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/pgalloc.h>
42 #include <asm/fixmap.h>
43 #include <asm/e820/api.h>
46 #include <asm/mmu_context.h>
47 #include <asm/proto.h>
49 #include <asm/sections.h>
50 #include <asm/kdebug.h>
52 #include <asm/set_memory.h>
54 #include <asm/uv/uv.h>
55 #include <asm/setup.h>
56 #include <asm/ftrace.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 static 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);
220 void arch_sync_kernel_mappings(unsigned long start, unsigned long end)
222 sync_global_pgds(start, end);
226 * NOTE: This function is marked __ref because it calls __init function
227 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
229 static __ref void *spp_getpage(void)
234 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
236 ptr = memblock_alloc(PAGE_SIZE, PAGE_SIZE);
238 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
239 panic("set_pte_phys: cannot allocate page data %s\n",
240 after_bootmem ? "after bootmem" : "");
243 pr_debug("spp_getpage %p\n", ptr);
248 static p4d_t *fill_p4d(pgd_t *pgd, unsigned long vaddr)
250 if (pgd_none(*pgd)) {
251 p4d_t *p4d = (p4d_t *)spp_getpage();
252 pgd_populate(&init_mm, pgd, p4d);
253 if (p4d != p4d_offset(pgd, 0))
254 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
255 p4d, p4d_offset(pgd, 0));
257 return p4d_offset(pgd, vaddr);
260 static pud_t *fill_pud(p4d_t *p4d, unsigned long vaddr)
262 if (p4d_none(*p4d)) {
263 pud_t *pud = (pud_t *)spp_getpage();
264 p4d_populate(&init_mm, p4d, pud);
265 if (pud != pud_offset(p4d, 0))
266 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
267 pud, pud_offset(p4d, 0));
269 return pud_offset(p4d, vaddr);
272 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
274 if (pud_none(*pud)) {
275 pmd_t *pmd = (pmd_t *) spp_getpage();
276 pud_populate(&init_mm, pud, pmd);
277 if (pmd != pmd_offset(pud, 0))
278 printk(KERN_ERR "PAGETABLE BUG #02! %p <-> %p\n",
279 pmd, pmd_offset(pud, 0));
281 return pmd_offset(pud, vaddr);
284 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
286 if (pmd_none(*pmd)) {
287 pte_t *pte = (pte_t *) spp_getpage();
288 pmd_populate_kernel(&init_mm, pmd, pte);
289 if (pte != pte_offset_kernel(pmd, 0))
290 printk(KERN_ERR "PAGETABLE BUG #03!\n");
292 return pte_offset_kernel(pmd, vaddr);
295 static void __set_pte_vaddr(pud_t *pud, unsigned long vaddr, pte_t new_pte)
297 pmd_t *pmd = fill_pmd(pud, vaddr);
298 pte_t *pte = fill_pte(pmd, vaddr);
300 set_pte(pte, new_pte);
303 * It's enough to flush this one mapping.
304 * (PGE mappings get flushed as well)
306 flush_tlb_one_kernel(vaddr);
309 void set_pte_vaddr_p4d(p4d_t *p4d_page, unsigned long vaddr, pte_t new_pte)
311 p4d_t *p4d = p4d_page + p4d_index(vaddr);
312 pud_t *pud = fill_pud(p4d, vaddr);
314 __set_pte_vaddr(pud, vaddr, new_pte);
317 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
319 pud_t *pud = pud_page + pud_index(vaddr);
321 __set_pte_vaddr(pud, vaddr, new_pte);
324 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
329 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
331 pgd = pgd_offset_k(vaddr);
332 if (pgd_none(*pgd)) {
334 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
338 p4d_page = p4d_offset(pgd, 0);
339 set_pte_vaddr_p4d(p4d_page, vaddr, pteval);
342 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
348 pgd = pgd_offset_k(vaddr);
349 p4d = fill_p4d(pgd, vaddr);
350 pud = fill_pud(p4d, vaddr);
351 return fill_pmd(pud, vaddr);
354 pte_t * __init populate_extra_pte(unsigned long vaddr)
358 pmd = populate_extra_pmd(vaddr);
359 return fill_pte(pmd, vaddr);
363 * Create large page table mappings for a range of physical addresses.
