2 * linux/arch/x86_64/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 2000 Pavel Machek <pavel@ucw.cz>
6 * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
18 #include <linux/swap.h>
19 #include <linux/smp.h>
20 #include <linux/init.h>
21 #include <linux/initrd.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.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/module.h>
31 #include <linux/memory.h>
32 #include <linux/memory_hotplug.h>
33 #include <linux/nmi.h>
34 #include <linux/gfp.h>
36 #include <asm/processor.h>
37 #include <asm/bios_ebda.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
40 #include <asm/pgalloc.h>
42 #include <asm/fixmap.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/cacheflush.h>
54 #include <asm/uv/uv.h>
55 #include <asm/setup.h>
57 #include "mm_internal.h"
59 static void ident_pmd_init(unsigned long pmd_flag, pmd_t *pmd_page,
60 unsigned long addr, unsigned long end)
63 for (; addr < end; addr += PMD_SIZE) {
64 pmd_t *pmd = pmd_page + pmd_index(addr);
66 if (!pmd_present(*pmd))
67 set_pmd(pmd, __pmd(addr | pmd_flag));
70 static int ident_pud_init(struct x86_mapping_info *info, pud_t *pud_page,
71 unsigned long addr, unsigned long end)
75 for (; addr < end; addr = next) {
76 pud_t *pud = pud_page + pud_index(addr);
79 next = (addr & PUD_MASK) + PUD_SIZE;
83 if (pud_present(*pud)) {
84 pmd = pmd_offset(pud, 0);
85 ident_pmd_init(info->pmd_flag, pmd, addr, next);
88 pmd = (pmd_t *)info->alloc_pgt_page(info->context);
91 ident_pmd_init(info->pmd_flag, pmd, addr, next);
92 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
98 int kernel_ident_mapping_init(struct x86_mapping_info *info, pgd_t *pgd_page,
99 unsigned long addr, unsigned long end)
103 int off = info->kernel_mapping ? pgd_index(__PAGE_OFFSET) : 0;
105 for (; addr < end; addr = next) {
106 pgd_t *pgd = pgd_page + pgd_index(addr) + off;
109 next = (addr & PGDIR_MASK) + PGDIR_SIZE;
113 if (pgd_present(*pgd)) {
114 pud = pud_offset(pgd, 0);
115 result = ident_pud_init(info, pud, addr, next);
121 pud = (pud_t *)info->alloc_pgt_page(info->context);
124 result = ident_pud_init(info, pud, addr, next);
127 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE));
133 static int __init parse_direct_gbpages_off(char *arg)
138 early_param("nogbpages", parse_direct_gbpages_off);
140 static int __init parse_direct_gbpages_on(char *arg)
145 early_param("gbpages", parse_direct_gbpages_on);
148 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
149 * physical space so we can cache the place of the first one and move
150 * around without checking the pgd every time.
153 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_IOMAP;
154 EXPORT_SYMBOL_GPL(__supported_pte_mask);
156 int force_personality32;
160 * Control non executable heap for 32bit processes.
161 * To control the stack too use noexec=off
163 * on PROT_READ does not imply PROT_EXEC for 32-bit processes (default)
164 * off PROT_READ implies PROT_EXEC
166 static int __init nonx32_setup(char *str)
168 if (!strcmp(str, "on"))
169 force_personality32 &= ~READ_IMPLIES_EXEC;
170 else if (!strcmp(str, "off"))
171 force_personality32 |= READ_IMPLIES_EXEC;
174 __setup("noexec32=", nonx32_setup);
177 * When memory was added/removed make sure all the processes MM have
178 * suitable PGD entries in the local PGD level page.
180 void sync_global_pgds(unsigned long start, unsigned long end)
182 unsigned long address;
184 for (address = start; address <= end; address += PGDIR_SIZE) {
185 const pgd_t *pgd_ref = pgd_offset_k(address);
188 if (pgd_none(*pgd_ref))
191 spin_lock(&pgd_lock);
192 list_for_each_entry(page, &pgd_list, lru) {
194 spinlock_t *pgt_lock;
196 pgd = (pgd_t *)page_address(page) + pgd_index(address);
197 /* the pgt_lock only for Xen */
198 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
202 set_pgd(pgd, *pgd_ref);
204 BUG_ON(pgd_page_vaddr(*pgd)
205 != pgd_page_vaddr(*pgd_ref));
207 spin_unlock(pgt_lock);
209 spin_unlock(&pgd_lock);
214 * NOTE: This function is marked __ref because it calls __init function
215 * (alloc_bootmem_pages). It's safe to do it ONLY when after_bootmem == 0.
