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);
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 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
285 struct page *new = pte_alloc_one(mm, address);
289 spin_lock(&mm->page_table_lock);
290 if (pmd_present(*pmd)) /* Another has populated it */
294 inc_page_state(nr_page_table_pages);
295 pmd_populate(mm, pmd, new);
297 spin_unlock(&mm->page_table_lock);
301 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
303 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
307 spin_lock(&init_mm.page_table_lock);
308 if (pmd_present(*pmd)) /* Another has populated it */
309 pte_free_kernel(new);
311 pmd_populate_kernel(&init_mm, pmd, new);
312 spin_unlock(&init_mm.page_table_lock);
316 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
319 add_mm_counter(mm, file_rss, file_rss);
321 add_mm_counter(mm, anon_rss, anon_rss);
325 * This function is called to print an error when a pte in a
326 * !VM_RESERVED region is found pointing to an invalid pfn (which
329 * The calling function must still handle the error.
331 void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
333 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
334 "vm_flags = %lx, vaddr = %lx\n",
335 (long long)pte_val(pte),
336 (vma->vm_mm == current->mm ? current->comm : "???"),
337 vma->vm_flags, vaddr);
342 * copy one vm_area from one task to the other. Assumes the page tables
343 * already present in the new task to be cleared in the whole range
344 * covered by this vma.
348 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
349 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
350 unsigned long addr, int *rss)
352 unsigned long vm_flags = vma->vm_flags;
353 pte_t pte = *src_pte;
357 /* pte contains position in swap or file, so copy. */
358 if (unlikely(!pte_present(pte))) {
359 if (!pte_file(pte)) {
360 swap_duplicate(pte_to_swp_entry(pte));
361 /* make sure dst_mm is on swapoff's mmlist. */
362 if (unlikely(list_empty(&dst_mm->mmlist))) {
363 spin_lock(&mmlist_lock);
364 list_add(&dst_mm->mmlist, &src_mm->mmlist);
365 spin_unlock(&mmlist_lock);
371 /* If the region is VM_RESERVED, the mapping is not
372 * mapped via rmap - duplicate the pte as is.
374 if (vm_flags & VM_RESERVED)
378 /* If the pte points outside of valid memory but
379 * the region is not VM_RESERVED, we have a problem.
381 if (unlikely(!pfn_valid(pfn))) {
382 print_bad_pte(vma, pte, addr);
383 goto out_set_pte; /* try to do something sane */
386 page = pfn_to_page(pfn);
389 * If it's a COW mapping, write protect it both
390 * in the parent and the child
392 if ((vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE) {
393 ptep_set_wrprotect(src_mm, addr, src_pte);
398 * If it's a shared mapping, mark it clean in
401 if (vm_flags & VM_SHARED)
402 pte = pte_mkclean(pte);
403 pte = pte_mkold(pte);
406 rss[!!PageAnon(page)]++;
409 set_pte_at(dst_mm, addr, dst_pte, pte);
412 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
413 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
414 unsigned long addr, unsigned long end)
416 pte_t *src_pte, *dst_pte;
417 spinlock_t *src_ptl, *dst_ptl;
423 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
426 src_pte = pte_offset_map_nested(src_pmd, addr);
427 src_ptl = &src_mm->page_table_lock;
432 * We are holding two locks at this point - either of them
433 * could generate latencies in another task on another CPU.
435 if (progress >= 32) {
437 if (need_resched() ||
438 need_lockbreak(src_ptl) ||
439 need_lockbreak(dst_ptl))
442 if (pte_none(*src_pte)) {
446 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
448 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
450 spin_unlock(src_ptl);
451 pte_unmap_nested(src_pte - 1);
452 add_mm_rss(dst_mm, rss[0], rss[1]);
453 pte_unmap_unlock(dst_pte - 1, dst_ptl);
460 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
461 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
462 unsigned long addr, unsigned long end)
464 pmd_t *src_pmd, *dst_pmd;
467 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
470 src_pmd = pmd_offset(src_pud, addr);
472 next = pmd_addr_end(addr, end);
473 if (pmd_none_or_clear_bad(src_pmd))
475 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
478 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
482 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
483 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
484 unsigned long addr, unsigned long end)
486 pud_t *src_pud, *dst_pud;
489 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
492 src_pud = pud_offset(src_pgd, addr);
494 next = pud_addr_end(addr, end);
495 if (pud_none_or_clear_bad(src_pud))
497 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
500 } while (dst_pud++, src_pud++, addr = next, addr != end);
504 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
505 struct vm_area_struct *vma)
507 pgd_t *src_pgd, *dst_pgd;
509 unsigned long addr = vma->vm_start;
510 unsigned long end = vma->vm_end;
513 * Don't copy ptes where a page fault will fill them correctly.
514 * Fork becomes much lighter when there are big shared or private
515 * readonly mappings. The tradeoff is that copy_page_range is more
516 * efficient than faulting.
518 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_RESERVED))) {
523 if (is_vm_hugetlb_page(vma))
524 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
526 dst_pgd = pgd_offset(dst_mm, addr);
527 src_pgd = pgd_offset(src_mm, addr);
529 next = pgd_addr_end(addr, end);
530 if (pgd_none_or_clear_bad(src_pgd))
532 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
535 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
539 static void zap_pte_range(struct mmu_gather *tlb,
540 struct vm_area_struct *vma, pmd_t *pmd,
541 unsigned long addr, unsigned long end,
542 struct zap_details *details)
544 struct mm_struct *mm = tlb->mm;
549 pte = pte_offset_map(pmd, addr);
554 if (pte_present(ptent)) {
555 struct page *page = NULL;
556 if (!(vma->vm_flags & VM_RESERVED)) {
557 unsigned long pfn = pte_pfn(ptent);
558 if (unlikely(!pfn_valid(pfn)))
559 print_bad_pte(vma, ptent, addr);
561 page = pfn_to_page(pfn);
563 if (unlikely(details) && page) {
565 * unmap_shared_mapping_pages() wants to
566 * invalidate cache without truncating:
567 * unmap shared but keep private pages.
