2 * mm/rmap.c - physical to virtual reverse mappings
4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
5 * Released under the General Public License (GPL).
7 * Simple, low overhead reverse mapping scheme.
8 * Please try to keep this thing as modular as possible.
10 * Provides methods for unmapping each kind of mapped page:
11 * the anon methods track anonymous pages, and
12 * the file methods track pages belonging to an inode.
14 * Original design by Rik van Riel <riel@conectiva.com.br> 2001
15 * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004
16 * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004
17 * Contributions by Hugh Dickins <hugh@veritas.com> 2003, 2004
21 * Lock ordering in mm:
23 * inode->i_mutex (while writing or truncating, not reading or faulting)
24 * inode->i_alloc_sem (vmtruncate_range)
26 * page->flags PG_locked (lock_page)
27 * mapping->i_mmap_lock
29 * mm->page_table_lock or pte_lock
30 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers)
34 * inode_lock (in set_page_dirty's __mark_inode_dirty)
35 * sb_lock (within inode_lock in fs/fs-writeback.c)
36 * mapping->tree_lock (widely used, in set_page_dirty,
37 * in arch-dependent flush_dcache_mmap_lock,
38 * within inode_lock in __sync_single_inode)
42 #include <linux/pagemap.h>
43 #include <linux/swap.h>
44 #include <linux/swapops.h>
45 #include <linux/slab.h>
46 #include <linux/init.h>
47 #include <linux/rmap.h>
48 #include <linux/rcupdate.h>
49 #include <linux/module.h>
50 #include <linux/kallsyms.h>
51 #include <linux/memcontrol.h>
53 #include <asm/tlbflush.h>
55 struct kmem_cache *anon_vma_cachep;
57 /* This must be called under the mmap_sem. */
58 int anon_vma_prepare(struct vm_area_struct *vma)
60 struct anon_vma *anon_vma = vma->anon_vma;
63 if (unlikely(!anon_vma)) {
64 struct mm_struct *mm = vma->vm_mm;
65 struct anon_vma *allocated, *locked;
67 anon_vma = find_mergeable_anon_vma(vma);
71 spin_lock(&locked->lock);
73 anon_vma = anon_vma_alloc();
74 if (unlikely(!anon_vma))
80 /* page_table_lock to protect against threads */
81 spin_lock(&mm->page_table_lock);
82 if (likely(!vma->anon_vma)) {
83 vma->anon_vma = anon_vma;
84 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
87 spin_unlock(&mm->page_table_lock);
90 spin_unlock(&locked->lock);
91 if (unlikely(allocated))
92 anon_vma_free(allocated);
97 void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next)
99 BUG_ON(vma->anon_vma != next->anon_vma);
100 list_del(&next->anon_vma_node);
103 void __anon_vma_link(struct vm_area_struct *vma)
105 struct anon_vma *anon_vma = vma->anon_vma;
108 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
111 void anon_vma_link(struct vm_area_struct *vma)
113 struct anon_vma *anon_vma = vma->anon_vma;
116 spin_lock(&anon_vma->lock);
117 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
118 spin_unlock(&anon_vma->lock);
122 void anon_vma_unlink(struct vm_area_struct *vma)
124 struct anon_vma *anon_vma = vma->anon_vma;
130 spin_lock(&anon_vma->lock);
131 list_del(&vma->anon_vma_node);
133 /* We must garbage collect the anon_vma if it's empty */
134 empty = list_empty(&anon_vma->head);
135 spin_unlock(&anon_vma->lock);
138 anon_vma_free(anon_vma);
141 static void anon_vma_ctor(struct kmem_cache *cachep, void *data)
143 struct anon_vma *anon_vma = data;
145 spin_lock_init(&anon_vma->lock);
146 INIT_LIST_HEAD(&anon_vma->head);
149 void __init anon_vma_init(void)
151 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
152 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor);
156 * Getting a lock on a stable anon_vma from a page off the LRU is
157 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
159 static struct anon_vma *page_lock_anon_vma(struct page *page)
161 struct anon_vma *anon_vma;
162 unsigned long anon_mapping;
165 anon_mapping = (unsigned long) page->mapping;
166 if (!(anon_mapping & PAGE_MAPPING_ANON))
168 if (!page_mapped(page))
171 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
172 spin_lock(&anon_vma->lock);
179 static void page_unlock_anon_vma(struct anon_vma *anon_vma)
181 spin_unlock(&anon_vma->lock);
186 * At what user virtual address is page expected in @vma?