217 static __ref void *spp_getpage(void)
222 ptr = (void *) get_zeroed_page(GFP_ATOMIC | __GFP_NOTRACK);
224 ptr = alloc_bootmem_pages(PAGE_SIZE);
226 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
227 panic("set_pte_phys: cannot allocate page data %s\n",
228 after_bootmem ? "after bootmem" : "");
231 pr_debug("spp_getpage %p\n", ptr);
236 static pud_t *fill_pud(pgd_t *pgd, unsigned long vaddr)
238 if (pgd_none(*pgd)) {
239 pud_t *pud = (pud_t *)spp_getpage();
240 pgd_populate(&init_mm, pgd, pud);
241 if (pud != pud_offset(pgd, 0))
242 printk(KERN_ERR "PAGETABLE BUG #00! %p <-> %p\n",
243 pud, pud_offset(pgd, 0));
245 return pud_offset(pgd, vaddr);
248 static pmd_t *fill_pmd(pud_t *pud, unsigned long vaddr)
250 if (pud_none(*pud)) {
251 pmd_t *pmd = (pmd_t *) spp_getpage();
252 pud_populate(&init_mm, pud, pmd);
253 if (pmd != pmd_offset(pud, 0))
254 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
255 pmd, pmd_offset(pud, 0));
257 return pmd_offset(pud, vaddr);
260 static pte_t *fill_pte(pmd_t *pmd, unsigned long vaddr)
262 if (pmd_none(*pmd)) {
263 pte_t *pte = (pte_t *) spp_getpage();
264 pmd_populate_kernel(&init_mm, pmd, pte);
265 if (pte != pte_offset_kernel(pmd, 0))
266 printk(KERN_ERR "PAGETABLE BUG #02!\n");
268 return pte_offset_kernel(pmd, vaddr);
271 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte)
277 pud = pud_page + pud_index(vaddr);
278 pmd = fill_pmd(pud, vaddr);
279 pte = fill_pte(pmd, vaddr);
281 set_pte(pte, new_pte);
284 * It's enough to flush this one mapping.
285 * (PGE mappings get flushed as well)
287 __flush_tlb_one(vaddr);
290 void set_pte_vaddr(unsigned long vaddr, pte_t pteval)
295 pr_debug("set_pte_vaddr %lx to %lx\n", vaddr, native_pte_val(pteval));
297 pgd = pgd_offset_k(vaddr);
298 if (pgd_none(*pgd)) {
300 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
303 pud_page = (pud_t*)pgd_page_vaddr(*pgd);
304 set_pte_vaddr_pud(pud_page, vaddr, pteval);
307 pmd_t * __init populate_extra_pmd(unsigned long vaddr)
312 pgd = pgd_offset_k(vaddr);
313 pud = fill_pud(pgd, vaddr);
314 return fill_pmd(pud, vaddr);
317 pte_t * __init populate_extra_pte(unsigned long vaddr)
321 pmd = populate_extra_pmd(vaddr);
322 return fill_pte(pmd, vaddr);
326 * Create large page table mappings for a range of physical addresses.
328 static void __init __init_extra_mapping(unsigned long phys, unsigned long size,
335 BUG_ON((phys & ~PMD_MASK) || (size & ~PMD_MASK));
336 for (; size; phys += PMD_SIZE, size -= PMD_SIZE) {
337 pgd = pgd_offset_k((unsigned long)__va(phys));
338 if (pgd_none(*pgd)) {
339 pud = (pud_t *) spp_getpage();
340 set_pgd(pgd, __pgd(__pa(pud) | _KERNPG_TABLE |
343 pud = pud_offset(pgd, (unsigned long)__va(phys));
344 if (pud_none(*pud)) {
345 pmd = (pmd_t *) spp_getpage();
346 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE |
349 pmd = pmd_offset(pud, phys);
350 BUG_ON(!pmd_none(*pmd));
351 set_pmd(pmd, __pmd(phys | pgprot_val(prot)));
355 void __init init_extra_mapping_wb(unsigned long phys, unsigned long size)
357 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE);
360 void __init init_extra_mapping_uc(unsigned long phys, unsigned long size)
362 __init_extra_mapping(phys, size, PAGE_KERNEL_LARGE_NOCACHE);
366 * The head.S code sets up the kernel high mapping:
368 * from __START_KERNEL_map to __START_KERNEL_map + size (== _end-_text)
370 * phys_addr holds the negative offset to the kernel, which is added
371 * to the compile time generated pmds. This results in invalid pmds up
372 * to the point where we hit the physaddr 0 mapping.