365 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
366 enum page_cache_mode cache)
374 pgprot_val(prot) = pgprot_val(PAGE_KERNEL_LARGE) |
375 protval_4k_2_large(cachemode2protval(cache));
376 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
377 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
378 pgd = pgd_offset_k((unsigned long)__va(phys));
379 if (pgd_none(*pgd)) {
380 p4d = (p4d_t *) spp_getpage();
381 set_pgd(pgd, __pgd(__pa(p4d) | _KERNPG_TABLE |
384 p4d = p4d_offset(pgd, (unsigned long)__va(phys));
385 if (p4d_none(*p4d)) {
386 pud = (pud_t *) spp_getpage();
387 set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE |
390 pud = pud_offset(p4d, (unsigned long)__va(phys));
391 if (pud_none(*pud)) {
392 pmd = (pmd_t *) spp_getpage();
393 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
396 pmd = pmd_offset(pud, phys);
397 BUG_ON(!pmd_none(*pmd));
398 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
402 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
404 __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_WB);
407 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
409 __init_extra_mapping(phys, size, _PAGE_CACHE_MODE_UC);
413 * The head.S code sets up the kernel high mapping:
415 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
417 * phys_base holds the negative offset to the kernel, which is added
418 * to the compile time generated pmds. This results in invalid pmds up
419 * to the point where we hit the physaddr 0 mapping.
421 * We limit the mappings to the region from _text to _brk_end. _brk_end
422 * is rounded up to the 2MB boundary. This catches the invalid pmds as
423 * well, as they are located before _text:
425 void __init cleanup_highmap(void)
427 unsigned long vaddr = __START_KERNEL_map;
428 unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
429 unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
430 pmd_t *pmd = level2_kernel_pgt;
433 * Native path, max_pfn_mapped is not set yet.
434 * Xen has valid max_pfn_mapped set in
435 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
438 vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
440 for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
443 if (vaddr < (unsigned long) _text || vaddr > end)
444 set_pmd(pmd, __pmd(0));
449 * Create PTE level page table mapping for physical addresses.
450 * It returns the last physical address mapped.
452 static unsigned long __meminit
453 phys_pte_init(pte_t *pte_page, unsigned long paddr, unsigned long paddr_end,
454 pgprot_t prot, bool init)
456 unsigned long pages = 0, paddr_next;
457 unsigned long paddr_last = paddr_end;
461 pte = pte_page + pte_index(paddr);
462 i = pte_index(paddr);
464 for (; i < PTRS_PER_PTE; i++, paddr = paddr_next, pte++) {
465 paddr_next = (paddr & PAGE_MASK) + PAGE_SIZE;
466 if (paddr >= paddr_end) {
467 if (!after_bootmem &&
468 !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
470 !e820__mapped_any(paddr & PAGE_MASK, paddr_next,
471 E820_TYPE_RESERVED_KERN))
472 set_pte_init(pte, __pte(0), init);
477 * We will re-use the existing mapping.
478 * Xen for example has some special requirements, like mapping
479 * pagetable pages as RO. So assume someone who pre-setup
480 * these mappings are more intelligent.
482 if (!pte_none(*pte)) {
489 pr_info(" pte=%p addr=%lx pte=%016lx\n", pte, paddr,
490 pfn_pte(paddr >> PAGE_SHIFT, PAGE_KERNEL).pte);
492 set_pte_init(pte, pfn_pte(paddr >> PAGE_SHIFT, prot), init);
493 paddr_last = (paddr & PAGE_MASK) + PAGE_SIZE;
496 update_page_count(PG_LEVEL_4K, pages);
502 * Create PMD level page table mapping for physical addresses. The virtual
503 * and physical address have to be aligned at this level.
504 * It returns the last physical address mapped.
506 static unsigned long __meminit
507 phys_pmd_init(pmd_t *pmd_page, unsigned long paddr, unsigned long paddr_end,
508 unsigned long page_size_mask, pgprot_t prot, bool init)
510 unsigned long pages = 0, paddr_next;
511 unsigned long paddr_last = paddr_end;
513 int i = pmd_index(paddr);
515 for (; i < PTRS_PER_PMD; i++, paddr = paddr_next) {
516 pmd_t *pmd = pmd_page + pmd_index(paddr);
518 pgprot_t new_prot = prot;
520 paddr_next = (paddr & PMD_MASK) + PMD_SIZE;
521 if (paddr >= paddr_end) {
522 if (!after_bootmem &&
523 !e820__mapped_any(paddr & PMD_MASK, paddr_next,
525 !e820__mapped_any(paddr & PMD_MASK, paddr_next,
526 E820_TYPE_RESERVED_KERN))
527 set_pmd_init(pmd, __pmd(0), init);
531 if (!pmd_none(*pmd)) {
532 if (!pmd_large(*pmd)) {
533 spin_lock(&init_mm.page_table_lock);
534 pte = (pte_t *)pmd_page_vaddr(*pmd);
535 paddr_last = phys_pte_init(pte, paddr,
538 spin_unlock(&init_mm.page_table_lock);
542 * If we are ok with PG_LEVEL_2M mapping, then we will
543 * use the existing mapping,
545 * Otherwise, we will split the large page mapping but
546 * use the same existing protection bits except for
547 * large page, so that we don't violate Intel's TLB
548 * Application note (317080) which says, while changing
549 * the page sizes, new and old translations should
550 * not differ with respect to page frame and
553 if (page_size_mask & (1 << PG_LEVEL_2M)) {
556 paddr_last = paddr_next;
559 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
562 if (page_size_mask & (1<<PG_LEVEL_2M)) {
564 spin_lock(&init_mm.page_table_lock);
565 set_pte_init((pte_t *)pmd,
566 pfn_pte((paddr & PMD_MASK) >> PAGE_SHIFT,
567 __pgprot(pgprot_val(prot) | _PAGE_PSE)),
569 spin_unlock(&init_mm.page_table_lock);
570 paddr_last = paddr_next;
574 pte = alloc_low_page();
575 paddr_last = phys_pte_init(pte, paddr, paddr_end, new_prot, init);
577 spin_lock(&init_mm.page_table_lock);
578 pmd_populate_kernel_init(&init_mm, pmd, pte, init);
579 spin_unlock(&init_mm.page_table_lock);
581 update_page_count(PG_LEVEL_2M, pages);
586 * Create PUD level page table mapping for physical addresses. The virtual
587 * and physical address do not have to be aligned at this level. KASLR can
588 * randomize virtual addresses up to this level.
