blob: 87f4d0f818194ea68cd323b1bcf84dd430b826a3 [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);
Minchan Kimbda807d2016-07-26 15:23:05 -0700171 /*
172 * We isolated non-lru movable page so here we can use
173 * __PageMovable because LRU page's mapping cannot have
174 * PAGE_MAPPING_MOVABLE.
175 */
Minchan Kimb1123ea62016-07-26 15:23:09 -0700176 if (unlikely(__PageMovable(page))) {
Minchan Kimbda807d2016-07-26 15:23:05 -0700177 VM_BUG_ON_PAGE(!PageIsolated(page), page);
178 lock_page(page);
179 if (PageMovable(page))
180 putback_movable_page(page);
181 else
182 __ClearPageIsolated(page);
183 unlock_page(page);
184 put_page(page);
185 } else {
Rafael Aquinibf6bddf2012-12-11 16:02:42 -0800186 putback_lru_page(page);
Ming Ling6afcf8e2016-12-12 16:42:26 -0800187 dec_node_page_state(page, NR_ISOLATED_ANON +
188 page_is_file_cache(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);
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700237 pte = pte_mkold(mk_pte(new, READ_ONCE(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;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800469 get_page(newpage); /* add cache reference */
Nicholas Piggin6326fec2016-12-25 13:00:29 +1000470 if (PageSwapBacked(page)) {
471 __SetPageSwapBacked(newpage);
472 if (PageSwapCache(page)) {
473 SetPageSwapCache(newpage);
474 set_page_private(newpage, page_private(page));
475 }
476 } else {
477 VM_BUG_ON_PAGE(PageSwapCache(page), page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800478 }
479
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800480 /* Move dirty while page refs frozen and newpage not yet exposed */
481 dirty = PageDirty(page);
482 if (dirty) {
483 ClearPageDirty(page);
484 SetPageDirty(newpage);
485 }
486
Johannes Weiner6d75f362016-12-12 16:43:43 -0800487 radix_tree_replace_slot(&mapping->page_tree, pslot, newpage);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800488
489 /*
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800490 * Drop cache reference from old page by unfreezing
491 * to one less reference.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800492 * We know this isn't the last reference.
493 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700494 page_ref_unfreeze(page, expected_count - 1);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800495
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800496 spin_unlock(&mapping->tree_lock);
497 /* Leave irq disabled to prevent preemption while updating stats */
498
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700499 /*
500 * If moved to a different zone then also account
501 * the page for that zone. Other VM counters will be
502 * taken care of when we establish references to the
503 * new page and drop references to the old page.
504 *
505 * Note that anonymous pages are accounted for
Mel Gorman4b9d0fa2016-07-28 15:46:17 -0700506 * via NR_FILE_PAGES and NR_ANON_MAPPED if they
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700507 * are mapped to swap space.
508 */
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800509 if (newzone != oldzone) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700510 __dec_node_state(oldzone->zone_pgdat, NR_FILE_PAGES);
511 __inc_node_state(newzone->zone_pgdat, NR_FILE_PAGES);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800512 if (PageSwapBacked(page) && !PageSwapCache(page)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700513 __dec_node_state(oldzone->zone_pgdat, NR_SHMEM);
514 __inc_node_state(newzone->zone_pgdat, NR_SHMEM);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800515 }
516 if (dirty && mapping_cap_account_dirty(mapping)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700517 __dec_node_state(oldzone->zone_pgdat, NR_FILE_DIRTY);
Mel Gorman5a1c84b2016-07-28 15:47:31 -0700518 __dec_zone_state(oldzone, NR_ZONE_WRITE_PENDING);
Mel Gorman11fb9982016-07-28 15:46:20 -0700519 __inc_node_state(newzone->zone_pgdat, NR_FILE_DIRTY);
Mel Gorman5a1c84b2016-07-28 15:47:31 -0700520 __inc_zone_state(newzone, NR_ZONE_WRITE_PENDING);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800521 }
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700522 }
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800523 local_irq_enable();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800524
Rafael Aquini78bd5202012-12-11 16:02:31 -0800525 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800526}
Richard Weinberger1118dce2016-06-16 23:26:14 +0200527EXPORT_SYMBOL(migrate_page_move_mapping);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800528
529/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900530 * The expected number of remaining references is the same as that
531 * of migrate_page_move_mapping().
532 */
533int migrate_huge_page_move_mapping(struct address_space *mapping,
534 struct page *newpage, struct page *page)
535{
536 int expected_count;
537 void **pslot;
538
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900539 spin_lock_irq(&mapping->tree_lock);
540
541 pslot = radix_tree_lookup_slot(&mapping->page_tree,
542 page_index(page));
543
544 expected_count = 2 + page_has_private(page);
545 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800546 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900547 spin_unlock_irq(&mapping->tree_lock);
548 return -EAGAIN;
549 }
550
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700551 if (!page_ref_freeze(page, expected_count)) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900552 spin_unlock_irq(&mapping->tree_lock);
553 return -EAGAIN;
554 }
555
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800556 newpage->index = page->index;
557 newpage->mapping = page->mapping;
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700558
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900559 get_page(newpage);
560
Johannes Weiner6d75f362016-12-12 16:43:43 -0800561 radix_tree_replace_slot(&mapping->page_tree, pslot, newpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900562
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700563 page_ref_unfreeze(page, expected_count - 1);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900564
565 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700566
Rafael Aquini78bd5202012-12-11 16:02:31 -0800567 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900568}
569
570/*
Dave Hansen30b0a102013-11-21 14:31:58 -0800571 * Gigantic pages are so large that we do not guarantee that page++ pointer
572 * arithmetic will work across the entire page. We need something more
573 * specialized.
574 */
575static void __copy_gigantic_page(struct page *dst, struct page *src,
576 int nr_pages)
577{
578 int i;
579 struct page *dst_base = dst;
580 struct page *src_base = src;
581
582 for (i = 0; i < nr_pages; ) {
583 cond_resched();
584 copy_highpage(dst, src);
585
586 i++;
587 dst = mem_map_next(dst, dst_base, i);
588 src = mem_map_next(src, src_base, i);
589 }
590}
591
592static void copy_huge_page(struct page *dst, struct page *src)
593{
594 int i;
595 int nr_pages;
596
597 if (PageHuge(src)) {
598 /* hugetlbfs page */
599 struct hstate *h = page_hstate(src);
600 nr_pages = pages_per_huge_page(h);
601
602 if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) {
603 __copy_gigantic_page(dst, src, nr_pages);
604 return;
605 }
606 } else {
607 /* thp page */
608 BUG_ON(!PageTransHuge(src));
609 nr_pages = hpage_nr_pages(src);
610 }
611
612 for (i = 0; i < nr_pages; i++) {
613 cond_resched();
614 copy_highpage(dst + i, src + i);
615 }
616}
617
618/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800619 * Copy the page to its new location
620 */
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900621void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800622{
Rik van Riel7851a452013-10-07 11:29:23 +0100623 int cpupid;
624
Mel Gormanb32967f2012-11-19 12:35:47 +0000625 if (PageHuge(page) || PageTransHuge(page))
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900626 copy_huge_page(newpage, page);
627 else
628 copy_highpage(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800629
630 if (PageError(page))
631 SetPageError(newpage);
632 if (PageReferenced(page))
633 SetPageReferenced(newpage);
634 if (PageUptodate(page))
635 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700636 if (TestClearPageActive(page)) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800637 VM_BUG_ON_PAGE(PageUnevictable(page), page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800638 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800639 } else if (TestClearPageUnevictable(page))
640 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800641 if (PageChecked(page))
642 SetPageChecked(newpage);
643 if (PageMappedToDisk(page))
644 SetPageMappedToDisk(newpage);
645
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800646 /* Move dirty on pages not done by migrate_page_move_mapping() */
647 if (PageDirty(page))
648 SetPageDirty(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800649
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700650 if (page_is_young(page))
651 set_page_young(newpage);
652 if (page_is_idle(page))
653 set_page_idle(newpage);
654
Rik van Riel7851a452013-10-07 11:29:23 +0100655 /*
656 * Copy NUMA information to the new page, to prevent over-eager
657 * future migrations of this same page.
658 */
659 cpupid = page_cpupid_xchg_last(page, -1);
660 page_cpupid_xchg_last(newpage, cpupid);
661
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800662 ksm_migrate_page(newpage, page);
Hugh Dickinsc8d65532013-02-22 16:35:10 -0800663 /*
664 * Please do not reorder this without considering how mm/ksm.c's
665 * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache().
666 */
Naoya Horiguchib3b3a992015-04-15 16:13:15 -0700667 if (PageSwapCache(page))
668 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800669 ClearPagePrivate(page);
670 set_page_private(page, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800671
672 /*
673 * If any waiters have accumulated on the new page then
674 * wake them up.
