blob: b810ac1359f0962868106a47f04ba4d91ed1042f [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>
Linus Torvalds61332dc2017-08-20 13:26:27 -070043#include <linux/ptrace.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080044
Michal Nazarewicz0d1836c2010-12-21 17:24:26 -080045#include <asm/tlbflush.h>
46
Mel Gorman7b2a2d42012-10-19 14:07:31 +010047#define CREATE_TRACE_POINTS
48#include <trace/events/migrate.h>
49
Christoph Lameterb20a3502006-03-22 00:09:12 -080050#include "internal.h"
51
Christoph Lameterb20a3502006-03-22 00:09:12 -080052/*
Christoph Lameter742755a2006-06-23 02:03:55 -070053 * migrate_prep() needs to be called before we start compiling a list of pages
Mel Gorman748446b2010-05-24 14:32:27 -070054 * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is
55 * undesirable, use migrate_prep_local()
Christoph Lameterb20a3502006-03-22 00:09:12 -080056 */
57int migrate_prep(void)
58{
Christoph Lameterb20a3502006-03-22 00:09:12 -080059 /*
60 * Clear the LRU lists so pages can be isolated.
61 * Note that pages may be moved off the LRU after we have
62 * drained them. Those pages will fail to migrate like other
63 * pages that may be busy.
64 */
65 lru_add_drain_all();
66
67 return 0;
68}
69
Mel Gorman748446b2010-05-24 14:32:27 -070070/* Do the necessary work of migrate_prep but not if it involves other CPUs */
71int migrate_prep_local(void)
72{
73 lru_add_drain();
74
75 return 0;
76}
77
Minchan Kimbda807d2016-07-26 15:23:05 -070078bool isolate_movable_page(struct page *page, isolate_mode_t mode)
79{
80 struct address_space *mapping;
81
82 /*
83 * Avoid burning cycles with pages that are yet under __free_pages(),
84 * or just got freed under us.
85 *
86 * In case we 'win' a race for a movable page being freed under us and
87 * raise its refcount preventing __free_pages() from doing its job
88 * the put_page() at the end of this block will take care of
89 * release this page, thus avoiding a nasty leakage.
90 */
91 if (unlikely(!get_page_unless_zero(page)))
92 goto out;
93
94 /*
95 * Check PageMovable before holding a PG_lock because page's owner
96 * assumes anybody doesn't touch PG_lock of newly allocated page
97 * so unconditionally grapping the lock ruins page's owner side.
98 */
99 if (unlikely(!__PageMovable(page)))
100 goto out_putpage;
101 /*
102 * As movable pages are not isolated from LRU lists, concurrent
103 * compaction threads can race against page migration functions
104 * as well as race against the releasing a page.
105 *
106 * In order to avoid having an already isolated movable page
107 * being (wrongly) re-isolated while it is under migration,
108 * or to avoid attempting to isolate pages being released,
109 * lets be sure we have the page lock
110 * before proceeding with the movable page isolation steps.
111 */
112 if (unlikely(!trylock_page(page)))
113 goto out_putpage;
114
115 if (!PageMovable(page) || PageIsolated(page))
116 goto out_no_isolated;
117
118 mapping = page_mapping(page);
119 VM_BUG_ON_PAGE(!mapping, page);
120
121 if (!mapping->a_ops->isolate_page(page, mode))
122 goto out_no_isolated;
123
124 /* Driver shouldn't use PG_isolated bit of page->flags */
125 WARN_ON_ONCE(PageIsolated(page));
126 __SetPageIsolated(page);
127 unlock_page(page);
128
129 return true;
130
131out_no_isolated:
132 unlock_page(page);
133out_putpage:
134 put_page(page);
135out:
136 return false;
137}
138
139/* It should be called on page which is PG_movable */
140void putback_movable_page(struct page *page)
141{
142 struct address_space *mapping;
143
144 VM_BUG_ON_PAGE(!PageLocked(page), page);
145 VM_BUG_ON_PAGE(!PageMovable(page), page);
146 VM_BUG_ON_PAGE(!PageIsolated(page), page);
147
148 mapping = page_mapping(page);
149 mapping->a_ops->putback_page(page);
150 __ClearPageIsolated(page);
151}
152
Christoph Lameterb20a3502006-03-22 00:09:12 -0800153/*
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800154 * Put previously isolated pages back onto the appropriate lists
155 * from where they were once taken off for compaction/migration.
156 *
Joonsoo Kim59c82b72014-01-21 15:51:17 -0800157 * This function shall be used whenever the isolated pageset has been
158 * built from lru, balloon, hugetlbfs page. See isolate_migratepages_range()
159 * and isolate_huge_page().
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800160 */
161void putback_movable_pages(struct list_head *l)
162{
163 struct page *page;
164 struct page *page2;
165
166 list_for_each_entry_safe(page, page2, l, lru) {
Naoya Horiguchi31caf662013-09-11 14:21:59 -0700167 if (unlikely(PageHuge(page))) {
168 putback_active_hugepage(page);
169 continue;
170 }
Rafael Aquini5733c7d2012-12-11 16:02:47 -0800171 list_del(&page->lru);
Minchan Kimbda807d2016-07-26 15:23:05 -0700172 /*
173 * We isolated non-lru movable page so here we can use
174 * __PageMovable because LRU page's mapping cannot have
175 * PAGE_MAPPING_MOVABLE.
