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)
287 spin_unlock(&mm->page_table_lock);
288 new = pte_alloc_one(mm, address);
289 spin_lock(&mm->page_table_lock);
293 if (pmd_present(*pmd)) /* Another has populated it */
297 inc_page_state(nr_page_table_pages);
298 pmd_populate(mm, pmd, new);
303 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
305 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
309 spin_lock(&init_mm.page_table_lock);
310 if (pmd_present(*pmd)) /* Another has populated it */
311 pte_free_kernel(new);
313 pmd_populate_kernel(&init_mm, pmd, new);
314 spin_unlock(&init_mm.page_table_lock);
318 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
321 add_mm_counter(mm, file_rss, file_rss);
323 add_mm_counter(mm, anon_rss, anon_rss);
327 * This function is called to print an error when a pte in a
328 * !VM_RESERVED region is found pointing to an invalid pfn (which
331 * The calling function must still handle the error.
333 void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
335 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
336 "vm_flags = %lx, vaddr = %lx\n",
337 (long long)pte_val(pte),
338 (vma->vm_mm == current->mm ? current->comm : "???"),
339 vma->vm_flags, vaddr);
344 * copy one vm_area from one task to the other. Assumes the page tables
345 * already present in the new task to be cleared in the whole range
346 * covered by this vma.
348 * dst->page_table_lock is held on entry and exit,
349 * but may be dropped within p[mg]d_alloc() and pte_alloc_map().
353 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
354 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
355 unsigned long addr, int *rss)
357 unsigned long vm_flags = vma->vm_flags;
358 pte_t pte = *src_pte;
362 /* pte contains position in swap or file, so copy. */
363 if (unlikely(!pte_present(pte))) {
364 if (!pte_file(pte)) {
365 swap_duplicate(pte_to_swp_entry(pte));
366 /* make sure dst_mm is on swapoff's mmlist. */
367 if (unlikely(list_empty(&dst_mm->mmlist))) {
368 spin_lock(&mmlist_lock);
369 list_add(&dst_mm->mmlist, &src_mm->mmlist);
370 spin_unlock(&mmlist_lock);
376 /* If the region is VM_RESERVED, the mapping is not
377 * mapped via rmap - duplicate the pte as is.
379 if (vm_flags & VM_RESERVED)
383 /* If the pte points outside of valid memory but
384 * the region is not VM_RESERVED, we have a problem.
386 if (unlikely(!pfn_valid(pfn))) {
387 print_bad_pte(vma, pte, addr);
388 goto out_set_pte; /* try to do something sane */
391 page = pfn_to_page(pfn);
394 * If it's a COW mapping, write protect it both
395 * in the parent and the child
397 if ((vm_flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE) {
398 ptep_set_wrprotect(src_mm, addr, src_pte);
403 * If it's a shared mapping, mark it clean in
406 if (vm_flags & VM_SHARED)
407 pte = pte_mkclean(pte);
408 pte = pte_mkold(pte);
411 rss[!!PageAnon(page)]++;
414 set_pte_at(dst_mm, addr, dst_pte, pte);
417 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
418 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
419 unsigned long addr, unsigned long end)
421 pte_t *src_pte, *dst_pte;
427 dst_pte = pte_alloc_map(dst_mm, dst_pmd, addr);
430 src_pte = pte_offset_map_nested(src_pmd, addr);
432 spin_lock(&src_mm->page_table_lock);
435 * We are holding two locks at this point - either of them
436 * could generate latencies in another task on another CPU.
438 if (progress >= 32) {
440 if (need_resched() ||
441 need_lockbreak(&src_mm->page_table_lock) ||
442 need_lockbreak(&dst_mm->page_table_lock))
445 if (pte_none(*src_pte)) {
449 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
451 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
452 spin_unlock(&src_mm->page_table_lock);
454 pte_unmap_nested(src_pte - 1);
455 pte_unmap(dst_pte - 1);
456 add_mm_rss(dst_mm, rss[0], rss[1]);
457 cond_resched_lock(&dst_mm->page_table_lock);
463 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
464 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
465 unsigned long addr, unsigned long end)
467 pmd_t *src_pmd, *dst_pmd;
470 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
473 src_pmd = pmd_offset(src_pud, addr);
475 next = pmd_addr_end(addr, end);
476 if (pmd_none_or_clear_bad(src_pmd))
478 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
481 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
485 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
486 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
487 unsigned long addr, unsigned long end)
489 pud_t *src_pud, *dst_pud;
492 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
495 src_pud = pud_offset(src_pgd, addr);
497 next = pud_addr_end(addr, end);
498 if (pud_none_or_clear_bad(src_pud))
500 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
503 } while (dst_pud++, src_pud++, addr = next, addr != end);
507 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
508 struct vm_area_struct *vma)
510 pgd_t *src_pgd, *dst_pgd;
512 unsigned long addr = vma->vm_start;
513 unsigned long end = vma->vm_end;
516 * Don't copy ptes where a page fault will fill them correctly.
517 * Fork becomes much lighter when there are big shared or private
518 * readonly mappings. The tradeoff is that copy_page_range is more
519 * efficient than faulting.
521 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_RESERVED))) {
526 if (is_vm_hugetlb_page(vma))
527 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
529 dst_pgd = pgd_offset(dst_mm, addr);
530 src_pgd = pgd_offset(src_mm, addr);
532 next = pgd_addr_end(addr, end);
533 if (pgd_none_or_clear_bad(src_pgd))
535 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
538 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
542 static void zap_pte_range(struct mmu_gather *tlb,
543 struct vm_area_struct *vma, pmd_t *pmd,
544 unsigned long addr, unsigned long end,
545 struct zap_details *details)
547 struct mm_struct *mm = tlb->mm;
552 pte = pte_offset_map(pmd, addr);
557 if (pte_present(ptent)) {
558 struct page *page = NULL;
559 if (!(vma->vm_flags & VM_RESERVED)) {
560 unsigned long pfn = pte_pfn(ptent);
561 if (unlikely(!pfn_valid(pfn)))
562 print_bad_pte(vma, ptent, addr);
564 page = pfn_to_page(pfn);
566 if (unlikely(details) && page) {
568 * unmap_shared_mapping_pages() wants to
569 * invalidate cache without truncating:
570 * unmap shared but keep private pages.