589 * It returns the last physical address mapped.
591 static unsigned long __meminit
592 phys_pud_init(pud_t *pud_page, unsigned long paddr, unsigned long paddr_end,
593 unsigned long page_size_mask, pgprot_t _prot, bool init)
595 unsigned long pages = 0, paddr_next;
596 unsigned long paddr_last = paddr_end;
597 unsigned long vaddr = (unsigned long)__va(paddr);
598 int i = pud_index(vaddr);
600 for (; i < PTRS_PER_PUD; i++, paddr = paddr_next) {
603 pgprot_t prot = _prot;
605 vaddr = (unsigned long)__va(paddr);
606 pud = pud_page + pud_index(vaddr);
607 paddr_next = (paddr & PUD_MASK) + PUD_SIZE;
609 if (paddr >= paddr_end) {
610 if (!after_bootmem &&
611 !e820__mapped_any(paddr & PUD_MASK, paddr_next,
613 !e820__mapped_any(paddr & PUD_MASK, paddr_next,
614 E820_TYPE_RESERVED_KERN))
615 set_pud_init(pud, __pud(0), init);
619 if (!pud_none(*pud)) {
620 if (!pud_large(*pud)) {
621 pmd = pmd_offset(pud, 0);
622 paddr_last = phys_pmd_init(pmd, paddr,
629 * If we are ok with PG_LEVEL_1G mapping, then we will
630 * use the existing mapping.
632 * Otherwise, we will split the gbpage mapping but use
633 * the same existing protection bits except for large
634 * page, so that we don't violate Intel's TLB
635 * Application note (317080) which says, while changing
636 * the page sizes, new and old translations should
637 * not differ with respect to page frame and
640 if (page_size_mask & (1 << PG_LEVEL_1G)) {
643 paddr_last = paddr_next;
646 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
649 if (page_size_mask & (1<<PG_LEVEL_1G)) {
651 spin_lock(&init_mm.page_table_lock);
653 prot = __pgprot(pgprot_val(prot) | __PAGE_KERNEL_LARGE);
655 set_pte_init((pte_t *)pud,
656 pfn_pte((paddr & PUD_MASK) >> PAGE_SHIFT,
659 spin_unlock(&init_mm.page_table_lock);
660 paddr_last = paddr_next;
664 pmd = alloc_low_page();
665 paddr_last = phys_pmd_init(pmd, paddr, paddr_end,
666 page_size_mask, prot, init);
668 spin_lock(&init_mm.page_table_lock);
669 pud_populate_init(&init_mm, pud, pmd, init);
670 spin_unlock(&init_mm.page_table_lock);
673 update_page_count(PG_LEVEL_1G, pages);
678 static unsigned long __meminit
679 phys_p4d_init(p4d_t *p4d_page, unsigned long paddr, unsigned long paddr_end,
680 unsigned long page_size_mask, pgprot_t prot, bool init)
682 unsigned long vaddr, vaddr_end, vaddr_next, paddr_next, paddr_last;
684 paddr_last = paddr_end;
685 vaddr = (unsigned long)__va(paddr);
686 vaddr_end = (unsigned long)__va(paddr_end);
688 if (!pgtable_l5_enabled())
689 return phys_pud_init((pud_t *) p4d_page, paddr, paddr_end,
690 page_size_mask, prot, init);
692 for (; vaddr < vaddr_end; vaddr = vaddr_next) {
693 p4d_t *p4d = p4d_page + p4d_index(vaddr);
696 vaddr_next = (vaddr & P4D_MASK) + P4D_SIZE;
699 if (paddr >= paddr_end) {
700 paddr_next = __pa(vaddr_next);
701 if (!after_bootmem &&
702 !e820__mapped_any(paddr & P4D_MASK, paddr_next,
704 !e820__mapped_any(paddr & P4D_MASK, paddr_next,