176 */
Minchan Kimb1123ea62016-07-26 15:23:09 -0700177 if (unlikely(__PageMovable(page))) {
Minchan Kimbda807d2016-07-26 15:23:05 -0700178 VM_BUG_ON_PAGE(!PageIsolated(page), page);
179 lock_page(page);
180 if (PageMovable(page))
181 putback_movable_page(page);
182 else
183 __ClearPageIsolated(page);
184 unlock_page(page);
185 put_page(page);
186 } else {
Ming Ling5d7d3622016-12-12 16:42:26 -0800187 dec_node_page_state(page, NR_ISOLATED_ANON +
188 page_is_file_cache(page));
Rabin Vincentc0a602ad2017-04-20 14:37:46 -0700189 putback_lru_page(page);
Minchan Kimbda807d2016-07-26 15:23:05 -0700190 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800191 }
Christoph Lameterb20a3502006-03-22 00:09:12 -0800192}
193
Christoph Lameter06972122006-06-23 02:03:35 -0700194/*
195 * Restore a potential migration pte to a working pte entry
196 */
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800197static int remove_migration_pte(struct page *new, struct vm_area_struct *vma,
198 unsigned long addr, void *old)
Christoph Lameter06972122006-06-23 02:03:35 -0700199{
200 struct mm_struct *mm = vma->vm_mm;
201 swp_entry_t entry;
Christoph Lameter06972122006-06-23 02:03:35 -0700202 pmd_t *pmd;
203 pte_t *ptep, pte;
204 spinlock_t *ptl;
205
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900206 if (unlikely(PageHuge(new))) {
207 ptep = huge_pte_offset(mm, addr);
208 if (!ptep)
209 goto out;
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800210 ptl = huge_pte_lockptr(hstate_vma(vma), mm, ptep);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900211 } else {
Bob Liu62190492012-12-11 16:00:37 -0800212 pmd = mm_find_pmd(mm, addr);
213 if (!pmd)
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900214 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700215
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900216 ptep = pte_offset_map(pmd, addr);
Christoph Lameter06972122006-06-23 02:03:35 -0700217
Hugh Dickins486cf462011-10-19 12:50:35 -0700218 /*
219 * Peek to check is_swap_pte() before taking ptlock? No, we
220 * can race mremap's move_ptes(), which skips anon_vma lock.
221 */
Christoph Lameter06972122006-06-23 02:03:35 -0700222
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900223 ptl = pte_lockptr(mm, pmd);
224 }
225
Christoph Lameter06972122006-06-23 02:03:35 -0700226 spin_lock(ptl);
227 pte = *ptep;
228 if (!is_swap_pte(pte))
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800229 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700230
231 entry = pte_to_swp_entry(pte);
232
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800233 if (!is_migration_entry(entry) ||
234 migration_entry_to_page(entry) != old)
235 goto unlock;
Christoph Lameter06972122006-06-23 02:03:35 -0700236
Christoph Lameter06972122006-06-23 02:03:35 -0700237 get_page(new);
Andrea Arcangeli6d2329f2016-10-07 17:01:22 -0700238 pte = pte_mkold(mk_pte(new, READ_ONCE(vma->vm_page_prot)));
Cyrill Gorcunovc3d16e12013-10-16 13:46:51 -0700239 if (pte_swp_soft_dirty(*ptep))
240 pte = pte_mksoft_dirty(pte);
Mel Gormand3cb8bf2014-10-02 19:47:41 +0100241
242 /* Recheck VMA as permissions can change since migration started */
Christoph Lameter06972122006-06-23 02:03:35 -0700243 if (is_write_migration_entry(entry))
Mel Gormand3cb8bf2014-10-02 19:47:41 +0100244 pte = maybe_mkwrite(pte, vma);
245
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200246#ifdef CONFIG_HUGETLB_PAGE
Tony Lube7517d2013-02-04 14:28:46 -0800247 if (PageHuge(new)) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900248 pte = pte_mkhuge(pte);
Tony Lube7517d2013-02-04 14:28:46 -0800249 pte = arch_make_huge_pte(pte, vma, new, 0);
250 }
Andi Kleen3ef8fd72010-10-11 16:03:21 +0200251#endif
Leonid Yegoshinc2cc4992013-05-24 15:55:18 -0700252 flush_dcache_page(new);
Christoph Lameter06972122006-06-23 02:03:35 -0700253 set_pte_at(mm, addr, ptep, pte);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700254
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900255 if (PageHuge(new)) {
256 if (PageAnon(new))
257 hugepage_add_anon_rmap(new, vma, addr);
258 else
Kirill A. Shutemov53f92632016-01-15 16:53:42 -0800259 page_dup_rmap(new, true);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900260 } else if (PageAnon(new))
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -0800261 page_add_anon_rmap(new, vma, addr, false);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700262 else
Kirill A. Shutemovdd78fed2016-07-26 15:25:26 -0700263 page_add_file_rmap(new, false);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700264
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700265 if (vma->vm_flags & VM_LOCKED && !PageTransCompound(new))
Hugh Dickins51afb122015-11-05 18:49:37 -0800266 mlock_vma_page(new);
267
Christoph Lameter04e62a22006-06-23 02:03:38 -0700268 /* No need to invalidate - it was non-present before */
Russell King4b3073e2009-12-18 16:40:18 +0000269 update_mmu_cache(vma, addr, ptep);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800270unlock:
Christoph Lameter06972122006-06-23 02:03:35 -0700271 pte_unmap_unlock(ptep, ptl);
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800272out:
273 return SWAP_AGAIN;
Christoph Lameter06972122006-06-23 02:03:35 -0700274}
275
276/*
Christoph Lameter04e62a22006-06-23 02:03:38 -0700277 * Get rid of all migration entries and replace them by
278 * references to the indicated page.