675 */
676 if (PageWriteback(newpage))
677 end_page_writeback(newpage);
Vlastimil Babkad435edc2016-03-15 14:56:15 -0700678
679 copy_page_owner(page, newpage);
Johannes Weiner74485cf2016-03-15 14:57:54 -0700680
681 mem_cgroup_migrate(page, newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800682}
Richard Weinberger1118dce2016-06-16 23:26:14 +0200683EXPORT_SYMBOL(migrate_page_copy);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800684
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700685/************************************************************
686 * Migration functions
687 ***********************************************************/
688
Christoph Lameterb20a3502006-03-22 00:09:12 -0800689/*
Minchan Kimbda807d2016-07-26 15:23:05 -0700690 * Common logic to directly migrate a single LRU page suitable for
David Howells266cf652009-04-03 16:42:36 +0100691 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800692 *
693 * Pages are locked upon entry and exit.
694 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700695int migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800696 struct page *newpage, struct page *page,
697 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800698{
699 int rc;
700
701 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
702
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500703 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800704
Rafael Aquini78bd5202012-12-11 16:02:31 -0800705 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800706 return rc;
707
708 migrate_page_copy(newpage, page);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800709 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800710}
711EXPORT_SYMBOL(migrate_page);
712
David Howells93614012006-09-30 20:45:40 +0200713#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800714/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700715 * Migration function for pages with buffers. This function can only be used
716 * if the underlying filesystem guarantees that no other references to "page"
717 * exist.
718 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700719int buffer_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800720 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700721{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700722 struct buffer_head *bh, *head;
723 int rc;
724
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700725 if (!page_has_buffers(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800726 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700727
728 head = page_buffers(page);
729
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500730 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode, 0);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700731
Rafael Aquini78bd5202012-12-11 16:02:31 -0800732 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700733 return rc;
734
Mel Gormanb969c4a2012-01-12 17:19:34 -0800735 /*
736 * In the async case, migrate_page_move_mapping locked the buffers
737 * with an IRQ-safe spinlock held. In the sync case, the buffers
738 * need to be locked now
739 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800740 if (mode != MIGRATE_ASYNC)
741 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700742
743 ClearPagePrivate(page);
744 set_page_private(newpage, page_private(page));
745 set_page_private(page, 0);
746 put_page(page);
747 get_page(newpage);
748
749 bh = head;
750 do {
751 set_bh_page(bh, newpage, bh_offset(bh));
752 bh = bh->b_this_page;
753
754 } while (bh != head);
755
756 SetPagePrivate(newpage);
757
758 migrate_page_copy(newpage, page);
759
760 bh = head;
761 do {
762 unlock_buffer(bh);
763 put_bh(bh);
764 bh = bh->b_this_page;
765
766 } while (bh != head);
767
Rafael Aquini78bd5202012-12-11 16:02:31 -0800768 return MIGRATEPAGE_SUCCESS;
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700769}
770EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200771#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700772
Christoph Lameter04e62a22006-06-23 02:03:38 -0700773/*
774 * Writeback a page to clean the dirty state
775 */
776static int writeout(struct address_space *mapping, struct page *page)
777{
778 struct writeback_control wbc = {
779 .sync_mode = WB_SYNC_NONE,
780 .nr_to_write = 1,
781 .range_start = 0,
782 .range_end = LLONG_MAX,
Christoph Lameter04e62a22006-06-23 02:03:38 -0700783 .for_reclaim = 1
784 };
785 int rc;
786
787 if (!mapping->a_ops->writepage)
788 /* No write method for the address space */
789 return -EINVAL;
790
791 if (!clear_page_dirty_for_io(page))
792 /* Someone else already triggered a write */
793 return -EAGAIN;
794
795 /*
796 * A dirty page may imply that the underlying filesystem has
797 * the page on some queue. So the page must be clean for
798 * migration. Writeout may mean we loose the lock and the
799 * page state is no longer what we checked for earlier.
800 * At this point we know that the migration attempt cannot
801 * be successful.
802 */
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700803 remove_migration_ptes(page, page, false);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700804
805 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700806
807 if (rc != AOP_WRITEPAGE_ACTIVATE)
808 /* unlocked. Relock */
809 lock_page(page);
810
Hugh Dickinsbda85502008-11-19 15:36:36 -0800811 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700812}
813
814/*
815 * Default handling if a filesystem does not provide a migration function.
816 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700817static int fallback_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800818 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700819{
Mel Gormanb969c4a2012-01-12 17:19:34 -0800820 if (PageDirty(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800821 /* Only writeback pages in full synchronous migration */
822 if (mode != MIGRATE_SYNC)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800823 return -EBUSY;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700824 return writeout(mapping, page);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800825 }
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700826
827 /*
828 * Buffers may be managed in a filesystem specific way.
829 * We must have no buffers or drop them.
830 */
David Howells266cf652009-04-03 16:42:36 +0100831 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700832 !try_to_release_page(page, GFP_KERNEL))
833 return -EAGAIN;
834
Mel Gormana6bc32b2012-01-12 17:19:43 -0800835 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700836}
837
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700838/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700839 * Move a page to a newly allocated page
840 * The page is locked and all ptes have been successfully removed.
841 *
842 * The new page will have replaced the old page if this function
843 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700844 *
845 * Return value:
846 * < 0 - error code
Rafael Aquini78bd5202012-12-11 16:02:31 -0800847 * MIGRATEPAGE_SUCCESS - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700848 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700849static int move_to_new_page(struct page *newpage, struct page *page,
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800850 enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700851{
852 struct address_space *mapping;
Minchan Kimbda807d2016-07-26 15:23:05 -0700853 int rc = -EAGAIN;
854 bool is_lru = !__PageMovable(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700855
Hugh Dickins7db76712015-11-05 18:49:49 -0800856 VM_BUG_ON_PAGE(!PageLocked(page), page);
857 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700858
Christoph Lametere24f0b82006-06-23 02:03:51 -0700859 mapping = page_mapping(page);
Minchan Kimbda807d2016-07-26 15:23:05 -0700860
861 if (likely(is_lru)) {
862 if (!mapping)
863 rc = migrate_page(mapping, newpage, page, mode);
864 else if (mapping->a_ops->migratepage)
865 /*
866 * Most pages have a mapping and most filesystems
867 * provide a migratepage callback. Anonymous pages
868 * are part of swap space which also has its own
869 * migratepage callback. This is the most common path
870 * for page migration.
871 */
872 rc = mapping->a_ops->migratepage(mapping, newpage,
873 page, mode);
874 else
875 rc = fallback_migrate_page(mapping, newpage,
876 page, mode);
877 } else {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700878 /*
Minchan Kimbda807d2016-07-26 15:23:05 -0700879 * In case of non-lru page, it could be released after
880 * isolation step. In that case, we shouldn't try migration.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700881 */
Minchan Kimbda807d2016-07-26 15:23:05 -0700882 VM_BUG_ON_PAGE(!PageIsolated(page), page);
883 if (!PageMovable(page)) {
884 rc = MIGRATEPAGE_SUCCESS;
885 __ClearPageIsolated(page);
886 goto out;
887 }
888
889 rc = mapping->a_ops->migratepage(mapping, newpage,
890 page, mode);
891 WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS &&
892 !PageIsolated(page));
893 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700894
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800895 /*
896 * When successful, old pagecache page->mapping must be cleared before
897 * page is freed; but stats require that PageAnon be left as PageAnon.
898 */
899 if (rc == MIGRATEPAGE_SUCCESS) {
Minchan Kimbda807d2016-07-26 15:23:05 -0700900 if (__PageMovable(page)) {
901 VM_BUG_ON_PAGE(!PageIsolated(page), page);
902
903 /*
904 * We clear PG_movable under page_lock so any compactor
905 * cannot try to migrate this page.
906 */
907 __ClearPageIsolated(page);
908 }
909
910 /*
911 * Anonymous and movable page->mapping will be cleard by
912 * free_pages_prepare so don't reset it here for keeping
913 * the type to work PageAnon, for example.
914 */
915 if (!PageMappingFlags(page))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800916 page->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700917 }
Minchan Kimbda807d2016-07-26 15:23:05 -0700918out:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700919 return rc;
920}
921
Minchan Kim0dabec92011-10-31 17:06:57 -0700922static int __unmap_and_move(struct page *page, struct page *newpage,
Hugh Dickins9c620e22013-02-22 16:35:14 -0800923 int force, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700924{
Minchan Kim0dabec92011-10-31 17:06:57 -0700925 int rc = -EAGAIN;
Hugh Dickins2ebba6b2014-12-12 16:56:19 -0800926 int page_was_mapped = 0;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700927 struct anon_vma *anon_vma = NULL;
Minchan Kimbda807d2016-07-26 15:23:05 -0700928 bool is_lru = !__PageMovable(page);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700929
Nick Piggin529ae9a2008-08-02 12:01:03 +0200930 if (!trylock_page(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800931 if (!force || mode == MIGRATE_ASYNC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700932 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800933
934 /*
935 * It's not safe for direct compaction to call lock_page.
936 * For example, during page readahead pages are added locked
937 * to the LRU. Later, when the IO completes the pages are
938 * marked uptodate and unlocked. However, the queueing
939 * could be merging multiple pages for one bio (e.g.