569 if (details->check_mapping &&
570 details->check_mapping != page->mapping)
573 * Each page->index must be checked when
574 * invalidating or truncating nonlinear.
576 if (details->nonlinear_vma &&
577 (page->index < details->first_index ||
578 page->index > details->last_index))
581 ptent = ptep_get_and_clear_full(mm, addr, pte,
583 tlb_remove_tlb_entry(tlb, pte, addr);
586 if (unlikely(details) && details->nonlinear_vma
587 && linear_page_index(details->nonlinear_vma,
588 addr) != page->index)
589 set_pte_at(mm, addr, pte,
590 pgoff_to_pte(page->index));
594 if (pte_dirty(ptent))
595 set_page_dirty(page);
596 if (pte_young(ptent))
597 mark_page_accessed(page);
600 page_remove_rmap(page);
601 tlb_remove_page(tlb, page);
605 * If details->check_mapping, we leave swap entries;
606 * if details->nonlinear_vma, we leave file entries.
608 if (unlikely(details))
610 if (!pte_file(ptent))
611 free_swap_and_cache(pte_to_swp_entry(ptent));
612 pte_clear_full(mm, addr, pte, tlb->fullmm);
613 } while (pte++, addr += PAGE_SIZE, addr != end);
615 add_mm_rss(mm, file_rss, anon_rss);
619 static inline void zap_pmd_range(struct mmu_gather *tlb,
620 struct vm_area_struct *vma, pud_t *pud,
621 unsigned long addr, unsigned long end,
622 struct zap_details *details)
627 pmd = pmd_offset(pud, addr);
629 next = pmd_addr_end(addr, end);
630 if (pmd_none_or_clear_bad(pmd))
632 zap_pte_range(tlb, vma, pmd, addr, next, details);
633 } while (pmd++, addr = next, addr != end);
636 static inline void zap_pud_range(struct mmu_gather *tlb,
637 struct vm_area_struct *vma, pgd_t *pgd,
638 unsigned long addr, unsigned long end,
639 struct zap_details *details)
644 pud = pud_offset(pgd, addr);
646 next = pud_addr_end(addr, end);
647 if (pud_none_or_clear_bad(pud))
649 zap_pmd_range(tlb, vma, pud, addr, next, details);
650 } while (pud++, addr = next, addr != end);
653 static void unmap_page_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
654 unsigned long addr, unsigned long end,
655 struct zap_details *details)
660 if (details && !details->check_mapping && !details->nonlinear_vma)
664 tlb_start_vma(tlb, vma);
665 pgd = pgd_offset(vma->vm_mm, addr);
667 next = pgd_addr_end(addr, end);
668 if (pgd_none_or_clear_bad(pgd))
670 zap_pud_range(tlb, vma, pgd, addr, next, details);
671 } while (pgd++, addr = next, addr != end);
672 tlb_end_vma(tlb, vma);
675 #ifdef CONFIG_PREEMPT
676 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
678 /* No preempt: go for improved straight-line efficiency */
679 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
683 * unmap_vmas - unmap a range of memory covered by a list of vma's
684 * @tlbp: address of the caller's struct mmu_gather
685 * @mm: the controlling mm_struct
686 * @vma: the starting vma
687 * @start_addr: virtual address at which to start unmapping
688 * @end_addr: virtual address at which to end unmapping
689 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
690 * @details: details of nonlinear truncation or shared cache invalidation
692 * Returns the end address of the unmapping (restart addr if interrupted).
694 * Unmap all pages in the vma list. Called under page_table_lock.
696 * We aim to not hold page_table_lock for too long (for scheduling latency
697 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
698 * return the ending mmu_gather to the caller.
700 * Only addresses between `start' and `end' will be unmapped.
702 * The VMA list must be sorted in ascending virtual address order.
704 * unmap_vmas() assumes that the caller will flush the whole unmapped address
705 * range after unmap_vmas() returns. So the only responsibility here is to
706 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
707 * drops the lock and schedules.
709 unsigned long unmap_vmas(struct mmu_gather **tlbp, struct mm_struct *mm,
710 struct vm_area_struct *vma, unsigned long start_addr,
711 unsigned long end_addr, unsigned long *nr_accounted,
712 struct zap_details *details)
714 unsigned long zap_bytes = ZAP_BLOCK_SIZE;
715 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
716 int tlb_start_valid = 0;
717 unsigned long start = start_addr;
718 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
719 int fullmm = (*tlbp)->fullmm;
721 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
724 start = max(vma->vm_start, start_addr);
725 if (start >= vma->vm_end)
727 end = min(vma->vm_end, end_addr);
728 if (end <= vma->vm_start)
731 if (vma->vm_flags & VM_ACCOUNT)
732 *nr_accounted += (end - start) >> PAGE_SHIFT;
734 while (start != end) {
737 if (!tlb_start_valid) {
742 if (is_vm_hugetlb_page(vma)) {
744 unmap_hugepage_range(vma, start, end);
746 block = min(zap_bytes, end - start);
747 unmap_page_range(*tlbp, vma, start,
748 start + block, details);
753 if ((long)zap_bytes > 0)
756 tlb_finish_mmu(*tlbp, tlb_start, start);
758 if (need_resched() ||
759 need_lockbreak(&mm->page_table_lock) ||
760 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
762 /* must reset count of rss freed */
763 *tlbp = tlb_gather_mmu(mm, fullmm);
766 spin_unlock(&mm->page_table_lock);
768 spin_lock(&mm->page_table_lock);
771 *tlbp = tlb_gather_mmu(mm, fullmm);
773 zap_bytes = ZAP_BLOCK_SIZE;
777 return start; /* which is now the end (or restart) address */
781 * zap_page_range - remove user pages in a given range
782 * @vma: vm_area_struct holding the applicable pages
783 * @address: starting address of pages to zap
784 * @size: number of bytes to zap
785 * @details: details of nonlinear truncation or shared cache invalidation
787 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
788 unsigned long size, struct zap_details *details)
790 struct mm_struct *mm = vma->vm_mm;
791 struct mmu_gather *tlb;
792 unsigned long end = address + size;
793 unsigned long nr_accounted = 0;
795 if (is_vm_hugetlb_page(vma)) {
796 zap_hugepage_range(vma, address, size);
801 spin_lock(&mm->page_table_lock);
802 tlb = tlb_gather_mmu(mm, 0);
803 update_hiwater_rss(mm);
804 end = unmap_vmas(&tlb, mm, vma, address, end, &nr_accounted, details);
805 tlb_finish_mmu(tlb, address, end);
806 spin_unlock(&mm->page_table_lock);
811 * Do a quick page-table lookup for a single page.