279 */
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700280void remove_migration_ptes(struct page *old, struct page *new, bool locked)
Christoph Lameter04e62a22006-06-23 02:03:38 -0700281{
Joonsoo Kim051ac832014-01-21 15:49:48 -0800282 struct rmap_walk_control rwc = {
283 .rmap_one = remove_migration_pte,
284 .arg = old,
285 };
286
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700287 if (locked)
288 rmap_walk_locked(new, &rwc);
289 else
290 rmap_walk(new, &rwc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700291}
292
293/*
Christoph Lameter06972122006-06-23 02:03:35 -0700294 * Something used the pte of a page under migration. We need to
295 * get to the page and wait until migration is finished.
296 * When we return from this function the fault will be retried.
Christoph Lameter06972122006-06-23 02:03:35 -0700297 */
Naoya Horiguchie66f17f2015-02-11 15:25:22 -0800298void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep,
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700299 spinlock_t *ptl)
Christoph Lameter06972122006-06-23 02:03:35 -0700300{
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700301 pte_t pte;
Christoph Lameter06972122006-06-23 02:03:35 -0700302 swp_entry_t entry;
303 struct page *page;
304
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700305 spin_lock(ptl);
Christoph Lameter06972122006-06-23 02:03:35 -0700306 pte = *ptep;
307 if (!is_swap_pte(pte))
308 goto out;
309
310 entry = pte_to_swp_entry(pte);
311 if (!is_migration_entry(entry))
312 goto out;
313
314 page = migration_entry_to_page(entry);
315
Nick Piggine2867812008-07-25 19:45:30 -0700316 /*
317 * Once radix-tree replacement of page migration started, page_count
318 * *must* be zero. And, we don't want to call wait_on_page_locked()
319 * against a page without get_page().
320 * So, we use get_page_unless_zero(), here. Even failed, page fault
321 * will occur again.
322 */
323 if (!get_page_unless_zero(page))
324 goto out;
Christoph Lameter06972122006-06-23 02:03:35 -0700325 pte_unmap_unlock(ptep, ptl);
326 wait_on_page_locked(page);
327 put_page(page);
328 return;
329out:
330 pte_unmap_unlock(ptep, ptl);
331}
332
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700333void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd,
334 unsigned long address)
335{
336 spinlock_t *ptl = pte_lockptr(mm, pmd);
337 pte_t *ptep = pte_offset_map(pmd, address);
338 __migration_entry_wait(mm, ptep, ptl);
339}
340
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800341void migration_entry_wait_huge(struct vm_area_struct *vma,
342 struct mm_struct *mm, pte_t *pte)
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700343{
Kirill A. Shutemovcb900f42013-11-14 14:31:02 -0800344 spinlock_t *ptl = huge_pte_lockptr(hstate_vma(vma), mm, pte);
Naoya Horiguchi30dad302013-06-12 14:05:04 -0700345 __migration_entry_wait(mm, pte, ptl);
346}
347
Mel Gormanb969c4a2012-01-12 17:19:34 -0800348#ifdef CONFIG_BLOCK
349/* Returns true if all buffers are successfully locked */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800350static bool buffer_migrate_lock_buffers(struct buffer_head *head,
351 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800352{
353 struct buffer_head *bh = head;
354
355 /* Simple case, sync compaction */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800356 if (mode != MIGRATE_ASYNC) {
Mel Gormanb969c4a2012-01-12 17:19:34 -0800357 do {
358 get_bh(bh);
359 lock_buffer(bh);
360 bh = bh->b_this_page;
361
362 } while (bh != head);
363
364 return true;
365 }
366
367 /* async case, we cannot block on lock_buffer so use trylock_buffer */
368 do {
369 get_bh(bh);
370 if (!trylock_buffer(bh)) {
371 /*
372 * We failed to lock the buffer and cannot stall in
373 * async migration. Release the taken locks
374 */
375 struct buffer_head *failed_bh = bh;
376 put_bh(failed_bh);
377 bh = head;
378 while (bh != failed_bh) {
379 unlock_buffer(bh);
380 put_bh(bh);
381 bh = bh->b_this_page;
382 }
383 return false;
384 }
385
386 bh = bh->b_this_page;
387 } while (bh != head);
388 return true;
389}
390#else
391static inline bool buffer_migrate_lock_buffers(struct buffer_head *head,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800392 enum migrate_mode mode)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800393{
394 return true;
395}
396#endif /* CONFIG_BLOCK */
397
Christoph Lameterb20a3502006-03-22 00:09:12 -0800398/*
Christoph Lameterc3fcf8a2006-06-23 02:03:32 -0700399 * Replace the page in the mapping.
