4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
41 #include <linux/kernel_stat.h>
43 #include <linux/hugetlb.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/highmem.h>
47 #include <linux/pagemap.h>
48 #include <linux/rmap.h>
49 #include <linux/module.h>
50 #include <linux/init.h>
52 #include <asm/pgalloc.h>
53 #include <asm/uaccess.h>
55 #include <asm/tlbflush.h>
56 #include <asm/pgtable.h>
58 #include <linux/swapops.h>
59 #include <linux/elf.h>
61 #ifndef CONFIG_NEED_MULTIPLE_NODES
62 /* use the per-pgdat data instead for discontigmem - mbligh */
63 unsigned long max_mapnr;
66 EXPORT_SYMBOL(max_mapnr);
67 EXPORT_SYMBOL(mem_map);
70 unsigned long num_physpages;
72 * A number of key systems in x86 including ioremap() rely on the assumption
73 * that high_memory defines the upper bound on direct map memory, then end
74 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
75 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
79 unsigned long vmalloc_earlyreserve;
81 EXPORT_SYMBOL(num_physpages);
82 EXPORT_SYMBOL(high_memory);
83 EXPORT_SYMBOL(vmalloc_earlyreserve);
86 * If a p?d_bad entry is found while walking page tables, report
87 * the error, before resetting entry to p?d_none. Usually (but
88 * very seldom) called out from the p?d_none_or_clear_bad macros.
91 void pgd_clear_bad(pgd_t *pgd)
97 void pud_clear_bad(pud_t *pud)
103 void pmd_clear_bad(pmd_t *pmd)
110 * Note: this doesn't free the actual pages themselves. That
111 * has been handled earlier when unmapping all the memory regions.
113 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
115 struct page *page = pmd_page(*pmd);
117 pte_free_tlb(tlb, page);
118 dec_page_state(nr_page_table_pages);
122 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
123 unsigned long addr, unsigned long end,
124 unsigned long floor, unsigned long ceiling)
131 pmd = pmd_offset(pud, addr);
133 next = pmd_addr_end(addr, end);
134 if (pmd_none_or_clear_bad(pmd))
136 free_pte_range(tlb, pmd);
137 } while (pmd++, addr = next, addr != end);
147 if (end - 1 > ceiling - 1)
150 pmd = pmd_offset(pud, start);
152 pmd_free_tlb(tlb, pmd);
155 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
156 unsigned long addr, unsigned long end,
157 unsigned long floor, unsigned long ceiling)
164 pud = pud_offset(pgd, addr);
166 next = pud_addr_end(addr, end);
167 if (pud_none_or_clear_bad(pud))
169 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
170 } while (pud++, addr = next, addr != end);
176 ceiling &= PGDIR_MASK;
180 if (end - 1 > ceiling - 1)
183 pud = pud_offset(pgd, start);
185 pud_free_tlb(tlb, pud);
189 * This function frees user-level page tables of a process.
191 * Must be called with pagetable lock held.
193 void free_pgd_range(struct mmu_gather **tlb,
194 unsigned long addr, unsigned long end,
195 unsigned long floor, unsigned long ceiling)
202 * The next few lines have given us lots of grief...
204 * Why are we testing PMD* at this top level? Because often
205 * there will be no work to do at all, and we'd prefer not to
206 * go all the way down to the bottom just to discover that.
208 * Why all these "- 1"s? Because 0 represents both the bottom
209 * of the address space and the top of it (using -1 for the
210 * top wouldn't help much: the masks would do the wrong thing).
211 * The rule is that addr 0 and floor 0 refer to the bottom of
212 * the address space, but end 0 and ceiling 0 refer to the top
213 * Comparisons need to use "end - 1" and "ceiling - 1" (though
214 * that end 0 case should be mythical).
216 * Wherever addr is brought up or ceiling brought down, we must
217 * be careful to reject "the opposite 0" before it confuses the
218 * subsequent tests. But what about where end is brought down
219 * by PMD_SIZE below? no, end can't go down to 0 there.
221 * Whereas we round start (addr) and ceiling down, by different
222 * masks at different levels, in order to test whether a table
223 * now has no other vmas using it, so can be freed, we don't
224 * bother to round floor or end up - the tests don't need that.
238 if (end - 1 > ceiling - 1)
244 pgd = pgd_offset((*tlb)->mm, addr);
246 next = pgd_addr_end(addr, end);
247 if (pgd_none_or_clear_bad(pgd))
249 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
250 } while (pgd++, addr = next, addr != end);
252 if (!tlb_is_full_mm(*tlb))
253 flush_tlb_pgtables((*tlb)->mm, start, end);
256 void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
257 unsigned long floor, unsigned long ceiling)
260 struct vm_area_struct *next = vma->vm_next;
261 unsigned long addr = vma->vm_start;
263 if (is_hugepage_only_range(vma->vm_mm, addr, HPAGE_SIZE)) {
264 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
265 floor, next? next->vm_start: ceiling);
268 * Optimization: gather nearby vmas into one call down
270 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
271 && !is_hugepage_only_range(vma->vm_mm, next->vm_start,
276 free_pgd_range(tlb, addr, vma->vm_end,
277 floor, next? next->vm_start: ceiling);
283 pte_t fastcall *pte_alloc_map(struct mm_struct *mm, pmd_t *pmd,
284 unsigned long address)
286 if (!pmd_present(*pmd)) {
289 spin_unlock(&mm->page_table_lock);
290 new = pte_alloc_one(mm, address);
291 spin_lock(&mm->page_table_lock);
295 * Because we dropped the lock, we should re-check the
296 * entry, as somebody else could have populated it..
298 if (pmd_present(*pmd)) {
303 inc_page_state(nr_page_table_pages);
304 pmd_populate(mm, pmd, new);
307 return pte_offset_map(pmd, address);
310 pte_t fastcall * pte_alloc_kernel(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
312 if (!pmd_present(*pmd)) {
315 spin_unlock(&mm->page_table_lock);
316 new = pte_alloc_one_kernel(mm, address);
317 spin_lock(&mm->page_table_lock);
322 * Because we dropped the lock, we should re-check the
323 * entry, as somebody else could have populated it..
325 if (pmd_present(*pmd)) {
326 pte_free_kernel(new);
329 pmd_populate_kernel(mm, pmd, new);
332 return pte_offset_kernel(pmd, address);
336 * copy one vm_area from one task to the other. Assumes the page tables
337 * already present in the new task to be cleared in the whole range
338 * covered by this vma.
340 * dst->page_table_lock is held on entry and exit,
341 * but may be dropped within p[mg]d_alloc() and pte_alloc_map().
