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_DISCONTIGMEM
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 static inline 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 /* Optimization: gather nearby vmas into a single call down */
264 while (next && next->vm_start <= vma->vm_end + PMD_SIZE) {
268 free_pgd_range(*tlb, addr, vma->vm_end,
269 floor, next? next->vm_start: ceiling);
274 pte_t fastcall * pte_alloc_map(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
276 if (!pmd_present(*pmd)) {
279 spin_unlock(&mm->page_table_lock);
280 new = pte_alloc_one(mm, address);
281 spin_lock(&mm->page_table_lock);
285 * Because we dropped the lock, we should re-check the
286 * entry, as somebody else could have populated it..
288 if (pmd_present(*pmd)) {
293 inc_page_state(nr_page_table_pages);
294 pmd_populate(mm, pmd, new);
297 return pte_offset_map(pmd, address);
300 pte_t fastcall * pte_alloc_kernel(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
302 if (!pmd_present(*pmd)) {
305 spin_unlock(&mm->page_table_lock);
306 new = pte_alloc_one_kernel(mm, address);
307 spin_lock(&mm->page_table_lock);
312 * Because we dropped the lock, we should re-check the
313 * entry, as somebody else could have populated it..
315 if (pmd_present(*pmd)) {
316 pte_free_kernel(new);
319 pmd_populate_kernel(mm, pmd, new);
322 return pte_offset_kernel(pmd, address);
326 * copy one vm_area from one task to the other. Assumes the page tables
327 * already present in the new task to be cleared in the whole range
328 * covered by this vma.
330 * dst->page_table_lock is held on entry and exit,
331 * but may be dropped within p[mg]d_alloc() and pte_alloc_map().
335 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
336 pte_t *dst_pte, pte_t *src_pte, unsigned long vm_flags,
339 pte_t pte = *src_pte;
343 /* pte contains position in swap or file, so copy. */
344 if (unlikely(!pte_present(pte))) {
345 if (!pte_file(pte)) {
346 swap_duplicate(pte_to_swp_entry(pte));
347 /* make sure dst_mm is on swapoff's mmlist. */
348 if (unlikely(list_empty(&dst_mm->mmlist))) {
349 spin_lock(&mmlist_lock);
350 list_add(&dst_mm->mmlist, &src_mm->mmlist);
351 spin_unlock(&mmlist_lock);
354 set_pte_at(dst_mm, addr, dst_pte, pte);
359 /* the pte points outside of valid memory, the
360 * mapping is assumed to be good, meaningful
361 * and not mapped via rmap - duplicate the
366 page = pfn_to_page(pfn);
368 if (!page || PageReserved(page)) {
369 set_pte_at(dst_mm, addr, dst_pte, pte);
374 * If it's a COW mapping, write protect it both
375 * in the parent and the child
377 if ((vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE) {
378 ptep_set_wrprotect(src_mm, addr, src_pte);
383 * If it's a shared mapping, mark it clean in
386 if (vm_flags & VM_SHARED)
387 pte = pte_mkclean(pte);
388 pte = pte_mkold(pte);
390 inc_mm_counter(dst_mm, rss);
392 inc_mm_counter(dst_mm, anon_rss);
393 set_pte_at(dst_mm, addr, dst_pte, pte);
397 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
398 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
399 unsigned long addr, unsigned long end)
401 pte_t *src_pte, *dst_pte;
402 unsigned long vm_flags = vma->vm_flags;
406 dst_pte = pte_alloc_map(dst_mm, dst_pmd, addr);
409 src_pte = pte_offset_map_nested(src_pmd, addr);
412 spin_lock(&src_mm->page_table_lock);
415 * We are holding two locks at this point - either of them
416 * could generate latencies in another task on another CPU.
418 if (progress >= 32 && (need_resched() ||
419 need_lockbreak(&src_mm->page_table_lock) ||
420 need_lockbreak(&dst_mm->page_table_lock)))
422 if (pte_none(*src_pte)) {
426 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vm_flags, addr);
428 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
429 spin_unlock(&src_mm->page_table_lock);
431 pte_unmap_nested(src_pte - 1);
432 pte_unmap(dst_pte - 1);
433 cond_resched_lock(&dst_mm->page_table_lock);
439 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
440 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
441 unsigned long addr, unsigned long end)
443 pmd_t *src_pmd, *dst_pmd;
446 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
449 src_pmd = pmd_offset(src_pud, addr);
451 next = pmd_addr_end(addr, end);
452 if (pmd_none_or_clear_bad(src_pmd))
454 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
457 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
461 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
462 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
463 unsigned long addr, unsigned long end)
465 pud_t *src_pud, *dst_pud;
468 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
471 src_pud = pud_offset(src_pgd, addr);
473 next = pud_addr_end(addr, end);
474 if (pud_none_or_clear_bad(src_pud))
476 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
479 } while (dst_pud++, src_pud++, addr = next, addr != end);
483 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
484 struct vm_area_struct *vma)
486 pgd_t *src_pgd, *dst_pgd;
488 unsigned long addr = vma->vm_start;
489 unsigned long end = vma->vm_end;
491 if (is_vm_hugetlb_page(vma))
492 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
494 dst_pgd = pgd_offset(dst_mm, addr);
495 src_pgd = pgd_offset(src_mm, addr);
497 next = pgd_addr_end(addr, end);
498 if (pgd_none_or_clear_bad(src_pgd))
500 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
503 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
507 static void zap_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
508 unsigned long addr, unsigned long end,
509 struct zap_details *details)
513 pte = pte_offset_map(pmd, addr);
518 if (pte_present(ptent)) {
519 struct page *page = NULL;
520 unsigned long pfn = pte_pfn(ptent);
521 if (pfn_valid(pfn)) {
522 page = pfn_to_page(pfn);
523 if (PageReserved(page))
526 if (unlikely(details) && page) {
528 * unmap_shared_mapping_pages() wants to
529 * invalidate cache without truncating:
530 * unmap shared but keep private pages.
532 if (details->check_mapping &&
533 details->check_mapping != page->mapping)
536 * Each page->index must be checked when
537 * invalidating or truncating nonlinear.