374 * We limit the mappings to the region from _text to _brk_end. _brk_end
375 * is rounded up to the 2MB boundary. This catches the invalid pmds as
376 * well, as they are located before _text:
378 void __init cleanup_highmap(void)
380 unsigned long vaddr = __START_KERNEL_map;
381 unsigned long vaddr_end = __START_KERNEL_map + KERNEL_IMAGE_SIZE;
382 unsigned long end = roundup((unsigned long)_brk_end, PMD_SIZE) - 1;
383 pmd_t *pmd = level2_kernel_pgt;
386 * Native path, max_pfn_mapped is not set yet.
387 * Xen has valid max_pfn_mapped set in
388 * arch/x86/xen/mmu.c:xen_setup_kernel_pagetable().
391 vaddr_end = __START_KERNEL_map + (max_pfn_mapped << PAGE_SHIFT);
393 for (; vaddr + PMD_SIZE - 1 < vaddr_end; pmd++, vaddr += PMD_SIZE) {
396 if (vaddr < (unsigned long) _text || vaddr > end)
397 set_pmd(pmd, __pmd(0));
401 static unsigned long __meminit
402 phys_pte_init(pte_t *pte_page, unsigned long addr, unsigned long end,
405 unsigned long pages = 0, next;
406 unsigned long last_map_addr = end;
409 pte_t *pte = pte_page + pte_index(addr);
411 for (i = pte_index(addr); i < PTRS_PER_PTE; i++, addr = next, pte++) {
412 next = (addr & PAGE_MASK) + PAGE_SIZE;
414 if (!after_bootmem &&
415 !e820_any_mapped(addr & PAGE_MASK, next, E820_RAM) &&
416 !e820_any_mapped(addr & PAGE_MASK, next, E820_RESERVED_KERN))
417 set_pte(pte, __pte(0));
422 * We will re-use the existing mapping.
423 * Xen for example has some special requirements, like mapping
424 * pagetable pages as RO. So assume someone who pre-setup
425 * these mappings are more intelligent.
434 printk(" pte=%p addr=%lx pte=%016lx\n",
435 pte, addr, pfn_pte(addr >> PAGE_SHIFT, PAGE_KERNEL).pte);
437 set_pte(pte, pfn_pte(addr >> PAGE_SHIFT, prot));
438 last_map_addr = (addr & PAGE_MASK) + PAGE_SIZE;
441 update_page_count(PG_LEVEL_4K, pages);
443 return last_map_addr;
446 static unsigned long __meminit
447 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end,
448 unsigned long page_size_mask, pgprot_t prot)
450 unsigned long pages = 0, next;
451 unsigned long last_map_addr = end;
453 int i = pmd_index(address);
455 for (; i < PTRS_PER_PMD; i++, address = next) {
456 pmd_t *pmd = pmd_page + pmd_index(address);
458 pgprot_t new_prot = prot;
460 next = (address & PMD_MASK) + PMD_SIZE;
461 if (address >= end) {
462 if (!after_bootmem &&
463 !e820_any_mapped(address & PMD_MASK, next, E820_RAM) &&
464 !e820_any_mapped(address & PMD_MASK, next, E820_RESERVED_KERN))
465 set_pmd(pmd, __pmd(0));
470 if (!pmd_large(*pmd)) {
471 spin_lock(&init_mm.page_table_lock);
472 pte = (pte_t *)pmd_page_vaddr(*pmd);
473 last_map_addr = phys_pte_init(pte, address,
475 spin_unlock(&init_mm.page_table_lock);
479 * If we are ok with PG_LEVEL_2M mapping, then we will
480 * use the existing mapping,
482 * Otherwise, we will split the large page mapping but
483 * use the same existing protection bits except for
484 * large page, so that we don't violate Intel's TLB
485 * Application note (317080) which says, while changing
486 * the page sizes, new and old translations should
487 * not differ with respect to page frame and
490 if (page_size_mask & (1 << PG_LEVEL_2M)) {
493 last_map_addr = next;
496 new_prot = pte_pgprot(pte_clrhuge(*(pte_t *)pmd));
499 if (page_size_mask & (1<<PG_LEVEL_2M)) {
501 spin_lock(&init_mm.page_table_lock);
502 set_pte((pte_t *)pmd,
503 pfn_pte((address & PMD_MASK) >> PAGE_SHIFT,
504 __pgprot(pgprot_val(prot) | _PAGE_PSE)));
505 spin_unlock(&init_mm.page_table_lock);
506 last_map_addr = next;
510 pte = alloc_low_page();
511 last_map_addr = phys_pte_init(pte, address, end, new_prot);
513 spin_lock(&init_mm.page_table_lock);
514 pmd_populate_kernel(&init_mm, pmd, pte);
515 spin_unlock(&init_mm.page_table_lock);
517 update_page_count(PG_LEVEL_2M, pages);
518 return last_map_addr;
521 static unsigned long __meminit
522 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end,
523 unsigned long page_size_mask)
525 unsigned long pages = 0, next;
526 unsigned long last_map_addr = end;
527 int i = pud_index(addr);
529 for (; i < PTRS_PER_PUD; i++, addr = next) {
530 pud_t *pud = pud_page + pud_index(addr);
532 pgprot_t prot = PAGE_KERNEL;
534 next = (addr & PUD_MASK) + PUD_SIZE;
536 if (!after_bootmem &&
537 !e820_any_mapped(addr & PUD_MASK, next, E820_RAM) &&
538 !e820_any_mapped(addr & PUD_MASK, next, E820_RESERVED_KERN))
539 set_pud(pud, __pud(0));
544 if (!pud_large(*pud)) {
545 pmd = pmd_offset(pud, 0);
546 last_map_addr = phys_pmd_init(pmd, addr, end,
547 page_size_mask, prot);
552 * If we are ok with PG_LEVEL_1G mapping, then we will
553 * use the existing mapping.