345 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
346 pte_t *dst_pte, pte_t *src_pte, unsigned long vm_flags,
349 pte_t pte = *src_pte;
353 /* pte contains position in swap or file, so copy. */
354 if (unlikely(!pte_present(pte))) {
355 if (!pte_file(pte)) {
356 swap_duplicate(pte_to_swp_entry(pte));
357 /* make sure dst_mm is on swapoff's mmlist. */
358 if (unlikely(list_empty(&dst_mm->mmlist))) {
359 spin_lock(&mmlist_lock);
360 list_add(&dst_mm->mmlist, &src_mm->mmlist);
361 spin_unlock(&mmlist_lock);
364 set_pte_at(dst_mm, addr, dst_pte, pte);
369 /* the pte points outside of valid memory, the
370 * mapping is assumed to be good, meaningful
371 * and not mapped via rmap - duplicate the
376 page = pfn_to_page(pfn);
378 if (!page || PageReserved(page)) {
379 set_pte_at(dst_mm, addr, dst_pte, pte);
384 * If it's a COW mapping, write protect it both
385 * in the parent and the child
387 if ((vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE) {
388 ptep_set_wrprotect(src_mm, addr, src_pte);
393 * If it's a shared mapping, mark it clean in
396 if (vm_flags & VM_SHARED)
397 pte = pte_mkclean(pte);
398 pte = pte_mkold(pte);
400 inc_mm_counter(dst_mm, rss);
402 inc_mm_counter(dst_mm, anon_rss);
403 set_pte_at(dst_mm, addr, dst_pte, pte);
407 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
408 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
409 unsigned long addr, unsigned long end)
411 pte_t *src_pte, *dst_pte;
412 unsigned long vm_flags = vma->vm_flags;
416 dst_pte = pte_alloc_map(dst_mm, dst_pmd, addr);
419 src_pte = pte_offset_map_nested(src_pmd, addr);
422 spin_lock(&src_mm->page_table_lock);
425 * We are holding two locks at this point - either of them
426 * could generate latencies in another task on another CPU.
428 if (progress >= 32 && (need_resched() ||
429 need_lockbreak(&src_mm->page_table_lock) ||
430 need_lockbreak(&dst_mm->page_table_lock)))
432 if (pte_none(*src_pte)) {
436 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vm_flags, addr);
438 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
439 spin_unlock(&src_mm->page_table_lock);
441 pte_unmap_nested(src_pte - 1);
442 pte_unmap(dst_pte - 1);
443 cond_resched_lock(&dst_mm->page_table_lock);
449 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
450 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
451 unsigned long addr, unsigned long end)
453 pmd_t *src_pmd, *dst_pmd;
456 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
459 src_pmd = pmd_offset(src_pud, addr);
461 next = pmd_addr_end(addr, end);
462 if (pmd_none_or_clear_bad(src_pmd))
464 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
467 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
471 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
472 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
473 unsigned long addr, unsigned long end)
475 pud_t *src_pud, *dst_pud;
478 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
481 src_pud = pud_offset(src_pgd, addr);
483 next = pud_addr_end(addr, end);
484 if (pud_none_or_clear_bad(src_pud))
486 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
489 } while (dst_pud++, src_pud++, addr = next, addr != end);
493 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
494 struct vm_area_struct *vma)
496 pgd_t *src_pgd, *dst_pgd;
498 unsigned long addr = vma->vm_start;
499 unsigned long end = vma->vm_end;
501 if (is_vm_hugetlb_page(vma))
502 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
504 dst_pgd = pgd_offset(dst_mm, addr);
505 src_pgd = pgd_offset(src_mm, addr);
507 next = pgd_addr_end(addr, end);
508 if (pgd_none_or_clear_bad(src_pgd))
510 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
513 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
517 static void zap_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
518 unsigned long addr, unsigned long end,
519 struct zap_details *details)
523 pte = pte_offset_map(pmd, addr);
528 if (pte_present(ptent)) {
529 struct page *page = NULL;
530 unsigned long pfn = pte_pfn(ptent);
531 if (pfn_valid(pfn)) {
532 page = pfn_to_page(pfn);
533 if (PageReserved(page))
536 if (unlikely(details) && page) {
538 * unmap_shared_mapping_pages() wants to
539 * invalidate cache without truncating:
540 * unmap shared but keep private pages.
542 if (details->check_mapping &&
543 details->check_mapping != page->mapping)
546 * Each page->index must be checked when
547 * invalidating or truncating nonlinear.
549 if (details->nonlinear_vma &&
550 (page->index < details->first_index ||
551 page->index > details->last_index))
554 ptent = ptep_get_and_clear(tlb->mm, addr, pte);
555 tlb_remove_tlb_entry(tlb, pte, addr);
558 if (unlikely(details) && details->nonlinear_vma
559 && linear_page_index(details->nonlinear_vma,
560 addr) != page->index)
561 set_pte_at(tlb->mm, addr, pte,
562 pgoff_to_pte(page->index));
563 if (pte_dirty(ptent))
564 set_page_dirty(page);
566 dec_mm_counter(tlb->mm, anon_rss);
567 else if (pte_young(ptent))
568 mark_page_accessed(page);
570 page_remove_rmap(page);
571 tlb_remove_page(tlb, page);
575 * If details->check_mapping, we leave swap entries;
576 * if details->nonlinear_vma, we leave file entries.
578 if (unlikely(details))
580 if (!pte_file(ptent))
581 free_swap_and_cache(pte_to_swp_entry(ptent));
582 pte_clear(tlb->mm, addr, pte);
583 } while (pte++, addr += PAGE_SIZE, addr != end);
587 static inline void zap_pmd_range(struct mmu_gather *tlb, pud_t *pud,
588 unsigned long addr, unsigned long end,
589 struct zap_details *details)
594 pmd = pmd_offset(pud, addr);
596 next = pmd_addr_end(addr, end);
597 if (pmd_none_or_clear_bad(pmd))
599 zap_pte_range(tlb, pmd, addr, next, details);
600 } while (pmd++, addr = next, addr != end);
603 static inline void zap_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
604 unsigned long addr, unsigned long end,
605 struct zap_details *details)
610 pud = pud_offset(pgd, addr);
612 next = pud_addr_end(addr, end);
613 if (pud_none_or_clear_bad(pud))
615 zap_pmd_range(tlb, pud, addr, next, details);
616 } while (pud++, addr = next, addr != end);
619 static void unmap_page_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
620 unsigned long addr, unsigned long end,
621 struct zap_details *details)
626 if (details && !details->check_mapping && !details->nonlinear_vma)
630 tlb_start_vma(tlb, vma);
631 pgd = pgd_offset(vma->vm_mm, addr);
633 next = pgd_addr_end(addr, end);
634 if (pgd_none_or_clear_bad(pgd))
636 zap_pud_range(tlb, pgd, addr, next, details);
637 } while (pgd++, addr = next, addr != end);
638 tlb_end_vma(tlb, vma);
641 #ifdef CONFIG_PREEMPT
642 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
644 /* No preempt: go for improved straight-line efficiency */
645 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
649 * unmap_vmas - unmap a range of memory covered by a list of vma's
650 * @tlbp: address of the caller's struct mmu_gather
651 * @mm: the controlling mm_struct
652 * @vma: the starting vma
653 * @start_addr: virtual address at which to start unmapping
654 * @end_addr: virtual address at which to end unmapping
655 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
656 * @details: details of nonlinear truncation or shared cache invalidation
658 * Returns the end address of the unmapping (restart addr if interrupted).