187 * Returns virtual address or -EFAULT if page's index/offset is not
188 * within the range mapped the @vma.
190 static inline unsigned long
191 vma_address(struct page *page, struct vm_area_struct *vma)
193 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
194 unsigned long address;
196 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
197 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
198 /* page should be within @vma mapping range */
205 * At what user virtual address is page expected in vma? checking that the
206 * page matches the vma: currently only used on anon pages, by unuse_vma;
208 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
210 if (PageAnon(page)) {
211 if ((void *)vma->anon_vma !=
212 (void *)page->mapping - PAGE_MAPPING_ANON)
214 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
216 vma->vm_file->f_mapping != page->mapping)
220 return vma_address(page, vma);
224 * Check that @page is mapped at @address into @mm.
226 * On success returns with pte mapped and locked.
228 pte_t *page_check_address(struct page *page, struct mm_struct *mm,
229 unsigned long address, spinlock_t **ptlp)
237 pgd = pgd_offset(mm, address);
238 if (!pgd_present(*pgd))
241 pud = pud_offset(pgd, address);
242 if (!pud_present(*pud))
245 pmd = pmd_offset(pud, address);
246 if (!pmd_present(*pmd))
249 pte = pte_offset_map(pmd, address);
250 /* Make a quick check before getting the lock */
251 if (!pte_present(*pte)) {
256 ptl = pte_lockptr(mm, pmd);
258 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
262 pte_unmap_unlock(pte, ptl);
267 * Subfunctions of page_referenced: page_referenced_one called
268 * repeatedly from either page_referenced_anon or page_referenced_file.
270 static int page_referenced_one(struct page *page,
271 struct vm_area_struct *vma, unsigned int *mapcount)
273 struct mm_struct *mm = vma->vm_mm;
274 unsigned long address;
279 address = vma_address(page, vma);
280 if (address == -EFAULT)
283 pte = page_check_address(page, mm, address, &ptl);
287 if (vma->vm_flags & VM_LOCKED) {
289 *mapcount = 1; /* break early from loop */
290 } else if (ptep_clear_flush_young(vma, address, pte))
293 /* Pretend the page is referenced if the task has the
294 swap token and is in the middle of a page fault. */
295 if (mm != current->mm && has_swap_token(mm) &&
296 rwsem_is_locked(&mm->mmap_sem))
300 pte_unmap_unlock(pte, ptl);
305 static int page_referenced_anon(struct page *page,
306 struct mem_cgroup *mem_cont)
308 unsigned int mapcount;
309 struct anon_vma *anon_vma;
310 struct vm_area_struct *vma;
313 anon_vma = page_lock_anon_vma(page);
317 mapcount = page_mapcount(page);
318 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
320 * If we are reclaiming on behalf of a cgroup, skip
321 * counting on behalf of references from different
324 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
326 referenced += page_referenced_one(page, vma, &mapcount);
331 page_unlock_anon_vma(anon_vma);
336 * page_referenced_file - referenced check for object-based rmap
337 * @page: the page we're checking references on.
339 * For an object-based mapped page, find all the places it is mapped and
340 * check/clear the referenced flag. This is done by following the page->mapping
341 * pointer, then walking the chain of vmas it holds. It returns the number
342 * of references it found.
344 * This function is only called from page_referenced for object-based pages.
346 static int page_referenced_file(struct page *page,
347 struct mem_cgroup *mem_cont)
349 unsigned int mapcount;
350 struct address_space *mapping = page->mapping;
351 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
352 struct vm_area_struct *vma;
353 struct prio_tree_iter iter;
357 * The caller's checks on page->mapping and !PageAnon have made
358 * sure that this is a file page: the check for page->mapping
359 * excludes the case just before it gets set on an anon page.
361 BUG_ON(PageAnon(page));
364 * The page lock not only makes sure that page->mapping cannot
365 * suddenly be NULLified by truncation, it makes sure that the
366 * structure at mapping cannot be freed and reused yet,
367 * so we can safely take mapping->i_mmap_lock.
369 BUG_ON(!PageLocked(page));
371 spin_lock(&mapping->i_mmap_lock);
374 * i_mmap_lock does not stabilize mapcount at all, but mapcount
375 * is more likely to be accurate if we note it after spinning.