555 * Otherwise, we will split the gbpage mapping but use
556 * the same existing protection bits except for large
557 * page, so that we don't violate Intel's TLB
558 * Application note (317080) which says, while changing
559 * the page sizes, new and old translations should
560 * not differ with respect to page frame and
563 if (page_size_mask & (1 << PG_LEVEL_1G)) {
566 last_map_addr = next;
569 prot = pte_pgprot(pte_clrhuge(*(pte_t *)pud));
572 if (page_size_mask & (1<<PG_LEVEL_1G)) {
574 spin_lock(&init_mm.page_table_lock);
575 set_pte((pte_t *)pud,
576 pfn_pte((addr & PUD_MASK) >> PAGE_SHIFT,
578 spin_unlock(&init_mm.page_table_lock);
579 last_map_addr = next;
583 pmd = alloc_low_page();
584 last_map_addr = phys_pmd_init(pmd, addr, end, page_size_mask,
587 spin_lock(&init_mm.page_table_lock);
588 pud_populate(&init_mm, pud, pmd);
589 spin_unlock(&init_mm.page_table_lock);
593 update_page_count(PG_LEVEL_1G, pages);
595 return last_map_addr;
598 unsigned long __meminit
599 kernel_physical_mapping_init(unsigned long start,
601 unsigned long page_size_mask)
603 bool pgd_changed = false;
604 unsigned long next, last_map_addr = end;
607 start = (unsigned long)__va(start);
608 end = (unsigned long)__va(end);
611 for (; start < end; start = next) {
612 pgd_t *pgd = pgd_offset_k(start);
615 next = (start & PGDIR_MASK) + PGDIR_SIZE;
618 pud = (pud_t *)pgd_page_vaddr(*pgd);
619 last_map_addr = phys_pud_init(pud, __pa(start),
620 __pa(end), page_size_mask);
624 pud = alloc_low_page();
625 last_map_addr = phys_pud_init(pud, __pa(start), __pa(end),
628 spin_lock(&init_mm.page_table_lock);
629 pgd_populate(&init_mm, pgd, pud);
630 spin_unlock(&init_mm.page_table_lock);
635 sync_global_pgds(addr, end - 1);
639 return last_map_addr;
643 void __init initmem_init(void)
645 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
649 void __init paging_init(void)
651 sparse_memory_present_with_active_regions(MAX_NUMNODES);
655 * clear the default setting with node 0
656 * note: don't use nodes_clear here, that is really clearing when
657 * numa support is not compiled in, and later node_set_state
658 * will not set it back.
660 node_clear_state(0, N_MEMORY);
661 if (N_MEMORY != N_NORMAL_MEMORY)
662 node_clear_state(0, N_NORMAL_MEMORY);
668 * Memory hotplug specific functions
670 #ifdef CONFIG_MEMORY_HOTPLUG
672 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need
675 static void update_end_of_memory_vars(u64 start, u64 size)
677 unsigned long end_pfn = PFN_UP(start + size);
679 if (end_pfn > max_pfn) {
681 max_low_pfn = end_pfn;
682 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1;
687 * Memory is added always to NORMAL zone. This means you will never get
688 * additional DMA/DMA32 memory.