660 * Unmap all pages in the vma list. Called under page_table_lock.
662 * We aim to not hold page_table_lock for too long (for scheduling latency
663 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
664 * return the ending mmu_gather to the caller.
666 * Only addresses between `start' and `end' will be unmapped.
668 * The VMA list must be sorted in ascending virtual address order.
670 * unmap_vmas() assumes that the caller will flush the whole unmapped address
671 * range after unmap_vmas() returns. So the only responsibility here is to
672 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
673 * drops the lock and schedules.
675 unsigned long unmap_vmas(struct mmu_gather **tlbp, struct mm_struct *mm,
676 struct vm_area_struct *vma, unsigned long start_addr,
677 unsigned long end_addr, unsigned long *nr_accounted,
678 struct zap_details *details)
680 unsigned long zap_bytes = ZAP_BLOCK_SIZE;
681 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
682 int tlb_start_valid = 0;
683 unsigned long start = start_addr;
684 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
685 int fullmm = tlb_is_full_mm(*tlbp);
687 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
690 start = max(vma->vm_start, start_addr);
691 if (start >= vma->vm_end)
693 end = min(vma->vm_end, end_addr);
694 if (end <= vma->vm_start)
697 if (vma->vm_flags & VM_ACCOUNT)
698 *nr_accounted += (end - start) >> PAGE_SHIFT;
700 while (start != end) {
703 if (!tlb_start_valid) {
708 if (is_vm_hugetlb_page(vma)) {
710 unmap_hugepage_range(vma, start, end);
712 block = min(zap_bytes, end - start);
713 unmap_page_range(*tlbp, vma, start,
714 start + block, details);
719 if ((long)zap_bytes > 0)
722 tlb_finish_mmu(*tlbp, tlb_start, start);
724 if (need_resched() ||
725 need_lockbreak(&mm->page_table_lock) ||
726 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
728 /* must reset count of rss freed */
729 *tlbp = tlb_gather_mmu(mm, fullmm);
732 spin_unlock(&mm->page_table_lock);
734 spin_lock(&mm->page_table_lock);
737 *tlbp = tlb_gather_mmu(mm, fullmm);
739 zap_bytes = ZAP_BLOCK_SIZE;
743 return start; /* which is now the end (or restart) address */
747 * zap_page_range - remove user pages in a given range
748 * @vma: vm_area_struct holding the applicable pages
749 * @address: starting address of pages to zap
750 * @size: number of bytes to zap
751 * @details: details of nonlinear truncation or shared cache invalidation
753 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
754 unsigned long size, struct zap_details *details)
756 struct mm_struct *mm = vma->vm_mm;
757 struct mmu_gather *tlb;
758 unsigned long end = address + size;
759 unsigned long nr_accounted = 0;
761 if (is_vm_hugetlb_page(vma)) {
762 zap_hugepage_range(vma, address, size);
767 spin_lock(&mm->page_table_lock);
768 tlb = tlb_gather_mmu(mm, 0);
769 end = unmap_vmas(&tlb, mm, vma, address, end, &nr_accounted, details);
770 tlb_finish_mmu(tlb, address, end);
771 spin_unlock(&mm->page_table_lock);
776 * Do a quick page-table lookup for a single page.
777 * mm->page_table_lock must be held.
779 static struct page *__follow_page(struct mm_struct *mm, unsigned long address,
780 int read, int write, int accessed)
789 page = follow_huge_addr(mm, address, write);
793 pgd = pgd_offset(mm, address);
794 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
797 pud = pud_offset(pgd, address);
798 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
801 pmd = pmd_offset(pud, address);
802 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
805 return follow_huge_pmd(mm, address, pmd, write);
807 ptep = pte_offset_map(pmd, address);
813 if (pte_present(pte)) {
814 if (write && !pte_write(pte))
816 if (read && !pte_read(pte))
819 if (pfn_valid(pfn)) {
820 page = pfn_to_page(pfn);
822 if (write && !pte_dirty(pte) &&!PageDirty(page))
823 set_page_dirty(page);
824 mark_page_accessed(page);
835 follow_page(struct mm_struct *mm, unsigned long address, int write)
837 return __follow_page(mm, address, 0, write, 1);
841 * check_user_page_readable() can be called frm niterrupt context by oprofile,
842 * so we need to avoid taking any non-irq-safe locks
844 int check_user_page_readable(struct mm_struct *mm, unsigned long address)
846 return __follow_page(mm, address, 1, 0, 0) != NULL;
848 EXPORT_SYMBOL(check_user_page_readable);
851 untouched_anonymous_page(struct mm_struct* mm, struct vm_area_struct *vma,
852 unsigned long address)
858 /* Check if the vma is for an anonymous mapping. */
859 if (vma->vm_ops && vma->vm_ops->nopage)
862 /* Check if page directory entry exists. */
863 pgd = pgd_offset(mm, address);
864 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
867 pud = pud_offset(pgd, address);
868 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
871 /* Check if page middle directory entry exists. */
872 pmd = pmd_offset(pud, address);
873 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
876 /* There is a pte slot for 'address' in 'mm'. */
880 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
881 unsigned long start, int len, int write, int force,
882 struct page **pages, struct vm_area_struct **vmas)
888 * Require read or write permissions.
889 * If 'force' is set, we only require the "MAY" flags.
891 flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
892 flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
896 struct vm_area_struct * vma;
898 vma = find_extend_vma(mm, start);
899 if (!vma && in_gate_area(tsk, start)) {
900 unsigned long pg = start & PAGE_MASK;
901 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
906 if (write) /* user gate pages are read-only */
907 return i ? : -EFAULT;
909 pgd = pgd_offset_k(pg);
911 pgd = pgd_offset_gate(mm, pg);
912 BUG_ON(pgd_none(*pgd));
913 pud = pud_offset(pgd, pg);
914 BUG_ON(pud_none(*pud));
915 pmd = pmd_offset(pud, pg);
917 return i ? : -EFAULT;
918 pte = pte_offset_map(pmd, pg);
919 if (pte_none(*pte)) {
921 return i ? : -EFAULT;
924 pages[i] = pte_page(*pte);
936 if (!vma || (vma->vm_flags & VM_IO)
937 || !(flags & vma->vm_flags))
938 return i ? : -EFAULT;
940 if (is_vm_hugetlb_page(vma)) {
941 i = follow_hugetlb_page(mm, vma, pages, vmas,
945 spin_lock(&mm->page_table_lock);
947 int write_access = write;
950 cond_resched_lock(&mm->page_table_lock);
951 while (!(page = follow_page(mm, start, write_access))) {
953 * Shortcut for anonymous pages. We don't want
954 * to force the creation of pages tables for
955 * insanely big anonymously mapped areas that
956 * nobody touched so far. This is important
957 * for doing a core dump for these mappings.