377 mapcount = page_mapcount(page);
379 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
381 * If we are reclaiming on behalf of a cgroup, skip
382 * counting on behalf of references from different
385 if (mem_cont && !mm_match_cgroup(vma->vm_mm, mem_cont))
387 if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE))
388 == (VM_LOCKED|VM_MAYSHARE)) {
392 referenced += page_referenced_one(page, vma, &mapcount);
397 spin_unlock(&mapping->i_mmap_lock);
402 * page_referenced - test if the page was referenced
403 * @page: the page to test
404 * @is_locked: caller holds lock on the page
406 * Quick test_and_clear_referenced for all mappings to a page,
407 * returns the number of ptes which referenced the page.
409 int page_referenced(struct page *page, int is_locked,
410 struct mem_cgroup *mem_cont)
414 if (page_test_and_clear_young(page))
417 if (TestClearPageReferenced(page))
420 if (page_mapped(page) && page->mapping) {
422 referenced += page_referenced_anon(page, mem_cont);
424 referenced += page_referenced_file(page, mem_cont);
425 else if (TestSetPageLocked(page))
430 page_referenced_file(page, mem_cont);
437 static int page_mkclean_one(struct page *page, struct vm_area_struct *vma)
439 struct mm_struct *mm = vma->vm_mm;
440 unsigned long address;
445 address = vma_address(page, vma);
446 if (address == -EFAULT)
449 pte = page_check_address(page, mm, address, &ptl);
453 if (pte_dirty(*pte) || pte_write(*pte)) {
456 flush_cache_page(vma, address, pte_pfn(*pte));
457 entry = ptep_clear_flush(vma, address, pte);
458 entry = pte_wrprotect(entry);
459 entry = pte_mkclean(entry);
460 set_pte_at(mm, address, pte, entry);
464 pte_unmap_unlock(pte, ptl);
469 static int page_mkclean_file(struct address_space *mapping, struct page *page)
471 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
472 struct vm_area_struct *vma;
473 struct prio_tree_iter iter;
476 BUG_ON(PageAnon(page));
478 spin_lock(&mapping->i_mmap_lock);
479 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
480 if (vma->vm_flags & VM_SHARED)
481 ret += page_mkclean_one(page, vma);
483 spin_unlock(&mapping->i_mmap_lock);
487 int page_mkclean(struct page *page)
491 BUG_ON(!PageLocked(page));
493 if (page_mapped(page)) {
494 struct address_space *mapping = page_mapping(page);
496 ret = page_mkclean_file(mapping, page);
497 if (page_test_dirty(page)) {
498 page_clear_dirty(page);
506 EXPORT_SYMBOL_GPL(page_mkclean);
509 * page_set_anon_rmap - setup new anonymous rmap
510 * @page: the page to add the mapping to
511 * @vma: the vm area in which the mapping is added
512 * @address: the user virtual address mapped
514 static void __page_set_anon_rmap(struct page *page,
515 struct vm_area_struct *vma, unsigned long address)
517 struct anon_vma *anon_vma = vma->anon_vma;
520 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
521 page->mapping = (struct address_space *) anon_vma;
523 page->index = linear_page_index(vma, address);
526 * nr_mapped state can be updated without turning off
527 * interrupts because it is not modified via interrupt.
529 __inc_zone_page_state(page, NR_ANON_PAGES);
533 * page_set_anon_rmap - sanity check anonymous rmap addition
534 * @page: the page to add the mapping to
535 * @vma: the vm area in which the mapping is added
536 * @address: the user virtual address mapped
538 static void __page_check_anon_rmap(struct page *page,
539 struct vm_area_struct *vma, unsigned long address)
541 #ifdef CONFIG_DEBUG_VM
543 * The page's anon-rmap details (mapping and index) are guaranteed to
544 * be set up correctly at this point.
546 * We have exclusion against page_add_anon_rmap because the caller
547 * always holds the page locked, except if called from page_dup_rmap,
548 * in which case the page is already known to be setup.
550 * We have exclusion against page_add_new_anon_rmap because those pages
551 * are initially only visible via the pagetables, and the pte is locked
552 * over the call to page_add_new_anon_rmap.