705 E820_TYPE_RESERVED_KERN))
706 set_p4d_init(p4d, __p4d(0), init);
710 if (!p4d_none(*p4d)) {
711 pud = pud_offset(p4d, 0);
712 paddr_last = phys_pud_init(pud, paddr, __pa(vaddr_end),
713 page_size_mask, prot, init);
717 pud = alloc_low_page();
718 paddr_last = phys_pud_init(pud, paddr, __pa(vaddr_end),
719 page_size_mask, prot, init);
721 spin_lock(&init_mm.page_table_lock);
722 p4d_populate_init(&init_mm, p4d, pud, init);
723 spin_unlock(&init_mm.page_table_lock);
729 static unsigned long __meminit
730 __kernel_physical_mapping_init(unsigned long paddr_start,
731 unsigned long paddr_end,
732 unsigned long page_size_mask,
733 pgprot_t prot, bool init)
735 bool pgd_changed = false;
736 unsigned long vaddr, vaddr_start, vaddr_end, vaddr_next, paddr_last;
738 paddr_last = paddr_end;
739 vaddr = (unsigned long)__va(paddr_start);
740 vaddr_end = (unsigned long)__va(paddr_end);
743 for (; vaddr < vaddr_end; vaddr = vaddr_next) {
744 pgd_t *pgd = pgd_offset_k(vaddr);
747 vaddr_next = (vaddr & PGDIR_MASK) + PGDIR_SIZE;
750 p4d = (p4d_t *)pgd_page_vaddr(*pgd);
751 paddr_last = phys_p4d_init(p4d, __pa(vaddr),
758 p4d = alloc_low_page();
759 paddr_last = phys_p4d_init(p4d, __pa(vaddr), __pa(vaddr_end),
760 page_size_mask, prot, init);
762 spin_lock(&init_mm.page_table_lock);
763 if (pgtable_l5_enabled())
764 pgd_populate_init(&init_mm, pgd, p4d, init);
766 p4d_populate_init(&init_mm, p4d_offset(pgd, vaddr),
767 (pud_t *) p4d, init);
769 spin_unlock(&init_mm.page_table_lock);
774 sync_global_pgds(vaddr_start, vaddr_end - 1);
781 * Create page table mapping for the physical memory for specific physical
782 * addresses. Note that it can only be used to populate non-present entries.
783 * The virtual and physical addresses have to be aligned on PMD level
784 * down. It returns the last physical address mapped.
786 unsigned long __meminit
787 kernel_physical_mapping_init(unsigned long paddr_start,
788 unsigned long paddr_end,
789 unsigned long page_size_mask, pgprot_t prot)
791 return __kernel_physical_mapping_init(paddr_start, paddr_end,
792 page_size_mask, prot, true);
796 * This function is similar to kernel_physical_mapping_init() above with the
797 * exception that it uses set_{pud,pmd}() instead of the set_{pud,pte}_safe()
798 * when updating the mapping. The caller is responsible to flush the TLBs after
799 * the function returns.
801 unsigned long __meminit
802 kernel_physical_mapping_change(unsigned long paddr_start,
803 unsigned long paddr_end,
804 unsigned long page_size_mask)
806 return __kernel_physical_mapping_init(paddr_start, paddr_end,
807 page_size_mask, PAGE_KERNEL,
812 void __init initmem_init(void)
814 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
818 void __init paging_init(void)
823 * clear the default setting with node 0
824 * note: don't use nodes_clear here, that is really clearing when
825 * numa support is not compiled in, and later node_set_state
826 * will not set it back.
