blob: f7ee04a5ae27a2934de9fab746667597bffafef3 [file] [log] [blame]
Christoph Lameterb20a3502006-03-22 00:09:12 -08001/*
Hugh Dickins14e0f9b2015-11-05 18:49:43 -08002 * Memory Migration functionality - linux/mm/migrate.c
Christoph Lameterb20a3502006-03-22 00:09:12 -08003 *
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>
Hugh Dickins42cb14b2015-11-05 18:50:05 -080033#include <linux/backing-dev.h>
Minchan Kimbda807d2016-07-26 15:23:05 -070034#include <linux/compaction.h>
Adrian Bunk4f5ca262008-07-23 21:27:02 -070035#include <linux/syscalls.h>
Naoya Horiguchi290408d2010-09-08 10:19:35 +090036#include <linux/hugetlb.h>
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -070037#include <linux/hugetlb_cgroup.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090038#include <linux/gfp.h>
Rafael Aquinibf6bddf2012-12-11 16:02:42 -080039#include <linux/balloon_compaction.h>
Mel Gormanf714f4f2013-12-18 17:08:33 -080040#include <linux/mmu_notifier.h>
Vladimir Davydov33c3fc72015-09-09 15:35:45 -070041#include <linux/page_idle.h>
Vlastimil Babkad435edc2016-03-15 14:56:15 -070042#include <linux/page_owner.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080043
Michal Nazarewicz0d1836c2010-12-21 17:24:26 -080044#include <asm/tlbflush.h>
45
Mel Gorman7b2a2d42012-10-19 14:07:31 +010046#define CREATE_TRACE_POINTS
47#include <trace/events/migrate.h>
48
Christoph Lameterb20a3502006-03-22 00:09:12 -080049#include "internal.h"
50
Christoph Lameterb20a3502006-03-22 00:09:12 -080051/*
Christoph Lameter742755a2006-06-23 02:03:55 -070052 * migrate_prep() needs to be called before we start compiling a list of pages
Mel Gorman748446b2010-05-24 14:32:27 -070053 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
54 * undesirable, use migrate_prep_local()
Christoph Lameterb20a3502006-03-22 00:09:12 -080055 */
56int migrate_prep(void)
57{
Christoph Lameterb20a3502006-03-22 00:09:12 -080058 /*
59 * Clear the LRU lists so pages can be isolated.
60 * Note that pages may be moved off the LRU after we have
61 * drained them. Those pages will fail to migrate like other
62 * pages that may be busy.
63 */
64 lru_add_drain_all();
65
66 return 0;
67}
68
Mel Gorman748446b2010-05-24 14:32:27 -070069/* Do the necessary work of migrate_prep but not if it involves other CPUs */
70int migrate_prep_local(void)
71{
72 lru_add_drain();
73
74 return 0;
75}
76
Minchan Kimbda807d2016-07-26 15:23:05 -070077bool isolate_movable_page(struct page *page, isolate_mode_t mode)
78{
79 struct address_space *mapping;
80
81 /*
82 * Avoid burning cycles with pages that are yet under __free_pages(),
83 * or just got freed under us.
84 *
85 * In case we 'win' a race for a movable page being freed under us and
86 * raise its refcount preventing __free_pages() from doing its job
87 * the put_page() at the end of this block will take care of
88 * release this page, thus avoiding a nasty leakage.
89 */
90 if (unlikely(!get_page_unless_zero(page)))
91 goto out;
92
93 /*
94 * Check PageMovable before holding a PG_lock because page's owner
95 * assumes anybody doesn't touch PG_lock of newly allocated page
96 * so unconditionally grapping the lock ruins page's owner side.
97 */
98 if (unlikely(!__PageMovable(page)))
99 goto out_putpage;
100 /*
101 * As movable pages are not isolated from LRU lists, concurrent
102 * compaction threads can race against page migration functions
103 * as well as race against the releasing a page.
104 *
105 * In order to avoid having an already isolated movable page
106 * being (wrongly) re-isolated while it is under migration,
107 * or to avoid attempting to isolate pages being released,
108 * lets be sure we have the page lock
109 * before proceeding with the movable page isolation steps.
110 */
111 if (unlikely(!trylock_page(page)))
112 goto out_putpage;
113
114 if (!PageMovable(page) || PageIsolated(page))
115 goto out_no_isolated;
116
117 mapping = page_mapping(page);
118 VM_BUG_ON_PAGE(!mapping, page);
119
120 if (!mapping->a_ops->isolate_page(page, mode))
121 goto out_no_isolated;
122
123 /* Driver shouldn't use PG_isolated bit of page->flags */
124 WARN_ON_ONCE(PageIsolated(page));
125 __SetPageIsolated(page);
126 unlock_page(page);
127
128 return true;
129
130out_no_isolated:
131 unlock_page(page);
132out_putpage:
133 put_page(page);
134out:
135 return false;
136}
137
138/* It should be called on page which is PG_movable */
139void putback_movable_page(struct page *page)
140{
141 struct address_space *mapping;
142
143 VM_BUG_ON_PAGE(!PageLocked(page), page);
144 VM_BUG_ON_PAGE(!PageMovable(page), page);
145 VM_BUG_ON_PAGE(!PageIsolated(page), page);
146
147 mapping = page_mapping(page);
148 mapping->a_ops->putback_page(page);
149 __ClearPageIsolated(page);
150}
151
Christoph Lameterb20a3502006-03-22 00:09:12 -0800152/*
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800153 * Put previously isolated pages back onto the appropriate lists
154 * from where they were once taken off for compaction/migration.
155 *
Joonsoo Kim59c82b72014-01-21 15:51:17 -0800156 * This function shall be used whenever the isolated pageset has been
157 * built from lru, balloon, hugetlbfs page. See isolate_migratepages_range()
158 * and isolate_huge_page().
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800159 */
160void putback_movable_pages(struct list_head *l)
161{
162 struct page *page;
163 struct page *page2;
164
165 list_for_each_entry_safe(page, page2, l, lru) {
Naoya Horiguchi31caf662013-09-11 14:21:59 -0700166 if (unlikely(PageHuge(page))) {
167 putback_active_hugepage(page);
168 continue;
169 }
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800170 list_del(&page->lru);
Mel Gorman599d0c92016-07-28 15:45:31 -0700171 dec_node_page_state(page, NR_ISOLATED_ANON +
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800172 page_is_file_cache(page));
Minchan Kimbda807d2016-07-26 15:23:05 -0700173 /*
174 * We isolated non-lru movable page so here we can use
175 * __PageMovable because LRU page's mapping cannot have
176 * PAGE_MAPPING_MOVABLE.
177 */
Minchan Kimb1123ea62016-07-26 15:23:09 -0700178 if (unlikely(__PageMovable(page))) {
Minchan Kimbda807d2016-07-26 15:23:05 -0700179 VM_BUG_ON_PAGE(!PageIsolated(page), page);
180 lock_page(page);
181 if (PageMovable(page))
182 putback_movable_page(page);
183 else
184 __ClearPageIsolated(page);
185 unlock_page(page);
186 put_page(page);
187 } else {
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800188 putback_lru_page(page);
Minchan Kimbda807d2016-07-26 15:23:05 -0700189 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800190 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800191}
192
Christoph Lameter06972122006-06-23 02:03:35 -0700193/*
194 * Restore a potential migration pte to a working pte entry
195 */
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800196static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
197 unsigned long addr, void *old)
Christoph Lameter06972122006-06-23 02:03:35 -0700198{
199 struct mm_struct *mm = vma->vm_mm;
200 swp_entry_t entry;
Christoph Lameter06972122006-06-23 02:03:35 -0700201 pmd_t *pmd;
202 pte_t *ptep, pte;
203 spinlock_t *ptl;
204
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900205 if (unlikely(PageHuge(new))) {
206 ptep = huge_pte_offset(mm, addr);
207 if (!ptep)
208 goto out;
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800209 ptl = huge_pte_lockptr(hstate_vma(vma), mm, ptep);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900210 } else {
Bob Liu62190492012-12-11 16:00:37 -0800211 pmd = mm_find_pmd(mm, addr);
212 if (!pmd)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900213 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700214
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900215 ptep = pte_offset_map(pmd, addr);
Christoph Lameter06972122006-06-23 02:03:35 -0700216
Hugh Dickins486cf462011-10-19 12:50:35 -0700217 /*
218 * Peek to check is_swap_pte() before taking ptlock? No, we
219 * can race mremap's move_ptes(), which skips anon_vma lock.
220 */
Christoph Lameter06972122006-06-23 02:03:35 -0700221
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900222 ptl = pte_lockptr(mm, pmd);
223 }
224
Christoph Lameter06972122006-06-23 02:03:35 -0700225 spin_lock(ptl);
226 pte = *ptep;
227 if (!is_swap_pte(pte))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800228 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700229
230 entry = pte_to_swp_entry(pte);
231
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800232 if (!is_migration_entry(entry) ||
233 migration_entry_to_page(entry) != old)
234 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700235
Christoph Lameter06972122006-06-23 02:03:35 -0700236 get_page(new);
237 pte = pte_mkold(mk_pte(new, vma->vm_page_prot));
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700238 if (pte_swp_soft_dirty(*ptep))
239 pte = pte_mksoft_dirty(pte);
Mel Gormand3cb8bf2014-10-02 19:47:41 +0100240
241 /* Recheck VMA as permissions can change since migration started */
Christoph Lameter06972122006-06-23 02:03:35 -0700242 if (is_write_migration_entry(entry))
Mel Gormand3cb8bf2014-10-02 19:47:41 +0100243 pte = maybe_mkwrite(pte, vma);
244
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200245#ifdef CONFIG_HUGETLB_PAGE
Tony Lube7517d2013-02-04 14:28:46 -0800246 if (PageHuge(new)) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900247 pte = pte_mkhuge(pte);
Tony Lube7517d2013-02-04 14:28:46 -0800248 pte = arch_make_huge_pte(pte, vma, new, 0);
249 }
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200250#endif
Leonid Yegoshinc2cc4992013-05-24 15:55:18 -0700251 flush_dcache_page(new);
Christoph Lameter06972122006-06-23 02:03:35 -0700252 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700253
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900254 if (PageHuge(new)) {
255 if (PageAnon(new))
256 hugepage_add_anon_rmap(new, vma, addr);
257 else
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800258 page_dup_rmap(new, true);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900259 } else if (PageAnon(new))
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -0800260 page_add_anon_rmap(new, vma, addr, false);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700261 else
Kirill A. Shutemovdd78fed2016-07-26 15:25:26 -0700262 page_add_file_rmap(new, false);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700263
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700264 if (vma->vm_flags & VM_LOCKED && !PageTransCompound(new))
Hugh Dickins51afb122015-11-05 18:49:37 -0800265 mlock_vma_page(new);
266
Christoph Lameter04e62a22006-06-23 02:03:38 -0700267 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +0000268 update_mmu_cache(vma, addr, ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800269unlock:
Christoph Lameter06972122006-06-23 02:03:35 -0700270 pte_unmap_unlock(ptep, ptl);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800271out:
272 return SWAP_AGAIN;
Christoph Lameter06972122006-06-23 02:03:35 -0700273}
274
275/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700276 * Get rid of all migration entries and replace them by
277 * references to the indicated page.