959 if (!write && untouched_anonymous_page(mm,vma,start)) {
960 page = ZERO_PAGE(start);
963 spin_unlock(&mm->page_table_lock);
964 switch (__handle_mm_fault(mm, vma, start,
968 * do_wp_page has broken COW when
969 * necessary, even if maybe_mkwrite
970 * decided not to set pte_write
980 case VM_FAULT_SIGBUS:
981 return i ? i : -EFAULT;
983 return i ? i : -ENOMEM;
987 spin_lock(&mm->page_table_lock);
991 flush_dcache_page(page);
992 if (!PageReserved(page))
993 page_cache_get(page);
1000 } while (len && start < vma->vm_end);
1001 spin_unlock(&mm->page_table_lock);
1005 EXPORT_SYMBOL(get_user_pages);
1007 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1008 unsigned long addr, unsigned long end, pgprot_t prot)
1012 pte = pte_alloc_map(mm, pmd, addr);
1016 pte_t zero_pte = pte_wrprotect(mk_pte(ZERO_PAGE(addr), prot));
1017 BUG_ON(!pte_none(*pte));
1018 set_pte_at(mm, addr, pte, zero_pte);
1019 } while (pte++, addr += PAGE_SIZE, addr != end);
1024 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1025 unsigned long addr, unsigned long end, pgprot_t prot)
1030 pmd = pmd_alloc(mm, pud, addr);
1034 next = pmd_addr_end(addr, end);
1035 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1037 } while (pmd++, addr = next, addr != end);
1041 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1042 unsigned long addr, unsigned long end, pgprot_t prot)
1047 pud = pud_alloc(mm, pgd, addr);
1051 next = pud_addr_end(addr, end);
1052 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1054 } while (pud++, addr = next, addr != end);
1058 int zeromap_page_range(struct vm_area_struct *vma,
1059 unsigned long addr, unsigned long size, pgprot_t prot)
1063 unsigned long end = addr + size;
1064 struct mm_struct *mm = vma->vm_mm;
1067 BUG_ON(addr >= end);
1068 pgd = pgd_offset(mm, addr);
1069 flush_cache_range(vma, addr, end);
1070 spin_lock(&mm->page_table_lock);
1072 next = pgd_addr_end(addr, end);
1073 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1076 } while (pgd++, addr = next, addr != end);
1077 spin_unlock(&mm->page_table_lock);
1082 * maps a range of physical memory into the requested pages. the old
1083 * mappings are removed. any references to nonexistent pages results
1084 * in null mappings (currently treated as "copy-on-access")
1086 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1087 unsigned long addr, unsigned long end,
1088 unsigned long pfn, pgprot_t prot)
1092 pte = pte_alloc_map(mm, pmd, addr);
1096 BUG_ON(!pte_none(*pte));
1097 if (!pfn_valid(pfn) || PageReserved(pfn_to_page(pfn)))
1098 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1100 } while (pte++, addr += PAGE_SIZE, addr != end);
1105 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1106 unsigned long addr, unsigned long end,
1107 unsigned long pfn, pgprot_t prot)
1112 pfn -= addr >> PAGE_SHIFT;
1113 pmd = pmd_alloc(mm, pud, addr);
1117 next = pmd_addr_end(addr, end);
1118 if (remap_pte_range(mm, pmd, addr, next,
1119 pfn + (addr >> PAGE_SHIFT), prot))
1121 } while (pmd++, addr = next, addr != end);
1125 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1126 unsigned long addr, unsigned long end,
1127 unsigned long pfn, pgprot_t prot)
1132 pfn -= addr >> PAGE_SHIFT;
1133 pud = pud_alloc(mm, pgd, addr);
1137 next = pud_addr_end(addr, end);
1138 if (remap_pmd_range(mm, pud, addr, next,
1139 pfn + (addr >> PAGE_SHIFT), prot))
1141 } while (pud++, addr = next, addr != end);
1145 /* Note: this is only safe if the mm semaphore is held when called. */
1146 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1147 unsigned long pfn, unsigned long size, pgprot_t prot)
1151 unsigned long end = addr + PAGE_ALIGN(size);
1152 struct mm_struct *mm = vma->vm_mm;
1156 * Physically remapped pages are special. Tell the
1157 * rest of the world about it:
1158 * VM_IO tells people not to look at these pages
1159 * (accesses can have side effects).
1160 * VM_RESERVED tells swapout not to try to touch
1163 vma->vm_flags |= VM_IO | VM_RESERVED;
1165 BUG_ON(addr >= end);
1166 pfn -= addr >> PAGE_SHIFT;
1167 pgd = pgd_offset(mm, addr);
1168 flush_cache_range(vma, addr, end);
1169 spin_lock(&mm->page_table_lock);
1171 next = pgd_addr_end(addr, end);
1172 err = remap_pud_range(mm, pgd, addr, next,
1173 pfn + (addr >> PAGE_SHIFT), prot);
1176 } while (pgd++, addr = next, addr != end);
1177 spin_unlock(&mm->page_table_lock);
1180 EXPORT_SYMBOL(remap_pfn_range);
1183 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1184 * servicing faults for write access. In the normal case, do always want
1185 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1186 * that do not have writing enabled, when used by access_process_vm.
1188 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1190 if (likely(vma->vm_flags & VM_WRITE))
1191 pte = pte_mkwrite(pte);
1196 * We hold the mm semaphore for reading and vma->vm_mm->page_table_lock
1198 static inline void break_cow(struct vm_area_struct * vma, struct page * new_page, unsigned long address,
1203 entry = maybe_mkwrite(pte_mkdirty(mk_pte(new_page, vma->vm_page_prot)),
1205 ptep_establish(vma, address, page_table, entry);
1206 update_mmu_cache(vma, address, entry);
1207 lazy_mmu_prot_update(entry);
1211 * This routine handles present pages, when users try to write
1212 * to a shared page. It is done by copying the page to a new address
1213 * and decrementing the shared-page counter for the old page.
1215 * Goto-purists beware: the only reason for goto's here is that it results
1216 * in better assembly code.. The "default" path will see no jumps at all.
1218 * Note that this routine assumes that the protection checks have been
1219 * done by the caller (the low-level page fault routine in most cases).
1220 * Thus we can safely just mark it writable once we've done any necessary
1223 * We also mark the page dirty at this point even though the page will
1224 * change only once the write actually happens. This avoids a few races,
1225 * and potentially makes it more efficient.
1227 * We hold the mm semaphore and the page_table_lock on entry and exit
1228 * with the page_table_lock released.