554 struct anon_vma *anon_vma = vma->anon_vma;
555 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
556 BUG_ON(page->mapping != (struct address_space *)anon_vma);
557 BUG_ON(page->index != linear_page_index(vma, address));
562 * page_add_anon_rmap - add pte mapping to an anonymous page
563 * @page: the page to add the mapping to
564 * @vma: the vm area in which the mapping is added
565 * @address: the user virtual address mapped
567 * The caller needs to hold the pte lock and the page must be locked.
569 void page_add_anon_rmap(struct page *page,
570 struct vm_area_struct *vma, unsigned long address)
572 VM_BUG_ON(!PageLocked(page));
573 VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
574 if (atomic_inc_and_test(&page->_mapcount))
575 __page_set_anon_rmap(page, vma, address);
577 __page_check_anon_rmap(page, vma, address);
579 * We unconditionally charged during prepare, we uncharge here
580 * This takes care of balancing the reference counts
582 mem_cgroup_uncharge_page(page);
587 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
588 * @page: the page to add the mapping to
589 * @vma: the vm area in which the mapping is added
590 * @address: the user virtual address mapped
592 * Same as page_add_anon_rmap but must only be called on *new* pages.
593 * This means the inc-and-test can be bypassed.
594 * Page does not have to be locked.
596 void page_add_new_anon_rmap(struct page *page,
597 struct vm_area_struct *vma, unsigned long address)
599 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
600 atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */
601 __page_set_anon_rmap(page, vma, address);
605 * page_add_file_rmap - add pte mapping to a file page
606 * @page: the page to add the mapping to
608 * The caller needs to hold the pte lock.
610 void page_add_file_rmap(struct page *page)
612 if (atomic_inc_and_test(&page->_mapcount))
613 __inc_zone_page_state(page, NR_FILE_MAPPED);
616 * We unconditionally charged during prepare, we uncharge here
617 * This takes care of balancing the reference counts
619 mem_cgroup_uncharge_page(page);
622 #ifdef CONFIG_DEBUG_VM
624 * page_dup_rmap - duplicate pte mapping to a page
625 * @page: the page to add the mapping to
627 * For copy_page_range only: minimal extract from page_add_file_rmap /
628 * page_add_anon_rmap, avoiding unnecessary tests (already checked) so it's
631 * The caller needs to hold the pte lock.
633 void page_dup_rmap(struct page *page, struct vm_area_struct *vma, unsigned long address)
635 BUG_ON(page_mapcount(page) == 0);
637 __page_check_anon_rmap(page, vma, address);
638 atomic_inc(&page->_mapcount);
643 * page_remove_rmap - take down pte mapping from a page
644 * @page: page to remove mapping from
646 * The caller needs to hold the pte lock.
648 void page_remove_rmap(struct page *page, struct vm_area_struct *vma)
650 if (atomic_add_negative(-1, &page->_mapcount)) {
651 if (unlikely(page_mapcount(page) < 0)) {
652 printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page));
653 printk (KERN_EMERG " page pfn = %lx\n", page_to_pfn(page));
654 printk (KERN_EMERG " page->flags = %lx\n", page->flags);
655 printk (KERN_EMERG " page->count = %x\n", page_count(page));
656 printk (KERN_EMERG " page->mapping = %p\n", page->mapping);
657 print_symbol (KERN_EMERG " vma->vm_ops = %s\n", (unsigned long)vma->vm_ops);
659 print_symbol (KERN_EMERG " vma->vm_ops->nopage = %s\n", (unsigned long)vma->vm_ops->nopage);
660 print_symbol (KERN_EMERG " vma->vm_ops->fault = %s\n", (unsigned long)vma->vm_ops->fault);
662 if (vma->vm_file && vma->vm_file->f_op)
663 print_symbol (KERN_EMERG " vma->vm_file->f_op->mmap = %s\n", (unsigned long)vma->vm_file->f_op->mmap);
668 * It would be tidy to reset the PageAnon mapping here,
669 * but that might overwrite a racing page_add_anon_rmap
670 * which increments mapcount after us but sets mapping
671 * before us: so leave the reset to free_hot_cold_page,
672 * and remember that it's only reliable while mapped.
673 * Leaving it set also helps swapoff to reinstate ptes
674 * faster for those pages still in swapcache.
676 if (page_test_dirty(page)) {
677 page_clear_dirty(page);
678 set_page_dirty(page);
680 mem_cgroup_uncharge_page(page);
682 __dec_zone_page_state(page,
683 PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED);
688 * Subfunctions of try_to_unmap: try_to_unmap_one called
689 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
691 static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
694 struct mm_struct *mm = vma->vm_mm;
695 unsigned long address;
699 int ret = SWAP_AGAIN;
701 address = vma_address(page, vma);
702 if (address == -EFAULT)
705 pte = page_check_address(page, mm, address, &ptl);
710 * If the page is mlock()d, we cannot swap it out.