940 * mpage_readpages). If an allocation happens for the
941 * second or third page, the process can end up locking
942 * the same page twice and deadlocking. Rather than
943 * trying to be clever about what pages can be locked,
944 * avoid the use of lock_page for direct compaction
945 * altogether.
946 */
947 if (current->flags & PF_MEMALLOC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700948 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800949
Christoph Lametere24f0b82006-06-23 02:03:51 -0700950 lock_page(page);
951 }
952
953 if (PageWriteback(page)) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700954 /*
Jianguo Wufed5b642013-04-29 15:07:58 -0700955 * Only in the case of a full synchronous migration is it
Mel Gormana6bc32b2012-01-12 17:19:43 -0800956 * necessary to wait for PageWriteback. In the async case,
957 * the retry loop is too short and in the sync-light case,
958 * the overhead of stalling is too much
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700959 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800960 if (mode != MIGRATE_SYNC) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700961 rc = -EBUSY;
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700962 goto out_unlock;
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700963 }
964 if (!force)
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700965 goto out_unlock;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700966 wait_on_page_writeback(page);
967 }
Hugh Dickins03f15c82015-11-05 18:49:56 -0800968
Christoph Lametere24f0b82006-06-23 02:03:51 -0700969 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700970 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
971 * we cannot notice that anon_vma is freed while we migrates a page.
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800972 * This get_anon_vma() delays freeing anon_vma pointer until the end
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700973 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700974 * File Caches may use write_page() or lock_page() in migration, then,
975 * just care Anon page here.
Hugh Dickins03f15c82015-11-05 18:49:56 -0800976 *
977 * Only page_get_anon_vma() understands the subtleties of
978 * getting a hold on an anon_vma from outside one of its mms.
979 * But if we cannot get anon_vma, then we won't need it anyway,
980 * because that implies that the anon page is no longer mapped
981 * (and cannot be remapped so long as we hold the page lock).
Christoph Lametere24f0b82006-06-23 02:03:51 -0700982 */
Hugh Dickins03f15c82015-11-05 18:49:56 -0800983 if (PageAnon(page) && !PageKsm(page))
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700984 anon_vma = page_get_anon_vma(page);
Shaohua Li62e1c552008-02-04 22:29:33 -0800985
Hugh Dickins7db76712015-11-05 18:49:49 -0800986 /*
987 * Block others from accessing the new page when we get around to
988 * establishing additional references. We are usually the only one
989 * holding a reference to newpage at this point. We used to have a BUG
990 * here if trylock_page(newpage) fails, but would like to allow for
991 * cases where there might be a race with the previous use of newpage.
992 * This is much like races on refcount of oldpage: just don't BUG().
993 */
994 if (unlikely(!trylock_page(newpage)))
995 goto out_unlock;
996
Minchan Kimbda807d2016-07-26 15:23:05 -0700997 if (unlikely(!is_lru)) {
998 rc = move_to_new_page(newpage, page, mode);
999 goto out_unlock_both;
1000 }
1001
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001002 /*
Shaohua Li62e1c552008-02-04 22:29:33 -08001003 * Corner case handling:
1004 * 1. When a new swap-cache page is read into, it is added to the LRU
1005 * and treated as swapcache but it has no rmap yet.
1006 * Calling try_to_unmap() against a page->mapping==NULL page will
1007 * trigger a BUG. So handle it here.
1008 * 2. An orphaned page (see truncate_complete_page) might have
1009 * fs-private metadata. The page can be picked up due to memory
1010 * offlining. Everywhere else except page reclaim, the page is
1011 * invisible to the vm, so the page can not be migrated. So try to
1012 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001013 */
Shaohua Li62e1c552008-02-04 22:29:33 -08001014 if (!page->mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -08001015 VM_BUG_ON_PAGE(PageAnon(page), page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -08001016 if (page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -08001017 try_to_free_buffers(page);
Hugh Dickins7db76712015-11-05 18:49:49 -08001018 goto out_unlock_both;
Shaohua Li62e1c552008-02-04 22:29:33 -08001019 }
Hugh Dickins7db76712015-11-05 18:49:49 -08001020 } else if (page_mapped(page)) {
1021 /* Establish migration ptes */
Hugh Dickins03f15c82015-11-05 18:49:56 -08001022 VM_BUG_ON_PAGE(PageAnon(page) && !PageKsm(page) && !anon_vma,
1023 page);
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001024 try_to_unmap(page,
Wanpeng Lida1b13c2015-09-08 15:03:27 -07001025 TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001026 page_was_mapped = 1;
1027 }
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001028
Christoph Lametere6a15302006-06-25 05:46:49 -07001029 if (!page_mapped(page))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001030 rc = move_to_new_page(newpage, page, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -07001031
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001032 if (page_was_mapped)
1033 remove_migration_ptes(page,
Kirill A. Shutemove3884662016-03-17 14:20:07 -07001034 rc == MIGRATEPAGE_SUCCESS ? newpage : page, false);
Mel Gorman3f6c8272010-05-24 14:32:17 -07001035
Hugh Dickins7db76712015-11-05 18:49:49 -08001036out_unlock_both:
1037 unlock_page(newpage);
1038out_unlock:
Mel Gorman3f6c8272010-05-24 14:32:17 -07001039 /* Drop an anon_vma reference if we took one */
Rik van Riel76545062010-08-09 17:18:41 -07001040 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -07001041 put_anon_vma(anon_vma);
Christoph Lametere24f0b82006-06-23 02:03:51 -07001042 unlock_page(page);
Minchan Kim0dabec92011-10-31 17:06:57 -07001043out:
Minchan Kimc6c919e2016-07-26 15:23:02 -07001044 /*
1045 * If migration is successful, decrease refcount of the newpage
1046 * which will not free the page because new page owner increased
1047 * refcounter. As well, if it is LRU page, add the page to LRU
1048 * list in here.
1049 */
1050 if (rc == MIGRATEPAGE_SUCCESS) {
Minchan Kimb1123ea62016-07-26 15:23:09 -07001051 if (unlikely(__PageMovable(newpage)))
Minchan Kimc6c919e2016-07-26 15:23:02 -07001052 put_page(newpage);
1053 else
1054 putback_lru_page(newpage);
1055 }
1056
Minchan Kim0dabec92011-10-31 17:06:57 -07001057 return rc;
1058}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001059
Minchan Kim0dabec92011-10-31 17:06:57 -07001060/*
Geert Uytterhoevenef2a5152015-04-14 15:44:22 -07001061 * gcc 4.7 and 4.8 on arm get an ICEs when inlining unmap_and_move(). Work
1062 * around it.
1063 */
1064#if (GCC_VERSION >= 40700 && GCC_VERSION < 40900) && defined(CONFIG_ARM)
1065#define ICE_noinline noinline
1066#else
1067#define ICE_noinline
1068#endif
1069
1070/*
Minchan Kim0dabec92011-10-31 17:06:57 -07001071 * Obtain the lock on page, remove all ptes and migrate the page
1072 * to the newly allocated page in newpage.
1073 */
Geert Uytterhoevenef2a5152015-04-14 15:44:22 -07001074static ICE_noinline int unmap_and_move(new_page_t get_new_page,
1075 free_page_t put_new_page,
1076 unsigned long private, struct page *page,
Naoya Horiguchiadd05ce2015-06-24 16:56:50 -07001077 int force, enum migrate_mode mode,
1078 enum migrate_reason reason)
Minchan Kim0dabec92011-10-31 17:06:57 -07001079{
Hugh Dickins2def7422015-11-05 18:49:46 -08001080 int rc = MIGRATEPAGE_SUCCESS;
Minchan Kim0dabec92011-10-31 17:06:57 -07001081 int *result = NULL;
Hugh Dickins2def7422015-11-05 18:49:46 -08001082 struct page *newpage;
Minchan Kim0dabec92011-10-31 17:06:57 -07001083
Hugh Dickins2def7422015-11-05 18:49:46 -08001084 newpage = get_new_page(page, private, &result);
Minchan Kim0dabec92011-10-31 17:06:57 -07001085 if (!newpage)
1086 return -ENOMEM;
1087
1088 if (page_count(page) == 1) {
1089 /* page was freed from under us. So we are done. */
Minchan Kimc6c919e2016-07-26 15:23:02 -07001090 ClearPageActive(page);
1091 ClearPageUnevictable(page);
Minchan Kimbda807d2016-07-26 15:23:05 -07001092 if (unlikely(__PageMovable(page))) {
1093 lock_page(page);
1094 if (!PageMovable(page))
1095 __ClearPageIsolated(page);
1096 unlock_page(page);
1097 }
Minchan Kimc6c919e2016-07-26 15:23:02 -07001098 if (put_new_page)
1099 put_new_page(newpage, private);
1100 else
1101 put_page(newpage);
Minchan Kim0dabec92011-10-31 17:06:57 -07001102 goto out;
1103 }
1104
Kirill A. Shutemov4d2fa962016-01-15 16:54:00 -08001105 if (unlikely(PageTransHuge(page))) {
1106 lock_page(page);
1107 rc = split_huge_page(page);
1108 unlock_page(page);
1109 if (rc)
Minchan Kim0dabec92011-10-31 17:06:57 -07001110 goto out;
Kirill A. Shutemov4d2fa962016-01-15 16:54:00 -08001111 }
Minchan Kim0dabec92011-10-31 17:06:57 -07001112
Hugh Dickins9c620e22013-02-22 16:35:14 -08001113 rc = __unmap_and_move(page, newpage, force, mode);
Minchan Kimc6c919e2016-07-26 15:23:02 -07001114 if (rc == MIGRATEPAGE_SUCCESS)
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001115 set_page_owner_migrate_reason(newpage, reason);
Rafael Aquinibf6bddf2012-12-11 16:02:42 -08001116
Minchan Kim0dabec92011-10-31 17:06:57 -07001117out:
Christoph Lametere24f0b82006-06-23 02:03:51 -07001118 if (rc != -EAGAIN) {
Minchan Kim0dabec92011-10-31 17:06:57 -07001119 /*
1120 * A page that has been migrated has all references
1121 * removed and will be freed. A page that has not been
1122 * migrated will have kepts its references and be
1123 * restored.