Christoph Lameter5b5c7122006-06-23 02:03:29 -0700400 *
401 * The number of remaining references must be:
402 * 1 for anonymous pages without a mapping
403 * 2 for pages with a mapping
David Howells266cf652009-04-03 16:42:36 +0100404 * 3 for pages with a mapping and PagePrivate/PagePrivate2 set.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800405 */
Gu Zheng36bc08c2013-07-16 17:56:16 +0800406int migrate_page_move_mapping(struct address_space *mapping,
Mel Gormanb969c4a2012-01-12 17:19:34 -0800407 struct page *newpage, struct page *page,
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500408 struct buffer_head *head, enum migrate_mode mode,
409 int extra_count)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800410{
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800411 struct zone *oldzone, *newzone;
412 int dirty;
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500413 int expected_count = 1 + extra_count;
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800414 void **pslot;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800415
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700416 if (!mapping) {
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700417 /* Anonymous page without mapping */
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500418 if (page_count(page) != expected_count)
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700419 return -EAGAIN;
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800420
421 /* No turning back from here */
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800422 newpage->index = page->index;
423 newpage->mapping = page->mapping;
424 if (PageSwapBacked(page))
Hugh Dickinsfa9949d2016-05-19 17:12:41 -0700425 __SetPageSwapBacked(newpage);
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800426
Rafael Aquini78bd5202012-12-11 16:02:31 -0800427 return MIGRATEPAGE_SUCCESS;
Christoph Lameter6c5240a2006-06-23 02:03:37 -0700428 }
429
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800430 oldzone = page_zone(page);
431 newzone = page_zone(newpage);
432
Nick Piggin19fd6232008-07-25 19:45:32 -0700433 spin_lock_irq(&mapping->tree_lock);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800434
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800435 pslot = radix_tree_lookup_slot(&mapping->page_tree,
436 page_index(page));
Christoph Lameterb20a3502006-03-22 00:09:12 -0800437
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500438 expected_count += 1 + page_has_private(page);
Nick Piggine2867812008-07-25 19:45:30 -0700439 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800440 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700441 spin_unlock_irq(&mapping->tree_lock);
Christoph Lametere23ca002006-04-10 22:52:57 -0700442 return -EAGAIN;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800443 }
444
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700445 if (!page_ref_freeze(page, expected_count)) {
Nick Piggin19fd6232008-07-25 19:45:32 -0700446 spin_unlock_irq(&mapping->tree_lock);
Nick Piggine2867812008-07-25 19:45:30 -0700447 return -EAGAIN;
448 }
449
Christoph Lameterb20a3502006-03-22 00:09:12 -0800450 /*
Mel Gormanb969c4a2012-01-12 17:19:34 -0800451 * In the async migration case of moving a page with buffers, lock the
452 * buffers using trylock before the mapping is moved. If the mapping
453 * was moved, we later failed to lock the buffers and could not move
454 * the mapping back due to an elevated page count, we would have to
455 * block waiting on other references to be dropped.
456 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800457 if (mode == MIGRATE_ASYNC && head &&
458 !buffer_migrate_lock_buffers(head, mode)) {
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700459 page_ref_unfreeze(page, expected_count);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800460 spin_unlock_irq(&mapping->tree_lock);
461 return -EAGAIN;
462 }
463
464 /*
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800465 * Now we know that no one else is looking at the page:
466 * no turning back from here.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800467 */
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800468 newpage->index = page->index;
469 newpage->mapping = page->mapping;
470 if (PageSwapBacked(page))
Hugh Dickinsfa9949d2016-05-19 17:12:41 -0700471 __SetPageSwapBacked(newpage);
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800472
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800473 get_page(newpage); /* add cache reference */
Christoph Lameterb20a3502006-03-22 00:09:12 -0800474 if (PageSwapCache(page)) {
475 SetPageSwapCache(newpage);
476 set_page_private(newpage, page_private(page));
477 }
478
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800479 /* Move dirty while page refs frozen and newpage not yet exposed */
480 dirty = PageDirty(page);
481 if (dirty) {
482 ClearPageDirty(page);
483 SetPageDirty(newpage);
484 }
485
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800486 radix_tree_replace_slot(pslot, newpage);
487
488 /*
Jacobo Giralt937a94c2012-01-10 15:07:11 -0800489 * Drop cache reference from old page by unfreezing
490 * to one less reference.
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800491 * We know this isn't the last reference.
492 */
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700493 page_ref_unfreeze(page, expected_count - 1);
Nick Piggin7cf9c2c2006-12-06 20:33:44 -0800494
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800495 spin_unlock(&mapping->tree_lock);
496 /* Leave irq disabled to prevent preemption while updating stats */
497
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700498 /*
499 * If moved to a different zone then also account
500 * the page for that zone. Other VM counters will be
501 * taken care of when we establish references to the
502 * new page and drop references to the old page.
