma(&init_mm))
1253 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR, PAGE_SIZE, KCORE_USER);
1255 mem_init_print_info(NULL);
1258 int kernel_set_to_readonly;
1260 void mark_rodata_ro(void)
1262 unsigned long start = PFN_ALIGN(_text);
1263 unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1264 unsigned long end = (unsigned long)__end_rodata_hpage_align;
1265 unsigned long text_end = PFN_ALIGN(_etext);
1266 unsigned long rodata_end = PFN_ALIGN(__end_rodata);
1267 unsigned long all_end;
1269 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1270 (end - start) >> 10);
1271 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1273 kernel_set_to_readonly = 1;
1276 * The rodata/data/bss/brk section (but not the kernel text!)
1277 * should also be not-executable.
1279 * We align all_end to PMD_SIZE because the existing mapping
1280 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1281 * split the PMD and the reminder between _brk_end and the end
1282 * of the PMD will remain mapped executable.
1284 * Any PMD which was setup after the one which covers _brk_end
1285 * has been zapped already via cleanup_highmem().
1287 all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1288 set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1290 #ifdef CONFIG_CPA_DEBUG
1291 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1292 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1294 printk(KERN_INFO "Testing CPA: again\n");
1295 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1298 free_kernel_image_pages("unused kernel image (text/rodata gap)",
1299 (void *)text_end, (void *)rodata_start);
1300 free_kernel_image_pages("unused kernel image (rodata/data gap)",
1301 (void *)rodata_end, (void *)_sdata);
1306 int kern_addr_valid(unsigned long addr)
1308 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1315 if (above != 0 && above != -1UL)
1318 pgd = pgd_offset_k(addr);
1322 p4d = p4d_offset(pgd, addr);
1326 pud = pud_offset(p4d, addr);
1330 if (pud_large(*pud))
1331 return pfn_valid(pud_pfn(*pud));
1333 pmd = pmd_offset(pud, addr);
1337 if (pmd_large(*pmd))
1338 return pfn_valid(pmd_pfn(*pmd));
1340 pte = pte_offset_kernel(pmd, addr);
1344 return pfn_valid(pte_pfn(*pte));
1348 * Block size is the minimum amount of memory which can be hotplugged or
1349 * hotremoved. It must be power of two and must be equal or larger than
1350 * MIN_MEMORY_BLOCK_SIZE.
1352 #define MAX_BLOCK_SIZE (2UL << 30)
1354 /* Amount of ram needed to start using large blocks */
1355 #define MEM_SIZE_FOR_LARGE_BLOCK (64UL << 30)
1357 /* Adjustable memory block size */
1358 static unsigned long set_memory_block_size;
1359 int __init set_memory_block_size_order(unsigned int order)
1361 unsigned long size = 1UL << order;
1363 if (size > MEM_SIZE_FOR_LARGE_BLOCK || size < MIN_MEMORY_BLOCK_SIZE)
1366 set_memory_block_size = size;
1370 static unsigned long probe_memory_block_size(void)
1372 unsigned long boot_mem_end = max_pfn << PAGE_SHIFT;
1375 /* If memory block size has been set, then use it */
1376 bz = set_memory_block_size;
1380 /* Use regular block if RAM is smaller than MEM_SIZE_FOR_LARGE_BLOCK */
1381 if (boot_mem_end < MEM_SIZE_FOR_LARGE_BLOCK) {
1382 bz = MIN_MEMORY_BLOCK_SIZE;
1386 /* Find the largest allowed block size that aligns to memory end */
1387 for (bz = MAX_BLOCK_SIZE; bz > MIN_MEMORY_BLOCK_SIZE; bz >>= 1) {
1388 if (IS_ALIGNED(boot_mem_end, bz))
1392 pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1397 static unsigned long memory_block_size_probed;
