blob: 3bbaf5d230b0104b3c41ce7cd012079e1bda3f1c [file] [log] [blame]
Christoph Lameterb20a3502006-03-22 00:09:12 -08001/*
2 * Memory Migration functionality - linux/mm/migration.c
3 *
4 * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter
5 *
6 * Page migration was first developed in the context of the memory hotplug
7 * project. The main authors of the migration code are:
8 *
9 * IWAMOTO Toshihiro <iwamoto@valinux.co.jp>
10 * Hirokazu Takahashi <taka@valinux.co.jp>
11 * Dave Hansen <haveblue@us.ibm.com>
Christoph Lametercde53532008-07-04 09:59:22 -070012 * Christoph Lameter
Christoph Lameterb20a3502006-03-22 00:09:12 -080013 */
14
15#include <linux/migrate.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040016#include <linux/export.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080017#include <linux/swap.h>
Christoph Lameter06972122006-06-23 02:03:35 -070018#include <linux/swapops.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080019#include <linux/pagemap.h>
Christoph Lametere23ca002006-04-10 22:52:57 -070020#include <linux/buffer_head.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080021#include <linux/mm_inline.h>
Pavel Emelyanovb4888932007-10-18 23:40:14 -070022#include <linux/nsproxy.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080023#include <linux/pagevec.h>
Hugh Dickinse9995ef2009-12-14 17:59:31 -080024#include <linux/ksm.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080025#include <linux/rmap.h>
26#include <linux/topology.h>
27#include <linux/cpu.h>
28#include <linux/cpuset.h>
Christoph Lameter04e62a22006-06-23 02:03:38 -070029#include <linux/writeback.h>
Christoph Lameter742755a2006-06-23 02:03:55 -070030#include <linux/mempolicy.h>
31#include <linux/vmalloc.h>
David Quigley86c3a762006-06-23 02:04:02 -070032#include <linux/security.h>
Balbir Singh8a9f3cc2008-02-07 00:13:53 -080033#include <linux/memcontrol.h>
Adrian Bunk4f5ca262008-07-23 21:27:02 -070034#include <linux/syscalls.h>
Naoya Horiguchi290408d2010-09-08 10:19:35 +090035#include <linux/hugetlb.h>
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -070036#include <linux/hugetlb_cgroup.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/gfp.h>
Rafael Aquinibf6bddf2012-12-11 16:02:42 -080038#include <linux/balloon_compaction.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080039
Michal Nazarewicz0d1836c2010-12-21 17:24:26 -080040#include <asm/tlbflush.h>
41
Mel Gorman7b2a2d42012-10-19 14:07:31 +010042#define CREATE_TRACE_POINTS
43#include <trace/events/migrate.h>
44
Christoph Lameterb20a3502006-03-22 00:09:12 -080045#include "internal.h"
46
Christoph Lameterb20a3502006-03-22 00:09:12 -080047/*
Christoph Lameter742755a2006-06-23 02:03:55 -070048 * migrate_prep() needs to be called before we start compiling a list of pages
Mel Gorman748446b2010-05-24 14:32:27 -070049 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
50 * undesirable, use migrate_prep_local()
Christoph Lameterb20a3502006-03-22 00:09:12 -080051 */
52int migrate_prep(void)
53{
Christoph Lameterb20a3502006-03-22 00:09:12 -080054 /*
55 * Clear the LRU lists so pages can be isolated.
56 * Note that pages may be moved off the LRU after we have
57 * drained them. Those pages will fail to migrate like other
58 * pages that may be busy.
59 */
60 lru_add_drain_all();
61
62 return 0;
63}
64
Mel Gorman748446b2010-05-24 14:32:27 -070065/* Do the necessary work of migrate_prep but not if it involves other CPUs */
66int migrate_prep_local(void)
67{
68 lru_add_drain();
69
70 return 0;
71}
72
Christoph Lameterb20a3502006-03-22 00:09:12 -080073/*
Lee Schermerhorn894bc312008-10-18 20:26:39 -070074 * Add isolated pages on the list back to the LRU under page lock
75 * to avoid leaking evictable pages back onto unevictable list.
Christoph Lameterb20a3502006-03-22 00:09:12 -080076 */
Minchan Kime13861d2010-05-24 14:31:59 -070077void putback_lru_pages(struct list_head *l)
Christoph Lameterb20a3502006-03-22 00:09:12 -080078{
79 struct page *page;
80 struct page *page2;
Christoph Lameterb20a3502006-03-22 00:09:12 -080081
82 list_for_each_entry_safe(page, page2, l, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -070083 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -070084 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -070085 page_is_file_cache(page));
Rafael Aquini5733c7d2012-12-11 16:02:47 -080086 putback_lru_page(page);
87 }
88}
89
90/*
91 * Put previously isolated pages back onto the appropriate lists
92 * from where they were once taken off for compaction/migration.
93 *
94 * This function shall be used instead of putback_lru_pages(),
95 * whenever the isolated pageset has been built by isolate_migratepages_range()
96 */
97void putback_movable_pages(struct list_head *l)
98{
99 struct page *page;
100 struct page *page2;
101
102 list_for_each_entry_safe(page, page2, l, lru) {
103 list_del(&page->lru);
104 dec_zone_page_state(page, NR_ISOLATED_ANON +
105 page_is_file_cache(page));
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800106 if (unlikely(balloon_page_movable(page)))
107 balloon_page_putback(page);
108 else
109 putback_lru_page(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800110 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800111}
112
Christoph Lameter06972122006-06-23 02:03:35 -0700113/*
114 * Restore a potential migration pte to a working pte entry
115 */
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800116static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
117 unsigned long addr, void *old)
Christoph Lameter06972122006-06-23 02:03:35 -0700118{
119 struct mm_struct *mm = vma->vm_mm;
120 swp_entry_t entry;
Christoph Lameter06972122006-06-23 02:03:35 -0700121 pmd_t *pmd;
122 pte_t *ptep, pte;
123 spinlock_t *ptl;
124
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900125 if (unlikely(PageHuge(new))) {
126 ptep = huge_pte_offset(mm, addr);
127 if (!ptep)
128 goto out;
129 ptl = &mm->page_table_lock;
130 } else {
Bob Liu62190492012-12-11 16:00:37 -0800131 pmd = mm_find_pmd(mm, addr);
132 if (!pmd)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900133 goto out;
Andrea Arcangeli500d65d2011-01-13 15:46:55 -0800134 if (pmd_trans_huge(*pmd))
135 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700136
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900137 ptep = pte_offset_map(pmd, addr);
Christoph Lameter06972122006-06-23 02:03:35 -0700138
Hugh Dickins486cf462011-10-19 12:50:35 -0700139 /*
140 * Peek to check is_swap_pte() before taking ptlock? No, we
141 * can race mremap's move_ptes(), which skips anon_vma lock.
142 */
Christoph Lameter06972122006-06-23 02:03:35 -0700143
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900144 ptl = pte_lockptr(mm, pmd);
145 }
146
Christoph Lameter06972122006-06-23 02:03:35 -0700147 spin_lock(ptl);
148 pte = *ptep;
149 if (!is_swap_pte(pte))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800150 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700151
152 entry = pte_to_swp_entry(pte);
153
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800154 if (!is_migration_entry(entry) ||
155 migration_entry_to_page(entry) != old)
156 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700157
Christoph Lameter06972122006-06-23 02:03:35 -0700158 get_page(new);
159 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
160 if (is_write_migration_entry(entry))
161 pte = pte_mkwrite(pte);
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200162#ifdef CONFIG_HUGETLB_PAGE
Tony Lube7517d2013-02-04 14:28:46 -0800163 if (PageHuge(new)) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900164 pte = pte_mkhuge(pte);
Tony Lube7517d2013-02-04 14:28:46 -0800165 pte = arch_make_huge_pte(pte, vma, new, 0);
166 }
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200167#endif
KAMEZAWA Hiroyuki97ee0522007-10-16 01:25:43 -0700168 flush_cache_page(vma, addr, pte_pfn(pte));
Christoph Lameter06972122006-06-23 02:03:35 -0700169 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700170
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900171 if (PageHuge(new)) {
172 if (PageAnon(new))
173 hugepage_add_anon_rmap(new, vma, addr);
174 else
175 page_dup_rmap(new);
176 } else if (PageAnon(new))
Christoph Lameter04e62a22006-06-23 02:03:38 -0700177 page_add_anon_rmap(new, vma, addr);
178 else
179 page_add_file_rmap(new);
180
181 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +0000182 update_mmu_cache(vma, addr, ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800183unlock:
Christoph Lameter06972122006-06-23 02:03:35 -0700184 pte_unmap_unlock(ptep, ptl);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800185out:
186 return SWAP_AGAIN;
Christoph Lameter06972122006-06-23 02:03:35 -0700187}
188
189/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700190 * Get rid of all migration entries and replace them by
191 * references to the indicated page.