812 * mm->page_table_lock must be held.
814 static struct page *__follow_page(struct mm_struct *mm, unsigned long address,
815 int read, int write, int accessed)
824 page = follow_huge_addr(mm, address, write);
828 pgd = pgd_offset(mm, address);
829 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
832 pud = pud_offset(pgd, address);
833 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
836 pmd = pmd_offset(pud, address);
837 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
840 return follow_huge_pmd(mm, address, pmd, write);
842 ptep = pte_offset_map(pmd, address);
848 if (pte_present(pte)) {
849 if (write && !pte_write(pte))
851 if (read && !pte_read(pte))
854 if (pfn_valid(pfn)) {
855 page = pfn_to_page(pfn);
857 if (write && !pte_dirty(pte) &&!PageDirty(page))
858 set_page_dirty(page);
859 mark_page_accessed(page);
870 follow_page(struct mm_struct *mm, unsigned long address, int write)
872 return __follow_page(mm, address, 0, write, 1);
876 * check_user_page_readable() can be called frm niterrupt context by oprofile,
877 * so we need to avoid taking any non-irq-safe locks
879 int check_user_page_readable(struct mm_struct *mm, unsigned long address)
881 return __follow_page(mm, address, 1, 0, 0) != NULL;
883 EXPORT_SYMBOL(check_user_page_readable);
886 untouched_anonymous_page(struct mm_struct* mm, struct vm_area_struct *vma,
887 unsigned long address)
893 /* Check if the vma is for an anonymous mapping. */
894 if (vma->vm_ops && vma->vm_ops->nopage)
897 /* Check if page directory entry exists. */
898 pgd = pgd_offset(mm, address);
899 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
902 pud = pud_offset(pgd, address);
903 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
906 /* Check if page middle directory entry exists. */
907 pmd = pmd_offset(pud, address);
908 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
911 /* There is a pte slot for 'address' in 'mm'. */
915 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
916 unsigned long start, int len, int write, int force,
917 struct page **pages, struct vm_area_struct **vmas)
923 * Require read or write permissions.
924 * If 'force' is set, we only require the "MAY" flags.
926 flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
927 flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
931 struct vm_area_struct * vma;
933 vma = find_extend_vma(mm, start);
934 if (!vma && in_gate_area(tsk, start)) {
935 unsigned long pg = start & PAGE_MASK;
936 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
941 if (write) /* user gate pages are read-only */
942 return i ? : -EFAULT;
944 pgd = pgd_offset_k(pg);
946 pgd = pgd_offset_gate(mm, pg);
947 BUG_ON(pgd_none(*pgd));
948 pud = pud_offset(pgd, pg);
949 BUG_ON(pud_none(*pud));
950 pmd = pmd_offset(pud, pg);
952 return i ? : -EFAULT;
953 pte = pte_offset_map(pmd, pg);
954 if (pte_none(*pte)) {
956 return i ? : -EFAULT;
959 pages[i] = pte_page(*pte);
971 if (!vma || (vma->vm_flags & (VM_IO | VM_RESERVED))
972 || !(flags & vma->vm_flags))
973 return i ? : -EFAULT;
975 if (is_vm_hugetlb_page(vma)) {
976 i = follow_hugetlb_page(mm, vma, pages, vmas,
980 spin_lock(&mm->page_table_lock);
982 int write_access = write;
985 cond_resched_lock(&mm->page_table_lock);
986 while (!(page = follow_page(mm, start, write_access))) {
990 * Shortcut for anonymous pages. We don't want
991 * to force the creation of pages tables for
992 * insanely big anonymously mapped areas that
993 * nobody touched so far. This is important
994 * for doing a core dump for these mappings.
996 if (!write && untouched_anonymous_page(mm,vma,start)) {
997 page = ZERO_PAGE(start);
1000 spin_unlock(&mm->page_table_lock);
1001 ret = __handle_mm_fault(mm, vma, start, write_access);
1004 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1005 * broken COW when necessary, even if maybe_mkwrite
1006 * decided not to set pte_write. We can thus safely do
1007 * subsequent page lookups as if they were reads.