572 if (details->check_mapping &&
573 details->check_mapping != page->mapping)
576 * Each page->index must be checked when
577 * invalidating or truncating nonlinear.
579 if (details->nonlinear_vma &&
580 (page->index < details->first_index ||
581 page->index > details->last_index))
584 ptent = ptep_get_and_clear_full(mm, addr, pte,
586 tlb_remove_tlb_entry(tlb, pte, addr);
589 if (unlikely(details) && details->nonlinear_vma
590 && linear_page_index(details->nonlinear_vma,
591 addr) != page->index)
592 set_pte_at(mm, addr, pte,
593 pgoff_to_pte(page->index));
597 if (pte_dirty(ptent))
598 set_page_dirty(page);
599 if (pte_young(ptent))
600 mark_page_accessed(page);
603 page_remove_rmap(page);
604 tlb_remove_page(tlb, page);
608 * If details->check_mapping, we leave swap entries;
609 * if details->nonlinear_vma, we leave file entries.
611 if (unlikely(details))
613 if (!pte_file(ptent))
614 free_swap_and_cache(pte_to_swp_entry(ptent));
615 pte_clear_full(mm, addr, pte, tlb->fullmm);
616 } while (pte++, addr += PAGE_SIZE, addr != end);
618 add_mm_rss(mm, file_rss, anon_rss);
622 static inline void zap_pmd_range(struct mmu_gather *tlb,
623 struct vm_area_struct *vma, pud_t *pud,
624 unsigned long addr, unsigned long end,
625 struct zap_details *details)
630 pmd = pmd_offset(pud, addr);
632 next = pmd_addr_end(addr, end);
633 if (pmd_none_or_clear_bad(pmd))
635 zap_pte_range(tlb, vma, pmd, addr, next, details);
636 } while (pmd++, addr = next, addr != end);
639 static inline void zap_pud_range(struct mmu_gather *tlb,
640 struct vm_area_struct *vma, pgd_t *pgd,
641 unsigned long addr, unsigned long end,
642 struct zap_details *details)
647 pud = pud_offset(pgd, addr);
649 next = pud_addr_end(addr, end);
650 if (pud_none_or_clear_bad(pud))
652 zap_pmd_range(tlb, vma, pud, addr, next, details);
653 } while (pud++, addr = next, addr != end);
656 static void unmap_page_range(struct mmu_gather *tlb, struct vm_area_struct *vma,
657 unsigned long addr, unsigned long end,
658 struct zap_details *details)
663 if (details && !details->check_mapping && !details->nonlinear_vma)
667 tlb_start_vma(tlb, vma);
668 pgd = pgd_offset(vma->vm_mm, addr);
670 next = pgd_addr_end(addr, end);
671 if (pgd_none_or_clear_bad(pgd))
673 zap_pud_range(tlb, vma, pgd, addr, next, details);
674 } while (pgd++, addr = next, addr != end);
675 tlb_end_vma(tlb, vma);
678 #ifdef CONFIG_PREEMPT
679 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
681 /* No preempt: go for improved straight-line efficiency */
682 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
686 * unmap_vmas - unmap a range of memory covered by a list of vma's
687 * @tlbp: address of the caller's struct mmu_gather
688 * @mm: the controlling mm_struct
689 * @vma: the starting vma
690 * @start_addr: virtual address at which to start unmapping
691 * @end_addr: virtual address at which to end unmapping
692 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
693 * @details: details of nonlinear truncation or shared cache invalidation
695 * Returns the end address of the unmapping (restart addr if interrupted).
697 * Unmap all pages in the vma list. Called under page_table_lock.
699 * We aim to not hold page_table_lock for too long (for scheduling latency
700 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
701 * return the ending mmu_gather to the caller.
703 * Only addresses between `start' and `end' will be unmapped.
705 * The VMA list must be sorted in ascending virtual address order.
707 * unmap_vmas() assumes that the caller will flush the whole unmapped address
708 * range after unmap_vmas() returns. So the only responsibility here is to
709 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
710 * drops the lock and schedules.
712 unsigned long unmap_vmas(struct mmu_gather **tlbp, struct mm_struct *mm,
713 struct vm_area_struct *vma, unsigned long start_addr,
714 unsigned long end_addr, unsigned long *nr_accounted,
715 struct zap_details *details)
717 unsigned long zap_bytes = ZAP_BLOCK_SIZE;
718 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
719 int tlb_start_valid = 0;
720 unsigned long start = start_addr;
721 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
722 int fullmm = (*tlbp)->fullmm;
724 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
727 start = max(vma->vm_start, start_addr);
728 if (start >= vma->vm_end)
730 end = min(vma->vm_end, end_addr);
731 if (end <= vma->vm_start)
734 if (vma->vm_flags & VM_ACCOUNT)
735 *nr_accounted += (end - start) >> PAGE_SHIFT;
737 while (start != end) {
740 if (!tlb_start_valid) {
745 if (is_vm_hugetlb_page(vma)) {
747 unmap_hugepage_range(vma, start, end);
749 block = min(zap_bytes, end - start);
750 unmap_page_range(*tlbp, vma, start,
751 start + block, details);
756 if ((long)zap_bytes > 0)
759 tlb_finish_mmu(*tlbp, tlb_start, start);
761 if (need_resched() ||
762 need_lockbreak(&mm->page_table_lock) ||
763 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
765 /* must reset count of rss freed */
766 *tlbp = tlb_gather_mmu(mm, fullmm);
769 spin_unlock(&mm->page_table_lock);
771 spin_lock(&mm->page_table_lock);
774 *tlbp = tlb_gather_mmu(mm, fullmm);
776 zap_bytes = ZAP_BLOCK_SIZE;
780 return start; /* which is now the end (or restart) address */
784 * zap_page_range - remove user pages in a given range
785 * @vma: vm_area_struct holding the applicable pages
786 * @address: starting address of pages to zap
787 * @size: number of bytes to zap
788 * @details: details of nonlinear truncation or shared cache invalidation
790 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
791 unsigned long size, struct zap_details *details)
793 struct mm_struct *mm = vma->vm_mm;
794 struct mmu_gather *tlb;
795 unsigned long end = address + size;
796 unsigned long nr_accounted = 0;
798 if (is_vm_hugetlb_page(vma)) {
799 zap_hugepage_range(vma, address, size);
804 spin_lock(&mm->page_table_lock);
805 tlb = tlb_gather_mmu(mm, 0);
806 update_hiwater_rss(mm);
807 end = unmap_vmas(&tlb, mm, vma, address, end, &nr_accounted, details);
808 tlb_finish_mmu(tlb, address, end);
809 spin_unlock(&mm->page_table_lock);
814 * Do a quick page-table lookup for a single page.