192 */
193static void remove_migration_ptes(struct page *old, struct page *new)
194{
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800195 rmap_walk(new, remove_migration_pte, old);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700196}
197
198/*
Christoph Lameter06972122006-06-23 02:03:35 -0700199 * Something used the pte of a page under migration. We need to
200 * get to the page and wait until migration is finished.
201 * When we return from this function the fault will be retried.
Christoph Lameter06972122006-06-23 02:03:35 -0700202 */
203void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
204 unsigned long address)
205{
206 pte_t *ptep, pte;
207 spinlock_t *ptl;
208 swp_entry_t entry;
209 struct page *page;
210
211 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
212 pte = *ptep;
213 if (!is_swap_pte(pte))
214 goto out;
215
216 entry = pte_to_swp_entry(pte);
217 if (!is_migration_entry(entry))
218 goto out;
219
220 page = migration_entry_to_page(entry);
221
Nick Piggine2867812008-07-25 19:45:30 -0700222 /*
223 * Once radix-tree replacement of page migration started, page_count
224 * *must* be zero. And, we don't want to call wait_on_page_locked()
225 * against a page without get_page().
226 * So, we use get_page_unless_zero(), here. Even failed, page fault
227 * will occur again.
228 */
229 if (!get_page_unless_zero(page))
230 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700231 pte_unmap_unlock(ptep, ptl);
232 wait_on_page_locked(page);
233 put_page(page);
234 return;
235out:
236 pte_unmap_unlock(ptep, ptl);
237}
238
Mel Gormanb969c4a2012-01-12 17:19:34 -0800239#ifdef CONFIG_BLOCK
240/* Returns true if all buffers are successfully locked */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800241static bool buffer_migrate_lock_buffers(struct buffer_head *head,
242 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800243{
244 struct buffer_head *bh = head;
245
246 /* Simple case, sync compaction */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800247 if (mode != MIGRATE_ASYNC) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800248 do {
249 get_bh(bh);
250 lock_buffer(bh);
251 bh = bh->b_this_page;
252
253 } while (bh != head);
254
255 return true;
256 }
257
258 /* async case, we cannot block on lock_buffer so use trylock_buffer */
259 do {
260 get_bh(bh);
261 if (!trylock_buffer(bh)) {
262 /*
263 * We failed to lock the buffer and cannot stall in
264 * async migration. Release the taken locks
265 */
266 struct buffer_head *failed_bh = bh;
267 put_bh(failed_bh);
268 bh = head;
269 while (bh != failed_bh) {
270 unlock_buffer(bh);
271 put_bh(bh);
272 bh = bh->b_this_page;
273 }
274 return false;
275 }
276
277 bh = bh->b_this_page;
278 } while (bh != head);
279 return true;
280}
281#else
282static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800283 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800284{
285 return true;
286}
287#endif /* CONFIG_BLOCK */
288
Christoph Lameterb20a3502006-03-22 00:09:12 -0800289/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700290 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700291 *
292 * The number of remaining references must be:
293 * 1 for anonymous pages without a mapping
294 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100295 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800296 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700297static int migrate_page_move_mapping(struct address_space *mapping,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800298 struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800299 struct buffer_head *head, enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800300{
Peter Zijlstra7039e1d2012-10-25 14:16:34 +0200301 int expected_count = 0;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800302 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800303
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700304 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700305 /* Anonymous page without mapping */
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700306 if (page_count(page) != 1)
307 return -EAGAIN;
Rafael Aquini78bd5202012-12-11 16:02:31 -0800308 return MIGRATEPAGE_SUCCESS;
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700309 }
310
Nick Piggin19fd6232008-07-25 19:45:32 -0700311 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800312
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800313 pslot = radix_tree_lookup_slot(&mapping->page_tree,
314 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800315
Johannes Weineredcf4742009-09-21 17:02:59 -0700316 expected_count = 2 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700317 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800318 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700319 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700320 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800321 }
322
Nick Piggine2867812008-07-25 19:45:30 -0700323 if (!page_freeze_refs(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700324 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700325 return -EAGAIN;
326 }
327
Christoph Lameterb20a3502006-03-22 00:09:12 -0800328 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800329 * In the async migration case of moving a page with buffers, lock the
330 * buffers using trylock before the mapping is moved. If the mapping
331 * was moved, we later failed to lock the buffers and could not move
332 * the mapping back due to an elevated page count, we would have to
333 * block waiting on other references to be dropped.
334 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800335 if (mode == MIGRATE_ASYNC && head &&
336 !buffer_migrate_lock_buffers(head, mode)) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800337 page_unfreeze_refs(page, expected_count);
338 spin_unlock_irq(&mapping->tree_lock);
339 return -EAGAIN;
340 }
341
342 /*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800343 * Now we know that no one else is looking at the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800344 */
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800345 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800346 if (PageSwapCache(page)) {
347 SetPageSwapCache(newpage);
348 set_page_private(newpage, page_private(page));
349 }
350
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800351 radix_tree_replace_slot(pslot, newpage);
352
353 /*
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800354 * Drop cache reference from old page by unfreezing
355 * to one less reference.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800356 * We know this isn't the last reference.
357 */
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800358 page_unfreeze_refs(page, expected_count - 1);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800359
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700360 /*
361 * If moved to a different zone then also account
362 * the page for that zone. Other VM counters will be
363 * taken care of when we establish references to the
364 * new page and drop references to the old page.
365 *
366 * Note that anonymous pages are accounted for
367 * via NR_FILE_PAGES and NR_ANON_PAGES if they
368 * are mapped to swap space.
369 */
370 __dec_zone_page_state(page, NR_FILE_PAGES);
371 __inc_zone_page_state(newpage, NR_FILE_PAGES);
Andrea Arcangeli99a15e22011-06-16 12:56:19 -0700372 if (!PageSwapCache(page) && PageSwapBacked(page)) {
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700373 __dec_zone_page_state(page, NR_SHMEM);
374 __inc_zone_page_state(newpage, NR_SHMEM);
375 }
Nick Piggin19fd6232008-07-25 19:45:32 -0700376 spin_unlock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800377
Rafael Aquini78bd5202012-12-11 16:02:31 -0800378 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800379}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800380
381/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900382 * The expected number of remaining references is the same as that
383 * of migrate_page_move_mapping().
384 */
385int migrate_huge_page_move_mapping(struct address_space *mapping,
386 struct page *newpage, struct page *page)
387{
388 int expected_count;
389 void **pslot;
390
391 if (!mapping) {
392 if (page_count(page) != 1)
393 return -EAGAIN;
Rafael Aquini78bd5202012-12-11 16:02:31 -0800394 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900395 }
396
397 spin_lock_irq(&mapping->tree_lock);
398
399 pslot = radix_tree_lookup_slot(&mapping->page_tree,
400 page_index(page));
401
402 expected_count = 2 + page_has_private(page);
403 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800404 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900405 spin_unlock_irq(&mapping->tree_lock);
406 return -EAGAIN;
407 }
408
409 if (!page_freeze_refs(page, expected_count)) {
410 spin_unlock_irq(&mapping->tree_lock);
411 return -EAGAIN;
412 }
413
414 get_page(newpage);
415
416 radix_tree_replace_slot(pslot, newpage);
417
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800418 page_unfreeze_refs(page, expected_count - 1);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900419
420 spin_unlock_irq(&mapping->tree_lock);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800421 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900422}
423
424/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800425 * Copy the page to its new location
426 */
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900427void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800428{
Mel Gormanb32967f2012-11-19 12:35:47 +0000429 if (PageHuge(page) || PageTransHuge(page))
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900430 copy_huge_page(newpage, page);
431 else
432 copy_highpage(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800433
434 if (PageError(page))
435 SetPageError(newpage);
436 if (PageReferenced(page))
437 SetPageReferenced(newpage);
438 if (PageUptodate(page))
439 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700440 if (TestClearPageActive(page)) {
441 VM_BUG_ON(PageUnevictable(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800442 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800443 } else if (TestClearPageUnevictable(page))
444 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800445 if (PageChecked(page))
446 SetPageChecked(newpage);
447 if (PageMappedToDisk(page))
448 SetPageMappedToDisk(newpage);
449
450 if (PageDirty(page)) {
451 clear_page_dirty_for_io(page);
Nick Piggin3a902c52008-04-30 00:55:16 -0700452 /*
453 * Want to mark the page and the radix tree as dirty, and
454 * redo the accounting that clear_page_dirty_for_io undid,
455 * but we can't use set_page_dirty because that function
456 * is actually a signal that all of the page has become dirty.