278 */
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700279void remove_migration_ptes(struct page *old, struct page *new, bool locked)
Christoph Lameter04e62a22006-06-23 02:03:38 -0700280{
Joonsoo Kim051ac832014-01-21 15:49:48 -0800281 struct rmap_walk_control rwc = {
282 .rmap_one = remove_migration_pte,
283 .arg = old,
284 };
285
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700286 if (locked)
287 rmap_walk_locked(new, &rwc);
288 else
289 rmap_walk(new, &rwc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700290}
291
292/*
Christoph Lameter06972122006-06-23 02:03:35 -0700293 * Something used the pte of a page under migration. We need to
294 * get to the page and wait until migration is finished.
295 * When we return from this function the fault will be retried.
Christoph Lameter06972122006-06-23 02:03:35 -0700296 */
Naoya Horiguchie66f17f2015-02-11 15:25:22 -0800297void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700298 spinlock_t *ptl)
Christoph Lameter06972122006-06-23 02:03:35 -0700299{
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700300 pte_t pte;
Christoph Lameter06972122006-06-23 02:03:35 -0700301 swp_entry_t entry;
302 struct page *page;
303
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700304 spin_lock(ptl);
Christoph Lameter06972122006-06-23 02:03:35 -0700305 pte = *ptep;
306 if (!is_swap_pte(pte))
307 goto out;
308
309 entry = pte_to_swp_entry(pte);
310 if (!is_migration_entry(entry))
311 goto out;
312
313 page = migration_entry_to_page(entry);
314
Nick Piggine2867812008-07-25 19:45:30 -0700315 /*
316 * Once radix-tree replacement of page migration started, page_count
317 * *must* be zero. And, we don't want to call wait_on_page_locked()
318 * against a page without get_page().
319 * So, we use get_page_unless_zero(), here. Even failed, page fault
320 * will occur again.
321 */
322 if (!get_page_unless_zero(page))
323 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700324 pte_unmap_unlock(ptep, ptl);
325 wait_on_page_locked(page);
326 put_page(page);
327 return;
328out:
329 pte_unmap_unlock(ptep, ptl);
330}
331
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700332void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
333 unsigned long address)
334{
335 spinlock_t *ptl = pte_lockptr(mm, pmd);
336 pte_t *ptep = pte_offset_map(pmd, address);
337 __migration_entry_wait(mm, ptep, ptl);
338}
339
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800340void migration_entry_wait_huge(struct vm_area_struct *vma,
341 struct mm_struct *mm, pte_t *pte)
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700342{
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800343 spinlock_t *ptl = huge_pte_lockptr(hstate_vma(vma), mm, pte);
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700344 __migration_entry_wait(mm, pte, ptl);
345}
346
Mel Gormanb969c4a2012-01-12 17:19:34 -0800347#ifdef CONFIG_BLOCK
348/* Returns true if all buffers are successfully locked */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800349static bool buffer_migrate_lock_buffers(struct buffer_head *head,
350 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800351{
352 struct buffer_head *bh = head;
353
354 /* Simple case, sync compaction */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800355 if (mode != MIGRATE_ASYNC) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800356 do {
357 get_bh(bh);
358 lock_buffer(bh);
359 bh = bh->b_this_page;
360
361 } while (bh != head);
362
363 return true;
364 }
365
366 /* async case, we cannot block on lock_buffer so use trylock_buffer */
367 do {
368 get_bh(bh);
369 if (!trylock_buffer(bh)) {
370 /*
371 * We failed to lock the buffer and cannot stall in
372 * async migration. Release the taken locks
373 */
374 struct buffer_head *failed_bh = bh;
375 put_bh(failed_bh);
376 bh = head;
377 while (bh != failed_bh) {
378 unlock_buffer(bh);
379 put_bh(bh);
380 bh = bh->b_this_page;
381 }
382 return false;
383 }
384
385 bh = bh->b_this_page;
386 } while (bh != head);
387 return true;
388}
389#else
390static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800391 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800392{
393 return true;
394}
395#endif /* CONFIG_BLOCK */
396
Christoph Lameterb20a3502006-03-22 00:09:12 -0800397/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700398 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700399 *
400 * The number of remaining references must be:
401 * 1 for anonymous pages without a mapping
402 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100403 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800404 */
Gu Zheng36bc08c2013-07-16 17:56:16 +0800405int migrate_page_move_mapping(struct address_space *mapping,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800406 struct page *newpage, struct page *page,
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500407 struct buffer_head *head, enum migrate_mode mode,
408 int extra_count)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800409{
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800410 struct zone *oldzone, *newzone;
411 int dirty;
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500412 int expected_count = 1 + extra_count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800413 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800414
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700415 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700416 /* Anonymous page without mapping */
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500417 if (page_count(page) != expected_count)
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700418 return -EAGAIN;
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800419
420 /* No turning back from here */
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800421 newpage->index = page->index;
422 newpage->mapping = page->mapping;
423 if (PageSwapBacked(page))
Hugh Dickinsfa9949d2016-05-19 17:12:41 -0700424 __SetPageSwapBacked(newpage);
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800425
Rafael Aquini78bd5202012-12-11 16:02:31 -0800426 return MIGRATEPAGE_SUCCESS;
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700427 }
428
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800429 oldzone = page_zone(page);
430 newzone = page_zone(newpage);
431
Nick Piggin19fd6232008-07-25 19:45:32 -0700432 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800433
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800434 pslot = radix_tree_lookup_slot(&mapping->page_tree,
435 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800436
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500437 expected_count += 1 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700438 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800439 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700440 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700441 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800442 }
443
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700444 if (!page_ref_freeze(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700445 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700446 return -EAGAIN;
447 }
448
Christoph Lameterb20a3502006-03-22 00:09:12 -0800449 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800450 * In the async migration case of moving a page with buffers, lock the
451 * buffers using trylock before the mapping is moved. If the mapping
452 * was moved, we later failed to lock the buffers and could not move
453 * the mapping back due to an elevated page count, we would have to
454 * block waiting on other references to be dropped.
455 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800456 if (mode == MIGRATE_ASYNC && head &&
457 !buffer_migrate_lock_buffers(head, mode)) {
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700458 page_ref_unfreeze(page, expected_count);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800459 spin_unlock_irq(&mapping->tree_lock);
460 return -EAGAIN;
461 }
462
463 /*
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800464 * Now we know that no one else is looking at the page:
465 * no turning back from here.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800466 */
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800467 newpage->index = page->index;
468 newpage->mapping = page->mapping;
469 if (PageSwapBacked(page))
Hugh Dickinsfa9949d2016-05-19 17:12:41 -0700470 __SetPageSwapBacked(newpage);
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800471
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800472 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800473 if (PageSwapCache(page)) {
474 SetPageSwapCache(newpage);
475 set_page_private(newpage, page_private(page));
476 }
477
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800478 /* Move dirty while page refs frozen and newpage not yet exposed */
479 dirty = PageDirty(page);
480 if (dirty) {
481 ClearPageDirty(page);
482 SetPageDirty(newpage);
483 }
484
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800485 radix_tree_replace_slot(pslot, newpage);
486
487 /*
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800488 * Drop cache reference from old page by unfreezing
489 * to one less reference.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800490 * We know this isn't the last reference.
491 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700492 page_ref_unfreeze(page, expected_count - 1);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800493
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800494 spin_unlock(&mapping->tree_lock);
495 /* Leave irq disabled to prevent preemption while updating stats */
496
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700497 /*
498 * If moved to a different zone then also account
499 * the page for that zone. Other VM counters will be
500 * taken care of when we establish references to the
501 * new page and drop references to the old page.
502 *
503 * Note that anonymous pages are accounted for
Mel Gorman4b9d0fa2016-07-28 15:46:17 -0700504 * via NR_FILE_PAGES and NR_ANON_MAPPED if they
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700505 * are mapped to swap space.
506 */
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800507 if (newzone != oldzone) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700508 __dec_node_state(oldzone->zone_pgdat, NR_FILE_PAGES);
509 __inc_node_state(newzone->zone_pgdat, NR_FILE_PAGES);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800510 if (PageSwapBacked(page) && !PageSwapCache(page)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700511 __dec_node_state(oldzone->zone_pgdat, NR_SHMEM);
512 __inc_node_state(newzone->zone_pgdat, NR_SHMEM);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800513 }
514 if (dirty && mapping_cap_account_dirty(mapping)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700515 __dec_node_state(oldzone->zone_pgdat, NR_FILE_DIRTY);
Mel Gorman5a1c84b2016-07-28 15:47:31 -0700516 __dec_zone_state(oldzone, NR_ZONE_WRITE_PENDING);
Mel Gorman11fb9982016-07-28 15:46:20 -0700517 __inc_node_state(newzone->zone_pgdat, NR_FILE_DIRTY);
Mel Gorman5a1c84b2016-07-28 15:47:31 -0700518 __inc_zone_state(newzone, NR_ZONE_WRITE_PENDING);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800519 }
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700520 }
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800521 local_irq_enable();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800522
Rafael Aquini78bd5202012-12-11 16:02:31 -0800523 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800524}
Richard Weinberger1118dce2016-06-16 23:26:14 +0200525EXPORT_SYMBOL(migrate_page_move_mapping);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800526
527/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900528 * The expected number of remaining references is the same as that
529 * of migrate_page_move_mapping().
530 */
531int migrate_huge_page_move_mapping(struct address_space *mapping,
532 struct page *newpage, struct page *page)
533{
534 int expected_count;
535 void **pslot;
536
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900537 spin_lock_irq(&mapping->tree_lock);
538
539 pslot = radix_tree_lookup_slot(&mapping->page_tree,
540 page_index(page));
541
542 expected_count = 2 + page_has_private(page);
543 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800544 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900545 spin_unlock_irq(&mapping->tree_lock);
546 return -EAGAIN;
547 }
548
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700549 if (!page_ref_freeze(page, expected_count)) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900550 spin_unlock_irq(&mapping->tree_lock);
551 return -EAGAIN;
552 }
553
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800554 newpage->index = page->index;
555 newpage->mapping = page->mapping;
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700556
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900557 get_page(newpage);
558
559 radix_tree_replace_slot(pslot, newpage);
560
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700561 page_ref_unfreeze(page, expected_count - 1);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900562
563 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700564
Rafael Aquini78bd5202012-12-11 16:02:31 -0800565 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900566}
567
568/*
Dave Hansen30b0a102013-11-21 14:31:58 -0800569 * Gigantic pages are so large that we do not guarantee that page++ pointer
570 * arithmetic will work across the entire page. We need something more
571 * specialized.
