ia64/linux-2.6.18-xen.hg

view mm/rmap.c @ 452:c7ed6fe5dca0

kexec: dont initialise regions in reserve_memory()

There is no need to initialise efi_memmap_res and boot_param_res in
reserve_memory() for the initial xen domain as it is done in
machine_kexec_setup_resources() using values from the kexec hypercall.

Signed-off-by: Simon Horman <horms@verge.net.au>
author Keir Fraser <keir.fraser@citrix.com>
date Thu Feb 28 10:55:18 2008 +0000 (2008-02-28)
parents 831230e53067
children
line source
1 /*
2 * mm/rmap.c - physical to virtual reverse mappings
3 *
4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
5 * Released under the General Public License (GPL).
6 *
7 * Simple, low overhead reverse mapping scheme.
8 * Please try to keep this thing as modular as possible.
9 *
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.
13 *
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
18 */
20 /*
21 * Lock ordering in mm:
22 *
23 * inode->i_mutex (while writing or truncating, not reading or faulting)
24 * inode->i_alloc_sem
25 *
26 * When a page fault occurs in writing from user to file, down_read
27 * of mmap_sem nests within i_mutex; in sys_msync, i_mutex nests within
28 * down_read of mmap_sem; i_mutex and down_write of mmap_sem are never
29 * taken together; in truncation, i_mutex is taken outermost.
30 *
31 * mm->mmap_sem
32 * page->flags PG_locked (lock_page)
33 * mapping->i_mmap_lock
34 * anon_vma->lock
35 * mm->page_table_lock or pte_lock
36 * zone->lru_lock (in mark_page_accessed, isolate_lru_page)
37 * swap_lock (in swap_duplicate, swap_info_get)
38 * mmlist_lock (in mmput, drain_mmlist and others)
39 * mapping->private_lock (in __set_page_dirty_buffers)
40 * inode_lock (in set_page_dirty's __mark_inode_dirty)
41 * sb_lock (within inode_lock in fs/fs-writeback.c)
42 * mapping->tree_lock (widely used, in set_page_dirty,
43 * in arch-dependent flush_dcache_mmap_lock,
44 * within inode_lock in __sync_single_inode)
45 */
47 #include <linux/mm.h>
48 #include <linux/pagemap.h>
49 #include <linux/swap.h>
50 #include <linux/swapops.h>
51 #include <linux/slab.h>
52 #include <linux/init.h>
53 #include <linux/rmap.h>
54 #include <linux/rcupdate.h>
55 #include <linux/module.h>
57 #include <asm/tlbflush.h>
59 struct kmem_cache *anon_vma_cachep;
61 static inline void validate_anon_vma(struct vm_area_struct *find_vma)
62 {
63 #ifdef CONFIG_DEBUG_VM
64 struct anon_vma *anon_vma = find_vma->anon_vma;
65 struct vm_area_struct *vma;
66 unsigned int mapcount = 0;
67 int found = 0;
69 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
70 mapcount++;
71 BUG_ON(mapcount > 100000);
72 if (vma == find_vma)
73 found = 1;
74 }
75 BUG_ON(!found);
76 #endif
77 }
79 /* This must be called under the mmap_sem. */
80 int anon_vma_prepare(struct vm_area_struct *vma)
81 {
82 struct anon_vma *anon_vma = vma->anon_vma;
84 might_sleep();
85 if (unlikely(!anon_vma)) {
86 struct mm_struct *mm = vma->vm_mm;
87 struct anon_vma *allocated, *locked;
89 anon_vma = find_mergeable_anon_vma(vma);
90 if (anon_vma) {
91 allocated = NULL;
92 locked = anon_vma;
93 spin_lock(&locked->lock);
94 } else {
95 anon_vma = anon_vma_alloc();
96 if (unlikely(!anon_vma))
97 return -ENOMEM;
98 allocated = anon_vma;
99 locked = NULL;
100 }
102 /* page_table_lock to protect against threads */
103 spin_lock(&mm->page_table_lock);
104 if (likely(!vma->anon_vma)) {
105 vma->anon_vma = anon_vma;
106 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
107 allocated = NULL;
108 }
109 spin_unlock(&mm->page_table_lock);
111 if (locked)
112 spin_unlock(&locked->lock);