1009 if (ret & VM_FAULT_WRITE)
1012 switch (ret & ~VM_FAULT_WRITE) {
1013 case VM_FAULT_MINOR:
1016 case VM_FAULT_MAJOR:
1019 case VM_FAULT_SIGBUS:
1020 return i ? i : -EFAULT;
1022 return i ? i : -ENOMEM;
1026 spin_lock(&mm->page_table_lock);
1030 flush_dcache_page(page);
1031 page_cache_get(page);
1038 } while (len && start < vma->vm_end);
1039 spin_unlock(&mm->page_table_lock);
1043 EXPORT_SYMBOL(get_user_pages);
1045 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1046 unsigned long addr, unsigned long end, pgprot_t prot)
1051 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1055 struct page *page = ZERO_PAGE(addr);
1056 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1057 page_cache_get(page);
1058 page_add_file_rmap(page);
1059 inc_mm_counter(mm, file_rss);
1060 BUG_ON(!pte_none(*pte));
1061 set_pte_at(mm, addr, pte, zero_pte);
1062 } while (pte++, addr += PAGE_SIZE, addr != end);
1063 pte_unmap_unlock(pte - 1, ptl);
1067 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1068 unsigned long addr, unsigned long end, pgprot_t prot)
1073 pmd = pmd_alloc(mm, pud, addr);
1077 next = pmd_addr_end(addr, end);
1078 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1080 } while (pmd++, addr = next, addr != end);
1084 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1085 unsigned long addr, unsigned long end, pgprot_t prot)
1090 pud = pud_alloc(mm, pgd, addr);
1094 next = pud_addr_end(addr, end);
1095 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1097 } while (pud++, addr = next, addr != end);
1101 int zeromap_page_range(struct vm_area_struct *vma,
1102 unsigned long addr, unsigned long size, pgprot_t prot)
1106 unsigned long end = addr + size;
1107 struct mm_struct *mm = vma->vm_mm;
1110 BUG_ON(addr >= end);
1111 pgd = pgd_offset(mm, addr);
1112 flush_cache_range(vma, addr, end);
1114 next = pgd_addr_end(addr, end);
1115 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1118 } while (pgd++, addr = next, addr != end);
1123 * maps a range of physical memory into the requested pages. the old
1124 * mappings are removed. any references to nonexistent pages results
1125 * in null mappings (currently treated as "copy-on-access")
1127 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1128 unsigned long addr, unsigned long end,
1129 unsigned long pfn, pgprot_t prot)
1134 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1138 BUG_ON(!pte_none(*pte));
1139 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1141 } while (pte++, addr += PAGE_SIZE, addr != end);
1142 pte_unmap_unlock(pte - 1, ptl);
1146 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1147 unsigned long addr, unsigned long end,
1148 unsigned long pfn, pgprot_t prot)
1153 pfn -= addr >> PAGE_SHIFT;
1154 pmd = pmd_alloc(mm, pud, addr);
1158 next = pmd_addr_end(addr, end);
1159 if (remap_pte_range(mm, pmd, addr, next,
1160 pfn + (addr >> PAGE_SHIFT), prot))
1162 } while (pmd++, addr = next, addr != end);
1166 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1167 unsigned long addr, unsigned long end,
1168 unsigned long pfn, pgprot_t prot)
1173 pfn -= addr >> PAGE_SHIFT;
1174 pud = pud_alloc(mm, pgd, addr);
1178 next = pud_addr_end(addr, end);
1179 if (remap_pmd_range(mm, pud, addr, next,
1180 pfn + (addr >> PAGE_SHIFT), prot))
1182 } while (pud++, addr = next, addr != end);
1186 /* Note: this is only safe if the mm semaphore is held when called. */
1187 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1188 unsigned long pfn, unsigned long size, pgprot_t prot)
1192 unsigned long end = addr + PAGE_ALIGN(size);
1193 struct mm_struct *mm = vma->vm_mm;
1197 * Physically remapped pages are special. Tell the
1198 * rest of the world about it:
1199 * VM_IO tells people not to look at these pages
1200 * (accesses can have side effects).
1201 * VM_RESERVED tells the core MM not to "manage" these pages
1202 * (e.g. refcount, mapcount, try to swap them out).
1204 vma->vm_flags |= VM_IO | VM_RESERVED;
1206 BUG_ON(addr >= end);
1207 pfn -= addr >> PAGE_SHIFT;
1208 pgd = pgd_offset(mm, addr);
1209 flush_cache_range(vma, addr, end);
1211 next = pgd_addr_end(addr, end);
1212 err = remap_pud_range(mm, pgd, addr, next,
1213 pfn + (addr >> PAGE_SHIFT), prot);
1216 } while (pgd++, addr = next, addr != end);
1219 EXPORT_SYMBOL(remap_pfn_range);
1222 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1223 * servicing faults for write access. In the normal case, do always want
1224 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1225 * that do not have writing enabled, when used by access_process_vm.
1227 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1229 if (likely(vma->vm_flags & VM_WRITE))
1230 pte = pte_mkwrite(pte);
1235 * This routine handles present pages, when users try to write
1236 * to a shared page. It is done by copying the page to a new address
1237 * and decrementing the shared-page counter for the old page.
1239 * Note that this routine assumes that the protection checks have been
1240 * done by the caller (the low-level page fault routine in most cases).
1241 * Thus we can safely just mark it writable once we've done any necessary
1244 * We also mark the page dirty at this point even though the page will
1245 * change only once the write actually happens. This avoids a few races,
1246 * and potentially makes it more efficient.
1248 * We hold the mm semaphore and the page_table_lock on entry and exit
1249 * with the page_table_lock released.
1251 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1252 unsigned long address, pte_t *page_table, pmd_t *pmd,
1255 struct page *old_page, *new_page;
1256 unsigned long pfn = pte_pfn(orig_pte);
1258 int ret = VM_FAULT_MINOR;
1260 BUG_ON(vma->vm_flags & VM_RESERVED);
1262 if (unlikely(!pfn_valid(pfn))) {
1264 * Page table corrupted: show pte and kill process.
1266 print_bad_pte(vma, orig_pte, address);
1270 old_page = pfn_to_page(pfn);
1272 if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
1273 int reuse = can_share_swap_page(old_page);
1274 unlock_page(old_page);
1276 flush_cache_page(vma, address, pfn);
1277 entry = pte_mkyoung(orig_pte);
1278 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1279 ptep_set_access_flags(vma, address, page_table, entry, 1);
1280 update_mmu_cache(vma, address, entry);
1281 lazy_mmu_prot_update(entry);
1282 ret |= VM_FAULT_WRITE;
1288 * Ok, we need to copy. Oh, well..