1230 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct * vma,
1231 unsigned long address, pte_t *page_table, pmd_t *pmd, pte_t pte)
1233 struct page *old_page, *new_page;
1234 unsigned long pfn = pte_pfn(pte);
1238 if (unlikely(!pfn_valid(pfn))) {
1240 * This should really halt the system so it can be debugged or
1241 * at least the kernel stops what it's doing before it corrupts
1242 * data, but for the moment just pretend this is OOM.
1244 pte_unmap(page_table);
1245 printk(KERN_ERR "do_wp_page: bogus page at address %08lx\n",
1247 spin_unlock(&mm->page_table_lock);
1248 return VM_FAULT_OOM;
1250 old_page = pfn_to_page(pfn);
1252 if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
1253 int reuse = can_share_swap_page(old_page);
1254 unlock_page(old_page);
1256 flush_cache_page(vma, address, pfn);
1257 entry = maybe_mkwrite(pte_mkyoung(pte_mkdirty(pte)),
1259 ptep_set_access_flags(vma, address, page_table, entry, 1);
1260 update_mmu_cache(vma, address, entry);
1261 lazy_mmu_prot_update(entry);
1262 pte_unmap(page_table);
1263 spin_unlock(&mm->page_table_lock);
1264 return VM_FAULT_MINOR|VM_FAULT_WRITE;
1267 pte_unmap(page_table);
1270 * Ok, we need to copy. Oh, well..
1272 if (!PageReserved(old_page))
1273 page_cache_get(old_page);
1274 spin_unlock(&mm->page_table_lock);
1276 if (unlikely(anon_vma_prepare(vma)))
1278 if (old_page == ZERO_PAGE(address)) {
1279 new_page = alloc_zeroed_user_highpage(vma, address);
1283 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1286 copy_user_highpage(new_page, old_page, address);
1289 * Re-check the pte - we dropped the lock
1291 ret = VM_FAULT_MINOR;
1292 spin_lock(&mm->page_table_lock);
1293 page_table = pte_offset_map(pmd, address);
1294 if (likely(pte_same(*page_table, pte))) {
1295 if (PageAnon(old_page))
1296 dec_mm_counter(mm, anon_rss);
1297 if (PageReserved(old_page))
1298 inc_mm_counter(mm, rss);
1300 page_remove_rmap(old_page);
1301 flush_cache_page(vma, address, pfn);
1302 break_cow(vma, new_page, address, page_table);
1303 lru_cache_add_active(new_page);
1304 page_add_anon_rmap(new_page, vma, address);
1306 /* Free the old page.. */
1307 new_page = old_page;
1308 ret |= VM_FAULT_WRITE;
1310 pte_unmap(page_table);
1311 page_cache_release(new_page);
1312 page_cache_release(old_page);
1313 spin_unlock(&mm->page_table_lock);
1317 page_cache_release(old_page);
1318 return VM_FAULT_OOM;
1322 * Helper functions for unmap_mapping_range().
1324 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1326 * We have to restart searching the prio_tree whenever we drop the lock,
1327 * since the iterator is only valid while the lock is held, and anyway
1328 * a later vma might be split and reinserted earlier while lock dropped.
1330 * The list of nonlinear vmas could be handled more efficiently, using
1331 * a placeholder, but handle it in the same way until a need is shown.
1332 * It is important to search the prio_tree before nonlinear list: a vma
1333 * may become nonlinear and be shifted from prio_tree to nonlinear list
1334 * while the lock is dropped; but never shifted from list to prio_tree.
1336 * In order to make forward progress despite restarting the search,
1337 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1338 * quickly skip it next time around. Since the prio_tree search only
1339 * shows us those vmas affected by unmapping the range in question, we
1340 * can't efficiently keep all vmas in step with mapping->truncate_count:
1341 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1342 * mapping->truncate_count and vma->vm_truncate_count are protected by
1345 * In order to make forward progress despite repeatedly restarting some
1346 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1347 * and restart from that address when we reach that vma again. It might
1348 * have been split or merged, shrunk or extended, but never shifted: so
1349 * restart_addr remains valid so long as it remains in the vma's range.
1350 * unmap_mapping_range forces truncate_count to leap over page-aligned
1351 * values so we can save vma's restart_addr in its truncate_count field.
1353 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1355 static void reset_vma_truncate_counts(struct address_space *mapping)
1357 struct vm_area_struct *vma;
1358 struct prio_tree_iter iter;
1360 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1361 vma->vm_truncate_count = 0;
1362 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1363 vma->vm_truncate_count = 0;
1366 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1367 unsigned long start_addr, unsigned long end_addr,
1368 struct zap_details *details)
1370 unsigned long restart_addr;
1374 restart_addr = vma->vm_truncate_count;
1375 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1376 start_addr = restart_addr;
1377 if (start_addr >= end_addr) {
1378 /* Top of vma has been split off since last time */
1379 vma->vm_truncate_count = details->truncate_count;
1384 restart_addr = zap_page_range(vma, start_addr,
1385 end_addr - start_addr, details);
1388 * We cannot rely on the break test in unmap_vmas:
1389 * on the one hand, we don't want to restart our loop
1390 * just because that broke out for the page_table_lock;
1391 * on the other hand, it does no test when vma is small.
1393 need_break = need_resched() ||
1394 need_lockbreak(details->i_mmap_lock);
1396 if (restart_addr >= end_addr) {
1397 /* We have now completed this vma: mark it so */
1398 vma->vm_truncate_count = details->truncate_count;
1402 /* Note restart_addr in vma's truncate_count field */
1403 vma->vm_truncate_count = restart_addr;
1408 spin_unlock(details->i_mmap_lock);
1410 spin_lock(details->i_mmap_lock);
1414 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1415 struct zap_details *details)
1417 struct vm_area_struct *vma;
1418 struct prio_tree_iter iter;
1419 pgoff_t vba, vea, zba, zea;
1422 vma_prio_tree_foreach(vma, &iter, root,
1423 details->first_index, details->last_index) {
1424 /* Skip quickly over those we have already dealt with */
1425 if (vma->vm_truncate_count == details->truncate_count)
1428 vba = vma->vm_pgoff;
1429 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1430 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1431 zba = details->first_index;
1434 zea = details->last_index;
1438 if (unmap_mapping_range_vma(vma,
1439 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1440 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1446 static inline void unmap_mapping_range_list(struct list_head *head,
1447 struct zap_details *details)
1449 struct vm_area_struct *vma;
1452 * In nonlinear VMAs there is no correspondence between virtual address
1453 * offset and file offset. So we must perform an exhaustive search
1454 * across *all* the pages in each nonlinear VMA, not just the pages
1455 * whose virtual address lies outside the file truncation point.
1458 list_for_each_entry(vma, head, shared.vm_set.list) {
1459 /* Skip quickly over those we have already dealt with */
1460 if (vma->vm_truncate_count == details->truncate_count)
1462 details->nonlinear_vma = vma;
1463 if (unmap_mapping_range_vma(vma, vma->vm_start,
1464 vma->vm_end, details) < 0)
1470 * unmap_mapping_range - unmap the portion of all mmaps
1471 * in the specified address_space corresponding to the specified
1472 * page range in the underlying file.