690 int arch_add_memory(int nid, u64 start, u64 size)
692 struct pglist_data *pgdat = NODE_DATA(nid);
693 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
694 unsigned long start_pfn = start >> PAGE_SHIFT;
695 unsigned long nr_pages = size >> PAGE_SHIFT;
698 init_memory_mapping(start, start + size);
700 ret = __add_pages(nid, zone, start_pfn, nr_pages);
703 /* update max_pfn, max_low_pfn and high_memory */
704 update_end_of_memory_vars(start, size);
708 EXPORT_SYMBOL_GPL(arch_add_memory);
710 #define PAGE_INUSE 0xFD
712 static void __meminit free_pagetable(struct page *page, int order)
715 bool bootmem = false;
717 unsigned int nr_pages = 1 << order;
719 /* bootmem page has reserved flag */
720 if (PageReserved(page)) {
721 __ClearPageReserved(page);
724 magic = (unsigned long)page->lru.next;
725 if (magic == SECTION_INFO || magic == MIX_SECTION_INFO) {
727 put_page_bootmem(page++);
729 __free_pages_bootmem(page, order);
731 free_pages((unsigned long)page_address(page), order);
734 * SECTION_INFO pages and MIX_SECTION_INFO pages
735 * are all allocated by bootmem.
738 zone = page_zone(page);
739 zone_span_writelock(zone);
740 zone->present_pages += nr_pages;
741 zone_span_writeunlock(zone);
742 totalram_pages += nr_pages;
746 static void __meminit free_pte_table(pte_t *pte_start, pmd_t *pmd)
751 for (i = 0; i < PTRS_PER_PTE; i++) {
757 /* free a pte talbe */
758 free_pagetable(pmd_page(*pmd), 0);
759 spin_lock(&init_mm.page_table_lock);
761 spin_unlock(&init_mm.page_table_lock);
764 static void __meminit free_pmd_table(pmd_t *pmd_start, pud_t *pud)
769 for (i = 0; i < PTRS_PER_PMD; i++) {
775 /* free a pmd talbe */
776 free_pagetable(pud_page(*pud), 0);
777 spin_lock(&init_mm.page_table_lock);
779 spin_unlock(&init_mm.page_table_lock);
782 /* Return true if pgd is changed, otherwise return false. */
783 static bool __meminit free_pud_table(pud_t *pud_start, pgd_t *pgd)
788 for (i = 0; i < PTRS_PER_PUD; i++) {
794 /* free a pud table */
795 free_pagetable(pgd_page(*pgd), 0);
796 spin_lock(&init_mm.page_table_lock);
798 spin_unlock(&init_mm.page_table_lock);
803 static void __meminit
804 remove_pte_table(pte_t *pte_start, unsigned long addr, unsigned long end,
807 unsigned long next, pages = 0;
810 phys_addr_t phys_addr;
812 pte = pte_start + pte_index(addr);
813 for (; addr < end; addr = next, pte++) {
814 next = (addr + PAGE_SIZE) & PAGE_MASK;
818 if (!pte_present(*pte))
822 * We mapped [0,1G) memory as identity mapping when
823 * initializing, in arch/x86/kernel/head_64.S. These
824 * pagetables cannot be removed.
826 phys_addr = pte_val(*pte) + (addr & PAGE_MASK);
827 if (phys_addr < (phys_addr_t)0x40000000)
830 if (IS_ALIGNED(addr, PAGE_SIZE) &&
831 IS_ALIGNED(next, PAGE_SIZE)) {
833 * Do not free direct mapping pages since they were
834 * freed when offlining, or simplely not in use.
837 free_pagetable(pte_page(*pte), 0);
839 spin_lock(&init_mm.page_table_lock);
840 pte_clear(&init_mm, addr, pte);
841 spin_unlock(&init_mm.page_table_lock);
843 /* For non-direct mapping, pages means nothing. */
847 * If we are here, we are freeing vmemmap pages since
848 * direct mapped memory ranges to be freed are aligned.