1290 page_cache_get(old_page);
1291 pte_unmap(page_table);
1292 spin_unlock(&mm->page_table_lock);
1294 if (unlikely(anon_vma_prepare(vma)))
1296 if (old_page == ZERO_PAGE(address)) {
1297 new_page = alloc_zeroed_user_highpage(vma, address);
1301 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1304 copy_user_highpage(new_page, old_page, address);
1308 * Re-check the pte - we dropped the lock
1310 spin_lock(&mm->page_table_lock);
1311 page_table = pte_offset_map(pmd, address);
1312 if (likely(pte_same(*page_table, orig_pte))) {
1313 page_remove_rmap(old_page);
1314 if (!PageAnon(old_page)) {
1315 inc_mm_counter(mm, anon_rss);
1316 dec_mm_counter(mm, file_rss);
1318 flush_cache_page(vma, address, pfn);
1319 entry = mk_pte(new_page, vma->vm_page_prot);
1320 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1321 ptep_establish(vma, address, page_table, entry);
1322 update_mmu_cache(vma, address, entry);
1323 lazy_mmu_prot_update(entry);
1325 lru_cache_add_active(new_page);
1326 page_add_anon_rmap(new_page, vma, address);
1328 /* Free the old page.. */
1329 new_page = old_page;
1330 ret |= VM_FAULT_WRITE;
1332 page_cache_release(new_page);
1333 page_cache_release(old_page);
1335 pte_unmap(page_table);
1336 spin_unlock(&mm->page_table_lock);
1339 page_cache_release(old_page);
1340 return VM_FAULT_OOM;
1344 * Helper functions for unmap_mapping_range().
1346 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1348 * We have to restart searching the prio_tree whenever we drop the lock,
1349 * since the iterator is only valid while the lock is held, and anyway
1350 * a later vma might be split and reinserted earlier while lock dropped.
1352 * The list of nonlinear vmas could be handled more efficiently, using
1353 * a placeholder, but handle it in the same way until a need is shown.
1354 * It is important to search the prio_tree before nonlinear list: a vma
1355 * may become nonlinear and be shifted from prio_tree to nonlinear list
1356 * while the lock is dropped; but never shifted from list to prio_tree.
1358 * In order to make forward progress despite restarting the search,
1359 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1360 * quickly skip it next time around. Since the prio_tree search only
1361 * shows us those vmas affected by unmapping the range in question, we
1362 * can't efficiently keep all vmas in step with mapping->truncate_count:
1363 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1364 * mapping->truncate_count and vma->vm_truncate_count are protected by
1367 * In order to make forward progress despite repeatedly restarting some
1368 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1369 * and restart from that address when we reach that vma again. It might
1370 * have been split or merged, shrunk or extended, but never shifted: so
1371 * restart_addr remains valid so long as it remains in the vma's range.
1372 * unmap_mapping_range forces truncate_count to leap over page-aligned
1373 * values so we can save vma's restart_addr in its truncate_count field.
1375 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1377 static void reset_vma_truncate_counts(struct address_space *mapping)
1379 struct vm_area_struct *vma;
1380 struct prio_tree_iter iter;
1382 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1383 vma->vm_truncate_count = 0;
1384 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1385 vma->vm_truncate_count = 0;
1388 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1389 unsigned long start_addr, unsigned long end_addr,
1390 struct zap_details *details)
1392 unsigned long restart_addr;
1396 restart_addr = vma->vm_truncate_count;
1397 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1398 start_addr = restart_addr;
1399 if (start_addr >= end_addr) {
1400 /* Top of vma has been split off since last time */
1401 vma->vm_truncate_count = details->truncate_count;
1406 restart_addr = zap_page_range(vma, start_addr,
1407 end_addr - start_addr, details);
1410 * We cannot rely on the break test in unmap_vmas:
1411 * on the one hand, we don't want to restart our loop
1412 * just because that broke out for the page_table_lock;
1413 * on the other hand, it does no test when vma is small.
1415 need_break = need_resched() ||
1416 need_lockbreak(details->i_mmap_lock);
1418 if (restart_addr >= end_addr) {
1419 /* We have now completed this vma: mark it so */
1420 vma->vm_truncate_count = details->truncate_count;
1424 /* Note restart_addr in vma's truncate_count field */
1425 vma->vm_truncate_count = restart_addr;
1430 spin_unlock(details->i_mmap_lock);
1432 spin_lock(details->i_mmap_lock);
1436 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1437 struct zap_details *details)
1439 struct vm_area_struct *vma;
1440 struct prio_tree_iter iter;
1441 pgoff_t vba, vea, zba, zea;
1444 vma_prio_tree_foreach(vma, &iter, root,
1445 details->first_index, details->last_index) {
1446 /* Skip quickly over those we have already dealt with */
1447 if (vma->vm_truncate_count == details->truncate_count)
1450 vba = vma->vm_pgoff;
1451 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1452 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1453 zba = details->first_index;
1456 zea = details->last_index;
1460 if (unmap_mapping_range_vma(vma,
1461 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1462 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1468 static inline void unmap_mapping_range_list(struct list_head *head,
1469 struct zap_details *details)
1471 struct vm_area_struct *vma;
1474 * In nonlinear VMAs there is no correspondence between virtual address
1475 * offset and file offset. So we must perform an exhaustive search
1476 * across *all* the pages in each nonlinear VMA, not just the pages
1477 * whose virtual address lies outside the file truncation point.
1480 list_for_each_entry(vma, head, shared.vm_set.list) {
1481 /* Skip quickly over those we have already dealt with */
1482 if (vma->vm_truncate_count == details->truncate_count)
1484 details->nonlinear_vma = vma;
1485 if (unmap_mapping_range_vma(vma, vma->vm_start,
1486 vma->vm_end, details) < 0)
1492 * unmap_mapping_range - unmap the portion of all mmaps
1493 * in the specified address_space corresponding to the specified
1494 * page range in the underlying file.
1495 * @mapping: the address space containing mmaps to be unmapped.
1496 * @holebegin: byte in first page to unmap, relative to the start of
1497 * the underlying file. This will be rounded down to a PAGE_SIZE
1498 * boundary. Note that this is different from vmtruncate(), which
1499 * must keep the partial page. In contrast, we must get rid of
1501 * @holelen: size of prospective hole in bytes. This will be rounded
1502 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1504 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1505 * but 0 when invalidating pagecache, don't throw away private data.