1473 * @mapping: the address space containing mmaps to be unmapped.
1474 * @holebegin: byte in first page to unmap, relative to the start of
1475 * the underlying file. This will be rounded down to a PAGE_SIZE
1476 * boundary. Note that this is different from vmtruncate(), which
1477 * must keep the partial page. In contrast, we must get rid of
1479 * @holelen: size of prospective hole in bytes. This will be rounded
1480 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1482 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1483 * but 0 when invalidating pagecache, don't throw away private data.
1485 void unmap_mapping_range(struct address_space *mapping,
1486 loff_t const holebegin, loff_t const holelen, int even_cows)
1488 struct zap_details details;
1489 pgoff_t hba = holebegin >> PAGE_SHIFT;
1490 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1492 /* Check for overflow. */
1493 if (sizeof(holelen) > sizeof(hlen)) {
1495 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1496 if (holeend & ~(long long)ULONG_MAX)
1497 hlen = ULONG_MAX - hba + 1;
1500 details.check_mapping = even_cows? NULL: mapping;
1501 details.nonlinear_vma = NULL;
1502 details.first_index = hba;
1503 details.last_index = hba + hlen - 1;
1504 if (details.last_index < details.first_index)
1505 details.last_index = ULONG_MAX;
1506 details.i_mmap_lock = &mapping->i_mmap_lock;
1508 spin_lock(&mapping->i_mmap_lock);
1510 /* serialize i_size write against truncate_count write */
1512 /* Protect against page faults, and endless unmapping loops */
1513 mapping->truncate_count++;
1515 * For archs where spin_lock has inclusive semantics like ia64
1516 * this smp_mb() will prevent to read pagetable contents
1517 * before the truncate_count increment is visible to
1521 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1522 if (mapping->truncate_count == 0)
1523 reset_vma_truncate_counts(mapping);
1524 mapping->truncate_count++;
1526 details.truncate_count = mapping->truncate_count;
1528 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1529 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1530 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1531 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1532 spin_unlock(&mapping->i_mmap_lock);
1534 EXPORT_SYMBOL(unmap_mapping_range);
1537 * Handle all mappings that got truncated by a "truncate()"
1540 * NOTE! We have to be ready to update the memory sharing
1541 * between the file and the memory map for a potential last
1542 * incomplete page. Ugly, but necessary.
1544 int vmtruncate(struct inode * inode, loff_t offset)
1546 struct address_space *mapping = inode->i_mapping;
1547 unsigned long limit;
1549 if (inode->i_size < offset)
1552 * truncation of in-use swapfiles is disallowed - it would cause
1553 * subsequent swapout to scribble on the now-freed blocks.
1555 if (IS_SWAPFILE(inode))
1557 i_size_write(inode, offset);
1558 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1559 truncate_inode_pages(mapping, offset);
1563 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1564 if (limit != RLIM_INFINITY && offset > limit)
1566 if (offset > inode->i_sb->s_maxbytes)
1568 i_size_write(inode, offset);
1571 if (inode->i_op && inode->i_op->truncate)
1572 inode->i_op->truncate(inode);
1575 send_sig(SIGXFSZ, current, 0);
1582 EXPORT_SYMBOL(vmtruncate);
1585 * Primitive swap readahead code. We simply read an aligned block of
1586 * (1 << page_cluster) entries in the swap area. This method is chosen
1587 * because it doesn't cost us any seek time. We also make sure to queue
1588 * the 'original' request together with the readahead ones...
1590 * This has been extended to use the NUMA policies from the mm triggering
1593 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1595 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1598 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1601 struct page *new_page;
1602 unsigned long offset;
1605 * Get the number of handles we should do readahead io to.
1607 num = valid_swaphandles(entry, &offset);
1608 for (i = 0; i < num; offset++, i++) {
1609 /* Ok, do the async read-ahead now */
1610 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1611 offset), vma, addr);
1614 page_cache_release(new_page);
1617 * Find the next applicable VMA for the NUMA policy.
1623 if (addr >= vma->vm_end) {
1625 next_vma = vma ? vma->vm_next : NULL;
1627 if (vma && addr < vma->vm_start)
1630 if (next_vma && addr >= next_vma->vm_start) {
1632 next_vma = vma->vm_next;
1637 lru_add_drain(); /* Push any new pages onto the LRU now */
1641 * We hold the mm semaphore and the page_table_lock on entry and
1642 * should release the pagetable lock on exit..
1644 static int do_swap_page(struct mm_struct * mm,
1645 struct vm_area_struct * vma, unsigned long address,
1646 pte_t *page_table, pmd_t *pmd, pte_t orig_pte, int write_access)
1649 swp_entry_t entry = pte_to_swp_entry(orig_pte);
1651 int ret = VM_FAULT_MINOR;
1653 pte_unmap(page_table);
1654 spin_unlock(&mm->page_table_lock);
1655 page = lookup_swap_cache(entry);
1657 swapin_readahead(entry, address, vma);
1658 page = read_swap_cache_async(entry, vma, address);
1661 * Back out if somebody else faulted in this pte while
1662 * we released the page table lock.
1664 spin_lock(&mm->page_table_lock);
1665 page_table = pte_offset_map(pmd, address);
1666 if (likely(pte_same(*page_table, orig_pte)))
1669 ret = VM_FAULT_MINOR;
1670 pte_unmap(page_table);
1671 spin_unlock(&mm->page_table_lock);
1675 /* Had to read the page from swap area: Major fault */
1676 ret = VM_FAULT_MAJOR;
1677 inc_page_state(pgmajfault);
1681 mark_page_accessed(page);
1685 * Back out if somebody else faulted in this pte while we
1686 * released the page table lock.
1688 spin_lock(&mm->page_table_lock);
1689 page_table = pte_offset_map(pmd, address);
1690 if (unlikely(!pte_same(*page_table, orig_pte))) {
1691 ret = VM_FAULT_MINOR;
1695 if (unlikely(!PageUptodate(page))) {
1696 ret = VM_FAULT_SIGBUS;
1700 /* The page isn't present yet, go ahead with the fault. */
1702 inc_mm_counter(mm, rss);
1703 pte = mk_pte(page, vma->vm_page_prot);
1704 if (write_access && can_share_swap_page(page)) {
1705 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
1709 flush_icache_page(vma, page);
1710 set_pte_at(mm, address, page_table, pte);
1711 page_add_anon_rmap(page, vma, address);
1715 remove_exclusive_swap_page(page);
1719 if (do_wp_page(mm, vma, address,
1720 page_table, pmd, pte) == VM_FAULT_OOM)
1725 /* No need to invalidate - it was non-present before */
1726 update_mmu_cache(vma, address, pte);
1727 lazy_mmu_prot_update(pte);
1728 pte_unmap(page_table);
1729 spin_unlock(&mm->page_table_lock);
1733 pte_unmap(page_table);
1734 spin_unlock(&mm->page_table_lock);
1736 page_cache_release(page);
1741 * We are called with the MM semaphore and page_table_lock
1742 * spinlock held to protect against concurrent faults in
1743 * multithreaded programs.