828 node_clear_state(0, N_MEMORY);
829 node_clear_state(0, N_NORMAL_MEMORY);
835 * Memory hotplug specific functions
837 #ifdef CONFIG_MEMORY_HOTPLUG
839 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
842 static void update_end_of_memory_vars(u64 start, u64 size)
844 unsigned long end_pfn = PFN_UP(start + size);
846 if (end_pfn > max_pfn) {
848 max_low_pfn = end_pfn;
849 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
853 int add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages,
854 struct mhp_params *params)
858 ret = __add_pages(nid, start_pfn, nr_pages, params);
861 /* update max_pfn, max_low_pfn and high_memory */
862 update_end_of_memory_vars(start_pfn << PAGE_SHIFT,
863 nr_pages << PAGE_SHIFT);
868 int arch_add_memory(int nid, u64 start, u64 size,
869 struct mhp_params *params)
871 unsigned long start_pfn = start >> PAGE_SHIFT;
872 unsigned long nr_pages = size >> PAGE_SHIFT;
874 init_memory_mapping(start, start + size, params->pgprot);
876 return add_pages(nid, start_pfn, nr_pages, params);
879 #define PAGE_INUSE 0xFD
881 static void __meminit free_pagetable(struct page *page, int order)
884 unsigned int nr_pages = 1 << order;
886 /* bootmem page has reserved flag */
887 if (PageReserved(page)) {
888 __ClearPageReserved(page);
890 magic = (unsigned long)page->freelist;
891 if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
893 put_page_bootmem(page++);
896 free_reserved_page(page++);
898 free_pages((unsigned long)page_address(page), order);
901 static void __meminit free_hugepage_table(struct page *page,
902 struct vmem_altmap *altmap)
905 vmem_altmap_free(altmap, PMD_SIZE / PAGE_SIZE);
907 free_pagetable(page, get_order(PMD_SIZE));
910 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
915 for (i = 0; i < PTRS_PER_PTE; i++) {
921 /* free a pte talbe */
922 free_pagetable(pmd_page(*pmd), 0);
923 spin_lock(&init_mm.page_table_lock);
925 spin_unlock(&init_mm.page_table_lock);
928 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
933 for (i = 0; i < PTRS_PER_PMD; i++) {
939 /* free a pmd talbe */
940 free_pagetable(pud_page(*pud), 0);
941 spin_lock(&init_mm.page_table_lock);
943 spin_unlock(&init_mm.page_table_lock);
946 static void __meminit free_pud_table(pud_t *pud_start, p4d_t *p4d)
951 for (i = 0; i < PTRS_PER_PUD; i++) {
957 /* free a pud talbe */
958 free_pagetable(p4d_page(*p4d), 0);
959 spin_lock(&init_mm.page_table_lock);
961 spin_unlock(&init_mm.page_table_lock);
964 static void __meminit
965 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
968 unsigned long next, pages = 0;
971 phys_addr_t phys_addr;
973 pte = pte_start + pte_index(addr);
974 for (; addr < end; addr = next, pte++) {
975 next = (addr + PAGE_SIZE) & PAGE_MASK;
979 if (!pte_present(*pte))
983 * We mapped [0,1G) memory as identity mapping when
984 * initializing, in arch/x86/kernel/head_64.S. These
985 * pagetables cannot be removed.
987 phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
988 if (phys_addr < (phys_addr_t)0x40000000)
991 if (PAGE_ALIGNED(addr) && PAGE_ALIGNED(next)) {
993 * Do not free direct mapping pages since they were
994 * freed when offlining, or simplely not in use.
997 free_pagetable(pte_page(*pte), 0);
999 spin_lock(&init_mm.page_table_lock);
1000 pte_clear(&init_mm, addr, pte);
1001 spin_unlock(&init_mm.page_table_lock);
1003 /* For non-direct mapping, pages means nothing. */
1007 * If we are here, we are freeing vmemmap pages since
1008 * direct mapped memory ranges to be freed are aligned.
1010 * If we are not removing the whole page, it means
1011 * other page structs in this page are being used and
1012 * we canot remove them. So fill the unused page_structs
1013 * with 0xFD, and remove the page when it is wholly
1016 memset((void *)addr, PAGE_INUSE, next - addr);
1018 page_addr = page_address(pte_page(*pte));
1019 if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
1020 free_pagetable(pte_page(*pte), 0);
1022 spin_lock(&init_mm.page_table_lock);
1023 pte_clear(&init_mm, addr, pte);
1024 spin_unlock(&init_mm.page_table_lock);
1029 /* Call free_pte_table() in remove_pmd_table(). */
1032 update_page_count(PG_LEVEL_4K, -pages);
1035 static void __meminit
1036 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
1037 bool direct, struct vmem_altmap *altmap)
1039 unsigned long next, pages = 0;
1044 pmd = pmd_start + pmd_index(addr);