539 if (details->nonlinear_vma &&
540 (page->index < details->first_index ||
541 page->index > details->last_index))
544 ptent = ptep_get_and_clear(tlb->mm, addr, pte);
545 tlb_remove_tlb_entry(tlb, pte, addr);
548 if (unlikely(details) && details->nonlinear_vma
549 && linear_page_index(details->nonlinear_vma,
550 addr) != page->index)
551 set_pte_at(tlb->mm, addr, pte,
552 pgoff_to_pte(page->index));
553 if (pte_dirty(ptent))
554 set_page_dirty(page);
556 dec_mm_counter(tlb->mm, anon_rss);
557 else if (pte_young(ptent))
558 mark_page_accessed(page);
560 page_remove_rmap(page);
561 tlb_remove_page(tlb, page);
565 * If details->check_mapping, we leave swap entries;
566 * if details->nonlinear_vma, we leave file entries.
568 if (unlikely(details))
570 if (!pte_file(ptent))
571 free_swap_and_cache(pte_to_swp_entry(ptent));
572 pte_clear(tlb->mm, addr, pte);
573 } while (pte++, addr += PAGE_SIZE, addr != end);
577 static inline void zap_pmd_range(struct mmu_gather *tlb, pud_t *pud,
578 unsigned long addr, unsigned long end,
579 struct zap_details *details)
584 pmd = pmd_offset(pud, addr);
586 next = pmd_addr_end(addr, end);
587 if (pmd_none_or_clear_bad(pmd))
589 zap_pte_range(tlb, pmd, addr, next, details);
590 } while (pmd++, addr = next, addr != end);
593 static inline void zap_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
594 unsigned long addr, unsigned long end,
595 struct zap_details *details)
600 pud = pud_offset(pgd, addr);
602 next = pud_addr_end(addr, end);
603 if (pud_none_or_clear_bad(pud))
605 zap_pmd_range(tlb, pud, addr, next, details);
606 } while (pud++, addr = next, addr != end);
609 static void unmap_page_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
610 unsigned long addr, unsigned long end,
611 struct zap_details *details)
616 if (details && !details->check_mapping && !details->nonlinear_vma)
620 tlb_start_vma(tlb, vma);
621 pgd = pgd_offset(vma->vm_mm, addr);
623 next = pgd_addr_end(addr, end);
624 if (pgd_none_or_clear_bad(pgd))
626 zap_pud_range(tlb, pgd, addr, next, details);
627 } while (pgd++, addr = next, addr != end);
628 tlb_end_vma(tlb, vma);
631 #ifdef CONFIG_PREEMPT
632 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
634 /* No preempt: go for improved straight-line efficiency */
635 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
639 * unmap_vmas - unmap a range of memory covered by a list of vma's
640 * @tlbp: address of the caller's struct mmu_gather
641 * @mm: the controlling mm_struct
642 * @vma: the starting vma
643 * @start_addr: virtual address at which to start unmapping
644 * @end_addr: virtual address at which to end unmapping
645 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
646 * @details: details of nonlinear truncation or shared cache invalidation
648 * Returns the number of vma's which were covered by the unmapping.
650 * Unmap all pages in the vma list. Called under page_table_lock.
652 * We aim to not hold page_table_lock for too long (for scheduling latency
653 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
654 * return the ending mmu_gather to the caller.
656 * Only addresses between `start' and `end' will be unmapped.
658 * The VMA list must be sorted in ascending virtual address order.
660 * unmap_vmas() assumes that the caller will flush the whole unmapped address
661 * range after unmap_vmas() returns. So the only responsibility here is to
662 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
663 * drops the lock and schedules.
665 int unmap_vmas(struct mmu_gather **tlbp, struct mm_struct *mm,
666 struct vm_area_struct *vma, unsigned long start_addr,
667 unsigned long end_addr, unsigned long *nr_accounted,
668 struct zap_details *details)
670 unsigned long zap_bytes = ZAP_BLOCK_SIZE;
671 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
672 int tlb_start_valid = 0;
674 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
675 int fullmm = tlb_is_full_mm(*tlbp);
677 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
681 start = max(vma->vm_start, start_addr);
682 if (start >= vma->vm_end)
684 end = min(vma->vm_end, end_addr);
685 if (end <= vma->vm_start)
688 if (vma->vm_flags & VM_ACCOUNT)
689 *nr_accounted += (end - start) >> PAGE_SHIFT;
692 while (start != end) {
695 if (!tlb_start_valid) {
700 if (is_vm_hugetlb_page(vma)) {
702 unmap_hugepage_range(vma, start, end);
704 block = min(zap_bytes, end - start);
705 unmap_page_range(*tlbp, vma, start,
706 start + block, details);
711 if ((long)zap_bytes > 0)
714 tlb_finish_mmu(*tlbp, tlb_start, start);
716 if (need_resched() ||
717 need_lockbreak(&mm->page_table_lock) ||
718 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
720 /* must reset count of rss freed */
721 *tlbp = tlb_gather_mmu(mm, fullmm);
722 details->break_addr = start;
725 spin_unlock(&mm->page_table_lock);
727 spin_lock(&mm->page_table_lock);
730 *tlbp = tlb_gather_mmu(mm, fullmm);
732 zap_bytes = ZAP_BLOCK_SIZE;
740 * zap_page_range - remove user pages in a given range
741 * @vma: vm_area_struct holding the applicable pages
742 * @address: starting address of pages to zap
743 * @size: number of bytes to zap
744 * @details: details of nonlinear truncation or shared cache invalidation
746 void zap_page_range(struct vm_area_struct *vma, unsigned long address,
747 unsigned long size, struct zap_details *details)
749 struct mm_struct *mm = vma->vm_mm;
750 struct mmu_gather *tlb;
751 unsigned long end = address + size;
752 unsigned long nr_accounted = 0;
754 if (is_vm_hugetlb_page(vma)) {
755 zap_hugepage_range(vma, address, size);
760 spin_lock(&mm->page_table_lock);
761 tlb = tlb_gather_mmu(mm, 0);
762 unmap_vmas(&tlb, mm, vma, address, end, &nr_accounted, details);
763 tlb_finish_mmu(tlb, address, end);
764 spin_unlock(&mm->page_table_lock);
768 * Do a quick page-table lookup for a single page.
769 * mm->page_table_lock must be held.