1746 do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
1747 pte_t *page_table, pmd_t *pmd, int write_access,
1751 struct page * page = ZERO_PAGE(addr);
1753 /* Read-only mapping of ZERO_PAGE. */
1754 entry = pte_wrprotect(mk_pte(ZERO_PAGE(addr), vma->vm_page_prot));
1756 /* ..except if it's a write access */
1758 /* Allocate our own private page. */
1759 pte_unmap(page_table);
1760 spin_unlock(&mm->page_table_lock);
1762 if (unlikely(anon_vma_prepare(vma)))
1764 page = alloc_zeroed_user_highpage(vma, addr);
1768 spin_lock(&mm->page_table_lock);
1769 page_table = pte_offset_map(pmd, addr);
1771 if (!pte_none(*page_table)) {
1772 pte_unmap(page_table);
1773 page_cache_release(page);
1774 spin_unlock(&mm->page_table_lock);
1777 inc_mm_counter(mm, rss);
1778 entry = maybe_mkwrite(pte_mkdirty(mk_pte(page,
1779 vma->vm_page_prot)),
1781 lru_cache_add_active(page);
1782 SetPageReferenced(page);
1783 page_add_anon_rmap(page, vma, addr);
1786 set_pte_at(mm, addr, page_table, entry);
1787 pte_unmap(page_table);
1789 /* No need to invalidate - it was non-present before */
1790 update_mmu_cache(vma, addr, entry);
1791 lazy_mmu_prot_update(entry);
1792 spin_unlock(&mm->page_table_lock);
1794 return VM_FAULT_MINOR;
1796 return VM_FAULT_OOM;
1800 * do_no_page() tries to create a new page mapping. It aggressively
1801 * tries to share with existing pages, but makes a separate copy if
1802 * the "write_access" parameter is true in order to avoid the next
1805 * As this is called only for pages that do not currently exist, we
1806 * do not need to flush old virtual caches or the TLB.
1808 * This is called with the MM semaphore held and the page table
1809 * spinlock held. Exit with the spinlock released.
1812 do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
1813 unsigned long address, int write_access, pte_t *page_table, pmd_t *pmd)
1815 struct page * new_page;
1816 struct address_space *mapping = NULL;
1818 unsigned int sequence = 0;
1819 int ret = VM_FAULT_MINOR;
1822 if (!vma->vm_ops || !vma->vm_ops->nopage)
1823 return do_anonymous_page(mm, vma, page_table,
1824 pmd, write_access, address);
1825 pte_unmap(page_table);
1826 spin_unlock(&mm->page_table_lock);
1829 mapping = vma->vm_file->f_mapping;
1830 sequence = mapping->truncate_count;
1831 smp_rmb(); /* serializes i_size against truncate_count */
1835 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
1837 * No smp_rmb is needed here as long as there's a full
1838 * spin_lock/unlock sequence inside the ->nopage callback
1839 * (for the pagecache lookup) that acts as an implicit
1840 * smp_mb() and prevents the i_size read to happen
1841 * after the next truncate_count read.
1844 /* no page was available -- either SIGBUS or OOM */
1845 if (new_page == NOPAGE_SIGBUS)
1846 return VM_FAULT_SIGBUS;
1847 if (new_page == NOPAGE_OOM)
1848 return VM_FAULT_OOM;
1851 * Should we do an early C-O-W break?
1853 if (write_access && !(vma->vm_flags & VM_SHARED)) {
1856 if (unlikely(anon_vma_prepare(vma)))
1858 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1861 copy_user_highpage(page, new_page, address);
1862 page_cache_release(new_page);
1867 spin_lock(&mm->page_table_lock);
1869 * For a file-backed vma, someone could have truncated or otherwise
1870 * invalidated this page. If unmap_mapping_range got called,
1871 * retry getting the page.
1873 if (mapping && unlikely(sequence != mapping->truncate_count)) {
1874 sequence = mapping->truncate_count;
1875 spin_unlock(&mm->page_table_lock);
1876 page_cache_release(new_page);
1879 page_table = pte_offset_map(pmd, address);
1882 * This silly early PAGE_DIRTY setting removes a race
1883 * due to the bad i386 page protection. But it's valid
1884 * for other architectures too.
1886 * Note that if write_access is true, we either now have
1887 * an exclusive copy of the page, or this is a shared mapping,
1888 * so we can make it writable and dirty to avoid having to
1889 * handle that later.
1891 /* Only go through if we didn't race with anybody else... */
1892 if (pte_none(*page_table)) {
1893 if (!PageReserved(new_page))
1894 inc_mm_counter(mm, rss);
1896 flush_icache_page(vma, new_page);
1897 entry = mk_pte(new_page, vma->vm_page_prot);
1899 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1900 set_pte_at(mm, address, page_table, entry);
1902 lru_cache_add_active(new_page);
1903 page_add_anon_rmap(new_page, vma, address);
1905 page_add_file_rmap(new_page);
1906 pte_unmap(page_table);
1908 /* One of our sibling threads was faster, back out. */
1909 pte_unmap(page_table);
1910 page_cache_release(new_page);
1911 spin_unlock(&mm->page_table_lock);
1915 /* no need to invalidate: a not-present page shouldn't be cached */
1916 update_mmu_cache(vma, address, entry);
1917 lazy_mmu_prot_update(entry);
1918 spin_unlock(&mm->page_table_lock);
1922 page_cache_release(new_page);
1928 * Fault of a previously existing named mapping. Repopulate the pte
1929 * from the encoded file_pte if possible. This enables swappable
1932 static int do_file_page(struct mm_struct * mm, struct vm_area_struct * vma,
1933 unsigned long address, int write_access, pte_t *pte, pmd_t *pmd)
1935 unsigned long pgoff;
1938 BUG_ON(!vma->vm_ops || !vma->vm_ops->nopage);
1940 * Fall back to the linear mapping if the fs does not support
1943 if (!vma->vm_ops || !vma->vm_ops->populate ||
1944 (write_access && !(vma->vm_flags & VM_SHARED))) {
1945 pte_clear(mm, address, pte);
1946 return do_no_page(mm, vma, address, write_access, pte, pmd);
1949 pgoff = pte_to_pgoff(*pte);
1952 spin_unlock(&mm->page_table_lock);
1954 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE, vma->vm_page_prot, pgoff, 0);
1956 return VM_FAULT_OOM;
1958 return VM_FAULT_SIGBUS;
1959 return VM_FAULT_MAJOR;
1963 * These routines also need to handle stuff like marking pages dirty
1964 * and/or accessed for architectures that don't do it in hardware (most
1965 * RISC architectures). The early dirtying is also good on the i386.