1507 void unmap_mapping_range(struct address_space *mapping,
1508 loff_t const holebegin, loff_t const holelen, int even_cows)
1510 struct zap_details details;
1511 pgoff_t hba = holebegin >> PAGE_SHIFT;
1512 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1514 /* Check for overflow. */
1515 if (sizeof(holelen) > sizeof(hlen)) {
1517 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1518 if (holeend & ~(long long)ULONG_MAX)
1519 hlen = ULONG_MAX - hba + 1;
1522 details.check_mapping = even_cows? NULL: mapping;
1523 details.nonlinear_vma = NULL;
1524 details.first_index = hba;
1525 details.last_index = hba + hlen - 1;
1526 if (details.last_index < details.first_index)
1527 details.last_index = ULONG_MAX;
1528 details.i_mmap_lock = &mapping->i_mmap_lock;
1530 spin_lock(&mapping->i_mmap_lock);
1532 /* serialize i_size write against truncate_count write */
1534 /* Protect against page faults, and endless unmapping loops */
1535 mapping->truncate_count++;
1537 * For archs where spin_lock has inclusive semantics like ia64
1538 * this smp_mb() will prevent to read pagetable contents
1539 * before the truncate_count increment is visible to
1543 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1544 if (mapping->truncate_count == 0)
1545 reset_vma_truncate_counts(mapping);
1546 mapping->truncate_count++;
1548 details.truncate_count = mapping->truncate_count;
1550 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1551 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1552 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1553 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1554 spin_unlock(&mapping->i_mmap_lock);
1556 EXPORT_SYMBOL(unmap_mapping_range);
1559 * Handle all mappings that got truncated by a "truncate()"
1562 * NOTE! We have to be ready to update the memory sharing
1563 * between the file and the memory map for a potential last
1564 * incomplete page. Ugly, but necessary.
1566 int vmtruncate(struct inode * inode, loff_t offset)
1568 struct address_space *mapping = inode->i_mapping;
1569 unsigned long limit;
1571 if (inode->i_size < offset)
1574 * truncation of in-use swapfiles is disallowed - it would cause
1575 * subsequent swapout to scribble on the now-freed blocks.
1577 if (IS_SWAPFILE(inode))
1579 i_size_write(inode, offset);
1580 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1581 truncate_inode_pages(mapping, offset);
1585 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1586 if (limit != RLIM_INFINITY && offset > limit)
1588 if (offset > inode->i_sb->s_maxbytes)
1590 i_size_write(inode, offset);
1593 if (inode->i_op && inode->i_op->truncate)
1594 inode->i_op->truncate(inode);
1597 send_sig(SIGXFSZ, current, 0);
1604 EXPORT_SYMBOL(vmtruncate);
1607 * Primitive swap readahead code. We simply read an aligned block of
1608 * (1 << page_cluster) entries in the swap area. This method is chosen
1609 * because it doesn't cost us any seek time. We also make sure to queue
1610 * the 'original' request together with the readahead ones...
1612 * This has been extended to use the NUMA policies from the mm triggering
1615 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1617 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1620 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1623 struct page *new_page;
1624 unsigned long offset;
1627 * Get the number of handles we should do readahead io to.
1629 num = valid_swaphandles(entry, &offset);
1630 for (i = 0; i < num; offset++, i++) {
1631 /* Ok, do the async read-ahead now */
1632 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1633 offset), vma, addr);
1636 page_cache_release(new_page);
1639 * Find the next applicable VMA for the NUMA policy.
1645 if (addr >= vma->vm_end) {
1647 next_vma = vma ? vma->vm_next : NULL;
1649 if (vma && addr < vma->vm_start)
1652 if (next_vma && addr >= next_vma->vm_start) {
1654 next_vma = vma->vm_next;
1659 lru_add_drain(); /* Push any new pages onto the LRU now */
1663 * We hold the mm semaphore and the page_table_lock on entry and
1664 * should release the pagetable lock on exit..
1666 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1667 unsigned long address, pte_t *page_table, pmd_t *pmd,
1668 int write_access, pte_t orig_pte)
1673 int ret = VM_FAULT_MINOR;
1675 pte_unmap(page_table);
1676 spin_unlock(&mm->page_table_lock);
1678 entry = pte_to_swp_entry(orig_pte);
1679 page = lookup_swap_cache(entry);
1681 swapin_readahead(entry, address, vma);
1682 page = read_swap_cache_async(entry, vma, address);
1685 * Back out if somebody else faulted in this pte while
1686 * we released the page table lock.
1688 spin_lock(&mm->page_table_lock);
1689 page_table = pte_offset_map(pmd, address);
1690 if (likely(pte_same(*page_table, orig_pte)))
1695 /* Had to read the page from swap area: Major fault */
1696 ret = VM_FAULT_MAJOR;
1697 inc_page_state(pgmajfault);
1701 mark_page_accessed(page);
1705 * Back out if somebody else faulted in this pte while we
1706 * released the page table lock.
1708 spin_lock(&mm->page_table_lock);
1709 page_table = pte_offset_map(pmd, address);
1710 if (unlikely(!pte_same(*page_table, orig_pte)))
1713 if (unlikely(!PageUptodate(page))) {
1714 ret = VM_FAULT_SIGBUS;
1718 /* The page isn't present yet, go ahead with the fault. */
1720 inc_mm_counter(mm, anon_rss);
1721 pte = mk_pte(page, vma->vm_page_prot);
1722 if (write_access && can_share_swap_page(page)) {
1723 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
1727 flush_icache_page(vma, page);
1728 set_pte_at(mm, address, page_table, pte);
1729 page_add_anon_rmap(page, vma, address);
1733 remove_exclusive_swap_page(page);
1737 if (do_wp_page(mm, vma, address,
1738 page_table, pmd, pte) == VM_FAULT_OOM)
1743 /* No need to invalidate - it was non-present before */
1744 update_mmu_cache(vma, address, pte);
1745 lazy_mmu_prot_update(pte);
1747 pte_unmap(page_table);
1748 spin_unlock(&mm->page_table_lock);
1752 pte_unmap(page_table);
1753 spin_unlock(&mm->page_table_lock);
1755 page_cache_release(page);
1760 * We are called with the MM semaphore and page_table_lock
1761 * spinlock held to protect against concurrent faults in
1762 * multithreaded programs.