772 __follow_page(struct mm_struct *mm, unsigned long address, int read, int write)
781 page = follow_huge_addr(mm, address, write);
785 pgd = pgd_offset(mm, address);
786 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
789 pud = pud_offset(pgd, address);
790 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
793 pmd = pmd_offset(pud, address);
794 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
797 return follow_huge_pmd(mm, address, pmd, write);
799 ptep = pte_offset_map(pmd, address);
805 if (pte_present(pte)) {
806 if (write && !pte_write(pte))
808 if (read && !pte_read(pte))
811 if (pfn_valid(pfn)) {
812 page = pfn_to_page(pfn);
813 if (write && !pte_dirty(pte) && !PageDirty(page))
814 set_page_dirty(page);
815 mark_page_accessed(page);
825 follow_page(struct mm_struct *mm, unsigned long address, int write)
827 return __follow_page(mm, address, /*read*/0, write);
831 check_user_page_readable(struct mm_struct *mm, unsigned long address)
833 return __follow_page(mm, address, /*read*/1, /*write*/0) != NULL;
836 EXPORT_SYMBOL(check_user_page_readable);
839 * Given a physical address, is there a useful struct page pointing to
840 * it? This may become more complex in the future if we start dealing
841 * with IO-aperture pages for direct-IO.
844 static inline struct page *get_page_map(struct page *page)
846 if (!pfn_valid(page_to_pfn(page)))
853 untouched_anonymous_page(struct mm_struct* mm, struct vm_area_struct *vma,
854 unsigned long address)
860 /* Check if the vma is for an anonymous mapping. */
861 if (vma->vm_ops && vma->vm_ops->nopage)
864 /* Check if page directory entry exists. */
865 pgd = pgd_offset(mm, address);
866 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
869 pud = pud_offset(pgd, address);
870 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
873 /* Check if page middle directory entry exists. */
874 pmd = pmd_offset(pud, address);
875 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
878 /* There is a pte slot for 'address' in 'mm'. */
883 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
884 unsigned long start, int len, int write, int force,
885 struct page **pages, struct vm_area_struct **vmas)
891 * Require read or write permissions.
892 * If 'force' is set, we only require the "MAY" flags.
894 flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
895 flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
899 struct vm_area_struct * vma;
901 vma = find_extend_vma(mm, start);
902 if (!vma && in_gate_area(tsk, start)) {
903 unsigned long pg = start & PAGE_MASK;
904 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
909 if (write) /* user gate pages are read-only */
910 return i ? : -EFAULT;
912 pgd = pgd_offset_k(pg);
914 pgd = pgd_offset_gate(mm, pg);
915 BUG_ON(pgd_none(*pgd));
916 pud = pud_offset(pgd, pg);
917 BUG_ON(pud_none(*pud));
918 pmd = pmd_offset(pud, pg);
919 BUG_ON(pmd_none(*pmd));
920 pte = pte_offset_map(pmd, pg);
921 BUG_ON(pte_none(*pte));
923 pages[i] = pte_page(*pte);
935 if (!vma || (vma->vm_flags & VM_IO)
936 || !(flags & vma->vm_flags))
937 return i ? : -EFAULT;
939 if (is_vm_hugetlb_page(vma)) {
940 i = follow_hugetlb_page(mm, vma, pages, vmas,
944 spin_lock(&mm->page_table_lock);
947 int lookup_write = write;
949 cond_resched_lock(&mm->page_table_lock);
950 while (!(map = follow_page(mm, start, lookup_write))) {
952 * Shortcut for anonymous pages. We don't want
953 * to force the creation of pages tables for
954 * insanly big anonymously mapped areas that
955 * nobody touched so far. This is important
956 * for doing a core dump for these mappings.
959 untouched_anonymous_page(mm,vma,start)) {
960 map = ZERO_PAGE(start);
963 spin_unlock(&mm->page_table_lock);
964 switch (handle_mm_fault(mm,vma,start,write)) {
971 case VM_FAULT_SIGBUS:
972 return i ? i : -EFAULT;
974 return i ? i : -ENOMEM;
979 * Now that we have performed a write fault
980 * and surely no longer have a shared page we
981 * shouldn't write, we shouldn't ignore an
982 * unwritable page in the page table if
983 * we are forcing write access.
985 lookup_write = write && !force;
986 spin_lock(&mm->page_table_lock);
989 pages[i] = get_page_map(map);
991 spin_unlock(&mm->page_table_lock);
993 page_cache_release(pages[i]);
997 flush_dcache_page(pages[i]);
998 if (!PageReserved(pages[i]))
999 page_cache_get(pages[i]);
1006 } while(len && start < vma->vm_end);
1007 spin_unlock(&mm->page_table_lock);
1013 EXPORT_SYMBOL(get_user_pages);
1015 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1016 unsigned long addr, unsigned long end, pgprot_t prot)
1020 pte = pte_alloc_map(mm, pmd, addr);
1024 pte_t zero_pte = pte_wrprotect(mk_pte(ZERO_PAGE(addr), prot));
1025 BUG_ON(!pte_none(*pte));
1026 set_pte_at(mm, addr, pte, zero_pte);
1027 } while (pte++, addr += PAGE_SIZE, addr != end);
1032 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1033 unsigned long addr, unsigned long end, pgprot_t prot)
1038 pmd = pmd_alloc(mm, pud, addr);
1042 next = pmd_addr_end(addr, end);
1043 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1045 } while (pmd++, addr = next, addr != end);
1049 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1050 unsigned long addr, unsigned long end, pgprot_t prot)
1055 pud = pud_alloc(mm, pgd, addr);
1059 next = pud_addr_end(addr, end);
1060 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1062 } while (pud++, addr = next, addr != end);
1066 int zeromap_page_range(struct vm_area_struct *vma,
1067 unsigned long addr, unsigned long size, pgprot_t prot)
1071 unsigned long end = addr + size;
1072 struct mm_struct *mm = vma->vm_mm;
1075 BUG_ON(addr >= end);
1076 pgd = pgd_offset(mm, addr);
1077 flush_cache_range(vma, addr, end);
1078 spin_lock(&mm->page_table_lock);
1080 next = pgd_addr_end(addr, end);
1081 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1084 } while (pgd++, addr = next, addr != end);
1085 spin_unlock(&mm->page_table_lock);
1090 * maps a range of physical memory into the requested pages. the old
1091 * mappings are removed. any references to nonexistent pages results
1092 * in null mappings (currently treated as "copy-on-access")
1094 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1095 unsigned long addr, unsigned long end,
1096 unsigned long pfn, pgprot_t prot)
1100 pte = pte_alloc_map(mm, pmd, addr);
1104 BUG_ON(!pte_none(*pte));
1105 if (!pfn_valid(pfn) || PageReserved(pfn_to_page(pfn)))
1106 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1108 } while (pte++, addr += PAGE_SIZE, addr != end);
1113 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1114 unsigned long addr, unsigned long end,
1115 unsigned long pfn, pgprot_t prot)
1120 pfn -= addr >> PAGE_SHIFT;
1121 pmd = pmd_alloc(mm, pud, addr);
1125 next = pmd_addr_end(addr, end);
1126 if (remap_pte_range(mm, pmd, addr, next,
1127 pfn + (addr >> PAGE_SHIFT), prot))
1129 } while (pmd++, addr = next, addr != end);
1133 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1134 unsigned long addr, unsigned long end,
1135 unsigned long pfn, pgprot_t prot)
1140 pfn -= addr >> PAGE_SHIFT;
1141 pud = pud_alloc(mm, pgd, addr);
1145 next = pud_addr_end(addr, end);
1146 if (remap_pmd_range(mm, pud, addr, next,
1147 pfn + (addr >> PAGE_SHIFT), prot))
1149 } while (pud++, addr = next, addr != end);
1153 /* Note: this is only safe if the mm semaphore is held when called. */
1154 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1155 unsigned long pfn, unsigned long size, pgprot_t prot)
1159 unsigned long end = addr + size;
1160 struct mm_struct *mm = vma->vm_mm;
1164 * Physically remapped pages are special. Tell the
1165 * rest of the world about it:
1166 * VM_IO tells people not to look at these pages
1167 * (accesses can have side effects).