815 * mm->page_table_lock must be held.
817 static struct page *__follow_page(struct mm_struct *mm, unsigned long address,
818 int read, int write, int accessed)
827 page = follow_huge_addr(mm, address, write);
831 pgd = pgd_offset(mm, address);
832 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
835 pud = pud_offset(pgd, address);
836 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
839 pmd = pmd_offset(pud, address);
840 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
843 return follow_huge_pmd(mm, address, pmd, write);
845 ptep = pte_offset_map(pmd, address);
851 if (pte_present(pte)) {
852 if (write && !pte_write(pte))
854 if (read && !pte_read(pte))
857 if (pfn_valid(pfn)) {
858 page = pfn_to_page(pfn);
860 if (write && !pte_dirty(pte) &&!PageDirty(page))
861 set_page_dirty(page);
862 mark_page_accessed(page);
873 follow_page(struct mm_struct *mm, unsigned long address, int write)
875 return __follow_page(mm, address, 0, write, 1);
879 * check_user_page_readable() can be called frm niterrupt context by oprofile,
880 * so we need to avoid taking any non-irq-safe locks
882 int check_user_page_readable(struct mm_struct *mm, unsigned long address)
884 return __follow_page(mm, address, 1, 0, 0) != NULL;
886 EXPORT_SYMBOL(check_user_page_readable);
889 untouched_anonymous_page(struct mm_struct* mm, struct vm_area_struct *vma,
890 unsigned long address)
896 /* Check if the vma is for an anonymous mapping. */
897 if (vma->vm_ops && vma->vm_ops->nopage)
900 /* Check if page directory entry exists. */
901 pgd = pgd_offset(mm, address);
902 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
905 pud = pud_offset(pgd, address);
906 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
909 /* Check if page middle directory entry exists. */
910 pmd = pmd_offset(pud, address);
911 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
914 /* There is a pte slot for 'address' in 'mm'. */
918 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
919 unsigned long start, int len, int write, int force,
920 struct page **pages, struct vm_area_struct **vmas)
926 * Require read or write permissions.
927 * If 'force' is set, we only require the "MAY" flags.
929 flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
930 flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
934 struct vm_area_struct * vma;
936 vma = find_extend_vma(mm, start);
937 if (!vma && in_gate_area(tsk, start)) {
938 unsigned long pg = start & PAGE_MASK;
939 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
944 if (write) /* user gate pages are read-only */
945 return i ? : -EFAULT;
947 pgd = pgd_offset_k(pg);
949 pgd = pgd_offset_gate(mm, pg);
950 BUG_ON(pgd_none(*pgd));
951 pud = pud_offset(pgd, pg);
952 BUG_ON(pud_none(*pud));
953 pmd = pmd_offset(pud, pg);
955 return i ? : -EFAULT;
956 pte = pte_offset_map(pmd, pg);
957 if (pte_none(*pte)) {
959 return i ? : -EFAULT;
962 pages[i] = pte_page(*pte);
974 if (!vma || (vma->vm_flags & (VM_IO | VM_RESERVED))
975 || !(flags & vma->vm_flags))
976 return i ? : -EFAULT;
978 if (is_vm_hugetlb_page(vma)) {
979 i = follow_hugetlb_page(mm, vma, pages, vmas,
983 spin_lock(&mm->page_table_lock);
985 int write_access = write;
988 cond_resched_lock(&mm->page_table_lock);
989 while (!(page = follow_page(mm, start, write_access))) {
993 * Shortcut for anonymous pages. We don't want
994 * to force the creation of pages tables for
995 * insanely big anonymously mapped areas that
996 * nobody touched so far. This is important
997 * for doing a core dump for these mappings.
999 if (!write && untouched_anonymous_page(mm,vma,start)) {
1000 page = ZERO_PAGE(start);
1003 spin_unlock(&mm->page_table_lock);
1004 ret = __handle_mm_fault(mm, vma, start, write_access);
1007 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1008 * broken COW when necessary, even if maybe_mkwrite
1009 * decided not to set pte_write. We can thus safely do
1010 * subsequent page lookups as if they were reads.