572 */
573static void __copy_gigantic_page(struct page *dst, struct page *src,
574 int nr_pages)
575{
576 int i;
577 struct page *dst_base = dst;
578 struct page *src_base = src;
579
580 for (i = 0; i < nr_pages; ) {
581 cond_resched();
582 copy_highpage(dst, src);
583
584 i++;
585 dst = mem_map_next(dst, dst_base, i);
586 src = mem_map_next(src, src_base, i);
587 }
588}
589
590static void copy_huge_page(struct page *dst, struct page *src)
591{
592 int i;
593 int nr_pages;
594
595 if (PageHuge(src)) {
596 /* hugetlbfs page */
597 struct hstate *h = page_hstate(src);
598 nr_pages = pages_per_huge_page(h);
599
600 if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) {
601 __copy_gigantic_page(dst, src, nr_pages);
602 return;
603 }
604 } else {
605 /* thp page */
606 BUG_ON(!PageTransHuge(src));
607 nr_pages = hpage_nr_pages(src);
608 }
609
610 for (i = 0; i < nr_pages; i++) {
611 cond_resched();
612 copy_highpage(dst + i, src + i);
613 }
614}
615
616/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800617 * Copy the page to its new location
618 */
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900619void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800620{
Rik van Riel7851a452013-10-07 11:29:23 +0100621 int cpupid;
622
Mel Gormanb32967f2012-11-19 12:35:47 +0000623 if (PageHuge(page) || PageTransHuge(page))
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900624 copy_huge_page(newpage, page);
625 else
626 copy_highpage(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800627
628 if (PageError(page))
629 SetPageError(newpage);
630 if (PageReferenced(page))
631 SetPageReferenced(newpage);
632 if (PageUptodate(page))
633 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700634 if (TestClearPageActive(page)) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800635 VM_BUG_ON_PAGE(PageUnevictable(page), page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800636 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800637 } else if (TestClearPageUnevictable(page))
638 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800639 if (PageChecked(page))
640 SetPageChecked(newpage);
641 if (PageMappedToDisk(page))
642 SetPageMappedToDisk(newpage);
643
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800644 /* Move dirty on pages not done by migrate_page_move_mapping() */
645 if (PageDirty(page))
646 SetPageDirty(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800647
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700648 if (page_is_young(page))
649 set_page_young(newpage);
650 if (page_is_idle(page))
651 set_page_idle(newpage);
652
Rik van Riel7851a452013-10-07 11:29:23 +0100653 /*
654 * Copy NUMA information to the new page, to prevent over-eager
655 * future migrations of this same page.
656 */
657 cpupid = page_cpupid_xchg_last(page, -1);
658 page_cpupid_xchg_last(newpage, cpupid);
659
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800660 ksm_migrate_page(newpage, page);
Hugh Dickinsc8d65532013-02-22 16:35:10 -0800661 /*
662 * Please do not reorder this without considering how mm/ksm.c's
663 * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache().
664 */
Naoya Horiguchib3b3a992015-04-15 16:13:15 -0700665 if (PageSwapCache(page))
666 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800667 ClearPagePrivate(page);
668 set_page_private(page, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800669
670 /*
671 * If any waiters have accumulated on the new page then
672 * wake them up.
673 */
674 if (PageWriteback(newpage))
675 end_page_writeback(newpage);
Vlastimil Babkad435edc2016-03-15 14:56:15 -0700676
677 copy_page_owner(page, newpage);
Johannes Weiner74485cf2016-03-15 14:57:54 -0700678
679 mem_cgroup_migrate(page, newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800680}
Richard Weinberger1118dce2016-06-16 23:26:14 +0200681EXPORT_SYMBOL(migrate_page_copy);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800682
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700683/************************************************************
684 * Migration functions
685 ***********************************************************/
686
Christoph Lameterb20a3502006-03-22 00:09:12 -0800687/*
Minchan Kimbda807d2016-07-26 15:23:05 -0700688 * Common logic to directly migrate a single LRU page suitable for
David Howells266cf652009-04-03 16:42:36 +0100689 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800690 *
691 * Pages are locked upon entry and exit.
692 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700693int migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800694 struct page *newpage, struct page *page,
695 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800696{
697 int rc;
698
699 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
700
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500701 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800702
Rafael Aquini78bd5202012-12-11 16:02:31 -0800703 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800704 return rc;
705
706 migrate_page_copy(newpage, page);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800707 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800708}
709EXPORT_SYMBOL(migrate_page);
710
David Howells93614012006-09-30 20:45:40 +0200711#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800712/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700713 * Migration function for pages with buffers. This function can only be used
714 * if the underlying filesystem guarantees that no other references to "page"
715 * exist.
716 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700717int buffer_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800718 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700719{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700720 struct buffer_head *bh, *head;
721 int rc;
722
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700723 if (!page_has_buffers(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800724 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700725
726 head = page_buffers(page);
727
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500728 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode, 0);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700729
Rafael Aquini78bd5202012-12-11 16:02:31 -0800730 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700731 return rc;
732
Mel Gormanb969c4a2012-01-12 17:19:34 -0800733 /*
734 * In the async case, migrate_page_move_mapping locked the buffers
735 * with an IRQ-safe spinlock held. In the sync case, the buffers
736 * need to be locked now
737 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800738 if (mode != MIGRATE_ASYNC)
739 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700740
741 ClearPagePrivate(page);
742 set_page_private(newpage, page_private(page));
743 set_page_private(page, 0);
744 put_page(page);
745 get_page(newpage);
746
747 bh = head;
748 do {
749 set_bh_page(bh, newpage, bh_offset(bh));
750 bh = bh->b_this_page;
751
752 } while (bh != head);
753
754 SetPagePrivate(newpage);
755
756 migrate_page_copy(newpage, page);
757
758 bh = head;
759 do {
760 unlock_buffer(bh);
761 put_bh(bh);
762 bh = bh->b_this_page;
763
764 } while (bh != head);
765
Rafael Aquini78bd5202012-12-11 16:02:31 -0800766 return MIGRATEPAGE_SUCCESS;
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700767}
768EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200769#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700770
Christoph Lameter04e62a22006-06-23 02:03:38 -0700771/*
772 * Writeback a page to clean the dirty state
773 */
774static int writeout(struct address_space *mapping, struct page *page)
775{
776 struct writeback_control wbc = {
777 .sync_mode = WB_SYNC_NONE,
778 .nr_to_write = 1,
779 .range_start = 0,
780 .range_end = LLONG_MAX,
Christoph Lameter04e62a22006-06-23 02:03:38 -0700781 .for_reclaim = 1
782 };
783 int rc;
784
785 if (!mapping->a_ops->writepage)
786 /* No write method for the address space */
787 return -EINVAL;
788
789 if (!clear_page_dirty_for_io(page))
790 /* Someone else already triggered a write */
791 return -EAGAIN;
792
793 /*
794 * A dirty page may imply that the underlying filesystem has
795 * the page on some queue. So the page must be clean for
796 * migration. Writeout may mean we loose the lock and the
797 * page state is no longer what we checked for earlier.
798 * At this point we know that the migration attempt cannot
799 * be successful.
800 */
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700801 remove_migration_ptes(page, page, false);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700802
803 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700804
805 if (rc != AOP_WRITEPAGE_ACTIVATE)
806 /* unlocked. Relock */
807 lock_page(page);
808
Hugh Dickinsbda85502008-11-19 15:36:36 -0800809 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700810}
811
812/*
813 * Default handling if a filesystem does not provide a migration function.
814 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700815static int fallback_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800816 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700817{
Mel Gormanb969c4a2012-01-12 17:19:34 -0800818 if (PageDirty(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800819 /* Only writeback pages in full synchronous migration */
820 if (mode != MIGRATE_SYNC)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800821 return -EBUSY;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700822 return writeout(mapping, page);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800823 }
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700824
825 /*
826 * Buffers may be managed in a filesystem specific way.
827 * We must have no buffers or drop them.
828 */
David Howells266cf652009-04-03 16:42:36 +0100829 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700830 !try_to_release_page(page, GFP_KERNEL))
831 return -EAGAIN;
832
Mel Gormana6bc32b2012-01-12 17:19:43 -0800833 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700834}
835
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700836/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700837 * Move a page to a newly allocated page
838 * The page is locked and all ptes have been successfully removed.
839 *
840 * The new page will have replaced the old page if this function
841 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700842 *
843 * Return value:
844 * < 0 - error code
Rafael Aquini78bd5202012-12-11 16:02:31 -0800845 * MIGRATEPAGE_SUCCESS - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700846 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700847static int move_to_new_page(struct page *newpage, struct page *page,
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800848 enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700849{
850 struct address_space *mapping;
Minchan Kimbda807d2016-07-26 15:23:05 -0700851 int rc = -EAGAIN;
852 bool is_lru = !__PageMovable(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700853
Hugh Dickins7db76712015-11-05 18:49:49 -0800854 VM_BUG_ON_PAGE(!PageLocked(page), page);
855 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700856
Christoph Lametere24f0b82006-06-23 02:03:51 -0700857 mapping = page_mapping(page);
Minchan Kimbda807d2016-07-26 15:23:05 -0700858
859 if (likely(is_lru)) {
860 if (!mapping)
861 rc = migrate_page(mapping, newpage, page, mode);
862 else if (mapping->a_ops->migratepage)
863 /*
864 * Most pages have a mapping and most filesystems
865 * provide a migratepage callback. Anonymous pages
866 * are part of swap space which also has its own
867 * migratepage callback. This is the most common path
868 * for page migration.
869 */
870 rc = mapping->a_ops->migratepage(mapping, newpage,
871 page, mode);
872 else
873 rc = fallback_migrate_page(mapping, newpage,
874 page, mode);
875 } else {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700876 /*
Minchan Kimbda807d2016-07-26 15:23:05 -0700877 * In case of non-lru page, it could be released after
878 * isolation step. In that case, we shouldn't try migration.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700879 */
Minchan Kimbda807d2016-07-26 15:23:05 -0700880 VM_BUG_ON_PAGE(!PageIsolated(page), page);
881 if (!PageMovable(page)) {
882 rc = MIGRATEPAGE_SUCCESS;
883 __ClearPageIsolated(page);
884 goto out;
885 }
886
887 rc = mapping->a_ops->migratepage(mapping, newpage,
888 page, mode);
889 WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS &&
890 !PageIsolated(page));
891 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700892
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800893 /*
894 * When successful, old pagecache page->mapping must be cleared before
895 * page is freed; but stats require that PageAnon be left as PageAnon.
896 */
897 if (rc == MIGRATEPAGE_SUCCESS) {
Minchan Kimbda807d2016-07-26 15:23:05 -0700898 if (__PageMovable(page)) {
899 VM_BUG_ON_PAGE(!PageIsolated(page), page);
900
901 /*
902 * We clear PG_movable under page_lock so any compactor
903 * cannot try to migrate this page.