711 * If it's recently referenced (perhaps page_referenced
712 * skipped over this mm) then we should reactivate it.
714 if (!migration && ((vma->vm_flags & VM_LOCKED) ||
715 (ptep_clear_flush_young(vma, address, pte)))) {
720 /* Nuke the page table entry. */
721 flush_cache_page(vma, address, page_to_pfn(page));
722 pteval = ptep_clear_flush(vma, address, pte);
724 /* Move the dirty bit to the physical page now the pte is gone. */
725 if (pte_dirty(pteval))
726 set_page_dirty(page);
728 /* Update high watermark before we lower rss */
729 update_hiwater_rss(mm);
731 if (PageAnon(page)) {
732 swp_entry_t entry = { .val = page_private(page) };
734 if (PageSwapCache(page)) {
736 * Store the swap location in the pte.
737 * See handle_pte_fault() ...
739 swap_duplicate(entry);
740 if (list_empty(&mm->mmlist)) {
741 spin_lock(&mmlist_lock);
742 if (list_empty(&mm->mmlist))
743 list_add(&mm->mmlist, &init_mm.mmlist);
744 spin_unlock(&mmlist_lock);
746 dec_mm_counter(mm, anon_rss);
747 #ifdef CONFIG_MIGRATION
750 * Store the pfn of the page in a special migration
751 * pte. do_swap_page() will wait until the migration
752 * pte is removed and then restart fault handling.
755 entry = make_migration_entry(page, pte_write(pteval));
758 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
759 BUG_ON(pte_file(*pte));
761 #ifdef CONFIG_MIGRATION
763 /* Establish migration entry for a file page */
765 entry = make_migration_entry(page, pte_write(pteval));
766 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
769 dec_mm_counter(mm, file_rss);
772 page_remove_rmap(page, vma);
773 page_cache_release(page);
776 pte_unmap_unlock(pte, ptl);
782 * objrmap doesn't work for nonlinear VMAs because the assumption that
783 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
784 * Consequently, given a particular page and its ->index, we cannot locate the
785 * ptes which are mapping that page without an exhaustive linear search.
787 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
788 * maps the file to which the target page belongs. The ->vm_private_data field
789 * holds the current cursor into that scan. Successive searches will circulate
790 * around the vma's virtual address space.
792 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
793 * more scanning pressure is placed against them as well. Eventually pages
794 * will become fully unmapped and are eligible for eviction.
796 * For very sparsely populated VMAs this is a little inefficient - chances are
797 * there there won't be many ptes located within the scan cluster. In this case
798 * maybe we could scan further - to the end of the pte page, perhaps.
800 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
801 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
803 static void try_to_unmap_cluster(unsigned long cursor,
804 unsigned int *mapcount, struct vm_area_struct *vma)
806 struct mm_struct *mm = vma->vm_mm;
814 unsigned long address;
817 address = (vma->vm_start + cursor) & CLUSTER_MASK;
818 end = address + CLUSTER_SIZE;
819 if (address < vma->vm_start)
820 address = vma->vm_start;
821 if (end > vma->vm_end)
824 pgd = pgd_offset(mm, address);
825 if (!pgd_present(*pgd))
828 pud = pud_offset(pgd, address);
829 if (!pud_present(*pud))
832 pmd = pmd_offset(pud, address);
833 if (!pmd_present(*pmd))
836 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
838 /* Update high watermark before we lower rss */
839 update_hiwater_rss(mm);
841 for (; address < end; pte++, address += PAGE_SIZE) {
842 if (!pte_present(*pte))
844 page = vm_normal_page(vma, address, *pte);
845 BUG_ON(!page || PageAnon(page));
847 if (ptep_clear_flush_young(vma, address, pte))
850 /* Nuke the page table entry. */
851 flush_cache_page(vma, address, pte_pfn(*pte));
852 pteval = ptep_clear_flush(vma, address, pte);
854 /* If nonlinear, store the file page offset in the pte. */
855 if (page->index != linear_page_index(vma, address))
856 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
858 /* Move the dirty bit to the physical page now the pte is gone. */
859 if (pte_dirty(pteval))
860 set_page_dirty(page);
862 page_remove_rmap(page, vma);
863 page_cache_release(page);
864 dec_mm_counter(mm, file_rss);
867 pte_unmap_unlock(pte - 1, ptl);
870 static int try_to_unmap_anon(struct page *page, int migration)
872 struct anon_vma *anon_vma;
873 struct vm_area_struct *vma;
874 int ret = SWAP_AGAIN;
876 anon_vma = page_lock_anon_vma(page);
880 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
881 ret = try_to_unmap_one(page, vma, migration);
882 if (ret == SWAP_FAIL || !page_mapped(page))
886 page_unlock_anon_vma(anon_vma);
891 * try_to_unmap_file - unmap file page using the object-based rmap method
892 * @page: the page to unmap
894 * Find all the mappings of a page using the mapping pointer and the vma chains
895 * contained in the address_space struct it points to.