1967 * There is also a hook called "update_mmu_cache()" that architectures
1968 * with external mmu caches can use to update those (ie the Sparc or
1969 * PowerPC hashed page tables that act as extended TLBs).
1971 * Note the "page_table_lock". It is to protect against kswapd removing
1972 * pages from under us. Note that kswapd only ever _removes_ pages, never
1973 * adds them. As such, once we have noticed that the page is not present,
1974 * we can drop the lock early.
1976 * The adding of pages is protected by the MM semaphore (which we hold),
1977 * so we don't need to worry about a page being suddenly been added into
1980 * We enter with the pagetable spinlock held, we are supposed to
1981 * release it when done.
1983 static inline int handle_pte_fault(struct mm_struct *mm,
1984 struct vm_area_struct * vma, unsigned long address,
1985 int write_access, pte_t *pte, pmd_t *pmd)
1990 if (!pte_present(entry)) {
1992 * If it truly wasn't present, we know that kswapd
1993 * and the PTE updates will not touch it later. So
1996 if (pte_none(entry))
1997 return do_no_page(mm, vma, address, write_access, pte, pmd);
1998 if (pte_file(entry))
1999 return do_file_page(mm, vma, address, write_access, pte, pmd);
2000 return do_swap_page(mm, vma, address, pte, pmd, entry, write_access);
2004 if (!pte_write(entry))
2005 return do_wp_page(mm, vma, address, pte, pmd, entry);
2006 entry = pte_mkdirty(entry);
2008 entry = pte_mkyoung(entry);
2009 ptep_set_access_flags(vma, address, pte, entry, write_access);
2010 update_mmu_cache(vma, address, entry);
2011 lazy_mmu_prot_update(entry);
2013 spin_unlock(&mm->page_table_lock);
2014 return VM_FAULT_MINOR;
2018 * By the time we get here, we already hold the mm semaphore
2020 int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct * vma,
2021 unsigned long address, int write_access)
2028 __set_current_state(TASK_RUNNING);
2030 inc_page_state(pgfault);
2032 if (is_vm_hugetlb_page(vma))
2033 return VM_FAULT_SIGBUS; /* mapping truncation does this. */
2036 * We need the page table lock to synchronize with kswapd
2037 * and the SMP-safe atomic PTE updates.
2039 pgd = pgd_offset(mm, address);
2040 spin_lock(&mm->page_table_lock);
2042 pud = pud_alloc(mm, pgd, address);
2046 pmd = pmd_alloc(mm, pud, address);
2050 pte = pte_alloc_map(mm, pmd, address);
2054 return handle_pte_fault(mm, vma, address, write_access, pte, pmd);
2057 spin_unlock(&mm->page_table_lock);
2058 return VM_FAULT_OOM;
2061 #ifndef __PAGETABLE_PUD_FOLDED
2063 * Allocate page upper directory.
2065 * We've already handled the fast-path in-line, and we own the
2068 pud_t fastcall *__pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2072 spin_unlock(&mm->page_table_lock);
2073 new = pud_alloc_one(mm, address);
2074 spin_lock(&mm->page_table_lock);
2079 * Because we dropped the lock, we should re-check the
2080 * entry, as somebody else could have populated it..
2082 if (pgd_present(*pgd)) {
2086 pgd_populate(mm, pgd, new);
2088 return pud_offset(pgd, address);
2090 #endif /* __PAGETABLE_PUD_FOLDED */
2092 #ifndef __PAGETABLE_PMD_FOLDED
2094 * Allocate page middle directory.
2096 * We've already handled the fast-path in-line, and we own the
2099 pmd_t fastcall *__pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2103 spin_unlock(&mm->page_table_lock);
2104 new = pmd_alloc_one(mm, address);
2105 spin_lock(&mm->page_table_lock);
2110 * Because we dropped the lock, we should re-check the
2111 * entry, as somebody else could have populated it..
2113 #ifndef __ARCH_HAS_4LEVEL_HACK
2114 if (pud_present(*pud)) {
2118 pud_populate(mm, pud, new);
2120 if (pgd_present(*pud)) {
2124 pgd_populate(mm, pud, new);
2125 #endif /* __ARCH_HAS_4LEVEL_HACK */
2128 return pmd_offset(pud, address);
2130 #endif /* __PAGETABLE_PMD_FOLDED */
2132 int make_pages_present(unsigned long addr, unsigned long end)
2134 int ret, len, write;
2135 struct vm_area_struct * vma;
2137 vma = find_vma(current->mm, addr);
2140 write = (vma->vm_flags & VM_WRITE) != 0;
2143 if (end > vma->vm_end)
2145 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2146 ret = get_user_pages(current, current->mm, addr,
2147 len, write, 0, NULL, NULL);
2150 return ret == len ? 0 : -1;
2154 * Map a vmalloc()-space virtual address to the physical page.
2156 struct page * vmalloc_to_page(void * vmalloc_addr)
2158 unsigned long addr = (unsigned long) vmalloc_addr;
2159 struct page *page = NULL;
2160 pgd_t *pgd = pgd_offset_k(addr);
2165 if (!pgd_none(*pgd)) {
2166 pud = pud_offset(pgd, addr);
2167 if (!pud_none(*pud)) {
2168 pmd = pmd_offset(pud, addr);
2169 if (!pmd_none(*pmd)) {
2170 ptep = pte_offset_map(pmd, addr);
2172 if (pte_present(pte))
2173 page = pte_page(pte);
2181 EXPORT_SYMBOL(vmalloc_to_page);
2184 * Map a vmalloc()-space virtual address to the physical page frame number.
2186 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2188 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2191 EXPORT_SYMBOL(vmalloc_to_pfn);
2194 * update_mem_hiwater
2195 * - update per process rss and vm high water data
2197 void update_mem_hiwater(struct task_struct *tsk)
2200 unsigned long rss = get_mm_counter(tsk->mm, rss);
2202 if (tsk->mm->hiwater_rss < rss)
2203 tsk->mm->hiwater_rss = rss;
2204 if (tsk->mm->hiwater_vm < tsk->mm->total_vm)
2205 tsk->mm->hiwater_vm = tsk->mm->total_vm;
2209 #if !defined(__HAVE_ARCH_GATE_AREA)
2211 #if defined(AT_SYSINFO_EHDR)
2212 struct vm_area_struct gate_vma;
2214 static int __init gate_vma_init(void)
2216 gate_vma.vm_mm = NULL;
2217 gate_vma.vm_start = FIXADDR_USER_START;
2218 gate_vma.vm_end = FIXADDR_USER_END;
2219 gate_vma.vm_page_prot = PAGE_READONLY;
2220 gate_vma.vm_flags = 0;
2223 __initcall(gate_vma_init);
2226 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2228 #ifdef AT_SYSINFO_EHDR
2235 int in_gate_area_no_task(unsigned long addr)
2237 #ifdef AT_SYSINFO_EHDR
2238 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2244 #endif /* __HAVE_ARCH_GATE_AREA */