1012 if (ret & VM_FAULT_WRITE)
1015 switch (ret & ~VM_FAULT_WRITE) {
1016 case VM_FAULT_MINOR:
1019 case VM_FAULT_MAJOR:
1022 case VM_FAULT_SIGBUS:
1023 return i ? i : -EFAULT;
1025 return i ? i : -ENOMEM;
1029 spin_lock(&mm->page_table_lock);
1033 flush_dcache_page(page);
1034 page_cache_get(page);
1041 } while (len && start < vma->vm_end);
1042 spin_unlock(&mm->page_table_lock);
1046 EXPORT_SYMBOL(get_user_pages);
1048 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1049 unsigned long addr, unsigned long end, pgprot_t prot)
1053 pte = pte_alloc_map(mm, pmd, addr);
1057 struct page *page = ZERO_PAGE(addr);
1058 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1059 page_cache_get(page);
1060 page_add_file_rmap(page);
1061 inc_mm_counter(mm, file_rss);
1062 BUG_ON(!pte_none(*pte));
1063 set_pte_at(mm, addr, pte, zero_pte);
1064 } while (pte++, addr += PAGE_SIZE, addr != end);
1069 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1070 unsigned long addr, unsigned long end, pgprot_t prot)
1075 pmd = pmd_alloc(mm, pud, addr);
1079 next = pmd_addr_end(addr, end);
1080 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1082 } while (pmd++, addr = next, addr != end);
1086 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1087 unsigned long addr, unsigned long end, pgprot_t prot)
1092 pud = pud_alloc(mm, pgd, addr);
1096 next = pud_addr_end(addr, end);
1097 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1099 } while (pud++, addr = next, addr != end);
1103 int zeromap_page_range(struct vm_area_struct *vma,
1104 unsigned long addr, unsigned long size, pgprot_t prot)
1108 unsigned long end = addr + size;
1109 struct mm_struct *mm = vma->vm_mm;
1112 BUG_ON(addr >= end);
1113 pgd = pgd_offset(mm, addr);
1114 flush_cache_range(vma, addr, end);
1115 spin_lock(&mm->page_table_lock);
1117 next = pgd_addr_end(addr, end);
1118 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1121 } while (pgd++, addr = next, addr != end);
1122 spin_unlock(&mm->page_table_lock);
1127 * maps a range of physical memory into the requested pages. the old
1128 * mappings are removed. any references to nonexistent pages results
1129 * in null mappings (currently treated as "copy-on-access")
1131 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1132 unsigned long addr, unsigned long end,
1133 unsigned long pfn, pgprot_t prot)
1137 pte = pte_alloc_map(mm, pmd, addr);
1141 BUG_ON(!pte_none(*pte));
1142 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1144 } while (pte++, addr += PAGE_SIZE, addr != end);
1149 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1150 unsigned long addr, unsigned long end,
1151 unsigned long pfn, pgprot_t prot)
1156 pfn -= addr >> PAGE_SHIFT;
1157 pmd = pmd_alloc(mm, pud, addr);
1161 next = pmd_addr_end(addr, end);
1162 if (remap_pte_range(mm, pmd, addr, next,
1163 pfn + (addr >> PAGE_SHIFT), prot))
1165 } while (pmd++, addr = next, addr != end);
1169 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1170 unsigned long addr, unsigned long end,
1171 unsigned long pfn, pgprot_t prot)
1176 pfn -= addr >> PAGE_SHIFT;
1177 pud = pud_alloc(mm, pgd, addr);
1181 next = pud_addr_end(addr, end);
1182 if (remap_pmd_range(mm, pud, addr, next,
1183 pfn + (addr >> PAGE_SHIFT), prot))
1185 } while (pud++, addr = next, addr != end);
1189 /* Note: this is only safe if the mm semaphore is held when called. */
1190 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1191 unsigned long pfn, unsigned long size, pgprot_t prot)
1195 unsigned long end = addr + PAGE_ALIGN(size);
1196 struct mm_struct *mm = vma->vm_mm;
1200 * Physically remapped pages are special. Tell the
1201 * rest of the world about it:
1202 * VM_IO tells people not to look at these pages
1203 * (accesses can have side effects).
1204 * VM_RESERVED tells the core MM not to "manage" these pages
1205 * (e.g. refcount, mapcount, try to swap them out).
1207 vma->vm_flags |= VM_IO | VM_RESERVED;
1209 BUG_ON(addr >= end);
1210 pfn -= addr >> PAGE_SHIFT;
1211 pgd = pgd_offset(mm, addr);
1212 flush_cache_range(vma, addr, end);
1213 spin_lock(&mm->page_table_lock);
1215 next = pgd_addr_end(addr, end);
1216 err = remap_pud_range(mm, pgd, addr, next,
1217 pfn + (addr >> PAGE_SHIFT), prot);
1220 } while (pgd++, addr = next, addr != end);
1221 spin_unlock(&mm->page_table_lock);
1224 EXPORT_SYMBOL(remap_pfn_range);
1227 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1228 * servicing faults for write access. In the normal case, do always want
1229 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1230 * that do not have writing enabled, when used by access_process_vm.
1232 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1234 if (likely(vma->vm_flags & VM_WRITE))
1235 pte = pte_mkwrite(pte);
1240 * This routine handles present pages, when users try to write
1241 * to a shared page. It is done by copying the page to a new address
1242 * and decrementing the shared-page counter for the old page.
1244 * Note that this routine assumes that the protection checks have been
1245 * done by the caller (the low-level page fault routine in most cases).
1246 * Thus we can safely just mark it writable once we've done any necessary
1249 * We also mark the page dirty at this point even though the page will
1250 * change only once the write actually happens. This avoids a few races,
1251 * and potentially makes it more efficient.
1253 * We hold the mm semaphore and the page_table_lock on entry and exit
1254 * with the page_table_lock released.
1256 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1257 unsigned long address, pte_t *page_table, pmd_t *pmd,
1260 struct page *old_page, *new_page;
1261 unsigned long pfn = pte_pfn(orig_pte);
1263 int ret = VM_FAULT_MINOR;
1265 BUG_ON(vma->vm_flags & VM_RESERVED);
1267 if (unlikely(!pfn_valid(pfn))) {
1269 * Page table corrupted: show pte and kill process.
1271 print_bad_pte(vma, orig_pte, address);
1275 old_page = pfn_to_page(pfn);
1277 if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
1278 int reuse = can_share_swap_page(old_page);
1279 unlock_page(old_page);
1281 flush_cache_page(vma, address, pfn);
1282 entry = pte_mkyoung(orig_pte);
1283 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1284 ptep_set_access_flags(vma, address, page_table, entry, 1);
1285 update_mmu_cache(vma, address, entry);
1286 lazy_mmu_prot_update(entry);
1287 ret |= VM_FAULT_WRITE;
1293 * Ok, we need to copy. Oh, well..
1295 page_cache_get(old_page);
1296 pte_unmap(page_table);
1297 spin_unlock(&mm->page_table_lock);
1299 if (unlikely(anon_vma_prepare(vma)))
1301 if (old_page == ZERO_PAGE(address)) {
1302 new_page = alloc_zeroed_user_highpage(vma, address);
1306 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1309 copy_user_highpage(new_page, old_page, address);
1313 * Re-check the pte - we dropped the lock
1315 spin_lock(&mm->page_table_lock);
1316 page_table = pte_offset_map(pmd, address);
1317 if (likely(pte_same(*page_table, orig_pte))) {
1318 page_remove_rmap(old_page);
1319 if (!PageAnon(old_page)) {
1320 inc_mm_counter(mm, anon_rss);
1321 dec_mm_counter(mm, file_rss);
1323 flush_cache_page(vma, address, pfn);
1324 entry = mk_pte(new_page, vma->vm_page_prot);
1325 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1326 ptep_establish(vma, address, page_table, entry);
1327 update_mmu_cache(vma, address, entry);
1328 lazy_mmu_prot_update(entry);
1330 lru_cache_add_active(new_page);
1331 page_add_anon_rmap(new_page, vma, address);
1333 /* Free the old page.. */
1334 new_page = old_page;
1335 ret |= VM_FAULT_WRITE;
1337 page_cache_release(new_page);
1338 page_cache_release(old_page);
1340 pte_unmap(page_table);
1341 spin_unlock(&mm->page_table_lock);
1344 page_cache_release(old_page);
1345 return VM_FAULT_OOM;
1349 * Helper functions for unmap_mapping_range().