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300457 * Whereas only part of our page may be dirty.
Nick Piggin3a902c52008-04-30 00:55:16 -0700458 */
Hugh Dickins752dc182012-06-02 00:27:47 -0700459 if (PageSwapBacked(page))
460 SetPageDirty(newpage);
461 else
462 __set_page_dirty_nobuffers(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800463 }
464
Nick Pigginb291f002008-10-18 20:26:44 -0700465 mlock_migrate_page(newpage, page);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800466 ksm_migrate_page(newpage, page);
Hugh Dickinsc8d65532013-02-22 16:35:10 -0800467 /*
468 * Please do not reorder this without considering how mm/ksm.c's
469 * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache().
470 */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800471 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800472 ClearPagePrivate(page);
473 set_page_private(page, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800474
475 /*
476 * If any waiters have accumulated on the new page then
477 * wake them up.
478 */
479 if (PageWriteback(newpage))
480 end_page_writeback(newpage);
481}
Christoph Lameterb20a3502006-03-22 00:09:12 -0800482
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700483/************************************************************
484 * Migration functions
485 ***********************************************************/
486
487/* Always fail migration. Used for mappings that are not movable */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700488int fail_migrate_page(struct address_space *mapping,
489 struct page *newpage, struct page *page)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700490{
491 return -EIO;
492}
493EXPORT_SYMBOL(fail_migrate_page);
494
Christoph Lameterb20a3502006-03-22 00:09:12 -0800495/*
496 * Common logic to directly migrate a single page suitable for
David Howells266cf652009-04-03 16:42:36 +0100497 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800498 *
499 * Pages are locked upon entry and exit.
500 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700501int migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800502 struct page *newpage, struct page *page,
503 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800504{
505 int rc;
506
507 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
508
Mel Gormana6bc32b2012-01-12 17:19:43 -0800509 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800510
Rafael Aquini78bd5202012-12-11 16:02:31 -0800511 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800512 return rc;
513
514 migrate_page_copy(newpage, page);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800515 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800516}
517EXPORT_SYMBOL(migrate_page);
518
David Howells93614012006-09-30 20:45:40 +0200519#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800520/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700521 * Migration function for pages with buffers. This function can only be used
522 * if the underlying filesystem guarantees that no other references to "page"
523 * exist.
524 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700525int buffer_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800526 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700527{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700528 struct buffer_head *bh, *head;
529 int rc;
530
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700531 if (!page_has_buffers(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800532 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700533
534 head = page_buffers(page);
535
Mel Gormana6bc32b2012-01-12 17:19:43 -0800536 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700537
Rafael Aquini78bd5202012-12-11 16:02:31 -0800538 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700539 return rc;
540
Mel Gormanb969c4a2012-01-12 17:19:34 -0800541 /*
542 * In the async case, migrate_page_move_mapping locked the buffers
543 * with an IRQ-safe spinlock held. In the sync case, the buffers
544 * need to be locked now
545 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800546 if (mode != MIGRATE_ASYNC)
547 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700548
549 ClearPagePrivate(page);
550 set_page_private(newpage, page_private(page));
551 set_page_private(page, 0);
552 put_page(page);
553 get_page(newpage);
554
555 bh = head;
556 do {
557 set_bh_page(bh, newpage, bh_offset(bh));
558 bh = bh->b_this_page;
559
560 } while (bh != head);
561
562 SetPagePrivate(newpage);
563
564 migrate_page_copy(newpage, page);
565
566 bh = head;
567 do {
568 unlock_buffer(bh);
569 put_bh(bh);
570 bh = bh->b_this_page;
571
572 } while (bh != head);
573
Rafael Aquini78bd5202012-12-11 16:02:31 -0800574 return MIGRATEPAGE_SUCCESS;
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700575}
576EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200577#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700578
Christoph Lameter04e62a22006-06-23 02:03:38 -0700579/*
580 * Writeback a page to clean the dirty state
581 */
582static int writeout(struct address_space *mapping, struct page *page)
583{
584 struct writeback_control wbc = {
585 .sync_mode = WB_SYNC_NONE,
586 .nr_to_write = 1,
587 .range_start = 0,
588 .range_end = LLONG_MAX,
Christoph Lameter04e62a22006-06-23 02:03:38 -0700589 .for_reclaim = 1
590 };
591 int rc;
592
593 if (!mapping->a_ops->writepage)
594 /* No write method for the address space */
595 return -EINVAL;
596
597 if (!clear_page_dirty_for_io(page))
598 /* Someone else already triggered a write */
599 return -EAGAIN;
600
601 /*
602 * A dirty page may imply that the underlying filesystem has
603 * the page on some queue. So the page must be clean for
604 * migration. Writeout may mean we loose the lock and the
605 * page state is no longer what we checked for earlier.
606 * At this point we know that the migration attempt cannot
607 * be successful.
608 */
609 remove_migration_ptes(page, page);
610
611 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700612
613 if (rc != AOP_WRITEPAGE_ACTIVATE)
614 /* unlocked. Relock */
615 lock_page(page);
616
Hugh Dickinsbda85502008-11-19 15:36:36 -0800617 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700618}
619
620/*
621 * Default handling if a filesystem does not provide a migration function.
622 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700623static int fallback_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800624 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700625{
Mel Gormanb969c4a2012-01-12 17:19:34 -0800626 if (PageDirty(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800627 /* Only writeback pages in full synchronous migration */
628 if (mode != MIGRATE_SYNC)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800629 return -EBUSY;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700630 return writeout(mapping, page);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800631 }
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700632
633 /*
634 * Buffers may be managed in a filesystem specific way.
635 * We must have no buffers or drop them.
636 */
David Howells266cf652009-04-03 16:42:36 +0100637 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700638 !try_to_release_page(page, GFP_KERNEL))
639 return -EAGAIN;
640
Mel Gormana6bc32b2012-01-12 17:19:43 -0800641 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700642}
643
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700644/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700645 * Move a page to a newly allocated page
646 * The page is locked and all ptes have been successfully removed.
647 *
648 * The new page will have replaced the old page if this function
649 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700650 *
651 * Return value:
652 * < 0 - error code
Rafael Aquini78bd5202012-12-11 16:02:31 -0800653 * MIGRATEPAGE_SUCCESS - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700654 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700655static int move_to_new_page(struct page *newpage, struct page *page,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800656 int remap_swapcache, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700657{
658 struct address_space *mapping;
659 int rc;
660
661 /*
662 * Block others from accessing the page when we get around to
663 * establishing additional references. We are the only one
664 * holding a reference to the new page at this point.
665 */
Nick Piggin529ae9a2008-08-02 12:01:03 +0200666 if (!trylock_page(newpage))
Christoph Lametere24f0b82006-06-23 02:03:51 -0700667 BUG();
668
669 /* Prepare mapping for the new page.*/
670 newpage->index = page->index;
671 newpage->mapping = page->mapping;
Rik van Rielb2e18532008-10-18 20:26:30 -0700672 if (PageSwapBacked(page))
673 SetPageSwapBacked(newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700674
675 mapping = page_mapping(page);
676 if (!mapping)
Mel Gormana6bc32b2012-01-12 17:19:43 -0800677 rc = migrate_page(mapping, newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800678 else if (mapping->a_ops->migratepage)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700679 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800680 * Most pages have a mapping and most filesystems provide a
681 * migratepage callback. Anonymous pages are part of swap
682 * space which also has its own migratepage callback. This
683 * is the most common path for page migration.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700684 */
Mel Gormanb969c4a2012-01-12 17:19:34 -0800685 rc = mapping->a_ops->migratepage(mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800686 newpage, page, mode);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800687 else
Mel Gormana6bc32b2012-01-12 17:19:43 -0800688 rc = fallback_migrate_page(mapping, newpage, page, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700689
Rafael Aquini78bd5202012-12-11 16:02:31 -0800690 if (rc != MIGRATEPAGE_SUCCESS) {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700691 newpage->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700692 } else {
693 if (remap_swapcache)
694 remove_migration_ptes(page, newpage);
Konstantin Khlebnikov35512ec2012-02-03 15:37:13 -0800695 page->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700696 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700697
698 unlock_page(newpage);
699
700 return rc;
701}
702
Minchan Kim0dabec92011-10-31 17:06:57 -0700703static int __unmap_and_move(struct page *page, struct page *newpage,
Hugh Dickins9c620e22013-02-22 16:35:14 -0800704 int force, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700705{
Minchan Kim0dabec92011-10-31 17:06:57 -0700706 int rc = -EAGAIN;
Mel Gorman3fe20112010-05-24 14:32:20 -0700707 int remap_swapcache = 1;
KAMEZAWA Hiroyuki56039ef2011-03-23 16:42:19 -0700708 struct mem_cgroup *mem;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700709 struct anon_vma *anon_vma = NULL;
Christoph Lameter95a402c2006-06-23 02:03:53 -0700710
Nick Piggin529ae9a2008-08-02 12:01:03 +0200711 if (!trylock_page(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800712 if (!force || mode == MIGRATE_ASYNC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700713 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800714
715 /*
716 * It's not safe for direct compaction to call lock_page.