1168 * VM_RESERVED tells swapout not to try to touch
1171 vma->vm_flags |= VM_IO | VM_RESERVED;
1173 BUG_ON(addr >= end);
1174 pfn -= addr >> PAGE_SHIFT;
1175 pgd = pgd_offset(mm, addr);
1176 flush_cache_range(vma, addr, end);
1177 spin_lock(&mm->page_table_lock);
1179 next = pgd_addr_end(addr, end);
1180 err = remap_pud_range(mm, pgd, addr, next,
1181 pfn + (addr >> PAGE_SHIFT), prot);
1184 } while (pgd++, addr = next, addr != end);
1185 spin_unlock(&mm->page_table_lock);
1188 EXPORT_SYMBOL(remap_pfn_range);
1191 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1192 * servicing faults for write access. In the normal case, do always want
1193 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1194 * that do not have writing enabled, when used by access_process_vm.
1196 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1198 if (likely(vma->vm_flags & VM_WRITE))
1199 pte = pte_mkwrite(pte);
1204 * We hold the mm semaphore for reading and vma->vm_mm->page_table_lock
1206 static inline void break_cow(struct vm_area_struct * vma, struct page * new_page, unsigned long address,
1211 entry = maybe_mkwrite(pte_mkdirty(mk_pte(new_page, vma->vm_page_prot)),
1213 ptep_establish(vma, address, page_table, entry);
1214 update_mmu_cache(vma, address, entry);
1215 lazy_mmu_prot_update(entry);
1219 * This routine handles present pages, when users try to write
1220 * to a shared page. It is done by copying the page to a new address
1221 * and decrementing the shared-page counter for the old page.
1223 * Goto-purists beware: the only reason for goto's here is that it results
1224 * in better assembly code.. The "default" path will see no jumps at all.
1226 * Note that this routine assumes that the protection checks have been
1227 * done by the caller (the low-level page fault routine in most cases).
1228 * Thus we can safely just mark it writable once we've done any necessary
1231 * We also mark the page dirty at this point even though the page will
1232 * change only once the write actually happens. This avoids a few races,
1233 * and potentially makes it more efficient.
1235 * We hold the mm semaphore and the page_table_lock on entry and exit
1236 * with the page_table_lock released.
1238 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct * vma,
1239 unsigned long address, pte_t *page_table, pmd_t *pmd, pte_t pte)
1241 struct page *old_page, *new_page;
1242 unsigned long pfn = pte_pfn(pte);
1245 if (unlikely(!pfn_valid(pfn))) {
1247 * This should really halt the system so it can be debugged or
1248 * at least the kernel stops what it's doing before it corrupts
1249 * data, but for the moment just pretend this is OOM.
1251 pte_unmap(page_table);
1252 printk(KERN_ERR "do_wp_page: bogus page at address %08lx\n",
1254 spin_unlock(&mm->page_table_lock);
1255 return VM_FAULT_OOM;
1257 old_page = pfn_to_page(pfn);
1259 if (!TestSetPageLocked(old_page)) {
1260 int reuse = can_share_swap_page(old_page);
1261 unlock_page(old_page);
1263 flush_cache_page(vma, address, pfn);
1264 entry = maybe_mkwrite(pte_mkyoung(pte_mkdirty(pte)),
1266 ptep_set_access_flags(vma, address, page_table, entry, 1);
1267 update_mmu_cache(vma, address, entry);
1268 lazy_mmu_prot_update(entry);
1269 pte_unmap(page_table);
1270 spin_unlock(&mm->page_table_lock);
1271 return VM_FAULT_MINOR;
1274 pte_unmap(page_table);
1277 * Ok, we need to copy. Oh, well..
1279 if (!PageReserved(old_page))
1280 page_cache_get(old_page);
1281 spin_unlock(&mm->page_table_lock);
1283 if (unlikely(anon_vma_prepare(vma)))
1285 if (old_page == ZERO_PAGE(address)) {
1286 new_page = alloc_zeroed_user_highpage(vma, address);
1290 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1293 copy_user_highpage(new_page, old_page, address);
1296 * Re-check the pte - we dropped the lock
1298 spin_lock(&mm->page_table_lock);
1299 page_table = pte_offset_map(pmd, address);
1300 if (likely(pte_same(*page_table, pte))) {
1301 if (PageAnon(old_page))
1302 dec_mm_counter(mm, anon_rss);
1303 if (PageReserved(old_page))
1304 inc_mm_counter(mm, rss);
1306 page_remove_rmap(old_page);
1307 flush_cache_page(vma, address, pfn);
1308 break_cow(vma, new_page, address, page_table);
1309 lru_cache_add_active(new_page);
1310 page_add_anon_rmap(new_page, vma, address);
1312 /* Free the old page.. */
1313 new_page = old_page;
1315 pte_unmap(page_table);
1316 page_cache_release(new_page);
1317 page_cache_release(old_page);
1318 spin_unlock(&mm->page_table_lock);
1319 return VM_FAULT_MINOR;
1322 page_cache_release(old_page);
1323 return VM_FAULT_OOM;
1327 * Helper functions for unmap_mapping_range().