904 */
905 __ClearPageIsolated(page);
906 }
907
908 /*
909 * Anonymous and movable page->mapping will be cleard by
910 * free_pages_prepare so don't reset it here for keeping
911 * the type to work PageAnon, for example.
912 */
913 if (!PageMappingFlags(page))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800914 page->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700915 }
Minchan Kimbda807d2016-07-26 15:23:05 -0700916out:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700917 return rc;
918}
919
Minchan Kim0dabec92011-10-31 17:06:57 -0700920static int __unmap_and_move(struct page *page, struct page *newpage,
Hugh Dickins9c620e22013-02-22 16:35:14 -0800921 int force, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700922{
Minchan Kim0dabec92011-10-31 17:06:57 -0700923 int rc = -EAGAIN;
Hugh Dickins2ebba6b2014-12-12 16:56:19 -0800924 int page_was_mapped = 0;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700925 struct anon_vma *anon_vma = NULL;
Minchan Kimbda807d2016-07-26 15:23:05 -0700926 bool is_lru = !__PageMovable(page);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700927
Nick Piggin529ae9a2008-08-02 12:01:03 +0200928 if (!trylock_page(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800929 if (!force || mode == MIGRATE_ASYNC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700930 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800931
932 /*
933 * It's not safe for direct compaction to call lock_page.
934 * For example, during page readahead pages are added locked
935 * to the LRU. Later, when the IO completes the pages are
936 * marked uptodate and unlocked. However, the queueing
937 * could be merging multiple pages for one bio (e.g.
938 * mpage_readpages). If an allocation happens for the
939 * second or third page, the process can end up locking
940 * the same page twice and deadlocking. Rather than
941 * trying to be clever about what pages can be locked,
942 * avoid the use of lock_page for direct compaction
943 * altogether.
944 */
945 if (current->flags & PF_MEMALLOC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700946 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800947
Christoph Lametere24f0b82006-06-23 02:03:51 -0700948 lock_page(page);
949 }
950
951 if (PageWriteback(page)) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700952 /*
Jianguo Wufed5b642013-04-29 15:07:58 -0700953 * Only in the case of a full synchronous migration is it
Mel Gormana6bc32b2012-01-12 17:19:43 -0800954 * necessary to wait for PageWriteback. In the async case,
955 * the retry loop is too short and in the sync-light case,
956 * the overhead of stalling is too much
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700957 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800958 if (mode != MIGRATE_SYNC) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700959 rc = -EBUSY;
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700960 goto out_unlock;
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700961 }
962 if (!force)
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700963 goto out_unlock;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700964 wait_on_page_writeback(page);
965 }
Hugh Dickins03f15c82015-11-05 18:49:56 -0800966
Christoph Lametere24f0b82006-06-23 02:03:51 -0700967 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700968 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
969 * we cannot notice that anon_vma is freed while we migrates a page.
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800970 * This get_anon_vma() delays freeing anon_vma pointer until the end
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700971 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700972 * File Caches may use write_page() or lock_page() in migration, then,
973 * just care Anon page here.
Hugh Dickins03f15c82015-11-05 18:49:56 -0800974 *
975 * Only page_get_anon_vma() understands the subtleties of
976 * getting a hold on an anon_vma from outside one of its mms.
977 * But if we cannot get anon_vma, then we won't need it anyway,
978 * because that implies that the anon page is no longer mapped
979 * (and cannot be remapped so long as we hold the page lock).
Christoph Lametere24f0b82006-06-23 02:03:51 -0700980 */
Hugh Dickins03f15c82015-11-05 18:49:56 -0800981 if (PageAnon(page) && !PageKsm(page))
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700982 anon_vma = page_get_anon_vma(page);
Shaohua Li62e1c552008-02-04 22:29:33 -0800983
Hugh Dickins7db76712015-11-05 18:49:49 -0800984 /*
985 * Block others from accessing the new page when we get around to
986 * establishing additional references. We are usually the only one
987 * holding a reference to newpage at this point. We used to have a BUG
988 * here if trylock_page(newpage) fails, but would like to allow for
989 * cases where there might be a race with the previous use of newpage.
990 * This is much like races on refcount of oldpage: just don't BUG().
991 */
992 if (unlikely(!trylock_page(newpage)))
993 goto out_unlock;
994
Minchan Kimbda807d2016-07-26 15:23:05 -0700995 if (unlikely(!is_lru)) {
996 rc = move_to_new_page(newpage, page, mode);
997 goto out_unlock_both;
998 }
999
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001000 /*
Shaohua Li62e1c552008-02-04 22:29:33 -08001001 * Corner case handling:
1002 * 1. When a new swap-cache page is read into, it is added to the LRU
1003 * and treated as swapcache but it has no rmap yet.
1004 * Calling try_to_unmap() against a page->mapping==NULL page will
1005 * trigger a BUG. So handle it here.
1006 * 2. An orphaned page (see truncate_complete_page) might have
1007 * fs-private metadata. The page can be picked up due to memory
1008 * offlining. Everywhere else except page reclaim, the page is
1009 * invisible to the vm, so the page can not be migrated. So try to
1010 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001011 */
Shaohua Li62e1c552008-02-04 22:29:33 -08001012 if (!page->mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -08001013 VM_BUG_ON_PAGE(PageAnon(page), page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -08001014 if (page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -08001015 try_to_free_buffers(page);
Hugh Dickins7db76712015-11-05 18:49:49 -08001016 goto out_unlock_both;
Shaohua Li62e1c552008-02-04 22:29:33 -08001017 }
Hugh Dickins7db76712015-11-05 18:49:49 -08001018 } else if (page_mapped(page)) {
1019 /* Establish migration ptes */
Hugh Dickins03f15c82015-11-05 18:49:56 -08001020 VM_BUG_ON_PAGE(PageAnon(page) && !PageKsm(page) && !anon_vma,
1021 page);
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001022 try_to_unmap(page,
Wanpeng Lida1b13c2015-09-08 15:03:27 -07001023 TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001024 page_was_mapped = 1;
1025 }
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001026
Christoph Lametere6a15302006-06-25 05:46:49 -07001027 if (!page_mapped(page))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001028 rc = move_to_new_page(newpage, page, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -07001029
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001030 if (page_was_mapped)
1031 remove_migration_ptes(page,
Kirill A. Shutemove3884662016-03-17 14:20:07 -07001032 rc == MIGRATEPAGE_SUCCESS ? newpage : page, false);
Mel Gorman3f6c8272010-05-24 14:32:17 -07001033
Hugh Dickins7db76712015-11-05 18:49:49 -08001034out_unlock_both:
1035 unlock_page(newpage);
1036out_unlock:
Mel Gorman3f6c8272010-05-24 14:32:17 -07001037 /* Drop an anon_vma reference if we took one */
Rik van Riel76545062010-08-09 17:18:41 -07001038 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -07001039 put_anon_vma(anon_vma);
Christoph Lametere24f0b82006-06-23 02:03:51 -07001040 unlock_page(page);
Minchan Kim0dabec92011-10-31 17:06:57 -07001041out:
Minchan Kimc6c919e2016-07-26 15:23:02 -07001042 /*
1043 * If migration is successful, decrease refcount of the newpage
1044 * which will not free the page because new page owner increased
1045 * refcounter. As well, if it is LRU page, add the page to LRU
1046 * list in here.
1047 */
1048 if (rc == MIGRATEPAGE_SUCCESS) {
Minchan Kimb1123ea62016-07-26 15:23:09 -07001049 if (unlikely(__PageMovable(newpage)))
Minchan Kimc6c919e2016-07-26 15:23:02 -07001050 put_page(newpage);
1051 else
1052 putback_lru_page(newpage);
1053 }
1054
Minchan Kim0dabec92011-10-31 17:06:57 -07001055 return rc;
1056}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001057
Minchan Kim0dabec92011-10-31 17:06:57 -07001058/*
Geert Uytterhoevenef2a5152015-04-14 15:44:22 -07001059 * gcc 4.7 and 4.8 on arm get an ICEs when inlining unmap_and_move(). Work
1060 * around it.
1061 */
1062#if (GCC_VERSION >= 40700 && GCC_VERSION < 40900) && defined(CONFIG_ARM)
1063#define ICE_noinline noinline
1064#else
1065#define ICE_noinline
1066#endif
1067
1068/*
Minchan Kim0dabec92011-10-31 17:06:57 -07001069 * Obtain the lock on page, remove all ptes and migrate the page
1070 * to the newly allocated page in newpage.
1071 */
Geert Uytterhoevenef2a5152015-04-14 15:44:22 -07001072static ICE_noinline int unmap_and_move(new_page_t get_new_page,
1073 free_page_t put_new_page,
1074 unsigned long private, struct page *page,
Naoya Horiguchiadd05ce2015-06-24 16:56:50 -07001075 int force, enum migrate_mode mode,
1076 enum migrate_reason reason)
Minchan Kim0dabec92011-10-31 17:06:57 -07001077{
Hugh Dickins2def7422015-11-05 18:49:46 -08001078 int rc = MIGRATEPAGE_SUCCESS;
Minchan Kim0dabec92011-10-31 17:06:57 -07001079 int *result = NULL;
Hugh Dickins2def7422015-11-05 18:49:46 -08001080 struct page *newpage;
Minchan Kim0dabec92011-10-31 17:06:57 -07001081
Hugh Dickins2def7422015-11-05 18:49:46 -08001082 newpage = get_new_page(page, private, &result);
Minchan Kim0dabec92011-10-31 17:06:57 -07001083 if (!newpage)
1084 return -ENOMEM;
1085
1086 if (page_count(page) == 1) {
1087 /* page was freed from under us. So we are done. */
Minchan Kimc6c919e2016-07-26 15:23:02 -07001088 ClearPageActive(page);
1089 ClearPageUnevictable(page);
Minchan Kimbda807d2016-07-26 15:23:05 -07001090 if (unlikely(__PageMovable(page))) {
1091 lock_page(page);
1092 if (!PageMovable(page))
1093 __ClearPageIsolated(page);
1094 unlock_page(page);
1095 }
Minchan Kimc6c919e2016-07-26 15:23:02 -07001096 if (put_new_page)
1097 put_new_page(newpage, private);
1098 else
1099 put_page(newpage);
Minchan Kim0dabec92011-10-31 17:06:57 -07001100 goto out;
1101 }
1102
Kirill A. Shutemov4d2fa962016-01-15 16:54:00 -08001103 if (unlikely(PageTransHuge(page))) {
1104 lock_page(page);
1105 rc = split_huge_page(page);
1106 unlock_page(page);
1107 if (rc)
Minchan Kim0dabec92011-10-31 17:06:57 -07001108 goto out;
Kirill A. Shutemov4d2fa962016-01-15 16:54:00 -08001109 }
Minchan Kim0dabec92011-10-31 17:06:57 -07001110
Hugh Dickins9c620e22013-02-22 16:35:14 -08001111 rc = __unmap_and_move(page, newpage, force, mode);
Minchan Kimc6c919e2016-07-26 15:23:02 -07001112 if (rc == MIGRATEPAGE_SUCCESS)
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001113 set_page_owner_migrate_reason(newpage, reason);
Rafael Aquinibf6bddf2012-12-11 16:02:42 -08001114
Minchan Kim0dabec92011-10-31 17:06:57 -07001115out:
Christoph Lametere24f0b82006-06-23 02:03:51 -07001116 if (rc != -EAGAIN) {
Minchan Kim0dabec92011-10-31 17:06:57 -07001117 /*
1118 * A page that has been migrated has all references
1119 * removed and will be freed. A page that has not been
1120 * migrated will have kepts its references and be
1121 * restored.