717 * For example, during page readahead pages are added locked
718 * to the LRU. Later, when the IO completes the pages are
719 * marked uptodate and unlocked. However, the queueing
720 * could be merging multiple pages for one bio (e.g.
721 * mpage_readpages). If an allocation happens for the
722 * second or third page, the process can end up locking
723 * the same page twice and deadlocking. Rather than
724 * trying to be clever about what pages can be locked,
725 * avoid the use of lock_page for direct compaction
726 * altogether.
727 */
728 if (current->flags & PF_MEMALLOC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700729 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800730
Christoph Lametere24f0b82006-06-23 02:03:51 -0700731 lock_page(page);
732 }
733
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800734 /* charge against new page */
Johannes Weiner0030f532012-07-31 16:45:25 -0700735 mem_cgroup_prepare_migration(page, newpage, &mem);
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800736
Christoph Lametere24f0b82006-06-23 02:03:51 -0700737 if (PageWriteback(page)) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700738 /*
Mel Gormana6bc32b2012-01-12 17:19:43 -0800739 * Only in the case of a full syncronous migration is it
740 * necessary to wait for PageWriteback. In the async case,
741 * the retry loop is too short and in the sync-light case,
742 * the overhead of stalling is too much
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700743 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800744 if (mode != MIGRATE_SYNC) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700745 rc = -EBUSY;
746 goto uncharge;
747 }
748 if (!force)
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800749 goto uncharge;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700750 wait_on_page_writeback(page);
751 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700752 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700753 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
754 * we cannot notice that anon_vma is freed while we migrates a page.
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800755 * This get_anon_vma() delays freeing anon_vma pointer until the end
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700756 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700757 * File Caches may use write_page() or lock_page() in migration, then,
758 * just care Anon page here.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700759 */
Hugh Dickinsb79bc0a2013-02-22 16:35:13 -0800760 if (PageAnon(page) && !PageKsm(page)) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800761 /*
Ingo Molnar4fc3f1d2012-12-02 19:56:50 +0000762 * Only page_lock_anon_vma_read() understands the subtleties of
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800763 * getting a hold on an anon_vma from outside one of its mms.
764 */
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700765 anon_vma = page_get_anon_vma(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800766 if (anon_vma) {
767 /*
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700768 * Anon page
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800769 */
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800770 } else if (PageSwapCache(page)) {
Mel Gorman3fe20112010-05-24 14:32:20 -0700771 /*
772 * We cannot be sure that the anon_vma of an unmapped
773 * swapcache page is safe to use because we don't
774 * know in advance if the VMA that this page belonged
775 * to still exists. If the VMA and others sharing the
776 * data have been freed, then the anon_vma could
777 * already be invalid.
778 *
779 * To avoid this possibility, swapcache pages get
780 * migrated but are not remapped when migration
781 * completes
782 */
783 remap_swapcache = 0;
784 } else {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800785 goto uncharge;
Mel Gorman3fe20112010-05-24 14:32:20 -0700786 }
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700787 }
Shaohua Li62e1c552008-02-04 22:29:33 -0800788
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800789 if (unlikely(balloon_page_movable(page))) {
790 /*
791 * A ballooned page does not need any special attention from
792 * physical to virtual reverse mapping procedures.
793 * Skip any attempt to unmap PTEs or to remap swap cache,
794 * in order to avoid burning cycles at rmap level, and perform
795 * the page migration right away (proteced by page lock).
796 */
797 rc = balloon_page_migrate(newpage, page, mode);
798 goto uncharge;
799 }
800
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700801 /*
Shaohua Li62e1c552008-02-04 22:29:33 -0800802 * Corner case handling:
803 * 1. When a new swap-cache page is read into, it is added to the LRU
804 * and treated as swapcache but it has no rmap yet.
805 * Calling try_to_unmap() against a page->mapping==NULL page will
806 * trigger a BUG. So handle it here.
807 * 2. An orphaned page (see truncate_complete_page) might have
808 * fs-private metadata. The page can be picked up due to memory
809 * offlining. Everywhere else except page reclaim, the page is
810 * invisible to the vm, so the page can not be migrated. So try to
811 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700812 */
Shaohua Li62e1c552008-02-04 22:29:33 -0800813 if (!page->mapping) {
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800814 VM_BUG_ON(PageAnon(page));
815 if (page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -0800816 try_to_free_buffers(page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800817 goto uncharge;
Shaohua Li62e1c552008-02-04 22:29:33 -0800818 }
Shaohua Liabfc3482009-09-21 17:01:19 -0700819 goto skip_unmap;
Shaohua Li62e1c552008-02-04 22:29:33 -0800820 }
821
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700822 /* Establish migration ptes or remove ptes */
Andi Kleen14fa31b2009-09-16 11:50:10 +0200823 try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700824
Shaohua Liabfc3482009-09-21 17:01:19 -0700825skip_unmap:
Christoph Lametere6a15302006-06-25 05:46:49 -0700826 if (!page_mapped(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800827 rc = move_to_new_page(newpage, page, remap_swapcache, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700828
Mel Gorman3fe20112010-05-24 14:32:20 -0700829 if (rc && remap_swapcache)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700830 remove_migration_ptes(page, page);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700831
832 /* Drop an anon_vma reference if we took one */
Rik van Riel76545062010-08-09 17:18:41 -0700833 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700834 put_anon_vma(anon_vma);
Mel Gorman3f6c8272010-05-24 14:32:17 -0700835
KAMEZAWA Hiroyuki01b1ae62009-01-07 18:07:50 -0800836uncharge:
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800837 mem_cgroup_end_migration(mem, page, newpage,
838 (rc == MIGRATEPAGE_SUCCESS ||
839 rc == MIGRATEPAGE_BALLOON_SUCCESS));
Christoph Lametere24f0b82006-06-23 02:03:51 -0700840 unlock_page(page);
Minchan Kim0dabec92011-10-31 17:06:57 -0700841out:
842 return rc;
843}
Christoph Lameter95a402c2006-06-23 02:03:53 -0700844
Minchan Kim0dabec92011-10-31 17:06:57 -0700845/*
846 * Obtain the lock on page, remove all ptes and migrate the page
847 * to the newly allocated page in newpage.
848 */
849static int unmap_and_move(new_page_t get_new_page, unsigned long private,
Hugh Dickins9c620e22013-02-22 16:35:14 -0800850 struct page *page, int force, enum migrate_mode mode)
Minchan Kim0dabec92011-10-31 17:06:57 -0700851{
852 int rc = 0;
853 int *result = NULL;
854 struct page *newpage = get_new_page(page, private, &result);
855
856 if (!newpage)
857 return -ENOMEM;
858
859 if (page_count(page) == 1) {
860 /* page was freed from under us. So we are done. */
861 goto out;
862 }
863
864 if (unlikely(PageTransHuge(page)))
865 if (unlikely(split_huge_page(page)))
866 goto out;
867
Hugh Dickins9c620e22013-02-22 16:35:14 -0800868 rc = __unmap_and_move(page, newpage, force, mode);
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800869
870 if (unlikely(rc == MIGRATEPAGE_BALLOON_SUCCESS)) {
871 /*
872 * A ballooned page has been migrated already.
873 * Now, it's the time to wrap-up counters,
874 * handle the page back to Buddy and return.
875 */
876 dec_zone_page_state(page, NR_ISOLATED_ANON +
877 page_is_file_cache(page));
878 balloon_page_free(page);
879 return MIGRATEPAGE_SUCCESS;
880 }
Minchan Kim0dabec92011-10-31 17:06:57 -0700881out:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700882 if (rc != -EAGAIN) {
Minchan Kim0dabec92011-10-31 17:06:57 -0700883 /*
884 * A page that has been migrated has all references
885 * removed and will be freed. A page that has not been
886 * migrated will have kepts its references and be
887 * restored.
888 */
889 list_del(&page->lru);
KOSAKI Motohiroa7312862009-09-21 17:01:37 -0700890 dec_zone_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -0700891 page_is_file_cache(page));
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700892 putback_lru_page(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700893 }
Christoph Lameter95a402c2006-06-23 02:03:53 -0700894 /*
895 * Move the new page to the LRU. If migration was not successful
896 * then this will free the page.