1124 */
1125 list_del(&page->lru);
Ming Ling6afcf8e2016-12-12 16:42:26 -08001126
1127 /*
1128 * Compaction can migrate also non-LRU pages which are
1129 * not accounted to NR_ISOLATED_*. They can be recognized
1130 * as __PageMovable
1131 */
1132 if (likely(!__PageMovable(page)))
1133 dec_node_page_state(page, NR_ISOLATED_ANON +
1134 page_is_file_cache(page));
Christoph Lametere24f0b82006-06-23 02:03:51 -07001135 }
David Rientjes68711a72014-06-04 16:08:25 -07001136
Christoph Lameter95a402c2006-06-23 02:03:53 -07001137 /*
Minchan Kimc6c919e2016-07-26 15:23:02 -07001138 * If migration is successful, releases reference grabbed during
1139 * isolation. Otherwise, restore the page to right list unless
1140 * we want to retry.
Christoph Lameter95a402c2006-06-23 02:03:53 -07001141 */
Minchan Kimc6c919e2016-07-26 15:23:02 -07001142 if (rc == MIGRATEPAGE_SUCCESS) {
1143 put_page(page);
1144 if (reason == MR_MEMORY_FAILURE) {
1145 /*
1146 * Set PG_HWPoison on just freed page
1147 * intentionally. Although it's rather weird,
1148 * it's how HWPoison flag works at the moment.
1149 */
1150 if (!test_set_page_hwpoison(page))
1151 num_poisoned_pages_inc();
1152 }
1153 } else {
Minchan Kimbda807d2016-07-26 15:23:05 -07001154 if (rc != -EAGAIN) {
1155 if (likely(!__PageMovable(page))) {
1156 putback_lru_page(page);
1157 goto put_new;
1158 }
1159
1160 lock_page(page);
1161 if (PageMovable(page))
1162 putback_movable_page(page);
1163 else
1164 __ClearPageIsolated(page);
1165 unlock_page(page);
1166 put_page(page);
1167 }
1168put_new:
Minchan Kimc6c919e2016-07-26 15:23:02 -07001169 if (put_new_page)
1170 put_new_page(newpage, private);
1171 else
1172 put_page(newpage);
1173 }
David Rientjes68711a72014-06-04 16:08:25 -07001174
Christoph Lameter742755a2006-06-23 02:03:55 -07001175 if (result) {
1176 if (rc)
1177 *result = rc;
1178 else
1179 *result = page_to_nid(newpage);
1180 }
Christoph Lametere24f0b82006-06-23 02:03:51 -07001181 return rc;
1182}
1183
1184/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001185 * Counterpart of unmap_and_move_page() for hugepage migration.
1186 *
1187 * This function doesn't wait the completion of hugepage I/O
1188 * because there is no race between I/O and migration for hugepage.
1189 * Note that currently hugepage I/O occurs only in direct I/O
1190 * where no lock is held and PG_writeback is irrelevant,
1191 * and writeback status of all subpages are counted in the reference
1192 * count of the head page (i.e. if all subpages of a 2MB hugepage are
1193 * under direct I/O, the reference of the head page is 512 and a bit more.)
1194 * This means that when we try to migrate hugepage whose subpages are
1195 * doing direct I/O, some references remain after try_to_unmap() and
1196 * hugepage migration fails without data corruption.
1197 *
1198 * There is also no race when direct I/O is issued on the page under migration,
1199 * because then pte is replaced with migration swap entry and direct I/O code
1200 * will wait in the page fault for migration to complete.
1201 */
1202static int unmap_and_move_huge_page(new_page_t get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001203 free_page_t put_new_page, unsigned long private,
1204 struct page *hpage, int force,
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001205 enum migrate_mode mode, int reason)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001206{
Hugh Dickins2def7422015-11-05 18:49:46 -08001207 int rc = -EAGAIN;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001208 int *result = NULL;
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001209 int page_was_mapped = 0;
Joonsoo Kim32665f22014-01-21 15:51:15 -08001210 struct page *new_hpage;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001211 struct anon_vma *anon_vma = NULL;
1212
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001213 /*
1214 * Movability of hugepages depends on architectures and hugepage size.
1215 * This check is necessary because some callers of hugepage migration
1216 * like soft offline and memory hotremove don't walk through page
1217 * tables or check whether the hugepage is pmd-based or not before
1218 * kicking migration.
1219 */
Naoya Horiguchi100873d2014-06-04 16:10:56 -07001220 if (!hugepage_migration_supported(page_hstate(hpage))) {
Joonsoo Kim32665f22014-01-21 15:51:15 -08001221 putback_active_hugepage(hpage);
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001222 return -ENOSYS;
Joonsoo Kim32665f22014-01-21 15:51:15 -08001223 }
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001224
Joonsoo Kim32665f22014-01-21 15:51:15 -08001225 new_hpage = get_new_page(hpage, private, &result);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001226 if (!new_hpage)
1227 return -ENOMEM;
1228
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001229 if (!trylock_page(hpage)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -08001230 if (!force || mode != MIGRATE_SYNC)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001231 goto out;
1232 lock_page(hpage);
1233 }
1234
Peter Zijlstra746b18d2011-05-24 17:12:10 -07001235 if (PageAnon(hpage))
1236 anon_vma = page_get_anon_vma(hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001237
Hugh Dickins7db76712015-11-05 18:49:49 -08001238 if (unlikely(!trylock_page(new_hpage)))
1239 goto put_anon;
1240
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001241 if (page_mapped(hpage)) {
1242 try_to_unmap(hpage,
1243 TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
1244 page_was_mapped = 1;
1245 }
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001246
1247 if (!page_mapped(hpage))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001248 rc = move_to_new_page(new_hpage, hpage, mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001249
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001250 if (page_was_mapped)
1251 remove_migration_ptes(hpage,
Kirill A. Shutemove3884662016-03-17 14:20:07 -07001252 rc == MIGRATEPAGE_SUCCESS ? new_hpage : hpage, false);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001253
Hugh Dickins7db76712015-11-05 18:49:49 -08001254 unlock_page(new_hpage);
1255
1256put_anon:
Hugh Dickinsfd4a4662011-01-13 15:47:31 -08001257 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -07001258 put_anon_vma(anon_vma);
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001259
Hugh Dickins2def7422015-11-05 18:49:46 -08001260 if (rc == MIGRATEPAGE_SUCCESS) {
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001261 hugetlb_cgroup_migrate(hpage, new_hpage);
Hugh Dickins2def7422015-11-05 18:49:46 -08001262 put_new_page = NULL;
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001263 set_page_owner_migrate_reason(new_hpage, reason);
Hugh Dickins2def7422015-11-05 18:49:46 -08001264 }
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001265
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001266 unlock_page(hpage);
Hillf Danton09761332011-12-08 14:34:20 -08001267out:
Naoya Horiguchib8ec1ce2013-09-11 14:22:01 -07001268 if (rc != -EAGAIN)
1269 putback_active_hugepage(hpage);
David Rientjes68711a72014-06-04 16:08:25 -07001270
1271 /*
1272 * If migration was not successful and there's a freeing callback, use
1273 * it. Otherwise, put_page() will drop the reference grabbed during
1274 * isolation.
1275 */
Hugh Dickins2def7422015-11-05 18:49:46 -08001276 if (put_new_page)
David Rientjes68711a72014-06-04 16:08:25 -07001277 put_new_page(new_hpage, private);
1278 else
Naoya Horiguchi3aaa76e2015-09-22 14:59:14 -07001279 putback_active_hugepage(new_hpage);
David Rientjes68711a72014-06-04 16:08:25 -07001280
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001281 if (result) {
1282 if (rc)
1283 *result = rc;
1284 else
1285 *result = page_to_nid(new_hpage);
1286 }
1287 return rc;
1288}
1289
1290/*
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001291 * migrate_pages - migrate the pages specified in a list, to the free pages
1292 * supplied as the target for the page migration
Christoph Lameterb20a3502006-03-22 00:09:12 -08001293 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001294 * @from: The list of pages to be migrated.