113 if (unlikely(allocated))
114 anon_vma_free(allocated);
115 }
116 return 0;
117 }
119 void __anon_vma_merge(struct vm_area_struct *vma, struct vm_area_struct *next)
120 {
121 BUG_ON(vma->anon_vma != next->anon_vma);
122 list_del(&next->anon_vma_node);
123 }
125 void __anon_vma_link(struct vm_area_struct *vma)
126 {
127 struct anon_vma *anon_vma = vma->anon_vma;
129 if (anon_vma) {
130 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
131 validate_anon_vma(vma);
132 }
133 }
135 void anon_vma_link(struct vm_area_struct *vma)
136 {
137 struct anon_vma *anon_vma = vma->anon_vma;
139 if (anon_vma) {
140 spin_lock(&anon_vma->lock);
141 list_add_tail(&vma->anon_vma_node, &anon_vma->head);
142 validate_anon_vma(vma);
143 spin_unlock(&anon_vma->lock);
144 }
145 }
147 void anon_vma_unlink(struct vm_area_struct *vma)
148 {
149 struct anon_vma *anon_vma = vma->anon_vma;
150 int empty;
152 if (!anon_vma)
153 return;
155 spin_lock(&anon_vma->lock);
156 validate_anon_vma(vma);
157 list_del(&vma->anon_vma_node);
159 /* We must garbage collect the anon_vma if it's empty */
160 empty = list_empty(&anon_vma->head);
161 spin_unlock(&anon_vma->lock);
163 if (empty)
164 anon_vma_free(anon_vma);
165 }
167 static void anon_vma_ctor(void *data, struct kmem_cache *cachep,
168 unsigned long flags)
169 {
170 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
171 SLAB_CTOR_CONSTRUCTOR) {
172 struct anon_vma *anon_vma = data;
174 spin_lock_init(&anon_vma->lock);
175 INIT_LIST_HEAD(&anon_vma->head);
176 }
177 }
179 void __init anon_vma_init(void)
180 {
181 anon_vma_cachep = kmem_cache_create("anon_vma", sizeof(struct anon_vma),
182 0, SLAB_DESTROY_BY_RCU|SLAB_PANIC, anon_vma_ctor, NULL);
183 }
185 /*
186 * Getting a lock on a stable anon_vma from a page off the LRU is
187 * tricky: page_lock_anon_vma rely on RCU to guard against the races.
188 */
189 static struct anon_vma *page_lock_anon_vma(struct page *page)
190 {
191 struct anon_vma *anon_vma = NULL;
192 unsigned long anon_mapping;
194 rcu_read_lock();
195 anon_mapping = (unsigned long) page->mapping;
196 if (!(anon_mapping & PAGE_MAPPING_ANON))
197 goto out;
198 if (!page_mapped(page))
199 goto out;
201 anon_vma = (struct anon_vma *) (anon_mapping - PAGE_MAPPING_ANON);
202 spin_lock(&anon_vma->lock);
203 out:
204 rcu_read_unlock();
205 return anon_vma;
206 }
208 /*
209 * At what user virtual address is page expected in vma?
210 */
211 static inline unsigned long
212 vma_address(struct page *page, struct vm_area_struct *vma)
213 {
214 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
215 unsigned long address;
217 address = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
218 if (unlikely(address < vma->vm_start || address >= vma->vm_end)) {
219 /* page should be within any vma from prio_tree_next */
220 BUG_ON(!PageAnon(page));
221 return -EFAULT;
222 }
223 return address;
224 }
226 /*
227 * At what user virtual address is page expected in vma? checking that the
228 * page matches the vma: currently only used on anon pages, by unuse_vma;
229 */
230 unsigned long page_address_in_vma(struct page *page, struct vm_area_struct *vma)
231 {
232 if (PageAnon(page)) {
233 if ((void *)vma->anon_vma !=
234 (void *)page->mapping - PAGE_MAPPING_ANON)
235 return -EFAULT;
236 } else if (page->mapping && !(vma->vm_flags & VM_NONLINEAR)) {
237 if (!vma->vm_file ||
238 vma->vm_file->f_mapping != page->mapping)
239 return -EFAULT;
240 } else
241 return -EFAULT;
242 return vma_address(page, vma);
243 }
245 /*
246 * Check that @page is mapped at @address into @mm.