850 * If we are not removing the whole page, it means
851 * other page structs in this page are being used and
852 * we canot remove them. So fill the unused page_structs
853 * with 0xFD, and remove the page when it is wholly
856 memset((void *)addr, PAGE_INUSE, next - addr);
858 page_addr = page_address(pte_page(*pte));
859 if (!memchr_inv(page_addr, PAGE_INUSE, PAGE_SIZE)) {
860 free_pagetable(pte_page(*pte), 0);
862 spin_lock(&init_mm.page_table_lock);
863 pte_clear(&init_mm, addr, pte);
864 spin_unlock(&init_mm.page_table_lock);
869 /* Call free_pte_table() in remove_pmd_table(). */
872 update_page_count(PG_LEVEL_4K, -pages);
875 static void __meminit
876 remove_pmd_table(pmd_t *pmd_start, unsigned long addr, unsigned long end,
879 unsigned long next, pages = 0;
884 pmd = pmd_start + pmd_index(addr);
885 for (; addr < end; addr = next, pmd++) {
886 next = pmd_addr_end(addr, end);
888 if (!pmd_present(*pmd))
891 if (pmd_large(*pmd)) {
892 if (IS_ALIGNED(addr, PMD_SIZE) &&
893 IS_ALIGNED(next, PMD_SIZE)) {
895 free_pagetable(pmd_page(*pmd),
896 get_order(PMD_SIZE));
898 spin_lock(&init_mm.page_table_lock);
900 spin_unlock(&init_mm.page_table_lock);
903 /* If here, we are freeing vmemmap pages. */
904 memset((void *)addr, PAGE_INUSE, next - addr);
906 page_addr = page_address(pmd_page(*pmd));
907 if (!memchr_inv(page_addr, PAGE_INUSE,
909 free_pagetable(pmd_page(*pmd),
910 get_order(PMD_SIZE));
912 spin_lock(&init_mm.page_table_lock);
914 spin_unlock(&init_mm.page_table_lock);
921 pte_base = (pte_t *)pmd_page_vaddr(*pmd);
922 remove_pte_table(pte_base, addr, next, direct);
923 free_pte_table(pte_base, pmd);
926 /* Call free_pmd_table() in remove_pud_table(). */
928 update_page_count(PG_LEVEL_2M, -pages);
931 static void __meminit
932 remove_pud_table(pud_t *pud_start, unsigned long addr, unsigned long end,
935 unsigned long next, pages = 0;
940 pud = pud_start + pud_index(addr);
941 for (; addr < end; addr = next, pud++) {
942 next = pud_addr_end(addr, end);
944 if (!pud_present(*pud))
947 if (pud_large(*pud)) {
948 if (IS_ALIGNED(addr, PUD_SIZE) &&
949 IS_ALIGNED(next, PUD_SIZE)) {
951 free_pagetable(pud_page(*pud),
952 get_order(PUD_SIZE));
954 spin_lock(&init_mm.page_table_lock);
956 spin_unlock(&init_mm.page_table_lock);
959 /* If here, we are freeing vmemmap pages. */
960 memset((void *)addr, PAGE_INUSE, next - addr);
962 page_addr = page_address(pud_page(*pud));
963 if (!memchr_inv(page_addr, PAGE_INUSE,
965 free_pagetable(pud_page(*pud),
966 get_order(PUD_SIZE));
968 spin_lock(&init_mm.page_table_lock);
970 spin_unlock(&init_mm.page_table_lock);
977 pmd_base = (pmd_t *)pud_page_vaddr(*pud);
978 remove_pmd_table(pmd_base, addr, next, direct);
979 free_pmd_table(pmd_base, pud);
983 update_page_count(PG_LEVEL_1G, -pages);
986 /* start and end are both virtual address. */
987 static void __meminit
988 remove_pagetable(unsigned long start, unsigned long end, bool direct)
993 bool pgd_changed = false;
995 for (; start < end; start = next) {
996 next = pgd_addr_end(start, end);
998 pgd = pgd_offset_k(start);
999 if (!pgd_present(*pgd))
1002 pud = (pud_t *)pgd_page_vaddr(*pgd);
1003 remove_pud_table(pud, start, next, direct);
1004 if (free_pud_table(pud, pgd))
1009 sync_global_pgds(start, end - 1);
1014 void __ref vmemmap_free(struct page *memmap, unsigned long nr_pages)
1016 unsigned long start = (unsigned long)memmap;
1017 unsigned long end = (unsigned long)(memmap + nr_pages);
1019 remove_pagetable(start, end, false);
1022 static void __meminit
1023 kernel_physical_mapping_remove(unsigned long start, unsigned long end)
1025 start = (unsigned long)__va(start);
1026 end = (unsigned long)__va(end);
1028 remove_pagetable(start, end, true);
1031 #ifdef CONFIG_MEMORY_HOTREMOVE
1032 int __ref arch_remove_memory(u64 start, u64 size)
1034 unsigned long start_pfn = start >> PAGE_SHIFT;
1035 unsigned long nr_pages = size >> PAGE_SHIFT;
1039 zone = page_zone(pfn_to_page(start_pfn));
1040 kernel_physical_mapping_remove(start, start + size);