1329 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1331 * We have to restart searching the prio_tree whenever we drop the lock,
1332 * since the iterator is only valid while the lock is held, and anyway
1333 * a later vma might be split and reinserted earlier while lock dropped.
1335 * The list of nonlinear vmas could be handled more efficiently, using
1336 * a placeholder, but handle it in the same way until a need is shown.
1337 * It is important to search the prio_tree before nonlinear list: a vma
1338 * may become nonlinear and be shifted from prio_tree to nonlinear list
1339 * while the lock is dropped; but never shifted from list to prio_tree.
1341 * In order to make forward progress despite restarting the search,
1342 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1343 * quickly skip it next time around. Since the prio_tree search only
1344 * shows us those vmas affected by unmapping the range in question, we
1345 * can't efficiently keep all vmas in step with mapping->truncate_count:
1346 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1347 * mapping->truncate_count and vma->vm_truncate_count are protected by
1350 * In order to make forward progress despite repeatedly restarting some
1351 * large vma, note the break_addr set by unmap_vmas when it breaks out:
1352 * and restart from that address when we reach that vma again. It might
1353 * have been split or merged, shrunk or extended, but never shifted: so
1354 * restart_addr remains valid so long as it remains in the vma's range.
1355 * unmap_mapping_range forces truncate_count to leap over page-aligned
1356 * values so we can save vma's restart_addr in its truncate_count field.
1358 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1360 static void reset_vma_truncate_counts(struct address_space *mapping)
1362 struct vm_area_struct *vma;
1363 struct prio_tree_iter iter;
1365 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1366 vma->vm_truncate_count = 0;
1367 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1368 vma->vm_truncate_count = 0;
1371 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1372 unsigned long start_addr, unsigned long end_addr,
1373 struct zap_details *details)
1375 unsigned long restart_addr;
1379 restart_addr = vma->vm_truncate_count;
1380 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1381 start_addr = restart_addr;
1382 if (start_addr >= end_addr) {
1383 /* Top of vma has been split off since last time */
1384 vma->vm_truncate_count = details->truncate_count;
1389 details->break_addr = end_addr;
1390 zap_page_range(vma, start_addr, end_addr - start_addr, details);
1393 * We cannot rely on the break test in unmap_vmas:
1394 * on the one hand, we don't want to restart our loop
1395 * just because that broke out for the page_table_lock;
1396 * on the other hand, it does no test when vma is small.
1398 need_break = need_resched() ||
1399 need_lockbreak(details->i_mmap_lock);
1401 if (details->break_addr >= end_addr) {
1402 /* We have now completed this vma: mark it so */
1403 vma->vm_truncate_count = details->truncate_count;
1407 /* Note restart_addr in vma's truncate_count field */
1408 vma->vm_truncate_count = details->break_addr;
1413 spin_unlock(details->i_mmap_lock);
1415 spin_lock(details->i_mmap_lock);
1419 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1420 struct zap_details *details)
1422 struct vm_area_struct *vma;
1423 struct prio_tree_iter iter;
1424 pgoff_t vba, vea, zba, zea;
1427 vma_prio_tree_foreach(vma, &iter, root,
1428 details->first_index, details->last_index) {
1429 /* Skip quickly over those we have already dealt with */
1430 if (vma->vm_truncate_count == details->truncate_count)
1433 vba = vma->vm_pgoff;
1434 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1435 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1436 zba = details->first_index;
1439 zea = details->last_index;
1443 if (unmap_mapping_range_vma(vma,
1444 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1445 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1451 static inline void unmap_mapping_range_list(struct list_head *head,
1452 struct zap_details *details)
1454 struct vm_area_struct *vma;
1457 * In nonlinear VMAs there is no correspondence between virtual address
1458 * offset and file offset. So we must perform an exhaustive search
1459 * across *all* the pages in each nonlinear VMA, not just the pages
1460 * whose virtual address lies outside the file truncation point.
1463 list_for_each_entry(vma, head, shared.vm_set.list) {
1464 /* Skip quickly over those we have already dealt with */
1465 if (vma->vm_truncate_count == details->truncate_count)
1467 details->nonlinear_vma = vma;
1468 if (unmap_mapping_range_vma(vma, vma->vm_start,
1469 vma->vm_end, details) < 0)
1475 * unmap_mapping_range - unmap the portion of all mmaps
1476 * in the specified address_space corresponding to the specified
1477 * page range in the underlying file.
1478 * @address_space: the address space containing mmaps to be unmapped.
1479 * @holebegin: byte in first page to unmap, relative to the start of
1480 * the underlying file. This will be rounded down to a PAGE_SIZE
1481 * boundary. Note that this is different from vmtruncate(), which
1482 * must keep the partial page. In contrast, we must get rid of
1484 * @holelen: size of prospective hole in bytes. This will be rounded
1485 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1487 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1488 * but 0 when invalidating pagecache, don't throw away private data.
1490 void unmap_mapping_range(struct address_space *mapping,
1491 loff_t const holebegin, loff_t const holelen, int even_cows)
1493 struct zap_details details;
1494 pgoff_t hba = holebegin >> PAGE_SHIFT;
1495 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1497 /* Check for overflow. */
1498 if (sizeof(holelen) > sizeof(hlen)) {
1500 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1501 if (holeend & ~(long long)ULONG_MAX)
1502 hlen = ULONG_MAX - hba + 1;
1505 details.check_mapping = even_cows? NULL: mapping;
1506 details.nonlinear_vma = NULL;
1507 details.first_index = hba;
1508 details.last_index = hba + hlen - 1;
1509 if (details.last_index < details.first_index)
1510 details.last_index = ULONG_MAX;
1511 details.i_mmap_lock = &mapping->i_mmap_lock;
1513 spin_lock(&mapping->i_mmap_lock);
1515 /* serialize i_size write against truncate_count write */
1517 /* Protect against page faults, and endless unmapping loops */
1518 mapping->truncate_count++;
1520 * For archs where spin_lock has inclusive semantics like ia64
1521 * this smp_mb() will prevent to read pagetable contents
1522 * before the truncate_count increment is visible to
1526 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1527 if (mapping->truncate_count == 0)
1528 reset_vma_truncate_counts(mapping);
1529 mapping->truncate_count++;
1531 details.truncate_count = mapping->truncate_count;
1533 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1534 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1535 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1536 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1537 spin_unlock(&mapping->i_mmap_lock);
1539 EXPORT_SYMBOL(unmap_mapping_range);
1542 * Handle all mappings that got truncated by a "truncate()"
1545 * NOTE! We have to be ready to update the memory sharing
1546 * between the file and the memory map for a potential last
1547 * incomplete page. Ugly, but necessary.