897 */
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700898 putback_lru_page(newpage);
Christoph Lameter742755a2006-06-23 02:03:55 -0700899 if (result) {
900 if (rc)
901 *result = rc;
902 else
903 *result = page_to_nid(newpage);
904 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700905 return rc;
906}
907
908/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900909 * Counterpart of unmap_and_move_page() for hugepage migration.
910 *
911 * This function doesn't wait the completion of hugepage I/O
912 * because there is no race between I/O and migration for hugepage.
913 * Note that currently hugepage I/O occurs only in direct I/O
914 * where no lock is held and PG_writeback is irrelevant,
915 * and writeback status of all subpages are counted in the reference
916 * count of the head page (i.e. if all subpages of a 2MB hugepage are
917 * under direct I/O, the reference of the head page is 512 and a bit more.)
918 * This means that when we try to migrate hugepage whose subpages are
919 * doing direct I/O, some references remain after try_to_unmap() and
920 * hugepage migration fails without data corruption.
921 *
922 * There is also no race when direct I/O is issued on the page under migration,
923 * because then pte is replaced with migration swap entry and direct I/O code
924 * will wait in the page fault for migration to complete.
925 */
926static int unmap_and_move_huge_page(new_page_t get_new_page,
927 unsigned long private, struct page *hpage,
Hugh Dickins9c620e22013-02-22 16:35:14 -0800928 int force, enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900929{
930 int rc = 0;
931 int *result = NULL;
932 struct page *new_hpage = get_new_page(hpage, private, &result);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900933 struct anon_vma *anon_vma = NULL;
934
935 if (!new_hpage)
936 return -ENOMEM;
937
938 rc = -EAGAIN;
939
940 if (!trylock_page(hpage)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800941 if (!force || mode != MIGRATE_SYNC)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900942 goto out;
943 lock_page(hpage);
944 }
945
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700946 if (PageAnon(hpage))
947 anon_vma = page_get_anon_vma(hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900948
949 try_to_unmap(hpage, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
950
951 if (!page_mapped(hpage))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800952 rc = move_to_new_page(new_hpage, hpage, 1, mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900953
954 if (rc)
955 remove_migration_ptes(hpage, hpage);
956
Hugh Dickinsfd4a4662011-01-13 15:47:31 -0800957 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -0700958 put_anon_vma(anon_vma);
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -0700959
960 if (!rc)
961 hugetlb_cgroup_migrate(hpage, new_hpage);
962
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900963 unlock_page(hpage);
Hillf Danton09761332011-12-08 14:34:20 -0800964out:
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900965 put_page(new_hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900966 if (result) {
967 if (rc)
968 *result = rc;
969 else
970 *result = page_to_nid(new_hpage);
971 }
972 return rc;
973}
974
975/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800976 * migrate_pages
977 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700978 * The function takes one list of pages to migrate and a function
979 * that determines from the page to be migrated and the private data
980 * the target of the move and allocates the page.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800981 *
982 * The function returns after 10 attempts or if no pages
983 * are movable anymore because to has become empty
Minchan Kimcf608ac2010-10-26 14:21:29 -0700984 * or no retryable pages exist anymore.
985 * Caller should call putback_lru_pages to return pages to the LRU
Minchan Kim28bd6572011-01-25 15:07:26 -0800986 * or free list only if ret != 0.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800987 *
Christoph Lameter95a402c2006-06-23 02:03:53 -0700988 * Return: Number of pages not migrated or error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800989 */
Hugh Dickins9c620e22013-02-22 16:35:14 -0800990int migrate_pages(struct list_head *from, new_page_t get_new_page,
991 unsigned long private, enum migrate_mode mode, int reason)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800992{
Christoph Lametere24f0b82006-06-23 02:03:51 -0700993 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800994 int nr_failed = 0;
Mel Gorman5647bc22012-10-19 10:46:20 +0100995 int nr_succeeded = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800996 int pass = 0;
997 struct page *page;
998 struct page *page2;
999 int swapwrite = current->flags & PF_SWAPWRITE;
1000 int rc;
1001
1002 if (!swapwrite)
1003 current->flags |= PF_SWAPWRITE;
1004
Christoph Lametere24f0b82006-06-23 02:03:51 -07001005 for(pass = 0; pass < 10 && retry; pass++) {
1006 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001007
Christoph Lametere24f0b82006-06-23 02:03:51 -07001008 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -07001009 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -08001010
Christoph Lameter95a402c2006-06-23 02:03:53 -07001011 rc = unmap_and_move(get_new_page, private,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001012 page, pass > 2, mode);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001013
Christoph Lametere24f0b82006-06-23 02:03:51 -07001014 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -07001015 case -ENOMEM:
1016 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001017 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001018 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001019 break;
Rafael Aquini78bd5202012-12-11 16:02:31 -08001020 case MIGRATEPAGE_SUCCESS:
Mel Gorman5647bc22012-10-19 10:46:20 +01001021 nr_succeeded++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001022 break;
1023 default:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001024 /* Permanent failure */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001025 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001026 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001027 }
Christoph Lameterb20a3502006-03-22 00:09:12 -08001028 }
1029 }
Rafael Aquini78bd5202012-12-11 16:02:31 -08001030 rc = nr_failed + retry;
Christoph Lameter95a402c2006-06-23 02:03:53 -07001031out:
Mel Gorman5647bc22012-10-19 10:46:20 +01001032 if (nr_succeeded)
1033 count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded);
1034 if (nr_failed)
1035 count_vm_events(PGMIGRATE_FAIL, nr_failed);
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001036 trace_mm_migrate_pages(nr_succeeded, nr_failed, mode, reason);
1037
Christoph Lameterb20a3502006-03-22 00:09:12 -08001038 if (!swapwrite)
1039 current->flags &= ~PF_SWAPWRITE;
1040
Rafael Aquini78bd5202012-12-11 16:02:31 -08001041 return rc;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001042}
1043
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001044int migrate_huge_page(struct page *hpage, new_page_t get_new_page,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001045 unsigned long private, enum migrate_mode mode)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001046{
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001047 int pass, rc;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001048
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001049 for (pass = 0; pass < 10; pass++) {
Hugh Dickins9c620e22013-02-22 16:35:14 -08001050 rc = unmap_and_move_huge_page(get_new_page, private,
1051 hpage, pass > 2, mode);
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001052 switch (rc) {
1053 case -ENOMEM:
1054 goto out;
1055 case -EAGAIN:
1056 /* try again */
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001057 cond_resched();
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001058 break;
Rafael Aquini78bd5202012-12-11 16:02:31 -08001059 case MIGRATEPAGE_SUCCESS:
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001060 goto out;
1061 default:
1062 rc = -EIO;
1063 goto out;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001064 }
1065 }
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001066out:
Aneesh Kumar K.V189ebff2012-07-31 16:42:06 -07001067 return rc;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001068}
1069
Christoph Lameter742755a2006-06-23 02:03:55 -07001070#ifdef CONFIG_NUMA
1071/*
1072 * Move a list of individual pages
1073 */
1074struct page_to_node {
1075 unsigned long addr;
1076 struct page *page;
1077 int node;
1078 int status;
1079};
1080
1081static struct page *new_page_node(struct page *p, unsigned long private,
1082 int **result)
1083{
1084 struct page_to_node *pm = (struct page_to_node *)private;
1085
1086 while (pm->node != MAX_NUMNODES && pm->page != p)
1087 pm++;
1088
1089 if (pm->node == MAX_NUMNODES)
1090 return NULL;
1091
1092 *result = &pm->status;
1093
Mel Gorman6484eb32009-06-16 15:31:54 -07001094 return alloc_pages_exact_node(pm->node,
Mel Gorman769848c2007-07-17 04:03:05 -07001095 GFP_HIGHUSER_MOVABLE | GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -07001096}
1097
1098/*
1099 * Move a set of pages as indicated in the pm array. The addr
1100 * field must be set to the virtual address of the page to be moved
1101 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001102 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -07001103 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001104static int do_move_page_to_node_array(struct mm_struct *mm,
1105 struct page_to_node *pm,
1106 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -07001107{
1108 int err;
1109 struct page_to_node *pp;
1110 LIST_HEAD(pagelist);
1111
1112 down_read(&mm->mmap_sem);
1113
1114 /*
1115 * Build a list of pages to migrate
1116 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001117 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1118 struct vm_area_struct *vma;
1119 struct page *page;
1120
Christoph Lameter742755a2006-06-23 02:03:55 -07001121 err = -EFAULT;
1122 vma = find_vma(mm, pp->addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001123 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -07001124 goto set_status;
1125
Andrea Arcangeli500d65d2011-01-13 15:46:55 -08001126 page = follow_page(vma, pp->addr, FOLL_GET|FOLL_SPLIT);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001127
1128 err = PTR_ERR(page);
1129 if (IS_ERR(page))
1130 goto set_status;
1131
Christoph Lameter742755a2006-06-23 02:03:55 -07001132 err = -ENOENT;
1133 if (!page)
1134 goto set_status;
1135
Hugh Dickins62b61f62009-12-14 17:59:33 -08001136 /* Use PageReserved to check for zero page */
Hugh Dickinsb79bc0a2013-02-22 16:35:13 -08001137 if (PageReserved(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001138 goto put_and_set;
1139
1140 pp->page = page;
1141 err = page_to_nid(page);
1142
1143 if (err == pp->node)
1144 /*
1145 * Node already in the right place
1146 */
1147 goto put_and_set;
1148
1149 err = -EACCES;
1150 if (page_mapcount(page) > 1 &&
1151 !migrate_all)
1152 goto put_and_set;
1153
Nick Piggin62695a82008-10-18 20:26:09 -07001154 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001155 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -07001156 list_add_tail(&page->lru, &pagelist);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001157 inc_zone_page_state(page, NR_ISOLATED_ANON +
1158 page_is_file_cache(page));
1159 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001160put_and_set:
1161 /*
1162 * Either remove the duplicate refcount from
1163 * isolate_lru_page() or drop the page ref if it was
1164 * not isolated.