1351 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1353 * We have to restart searching the prio_tree whenever we drop the lock,
1354 * since the iterator is only valid while the lock is held, and anyway
1355 * a later vma might be split and reinserted earlier while lock dropped.
1357 * The list of nonlinear vmas could be handled more efficiently, using
1358 * a placeholder, but handle it in the same way until a need is shown.
1359 * It is important to search the prio_tree before nonlinear list: a vma
1360 * may become nonlinear and be shifted from prio_tree to nonlinear list
1361 * while the lock is dropped; but never shifted from list to prio_tree.
1363 * In order to make forward progress despite restarting the search,
1364 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1365 * quickly skip it next time around. Since the prio_tree search only
1366 * shows us those vmas affected by unmapping the range in question, we
1367 * can't efficiently keep all vmas in step with mapping->truncate_count:
1368 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1369 * mapping->truncate_count and vma->vm_truncate_count are protected by
1372 * In order to make forward progress despite repeatedly restarting some
1373 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1374 * and restart from that address when we reach that vma again. It might
1375 * have been split or merged, shrunk or extended, but never shifted: so
1376 * restart_addr remains valid so long as it remains in the vma's range.
1377 * unmap_mapping_range forces truncate_count to leap over page-aligned
1378 * values so we can save vma's restart_addr in its truncate_count field.
1380 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1382 static void reset_vma_truncate_counts(struct address_space *mapping)
1384 struct vm_area_struct *vma;
1385 struct prio_tree_iter iter;
1387 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1388 vma->vm_truncate_count = 0;
1389 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1390 vma->vm_truncate_count = 0;
1393 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1394 unsigned long start_addr, unsigned long end_addr,
1395 struct zap_details *details)
1397 unsigned long restart_addr;
1401 restart_addr = vma->vm_truncate_count;
1402 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1403 start_addr = restart_addr;
1404 if (start_addr >= end_addr) {
1405 /* Top of vma has been split off since last time */
1406 vma->vm_truncate_count = details->truncate_count;
1411 restart_addr = zap_page_range(vma, start_addr,
1412 end_addr - start_addr, details);
1415 * We cannot rely on the break test in unmap_vmas:
1416 * on the one hand, we don't want to restart our loop
1417 * just because that broke out for the page_table_lock;
1418 * on the other hand, it does no test when vma is small.
1420 need_break = need_resched() ||
1421 need_lockbreak(details->i_mmap_lock);
1423 if (restart_addr >= end_addr) {
1424 /* We have now completed this vma: mark it so */
1425 vma->vm_truncate_count = details->truncate_count;
1429 /* Note restart_addr in vma's truncate_count field */
1430 vma->vm_truncate_count = restart_addr;
1435 spin_unlock(details->i_mmap_lock);
1437 spin_lock(details->i_mmap_lock);
1441 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1442 struct zap_details *details)
1444 struct vm_area_struct *vma;
1445 struct prio_tree_iter iter;
1446 pgoff_t vba, vea, zba, zea;
1449 vma_prio_tree_foreach(vma, &iter, root,
1450 details->first_index, details->last_index) {
1451 /* Skip quickly over those we have already dealt with */
1452 if (vma->vm_truncate_count == details->truncate_count)
1455 vba = vma->vm_pgoff;
1456 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1457 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1458 zba = details->first_index;
1461 zea = details->last_index;
1465 if (unmap_mapping_range_vma(vma,
1466 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1467 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1473 static inline void unmap_mapping_range_list(struct list_head *head,
1474 struct zap_details *details)
1476 struct vm_area_struct *vma;
1479 * In nonlinear VMAs there is no correspondence between virtual address
1480 * offset and file offset. So we must perform an exhaustive search
1481 * across *all* the pages in each nonlinear VMA, not just the pages
1482 * whose virtual address lies outside the file truncation point.
1485 list_for_each_entry(vma, head, shared.vm_set.list) {
1486 /* Skip quickly over those we have already dealt with */
1487 if (vma->vm_truncate_count == details->truncate_count)
1489 details->nonlinear_vma = vma;
1490 if (unmap_mapping_range_vma(vma, vma->vm_start,
1491 vma->vm_end, details) < 0)
1497 * unmap_mapping_range - unmap the portion of all mmaps
1498 * in the specified address_space corresponding to the specified
1499 * page range in the underlying file.
1500 * @mapping: the address space containing mmaps to be unmapped.
1501 * @holebegin: byte in first page to unmap, relative to the start of
1502 * the underlying file. This will be rounded down to a PAGE_SIZE
1503 * boundary. Note that this is different from vmtruncate(), which
1504 * must keep the partial page. In contrast, we must get rid of
1506 * @holelen: size of prospective hole in bytes. This will be rounded
1507 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1509 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1510 * but 0 when invalidating pagecache, don't throw away private data.