1764 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
1765 unsigned long address, pte_t *page_table, pmd_t *pmd,
1768 struct page *page = ZERO_PAGE(addr);
1771 /* Mapping of ZERO_PAGE - vm_page_prot is readonly */
1772 entry = mk_pte(page, vma->vm_page_prot);
1775 /* Allocate our own private page. */
1776 pte_unmap(page_table);
1777 spin_unlock(&mm->page_table_lock);
1779 if (unlikely(anon_vma_prepare(vma)))
1781 page = alloc_zeroed_user_highpage(vma, address);
1785 spin_lock(&mm->page_table_lock);
1786 page_table = pte_offset_map(pmd, address);
1788 if (!pte_none(*page_table)) {
1789 page_cache_release(page);
1792 inc_mm_counter(mm, anon_rss);
1793 entry = mk_pte(page, vma->vm_page_prot);
1794 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1795 lru_cache_add_active(page);
1796 SetPageReferenced(page);
1797 page_add_anon_rmap(page, vma, address);
1799 inc_mm_counter(mm, file_rss);
1800 page_add_file_rmap(page);
1801 page_cache_get(page);
1804 set_pte_at(mm, address, page_table, entry);
1806 /* No need to invalidate - it was non-present before */
1807 update_mmu_cache(vma, address, entry);
1808 lazy_mmu_prot_update(entry);
1810 pte_unmap(page_table);
1811 spin_unlock(&mm->page_table_lock);
1812 return VM_FAULT_MINOR;
1814 return VM_FAULT_OOM;
1818 * do_no_page() tries to create a new page mapping. It aggressively
1819 * tries to share with existing pages, but makes a separate copy if
1820 * the "write_access" parameter is true in order to avoid the next
1823 * As this is called only for pages that do not currently exist, we
1824 * do not need to flush old virtual caches or the TLB.
1826 * This is called with the MM semaphore held and the page table
1827 * spinlock held. Exit with the spinlock released.
1829 static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
1830 unsigned long address, pte_t *page_table, pmd_t *pmd,
1833 struct page *new_page;
1834 struct address_space *mapping = NULL;
1836 unsigned int sequence = 0;
1837 int ret = VM_FAULT_MINOR;
1840 pte_unmap(page_table);
1841 spin_unlock(&mm->page_table_lock);
1844 mapping = vma->vm_file->f_mapping;
1845 sequence = mapping->truncate_count;
1846 smp_rmb(); /* serializes i_size against truncate_count */
1849 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
1851 * No smp_rmb is needed here as long as there's a full
1852 * spin_lock/unlock sequence inside the ->nopage callback
1853 * (for the pagecache lookup) that acts as an implicit
1854 * smp_mb() and prevents the i_size read to happen
1855 * after the next truncate_count read.
1858 /* no page was available -- either SIGBUS or OOM */
1859 if (new_page == NOPAGE_SIGBUS)
1860 return VM_FAULT_SIGBUS;
1861 if (new_page == NOPAGE_OOM)
1862 return VM_FAULT_OOM;
1865 * Should we do an early C-O-W break?
1867 if (write_access && !(vma->vm_flags & VM_SHARED)) {
1870 if (unlikely(anon_vma_prepare(vma)))
1872 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1875 copy_user_highpage(page, new_page, address);
1876 page_cache_release(new_page);
1881 spin_lock(&mm->page_table_lock);
1883 * For a file-backed vma, someone could have truncated or otherwise
1884 * invalidated this page. If unmap_mapping_range got called,
1885 * retry getting the page.
1887 if (mapping && unlikely(sequence != mapping->truncate_count)) {
1888 spin_unlock(&mm->page_table_lock);
1889 page_cache_release(new_page);
1891 sequence = mapping->truncate_count;
1895 page_table = pte_offset_map(pmd, address);
1898 * This silly early PAGE_DIRTY setting removes a race
1899 * due to the bad i386 page protection. But it's valid
1900 * for other architectures too.
1902 * Note that if write_access is true, we either now have
1903 * an exclusive copy of the page, or this is a shared mapping,
1904 * so we can make it writable and dirty to avoid having to
1905 * handle that later.
1907 /* Only go through if we didn't race with anybody else... */
1908 if (pte_none(*page_table)) {
1909 flush_icache_page(vma, new_page);
1910 entry = mk_pte(new_page, vma->vm_page_prot);
1912 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1913 set_pte_at(mm, address, page_table, entry);
1915 inc_mm_counter(mm, anon_rss);
1916 lru_cache_add_active(new_page);
1917 page_add_anon_rmap(new_page, vma, address);
1918 } else if (!(vma->vm_flags & VM_RESERVED)) {
1919 inc_mm_counter(mm, file_rss);
1920 page_add_file_rmap(new_page);
1923 /* One of our sibling threads was faster, back out. */
1924 page_cache_release(new_page);
1928 /* no need to invalidate: a not-present page shouldn't be cached */
1929 update_mmu_cache(vma, address, entry);
1930 lazy_mmu_prot_update(entry);
1932 pte_unmap(page_table);
1933 spin_unlock(&mm->page_table_lock);
1936 page_cache_release(new_page);
1937 return VM_FAULT_OOM;
1941 * Fault of a previously existing named mapping. Repopulate the pte
1942 * from the encoded file_pte if possible. This enables swappable
1945 static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
1946 unsigned long address, pte_t *page_table, pmd_t *pmd,
1947 int write_access, pte_t orig_pte)
1952 pte_unmap(page_table);
1953 spin_unlock(&mm->page_table_lock);
1955 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
1957 * Page table corrupted: show pte and kill process.
1959 print_bad_pte(vma, orig_pte, address);
1960 return VM_FAULT_OOM;
1962 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
1964 pgoff = pte_to_pgoff(orig_pte);
1965 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
1966 vma->vm_page_prot, pgoff, 0);
1968 return VM_FAULT_OOM;
1970 return VM_FAULT_SIGBUS;
1971 return VM_FAULT_MAJOR;
1975 * These routines also need to handle stuff like marking pages dirty
1976 * and/or accessed for architectures that don't do it in hardware (most
1977 * RISC architectures). The early dirtying is also good on the i386.