503 *
504 * Note that anonymous pages are accounted for
Mel Gorman4b9d0fa2016-07-28 15:46:17 -0700505 * via NR_FILE_PAGES and NR_ANON_MAPPED if they
Christoph Lameter0e8c7d02007-04-23 14:41:09 -0700506 * are mapped to swap space.
507 */
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800508 if (newzone != oldzone) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700509 __dec_node_state(oldzone->zone_pgdat, NR_FILE_PAGES);
510 __inc_node_state(newzone->zone_pgdat, NR_FILE_PAGES);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800511 if (PageSwapBacked(page) && !PageSwapCache(page)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700512 __dec_node_state(oldzone->zone_pgdat, NR_SHMEM);
513 __inc_node_state(newzone->zone_pgdat, NR_SHMEM);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800514 }
515 if (dirty && mapping_cap_account_dirty(mapping)) {
Mel Gorman11fb9982016-07-28 15:46:20 -0700516 __dec_node_state(oldzone->zone_pgdat, NR_FILE_DIRTY);
Mel Gorman5a1c84b2016-07-28 15:47:31 -0700517 __dec_zone_state(oldzone, NR_ZONE_WRITE_PENDING);
Mel Gorman11fb9982016-07-28 15:46:20 -0700518 __inc_node_state(newzone->zone_pgdat, NR_FILE_DIRTY);
Mel Gorman5a1c84b2016-07-28 15:47:31 -0700519 __inc_zone_state(newzone, NR_ZONE_WRITE_PENDING);
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800520 }
KOSAKI Motohiro4b021082009-09-21 17:01:33 -0700521 }
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800522 local_irq_enable();
Christoph Lameterb20a3502006-03-22 00:09:12 -0800523
Rafael Aquini78bd5202012-12-11 16:02:31 -0800524 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800525}
Richard Weinberger1118dce2016-06-16 23:26:14 +0200526EXPORT_SYMBOL(migrate_page_move_mapping);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800527
528/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900529 * The expected number of remaining references is the same as that
530 * of migrate_page_move_mapping().
531 */
532int migrate_huge_page_move_mapping(struct address_space *mapping,
533 struct page *newpage, struct page *page)
534{
535 int expected_count;
536 void **pslot;
537
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900538 spin_lock_irq(&mapping->tree_lock);
539
540 pslot = radix_tree_lookup_slot(&mapping->page_tree,
541 page_index(page));
542
543 expected_count = 2 + page_has_private(page);
544 if (page_count(page) != expected_count ||
Mel Gorman29c1f672011-01-13 15:47:21 -0800545 radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900546 spin_unlock_irq(&mapping->tree_lock);
547 return -EAGAIN;
548 }
549
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700550 if (!page_ref_freeze(page, expected_count)) {
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900551 spin_unlock_irq(&mapping->tree_lock);
552 return -EAGAIN;
553 }
554
Hugh Dickinscf4b7692015-11-05 18:50:02 -0800555 newpage->index = page->index;
556 newpage->mapping = page->mapping;
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700557
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900558 get_page(newpage);
559
560 radix_tree_replace_slot(pslot, newpage);
561
Joonsoo Kimfe896d12016-03-17 14:19:26 -0700562 page_ref_unfreeze(page, expected_count - 1);
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900563
564 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6a93ca82016-03-15 14:57:19 -0700565
Rafael Aquini78bd5202012-12-11 16:02:31 -0800566 return MIGRATEPAGE_SUCCESS;
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900567}
568
569/*
Dave Hansen30b0a102013-11-21 14:31:58 -0800570 * Gigantic pages are so large that we do not guarantee that page++ pointer
571 * arithmetic will work across the entire page. We need something more
572 * specialized.