1512 void unmap_mapping_range(struct address_space *mapping,
1513 loff_t const holebegin, loff_t const holelen, int even_cows)
1515 struct zap_details details;
1516 pgoff_t hba = holebegin >> PAGE_SHIFT;
1517 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1519 /* Check for overflow. */
1520 if (sizeof(holelen) > sizeof(hlen)) {
1522 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1523 if (holeend & ~(long long)ULONG_MAX)
1524 hlen = ULONG_MAX - hba + 1;
1527 details.check_mapping = even_cows? NULL: mapping;
1528 details.nonlinear_vma = NULL;
1529 details.first_index = hba;
1530 details.last_index = hba + hlen - 1;
1531 if (details.last_index < details.first_index)
1532 details.last_index = ULONG_MAX;
1533 details.i_mmap_lock = &mapping->i_mmap_lock;
1535 spin_lock(&mapping->i_mmap_lock);
1537 /* serialize i_size write against truncate_count write */
1539 /* Protect against page faults, and endless unmapping loops */
1540 mapping->truncate_count++;
1542 * For archs where spin_lock has inclusive semantics like ia64
1543 * this smp_mb() will prevent to read pagetable contents
1544 * before the truncate_count increment is visible to
1548 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1549 if (mapping->truncate_count == 0)
1550 reset_vma_truncate_counts(mapping);
1551 mapping->truncate_count++;
1553 details.truncate_count = mapping->truncate_count;
1555 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1556 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1557 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1558 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1559 spin_unlock(&mapping->i_mmap_lock);
1561 EXPORT_SYMBOL(unmap_mapping_range);
1564 * Handle all mappings that got truncated by a "truncate()"
1567 * NOTE! We have to be ready to update the memory sharing
1568 * between the file and the memory map for a potential last
1569 * incomplete page. Ugly, but necessary.
1571 int vmtruncate(struct inode * inode, loff_t offset)
1573 struct address_space *mapping = inode->i_mapping;
1574 unsigned long limit;
1576 if (inode->i_size < offset)
1579 * truncation of in-use swapfiles is disallowed - it would cause
1580 * subsequent swapout to scribble on the now-freed blocks.
1582 if (IS_SWAPFILE(inode))
1584 i_size_write(inode, offset);
1585 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1586 truncate_inode_pages(mapping, offset);
1590 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1591 if (limit != RLIM_INFINITY && offset > limit)
1593 if (offset > inode->i_sb->s_maxbytes)
1595 i_size_write(inode, offset);
1598 if (inode->i_op && inode->i_op->truncate)
1599 inode->i_op->truncate(inode);
1602 send_sig(SIGXFSZ, current, 0);
1609 EXPORT_SYMBOL(vmtruncate);
1612 * Primitive swap readahead code. We simply read an aligned block of
1613 * (1 << page_cluster) entries in the swap area. This method is chosen
1614 * because it doesn't cost us any seek time. We also make sure to queue
1615 * the 'original' request together with the readahead ones...
1617 * This has been extended to use the NUMA policies from the mm triggering
1620 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1622 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1625 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1628 struct page *new_page;
1629 unsigned long offset;
1632 * Get the number of handles we should do readahead io to.
1634 num = valid_swaphandles(entry, &offset);
1635 for (i = 0; i < num; offset++, i++) {
1636 /* Ok, do the async read-ahead now */
1637 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
1638 offset), vma, addr);
1641 page_cache_release(new_page);
1644 * Find the next applicable VMA for the NUMA policy.
1650 if (addr >= vma->vm_end) {
1652 next_vma = vma ? vma->vm_next : NULL;
1654 if (vma && addr < vma->vm_start)
1657 if (next_vma && addr >= next_vma->vm_start) {
1659 next_vma = vma->vm_next;
1664 lru_add_drain(); /* Push any new pages onto the LRU now */
1668 * We hold the mm semaphore and the page_table_lock on entry and
1669 * should release the pagetable lock on exit..
1671 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
1672 unsigned long address, pte_t *page_table, pmd_t *pmd,
1673 int write_access, pte_t orig_pte)
1678 int ret = VM_FAULT_MINOR;
1680 pte_unmap(page_table);
1681 spin_unlock(&mm->page_table_lock);
1683 entry = pte_to_swp_entry(orig_pte);
1684 page = lookup_swap_cache(entry);
1686 swapin_readahead(entry, address, vma);
1687 page = read_swap_cache_async(entry, vma, address);
1690 * Back out if somebody else faulted in this pte while
1691 * we released the page table lock.
1693 spin_lock(&mm->page_table_lock);
1694 page_table = pte_offset_map(pmd, address);
1695 if (likely(pte_same(*page_table, orig_pte)))
1700 /* Had to read the page from swap area: Major fault */
1701 ret = VM_FAULT_MAJOR;
1702 inc_page_state(pgmajfault);
1706 mark_page_accessed(page);
1710 * Back out if somebody else faulted in this pte while we
1711 * released the page table lock.
1713 spin_lock(&mm->page_table_lock);
1714 page_table = pte_offset_map(pmd, address);
1715 if (unlikely(!pte_same(*page_table, orig_pte)))
1718 if (unlikely(!PageUptodate(page))) {
1719 ret = VM_FAULT_SIGBUS;
1723 /* The page isn't present yet, go ahead with the fault. */
1725 inc_mm_counter(mm, anon_rss);
1726 pte = mk_pte(page, vma->vm_page_prot);
1727 if (write_access && can_share_swap_page(page)) {
1728 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
1732 flush_icache_page(vma, page);
1733 set_pte_at(mm, address, page_table, pte);
1734 page_add_anon_rmap(page, vma, address);
1738 remove_exclusive_swap_page(page);
1742 if (do_wp_page(mm, vma, address,
1743 page_table, pmd, pte) == VM_FAULT_OOM)
1748 /* No need to invalidate - it was non-present before */
1749 update_mmu_cache(vma, address, pte);
1750 lazy_mmu_prot_update(pte);
1752 pte_unmap(page_table);
1753 spin_unlock(&mm->page_table_lock);
1757 pte_unmap(page_table);
1758 spin_unlock(&mm->page_table_lock);
1760 page_cache_release(page);
1765 * We are called with the MM semaphore and page_table_lock
1766 * spinlock held to protect against concurrent faults in
1767 * multithreaded programs.