1165 */
1166 put_page(page);
1167set_status:
1168 pp->status = err;
1169 }
1170
Brice Gogline78bbfa2008-10-18 20:27:15 -07001171 err = 0;
Minchan Kimcf608ac2010-10-26 14:21:29 -07001172 if (!list_empty(&pagelist)) {
Christoph Lameter742755a2006-06-23 02:03:55 -07001173 err = migrate_pages(&pagelist, new_page_node,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001174 (unsigned long)pm, MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001175 if (err)
1176 putback_lru_pages(&pagelist);
1177 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001178
1179 up_read(&mm->mmap_sem);
1180 return err;
1181}
1182
1183/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001184 * Migrate an array of page address onto an array of nodes and fill
1185 * the corresponding array of status.
1186 */
Christoph Lameter3268c632012-03-21 16:34:06 -07001187static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001188 unsigned long nr_pages,
1189 const void __user * __user *pages,
1190 const int __user *nodes,
1191 int __user *status, int flags)
1192{
Brice Goglin3140a222009-01-06 14:38:57 -08001193 struct page_to_node *pm;
Brice Goglin3140a222009-01-06 14:38:57 -08001194 unsigned long chunk_nr_pages;
1195 unsigned long chunk_start;
1196 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001197
Brice Goglin3140a222009-01-06 14:38:57 -08001198 err = -ENOMEM;
1199 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1200 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001201 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -07001202
1203 migrate_prep();
1204
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001205 /*
Brice Goglin3140a222009-01-06 14:38:57 -08001206 * Store a chunk of page_to_node array in a page,
1207 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001208 */
Brice Goglin3140a222009-01-06 14:38:57 -08001209 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001210
Brice Goglin3140a222009-01-06 14:38:57 -08001211 for (chunk_start = 0;
1212 chunk_start < nr_pages;
1213 chunk_start += chunk_nr_pages) {
1214 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001215
Brice Goglin3140a222009-01-06 14:38:57 -08001216 if (chunk_start + chunk_nr_pages > nr_pages)
1217 chunk_nr_pages = nr_pages - chunk_start;
1218
1219 /* fill the chunk pm with addrs and nodes from user-space */
1220 for (j = 0; j < chunk_nr_pages; j++) {
1221 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001222 int node;
1223
Brice Goglin3140a222009-01-06 14:38:57 -08001224 err = -EFAULT;
1225 if (get_user(p, pages + j + chunk_start))
1226 goto out_pm;
1227 pm[j].addr = (unsigned long) p;
1228
1229 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001230 goto out_pm;
1231
1232 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -08001233 if (node < 0 || node >= MAX_NUMNODES)
1234 goto out_pm;
1235
Lai Jiangshan389162c2012-12-12 13:51:30 -08001236 if (!node_state(node, N_MEMORY))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001237 goto out_pm;
1238
1239 err = -EACCES;
1240 if (!node_isset(node, task_nodes))
1241 goto out_pm;
1242
Brice Goglin3140a222009-01-06 14:38:57 -08001243 pm[j].node = node;
1244 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001245
Brice Goglin3140a222009-01-06 14:38:57 -08001246 /* End marker for this chunk */
1247 pm[chunk_nr_pages].node = MAX_NUMNODES;
1248
1249 /* Migrate this chunk */
1250 err = do_move_page_to_node_array(mm, pm,
1251 flags & MPOL_MF_MOVE_ALL);
1252 if (err < 0)
1253 goto out_pm;
1254
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001255 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -08001256 for (j = 0; j < chunk_nr_pages; j++)
1257 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001258 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -08001259 goto out_pm;
1260 }
1261 }
1262 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001263
1264out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -08001265 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001266out:
1267 return err;
1268}
1269
1270/*
Brice Goglin2f007e72008-10-18 20:27:16 -07001271 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -07001272 */
Brice Goglin80bba122008-12-09 13:14:23 -08001273static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1274 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001275{
Brice Goglin2f007e72008-10-18 20:27:16 -07001276 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001277
Christoph Lameter742755a2006-06-23 02:03:55 -07001278 down_read(&mm->mmap_sem);
1279
Brice Goglin2f007e72008-10-18 20:27:16 -07001280 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001281 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001282 struct vm_area_struct *vma;
1283 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001284 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001285
1286 vma = find_vma(mm, addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001287 if (!vma || addr < vma->vm_start)
Christoph Lameter742755a2006-06-23 02:03:55 -07001288 goto set_status;
1289
Brice Goglin2f007e72008-10-18 20:27:16 -07001290 page = follow_page(vma, addr, 0);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001291
1292 err = PTR_ERR(page);
1293 if (IS_ERR(page))
1294 goto set_status;
1295
Christoph Lameter742755a2006-06-23 02:03:55 -07001296 err = -ENOENT;
1297 /* Use PageReserved to check for zero page */
Hugh Dickinsb79bc0a2013-02-22 16:35:13 -08001298 if (!page || PageReserved(page))
Christoph Lameter742755a2006-06-23 02:03:55 -07001299 goto set_status;
1300
1301 err = page_to_nid(page);
1302set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001303 *status = err;
1304
1305 pages++;
1306 status++;
1307 }
1308
1309 up_read(&mm->mmap_sem);
1310}
1311
1312/*
1313 * Determine the nodes of a user array of pages and store it in
1314 * a user array of status.
1315 */
1316static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1317 const void __user * __user *pages,
1318 int __user *status)
1319{
1320#define DO_PAGES_STAT_CHUNK_NR 16
1321 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1322 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
Brice Goglin80bba122008-12-09 13:14:23 -08001323
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001324 while (nr_pages) {
1325 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001326
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001327 chunk_nr = nr_pages;
1328 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1329 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1330
1331 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1332 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001333
1334 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1335
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001336 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1337 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001338
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001339 pages += chunk_nr;
1340 status += chunk_nr;
1341 nr_pages -= chunk_nr;
1342 }
1343 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001344}
1345
1346/*
1347 * Move a list of pages in the address space of the currently executing
1348 * process.
1349 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001350SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1351 const void __user * __user *, pages,
1352 const int __user *, nodes,
1353 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001354{
David Howellsc69e8d92008-11-14 10:39:19 +11001355 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001356 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001357 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001358 int err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001359 nodemask_t task_nodes;
Christoph Lameter742755a2006-06-23 02:03:55 -07001360
1361 /* Check flags */
1362 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1363 return -EINVAL;
1364
1365 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1366 return -EPERM;
1367
1368 /* Find the mm_struct */
Greg Thelena879bf52011-02-25 14:44:13 -08001369 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001370 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001371 if (!task) {
Greg Thelena879bf52011-02-25 14:44:13 -08001372 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001373 return -ESRCH;
1374 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001375 get_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001376
1377 /*
1378 * Check if this process has the right to modify the specified
1379 * process. The right exists if the process has administrative
1380 * capabilities, superuser privileges or the same
1381 * userid as the target process.