247 *
248 * On success returns with pte mapped and locked.
249 */
250 pte_t *page_check_address(struct page *page, struct mm_struct *mm,
251 unsigned long address, spinlock_t **ptlp)
252 {
253 pgd_t *pgd;
254 pud_t *pud;
255 pmd_t *pmd;
256 pte_t *pte;
257 spinlock_t *ptl;
259 pgd = pgd_offset(mm, address);
260 if (!pgd_present(*pgd))
261 return NULL;
263 pud = pud_offset(pgd, address);
264 if (!pud_present(*pud))
265 return NULL;
267 pmd = pmd_offset(pud, address);
268 if (!pmd_present(*pmd))
269 return NULL;
271 pte = pte_offset_map(pmd, address);
272 /* Make a quick check before getting the lock */
273 if (!pte_present(*pte)) {
274 pte_unmap(pte);
275 return NULL;
276 }
278 ptl = pte_lockptr(mm, pmd);
279 spin_lock(ptl);
280 if (pte_present(*pte) && page_to_pfn(page) == pte_pfn(*pte)) {
281 *ptlp = ptl;
282 return pte;
283 }
284 pte_unmap_unlock(pte, ptl);
285 return NULL;
286 }
288 /*
289 * Subfunctions of page_referenced: page_referenced_one called
290 * repeatedly from either page_referenced_anon or page_referenced_file.
291 */
292 static int page_referenced_one(struct page *page,
293 struct vm_area_struct *vma, unsigned int *mapcount)
294 {
295 struct mm_struct *mm = vma->vm_mm;
296 unsigned long address;
297 pte_t *pte;
298 spinlock_t *ptl;
299 int referenced = 0;
301 address = vma_address(page, vma);
302 if (address == -EFAULT)
303 goto out;
305 pte = page_check_address(page, mm, address, &ptl);
306 if (!pte)
307 goto out;
309 if (ptep_clear_flush_young(vma, address, pte))
310 referenced++;
312 /* Pretend the page is referenced if the task has the
313 swap token and is in the middle of a page fault. */
314 if (mm != current->mm && has_swap_token(mm) &&
315 rwsem_is_locked(&mm->mmap_sem))
316 referenced++;
318 (*mapcount)--;
319 pte_unmap_unlock(pte, ptl);
320 out:
321 return referenced;
322 }
324 static int page_referenced_anon(struct page *page)
325 {
326 unsigned int mapcount;
327 struct anon_vma *anon_vma;
328 struct vm_area_struct *vma;
329 int referenced = 0;
331 anon_vma = page_lock_anon_vma(page);
332 if (!anon_vma)
333 return referenced;
335 mapcount = page_mapcount(page);
336 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
337 referenced += page_referenced_one(page, vma, &mapcount);
338 if (!mapcount)
339 break;
340 }
341 spin_unlock(&anon_vma->lock);
342 return referenced;
343 }
345 /**
346 * page_referenced_file - referenced check for object-based rmap
347 * @page: the page we're checking references on.
348 *
349 * For an object-based mapped page, find all the places it is mapped and
350 * check/clear the referenced flag. This is done by following the page->mapping
351 * pointer, then walking the chain of vmas it holds. It returns the number
352 * of references it found.
353 *
354 * This function is only called from page_referenced for object-based pages.
355 */
356 static int page_referenced_file(struct page *page)
357 {
358 unsigned int mapcount;
359 struct address_space *mapping = page->mapping;
360 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
361 struct vm_area_struct *vma;
362 struct prio_tree_iter iter;
363 int referenced = 0;
365 /*
366 * The caller's checks on page->mapping and !PageAnon have made
367 * sure that this is a file page: the check for page->mapping
368 * excludes the case just before it gets set on an anon page.
369 */
370 BUG_ON(PageAnon(page));
372 /*
373 * The page lock not only makes sure that page->mapping cannot
374 * suddenly be NULLified by truncation, it makes sure that the
375 * structure at mapping cannot be freed and reused yet,
376 * so we can safely take mapping->i_mmap_lock.