1769 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
1770 unsigned long address, pte_t *page_table, pmd_t *pmd,
1773 struct page *page = ZERO_PAGE(addr);
1776 /* Mapping of ZERO_PAGE - vm_page_prot is readonly */
1777 entry = mk_pte(page, vma->vm_page_prot);
1780 /* Allocate our own private page. */
1781 pte_unmap(page_table);
1782 spin_unlock(&mm->page_table_lock);
1784 if (unlikely(anon_vma_prepare(vma)))
1786 page = alloc_zeroed_user_highpage(vma, address);
1790 spin_lock(&mm->page_table_lock);
1791 page_table = pte_offset_map(pmd, address);
1793 if (!pte_none(*page_table)) {
1794 page_cache_release(page);
1797 inc_mm_counter(mm, anon_rss);
1798 entry = mk_pte(page, vma->vm_page_prot);
1799 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1800 lru_cache_add_active(page);
1801 SetPageReferenced(page);
1802 page_add_anon_rmap(page, vma, address);
1804 inc_mm_counter(mm, file_rss);
1805 page_add_file_rmap(page);
1806 page_cache_get(page);
1809 set_pte_at(mm, address, page_table, entry);
1811 /* No need to invalidate - it was non-present before */
1812 update_mmu_cache(vma, address, entry);
1813 lazy_mmu_prot_update(entry);
1815 pte_unmap(page_table);
1816 spin_unlock(&mm->page_table_lock);
1817 return VM_FAULT_MINOR;
1819 return VM_FAULT_OOM;
1823 * do_no_page() tries to create a new page mapping. It aggressively
1824 * tries to share with existing pages, but makes a separate copy if
1825 * the "write_access" parameter is true in order to avoid the next
1828 * As this is called only for pages that do not currently exist, we
1829 * do not need to flush old virtual caches or the TLB.
1831 * This is called with the MM semaphore held and the page table
1832 * spinlock held. Exit with the spinlock released.
1834 static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
1835 unsigned long address, pte_t *page_table, pmd_t *pmd,
1838 struct page *new_page;
1839 struct address_space *mapping = NULL;
1841 unsigned int sequence = 0;
1842 int ret = VM_FAULT_MINOR;
1845 pte_unmap(page_table);
1846 spin_unlock(&mm->page_table_lock);
1849 mapping = vma->vm_file->f_mapping;
1850 sequence = mapping->truncate_count;
1851 smp_rmb(); /* serializes i_size against truncate_count */
1854 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
1856 * No smp_rmb is needed here as long as there's a full
1857 * spin_lock/unlock sequence inside the ->nopage callback
1858 * (for the pagecache lookup) that acts as an implicit
1859 * smp_mb() and prevents the i_size read to happen
1860 * after the next truncate_count read.
1863 /* no page was available -- either SIGBUS or OOM */
1864 if (new_page == NOPAGE_SIGBUS)
1865 return VM_FAULT_SIGBUS;
1866 if (new_page == NOPAGE_OOM)
1867 return VM_FAULT_OOM;
1870 * Should we do an early C-O-W break?
1872 if (write_access && !(vma->vm_flags & VM_SHARED)) {
1875 if (unlikely(anon_vma_prepare(vma)))
1877 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1880 copy_user_highpage(page, new_page, address);
1881 page_cache_release(new_page);
1886 spin_lock(&mm->page_table_lock);
1888 * For a file-backed vma, someone could have truncated or otherwise
1889 * invalidated this page. If unmap_mapping_range got called,
1890 * retry getting the page.
1892 if (mapping && unlikely(sequence != mapping->truncate_count)) {
1893 spin_unlock(&mm->page_table_lock);
1894 page_cache_release(new_page);
1896 sequence = mapping->truncate_count;
1900 page_table = pte_offset_map(pmd, address);
1903 * This silly early PAGE_DIRTY setting removes a race
1904 * due to the bad i386 page protection. But it's valid
1905 * for other architectures too.
1907 * Note that if write_access is true, we either now have
1908 * an exclusive copy of the page, or this is a shared mapping,
1909 * so we can make it writable and dirty to avoid having to
1910 * handle that later.
1912 /* Only go through if we didn't race with anybody else... */
1913 if (pte_none(*page_table)) {
1914 flush_icache_page(vma, new_page);
1915 entry = mk_pte(new_page, vma->vm_page_prot);
1917 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1918 set_pte_at(mm, address, page_table, entry);
1920 inc_mm_counter(mm, anon_rss);
1921 lru_cache_add_active(new_page);
1922 page_add_anon_rmap(new_page, vma, address);
1923 } else if (!(vma->vm_flags & VM_RESERVED)) {
1924 inc_mm_counter(mm, file_rss);
1925 page_add_file_rmap(new_page);
1928 /* One of our sibling threads was faster, back out. */
1929 page_cache_release(new_page);
1933 /* no need to invalidate: a not-present page shouldn't be cached */
1934 update_mmu_cache(vma, address, entry);
1935 lazy_mmu_prot_update(entry);
1937 pte_unmap(page_table);
1938 spin_unlock(&mm->page_table_lock);
1941 page_cache_release(new_page);
1942 return VM_FAULT_OOM;
1946 * Fault of a previously existing named mapping. Repopulate the pte
1947 * from the encoded file_pte if possible. This enables swappable
1950 static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
1951 unsigned long address, pte_t *page_table, pmd_t *pmd,
1952 int write_access, pte_t orig_pte)
1957 pte_unmap(page_table);
1958 spin_unlock(&mm->page_table_lock);
1960 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
1962 * Page table corrupted: show pte and kill process.
1964 print_bad_pte(vma, orig_pte, address);
1965 return VM_FAULT_OOM;
1967 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
1969 pgoff = pte_to_pgoff(orig_pte);
1970 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
1971 vma->vm_page_prot, pgoff, 0);
1973 return VM_FAULT_OOM;
1975 return VM_FAULT_SIGBUS;
1976 return VM_FAULT_MAJOR;
1980 * These routines also need to handle stuff like marking pages dirty
1981 * and/or accessed for architectures that don't do it in hardware (most
1982 * RISC architectures). The early dirtying is also good on the i386.
1984 * There is also a hook called "update_mmu_cache()" that architectures
1985 * with external mmu caches can use to update those (ie the Sparc or
1986 * PowerPC hashed page tables that act as extended TLBs).
1988 * Note the "page_table_lock". It is to protect against kswapd removing
1989 * pages from under us. Note that kswapd only ever _removes_ pages, never
1990 * adds them. As such, once we have noticed that the page is not present,
1991 * we can drop the lock early.
1993 * The adding of pages is protected by the MM semaphore (which we hold),
1994 * so we don't need to worry about a page being suddenly been added into
1997 * We enter with the pagetable spinlock held, we are supposed to
1998 * release it when done.