1295 * @get_new_page: The function used to allocate free pages to be used
1296 * as the target of the page migration.
David Rientjes68711a72014-06-04 16:08:25 -07001297 * @put_new_page: The function used to free target pages if migration
1298 * fails, or NULL if no special handling is necessary.
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001299 * @private: Private data to be passed on to get_new_page()
1300 * @mode: The migration mode that specifies the constraints for
1301 * page migration, if any.
1302 * @reason: The reason for page migration.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001303 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001304 * The function returns after 10 attempts or if no pages are movable any more
1305 * because the list has become empty or no retryable pages exist any more.
Hugh Dickins14e0f9b2015-11-05 18:49:43 -08001306 * The caller should call putback_movable_pages() to return pages to the LRU
Minchan Kim28bd6572011-01-25 15:07:26 -08001307 * or free list only if ret != 0.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001308 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001309 * Returns the number of pages that were not migrated, or an error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001310 */
Hugh Dickins9c620e22013-02-22 16:35:14 -08001311int migrate_pages(struct list_head *from, new_page_t get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001312 free_page_t put_new_page, unsigned long private,
1313 enum migrate_mode mode, int reason)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001314{
Christoph Lametere24f0b82006-06-23 02:03:51 -07001315 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001316 int nr_failed = 0;
Mel Gorman5647bc22012-10-19 10:46:20 +01001317 int nr_succeeded = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001318 int pass = 0;
1319 struct page *page;
1320 struct page *page2;
1321 int swapwrite = current->flags & PF_SWAPWRITE;
1322 int rc;
1323
1324 if (!swapwrite)
1325 current->flags |= PF_SWAPWRITE;
1326
Christoph Lametere24f0b82006-06-23 02:03:51 -07001327 for(pass = 0; pass < 10 && retry; pass++) {
1328 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001329
Christoph Lametere24f0b82006-06-23 02:03:51 -07001330 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -07001331 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -08001332
Naoya Horiguchi31caf662013-09-11 14:21:59 -07001333 if (PageHuge(page))
1334 rc = unmap_and_move_huge_page(get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001335 put_new_page, private, page,
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001336 pass > 2, mode, reason);
Naoya Horiguchi31caf662013-09-11 14:21:59 -07001337 else
David Rientjes68711a72014-06-04 16:08:25 -07001338 rc = unmap_and_move(get_new_page, put_new_page,
Naoya Horiguchiadd05ce2015-06-24 16:56:50 -07001339 private, page, pass > 2, mode,
1340 reason);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001341
Christoph Lametere24f0b82006-06-23 02:03:51 -07001342 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -07001343 case -ENOMEM:
David Rientjesdfef2ef2016-05-20 16:59:05 -07001344 nr_failed++;
Christoph Lameter95a402c2006-06-23 02:03:53 -07001345 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001346 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001347 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001348 break;
Rafael Aquini78bd5202012-12-11 16:02:31 -08001349 case MIGRATEPAGE_SUCCESS:
Mel Gorman5647bc22012-10-19 10:46:20 +01001350 nr_succeeded++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001351 break;
1352 default:
Naoya Horiguchi354a3362014-01-21 15:51:14 -08001353 /*
1354 * Permanent failure (-EBUSY, -ENOSYS, etc.):
1355 * unlike -EAGAIN case, the failed page is
1356 * removed from migration page list and not
1357 * retried in the next outer loop.
1358 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001359 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001360 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001361 }
Christoph Lameterb20a3502006-03-22 00:09:12 -08001362 }
1363 }
Vlastimil Babkaf2f81fb2015-11-05 18:47:03 -08001364 nr_failed += retry;
1365 rc = nr_failed;
Christoph Lameter95a402c2006-06-23 02:03:53 -07001366out:
Mel Gorman5647bc22012-10-19 10:46:20 +01001367 if (nr_succeeded)
1368 count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded);
1369 if (nr_failed)
1370 count_vm_events(PGMIGRATE_FAIL, nr_failed);
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001371 trace_mm_migrate_pages(nr_succeeded, nr_failed, mode, reason);
1372
Christoph Lameterb20a3502006-03-22 00:09:12 -08001373 if (!swapwrite)
1374 current->flags &= ~PF_SWAPWRITE;
1375
Rafael Aquini78bd5202012-12-11 16:02:31 -08001376 return rc;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001377}
1378
Christoph Lameter742755a2006-06-23 02:03:55 -07001379#ifdef CONFIG_NUMA
1380/*
1381 * Move a list of individual pages
1382 */
1383struct page_to_node {
1384 unsigned long addr;
1385 struct page *page;
1386 int node;
1387 int status;
1388};
1389
1390static struct page *new_page_node(struct page *p, unsigned long private,
1391 int **result)
1392{
1393 struct page_to_node *pm = (struct page_to_node *)private;
1394
1395 while (pm->node != MAX_NUMNODES && pm->page != p)
1396 pm++;
1397
1398 if (pm->node == MAX_NUMNODES)
1399 return NULL;
1400
1401 *result = &pm->status;
1402
Naoya Horiguchie632a932013-09-11 14:22:04 -07001403 if (PageHuge(p))
1404 return alloc_huge_page_node(page_hstate(compound_head(p)),
1405 pm->node);
1406 else
Vlastimil Babka96db8002015-09-08 15:03:50 -07001407 return __alloc_pages_node(pm->node,
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001408 GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -07001409}
1410
1411/*
1412 * Move a set of pages as indicated in the pm array. The addr
1413 * field must be set to the virtual address of the page to be moved
1414 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001415 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -07001416 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001417static int do_move_page_to_node_array(struct mm_struct *mm,
1418 struct page_to_node *pm,
1419 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -07001420{
1421 int err;
1422 struct page_to_node *pp;
1423 LIST_HEAD(pagelist);
1424
1425 down_read(&mm->mmap_sem);
1426
1427 /*
1428 * Build a list of pages to migrate
1429 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001430 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1431 struct vm_area_struct *vma;
1432 struct page *page;
1433
Christoph Lameter742755a2006-06-23 02:03:55 -07001434 err = -EFAULT;
1435 vma = find_vma(mm, pp->addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001436 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -07001437 goto set_status;
1438
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001439 /* FOLL_DUMP to ignore special (like zero) pages */
1440 page = follow_page(vma, pp->addr,
1441 FOLL_GET | FOLL_SPLIT | FOLL_DUMP);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001442
1443 err = PTR_ERR(page);
1444 if (IS_ERR(page))
1445 goto set_status;
1446
Christoph Lameter742755a2006-06-23 02:03:55 -07001447 err = -ENOENT;
1448 if (!page)
1449 goto set_status;
1450
Christoph Lameter742755a2006-06-23 02:03:55 -07001451 pp->page = page;
1452 err = page_to_nid(page);
1453
1454 if (err == pp->node)
1455 /*
1456 * Node already in the right place
1457 */
1458 goto put_and_set;
1459
1460 err = -EACCES;
1461 if (page_mapcount(page) > 1 &&
1462 !migrate_all)
1463 goto put_and_set;
1464
Naoya Horiguchie632a932013-09-11 14:22:04 -07001465 if (PageHuge(page)) {
Naoya Horiguchie66f17f2015-02-11 15:25:22 -08001466 if (PageHead(page))
1467 isolate_huge_page(page, &pagelist);
Naoya Horiguchie632a932013-09-11 14:22:04 -07001468 goto put_and_set;
1469 }
1470
Nick Piggin62695a82008-10-18 20:26:09 -07001471 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001472 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -07001473 list_add_tail(&page->lru, &pagelist);
Mel Gorman599d0c92016-07-28 15:45:31 -07001474 inc_node_page_state(page, NR_ISOLATED_ANON +
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001475 page_is_file_cache(page));
1476 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001477put_and_set:
1478 /*
1479 * Either remove the duplicate refcount from
1480 * isolate_lru_page() or drop the page ref if it was
1481 * not isolated.
1482 */
1483 put_page(page);
1484set_status:
1485 pp->status = err;
1486 }
1487
Brice Gogline78bbfa2008-10-18 20:27:15 -07001488 err = 0;
Minchan Kimcf608ac2010-10-26 14:21:29 -07001489 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001490 err = migrate_pages(&pagelist, new_page_node, NULL,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001491 (unsigned long)pm, MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001492 if (err)
Naoya Horiguchie632a932013-09-11 14:22:04 -07001493 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001494 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001495
1496 up_read(&mm->mmap_sem);
1497 return err;
1498}
1499
1500/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001501 * Migrate an array of page address onto an array of nodes and fill
1502 * the corresponding array of status.