1122 */
1123 list_del(&page->lru);
Mel Gorman599d0c92016-07-28 15:45:31 -07001124 dec_node_page_state(page, NR_ISOLATED_ANON +
Johannes Weiner6c0b1352009-09-21 17:02:59 -07001125 page_is_file_cache(page));
Christoph Lametere24f0b82006-06-23 02:03:51 -07001126 }
David Rientjes68711a72014-06-04 16:08:25 -07001127
Christoph Lameter95a402c2006-06-23 02:03:53 -07001128 /*
Minchan Kimc6c919e2016-07-26 15:23:02 -07001129 * If migration is successful, releases reference grabbed during
1130 * isolation. Otherwise, restore the page to right list unless
1131 * we want to retry.
Christoph Lameter95a402c2006-06-23 02:03:53 -07001132 */
Minchan Kimc6c919e2016-07-26 15:23:02 -07001133 if (rc == MIGRATEPAGE_SUCCESS) {
1134 put_page(page);
1135 if (reason == MR_MEMORY_FAILURE) {
1136 /*
1137 * Set PG_HWPoison on just freed page
1138 * intentionally. Although it's rather weird,
1139 * it's how HWPoison flag works at the moment.
1140 */
1141 if (!test_set_page_hwpoison(page))
1142 num_poisoned_pages_inc();
1143 }
1144 } else {
Minchan Kimbda807d2016-07-26 15:23:05 -07001145 if (rc != -EAGAIN) {
1146 if (likely(!__PageMovable(page))) {
1147 putback_lru_page(page);
1148 goto put_new;
1149 }
1150
1151 lock_page(page);
1152 if (PageMovable(page))
1153 putback_movable_page(page);
1154 else
1155 __ClearPageIsolated(page);
1156 unlock_page(page);
1157 put_page(page);
1158 }
1159put_new:
Minchan Kimc6c919e2016-07-26 15:23:02 -07001160 if (put_new_page)
1161 put_new_page(newpage, private);
1162 else
1163 put_page(newpage);
1164 }
David Rientjes68711a72014-06-04 16:08:25 -07001165
Christoph Lameter742755a2006-06-23 02:03:55 -07001166 if (result) {
1167 if (rc)
1168 *result = rc;
1169 else
1170 *result = page_to_nid(newpage);
1171 }
Christoph Lametere24f0b82006-06-23 02:03:51 -07001172 return rc;
1173}
1174
1175/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001176 * Counterpart of unmap_and_move_page() for hugepage migration.
1177 *
1178 * This function doesn't wait the completion of hugepage I/O
1179 * because there is no race between I/O and migration for hugepage.
1180 * Note that currently hugepage I/O occurs only in direct I/O
1181 * where no lock is held and PG_writeback is irrelevant,
1182 * and writeback status of all subpages are counted in the reference
1183 * count of the head page (i.e. if all subpages of a 2MB hugepage are
1184 * under direct I/O, the reference of the head page is 512 and a bit more.)
1185 * This means that when we try to migrate hugepage whose subpages are
1186 * doing direct I/O, some references remain after try_to_unmap() and
1187 * hugepage migration fails without data corruption.
1188 *
1189 * There is also no race when direct I/O is issued on the page under migration,
1190 * because then pte is replaced with migration swap entry and direct I/O code
1191 * will wait in the page fault for migration to complete.
1192 */
1193static int unmap_and_move_huge_page(new_page_t get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001194 free_page_t put_new_page, unsigned long private,
1195 struct page *hpage, int force,
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001196 enum migrate_mode mode, int reason)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001197{
Hugh Dickins2def7422015-11-05 18:49:46 -08001198 int rc = -EAGAIN;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001199 int *result = NULL;
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001200 int page_was_mapped = 0;
Joonsoo Kim32665f22014-01-21 15:51:15 -08001201 struct page *new_hpage;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001202 struct anon_vma *anon_vma = NULL;
1203
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001204 /*
1205 * Movability of hugepages depends on architectures and hugepage size.
1206 * This check is necessary because some callers of hugepage migration
1207 * like soft offline and memory hotremove don't walk through page
1208 * tables or check whether the hugepage is pmd-based or not before
1209 * kicking migration.
1210 */
Naoya Horiguchi100873d2014-06-04 16:10:56 -07001211 if (!hugepage_migration_supported(page_hstate(hpage))) {
Joonsoo Kim32665f22014-01-21 15:51:15 -08001212 putback_active_hugepage(hpage);
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001213 return -ENOSYS;
Joonsoo Kim32665f22014-01-21 15:51:15 -08001214 }
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001215
Joonsoo Kim32665f22014-01-21 15:51:15 -08001216 new_hpage = get_new_page(hpage, private, &result);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001217 if (!new_hpage)
1218 return -ENOMEM;
1219
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001220 if (!trylock_page(hpage)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -08001221 if (!force || mode != MIGRATE_SYNC)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001222 goto out;
1223 lock_page(hpage);
1224 }
1225
Peter Zijlstra746b18d2011-05-24 17:12:10 -07001226 if (PageAnon(hpage))
1227 anon_vma = page_get_anon_vma(hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001228
Hugh Dickins7db76712015-11-05 18:49:49 -08001229 if (unlikely(!trylock_page(new_hpage)))
1230 goto put_anon;
1231
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001232 if (page_mapped(hpage)) {
1233 try_to_unmap(hpage,
1234 TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
1235 page_was_mapped = 1;
1236 }
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001237
1238 if (!page_mapped(hpage))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001239 rc = move_to_new_page(new_hpage, hpage, mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001240
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001241 if (page_was_mapped)
1242 remove_migration_ptes(hpage,
Kirill A. Shutemove3884662016-03-17 14:20:07 -07001243 rc == MIGRATEPAGE_SUCCESS ? new_hpage : hpage, false);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001244
Hugh Dickins7db76712015-11-05 18:49:49 -08001245 unlock_page(new_hpage);
1246
1247put_anon:
Hugh Dickinsfd4a4662011-01-13 15:47:31 -08001248 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -07001249 put_anon_vma(anon_vma);
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001250
Hugh Dickins2def7422015-11-05 18:49:46 -08001251 if (rc == MIGRATEPAGE_SUCCESS) {
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001252 hugetlb_cgroup_migrate(hpage, new_hpage);
Hugh Dickins2def7422015-11-05 18:49:46 -08001253 put_new_page = NULL;
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001254 set_page_owner_migrate_reason(new_hpage, reason);
Hugh Dickins2def7422015-11-05 18:49:46 -08001255 }
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001256
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001257 unlock_page(hpage);
Hillf Danton09761332011-12-08 14:34:20 -08001258out:
Naoya Horiguchib8ec1ce2013-09-11 14:22:01 -07001259 if (rc != -EAGAIN)
1260 putback_active_hugepage(hpage);
David Rientjes68711a72014-06-04 16:08:25 -07001261
1262 /*
1263 * If migration was not successful and there's a freeing callback, use
1264 * it. Otherwise, put_page() will drop the reference grabbed during
1265 * isolation.
1266 */
Hugh Dickins2def7422015-11-05 18:49:46 -08001267 if (put_new_page)
David Rientjes68711a72014-06-04 16:08:25 -07001268 put_new_page(new_hpage, private);
1269 else
Naoya Horiguchi3aaa76e2015-09-22 14:59:14 -07001270 putback_active_hugepage(new_hpage);
David Rientjes68711a72014-06-04 16:08:25 -07001271
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001272 if (result) {
1273 if (rc)
1274 *result = rc;
1275 else
1276 *result = page_to_nid(new_hpage);
1277 }
1278 return rc;
1279}
1280
1281/*
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001282 * migrate_pages - migrate the pages specified in a list, to the free pages
1283 * supplied as the target for the page migration
Christoph Lameterb20a3502006-03-22 00:09:12 -08001284 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001285 * @from: The list of pages to be migrated.
1286 * @get_new_page: The function used to allocate free pages to be used
1287 * as the target of the page migration.
David Rientjes68711a72014-06-04 16:08:25 -07001288 * @put_new_page: The function used to free target pages if migration
1289 * fails, or NULL if no special handling is necessary.
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001290 * @private: Private data to be passed on to get_new_page()
1291 * @mode: The migration mode that specifies the constraints for
1292 * page migration, if any.
1293 * @reason: The reason for page migration.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001294 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001295 * The function returns after 10 attempts or if no pages are movable any more
1296 * because the list has become empty or no retryable pages exist any more.
Hugh Dickins14e0f9b2015-11-05 18:49:43 -08001297 * The caller should call putback_movable_pages() to return pages to the LRU
Minchan Kim28bd6572011-01-25 15:07:26 -08001298 * or free list only if ret != 0.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001299 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001300 * Returns the number of pages that were not migrated, or an error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001301 */
Hugh Dickins9c620e22013-02-22 16:35:14 -08001302int migrate_pages(struct list_head *from, new_page_t get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001303 free_page_t put_new_page, unsigned long private,
1304 enum migrate_mode mode, int reason)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001305{
Christoph Lametere24f0b82006-06-23 02:03:51 -07001306 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001307 int nr_failed = 0;
Mel Gorman5647bc22012-10-19 10:46:20 +01001308 int nr_succeeded = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001309 int pass = 0;
1310 struct page *page;
1311 struct page *page2;
1312 int swapwrite = current->flags & PF_SWAPWRITE;
1313 int rc;
1314
1315 if (!swapwrite)
1316 current->flags |= PF_SWAPWRITE;
1317
Christoph Lametere24f0b82006-06-23 02:03:51 -07001318 for(pass = 0; pass < 10 && retry; pass++) {
1319 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001320
Christoph Lametere24f0b82006-06-23 02:03:51 -07001321 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -07001322 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -08001323
Naoya Horiguchi31caf662013-09-11 14:21:59 -07001324 if (PageHuge(page))
1325 rc = unmap_and_move_huge_page(get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001326 put_new_page, private, page,
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001327 pass > 2, mode, reason);
Naoya Horiguchi31caf662013-09-11 14:21:59 -07001328 else
David Rientjes68711a72014-06-04 16:08:25 -07001329 rc = unmap_and_move(get_new_page, put_new_page,
Naoya Horiguchiadd05ce2015-06-24 16:56:50 -07001330 private, page, pass > 2, mode,
1331 reason);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001332
Christoph Lametere24f0b82006-06-23 02:03:51 -07001333 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -07001334 case -ENOMEM:
David Rientjesdfef2ef2016-05-20 16:59:05 -07001335 nr_failed++;
Christoph Lameter95a402c2006-06-23 02:03:53 -07001336 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001337 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001338 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001339 break;
Rafael Aquini78bd5202012-12-11 16:02:31 -08001340 case MIGRATEPAGE_SUCCESS:
Mel Gorman5647bc22012-10-19 10:46:20 +01001341 nr_succeeded++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001342 break;
1343 default:
Naoya Horiguchi354a3362014-01-21 15:51:14 -08001344 /*
1345 * Permanent failure (-EBUSY, -ENOSYS, etc.):
1346 * unlike -EAGAIN case, the failed page is
1347 * removed from migration page list and not
1348 * retried in the next outer loop.