1041 ret = __remove_pages(zone, start_pfn, nr_pages);
1047 #endif /* CONFIG_MEMORY_HOTPLUG */
1049 static struct kcore_list kcore_vsyscall;
1051 static void __init register_page_bootmem_info(void)
1056 for_each_online_node(i)
1057 register_page_bootmem_info_node(NODE_DATA(i));
1061 void __init mem_init(void)
1063 long codesize, reservedpages, datasize, initsize;
1064 unsigned long absent_pages;
1068 /* clear_bss() already clear the empty_zero_page */
1072 /* this will put all low memory onto the freelists */
1073 register_page_bootmem_info();
1074 totalram_pages = free_all_bootmem();
1076 absent_pages = absent_pages_in_range(0, max_pfn);
1077 reservedpages = max_pfn - totalram_pages - absent_pages;
1080 codesize = (unsigned long) &_etext - (unsigned long) &_text;
1081 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
1082 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
1084 /* Register memory areas for /proc/kcore */
1085 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
1086 VSYSCALL_END - VSYSCALL_START, KCORE_OTHER);
1088 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
1089 "%ldk absent, %ldk reserved, %ldk data, %ldk init)\n",
1090 nr_free_pages() << (PAGE_SHIFT-10),
1091 max_pfn << (PAGE_SHIFT-10),
1093 absent_pages << (PAGE_SHIFT-10),
1094 reservedpages << (PAGE_SHIFT-10),
1099 #ifdef CONFIG_DEBUG_RODATA
1100 const int rodata_test_data = 0xC3;
1101 EXPORT_SYMBOL_GPL(rodata_test_data);
1103 int kernel_set_to_readonly;
1105 void set_kernel_text_rw(void)
1107 unsigned long start = PFN_ALIGN(_text);
1108 unsigned long end = PFN_ALIGN(__stop___ex_table);
1110 if (!kernel_set_to_readonly)
1113 pr_debug("Set kernel text: %lx - %lx for read write\n",
1117 * Make the kernel identity mapping for text RW. Kernel text
1118 * mapping will always be RO. Refer to the comment in
1119 * static_protections() in pageattr.c
1121 set_memory_rw(start, (end - start) >> PAGE_SHIFT);
1124 void set_kernel_text_ro(void)
1126 unsigned long start = PFN_ALIGN(_text);
1127 unsigned long end = PFN_ALIGN(__stop___ex_table);
1129 if (!kernel_set_to_readonly)
1132 pr_debug("Set kernel text: %lx - %lx for read only\n",
1136 * Set the kernel identity mapping for text RO.
1138 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1141 void mark_rodata_ro(void)
1143 unsigned long start = PFN_ALIGN(_text);
1144 unsigned long rodata_start = PFN_ALIGN(__start_rodata);
1145 unsigned long end = (unsigned long) &__end_rodata_hpage_align;
1146 unsigned long text_end = PFN_ALIGN(&__stop___ex_table);
1147 unsigned long rodata_end = PFN_ALIGN(&__end_rodata);
1148 unsigned long all_end = PFN_ALIGN(&_end);
1150 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1151 (end - start) >> 10);
1152 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
1154 kernel_set_to_readonly = 1;
1157 * The rodata/data/bss/brk section (but not the kernel text!)
1158 * should also be not-executable.
1160 set_memory_nx(rodata_start, (all_end - rodata_start) >> PAGE_SHIFT);
1164 #ifdef CONFIG_CPA_DEBUG
1165 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
1166 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
1168 printk(KERN_INFO "Testing CPA: again\n");
1169 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
1172 free_init_pages("unused kernel memory",
1173 (unsigned long) __va(__pa_symbol(text_end)),
1174 (unsigned long) __va(__pa_symbol(rodata_start)));
1176 free_init_pages("unused kernel memory",
1177 (unsigned long) __va(__pa_symbol(rodata_end)),
1178 (unsigned long) __va(__pa_symbol(_sdata)));
1183 int kern_addr_valid(unsigned long addr)
1185 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
1191 if (above != 0 && above != -1UL)
1194 pgd = pgd_offset_k(addr);
1198 pud = pud_offset(pgd, addr);
1202 if (pud_large(*pud))
1203 return pfn_valid(pud_pfn(*pud));
1205 pmd = pmd_offset(pud, addr);
1209 if (pmd_large(*pmd))
1210 return pfn_valid(pmd_pfn(*pmd));
1212 pte = pte_offset_kernel(pmd, addr);
1216 return pfn_valid(pte_pfn(*pte));
1220 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