1045 for (; addr < end; addr = next, pmd++) {
1046 next = pmd_addr_end(addr, end);
1048 if (!pmd_present(*pmd))
1051 if (pmd_large(*pmd)) {
1052 if (IS_ALIGNED(addr, PMD_SIZE) &&
1053 IS_ALIGNED(next, PMD_SIZE)) {
1055 free_hugepage_table(pmd_page(*pmd),
1058 spin_lock(&init_mm.page_table_lock);
1060 spin_unlock(&init_mm.page_table_lock);
1063 /* If here, we are freeing vmemmap pages. */
1064 memset((void *)addr, PAGE_INUSE, next - addr);
1066 page_addr = page_address(pmd_page(*pmd));
1067 if (!memchr_inv(page_addr, PAGE_INUSE,
1069 free_hugepage_table(pmd_page(*pmd),
1072 spin_lock(&init_mm.page_table_lock);
1074 spin_unlock(&init_mm.page_table_lock);
1081 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
1082 remove_pte_table(pte_base, addr, next, direct);
1083 free_pte_table(pte_base, pmd);
1086 /* Call free_pmd_table() in remove_pud_table(). */
1088 update_page_count(PG_LEVEL_2M, -pages);
1091 static void __meminit
1092 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
1093 struct vmem_altmap *altmap, bool direct)
1095 unsigned long next, pages = 0;
1100 pud = pud_start + pud_index(addr);
1101 for (; addr < end; addr = next, pud++) {
1102 next = pud_addr_end(addr, end);
1104 if (!pud_present(*pud))
1107 if (pud_large(*pud)) {
1108 if (IS_ALIGNED(addr, PUD_SIZE) &&
1109 IS_ALIGNED(next, PUD_SIZE)) {
1111 free_pagetable(pud_page(*pud),
1112 get_order(PUD_SIZE));
1114 spin_lock(&init_mm.page_table_lock);
1116 spin_unlock(&init_mm.page_table_lock);
1119 /* If here, we are freeing vmemmap pages. */
1120 memset((void *)addr, PAGE_INUSE, next - addr);
1122 page_addr = page_address(pud_page(*pud));
1123 if (!memchr_inv(page_addr, PAGE_INUSE,
1125 free_pagetable(pud_page(*pud),
1126 get_order(PUD_SIZE));
1128 spin_lock(&init_mm.page_table_lock);
1130 spin_unlock(&init_mm.page_table_lock);
1137 pmd_base = pmd_offset(pud, 0);
1138 remove_pmd_table(pmd_base, addr, next, direct, altmap);
1139 free_pmd_table(pmd_base, pud);
1143 update_page_count(PG_LEVEL_1G, -pages);
1146 static void __meminit
1147 remove_p4d_table(p4d_t *p4d_start, unsigned long addr, unsigned long end,
1148 struct vmem_altmap *altmap, bool direct)
1150 unsigned long next, pages = 0;
1154 p4d = p4d_start + p4d_index(addr);
1155 for (; addr < end; addr = next, p4d++) {
1156 next = p4d_addr_end(addr, end);
1158 if (!p4d_present(*p4d))
1161 BUILD_BUG_ON(p4d_large(*p4d));
1163 pud_base = pud_offset(p4d, 0);
1164 remove_pud_table(pud_base, addr, next, altmap, direct);
1166 * For 4-level page tables we do not want to free PUDs, but in the
1167 * 5-level case we should free them. This code will have to change
1168 * to adapt for boot-time switching between 4 and 5 level page tables.
1170 if (pgtable_l5_enabled())
1171 free_pud_table(pud_base, p4d);
1175 update_page_count(PG_LEVEL_512G, -pages);
1178 /* start and end are both virtual address. */
1179 static void __meminit
1180 remove_pagetable(unsigned long start, unsigned long end, bool direct,
1181 struct vmem_altmap *altmap)
1188 for (addr = start; addr < end; addr = next) {
1189 next = pgd_addr_end(addr, end);
1191 pgd = pgd_offset_k(addr);
1192 if (!pgd_present(*pgd))
1195 p4d = p4d_offset(pgd, 0);
1196 remove_p4d_table(p4d, addr, next, altmap, direct);
1202 void __ref vmemmap_free(unsigned long start, unsigned long end,
1203 struct vmem_altmap *altmap)
1205 remove_pagetable(start, end, false, altmap);
1208 static void __meminit
1209 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
1211 start = (unsigned long)__va(start);
1212 end = (unsigned long)__va(end);
1214 remove_pagetable(start, end, true, NULL);
1217 void __ref arch_remove_memory(int nid, u64 start, u64 size,
1218 struct vmem_altmap *altmap)
1220 unsigned long start_pfn = start >> PAGE_SHIFT;
1221 unsigned long nr_pages = size >> PAGE_SHIFT;
1223 __remove_pages(start_pfn, nr_pages, altmap);
1224 kernel_physical_mapping_remove(start, start + size);
1226 #endif /* CONFIG_MEMORY_HOTPLUG */
1228 static struct kcore_list kcore_vsyscall;
1230 static void __init register_page_bootmem_info(void)
1235 for_each_online_node(i)
1236 register_page_bootmem_info_node(NODE_DATA(i));
1241 * Pre-allocates page-table pages for the vmalloc area in the kernel page-table.
1242 * Only the level which needs to be synchronized between all page-tables is
1243 * allocated because the synchronization can be expensive.