1349 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001350 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001351 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001352 }
Christoph Lameterb20a3502006-03-22 00:09:12 -08001353 }
1354 }
Vlastimil Babkaf2f81fb2015-11-05 18:47:03 -08001355 nr_failed += retry;
1356 rc = nr_failed;
Christoph Lameter95a402c2006-06-23 02:03:53 -07001357out:
Mel Gorman5647bc22012-10-19 10:46:20 +01001358 if (nr_succeeded)
1359 count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded);
1360 if (nr_failed)
1361 count_vm_events(PGMIGRATE_FAIL, nr_failed);
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001362 trace_mm_migrate_pages(nr_succeeded, nr_failed, mode, reason);
1363
Christoph Lameterb20a3502006-03-22 00:09:12 -08001364 if (!swapwrite)
1365 current->flags &= ~PF_SWAPWRITE;
1366
Rafael Aquini78bd5202012-12-11 16:02:31 -08001367 return rc;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001368}
1369
Christoph Lameter742755a2006-06-23 02:03:55 -07001370#ifdef CONFIG_NUMA
1371/*
1372 * Move a list of individual pages
1373 */
1374struct page_to_node {
1375 unsigned long addr;
1376 struct page *page;
1377 int node;
1378 int status;
1379};
1380
1381static struct page *new_page_node(struct page *p, unsigned long private,
1382 int **result)
1383{
1384 struct page_to_node *pm = (struct page_to_node *)private;
1385
1386 while (pm->node != MAX_NUMNODES && pm->page != p)
1387 pm++;
1388
1389 if (pm->node == MAX_NUMNODES)
1390 return NULL;
1391
1392 *result = &pm->status;
1393
Naoya Horiguchie632a932013-09-11 14:22:04 -07001394 if (PageHuge(p))
1395 return alloc_huge_page_node(page_hstate(compound_head(p)),
1396 pm->node);
1397 else
Vlastimil Babka96db8002015-09-08 15:03:50 -07001398 return __alloc_pages_node(pm->node,
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001399 GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -07001400}
1401
1402/*
1403 * Move a set of pages as indicated in the pm array. The addr
1404 * field must be set to the virtual address of the page to be moved
1405 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001406 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -07001407 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001408static int do_move_page_to_node_array(struct mm_struct *mm,
1409 struct page_to_node *pm,
1410 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -07001411{
1412 int err;
1413 struct page_to_node *pp;
1414 LIST_HEAD(pagelist);
1415
1416 down_read(&mm->mmap_sem);
1417
1418 /*
1419 * Build a list of pages to migrate
1420 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001421 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1422 struct vm_area_struct *vma;
1423 struct page *page;
1424
Christoph Lameter742755a2006-06-23 02:03:55 -07001425 err = -EFAULT;
1426 vma = find_vma(mm, pp->addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001427 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -07001428 goto set_status;
1429
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001430 /* FOLL_DUMP to ignore special (like zero) pages */
1431 page = follow_page(vma, pp->addr,
1432 FOLL_GET | FOLL_SPLIT | FOLL_DUMP);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001433
1434 err = PTR_ERR(page);
1435 if (IS_ERR(page))
1436 goto set_status;
1437
Christoph Lameter742755a2006-06-23 02:03:55 -07001438 err = -ENOENT;
1439 if (!page)
1440 goto set_status;
1441
Christoph Lameter742755a2006-06-23 02:03:55 -07001442 pp->page = page;
1443 err = page_to_nid(page);
1444
1445 if (err == pp->node)
1446 /*
1447 * Node already in the right place
1448 */
1449 goto put_and_set;
1450
1451 err = -EACCES;
1452 if (page_mapcount(page) > 1 &&
1453 !migrate_all)
1454 goto put_and_set;
1455
Naoya Horiguchie632a932013-09-11 14:22:04 -07001456 if (PageHuge(page)) {
Naoya Horiguchie66f17f2015-02-11 15:25:22 -08001457 if (PageHead(page))
1458 isolate_huge_page(page, &pagelist);
Naoya Horiguchie632a932013-09-11 14:22:04 -07001459 goto put_and_set;
1460 }
1461
Nick Piggin62695a82008-10-18 20:26:09 -07001462 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001463 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -07001464 list_add_tail(&page->lru, &pagelist);
Mel Gorman599d0c92016-07-28 15:45:31 -07001465 inc_node_page_state(page, NR_ISOLATED_ANON +
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001466 page_is_file_cache(page));
1467 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001468put_and_set:
1469 /*
1470 * Either remove the duplicate refcount from
1471 * isolate_lru_page() or drop the page ref if it was
1472 * not isolated.
1473 */
1474 put_page(page);
1475set_status:
1476 pp->status = err;
1477 }
1478
Brice Gogline78bbfa2008-10-18 20:27:15 -07001479 err = 0;
Minchan Kimcf608ac2010-10-26 14:21:29 -07001480 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001481 err = migrate_pages(&pagelist, new_page_node, NULL,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001482 (unsigned long)pm, MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001483 if (err)
Naoya Horiguchie632a932013-09-11 14:22:04 -07001484 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001485 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001486
1487 up_read(&mm->mmap_sem);
1488 return err;
1489}
1490
1491/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001492 * Migrate an array of page address onto an array of nodes and fill
1493 * the corresponding array of status.
1494 */
Christoph Lameter3268c632012-03-21 16:34:06 -07001495static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001496 unsigned long nr_pages,
1497 const void __user * __user *pages,
1498 const int __user *nodes,
1499 int __user *status, int flags)
1500{
Brice Goglin3140a222009-01-06 14:38:57 -08001501 struct page_to_node *pm;
Brice Goglin3140a222009-01-06 14:38:57 -08001502 unsigned long chunk_nr_pages;
1503 unsigned long chunk_start;
1504 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001505
Brice Goglin3140a222009-01-06 14:38:57 -08001506 err = -ENOMEM;
1507 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1508 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001509 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -07001510
1511 migrate_prep();
1512
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001513 /*
Brice Goglin3140a222009-01-06 14:38:57 -08001514 * Store a chunk of page_to_node array in a page,
1515 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001516 */
Brice Goglin3140a222009-01-06 14:38:57 -08001517 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001518
Brice Goglin3140a222009-01-06 14:38:57 -08001519 for (chunk_start = 0;
1520 chunk_start < nr_pages;
1521 chunk_start += chunk_nr_pages) {
1522 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001523
Brice Goglin3140a222009-01-06 14:38:57 -08001524 if (chunk_start + chunk_nr_pages > nr_pages)
1525 chunk_nr_pages = nr_pages - chunk_start;
1526
1527 /* fill the chunk pm with addrs and nodes from user-space */
1528 for (j = 0; j < chunk_nr_pages; j++) {
1529 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001530 int node;
1531
Brice Goglin3140a222009-01-06 14:38:57 -08001532 err = -EFAULT;
1533 if (get_user(p, pages + j + chunk_start))
1534 goto out_pm;
1535 pm[j].addr = (unsigned long) p;
1536
1537 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001538 goto out_pm;
1539
1540 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -08001541 if (node < 0 || node >= MAX_NUMNODES)
1542 goto out_pm;
1543
Lai Jiangshan389162c2012-12-12 13:51:30 -08001544 if (!node_state(node, N_MEMORY))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001545 goto out_pm;
1546
1547 err = -EACCES;
1548 if (!node_isset(node, task_nodes))
1549 goto out_pm;
1550
Brice Goglin3140a222009-01-06 14:38:57 -08001551 pm[j].node = node;
1552 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001553
Brice Goglin3140a222009-01-06 14:38:57 -08001554 /* End marker for this chunk */
1555 pm[chunk_nr_pages].node = MAX_NUMNODES;
1556
1557 /* Migrate this chunk */
1558 err = do_move_page_to_node_array(mm, pm,
1559 flags & MPOL_MF_MOVE_ALL);
1560 if (err < 0)
1561 goto out_pm;
1562
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001563 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -08001564 for (j = 0; j < chunk_nr_pages; j++)
1565 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001566 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -08001567 goto out_pm;
1568 }
1569 }
1570 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001571
1572out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -08001573 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001574out:
1575 return err;
1576}
1577
1578/*
Brice Goglin2f007e72008-10-18 20:27:16 -07001579 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -07001580 */
Brice Goglin80bba122008-12-09 13:14:23 -08001581static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1582 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001583{
Brice Goglin2f007e72008-10-18 20:27:16 -07001584 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001585
Christoph Lameter742755a2006-06-23 02:03:55 -07001586 down_read(&mm->mmap_sem);
1587
Brice Goglin2f007e72008-10-18 20:27:16 -07001588 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001589 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001590 struct vm_area_struct *vma;
1591 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001592 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001593
1594 vma = find_vma(mm, addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001595 if (!vma || addr < vma->vm_start)
Christoph Lameter742755a2006-06-23 02:03:55 -07001596 goto set_status;
1597
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001598 /* FOLL_DUMP to ignore special (like zero) pages */
1599 page = follow_page(vma, addr, FOLL_DUMP);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001600
1601 err = PTR_ERR(page);
1602 if (IS_ERR(page))
1603 goto set_status;
1604
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001605 err = page ? page_to_nid(page) : -ENOENT;
Christoph Lameter742755a2006-06-23 02:03:55 -07001606set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001607 *status = err;
1608
1609 pages++;
1610 status++;
1611 }
1612
1613 up_read(&mm->mmap_sem);
1614}
1615
1616/*
1617 * Determine the nodes of a user array of pages and store it in
1618 * a user array of status.
1619 */
1620static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1621 const void __user * __user *pages,
1622 int __user *status)
1623{
1624#define DO_PAGES_STAT_CHUNK_NR 16
1625 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1626 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
Brice Goglin80bba122008-12-09 13:14:23 -08001627
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001628 while (nr_pages) {
1629 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001630
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001631 chunk_nr = nr_pages;
1632 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1633 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1634
1635 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1636 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001637
1638 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1639
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001640 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1641 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001642
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001643 pages += chunk_nr;
1644 status += chunk_nr;
1645 nr_pages -= chunk_nr;
1646 }
1647 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001648}
1649
1650/*
1651 * Move a list of pages in the address space of the currently executing
1652 * process.