1221 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
1222 * not need special handling anymore:
1224 static struct vm_area_struct gate_vma = {
1225 .vm_start = VSYSCALL_START,
1226 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
1227 .vm_page_prot = PAGE_READONLY_EXEC,
1228 .vm_flags = VM_READ | VM_EXEC
1231 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
1233 #ifdef CONFIG_IA32_EMULATION
1234 if (!mm || mm->context.ia32_compat)
1240 int in_gate_area(struct mm_struct *mm, unsigned long addr)
1242 struct vm_area_struct *vma = get_gate_vma(mm);
1247 return (addr >= vma->vm_start) && (addr < vma->vm_end);
1251 * Use this when you have no reliable mm, typically from interrupt
1252 * context. It is less reliable than using a task's mm and may give
1255 int in_gate_area_no_mm(unsigned long addr)
1257 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
1260 const char *arch_vma_name(struct vm_area_struct *vma)
1262 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
1264 if (vma == &gate_vma)
1265 return "[vsyscall]";
1269 #ifdef CONFIG_X86_UV
1270 unsigned long memory_block_size_bytes(void)
1272 if (is_uv_system()) {
1273 printk(KERN_INFO "UV: memory block size 2GB\n");
1274 return 2UL * 1024 * 1024 * 1024;
1276 return MIN_MEMORY_BLOCK_SIZE;
1280 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1282 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
1284 static long __meminitdata addr_start, addr_end;
1285 static void __meminitdata *p_start, *p_end;
1286 static int __meminitdata node_start;
1289 vmemmap_populate(struct page *start_page, unsigned long size, int node)
1291 unsigned long addr = (unsigned long)start_page;
1292 unsigned long end = (unsigned long)(start_page + size);
1298 for (; addr < end; addr = next) {
1301 pgd = vmemmap_pgd_populate(addr, node);
1305 pud = vmemmap_pud_populate(pgd, addr, node);
1310 next = (addr + PAGE_SIZE) & PAGE_MASK;
1311 pmd = vmemmap_pmd_populate(pud, addr, node);
1316 p = vmemmap_pte_populate(pmd, addr, node);
1321 addr_end = addr + PAGE_SIZE;
1322 p_end = p + PAGE_SIZE;
1324 next = pmd_addr_end(addr, end);
1326 pmd = pmd_offset(pud, addr);
1327 if (pmd_none(*pmd)) {
1330 p = vmemmap_alloc_block_buf(PMD_SIZE, node);
1334 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
1336 set_pmd(pmd, __pmd(pte_val(entry)));
1338 /* check to see if we have contiguous blocks */
1339 if (p_end != p || node_start != node) {
1341 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1342 addr_start, addr_end-1, p_start, p_end-1, node_start);
1348 addr_end = addr + PMD_SIZE;
1349 p_end = p + PMD_SIZE;
1351 vmemmap_verify((pte_t *)pmd, node, addr, next);
1355 sync_global_pgds((unsigned long)start_page, end - 1);
1359 #if defined(CONFIG_MEMORY_HOTPLUG_SPARSE) && defined(CONFIG_HAVE_BOOTMEM_INFO_NODE)
1360 void register_page_bootmem_memmap(unsigned long section_nr,
1361 struct page *start_page, unsigned long size)
1363 unsigned long addr = (unsigned long)start_page;
1364 unsigned long end = (unsigned long)(start_page + size);
1369 unsigned int nr_pages;
1372 for (; addr < end; addr = next) {
1375 pgd = pgd_offset_k(addr);
1376 if (pgd_none(*pgd)) {
1377 next = (addr + PAGE_SIZE) & PAGE_MASK;
1380 get_page_bootmem(section_nr, pgd_page(*pgd), MIX_SECTION_INFO);
1382 pud = pud_offset(pgd, addr);
1383 if (pud_none(*pud)) {
1384 next = (addr + PAGE_SIZE) & PAGE_MASK;
1387 get_page_bootmem(section_nr, pud_page(*pud), MIX_SECTION_INFO);
1390 next = (addr + PAGE_SIZE) & PAGE_MASK;
1391 pmd = pmd_offset(pud, addr);
1394 get_page_bootmem(section_nr, pmd_page(*pmd),
1397 pte = pte_offset_kernel(pmd, addr);
1400 get_page_bootmem(section_nr, pte_page(*pte),
1403 next = pmd_addr_end(addr, end);
1405 pmd = pmd_offset(pud, addr);
1409 nr_pages = 1 << (get_order(PMD_SIZE));
1410 page = pmd_page(*pmd);
1412 get_page_bootmem(section_nr, page++,
1419 void __meminit vmemmap_populate_print_last(void)
1422 printk(KERN_DEBUG " [%lx-%lx] PMD -> [%p-%p] on node %d\n",
1423 addr_start, addr_end-1, p_start, p_end-1, node_start);