377 */
378 BUG_ON(!PageLocked(page));
380 spin_lock(&mapping->i_mmap_lock);
382 /*
383 * i_mmap_lock does not stabilize mapcount at all, but mapcount
384 * is more likely to be accurate if we note it after spinning.
385 */
386 mapcount = page_mapcount(page);
388 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
389 if ((vma->vm_flags & (VM_LOCKED|VM_MAYSHARE))
390 == (VM_LOCKED|VM_MAYSHARE)) {
391 referenced++;
392 break;
393 }
394 referenced += page_referenced_one(page, vma, &mapcount);
395 if (!mapcount)
396 break;
397 }
399 spin_unlock(&mapping->i_mmap_lock);
400 return referenced;
401 }
403 /**
404 * page_referenced - test if the page was referenced
405 * @page: the page to test
406 * @is_locked: caller holds lock on the page
407 *
408 * Quick test_and_clear_referenced for all mappings to a page,
409 * returns the number of ptes which referenced the page.
410 */
411 int page_referenced(struct page *page, int is_locked)
412 {
413 int referenced = 0;
415 if (page_test_and_clear_young(page))
416 referenced++;
418 if (TestClearPageReferenced(page))
419 referenced++;
421 if (page_mapped(page) && page->mapping) {
422 if (PageAnon(page))
423 referenced += page_referenced_anon(page);
424 else if (is_locked)
425 referenced += page_referenced_file(page);
426 else if (TestSetPageLocked(page))
427 referenced++;
428 else {
429 if (page->mapping)
430 referenced += page_referenced_file(page);
431 unlock_page(page);
432 }
433 }
434 return referenced;
435 }
437 /**
438 * page_set_anon_rmap - setup new anonymous rmap
439 * @page: the page to add the mapping to
440 * @vma: the vm area in which the mapping is added
441 * @address: the user virtual address mapped
442 */
443 static void __page_set_anon_rmap(struct page *page,
444 struct vm_area_struct *vma, unsigned long address)
445 {
446 struct anon_vma *anon_vma = vma->anon_vma;
448 BUG_ON(!anon_vma);
449 anon_vma = (void *) anon_vma + PAGE_MAPPING_ANON;
450 page->mapping = (struct address_space *) anon_vma;
452 page->index = linear_page_index(vma, address);
454 /*
455 * nr_mapped state can be updated without turning off
456 * interrupts because it is not modified via interrupt.
457 */
458 __inc_zone_page_state(page, NR_ANON_PAGES);
459 }
461 /**
462 * page_add_anon_rmap - add pte mapping to an anonymous page
463 * @page: the page to add the mapping to
464 * @vma: the vm area in which the mapping is added
465 * @address: the user virtual address mapped
466 *
467 * The caller needs to hold the pte lock.
468 */
469 void page_add_anon_rmap(struct page *page,
470 struct vm_area_struct *vma, unsigned long address)
471 {
472 if (atomic_inc_and_test(&page->_mapcount))
473 __page_set_anon_rmap(page, vma, address);
474 /* else checking page index and mapping is racy */
475 }
477 /*
478 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
479 * @page: the page to add the mapping to
480 * @vma: the vm area in which the mapping is added
481 * @address: the user virtual address mapped
482 *
483 * Same as page_add_anon_rmap but must only be called on *new* pages.
484 * This means the inc-and-test can be bypassed.
485 */
486 void page_add_new_anon_rmap(struct page *page,
487 struct vm_area_struct *vma, unsigned long address)
488 {
489 atomic_set(&page->_mapcount, 0); /* elevate count by 1 (starts at -1) */
490 __page_set_anon_rmap(page, vma, address);
491 }
493 /**
494 * page_add_file_rmap - add pte mapping to a file page
495 * @page: the page to add the mapping to
496 *
497 * The caller needs to hold the pte lock.
498 */
499 void page_add_file_rmap(struct page *page)
500 {
501 if (atomic_inc_and_test(&page->_mapcount))
502 __inc_zone_page_state(page, NR_FILE_MAPPED);
503 }
505 /**
506 * page_remove_rmap - take down pte mapping from a page
507 * @page: page to remove mapping from
508 *
509 * The caller needs to hold the pte lock.