1653 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001654SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1655 const void __user * __user *, pages,
1656 const int __user *, nodes,
1657 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001658{
David Howellsc69e8d92008-11-14 10:39:19 +11001659 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001660 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001661 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001662 int err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001663 nodemask_t task_nodes;
Christoph Lameter742755a2006-06-23 02:03:55 -07001664
1665 /* Check flags */
1666 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1667 return -EINVAL;
1668
1669 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1670 return -EPERM;
1671
1672 /* Find the mm_struct */
Greg Thelena879bf52011-02-25 14:44:13 -08001673 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001674 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001675 if (!task) {
Greg Thelena879bf52011-02-25 14:44:13 -08001676 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001677 return -ESRCH;
1678 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001679 get_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001680
1681 /*
1682 * Check if this process has the right to modify the specified
1683 * process. The right exists if the process has administrative
1684 * capabilities, superuser privileges or the same
1685 * userid as the target process.
1686 */
David Howellsc69e8d92008-11-14 10:39:19 +11001687 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001688 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1689 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001690 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001691 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001692 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001693 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001694 }
David Howellsc69e8d92008-11-14 10:39:19 +11001695 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001696
David Quigley86c3a762006-06-23 02:04:02 -07001697 err = security_task_movememory(task);
1698 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001699 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001700
Christoph Lameter3268c632012-03-21 16:34:06 -07001701 task_nodes = cpuset_mems_allowed(task);
1702 mm = get_task_mm(task);
1703 put_task_struct(task);
1704
Sasha Levin6e8b09e2012-04-25 16:01:53 -07001705 if (!mm)
1706 return -EINVAL;
1707
1708 if (nodes)
1709 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1710 nodes, status, flags);
1711 else
1712 err = do_pages_stat(mm, nr_pages, pages, status);
Christoph Lameter3268c632012-03-21 16:34:06 -07001713
1714 mmput(mm);
1715 return err;
David Quigley86c3a762006-06-23 02:04:02 -07001716
Christoph Lameter742755a2006-06-23 02:03:55 -07001717out:
Christoph Lameter3268c632012-03-21 16:34:06 -07001718 put_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001719 return err;
1720}
Christoph Lameter742755a2006-06-23 02:03:55 -07001721
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001722#ifdef CONFIG_NUMA_BALANCING
1723/*
1724 * Returns true if this is a safe migration target node for misplaced NUMA
1725 * pages. Currently it only checks the watermarks which crude
1726 */
1727static bool migrate_balanced_pgdat(struct pglist_data *pgdat,
Mel Gorman3abef4e2013-02-22 16:34:27 -08001728 unsigned long nr_migrate_pages)
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001729{
1730 int z;
Mel Gorman599d0c92016-07-28 15:45:31 -07001731
1732 if (!pgdat_reclaimable(pgdat))
1733 return false;
1734
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001735 for (z = pgdat->nr_zones - 1; z >= 0; z--) {
1736 struct zone *zone = pgdat->node_zones + z;
1737
1738 if (!populated_zone(zone))
1739 continue;
1740
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001741 /* Avoid waking kswapd by allocating pages_to_migrate pages. */
1742 if (!zone_watermark_ok(zone, 0,
1743 high_wmark_pages(zone) +
1744 nr_migrate_pages,
1745 0, 0))
1746 continue;
1747 return true;
1748 }
1749 return false;
1750}
1751
1752static struct page *alloc_misplaced_dst_page(struct page *page,
1753 unsigned long data,
1754 int **result)
1755{
1756 int nid = (int) data;
1757 struct page *newpage;
1758
Vlastimil Babka96db8002015-09-08 15:03:50 -07001759 newpage = __alloc_pages_node(nid,
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001760 (GFP_HIGHUSER_MOVABLE |
1761 __GFP_THISNODE | __GFP_NOMEMALLOC |
1762 __GFP_NORETRY | __GFP_NOWARN) &
Mel Gorman8479eba2016-02-26 15:19:31 -08001763 ~__GFP_RECLAIM, 0);
Hillf Dantonbac03822012-11-27 14:46:24 +00001764
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001765 return newpage;
1766}
1767
1768/*
Mel Gormana8f60772012-11-14 21:41:46 +00001769 * page migration rate limiting control.
1770 * Do not migrate more than @pages_to_migrate in a @migrate_interval_millisecs
1771 * window of time. Default here says do not migrate more than 1280M per second.
1772 */
1773static unsigned int migrate_interval_millisecs __read_mostly = 100;
1774static unsigned int ratelimit_pages __read_mostly = 128 << (20 - PAGE_SHIFT);
1775
Mel Gormanb32967f2012-11-19 12:35:47 +00001776/* Returns true if the node is migrate rate-limited after the update */
Mel Gorman1c30e012014-01-21 15:50:58 -08001777static bool numamigrate_update_ratelimit(pg_data_t *pgdat,
1778 unsigned long nr_pages)
Mel Gormanb32967f2012-11-19 12:35:47 +00001779{
Mel Gormanb32967f2012-11-19 12:35:47 +00001780 /*
1781 * Rate-limit the amount of data that is being migrated to a node.
1782 * Optimal placement is no good if the memory bus is saturated and
1783 * all the time is being spent migrating!
1784 */
Mel Gormanb32967f2012-11-19 12:35:47 +00001785 if (time_after(jiffies, pgdat->numabalancing_migrate_next_window)) {
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001786 spin_lock(&pgdat->numabalancing_migrate_lock);
Mel Gormanb32967f2012-11-19 12:35:47 +00001787 pgdat->numabalancing_migrate_nr_pages = 0;
1788 pgdat->numabalancing_migrate_next_window = jiffies +
1789 msecs_to_jiffies(migrate_interval_millisecs);
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001790 spin_unlock(&pgdat->numabalancing_migrate_lock);
Mel Gormanb32967f2012-11-19 12:35:47 +00001791 }
Mel Gormanaf1839d2014-01-21 15:51:01 -08001792 if (pgdat->numabalancing_migrate_nr_pages > ratelimit_pages) {
1793 trace_mm_numa_migrate_ratelimit(current, pgdat->node_id,
1794 nr_pages);
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001795 return true;
Mel Gormanaf1839d2014-01-21 15:51:01 -08001796 }
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001797
1798 /*
1799 * This is an unlocked non-atomic update so errors are possible.
1800 * The consequences are failing to migrate when we potentiall should
1801 * have which is not severe enough to warrant locking. If it is ever
1802 * a problem, it can be converted to a per-cpu counter.
1803 */
1804 pgdat->numabalancing_migrate_nr_pages += nr_pages;
1805 return false;
Mel Gormanb32967f2012-11-19 12:35:47 +00001806}
1807
Mel Gorman1c30e012014-01-21 15:50:58 -08001808static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page)
Mel Gormanb32967f2012-11-19 12:35:47 +00001809{
Hugh Dickins340ef392013-02-22 16:34:33 -08001810 int page_lru;
Mel Gormanb32967f2012-11-19 12:35:47 +00001811
Sasha Levin309381fea2014-01-23 15:52:54 -08001812 VM_BUG_ON_PAGE(compound_order(page) && !PageTransHuge(page), page);
Mel Gorman3abef4e2013-02-22 16:34:27 -08001813
Mel Gormanb32967f2012-11-19 12:35:47 +00001814 /* Avoid migrating to a node that is nearly full */
Hugh Dickins340ef392013-02-22 16:34:33 -08001815 if (!migrate_balanced_pgdat(pgdat, 1UL << compound_order(page)))
1816 return 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001817
Hugh Dickins340ef392013-02-22 16:34:33 -08001818 if (isolate_lru_page(page))
1819 return 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001820
1821 /*
Hugh Dickins340ef392013-02-22 16:34:33 -08001822 * migrate_misplaced_transhuge_page() skips page migration's usual
1823 * check on page_count(), so we must do it here, now that the page
1824 * has been isolated: a GUP pin, or any other pin, prevents migration.
1825 * The expected page count is 3: 1 for page's mapcount and 1 for the
1826 * caller's pin and 1 for the reference taken by isolate_lru_page().
1827 */
1828 if (PageTransHuge(page) && page_count(page) != 3) {
1829 putback_lru_page(page);
1830 return 0;
1831 }
1832
1833 page_lru = page_is_file_cache(page);
Mel Gorman599d0c92016-07-28 15:45:31 -07001834 mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + page_lru,
Hugh Dickins340ef392013-02-22 16:34:33 -08001835 hpage_nr_pages(page));
1836
1837 /*
1838 * Isolating the page has taken another reference, so the
1839 * caller's reference can be safely dropped without the page
1840 * disappearing underneath us during migration.
Mel Gormanb32967f2012-11-19 12:35:47 +00001841 */
1842 put_page(page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001843 return 1;
Mel Gormanb32967f2012-11-19 12:35:47 +00001844}
1845
Mel Gormande466bd2013-12-18 17:08:42 -08001846bool pmd_trans_migrating(pmd_t pmd)
1847{
1848 struct page *page = pmd_page(pmd);
1849 return PageLocked(page);
1850}
1851
Mel Gormana8f60772012-11-14 21:41:46 +00001852/*
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001853 * Attempt to migrate a misplaced page to the specified destination
1854 * node. Caller is expected to have an elevated reference count on
1855 * the page that will be dropped by this function before returning.
1856 */
Mel Gorman1bc115d2013-10-07 11:29:05 +01001857int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma,
1858 int node)
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001859{
Mel Gormana8f60772012-11-14 21:41:46 +00001860 pg_data_t *pgdat = NODE_DATA(node);
Hugh Dickins340ef392013-02-22 16:34:33 -08001861 int isolated;
Mel Gormanb32967f2012-11-19 12:35:47 +00001862 int nr_remaining;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001863 LIST_HEAD(migratepages);
1864
1865 /*
Mel Gorman1bc115d2013-10-07 11:29:05 +01001866 * Don't migrate file pages that are mapped in multiple processes
1867 * with execute permissions as they are probably shared libraries.
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001868 */
Mel Gorman1bc115d2013-10-07 11:29:05 +01001869 if (page_mapcount(page) != 1 && page_is_file_cache(page) &&
1870 (vma->vm_flags & VM_EXEC))
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001871 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001872
Mel Gormana8f60772012-11-14 21:41:46 +00001873 /*
1874 * Rate-limit the amount of data that is being migrated to a node.
1875 * Optimal placement is no good if the memory bus is saturated and
1876 * all the time is being spent migrating!