1503 */
Christoph Lameter3268c632012-03-21 16:34:06 -07001504static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001505 unsigned long nr_pages,
1506 const void __user * __user *pages,
1507 const int __user *nodes,
1508 int __user *status, int flags)
1509{
Brice Goglin3140a222009-01-06 14:38:57 -08001510 struct page_to_node *pm;
Brice Goglin3140a222009-01-06 14:38:57 -08001511 unsigned long chunk_nr_pages;
1512 unsigned long chunk_start;
1513 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001514
Brice Goglin3140a222009-01-06 14:38:57 -08001515 err = -ENOMEM;
1516 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1517 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001518 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -07001519
1520 migrate_prep();
1521
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001522 /*
Brice Goglin3140a222009-01-06 14:38:57 -08001523 * Store a chunk of page_to_node array in a page,
1524 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001525 */
Brice Goglin3140a222009-01-06 14:38:57 -08001526 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001527
Brice Goglin3140a222009-01-06 14:38:57 -08001528 for (chunk_start = 0;
1529 chunk_start < nr_pages;
1530 chunk_start += chunk_nr_pages) {
1531 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001532
Brice Goglin3140a222009-01-06 14:38:57 -08001533 if (chunk_start + chunk_nr_pages > nr_pages)
1534 chunk_nr_pages = nr_pages - chunk_start;
1535
1536 /* fill the chunk pm with addrs and nodes from user-space */
1537 for (j = 0; j < chunk_nr_pages; j++) {
1538 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001539 int node;
1540
Brice Goglin3140a222009-01-06 14:38:57 -08001541 err = -EFAULT;
1542 if (get_user(p, pages + j + chunk_start))
1543 goto out_pm;
1544 pm[j].addr = (unsigned long) p;
1545
1546 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001547 goto out_pm;
1548
1549 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -08001550 if (node < 0 || node >= MAX_NUMNODES)
1551 goto out_pm;
1552
Lai Jiangshan389162c2012-12-12 13:51:30 -08001553 if (!node_state(node, N_MEMORY))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001554 goto out_pm;
1555
1556 err = -EACCES;
1557 if (!node_isset(node, task_nodes))
1558 goto out_pm;
1559
Brice Goglin3140a222009-01-06 14:38:57 -08001560 pm[j].node = node;
1561 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001562
Brice Goglin3140a222009-01-06 14:38:57 -08001563 /* End marker for this chunk */
1564 pm[chunk_nr_pages].node = MAX_NUMNODES;
1565
1566 /* Migrate this chunk */
1567 err = do_move_page_to_node_array(mm, pm,
1568 flags & MPOL_MF_MOVE_ALL);
1569 if (err < 0)
1570 goto out_pm;
1571
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001572 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -08001573 for (j = 0; j < chunk_nr_pages; j++)
1574 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001575 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -08001576 goto out_pm;
1577 }
1578 }
1579 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001580
1581out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -08001582 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001583out:
1584 return err;
1585}
1586
1587/*
Brice Goglin2f007e72008-10-18 20:27:16 -07001588 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -07001589 */
Brice Goglin80bba122008-12-09 13:14:23 -08001590static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1591 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001592{
Brice Goglin2f007e72008-10-18 20:27:16 -07001593 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001594
Christoph Lameter742755a2006-06-23 02:03:55 -07001595 down_read(&mm->mmap_sem);
1596
Brice Goglin2f007e72008-10-18 20:27:16 -07001597 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001598 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001599 struct vm_area_struct *vma;
1600 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001601 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001602
1603 vma = find_vma(mm, addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001604 if (!vma || addr < vma->vm_start)
Christoph Lameter742755a2006-06-23 02:03:55 -07001605 goto set_status;
1606
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001607 /* FOLL_DUMP to ignore special (like zero) pages */
1608 page = follow_page(vma, addr, FOLL_DUMP);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001609
1610 err = PTR_ERR(page);
1611 if (IS_ERR(page))
1612 goto set_status;
1613
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001614 err = page ? page_to_nid(page) : -ENOENT;
Christoph Lameter742755a2006-06-23 02:03:55 -07001615set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001616 *status = err;
1617
1618 pages++;
1619 status++;
1620 }
1621
1622 up_read(&mm->mmap_sem);
1623}
1624
1625/*
1626 * Determine the nodes of a user array of pages and store it in
1627 * a user array of status.
1628 */
1629static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1630 const void __user * __user *pages,
1631 int __user *status)
1632{
1633#define DO_PAGES_STAT_CHUNK_NR 16
1634 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1635 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
Brice Goglin80bba122008-12-09 13:14:23 -08001636
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001637 while (nr_pages) {
1638 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001639
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001640 chunk_nr = nr_pages;
1641 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1642 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1643
1644 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1645 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001646
1647 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1648
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001649 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1650 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001651
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001652 pages += chunk_nr;
1653 status += chunk_nr;
1654 nr_pages -= chunk_nr;
1655 }
1656 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001657}
1658
1659/*
1660 * Move a list of pages in the address space of the currently executing
1661 * process.
1662 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001663SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1664 const void __user * __user *, pages,
1665 const int __user *, nodes,
1666 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001667{
David Howellsc69e8d92008-11-14 10:39:19 +11001668 const struct cred *cred = current_cred(), *tcred;
Christoph Lameter742755a2006-06-23 02:03:55 -07001669 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001670 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001671 int err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001672 nodemask_t task_nodes;
Christoph Lameter742755a2006-06-23 02:03:55 -07001673
1674 /* Check flags */
1675 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1676 return -EINVAL;
1677
1678 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1679 return -EPERM;
1680
1681 /* Find the mm_struct */
Greg Thelena879bf52011-02-25 14:44:13 -08001682 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001683 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001684 if (!task) {
Greg Thelena879bf52011-02-25 14:44:13 -08001685 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001686 return -ESRCH;
1687 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001688 get_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001689
1690 /*
1691 * Check if this process has the right to modify the specified
1692 * process. The right exists if the process has administrative
1693 * capabilities, superuser privileges or the same
1694 * userid as the target process.
1695 */
David Howellsc69e8d92008-11-14 10:39:19 +11001696 tcred = __task_cred(task);
Eric W. Biedermanb38a86e2012-03-12 15:48:24 -07001697 if (!uid_eq(cred->euid, tcred->suid) && !uid_eq(cred->euid, tcred->uid) &&
1698 !uid_eq(cred->uid, tcred->suid) && !uid_eq(cred->uid, tcred->uid) &&
Christoph Lameter742755a2006-06-23 02:03:55 -07001699 !capable(CAP_SYS_NICE)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001700 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001701 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001702 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001703 }
David Howellsc69e8d92008-11-14 10:39:19 +11001704 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001705
David Quigley86c3a762006-06-23 02:04:02 -07001706 err = security_task_movememory(task);
1707 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001708 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001709
Christoph Lameter3268c632012-03-21 16:34:06 -07001710 task_nodes = cpuset_mems_allowed(task);
1711 mm = get_task_mm(task);
1712 put_task_struct(task);
1713
Sasha Levin6e8b09e2012-04-25 16:01:53 -07001714 if (!mm)
1715 return -EINVAL;
1716
1717 if (nodes)
1718 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1719 nodes, status, flags);
1720 else
1721 err = do_pages_stat(mm, nr_pages, pages, status);
Christoph Lameter3268c632012-03-21 16:34:06 -07001722
1723 mmput(mm);
1724 return err;
David Quigley86c3a762006-06-23 02:04:02 -07001725
Christoph Lameter742755a2006-06-23 02:03:55 -07001726out:
Christoph Lameter3268c632012-03-21 16:34:06 -07001727 put_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001728 return err;
1729}
Christoph Lameter742755a2006-06-23 02:03:55 -07001730
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001731#ifdef CONFIG_NUMA_BALANCING
1732/*
1733 * Returns true if this is a safe migration target node for misplaced NUMA
1734 * pages. Currently it only checks the watermarks which crude
1735 */
1736static bool migrate_balanced_pgdat(struct pglist_data *pgdat,
Mel Gorman3abef4e2013-02-22 16:34:27 -08001737 unsigned long nr_migrate_pages)
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001738{
1739 int z;
Mel Gorman599d0c92016-07-28 15:45:31 -07001740
1741 if (!pgdat_reclaimable(pgdat))
1742 return false;
1743
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001744 for (z = pgdat->nr_zones - 1; z >= 0; z--) {
1745 struct zone *zone = pgdat->node_zones + z;
1746
1747 if (!populated_zone(zone))
1748 continue;
1749
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001750 /* Avoid waking kswapd by allocating pages_to_migrate pages. */
1751 if (!zone_watermark_ok(zone, 0,
1752 high_wmark_pages(zone) +
1753 nr_migrate_pages,
1754 0, 0))
1755 continue;
1756 return true;
1757 }
1758 return false;
1759}
1760
1761static struct page *alloc_misplaced_dst_page(struct page *page,
1762 unsigned long data,
1763 int **result)
1764{
1765 int nid = (int) data;
1766 struct page *newpage;
1767
Vlastimil Babka96db8002015-09-08 15:03:50 -07001768 newpage = __alloc_pages_node(nid,
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001769 (GFP_HIGHUSER_MOVABLE |
1770 __GFP_THISNODE | __GFP_NOMEMALLOC |
1771 __GFP_NORETRY | __GFP_NOWARN) &
Mel Gorman8479eba2016-02-26 15:19:31 -08001772 ~__GFP_RECLAIM, 0);
Hillf Dantonbac03822012-11-27 14:46:24 +00001773
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001774 return newpage;
1775}
1776
1777/*
Mel Gormana8f60772012-11-14 21:41:46 +00001778 * page migration rate limiting control.