510 */
511 void page_remove_rmap(struct page *page)
512 {
513 if (atomic_add_negative(-1, &page->_mapcount)) {
514 #ifdef CONFIG_DEBUG_VM
515 if (unlikely(page_mapcount(page) < 0)) {
516 printk (KERN_EMERG "Eeek! page_mapcount(page) went negative! (%d)\n", page_mapcount(page));
517 printk (KERN_EMERG " page->flags = %lx\n", page->flags);
518 printk (KERN_EMERG " page->count = %x\n", page_count(page));
519 printk (KERN_EMERG " page->mapping = %p\n", page->mapping);
520 }
521 #endif
522 BUG_ON(page_mapcount(page) < 0);
523 /*
524 * It would be tidy to reset the PageAnon mapping here,
525 * but that might overwrite a racing page_add_anon_rmap
526 * which increments mapcount after us but sets mapping
527 * before us: so leave the reset to free_hot_cold_page,
528 * and remember that it's only reliable while mapped.
529 * Leaving it set also helps swapoff to reinstate ptes
530 * faster for those pages still in swapcache.
531 */
532 if (page_test_and_clear_dirty(page))
533 set_page_dirty(page);
534 __dec_zone_page_state(page,
535 PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED);
536 }
537 }
539 /*
540 * Subfunctions of try_to_unmap: try_to_unmap_one called
541 * repeatedly from either try_to_unmap_anon or try_to_unmap_file.
542 */
543 static int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
544 int migration)
545 {
546 struct mm_struct *mm = vma->vm_mm;
547 unsigned long address;
548 pte_t *pte;
549 pte_t pteval;
550 spinlock_t *ptl;
551 int ret = SWAP_AGAIN;
553 address = vma_address(page, vma);
554 if (address == -EFAULT)
555 goto out;
557 pte = page_check_address(page, mm, address, &ptl);
558 if (!pte)
559 goto out;
561 /*
562 * If the page is mlock()d, we cannot swap it out.
563 * If it's recently referenced (perhaps page_referenced
564 * skipped over this mm) then we should reactivate it.
565 */
566 if (!migration && ((vma->vm_flags & VM_LOCKED) ||
567 (ptep_clear_flush_young(vma, address, pte)))) {
568 ret = SWAP_FAIL;
569 goto out_unmap;
570 }
572 /* Nuke the page table entry. */
573 flush_cache_page(vma, address, page_to_pfn(page));
574 pteval = ptep_clear_flush(vma, address, pte);
576 /* Move the dirty bit to the physical page now the pte is gone. */
577 if (pte_dirty(pteval))
578 set_page_dirty(page);
580 /* Update high watermark before we lower rss */
581 update_hiwater_rss(mm);
583 if (PageAnon(page)) {
584 swp_entry_t entry = { .val = page_private(page) };
586 if (PageSwapCache(page)) {
587 /*
588 * Store the swap location in the pte.
589 * See handle_pte_fault() ...
590 */
591 swap_duplicate(entry);
592 if (list_empty(&mm->mmlist)) {
593 spin_lock(&mmlist_lock);
594 if (list_empty(&mm->mmlist))
595 list_add(&mm->mmlist, &init_mm.mmlist);
596 spin_unlock(&mmlist_lock);
597 }
598 dec_mm_counter(mm, anon_rss);
599 #ifdef CONFIG_MIGRATION
600 } else {
601 /*
602 * Store the pfn of the page in a special migration
603 * pte. do_swap_page() will wait until the migration
604 * pte is removed and then restart fault handling.
605 */
606 BUG_ON(!migration);
607 entry = make_migration_entry(page, pte_write(pteval));
608 #endif
609 }
610 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
611 BUG_ON(pte_file(*pte));
612 } else
613 #ifdef CONFIG_MIGRATION
614 if (migration) {
615 /* Establish migration entry for a file page */
616 swp_entry_t entry;
617 entry = make_migration_entry(page, pte_write(pteval));
618 set_pte_at(mm, address, pte, swp_entry_to_pte(entry));
619 } else
620 #endif
621 dec_mm_counter(mm, file_rss);
624 page_remove_rmap(page);
625 page_cache_release(page);
627 out_unmap:
628 pte_unmap_unlock(pte, ptl);
629 out:
630 return ret;
631 }
633 /*
634 * objrmap doesn't work for nonlinear VMAs because the assumption that
635 * offset-into-file correlates with offset-into-virtual-addresses does not hold.