1549 int vmtruncate(struct inode * inode, loff_t offset)
1551 struct address_space *mapping = inode->i_mapping;
1552 unsigned long limit;
1554 if (inode->i_size < offset)
1557 * truncation of in-use swapfiles is disallowed - it would cause
1558 * subsequent swapout to scribble on the now-freed blocks.
1560 if (IS_SWAPFILE(inode))
1562 i_size_write(inode, offset);
1563 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1564 truncate_inode_pages(mapping, offset);
1568 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1569 if (limit != RLIM_INFINITY && offset > limit)
1571 if (offset > inode->i_sb->s_maxbytes)
1573 i_size_write(inode, offset);
1576 if (inode->i_op && inode->i_op->truncate)
1577 inode->i_op->truncate(inode);
1580 send_sig(SIGXFSZ, current, 0);
1587 EXPORT_SYMBOL(vmtruncate);
1590 * Primitive swap readahead code. We simply read an aligned block of
1591 * (1 << page_cluster) entries in the swap area. This method is chosen
1592 * because it doesn't cost us any seek time. We also make sure to queue
1593 * the 'original' request together with the readahead ones...
1595 * This has been extended to use the NUMA policies from the mm triggering
1598 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1600 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1603 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1606 struct page *new_page;
1607 unsigned long offset;
1610 * Get the number of handles we should do readahead io to.
1612 num = valid_swaphandles(entry, &offset);
1613 for (i = 0; i < num; offset++, i++) {
1614 /* Ok, do the async read-ahead now */
1615 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1616 offset), vma, addr);
1619 page_cache_release(new_page);
1622 * Find the next applicable VMA for the NUMA policy.
1628 if (addr >= vma->vm_end) {
1630 next_vma = vma ? vma->vm_next : NULL;
1632 if (vma && addr < vma->vm_start)
1635 if (next_vma && addr >= next_vma->vm_start) {
1637 next_vma = vma->vm_next;
1642 lru_add_drain(); /* Push any new pages onto the LRU now */
1646 * We hold the mm semaphore and the page_table_lock on entry and
1647 * should release the pagetable lock on exit..
1649 static int do_swap_page(struct mm_struct * mm,
1650 struct vm_area_struct * vma, unsigned long address,
1651 pte_t *page_table, pmd_t *pmd, pte_t orig_pte, int write_access)
1654 swp_entry_t entry = pte_to_swp_entry(orig_pte);
1656 int ret = VM_FAULT_MINOR;
1658 pte_unmap(page_table);
1659 spin_unlock(&mm->page_table_lock);
1660 page = lookup_swap_cache(entry);
1662 swapin_readahead(entry, address, vma);
1663 page = read_swap_cache_async(entry, vma, address);
1666 * Back out if somebody else faulted in this pte while
1667 * we released the page table lock.
1669 spin_lock(&mm->page_table_lock);
1670 page_table = pte_offset_map(pmd, address);
1671 if (likely(pte_same(*page_table, orig_pte)))
1674 ret = VM_FAULT_MINOR;
1675 pte_unmap(page_table);
1676 spin_unlock(&mm->page_table_lock);
1680 /* Had to read the page from swap area: Major fault */
1681 ret = VM_FAULT_MAJOR;
1682 inc_page_state(pgmajfault);
1686 mark_page_accessed(page);
1690 * Back out if somebody else faulted in this pte while we
1691 * released the page table lock.
1693 spin_lock(&mm->page_table_lock);
1694 page_table = pte_offset_map(pmd, address);
1695 if (unlikely(!pte_same(*page_table, orig_pte))) {
1696 pte_unmap(page_table);
1697 spin_unlock(&mm->page_table_lock);
1699 page_cache_release(page);
1700 ret = VM_FAULT_MINOR;
1704 /* The page isn't present yet, go ahead with the fault. */
1708 remove_exclusive_swap_page(page);
1710 inc_mm_counter(mm, rss);
1711 pte = mk_pte(page, vma->vm_page_prot);
1712 if (write_access && can_share_swap_page(page)) {
1713 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
1718 flush_icache_page(vma, page);
1719 set_pte_at(mm, address, page_table, pte);
1720 page_add_anon_rmap(page, vma, address);
1723 if (do_wp_page(mm, vma, address,
1724 page_table, pmd, pte) == VM_FAULT_OOM)
1729 /* No need to invalidate - it was non-present before */
1730 update_mmu_cache(vma, address, pte);
1731 lazy_mmu_prot_update(pte);
1732 pte_unmap(page_table);
1733 spin_unlock(&mm->page_table_lock);
1739 * We are called with the MM semaphore and page_table_lock
1740 * spinlock held to protect against concurrent faults in
1741 * multithreaded programs.