1779 * Do not migrate more than @pages_to_migrate in a @migrate_interval_millisecs
1780 * window of time. Default here says do not migrate more than 1280M per second.
1781 */
1782static unsigned int migrate_interval_millisecs __read_mostly = 100;
1783static unsigned int ratelimit_pages __read_mostly = 128 << (20 - PAGE_SHIFT);
1784
Mel Gormanb32967f2012-11-19 12:35:47 +00001785/* Returns true if the node is migrate rate-limited after the update */
Mel Gorman1c30e012014-01-21 15:50:58 -08001786static bool numamigrate_update_ratelimit(pg_data_t *pgdat,
1787 unsigned long nr_pages)
Mel Gormanb32967f2012-11-19 12:35:47 +00001788{
Mel Gormanb32967f2012-11-19 12:35:47 +00001789 /*
1790 * Rate-limit the amount of data that is being migrated to a node.
1791 * Optimal placement is no good if the memory bus is saturated and
1792 * all the time is being spent migrating!
1793 */
Mel Gormanb32967f2012-11-19 12:35:47 +00001794 if (time_after(jiffies, pgdat->numabalancing_migrate_next_window)) {
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001795 spin_lock(&pgdat->numabalancing_migrate_lock);
Mel Gormanb32967f2012-11-19 12:35:47 +00001796 pgdat->numabalancing_migrate_nr_pages = 0;
1797 pgdat->numabalancing_migrate_next_window = jiffies +
1798 msecs_to_jiffies(migrate_interval_millisecs);
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001799 spin_unlock(&pgdat->numabalancing_migrate_lock);
Mel Gormanb32967f2012-11-19 12:35:47 +00001800 }
Mel Gormanaf1839d2014-01-21 15:51:01 -08001801 if (pgdat->numabalancing_migrate_nr_pages > ratelimit_pages) {
1802 trace_mm_numa_migrate_ratelimit(current, pgdat->node_id,
1803 nr_pages);
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001804 return true;
Mel Gormanaf1839d2014-01-21 15:51:01 -08001805 }
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001806
1807 /*
1808 * This is an unlocked non-atomic update so errors are possible.
1809 * The consequences are failing to migrate when we potentiall should
1810 * have which is not severe enough to warrant locking. If it is ever
1811 * a problem, it can be converted to a per-cpu counter.
1812 */
1813 pgdat->numabalancing_migrate_nr_pages += nr_pages;
1814 return false;
Mel Gormanb32967f2012-11-19 12:35:47 +00001815}
1816
Mel Gorman1c30e012014-01-21 15:50:58 -08001817static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page)
Mel Gormanb32967f2012-11-19 12:35:47 +00001818{
Hugh Dickins340ef392013-02-22 16:34:33 -08001819 int page_lru;
Mel Gormanb32967f2012-11-19 12:35:47 +00001820
Sasha Levin309381fea2014-01-23 15:52:54 -08001821 VM_BUG_ON_PAGE(compound_order(page) && !PageTransHuge(page), page);
Mel Gorman3abef4e2013-02-22 16:34:27 -08001822
Mel Gormanb32967f2012-11-19 12:35:47 +00001823 /* Avoid migrating to a node that is nearly full */
Hugh Dickins340ef392013-02-22 16:34:33 -08001824 if (!migrate_balanced_pgdat(pgdat, 1UL << compound_order(page)))
1825 return 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001826
Hugh Dickins340ef392013-02-22 16:34:33 -08001827 if (isolate_lru_page(page))
1828 return 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001829
1830 /*
Hugh Dickins340ef392013-02-22 16:34:33 -08001831 * migrate_misplaced_transhuge_page() skips page migration's usual
1832 * check on page_count(), so we must do it here, now that the page
1833 * has been isolated: a GUP pin, or any other pin, prevents migration.
1834 * The expected page count is 3: 1 for page's mapcount and 1 for the
1835 * caller's pin and 1 for the reference taken by isolate_lru_page().
1836 */
1837 if (PageTransHuge(page) && page_count(page) != 3) {
1838 putback_lru_page(page);
1839 return 0;
1840 }
1841
1842 page_lru = page_is_file_cache(page);
Mel Gorman599d0c92016-07-28 15:45:31 -07001843 mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + page_lru,
Hugh Dickins340ef392013-02-22 16:34:33 -08001844 hpage_nr_pages(page));
1845
1846 /*
1847 * Isolating the page has taken another reference, so the
1848 * caller's reference can be safely dropped without the page
1849 * disappearing underneath us during migration.
Mel Gormanb32967f2012-11-19 12:35:47 +00001850 */
1851 put_page(page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001852 return 1;
Mel Gormanb32967f2012-11-19 12:35:47 +00001853}
1854
Mel Gormande466bd2013-12-18 17:08:42 -08001855bool pmd_trans_migrating(pmd_t pmd)
1856{
1857 struct page *page = pmd_page(pmd);
1858 return PageLocked(page);
1859}
1860
Mel Gormana8f60772012-11-14 21:41:46 +00001861/*
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001862 * Attempt to migrate a misplaced page to the specified destination
1863 * node. Caller is expected to have an elevated reference count on
1864 * the page that will be dropped by this function before returning.
1865 */
Mel Gorman1bc115d2013-10-07 11:29:05 +01001866int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma,
1867 int node)
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001868{
Mel Gormana8f60772012-11-14 21:41:46 +00001869 pg_data_t *pgdat = NODE_DATA(node);
Hugh Dickins340ef392013-02-22 16:34:33 -08001870 int isolated;
Mel Gormanb32967f2012-11-19 12:35:47 +00001871 int nr_remaining;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001872 LIST_HEAD(migratepages);
1873
1874 /*
Mel Gorman1bc115d2013-10-07 11:29:05 +01001875 * Don't migrate file pages that are mapped in multiple processes
1876 * with execute permissions as they are probably shared libraries.
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001877 */
Mel Gorman1bc115d2013-10-07 11:29:05 +01001878 if (page_mapcount(page) != 1 && page_is_file_cache(page) &&
1879 (vma->vm_flags & VM_EXEC))
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001880 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001881
Mel Gormana8f60772012-11-14 21:41:46 +00001882 /*
1883 * Rate-limit the amount of data that is being migrated to a node.
1884 * Optimal placement is no good if the memory bus is saturated and
1885 * all the time is being spent migrating!
1886 */
Hugh Dickins340ef392013-02-22 16:34:33 -08001887 if (numamigrate_update_ratelimit(pgdat, 1))
Mel Gormana8f60772012-11-14 21:41:46 +00001888 goto out;
Mel Gormana8f60772012-11-14 21:41:46 +00001889
Mel Gormanb32967f2012-11-19 12:35:47 +00001890 isolated = numamigrate_isolate_page(pgdat, page);
1891 if (!isolated)
1892 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001893
Mel Gormanb32967f2012-11-19 12:35:47 +00001894 list_add(&page->lru, &migratepages);
Hugh Dickins9c620e22013-02-22 16:35:14 -08001895 nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page,
David Rientjes68711a72014-06-04 16:08:25 -07001896 NULL, node, MIGRATE_ASYNC,
1897 MR_NUMA_MISPLACED);
Mel Gormanb32967f2012-11-19 12:35:47 +00001898 if (nr_remaining) {
Joonsoo Kim59c82b72014-01-21 15:51:17 -08001899 if (!list_empty(&migratepages)) {
1900 list_del(&page->lru);
Mel Gorman599d0c92016-07-28 15:45:31 -07001901 dec_node_page_state(page, NR_ISOLATED_ANON +
Joonsoo Kim59c82b72014-01-21 15:51:17 -08001902 page_is_file_cache(page));
1903 putback_lru_page(page);
1904 }
Mel Gormanb32967f2012-11-19 12:35:47 +00001905 isolated = 0;
1906 } else
1907 count_vm_numa_event(NUMA_PAGE_MIGRATE);
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001908 BUG_ON(!list_empty(&migratepages));
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001909 return isolated;
Hugh Dickins340ef392013-02-22 16:34:33 -08001910
1911out:
1912 put_page(page);
1913 return 0;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001914}
Mel Gorman220018d2012-12-05 09:32:56 +00001915#endif /* CONFIG_NUMA_BALANCING */
Mel Gormanb32967f2012-11-19 12:35:47 +00001916
Mel Gorman220018d2012-12-05 09:32:56 +00001917#if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
Hugh Dickins340ef392013-02-22 16:34:33 -08001918/*
1919 * Migrates a THP to a given target node. page must be locked and is unlocked
1920 * before returning.