1245 static void __init preallocate_vmalloc_pages(void)
1250 for (addr = VMALLOC_START; addr <= VMALLOC_END; addr = ALIGN(addr + 1, PGDIR_SIZE)) {
1251 pgd_t *pgd = pgd_offset_k(addr);
1256 p4d = p4d_alloc(&init_mm, pgd, addr);
1261 * With 5-level paging the P4D level is not folded. So the PGDs
1262 * are now populated and there is no need to walk down to the
1265 if (pgtable_l5_enabled())
1269 pud = pud_alloc(&init_mm, p4d, addr);
1279 * The pages have to be there now or they will be missing in
1280 * process page-tables later.
1282 panic("Failed to pre-allocate %s pages for vmalloc area\n", lvl);
1285 void __init mem_init(void)
1289 /* clear_bss() already clear the empty_zero_page */
1291 /* this will put all memory onto the freelists */
1292 memblock_free_all();
1294 x86_init.hyper.init_after_bootmem();
1297 * Must be done after boot memory is put on freelist, because here we
1298 * might set fields in deferred struct pages that have not yet been
1299 * initialized, and memblock_free_all() initializes all the reserved
1300 * deferred pages for us.
1302 register_page_bootmem_info();
1304 /* Register memory areas for /proc/kcore */
1305 if (get_gate_vma(&init_mm))
1306 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_ADDR, PAGE_SIZE, KCORE_USER);
1308 preallocate_vmalloc_pages();
1310 mem_init_print_info(NULL);
1313 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1314 int __init deferred_page_init_max_threads(const struct cpumask *node_cpumask)
1317 * More CPUs always led to greater speedups on tested systems, up to
1318 * all the nodes' CPUs. Use all since the system is otherwise idle
1321 return max_t(int, cpumask_weight(node_cpumask), 1);
1325 int kernel_set_to_readonly;
1327 void mark_rodata_ro(void)
1329 unsigned long start = PFN_ALIGN(_text);
1330 unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1331 unsigned long end = (unsigned long)__end_rodata_hpage_align;
1332 unsigned long text_end = PFN_ALIGN(_etext);
1333 unsigned long rodata_end = PFN_ALIGN(__end_rodata);
1334 unsigned long all_end;
1336 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1337 (end - start) >> 10);
1338 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1340 kernel_set_to_readonly = 1;
1343 * The rodata/data/bss/brk section (but not the kernel text!)
1344 * should also be not-executable.
1346 * We align all_end to PMD_SIZE because the existing mapping
1347 * is a full PMD. If we would align _brk_end to PAGE_SIZE we
1348 * split the PMD and the reminder between _brk_end and the end
1349 * of the PMD will remain mapped executable.
1351 * Any PMD which was setup after the one which covers _brk_end
1352 * has been zapped already via cleanup_highmem().
1354 all_end = roundup((unsigned long)_brk_end, PMD_SIZE);
1355 set_memory_nx(text_end, (all_end - text_end) >> PAGE_SHIFT);
1357 set_ftrace_ops_ro();
1359 #ifdef CONFIG_CPA_DEBUG
1360 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1361 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1363 printk(KERN_INFO "Testing CPA: again\n");
1364 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1367 free_kernel_image_pages("unused kernel image (text/rodata gap)",
1368 (void *)text_end, (void *)rodata_start);
1369 free_kernel_image_pages("unused kernel image (rodata/data gap)",
1370 (void *)rodata_end, (void *)_sdata);
1375 int kern_addr_valid(unsigned long addr)
1377 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1384 if (above != 0 && above != -1UL)
1387 pgd = pgd_offset_k(addr);
1391 p4d = p4d_offset(pgd, addr);
1395 pud = pud_offset(p4d, addr);
1399 if (pud_large(*pud))
1400 return pfn_valid(pud_pfn(*pud));
1402 pmd = pmd_offset(pud, addr);
1406 if (pmd_large(*pmd))
1407 return pfn_valid(pmd_pfn(*pmd));
1409 pte = pte_offset_kernel(pmd, addr);
1413 return pfn_valid(pte_pfn(*pte));
1417 * Block size is the minimum amount of memory which can be hotplugged or
1418 * hotremoved. It must be power of two and must be equal or larger than
1419 * MIN_MEMORY_BLOCK_SIZE.