573 */
574static void __copy_gigantic_page(struct page *dst, struct page *src,
575 int nr_pages)
576{
577 int i;
578 struct page *dst_base = dst;
579 struct page *src_base = src;
580
581 for (i = 0; i < nr_pages; ) {
582 cond_resched();
583 copy_highpage(dst, src);
584
585 i++;
586 dst = mem_map_next(dst, dst_base, i);
587 src = mem_map_next(src, src_base, i);
588 }
589}
590
591static void copy_huge_page(struct page *dst, struct page *src)
592{
593 int i;
594 int nr_pages;
595
596 if (PageHuge(src)) {
597 /* hugetlbfs page */
598 struct hstate *h = page_hstate(src);
599 nr_pages = pages_per_huge_page(h);
600
601 if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) {
602 __copy_gigantic_page(dst, src, nr_pages);
603 return;
604 }
605 } else {
606 /* thp page */
607 BUG_ON(!PageTransHuge(src));
608 nr_pages = hpage_nr_pages(src);
609 }
610
611 for (i = 0; i < nr_pages; i++) {
612 cond_resched();
613 copy_highpage(dst + i, src + i);
614 }
615}
616
617/*
Christoph Lameterb20a3502006-03-22 00:09:12 -0800618 * Copy the page to its new location
619 */
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900620void migrate_page_copy(struct page *newpage, struct page *page)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800621{
Rik van Riel7851a452013-10-07 11:29:23 +0100622 int cpupid;
623
Mel Gormanb32967f2012-11-19 12:35:47 +0000624 if (PageHuge(page) || PageTransHuge(page))
Naoya Horiguchi290408d2010-09-08 10:19:35 +0900625 copy_huge_page(newpage, page);
626 else
627 copy_highpage(newpage, page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800628
629 if (PageError(page))
630 SetPageError(newpage);
631 if (PageReferenced(page))
632 SetPageReferenced(newpage);
633 if (PageUptodate(page))
634 SetPageUptodate(newpage);
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700635 if (TestClearPageActive(page)) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800636 VM_BUG_ON_PAGE(PageUnevictable(page), page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800637 SetPageActive(newpage);
Lee Schermerhorn418b27e2009-12-14 17:59:54 -0800638 } else if (TestClearPageUnevictable(page))
639 SetPageUnevictable(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800640 if (PageChecked(page))
641 SetPageChecked(newpage);
642 if (PageMappedToDisk(page))
643 SetPageMappedToDisk(newpage);
644
Hugh Dickins42cb14b2015-11-05 18:50:05 -0800645 /* Move dirty on pages not done by migrate_page_move_mapping() */
646 if (PageDirty(page))
647 SetPageDirty(newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800648
Vladimir Davydov33c3fc72015-09-09 15:35:45 -0700649 if (page_is_young(page))
650 set_page_young(newpage);
651 if (page_is_idle(page))
652 set_page_idle(newpage);
653
Rik van Riel7851a452013-10-07 11:29:23 +0100654 /*
655 * Copy NUMA information to the new page, to prevent over-eager
656 * future migrations of this same page.
657 */
658 cpupid = page_cpupid_xchg_last(page, -1);
659 page_cpupid_xchg_last(newpage, cpupid);
660
Hugh Dickinse9995ef2009-12-14 17:59:31 -0800661 ksm_migrate_page(newpage, page);
Hugh Dickinsc8d65532013-02-22 16:35:10 -0800662 /*
663 * Please do not reorder this without considering how mm/ksm.c's
664 * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache().
665 */
Naoya Horiguchib3b3a992015-04-15 16:13:15 -0700666 if (PageSwapCache(page))
667 ClearPageSwapCache(page);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800668 ClearPagePrivate(page);
669 set_page_private(page, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800670
671 /*
672 * If any waiters have accumulated on the new page then
673 * wake them up.
674 */
675 if (PageWriteback(newpage))
676 end_page_writeback(newpage);
Vlastimil Babkad435edc2016-03-15 14:56:15 -0700677
678 copy_page_owner(page, newpage);
Johannes Weiner74485cf2016-03-15 14:57:54 -0700679
680 mem_cgroup_migrate(page, newpage);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800681}
Richard Weinberger1118dce2016-06-16 23:26:14 +0200682EXPORT_SYMBOL(migrate_page_copy);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800683
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700684/************************************************************
685 * Migration functions
686 ***********************************************************/
687
Christoph Lameterb20a3502006-03-22 00:09:12 -0800688/*
Minchan Kimbda807d2016-07-26 15:23:05 -0700689 * Common logic to directly migrate a single LRU page suitable for
David Howells266cf652009-04-03 16:42:36 +0100690 * pages that do not use PagePrivate/PagePrivate2.
Christoph Lameterb20a3502006-03-22 00:09:12 -0800691 *
692 * Pages are locked upon entry and exit.
693 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700694int migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800695 struct page *newpage, struct page *page,
696 enum migrate_mode mode)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800697{
698 int rc;
699
700 BUG_ON(PageWriteback(page)); /* Writeback must be complete */
701
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500702 rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0);
Christoph Lameterb20a3502006-03-22 00:09:12 -0800703
Rafael Aquini78bd5202012-12-11 16:02:31 -0800704 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameterb20a3502006-03-22 00:09:12 -0800705 return rc;
706
707 migrate_page_copy(newpage, page);
Rafael Aquini78bd5202012-12-11 16:02:31 -0800708 return MIGRATEPAGE_SUCCESS;
Christoph Lameterb20a3502006-03-22 00:09:12 -0800709}
710EXPORT_SYMBOL(migrate_page);
711
David Howells93614012006-09-30 20:45:40 +0200712#ifdef CONFIG_BLOCK
Christoph Lameterb20a3502006-03-22 00:09:12 -0800713/*
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700714 * Migration function for pages with buffers. This function can only be used
715 * if the underlying filesystem guarantees that no other references to "page"
716 * exist.