1382 */
David Howellsc69e8d92008-11-14 10:39:19 +11001383 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001384 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1385 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001386 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001387 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001388 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001389 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001390 }
David Howellsc69e8d92008-11-14 10:39:19 +11001391 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001392
David Quigley86c3a762006-06-23 02:04:02 -07001393 err = security_task_movememory(task);
1394 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001395 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001396
Christoph Lameter3268c632012-03-21 16:34:06 -07001397 task_nodes = cpuset_mems_allowed(task);
1398 mm = get_task_mm(task);
1399 put_task_struct(task);
1400
Sasha Levin6e8b09e2012-04-25 16:01:53 -07001401 if (!mm)
1402 return -EINVAL;
1403
1404 if (nodes)
1405 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1406 nodes, status, flags);
1407 else
1408 err = do_pages_stat(mm, nr_pages, pages, status);
Christoph Lameter3268c632012-03-21 16:34:06 -07001409
1410 mmput(mm);
1411 return err;
David Quigley86c3a762006-06-23 02:04:02 -07001412
Christoph Lameter742755a2006-06-23 02:03:55 -07001413out:
Christoph Lameter3268c632012-03-21 16:34:06 -07001414 put_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001415 return err;
1416}
Christoph Lameter742755a2006-06-23 02:03:55 -07001417
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001418/*
1419 * Call migration functions in the vma_ops that may prepare
1420 * memory in a vm for migration. migration functions may perform
1421 * the migration for vmas that do not have an underlying page struct.
1422 */
1423int migrate_vmas(struct mm_struct *mm, const nodemask_t *to,
1424 const nodemask_t *from, unsigned long flags)
1425{
1426 struct vm_area_struct *vma;
1427 int err = 0;
1428
Daisuke Nishimura1001c9f2009-02-11 13:04:18 -08001429 for (vma = mm->mmap; vma && !err; vma = vma->vm_next) {
Christoph Lameter7b2259b2006-06-25 05:46:48 -07001430 if (vma->vm_ops && vma->vm_ops->migrate) {
1431 err = vma->vm_ops->migrate(vma, to, from, flags);
1432 if (err)
1433 break;
1434 }
1435 }
1436 return err;
1437}
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001438
1439#ifdef CONFIG_NUMA_BALANCING
1440/*
1441 * Returns true if this is a safe migration target node for misplaced NUMA
1442 * pages. Currently it only checks the watermarks which crude
1443 */
1444static bool migrate_balanced_pgdat(struct pglist_data *pgdat,
Mel Gorman3abef4e2013-02-22 16:34:27 -08001445 unsigned long nr_migrate_pages)
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001446{
1447 int z;
1448 for (z = pgdat->nr_zones - 1; z >= 0; z--) {
1449 struct zone *zone = pgdat->node_zones + z;
1450
1451 if (!populated_zone(zone))
1452 continue;
1453
1454 if (zone->all_unreclaimable)
1455 continue;
1456
1457 /* Avoid waking kswapd by allocating pages_to_migrate pages. */
1458 if (!zone_watermark_ok(zone, 0,
1459 high_wmark_pages(zone) +
1460 nr_migrate_pages,
1461 0, 0))
1462 continue;
1463 return true;
1464 }
1465 return false;
1466}
1467
1468static struct page *alloc_misplaced_dst_page(struct page *page,
1469 unsigned long data,
1470 int **result)
1471{
1472 int nid = (int) data;
1473 struct page *newpage;
1474
1475 newpage = alloc_pages_exact_node(nid,
1476 (GFP_HIGHUSER_MOVABLE | GFP_THISNODE |
1477 __GFP_NOMEMALLOC | __GFP_NORETRY |
1478 __GFP_NOWARN) &
1479 ~GFP_IOFS, 0);
Hillf Dantonbac03822012-11-27 14:46:24 +00001480 if (newpage)
Mel Gorman22b751c2013-02-22 16:34:59 -08001481 page_nid_xchg_last(newpage, page_nid_last(page));
Hillf Dantonbac03822012-11-27 14:46:24 +00001482
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001483 return newpage;
1484}
1485
1486/*
Mel Gormana8f60772012-11-14 21:41:46 +00001487 * page migration rate limiting control.
1488 * Do not migrate more than @pages_to_migrate in a @migrate_interval_millisecs
1489 * window of time. Default here says do not migrate more than 1280M per second.
Mel Gormane14808b2012-11-19 10:59:15 +00001490 * If a node is rate-limited then PTE NUMA updates are also rate-limited. However
1491 * as it is faults that reset the window, pte updates will happen unconditionally
1492 * if there has not been a fault since @pteupdate_interval_millisecs after the
1493 * throttle window closed.
Mel Gormana8f60772012-11-14 21:41:46 +00001494 */
1495static unsigned int migrate_interval_millisecs __read_mostly = 100;
Mel Gormane14808b2012-11-19 10:59:15 +00001496static unsigned int pteupdate_interval_millisecs __read_mostly = 1000;
Mel Gormana8f60772012-11-14 21:41:46 +00001497static unsigned int ratelimit_pages __read_mostly = 128 << (20 - PAGE_SHIFT);
1498
Mel Gormane14808b2012-11-19 10:59:15 +00001499/* Returns true if NUMA migration is currently rate limited */
1500bool migrate_ratelimited(int node)
1501{
1502 pg_data_t *pgdat = NODE_DATA(node);
1503
1504 if (time_after(jiffies, pgdat->numabalancing_migrate_next_window +
1505 msecs_to_jiffies(pteupdate_interval_millisecs)))
1506 return false;
1507
1508 if (pgdat->numabalancing_migrate_nr_pages < ratelimit_pages)
1509 return false;
1510
1511 return true;
1512}
1513
Mel Gormanb32967f2012-11-19 12:35:47 +00001514/* Returns true if the node is migrate rate-limited after the update */
Mel Gormand28d43352012-11-29 09:24:36 +00001515bool numamigrate_update_ratelimit(pg_data_t *pgdat, unsigned long nr_pages)
Mel Gormanb32967f2012-11-19 12:35:47 +00001516{
1517 bool rate_limited = false;
1518
1519 /*
1520 * Rate-limit the amount of data that is being migrated to a node.
1521 * Optimal placement is no good if the memory bus is saturated and
1522 * all the time is being spent migrating!
1523 */
1524 spin_lock(&pgdat->numabalancing_migrate_lock);
1525 if (time_after(jiffies, pgdat->numabalancing_migrate_next_window)) {
1526 pgdat->numabalancing_migrate_nr_pages = 0;
1527 pgdat->numabalancing_migrate_next_window = jiffies +
1528 msecs_to_jiffies(migrate_interval_millisecs);
1529 }
1530 if (pgdat->numabalancing_migrate_nr_pages > ratelimit_pages)
1531 rate_limited = true;
1532 else
Mel Gormand28d43352012-11-29 09:24:36 +00001533 pgdat->numabalancing_migrate_nr_pages += nr_pages;
Mel Gormanb32967f2012-11-19 12:35:47 +00001534 spin_unlock(&pgdat->numabalancing_migrate_lock);
1535
1536 return rate_limited;
1537}
1538
1539int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page)
1540{
Hugh Dickins340ef392013-02-22 16:34:33 -08001541 int page_lru;
Mel Gormanb32967f2012-11-19 12:35:47 +00001542
Mel Gorman3abef4e2013-02-22 16:34:27 -08001543 VM_BUG_ON(compound_order(page) && !PageTransHuge(page));
1544
Mel Gormanb32967f2012-11-19 12:35:47 +00001545 /* Avoid migrating to a node that is nearly full */
Hugh Dickins340ef392013-02-22 16:34:33 -08001546 if (!migrate_balanced_pgdat(pgdat, 1UL << compound_order(page)))
1547 return 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001548
Hugh Dickins340ef392013-02-22 16:34:33 -08001549 if (isolate_lru_page(page))
1550 return 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001551
1552 /*
Hugh Dickins340ef392013-02-22 16:34:33 -08001553 * migrate_misplaced_transhuge_page() skips page migration's usual
1554 * check on page_count(), so we must do it here, now that the page
1555 * has been isolated: a GUP pin, or any other pin, prevents migration.
1556 * The expected page count is 3: 1 for page's mapcount and 1 for the
1557 * caller's pin and 1 for the reference taken by isolate_lru_page().
1558 */
1559 if (PageTransHuge(page) && page_count(page) != 3) {
1560 putback_lru_page(page);
1561 return 0;
1562 }
1563
1564 page_lru = page_is_file_cache(page);
1565 mod_zone_page_state(page_zone(page), NR_ISOLATED_ANON + page_lru,
1566 hpage_nr_pages(page));
1567
1568 /*
1569 * Isolating the page has taken another reference, so the
1570 * caller's reference can be safely dropped without the page
1571 * disappearing underneath us during migration.