1744 do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
1745 pte_t *page_table, pmd_t *pmd, int write_access,
1749 struct page * page = ZERO_PAGE(addr);
1751 /* Read-only mapping of ZERO_PAGE. */
1752 entry = pte_wrprotect(mk_pte(ZERO_PAGE(addr), vma->vm_page_prot));
1754 /* ..except if it's a write access */
1756 /* Allocate our own private page. */
1757 pte_unmap(page_table);
1758 spin_unlock(&mm->page_table_lock);
1760 if (unlikely(anon_vma_prepare(vma)))
1762 page = alloc_zeroed_user_highpage(vma, addr);
1766 spin_lock(&mm->page_table_lock);
1767 page_table = pte_offset_map(pmd, addr);
1769 if (!pte_none(*page_table)) {
1770 pte_unmap(page_table);
1771 page_cache_release(page);
1772 spin_unlock(&mm->page_table_lock);
1775 inc_mm_counter(mm, rss);
1776 entry = maybe_mkwrite(pte_mkdirty(mk_pte(page,
1777 vma->vm_page_prot)),
1779 lru_cache_add_active(page);
1780 SetPageReferenced(page);
1781 page_add_anon_rmap(page, vma, addr);
1784 set_pte_at(mm, addr, page_table, entry);
1785 pte_unmap(page_table);
1787 /* No need to invalidate - it was non-present before */
1788 update_mmu_cache(vma, addr, entry);
1789 lazy_mmu_prot_update(entry);
1790 spin_unlock(&mm->page_table_lock);
1792 return VM_FAULT_MINOR;
1794 return VM_FAULT_OOM;
1798 * do_no_page() tries to create a new page mapping. It aggressively
1799 * tries to share with existing pages, but makes a separate copy if
1800 * the "write_access" parameter is true in order to avoid the next
1803 * As this is called only for pages that do not currently exist, we
1804 * do not need to flush old virtual caches or the TLB.
1806 * This is called with the MM semaphore held and the page table
1807 * spinlock held. Exit with the spinlock released.
1810 do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
1811 unsigned long address, int write_access, pte_t *page_table, pmd_t *pmd)
1813 struct page * new_page;
1814 struct address_space *mapping = NULL;
1816 unsigned int sequence = 0;
1817 int ret = VM_FAULT_MINOR;
1820 if (!vma->vm_ops || !vma->vm_ops->nopage)
1821 return do_anonymous_page(mm, vma, page_table,
1822 pmd, write_access, address);
1823 pte_unmap(page_table);
1824 spin_unlock(&mm->page_table_lock);
1827 mapping = vma->vm_file->f_mapping;
1828 sequence = mapping->truncate_count;
1829 smp_rmb(); /* serializes i_size against truncate_count */
1833 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
1835 * No smp_rmb is needed here as long as there's a full
1836 * spin_lock/unlock sequence inside the ->nopage callback
1837 * (for the pagecache lookup) that acts as an implicit
1838 * smp_mb() and prevents the i_size read to happen
1839 * after the next truncate_count read.
1842 /* no page was available -- either SIGBUS or OOM */
1843 if (new_page == NOPAGE_SIGBUS)
1844 return VM_FAULT_SIGBUS;
1845 if (new_page == NOPAGE_OOM)
1846 return VM_FAULT_OOM;
1849 * Should we do an early C-O-W break?
1851 if (write_access && !(vma->vm_flags & VM_SHARED)) {
1854 if (unlikely(anon_vma_prepare(vma)))
1856 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1859 copy_user_highpage(page, new_page, address);
1860 page_cache_release(new_page);
1865 spin_lock(&mm->page_table_lock);
1867 * For a file-backed vma, someone could have truncated or otherwise
1868 * invalidated this page. If unmap_mapping_range got called,
1869 * retry getting the page.
1871 if (mapping && unlikely(sequence != mapping->truncate_count)) {
1872 sequence = mapping->truncate_count;
1873 spin_unlock(&mm->page_table_lock);
1874 page_cache_release(new_page);
1877 page_table = pte_offset_map(pmd, address);
1880 * This silly early PAGE_DIRTY setting removes a race
1881 * due to the bad i386 page protection. But it's valid
1882 * for other architectures too.
1884 * Note that if write_access is true, we either now have
1885 * an exclusive copy of the page, or this is a shared mapping,
1886 * so we can make it writable and dirty to avoid having to
1887 * handle that later.
1889 /* Only go through if we didn't race with anybody else... */
1890 if (pte_none(*page_table)) {
1891 if (!PageReserved(new_page))
1892 inc_mm_counter(mm, rss);
1894 flush_icache_page(vma, new_page);
1895 entry = mk_pte(new_page, vma->vm_page_prot);
1897 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1898 set_pte_at(mm, address, page_table, entry);
1900 lru_cache_add_active(new_page);
1901 page_add_anon_rmap(new_page, vma, address);
1903 page_add_file_rmap(new_page);
1904 pte_unmap(page_table);
1906 /* One of our sibling threads was faster, back out. */
1907 pte_unmap(page_table);
1908 page_cache_release(new_page);
1909 spin_unlock(&mm->page_table_lock);
1913 /* no need to invalidate: a not-present page shouldn't be cached */
1914 update_mmu_cache(vma, address, entry);
1915 lazy_mmu_prot_update(entry);
1916 spin_unlock(&mm->page_table_lock);
1920 page_cache_release(new_page);
1926 * Fault of a previously existing named mapping. Repopulate the pte
1927 * from the encoded file_pte if possible. This enables swappable
1930 static int do_file_page(struct mm_struct * mm, struct vm_area_struct * vma,
1931 unsigned long address, int write_access, pte_t *pte, pmd_t *pmd)
1933 unsigned long pgoff;
1936 BUG_ON(!vma->vm_ops || !vma->vm_ops->nopage);
1938 * Fall back to the linear mapping if the fs does not support
1941 if (!vma->vm_ops || !vma->vm_ops->populate ||
1942 (write_access && !(vma->vm_flags & VM_SHARED))) {
1943 pte_clear(mm, address, pte);
1944 return do_no_page(mm, vma, address, write_access, pte, pmd);
1947 pgoff = pte_to_pgoff(*pte);
1950 spin_unlock(&mm->page_table_lock);
1952 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE, vma->vm_page_prot, pgoff, 0);
1954 return VM_FAULT_OOM;
1956 return VM_FAULT_SIGBUS;
1957 return VM_FAULT_MAJOR;
1961 * These routines also need to handle stuff like marking pages dirty
1962 * and/or accessed for architectures that don't do it in hardware (most
1963 * RISC architectures). The early dirtying is also good on the i386.
1965 * There is also a hook called "update_mmu_cache()" that architectures
1966 * with external mmu caches can use to update those (ie the Sparc or
1967 * PowerPC hashed page tables that act as extended TLBs).
1969 * Note the "page_table_lock". It is to protect against kswapd removing
1970 * pages from under us. Note that kswapd only ever _removes_ pages, never
1971 * adds them. As such, once we have noticed that the page is not present,
1972 * we can drop the lock early.
1974 * The adding of pages is protected by the MM semaphore (which we hold),
1975 * so we don't need to worry about a page being suddenly been added into
1978 * We enter with the pagetable spinlock held, we are supposed to
1979 * release it when done.