1421 #define MAX_BLOCK_SIZE (2UL << 30)
1423 /* Amount of ram needed to start using large blocks */
1424 #define MEM_SIZE_FOR_LARGE_BLOCK (64UL << 30)
1426 /* Adjustable memory block size */
1427 static unsigned long set_memory_block_size;
1428 int __init set_memory_block_size_order(unsigned int order)
1430 unsigned long size = 1UL << order;
1432 if (size > MEM_SIZE_FOR_LARGE_BLOCK || size < MIN_MEMORY_BLOCK_SIZE)
1435 set_memory_block_size = size;
1439 static unsigned long probe_memory_block_size(void)
1441 unsigned long boot_mem_end = max_pfn << PAGE_SHIFT;
1444 /* If memory block size has been set, then use it */
1445 bz = set_memory_block_size;
1449 /* Use regular block if RAM is smaller than MEM_SIZE_FOR_LARGE_BLOCK */
1450 if (boot_mem_end < MEM_SIZE_FOR_LARGE_BLOCK) {
1451 bz = MIN_MEMORY_BLOCK_SIZE;
1455 /* Find the largest allowed block size that aligns to memory end */
1456 for (bz = MAX_BLOCK_SIZE; bz > MIN_MEMORY_BLOCK_SIZE; bz >>= 1) {
1457 if (IS_ALIGNED(boot_mem_end, bz))
1461 pr_info("x86/mm: Memory block size: %ldMB\n", bz >> 20);
1466 static unsigned long memory_block_size_probed;
1467 unsigned long memory_block_size_bytes(void)
1469 if (!memory_block_size_probed)
1470 memory_block_size_probed = probe_memory_block_size();
1472 return memory_block_size_probed;
1475 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1477 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1479 static long __meminitdata addr_start, addr_end;
1480 static void __meminitdata *p_start, *p_end;
1481 static int __meminitdata node_start;
1483 static int __meminit vmemmap_populate_hugepages(unsigned long start,
1484 unsigned long end, int node, struct vmem_altmap *altmap)
1493 for (addr = start; addr < end; addr = next) {
1494 next = pmd_addr_end(addr, end);
1496 pgd = vmemmap_pgd_populate(addr, node);
1500 p4d = vmemmap_p4d_populate(pgd, addr, node);
1504 pud = vmemmap_pud_populate(p4d, addr, node);
1508 pmd = pmd_offset(pud, addr);
1509 if (pmd_none(*pmd)) {
1512 p = vmemmap_alloc_block_buf(PMD_SIZE, node, altmap);
1516 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1518 set_pmd(pmd, __pmd(pte_val(entry)));
1520 /* check to see if we have contiguous blocks */
1521 if (p_end != p || node_start != node) {
1523 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1524 addr_start, addr_end-1, p_start, p_end-1, node_start);
1530 addr_end = addr + PMD_SIZE;
1531 p_end = p + PMD_SIZE;
1534 return -ENOMEM; /* no fallback */
1535 } else if (pmd_large(*pmd)) {
1536 vmemmap_verify((pte_t *)pmd, node, addr, next);
1539 if (vmemmap_populate_basepages(addr, next, node, NULL))
1545 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
1546 struct vmem_altmap *altmap)
1550 if (end - start < PAGES_PER_SECTION * sizeof(struct page))
1551 err = vmemmap_populate_basepages(start, end, node, NULL);
1552 else if (boot_cpu_has(X86_FEATURE_PSE))
1553 err = vmemmap_populate_hugepages(start, end, node, altmap);
1555 pr_err_once("%s: no cpu support for altmap allocations\n",
1559 err = vmemmap_populate_basepages(start, end, node, NULL);
1561 sync_global_pgds(start, end - 1);
1565 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1566 void register_page_bootmem_memmap(unsigned long section_nr,
1567 struct page *start_page, unsigned long nr_pages)
1569 unsigned long addr = (unsigned long)start_page;
1570 unsigned long end = (unsigned long)(start_page + nr_pages);
1576 unsigned int nr_pmd_pages;
1579 for (; addr < end; addr = next) {
1582 pgd = pgd_offset_k(addr);
1583 if (pgd_none(*pgd)) {
1584 next = (addr + PAGE_SIZE) & PAGE_MASK;
1587 get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1589 p4d = p4d_offset(pgd, addr);
1590 if (p4d_none(*p4d)) {
1591 next = (addr + PAGE_SIZE) & PAGE_MASK;
1594 get_page_bootmem(section_nr, p4d_page(*p4d), MIX_SECTION_INFO);
1596 pud = pud_offset(p4d, addr);
1597 if (pud_none(*pud)) {
1598 next = (addr + PAGE_SIZE) & PAGE_MASK;
1601 get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1603 if (!boot_cpu_has(X86_FEATURE_PSE)) {
1604 next = (addr + PAGE_SIZE) & PAGE_MASK;
1605 pmd = pmd_offset(pud, addr);
1608 get_page_bootmem(section_nr, pmd_page(*pmd),
1611 pte = pte_offset_kernel(pmd, addr);
1614 get_page_bootmem(section_nr, pte_page(*pte),
1617 next = pmd_addr_end(addr, end);
1619 pmd = pmd_offset(pud, addr);
1623 nr_pmd_pages = 1 << get_order(PMD_SIZE);
1624 page = pmd_page(*pmd);
1625 while (nr_pmd_pages--)
1626 get_page_bootmem(section_nr, page++,
1633 void __meminit vmemmap_populate_print_last(void)
1636 pr_debug(" [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1637 addr_start, addr_end-1, p_start, p_end-1, node_start);