636 * Consequently, given a particular page and its ->index, we cannot locate the
637 * ptes which are mapping that page without an exhaustive linear search.
638 *
639 * So what this code does is a mini "virtual scan" of each nonlinear VMA which
640 * maps the file to which the target page belongs. The ->vm_private_data field
641 * holds the current cursor into that scan. Successive searches will circulate
642 * around the vma's virtual address space.
643 *
644 * So as more replacement pressure is applied to the pages in a nonlinear VMA,
645 * more scanning pressure is placed against them as well. Eventually pages
646 * will become fully unmapped and are eligible for eviction.
647 *
648 * For very sparsely populated VMAs this is a little inefficient - chances are
649 * there there won't be many ptes located within the scan cluster. In this case
650 * maybe we could scan further - to the end of the pte page, perhaps.
651 */
652 #define CLUSTER_SIZE min(32*PAGE_SIZE, PMD_SIZE)
653 #define CLUSTER_MASK (~(CLUSTER_SIZE - 1))
655 static void try_to_unmap_cluster(unsigned long cursor,
656 unsigned int *mapcount, struct vm_area_struct *vma)
657 {
658 struct mm_struct *mm = vma->vm_mm;
659 pgd_t *pgd;
660 pud_t *pud;
661 pmd_t *pmd;
662 pte_t *pte;
663 pte_t pteval;
664 spinlock_t *ptl;
665 struct page *page;
666 unsigned long address;
667 unsigned long end;
669 address = (vma->vm_start + cursor) & CLUSTER_MASK;
670 end = address + CLUSTER_SIZE;
671 if (address < vma->vm_start)
672 address = vma->vm_start;
673 if (end > vma->vm_end)
674 end = vma->vm_end;
676 pgd = pgd_offset(mm, address);
677 if (!pgd_present(*pgd))
678 return;
680 pud = pud_offset(pgd, address);
681 if (!pud_present(*pud))
682 return;
684 pmd = pmd_offset(pud, address);
685 if (!pmd_present(*pmd))
686 return;
688 pte = pte_offset_map_lock(mm, pmd, address, &ptl);
690 /* Update high watermark before we lower rss */
691 update_hiwater_rss(mm);
693 for (; address < end; pte++, address += PAGE_SIZE) {
694 if (!pte_present(*pte))
695 continue;
696 page = vm_normal_page(vma, address, *pte);
697 BUG_ON(!page || PageAnon(page));
699 if (ptep_clear_flush_young(vma, address, pte))
700 continue;
702 /* Nuke the page table entry. */
703 flush_cache_page(vma, address, pte_pfn(*pte));
704 pteval = ptep_clear_flush(vma, address, pte);
706 /* If nonlinear, store the file page offset in the pte. */
707 if (page->index != linear_page_index(vma, address))
708 set_pte_at(mm, address, pte, pgoff_to_pte(page->index));
710 /* Move the dirty bit to the physical page now the pte is gone. */
711 if (pte_dirty(pteval))
712 set_page_dirty(page);
714 page_remove_rmap(page);
715 page_cache_release(page);
716 dec_mm_counter(mm, file_rss);
717 (*mapcount)--;
718 }
719 pte_unmap_unlock(pte - 1, ptl);
720 }
722 static int try_to_unmap_anon(struct page *page, int migration)
723 {
724 struct anon_vma *anon_vma;
725 struct vm_area_struct *vma;
726 int ret = SWAP_AGAIN;
728 anon_vma = page_lock_anon_vma(page);
729 if (!anon_vma)
730 return ret;
732 list_for_each_entry(vma, &anon_vma->head, anon_vma_node) {
733 ret = try_to_unmap_one(page, vma, migration);
734 if (ret == SWAP_FAIL || !page_mapped(page))
735 break;
736 }
737 spin_unlock(&anon_vma->lock);
738 return ret;
739 }
741 /**
742 * try_to_unmap_file - unmap file page using the object-based rmap method
743 * @page: the page to unmap
744 *
745 * Find all the mappings of a page using the mapping pointer and the vma chains
746 * contained in the address_space struct it points to.