1921 */
Mel Gormanb32967f2012-11-19 12:35:47 +00001922int migrate_misplaced_transhuge_page(struct mm_struct *mm,
1923 struct vm_area_struct *vma,
1924 pmd_t *pmd, pmd_t entry,
1925 unsigned long address,
1926 struct page *page, int node)
1927{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001928 spinlock_t *ptl;
Mel Gormanb32967f2012-11-19 12:35:47 +00001929 pg_data_t *pgdat = NODE_DATA(node);
1930 int isolated = 0;
1931 struct page *new_page = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00001932 int page_lru = page_is_file_cache(page);
Mel Gormanf714f4f2013-12-18 17:08:33 -08001933 unsigned long mmun_start = address & HPAGE_PMD_MASK;
1934 unsigned long mmun_end = mmun_start + HPAGE_PMD_SIZE;
Mel Gorman2b4847e2013-12-18 17:08:32 -08001935 pmd_t orig_entry;
Mel Gormanb32967f2012-11-19 12:35:47 +00001936
1937 /*
Mel Gormanb32967f2012-11-19 12:35:47 +00001938 * Rate-limit the amount of data that is being migrated to a node.
1939 * Optimal placement is no good if the memory bus is saturated and
1940 * all the time is being spent migrating!
1941 */
Mel Gormand28d43352012-11-29 09:24:36 +00001942 if (numamigrate_update_ratelimit(pgdat, HPAGE_PMD_NR))
Mel Gormanb32967f2012-11-19 12:35:47 +00001943 goto out_dropref;
1944
1945 new_page = alloc_pages_node(node,
Vlastimil Babka25160352016-07-28 15:49:25 -07001946 (GFP_TRANSHUGE_LIGHT | __GFP_THISNODE),
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001947 HPAGE_PMD_ORDER);
Hugh Dickins340ef392013-02-22 16:34:33 -08001948 if (!new_page)
1949 goto out_fail;
Kirill A. Shutemov9a982252016-01-15 16:54:17 -08001950 prep_transhuge_page(new_page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001951
Mel Gormanb32967f2012-11-19 12:35:47 +00001952 isolated = numamigrate_isolate_page(pgdat, page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001953 if (!isolated) {
Mel Gormanb32967f2012-11-19 12:35:47 +00001954 put_page(new_page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001955 goto out_fail;
Mel Gormanb32967f2012-11-19 12:35:47 +00001956 }
Aneesh Kumar K.V458aa762016-03-17 14:18:56 -07001957 /*
1958 * We are not sure a pending tlb flush here is for a huge page
1959 * mapping or not. Hence use the tlb range variant
1960 */
Mel Gormanb0943d62013-12-18 17:08:46 -08001961 if (mm_tlb_flush_pending(mm))
1962 flush_tlb_range(vma, mmun_start, mmun_end);
1963
Mel Gormanb32967f2012-11-19 12:35:47 +00001964 /* Prepare a page as a migration target */
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001965 __SetPageLocked(new_page);
Hugh Dickinsfa9949d2016-05-19 17:12:41 -07001966 __SetPageSwapBacked(new_page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001967
1968 /* anon mapping, we can simply copy page->mapping to the new page: */
1969 new_page->mapping = page->mapping;
1970 new_page->index = page->index;
1971 migrate_page_copy(new_page, page);
1972 WARN_ON(PageLRU(new_page));
1973
1974 /* Recheck the target PMD */
Mel Gormanf714f4f2013-12-18 17:08:33 -08001975 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001976 ptl = pmd_lock(mm, pmd);
Mel Gorman2b4847e2013-12-18 17:08:32 -08001977 if (unlikely(!pmd_same(*pmd, entry) || page_count(page) != 2)) {
1978fail_putback:
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001979 spin_unlock(ptl);
Mel Gormanf714f4f2013-12-18 17:08:33 -08001980 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Mel Gormanb32967f2012-11-19 12:35:47 +00001981
1982 /* Reverse changes made by migrate_page_copy() */
1983 if (TestClearPageActive(new_page))
1984 SetPageActive(page);
1985 if (TestClearPageUnevictable(new_page))
1986 SetPageUnevictable(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001987
1988 unlock_page(new_page);
1989 put_page(new_page); /* Free it */
1990
Mel Gormana54a4072013-10-07 11:28:46 +01001991 /* Retake the callers reference and putback on LRU */
1992 get_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001993 putback_lru_page(page);
Mel Gorman599d0c92016-07-28 15:45:31 -07001994 mod_node_page_state(page_pgdat(page),
Mel Gormana54a4072013-10-07 11:28:46 +01001995 NR_ISOLATED_ANON + page_lru, -HPAGE_PMD_NR);
Mel Gormaneb4489f62013-12-18 17:08:39 -08001996
1997 goto out_unlock;
Mel Gormanb32967f2012-11-19 12:35:47 +00001998 }
1999
Mel Gorman2b4847e2013-12-18 17:08:32 -08002000 orig_entry = *pmd;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07002001 entry = mk_huge_pmd(new_page, vma->vm_page_prot);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002002 entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
Mel Gormanb32967f2012-11-19 12:35:47 +00002003
Mel Gorman2b4847e2013-12-18 17:08:32 -08002004 /*
2005 * Clear the old entry under pagetable lock and establish the new PTE.
2006 * Any parallel GUP will either observe the old page blocking on the
2007 * page lock, block on the page table lock or observe the new page.
2008 * The SetPageUptodate on the new page and page_add_new_anon_rmap
2009 * guarantee the copy is visible before the pagetable update.
2010 */
Mel Gormanf714f4f2013-12-18 17:08:33 -08002011 flush_cache_range(vma, mmun_start, mmun_end);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002012 page_add_anon_rmap(new_page, vma, mmun_start, true);
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -07002013 pmdp_huge_clear_flush_notify(vma, mmun_start, pmd);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002014 set_pmd_at(mm, mmun_start, pmd, entry);
Stephen Rothwellce4a9cc2012-12-10 19:50:57 +11002015 update_mmu_cache_pmd(vma, address, &entry);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002016
2017 if (page_count(page) != 2) {
Mel Gormanf714f4f2013-12-18 17:08:33 -08002018 set_pmd_at(mm, mmun_start, pmd, orig_entry);
Aneesh Kumar K.V458aa762016-03-17 14:18:56 -07002019 flush_pmd_tlb_range(vma, mmun_start, mmun_end);
Joerg Roedel34ee6452014-11-13 13:46:09 +11002020 mmu_notifier_invalidate_range(mm, mmun_start, mmun_end);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002021 update_mmu_cache_pmd(vma, address, &entry);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002022 page_remove_rmap(new_page, true);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002023 goto fail_putback;
2024 }
2025
Hugh Dickins51afb122015-11-05 18:49:37 -08002026 mlock_migrate_page(new_page, page);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002027 page_remove_rmap(page, true);
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07002028 set_page_owner_migrate_reason(new_page, MR_NUMA_MISPLACED);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002029
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08002030 spin_unlock(ptl);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002031 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Mel Gormanb32967f2012-11-19 12:35:47 +00002032
Mel Gorman11de9922014-06-04 16:07:41 -07002033 /* Take an "isolate" reference and put new page on the LRU. */
2034 get_page(new_page);
2035 putback_lru_page(new_page);
2036
Mel Gormanb32967f2012-11-19 12:35:47 +00002037 unlock_page(new_page);
2038 unlock_page(page);
2039 put_page(page); /* Drop the rmap reference */
2040 put_page(page); /* Drop the LRU isolation reference */
2041
2042 count_vm_events(PGMIGRATE_SUCCESS, HPAGE_PMD_NR);
2043 count_vm_numa_events(NUMA_PAGE_MIGRATE, HPAGE_PMD_NR);
2044
Mel Gorman599d0c92016-07-28 15:45:31 -07002045 mod_node_page_state(page_pgdat(page),
Mel Gormanb32967f2012-11-19 12:35:47 +00002046 NR_ISOLATED_ANON + page_lru,
2047 -HPAGE_PMD_NR);
2048 return isolated;
2049
Hugh Dickins340ef392013-02-22 16:34:33 -08002050out_fail:
2051 count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR);
Mel Gormanb32967f2012-11-19 12:35:47 +00002052out_dropref:
Mel Gorman2b4847e2013-12-18 17:08:32 -08002053 ptl = pmd_lock(mm, pmd);
2054 if (pmd_same(*pmd, entry)) {
Mel Gorman4d942462015-02-12 14:58:28 -08002055 entry = pmd_modify(entry, vma->vm_page_prot);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002056 set_pmd_at(mm, mmun_start, pmd, entry);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002057 update_mmu_cache_pmd(vma, address, &entry);
2058 }
2059 spin_unlock(ptl);
Mel Gormana54a4072013-10-07 11:28:46 +01002060
Mel Gormaneb4489f62013-12-18 17:08:39 -08002061out_unlock:
Hugh Dickins340ef392013-02-22 16:34:33 -08002062 unlock_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00002063 put_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00002064 return 0;
2065}
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02002066#endif /* CONFIG_NUMA_BALANCING */
2067
2068#endif /* CONFIG_NUMA */