1398 unsigned long memory_block_size_bytes(void)
1400 if (!memory_block_size_probed)
1401 memory_block_size_probed = probe_memory_block_size();
1403 return memory_block_size_probed;
1406 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1408 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1410 static long __meminitdata addr_start, addr_end;
1411 static void __meminitdata *p_start, *p_end;
1412 static int __meminitdata node_start;
1414 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1415 unsigned long end, int node, struct vmem_altmap *altmap)
1424 for (addr = start; addr < end; addr = next) {
1425 next = pmd_addr_end(addr, end);
1427 pgd = vmemmap_pgd_populate(addr, node);
1431 p4d = vmemmap_p4d_populate(pgd, addr, node);
1435 pud = vmemmap_pud_populate(p4d, addr, node);
1439 pmd = pmd_offset(pud, addr);
1440 if (pmd_none(*pmd)) {
1444 p = altmap_alloc_block_buf(PMD_SIZE, altmap);
1446 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
1450 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1452 set_pmd(pmd, __pmd(pte_val(entry)));
1454 /* check to see if we have contiguous blocks */
1455 if (p_end != p || node_start != node) {
1457 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1458 addr_start, addr_end-1, p_start, p_end-1, node_start);
1464 addr_end = addr + PMD_SIZE;
1465 p_end = p + PMD_SIZE;
1468 return -ENOMEM; /* no fallback */
1469 } else if (pmd_large(*pmd)) {
1470 vmemmap_verify((pte_t *)pmd, node, addr, next);
1473 if (vmemmap_populate_basepages(addr, next, node))
1479 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1480 struct vmem_altmap *altmap)
1484 if (end - start < PAGES_PER_SECTION * sizeof(struct page))
1485 err = vmemmap_populate_basepages(start, end, node);
1486 else if (boot_cpu_has(X86_FEATURE_PSE))
1487 err = vmemmap_populate_hugepages(start, end, node, altmap);
1489 pr_err_once("%s: no cpu support for altmap allocations\n",
1493 err = vmemmap_populate_basepages(start, end, node);
1495 sync_global_pgds(start, end - 1);
1499 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1500 void register_page_bootmem_memmap(unsigned long section_nr,
1501 struct page *start_page, unsigned long nr_pages)
1503 unsigned long addr = (unsigned long)start_page;
1504 unsigned long end = (unsigned long)(start_page + nr_pages);
1510 unsigned int nr_pmd_pages;
1513 for (; addr < end; addr = next) {
1516 pgd = pgd_offset_k(addr);
1517 if (pgd_none(*pgd)) {
1518 next = (addr + PAGE_SIZE) & PAGE_MASK;
1521 get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1523 p4d = p4d_offset(pgd, addr);
1524 if (p4d_none(*p4d)) {
1525 next = (addr + PAGE_SIZE) & PAGE_MASK;
1528 get_page_bootmem(section_nr, p4d_page(*p4d), MIX_SECTION_INFO);
1530 pud = pud_offset(p4d, addr);
1531 if (pud_none(*pud)) {
1532 next = (addr + PAGE_SIZE) & PAGE_MASK;
1535 get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1537 if (!boot_cpu_has(X86_FEATURE_PSE)) {
1538 next = (addr + PAGE_SIZE) & PAGE_MASK;
1539 pmd = pmd_offset(pud, addr);
1542 get_page_bootmem(section_nr, pmd_page(*pmd),
1545 pte = pte_offset_kernel(pmd, addr);
1548 get_page_bootmem(section_nr, pte_page(*pte),
1551 next = pmd_addr_end(addr, end);
1553 pmd = pmd_offset(pud, addr);
1557 nr_pmd_pages = 1 << get_order(PMD_SIZE);
1558 page = pmd_page(*pmd);
1559 while (nr_pmd_pages--)
1560 get_page_bootmem(section_nr, page++,
1567 void __meminit vmemmap_populate_print_last(void)
1570 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1571 addr_start, addr_end-1, p_start, p_end-1, node_start);