747 *
748 * This function is only called from try_to_unmap for object-based pages.
749 */
750 static int try_to_unmap_file(struct page *page, int migration)
751 {
752 struct address_space *mapping = page->mapping;
753 pgoff_t pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
754 struct vm_area_struct *vma;
755 struct prio_tree_iter iter;
756 int ret = SWAP_AGAIN;
757 unsigned long cursor;
758 unsigned long max_nl_cursor = 0;
759 unsigned long max_nl_size = 0;
760 unsigned int mapcount;
762 spin_lock(&mapping->i_mmap_lock);
763 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
764 ret = try_to_unmap_one(page, vma, migration);
765 if (ret == SWAP_FAIL || !page_mapped(page))
766 goto out;
767 }
769 if (list_empty(&mapping->i_mmap_nonlinear))
770 goto out;
772 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
773 shared.vm_set.list) {
774 if ((vma->vm_flags & VM_LOCKED) && !migration)
775 continue;
776 cursor = (unsigned long) vma->vm_private_data;
777 if (cursor > max_nl_cursor)
778 max_nl_cursor = cursor;
779 cursor = vma->vm_end - vma->vm_start;
780 if (cursor > max_nl_size)
781 max_nl_size = cursor;
782 }
784 if (max_nl_size == 0) { /* any nonlinears locked or reserved */
785 ret = SWAP_FAIL;
786 goto out;
787 }
789 /*
790 * We don't try to search for this page in the nonlinear vmas,
791 * and page_referenced wouldn't have found it anyway. Instead
792 * just walk the nonlinear vmas trying to age and unmap some.
793 * The mapcount of the page we came in with is irrelevant,
794 * but even so use it as a guide to how hard we should try?
795 */
796 mapcount = page_mapcount(page);
797 if (!mapcount)
798 goto out;
799 cond_resched_lock(&mapping->i_mmap_lock);
801 max_nl_size = (max_nl_size + CLUSTER_SIZE - 1) & CLUSTER_MASK;
802 if (max_nl_cursor == 0)
803 max_nl_cursor = CLUSTER_SIZE;
805 do {
806 list_for_each_entry(vma, &mapping->i_mmap_nonlinear,
807 shared.vm_set.list) {
808 if ((vma->vm_flags & VM_LOCKED) && !migration)
809 continue;
810 cursor = (unsigned long) vma->vm_private_data;
811 while ( cursor < max_nl_cursor &&
812 cursor < vma->vm_end - vma->vm_start) {
813 try_to_unmap_cluster(cursor, &mapcount, vma);
814 cursor += CLUSTER_SIZE;
815 vma->vm_private_data = (void *) cursor;
816 if ((int)mapcount <= 0)
817 goto out;
818 }
819 vma->vm_private_data = (void *) max_nl_cursor;
820 }
821 cond_resched_lock(&mapping->i_mmap_lock);
822 max_nl_cursor += CLUSTER_SIZE;
823 } while (max_nl_cursor <= max_nl_size);
825 /*
826 * Don't loop forever (perhaps all the remaining pages are
827 * in locked vmas). Reset cursor on all unreserved nonlinear
828 * vmas, now forgetting on which ones it had fallen behind.
829 */
830 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
831 vma->vm_private_data = NULL;
832 out:
833 spin_unlock(&mapping->i_mmap_lock);
834 return ret;
835 }
837 /**
838 * try_to_unmap - try to remove all page table mappings to a page
839 * @page: the page to get unmapped
840 *
841 * Tries to remove all the page table entries which are mapping this
842 * page, used in the pageout path. Caller must hold the page lock.
843 * Return values are:
844 *
845 * SWAP_SUCCESS - we succeeded in removing all mappings
846 * SWAP_AGAIN - we missed a mapping, try again later
847 * SWAP_FAIL - the page is unswappable
848 */
849 int try_to_unmap(struct page *page, int migration)
850 {
851 int ret;
853 BUG_ON(!PageLocked(page));
855 if (PageAnon(page))
856 ret = try_to_unmap_anon(page, migration);
857 else
858 ret = try_to_unmap_file(page, migration);
860 if (!page_mapped(page))
861 ret = SWAP_SUCCESS;
862 return ret;
863 }