717 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -0700718int buffer_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800719 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700720{
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700721 struct buffer_head *bh, *head;
722 int rc;
723
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700724 if (!page_has_buffers(page))
Mel Gormana6bc32b2012-01-12 17:19:43 -0800725 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700726
727 head = page_buffers(page);
728
Benjamin LaHaise8e321fe2013-12-21 17:56:08 -0500729 rc = migrate_page_move_mapping(mapping, newpage, page, head, mode, 0);
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700730
Rafael Aquini78bd5202012-12-11 16:02:31 -0800731 if (rc != MIGRATEPAGE_SUCCESS)
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700732 return rc;
733
Mel Gormanb969c4a2012-01-12 17:19:34 -0800734 /*
735 * In the async case, migrate_page_move_mapping locked the buffers
736 * with an IRQ-safe spinlock held. In the sync case, the buffers
737 * need to be locked now
738 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800739 if (mode != MIGRATE_ASYNC)
740 BUG_ON(!buffer_migrate_lock_buffers(head, mode));
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700741
742 ClearPagePrivate(page);
743 set_page_private(newpage, page_private(page));
744 set_page_private(page, 0);
745 put_page(page);
746 get_page(newpage);
747
748 bh = head;
749 do {
750 set_bh_page(bh, newpage, bh_offset(bh));
751 bh = bh->b_this_page;
752
753 } while (bh != head);
754
755 SetPagePrivate(newpage);
756
757 migrate_page_copy(newpage, page);
758
759 bh = head;
760 do {
761 unlock_buffer(bh);
762 put_bh(bh);
763 bh = bh->b_this_page;
764
765 } while (bh != head);
766
Rafael Aquini78bd5202012-12-11 16:02:31 -0800767 return MIGRATEPAGE_SUCCESS;
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700768}
769EXPORT_SYMBOL(buffer_migrate_page);
David Howells93614012006-09-30 20:45:40 +0200770#endif
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700771
Christoph Lameter04e62a22006-06-23 02:03:38 -0700772/*
773 * Writeback a page to clean the dirty state
774 */
775static int writeout(struct address_space *mapping, struct page *page)
776{
777 struct writeback_control wbc = {
778 .sync_mode = WB_SYNC_NONE,
779 .nr_to_write = 1,
780 .range_start = 0,
781 .range_end = LLONG_MAX,
Christoph Lameter04e62a22006-06-23 02:03:38 -0700782 .for_reclaim = 1
783 };
784 int rc;
785
786 if (!mapping->a_ops->writepage)
787 /* No write method for the address space */
788 return -EINVAL;
789
790 if (!clear_page_dirty_for_io(page))
791 /* Someone else already triggered a write */
792 return -EAGAIN;
793
794 /*
795 * A dirty page may imply that the underlying filesystem has
796 * the page on some queue. So the page must be clean for
797 * migration. Writeout may mean we loose the lock and the
798 * page state is no longer what we checked for earlier.
799 * At this point we know that the migration attempt cannot
800 * be successful.
801 */
Kirill A. Shutemove3884662016-03-17 14:20:07 -0700802 remove_migration_ptes(page, page, false);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700803
804 rc = mapping->a_ops->writepage(page, &wbc);
Christoph Lameter04e62a22006-06-23 02:03:38 -0700805
806 if (rc != AOP_WRITEPAGE_ACTIVATE)
807 /* unlocked. Relock */
808 lock_page(page);
809
Hugh Dickinsbda85502008-11-19 15:36:36 -0800810 return (rc < 0) ? -EIO : -EAGAIN;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700811}
812
813/*
814 * Default handling if a filesystem does not provide a migration function.
815 */
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700816static int fallback_migrate_page(struct address_space *mapping,
Mel Gormana6bc32b2012-01-12 17:19:43 -0800817 struct page *newpage, struct page *page, enum migrate_mode mode)
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700818{
Mel Gormanb969c4a2012-01-12 17:19:34 -0800819 if (PageDirty(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800820 /* Only writeback pages in full synchronous migration */
821 if (mode != MIGRATE_SYNC)
Mel Gormanb969c4a2012-01-12 17:19:34 -0800822 return -EBUSY;
Christoph Lameter04e62a22006-06-23 02:03:38 -0700823 return writeout(mapping, page);
Mel Gormanb969c4a2012-01-12 17:19:34 -0800824 }
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700825
826 /*
827 * Buffers may be managed in a filesystem specific way.
828 * We must have no buffers or drop them.
829 */
David Howells266cf652009-04-03 16:42:36 +0100830 if (page_has_private(page) &&
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700831 !try_to_release_page(page, GFP_KERNEL))
832 return -EAGAIN;
833
Mel Gormana6bc32b2012-01-12 17:19:43 -0800834 return migrate_page(mapping, newpage, page, mode);
Christoph Lameter8351a6e2006-06-23 02:03:33 -0700835}
836
Christoph Lameter1d8b85c2006-06-23 02:03:28 -0700837/*
Christoph Lametere24f0b82006-06-23 02:03:51 -0700838 * Move a page to a newly allocated page
839 * The page is locked and all ptes have been successfully removed.
840 *
841 * The new page will have replaced the old page if this function
842 * is successful.