897 * This function is only called from try_to_unmap for object-based pages.
899 static int try_to_unmap_file(struct page *page, int migration)
901 struct address_space *mapping = page->mapping;
902 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
903 struct vm_area_struct *vma;
904 struct prio_tree_iter iter;
905 int ret = SWAP_AGAIN;
906 unsigned long cursor;
907 unsigned long max_nl_cursor = 0;
908 unsigned long max_nl_size = 0;
909 unsigned int mapcount;
911 spin_lock(&mapping->i_mmap_lock);
912 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
913 ret = try_to_unmap_one(page, vma, migration);
914 if (ret == SWAP_FAIL || !page_mapped(page))
918 if (list_empty(&mapping->i_mmap_nonlinear))
921 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
922 shared.vm_set.list) {
923 if ((vma->vm_flags & VM_LOCKED) && !migration)
925 cursor = (unsigned long) vma->vm_private_data;
926 if (cursor > max_nl_cursor)
927 max_nl_cursor = cursor;
928 cursor = vma->vm_end - vma->vm_start;
929 if (cursor > max_nl_size)
930 max_nl_size = cursor;
933 if (max_nl_size == 0) { /* any nonlinears locked or reserved */
939 * We don't try to search for this page in the nonlinear vmas,
940 * and page_referenced wouldn't have found it anyway. Instead
941 * just walk the nonlinear vmas trying to age and unmap some.
942 * The mapcount of the page we came in with is irrelevant,
943 * but even so use it as a guide to how hard we should try?
945 mapcount = page_mapcount(page);
948 cond_resched_lock(&mapping->i_mmap_lock);
950 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
951 if (max_nl_cursor == 0)
952 max_nl_cursor = CLUSTER_SIZE;
955 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
956 shared.vm_set.list) {
957 if ((vma->vm_flags & VM_LOCKED) && !migration)
959 cursor = (unsigned long) vma->vm_private_data;
960 while ( cursor < max_nl_cursor &&
961 cursor < vma->vm_end - vma->vm_start) {
962 try_to_unmap_cluster(cursor, &mapcount, vma);
963 cursor += CLUSTER_SIZE;
964 vma->vm_private_data = (void *) cursor;
965 if ((int)mapcount <= 0)
968 vma->vm_private_data = (void *) max_nl_cursor;
970 cond_resched_lock(&mapping->i_mmap_lock);
971 max_nl_cursor += CLUSTER_SIZE;
972 } while (max_nl_cursor <= max_nl_size);
975 * Don't loop forever (perhaps all the remaining pages are
976 * in locked vmas). Reset cursor on all unreserved nonlinear
977 * vmas, now forgetting on which ones it had fallen behind.
979 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
980 vma->vm_private_data = NULL;
982 spin_unlock(&mapping->i_mmap_lock);
987 * try_to_unmap - try to remove all page table mappings to a page
988 * @page: the page to get unmapped
990 * Tries to remove all the page table entries which are mapping this
991 * page, used in the pageout path. Caller must hold the page lock.
994 * SWAP_SUCCESS - we succeeded in removing all mappings
995 * SWAP_AGAIN - we missed a mapping, try again later
996 * SWAP_FAIL - the page is unswappable
998 int try_to_unmap(struct page *page, int migration)
1002 BUG_ON(!PageLocked(page));
1005 ret = try_to_unmap_anon(page, migration);
1007 ret = try_to_unmap_file(page, migration);
1009 if (!page_mapped(page))