Mel Gormanb32967f2012-11-19 12:35:47 +00001572 */
1573 put_page(page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001574 return 1;
Mel Gormanb32967f2012-11-19 12:35:47 +00001575}
1576
Mel Gormana8f60772012-11-14 21:41:46 +00001577/*
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001578 * Attempt to migrate a misplaced page to the specified destination
1579 * node. Caller is expected to have an elevated reference count on
1580 * the page that will be dropped by this function before returning.
1581 */
1582int migrate_misplaced_page(struct page *page, int node)
1583{
Mel Gormana8f60772012-11-14 21:41:46 +00001584 pg_data_t *pgdat = NODE_DATA(node);
Hugh Dickins340ef392013-02-22 16:34:33 -08001585 int isolated;
Mel Gormanb32967f2012-11-19 12:35:47 +00001586 int nr_remaining;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001587 LIST_HEAD(migratepages);
1588
1589 /*
1590 * Don't migrate pages that are mapped in multiple processes.
1591 * TODO: Handle false sharing detection instead of this hammer
1592 */
Hugh Dickins340ef392013-02-22 16:34:33 -08001593 if (page_mapcount(page) != 1)
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001594 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001595
Mel Gormana8f60772012-11-14 21:41:46 +00001596 /*
1597 * Rate-limit the amount of data that is being migrated to a node.
1598 * Optimal placement is no good if the memory bus is saturated and
1599 * all the time is being spent migrating!
1600 */
Hugh Dickins340ef392013-02-22 16:34:33 -08001601 if (numamigrate_update_ratelimit(pgdat, 1))
Mel Gormana8f60772012-11-14 21:41:46 +00001602 goto out;
Mel Gormana8f60772012-11-14 21:41:46 +00001603
Mel Gormanb32967f2012-11-19 12:35:47 +00001604 isolated = numamigrate_isolate_page(pgdat, page);
1605 if (!isolated)
1606 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001607
Mel Gormanb32967f2012-11-19 12:35:47 +00001608 list_add(&page->lru, &migratepages);
Hugh Dickins9c620e22013-02-22 16:35:14 -08001609 nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page,
1610 node, MIGRATE_ASYNC, MR_NUMA_MISPLACED);
Mel Gormanb32967f2012-11-19 12:35:47 +00001611 if (nr_remaining) {
1612 putback_lru_pages(&migratepages);
1613 isolated = 0;
1614 } else
1615 count_vm_numa_event(NUMA_PAGE_MIGRATE);
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001616 BUG_ON(!list_empty(&migratepages));
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001617 return isolated;
Hugh Dickins340ef392013-02-22 16:34:33 -08001618
1619out:
1620 put_page(page);
1621 return 0;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001622}
Mel Gorman220018d2012-12-05 09:32:56 +00001623#endif /* CONFIG_NUMA_BALANCING */
Mel Gormanb32967f2012-11-19 12:35:47 +00001624
Mel Gorman220018d2012-12-05 09:32:56 +00001625#if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
Hugh Dickins340ef392013-02-22 16:34:33 -08001626/*
1627 * Migrates a THP to a given target node. page must be locked and is unlocked
1628 * before returning.
1629 */
Mel Gormanb32967f2012-11-19 12:35:47 +00001630int migrate_misplaced_transhuge_page(struct mm_struct *mm,
1631 struct vm_area_struct *vma,
1632 pmd_t *pmd, pmd_t entry,
1633 unsigned long address,
1634 struct page *page, int node)
1635{
1636 unsigned long haddr = address & HPAGE_PMD_MASK;
1637 pg_data_t *pgdat = NODE_DATA(node);
1638 int isolated = 0;
1639 struct page *new_page = NULL;
1640 struct mem_cgroup *memcg = NULL;
1641 int page_lru = page_is_file_cache(page);
1642
1643 /*
1644 * Don't migrate pages that are mapped in multiple processes.
1645 * TODO: Handle false sharing detection instead of this hammer
1646 */
1647 if (page_mapcount(page) != 1)
1648 goto out_dropref;
1649
1650 /*
1651 * Rate-limit the amount of data that is being migrated to a node.
1652 * Optimal placement is no good if the memory bus is saturated and
1653 * all the time is being spent migrating!
1654 */
Mel Gormand28d43352012-11-29 09:24:36 +00001655 if (numamigrate_update_ratelimit(pgdat, HPAGE_PMD_NR))
Mel Gormanb32967f2012-11-19 12:35:47 +00001656 goto out_dropref;
1657
1658 new_page = alloc_pages_node(node,
1659 (GFP_TRANSHUGE | GFP_THISNODE) & ~__GFP_WAIT, HPAGE_PMD_ORDER);
Hugh Dickins340ef392013-02-22 16:34:33 -08001660 if (!new_page)
1661 goto out_fail;
1662
Mel Gorman22b751c2013-02-22 16:34:59 -08001663 page_nid_xchg_last(new_page, page_nid_last(page));
Mel Gormanb32967f2012-11-19 12:35:47 +00001664
1665 isolated = numamigrate_isolate_page(pgdat, page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001666 if (!isolated) {
Mel Gormanb32967f2012-11-19 12:35:47 +00001667 put_page(new_page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001668 goto out_fail;
Mel Gormanb32967f2012-11-19 12:35:47 +00001669 }
1670
1671 /* Prepare a page as a migration target */
1672 __set_page_locked(new_page);
1673 SetPageSwapBacked(new_page);
1674
1675 /* anon mapping, we can simply copy page->mapping to the new page: */
1676 new_page->mapping = page->mapping;
1677 new_page->index = page->index;
1678 migrate_page_copy(new_page, page);
1679 WARN_ON(PageLRU(new_page));
1680
1681 /* Recheck the target PMD */
1682 spin_lock(&mm->page_table_lock);
1683 if (unlikely(!pmd_same(*pmd, entry))) {
1684 spin_unlock(&mm->page_table_lock);
1685
1686 /* Reverse changes made by migrate_page_copy() */
1687 if (TestClearPageActive(new_page))
1688 SetPageActive(page);
1689 if (TestClearPageUnevictable(new_page))
1690 SetPageUnevictable(page);
1691 mlock_migrate_page(page, new_page);
1692
1693 unlock_page(new_page);
1694 put_page(new_page); /* Free it */
1695
1696 unlock_page(page);
1697 putback_lru_page(page);
1698
1699 count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR);
Hugh Dickins340ef392013-02-22 16:34:33 -08001700 isolated = 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001701 goto out;
1702 }
1703
1704 /*
1705 * Traditional migration needs to prepare the memcg charge
1706 * transaction early to prevent the old page from being
1707 * uncharged when installing migration entries. Here we can
1708 * save the potential rollback and start the charge transfer
1709 * only when migration is already known to end successfully.
1710 */
1711 mem_cgroup_prepare_migration(page, new_page, &memcg);
1712
1713 entry = mk_pmd(new_page, vma->vm_page_prot);
1714 entry = pmd_mknonnuma(entry);
1715 entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
1716 entry = pmd_mkhuge(entry);
1717
1718 page_add_new_anon_rmap(new_page, vma, haddr);
1719
1720 set_pmd_at(mm, haddr, pmd, entry);
Stephen Rothwellce4a9cc2012-12-10 19:50:57 +11001721 update_mmu_cache_pmd(vma, address, &entry);
Mel Gormanb32967f2012-11-19 12:35:47 +00001722 page_remove_rmap(page);
1723 /*
1724 * Finish the charge transaction under the page table lock to
1725 * prevent split_huge_page() from dividing up the charge
1726 * before it's fully transferred to the new page.
1727 */
1728 mem_cgroup_end_migration(memcg, page, new_page, true);
1729 spin_unlock(&mm->page_table_lock);
1730
1731 unlock_page(new_page);
1732 unlock_page(page);
1733 put_page(page); /* Drop the rmap reference */
1734 put_page(page); /* Drop the LRU isolation reference */
1735
1736 count_vm_events(PGMIGRATE_SUCCESS, HPAGE_PMD_NR);
1737 count_vm_numa_events(NUMA_PAGE_MIGRATE, HPAGE_PMD_NR);
1738
1739out:
1740 mod_zone_page_state(page_zone(page),
1741 NR_ISOLATED_ANON + page_lru,
1742 -HPAGE_PMD_NR);
1743 return isolated;
1744
Hugh Dickins340ef392013-02-22 16:34:33 -08001745out_fail:
1746 count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR);
Mel Gormanb32967f2012-11-19 12:35:47 +00001747out_dropref:
Hugh Dickins340ef392013-02-22 16:34:33 -08001748 unlock_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001749 put_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001750 return 0;
1751}
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001752#endif /* CONFIG_NUMA_BALANCING */
1753
1754#endif /* CONFIG_NUMA */