2000 static inline int handle_pte_fault(struct mm_struct *mm,
2001 struct vm_area_struct *vma, unsigned long address,
2002 pte_t *pte, pmd_t *pmd, int write_access)
2007 if (!pte_present(entry)) {
2008 if (pte_none(entry)) {
2009 if (!vma->vm_ops || !vma->vm_ops->nopage)
2010 return do_anonymous_page(mm, vma, address,
2011 pte, pmd, write_access);
2012 return do_no_page(mm, vma, address,
2013 pte, pmd, write_access);
2015 if (pte_file(entry))
2016 return do_file_page(mm, vma, address,
2017 pte, pmd, write_access, entry);
2018 return do_swap_page(mm, vma, address,
2019 pte, pmd, write_access, entry);
2023 if (!pte_write(entry))
2024 return do_wp_page(mm, vma, address, pte, pmd, entry);
2025 entry = pte_mkdirty(entry);
2027 entry = pte_mkyoung(entry);
2028 ptep_set_access_flags(vma, address, pte, entry, write_access);
2029 update_mmu_cache(vma, address, entry);
2030 lazy_mmu_prot_update(entry);
2032 spin_unlock(&mm->page_table_lock);
2033 return VM_FAULT_MINOR;
2037 * By the time we get here, we already hold the mm semaphore
2039 int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2040 unsigned long address, int write_access)
2047 __set_current_state(TASK_RUNNING);
2049 inc_page_state(pgfault);
2051 if (unlikely(is_vm_hugetlb_page(vma)))
2052 return hugetlb_fault(mm, vma, address, write_access);
2055 * We need the page table lock to synchronize with kswapd
2056 * and the SMP-safe atomic PTE updates.
2058 pgd = pgd_offset(mm, address);
2059 spin_lock(&mm->page_table_lock);
2061 pud = pud_alloc(mm, pgd, address);
2065 pmd = pmd_alloc(mm, pud, address);
2069 pte = pte_alloc_map(mm, pmd, address);
2073 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2076 spin_unlock(&mm->page_table_lock);
2077 return VM_FAULT_OOM;
2080 #ifndef __PAGETABLE_PUD_FOLDED
2082 * Allocate page upper directory.
2083 * We've already handled the fast-path in-line.
2085 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2089 if (mm != &init_mm) /* Temporary bridging hack */
2090 spin_unlock(&mm->page_table_lock);
2091 new = pud_alloc_one(mm, address);
2093 if (mm != &init_mm) /* Temporary bridging hack */
2094 spin_lock(&mm->page_table_lock);
2098 spin_lock(&mm->page_table_lock);
2099 if (pgd_present(*pgd)) /* Another has populated it */
2102 pgd_populate(mm, pgd, new);
2103 if (mm == &init_mm) /* Temporary bridging hack */
2104 spin_unlock(&mm->page_table_lock);
2107 #endif /* __PAGETABLE_PUD_FOLDED */
2109 #ifndef __PAGETABLE_PMD_FOLDED
2111 * Allocate page middle directory.
2112 * We've already handled the fast-path in-line.
2114 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2118 if (mm != &init_mm) /* Temporary bridging hack */
2119 spin_unlock(&mm->page_table_lock);
2120 new = pmd_alloc_one(mm, address);
2122 if (mm != &init_mm) /* Temporary bridging hack */
2123 spin_lock(&mm->page_table_lock);
2127 spin_lock(&mm->page_table_lock);
2128 #ifndef __ARCH_HAS_4LEVEL_HACK
2129 if (pud_present(*pud)) /* Another has populated it */
2132 pud_populate(mm, pud, new);
2134 if (pgd_present(*pud)) /* Another has populated it */
2137 pgd_populate(mm, pud, new);
2138 #endif /* __ARCH_HAS_4LEVEL_HACK */
2139 if (mm == &init_mm) /* Temporary bridging hack */
2140 spin_unlock(&mm->page_table_lock);
2143 #endif /* __PAGETABLE_PMD_FOLDED */
2145 int make_pages_present(unsigned long addr, unsigned long end)
2147 int ret, len, write;
2148 struct vm_area_struct * vma;
2150 vma = find_vma(current->mm, addr);
2153 write = (vma->vm_flags & VM_WRITE) != 0;
2156 if (end > vma->vm_end)
2158 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2159 ret = get_user_pages(current, current->mm, addr,
2160 len, write, 0, NULL, NULL);
2163 return ret == len ? 0 : -1;
2167 * Map a vmalloc()-space virtual address to the physical page.
2169 struct page * vmalloc_to_page(void * vmalloc_addr)
2171 unsigned long addr = (unsigned long) vmalloc_addr;
2172 struct page *page = NULL;
2173 pgd_t *pgd = pgd_offset_k(addr);
2178 if (!pgd_none(*pgd)) {
2179 pud = pud_offset(pgd, addr);
2180 if (!pud_none(*pud)) {
2181 pmd = pmd_offset(pud, addr);
2182 if (!pmd_none(*pmd)) {
2183 ptep = pte_offset_map(pmd, addr);
2185 if (pte_present(pte))
2186 page = pte_page(pte);
2194 EXPORT_SYMBOL(vmalloc_to_page);
2197 * Map a vmalloc()-space virtual address to the physical page frame number.
2199 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2201 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2204 EXPORT_SYMBOL(vmalloc_to_pfn);
2206 #if !defined(__HAVE_ARCH_GATE_AREA)
2208 #if defined(AT_SYSINFO_EHDR)
2209 static struct vm_area_struct gate_vma;
2211 static int __init gate_vma_init(void)
2213 gate_vma.vm_mm = NULL;
2214 gate_vma.vm_start = FIXADDR_USER_START;
2215 gate_vma.vm_end = FIXADDR_USER_END;
2216 gate_vma.vm_page_prot = PAGE_READONLY;
2217 gate_vma.vm_flags = VM_RESERVED;
2220 __initcall(gate_vma_init);
2223 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2225 #ifdef AT_SYSINFO_EHDR
2232 int in_gate_area_no_task(unsigned long addr)
2234 #ifdef AT_SYSINFO_EHDR
2235 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2241 #endif /* __HAVE_ARCH_GATE_AREA */