1979 * There is also a hook called "update_mmu_cache()" that architectures
1980 * with external mmu caches can use to update those (ie the Sparc or
1981 * PowerPC hashed page tables that act as extended TLBs).
1983 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1984 * but allow concurrent faults), and pte mapped but not yet locked.
1985 * We return with mmap_sem still held, but pte unmapped and unlocked.
1987 static inline int handle_pte_fault(struct mm_struct *mm,
1988 struct vm_area_struct *vma, unsigned long address,
1989 pte_t *pte, pmd_t *pmd, int write_access)
1993 spin_lock(&mm->page_table_lock);
1995 if (!pte_present(entry)) {
1996 if (pte_none(entry)) {
1997 if (!vma->vm_ops || !vma->vm_ops->nopage)
1998 return do_anonymous_page(mm, vma, address,
1999 pte, pmd, write_access);
2000 return do_no_page(mm, vma, address,
2001 pte, pmd, write_access);
2003 if (pte_file(entry))
2004 return do_file_page(mm, vma, address,
2005 pte, pmd, write_access, entry);
2006 return do_swap_page(mm, vma, address,
2007 pte, pmd, write_access, entry);
2011 if (!pte_write(entry))
2012 return do_wp_page(mm, vma, address, pte, pmd, entry);
2013 entry = pte_mkdirty(entry);
2015 entry = pte_mkyoung(entry);
2016 ptep_set_access_flags(vma, address, pte, entry, write_access);
2017 update_mmu_cache(vma, address, entry);
2018 lazy_mmu_prot_update(entry);
2020 spin_unlock(&mm->page_table_lock);
2021 return VM_FAULT_MINOR;
2025 * By the time we get here, we already hold the mm semaphore
2027 int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2028 unsigned long address, int write_access)
2035 __set_current_state(TASK_RUNNING);
2037 inc_page_state(pgfault);
2039 if (unlikely(is_vm_hugetlb_page(vma)))
2040 return hugetlb_fault(mm, vma, address, write_access);
2042 pgd = pgd_offset(mm, address);
2043 pud = pud_alloc(mm, pgd, address);
2045 return VM_FAULT_OOM;
2046 pmd = pmd_alloc(mm, pud, address);
2048 return VM_FAULT_OOM;
2049 pte = pte_alloc_map(mm, pmd, address);
2051 return VM_FAULT_OOM;
2053 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2056 #ifndef __PAGETABLE_PUD_FOLDED
2058 * Allocate page upper directory.
2059 * We've already handled the fast-path in-line.
2061 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2063 pud_t *new = pud_alloc_one(mm, address);
2067 spin_lock(&mm->page_table_lock);
2068 if (pgd_present(*pgd)) /* Another has populated it */
2071 pgd_populate(mm, pgd, new);
2072 spin_unlock(&mm->page_table_lock);
2075 #endif /* __PAGETABLE_PUD_FOLDED */
2077 #ifndef __PAGETABLE_PMD_FOLDED
2079 * Allocate page middle directory.
2080 * We've already handled the fast-path in-line.
2082 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2084 pmd_t *new = pmd_alloc_one(mm, address);
2088 spin_lock(&mm->page_table_lock);
2089 #ifndef __ARCH_HAS_4LEVEL_HACK
2090 if (pud_present(*pud)) /* Another has populated it */
2093 pud_populate(mm, pud, new);
2095 if (pgd_present(*pud)) /* Another has populated it */
2098 pgd_populate(mm, pud, new);
2099 #endif /* __ARCH_HAS_4LEVEL_HACK */
2100 spin_unlock(&mm->page_table_lock);
2103 #endif /* __PAGETABLE_PMD_FOLDED */
2105 int make_pages_present(unsigned long addr, unsigned long end)
2107 int ret, len, write;
2108 struct vm_area_struct * vma;
2110 vma = find_vma(current->mm, addr);
2113 write = (vma->vm_flags & VM_WRITE) != 0;
2116 if (end > vma->vm_end)
2118 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2119 ret = get_user_pages(current, current->mm, addr,
2120 len, write, 0, NULL, NULL);
2123 return ret == len ? 0 : -1;
2127 * Map a vmalloc()-space virtual address to the physical page.
2129 struct page * vmalloc_to_page(void * vmalloc_addr)
2131 unsigned long addr = (unsigned long) vmalloc_addr;
2132 struct page *page = NULL;
2133 pgd_t *pgd = pgd_offset_k(addr);
2138 if (!pgd_none(*pgd)) {
2139 pud = pud_offset(pgd, addr);
2140 if (!pud_none(*pud)) {
2141 pmd = pmd_offset(pud, addr);
2142 if (!pmd_none(*pmd)) {
2143 ptep = pte_offset_map(pmd, addr);
2145 if (pte_present(pte))
2146 page = pte_page(pte);
2154 EXPORT_SYMBOL(vmalloc_to_page);
2157 * Map a vmalloc()-space virtual address to the physical page frame number.
2159 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2161 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2164 EXPORT_SYMBOL(vmalloc_to_pfn);
2166 #if !defined(__HAVE_ARCH_GATE_AREA)
2168 #if defined(AT_SYSINFO_EHDR)
2169 static struct vm_area_struct gate_vma;
2171 static int __init gate_vma_init(void)
2173 gate_vma.vm_mm = NULL;
2174 gate_vma.vm_start = FIXADDR_USER_START;
2175 gate_vma.vm_end = FIXADDR_USER_END;
2176 gate_vma.vm_page_prot = PAGE_READONLY;
2177 gate_vma.vm_flags = VM_RESERVED;
2180 __initcall(gate_vma_init);
2183 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2185 #ifdef AT_SYSINFO_EHDR
2192 int in_gate_area_no_task(unsigned long addr)
2194 #ifdef AT_SYSINFO_EHDR
2195 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2201 #endif /* __HAVE_ARCH_GATE_AREA */