1981 static inline int handle_pte_fault(struct mm_struct *mm,
1982 struct vm_area_struct * vma, unsigned long address,
1983 int write_access, pte_t *pte, pmd_t *pmd)
1988 if (!pte_present(entry)) {
1990 * If it truly wasn't present, we know that kswapd
1991 * and the PTE updates will not touch it later. So
1994 if (pte_none(entry))
1995 return do_no_page(mm, vma, address, write_access, pte, pmd);
1996 if (pte_file(entry))
1997 return do_file_page(mm, vma, address, write_access, pte, pmd);
1998 return do_swap_page(mm, vma, address, pte, pmd, entry, write_access);
2002 if (!pte_write(entry))
2003 return do_wp_page(mm, vma, address, pte, pmd, entry);
2005 entry = pte_mkdirty(entry);
2007 entry = pte_mkyoung(entry);
2008 ptep_set_access_flags(vma, address, pte, entry, write_access);
2009 update_mmu_cache(vma, address, entry);
2010 lazy_mmu_prot_update(entry);
2012 spin_unlock(&mm->page_table_lock);
2013 return VM_FAULT_MINOR;
2017 * By the time we get here, we already hold the mm semaphore
2019 int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct * vma,
2020 unsigned long address, int write_access)
2027 __set_current_state(TASK_RUNNING);
2029 inc_page_state(pgfault);
2031 if (is_vm_hugetlb_page(vma))
2032 return VM_FAULT_SIGBUS; /* mapping truncation does this. */
2035 * We need the page table lock to synchronize with kswapd
2036 * and the SMP-safe atomic PTE updates.
2038 pgd = pgd_offset(mm, address);
2039 spin_lock(&mm->page_table_lock);
2041 pud = pud_alloc(mm, pgd, address);
2045 pmd = pmd_alloc(mm, pud, address);
2049 pte = pte_alloc_map(mm, pmd, address);
2053 return handle_pte_fault(mm, vma, address, write_access, pte, pmd);
2056 spin_unlock(&mm->page_table_lock);
2057 return VM_FAULT_OOM;
2060 #ifndef __PAGETABLE_PUD_FOLDED
2062 * Allocate page upper directory.
2064 * We've already handled the fast-path in-line, and we own the
2067 pud_t fastcall *__pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2071 spin_unlock(&mm->page_table_lock);
2072 new = pud_alloc_one(mm, address);
2073 spin_lock(&mm->page_table_lock);
2078 * Because we dropped the lock, we should re-check the
2079 * entry, as somebody else could have populated it..
2081 if (pgd_present(*pgd)) {
2085 pgd_populate(mm, pgd, new);
2087 return pud_offset(pgd, address);
2089 #endif /* __PAGETABLE_PUD_FOLDED */
2091 #ifndef __PAGETABLE_PMD_FOLDED
2093 * Allocate page middle directory.
2095 * We've already handled the fast-path in-line, and we own the
2098 pmd_t fastcall *__pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2102 spin_unlock(&mm->page_table_lock);
2103 new = pmd_alloc_one(mm, address);
2104 spin_lock(&mm->page_table_lock);
2109 * Because we dropped the lock, we should re-check the
2110 * entry, as somebody else could have populated it..
2112 #ifndef __ARCH_HAS_4LEVEL_HACK
2113 if (pud_present(*pud)) {
2117 pud_populate(mm, pud, new);
2119 if (pgd_present(*pud)) {
2123 pgd_populate(mm, pud, new);
2124 #endif /* __ARCH_HAS_4LEVEL_HACK */
2127 return pmd_offset(pud, address);
2129 #endif /* __PAGETABLE_PMD_FOLDED */
2131 int make_pages_present(unsigned long addr, unsigned long end)
2133 int ret, len, write;
2134 struct vm_area_struct * vma;
2136 vma = find_vma(current->mm, addr);
2139 write = (vma->vm_flags & VM_WRITE) != 0;
2142 if (end > vma->vm_end)
2144 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2145 ret = get_user_pages(current, current->mm, addr,
2146 len, write, 0, NULL, NULL);
2149 return ret == len ? 0 : -1;
2153 * Map a vmalloc()-space virtual address to the physical page.
2155 struct page * vmalloc_to_page(void * vmalloc_addr)
2157 unsigned long addr = (unsigned long) vmalloc_addr;
2158 struct page *page = NULL;
2159 pgd_t *pgd = pgd_offset_k(addr);
2164 if (!pgd_none(*pgd)) {
2165 pud = pud_offset(pgd, addr);
2166 if (!pud_none(*pud)) {
2167 pmd = pmd_offset(pud, addr);
2168 if (!pmd_none(*pmd)) {
2169 ptep = pte_offset_map(pmd, addr);
2171 if (pte_present(pte))
2172 page = pte_page(pte);
2180 EXPORT_SYMBOL(vmalloc_to_page);
2183 * Map a vmalloc()-space virtual address to the physical page frame number.
2185 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2187 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2190 EXPORT_SYMBOL(vmalloc_to_pfn);
2193 * update_mem_hiwater
2194 * - update per process rss and vm high water data
2196 void update_mem_hiwater(struct task_struct *tsk)
2199 unsigned long rss = get_mm_counter(tsk->mm, rss);
2201 if (tsk->mm->hiwater_rss < rss)
2202 tsk->mm->hiwater_rss = rss;
2203 if (tsk->mm->hiwater_vm < tsk->mm->total_vm)
2204 tsk->mm->hiwater_vm = tsk->mm->total_vm;
2208 #if !defined(__HAVE_ARCH_GATE_AREA)
2210 #if defined(AT_SYSINFO_EHDR)
2211 struct vm_area_struct gate_vma;
2213 static int __init gate_vma_init(void)
2215 gate_vma.vm_mm = NULL;
2216 gate_vma.vm_start = FIXADDR_USER_START;
2217 gate_vma.vm_end = FIXADDR_USER_END;
2218 gate_vma.vm_page_prot = PAGE_READONLY;
2219 gate_vma.vm_flags = 0;
2222 __initcall(gate_vma_init);
2225 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2227 #ifdef AT_SYSINFO_EHDR
2234 int in_gate_area_no_task(unsigned long addr)
2236 #ifdef AT_SYSINFO_EHDR
2237 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2243 #endif /* __HAVE_ARCH_GATE_AREA */