1877 */
Hugh Dickins340ef392013-02-22 16:34:33 -08001878 if (numamigrate_update_ratelimit(pgdat, 1))
Mel Gormana8f60772012-11-14 21:41:46 +00001879 goto out;
Mel Gormana8f60772012-11-14 21:41:46 +00001880
Mel Gormanb32967f2012-11-19 12:35:47 +00001881 isolated = numamigrate_isolate_page(pgdat, page);
1882 if (!isolated)
1883 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001884
Mel Gormanb32967f2012-11-19 12:35:47 +00001885 list_add(&page->lru, &migratepages);
Hugh Dickins9c620e22013-02-22 16:35:14 -08001886 nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page,
David Rientjes68711a72014-06-04 16:08:25 -07001887 NULL, node, MIGRATE_ASYNC,
1888 MR_NUMA_MISPLACED);
Mel Gormanb32967f2012-11-19 12:35:47 +00001889 if (nr_remaining) {
Joonsoo Kim59c82b72014-01-21 15:51:17 -08001890 if (!list_empty(&migratepages)) {
1891 list_del(&page->lru);
Mel Gorman599d0c92016-07-28 15:45:31 -07001892 dec_node_page_state(page, NR_ISOLATED_ANON +
Joonsoo Kim59c82b72014-01-21 15:51:17 -08001893 page_is_file_cache(page));
1894 putback_lru_page(page);
1895 }
Mel Gormanb32967f2012-11-19 12:35:47 +00001896 isolated = 0;
1897 } else
1898 count_vm_numa_event(NUMA_PAGE_MIGRATE);
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001899 BUG_ON(!list_empty(&migratepages));
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001900 return isolated;
Hugh Dickins340ef392013-02-22 16:34:33 -08001901
1902out:
1903 put_page(page);
1904 return 0;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001905}
Mel Gorman220018d2012-12-05 09:32:56 +00001906#endif /* CONFIG_NUMA_BALANCING */
Mel Gormanb32967f2012-11-19 12:35:47 +00001907
Mel Gorman220018d2012-12-05 09:32:56 +00001908#if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
Hugh Dickins340ef392013-02-22 16:34:33 -08001909/*
1910 * Migrates a THP to a given target node. page must be locked and is unlocked
1911 * before returning.
1912 */
Mel Gormanb32967f2012-11-19 12:35:47 +00001913int migrate_misplaced_transhuge_page(struct mm_struct *mm,
1914 struct vm_area_struct *vma,
1915 pmd_t *pmd, pmd_t entry,
1916 unsigned long address,
1917 struct page *page, int node)
1918{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001919 spinlock_t *ptl;
Mel Gormanb32967f2012-11-19 12:35:47 +00001920 pg_data_t *pgdat = NODE_DATA(node);
1921 int isolated = 0;
1922 struct page *new_page = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00001923 int page_lru = page_is_file_cache(page);
Mel Gormanf714f4f2013-12-18 17:08:33 -08001924 unsigned long mmun_start = address & HPAGE_PMD_MASK;
1925 unsigned long mmun_end = mmun_start + HPAGE_PMD_SIZE;
Mel Gorman2b4847e2013-12-18 17:08:32 -08001926 pmd_t orig_entry;
Mel Gormanb32967f2012-11-19 12:35:47 +00001927
1928 /*
Mel Gormanb32967f2012-11-19 12:35:47 +00001929 * Rate-limit the amount of data that is being migrated to a node.
1930 * Optimal placement is no good if the memory bus is saturated and
1931 * all the time is being spent migrating!
1932 */
Mel Gormand28d43352012-11-29 09:24:36 +00001933 if (numamigrate_update_ratelimit(pgdat, HPAGE_PMD_NR))
Mel Gormanb32967f2012-11-19 12:35:47 +00001934 goto out_dropref;
1935
1936 new_page = alloc_pages_node(node,
Vlastimil Babka25160352016-07-28 15:49:25 -07001937 (GFP_TRANSHUGE_LIGHT | __GFP_THISNODE),
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001938 HPAGE_PMD_ORDER);
Hugh Dickins340ef392013-02-22 16:34:33 -08001939 if (!new_page)
1940 goto out_fail;
Kirill A. Shutemov9a982252016-01-15 16:54:17 -08001941 prep_transhuge_page(new_page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001942
Mel Gormanb32967f2012-11-19 12:35:47 +00001943 isolated = numamigrate_isolate_page(pgdat, page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001944 if (!isolated) {
Mel Gormanb32967f2012-11-19 12:35:47 +00001945 put_page(new_page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001946 goto out_fail;
Mel Gormanb32967f2012-11-19 12:35:47 +00001947 }
Aneesh Kumar K.V458aa762016-03-17 14:18:56 -07001948 /*
1949 * We are not sure a pending tlb flush here is for a huge page
1950 * mapping or not. Hence use the tlb range variant
1951 */
Mel Gormanb0943d62013-12-18 17:08:46 -08001952 if (mm_tlb_flush_pending(mm))
1953 flush_tlb_range(vma, mmun_start, mmun_end);
1954
Mel Gormanb32967f2012-11-19 12:35:47 +00001955 /* Prepare a page as a migration target */
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001956 __SetPageLocked(new_page);
Hugh Dickinsfa9949d2016-05-19 17:12:41 -07001957 __SetPageSwapBacked(new_page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001958
1959 /* anon mapping, we can simply copy page->mapping to the new page: */
1960 new_page->mapping = page->mapping;
1961 new_page->index = page->index;
1962 migrate_page_copy(new_page, page);
1963 WARN_ON(PageLRU(new_page));
1964
1965 /* Recheck the target PMD */
Mel Gormanf714f4f2013-12-18 17:08:33 -08001966 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001967 ptl = pmd_lock(mm, pmd);
Mel Gorman2b4847e2013-12-18 17:08:32 -08001968 if (unlikely(!pmd_same(*pmd, entry) || page_count(page) != 2)) {
1969fail_putback:
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001970 spin_unlock(ptl);
Mel Gormanf714f4f2013-12-18 17:08:33 -08001971 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Mel Gormanb32967f2012-11-19 12:35:47 +00001972
1973 /* Reverse changes made by migrate_page_copy() */
1974 if (TestClearPageActive(new_page))
1975 SetPageActive(page);
1976 if (TestClearPageUnevictable(new_page))
1977 SetPageUnevictable(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001978
1979 unlock_page(new_page);
1980 put_page(new_page); /* Free it */
1981
Mel Gormana54a4072013-10-07 11:28:46 +01001982 /* Retake the callers reference and putback on LRU */
1983 get_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001984 putback_lru_page(page);
Mel Gorman599d0c92016-07-28 15:45:31 -07001985 mod_node_page_state(page_pgdat(page),
Mel Gormana54a4072013-10-07 11:28:46 +01001986 NR_ISOLATED_ANON + page_lru, -HPAGE_PMD_NR);
Mel Gormaneb4489f62013-12-18 17:08:39 -08001987
1988 goto out_unlock;
Mel Gormanb32967f2012-11-19 12:35:47 +00001989 }
1990
Mel Gorman2b4847e2013-12-18 17:08:32 -08001991 orig_entry = *pmd;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07001992 entry = mk_huge_pmd(new_page, vma->vm_page_prot);
Mel Gorman2b4847e2013-12-18 17:08:32 -08001993 entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
Mel Gormanb32967f2012-11-19 12:35:47 +00001994
Mel Gorman2b4847e2013-12-18 17:08:32 -08001995 /*
1996 * Clear the old entry under pagetable lock and establish the new PTE.
1997 * Any parallel GUP will either observe the old page blocking on the
1998 * page lock, block on the page table lock or observe the new page.
1999 * The SetPageUptodate on the new page and page_add_new_anon_rmap
2000 * guarantee the copy is visible before the pagetable update.
2001 */
Mel Gormanf714f4f2013-12-18 17:08:33 -08002002 flush_cache_range(vma, mmun_start, mmun_end);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002003 page_add_anon_rmap(new_page, vma, mmun_start, true);
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -07002004 pmdp_huge_clear_flush_notify(vma, mmun_start, pmd);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002005 set_pmd_at(mm, mmun_start, pmd, entry);
Stephen Rothwellce4a9cc2012-12-10 19:50:57 +11002006 update_mmu_cache_pmd(vma, address, &entry);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002007
2008 if (page_count(page) != 2) {
Mel Gormanf714f4f2013-12-18 17:08:33 -08002009 set_pmd_at(mm, mmun_start, pmd, orig_entry);
Aneesh Kumar K.V458aa762016-03-17 14:18:56 -07002010 flush_pmd_tlb_range(vma, mmun_start, mmun_end);
Joerg Roedel34ee6452014-11-13 13:46:09 +11002011 mmu_notifier_invalidate_range(mm, mmun_start, mmun_end);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002012 update_mmu_cache_pmd(vma, address, &entry);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002013 page_remove_rmap(new_page, true);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002014 goto fail_putback;
2015 }
2016
Hugh Dickins51afb122015-11-05 18:49:37 -08002017 mlock_migrate_page(new_page, page);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002018 page_remove_rmap(page, true);
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07002019 set_page_owner_migrate_reason(new_page, MR_NUMA_MISPLACED);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002020
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08002021 spin_unlock(ptl);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002022 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Mel Gormanb32967f2012-11-19 12:35:47 +00002023
Mel Gorman11de9922014-06-04 16:07:41 -07002024 /* Take an "isolate" reference and put new page on the LRU. */
2025 get_page(new_page);
2026 putback_lru_page(new_page);
2027
Mel Gormanb32967f2012-11-19 12:35:47 +00002028 unlock_page(new_page);
2029 unlock_page(page);
2030 put_page(page); /* Drop the rmap reference */
2031 put_page(page); /* Drop the LRU isolation reference */
2032
2033 count_vm_events(PGMIGRATE_SUCCESS, HPAGE_PMD_NR);
2034 count_vm_numa_events(NUMA_PAGE_MIGRATE, HPAGE_PMD_NR);
2035
Mel Gorman599d0c92016-07-28 15:45:31 -07002036 mod_node_page_state(page_pgdat(page),
Mel Gormanb32967f2012-11-19 12:35:47 +00002037 NR_ISOLATED_ANON + page_lru,
2038 -HPAGE_PMD_NR);
2039 return isolated;
2040
Hugh Dickins340ef392013-02-22 16:34:33 -08002041out_fail:
2042 count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR);
Mel Gormanb32967f2012-11-19 12:35:47 +00002043out_dropref:
Mel Gorman2b4847e2013-12-18 17:08:32 -08002044 ptl = pmd_lock(mm, pmd);
2045 if (pmd_same(*pmd, entry)) {
Mel Gorman4d942462015-02-12 14:58:28 -08002046 entry = pmd_modify(entry, vma->vm_page_prot);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002047 set_pmd_at(mm, mmun_start, pmd, entry);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002048 update_mmu_cache_pmd(vma, address, &entry);
2049 }
2050 spin_unlock(ptl);
Mel Gormana54a4072013-10-07 11:28:46 +01002051
Mel Gormaneb4489f62013-12-18 17:08:39 -08002052out_unlock:
Hugh Dickins340ef392013-02-22 16:34:33 -08002053 unlock_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00002054 put_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00002055 return 0;
2056}
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02002057#endif /* CONFIG_NUMA_BALANCING */
2058
2059#endif /* CONFIG_NUMA */