Lee Schermerhorn894bc312008-10-18 20:26:39 -0700843 *
844 * Return value:
845 * < 0 - error code
Rafael Aquini78bd5202012-12-11 16:02:31 -0800846 * MIGRATEPAGE_SUCCESS - success
Christoph Lametere24f0b82006-06-23 02:03:51 -0700847 */
Mel Gorman3fe20112010-05-24 14:32:20 -0700848static int move_to_new_page(struct page *newpage, struct page *page,
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800849 enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700850{
851 struct address_space *mapping;
Minchan Kimbda807d2016-07-26 15:23:05 -0700852 int rc = -EAGAIN;
853 bool is_lru = !__PageMovable(page);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700854
Hugh Dickins7db76712015-11-05 18:49:49 -0800855 VM_BUG_ON_PAGE(!PageLocked(page), page);
856 VM_BUG_ON_PAGE(!PageLocked(newpage), newpage);
Christoph Lametere24f0b82006-06-23 02:03:51 -0700857
Christoph Lametere24f0b82006-06-23 02:03:51 -0700858 mapping = page_mapping(page);
Minchan Kimbda807d2016-07-26 15:23:05 -0700859
860 if (likely(is_lru)) {
861 if (!mapping)
862 rc = migrate_page(mapping, newpage, page, mode);
863 else if (mapping->a_ops->migratepage)
864 /*
865 * Most pages have a mapping and most filesystems
866 * provide a migratepage callback. Anonymous pages
867 * are part of swap space which also has its own
868 * migratepage callback. This is the most common path
869 * for page migration.
870 */
871 rc = mapping->a_ops->migratepage(mapping, newpage,
872 page, mode);
873 else
874 rc = fallback_migrate_page(mapping, newpage,
875 page, mode);
876 } else {
Christoph Lametere24f0b82006-06-23 02:03:51 -0700877 /*
Minchan Kimbda807d2016-07-26 15:23:05 -0700878 * In case of non-lru page, it could be released after
879 * isolation step. In that case, we shouldn't try migration.
Christoph Lametere24f0b82006-06-23 02:03:51 -0700880 */
Minchan Kimbda807d2016-07-26 15:23:05 -0700881 VM_BUG_ON_PAGE(!PageIsolated(page), page);
882 if (!PageMovable(page)) {
883 rc = MIGRATEPAGE_SUCCESS;
884 __ClearPageIsolated(page);
885 goto out;
886 }
887
888 rc = mapping->a_ops->migratepage(mapping, newpage,
889 page, mode);
890 WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS &&
891 !PageIsolated(page));
892 }
Christoph Lametere24f0b82006-06-23 02:03:51 -0700893
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800894 /*
895 * When successful, old pagecache page->mapping must be cleared before
896 * page is freed; but stats require that PageAnon be left as PageAnon.
897 */
898 if (rc == MIGRATEPAGE_SUCCESS) {
Minchan Kimbda807d2016-07-26 15:23:05 -0700899 if (__PageMovable(page)) {
900 VM_BUG_ON_PAGE(!PageIsolated(page), page);
901
902 /*
903 * We clear PG_movable under page_lock so any compactor
904 * cannot try to migrate this page.
905 */
906 __ClearPageIsolated(page);
907 }
908
909 /*
910 * Anonymous and movable page->mapping will be cleard by
911 * free_pages_prepare so don't reset it here for keeping
912 * the type to work PageAnon, for example.
913 */
914 if (!PageMappingFlags(page))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -0800915 page->mapping = NULL;
Mel Gorman3fe20112010-05-24 14:32:20 -0700916 }
Minchan Kimbda807d2016-07-26 15:23:05 -0700917out:
Christoph Lametere24f0b82006-06-23 02:03:51 -0700918 return rc;
919}
920
Minchan Kim0dabec92011-10-31 17:06:57 -0700921static int __unmap_and_move(struct page *page, struct page *newpage,
Hugh Dickins9c620e22013-02-22 16:35:14 -0800922 int force, enum migrate_mode mode)
Christoph Lametere24f0b82006-06-23 02:03:51 -0700923{
Minchan Kim0dabec92011-10-31 17:06:57 -0700924 int rc = -EAGAIN;
Hugh Dickins2ebba6b2014-12-12 16:56:19 -0800925 int page_was_mapped = 0;
Mel Gorman3f6c8272010-05-24 14:32:17 -0700926 struct anon_vma *anon_vma = NULL;
Minchan Kimbda807d2016-07-26 15:23:05 -0700927 bool is_lru = !__PageMovable(page);
Christoph Lameter95a402c2006-06-23 02:03:53 -0700928
Nick Piggin529ae9a2008-08-02 12:01:03 +0200929 if (!trylock_page(page)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -0800930 if (!force || mode == MIGRATE_ASYNC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700931 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800932
933 /*
934 * It's not safe for direct compaction to call lock_page.
935 * For example, during page readahead pages are added locked
936 * to the LRU. Later, when the IO completes the pages are
937 * marked uptodate and unlocked. However, the queueing
938 * could be merging multiple pages for one bio (e.g.
939 * mpage_readpages). If an allocation happens for the
940 * second or third page, the process can end up locking
941 * the same page twice and deadlocking. Rather than
942 * trying to be clever about what pages can be locked,
943 * avoid the use of lock_page for direct compaction
944 * altogether.
945 */
946 if (current->flags & PF_MEMALLOC)
Minchan Kim0dabec92011-10-31 17:06:57 -0700947 goto out;
Mel Gorman3e7d3442011-01-13 15:45:56 -0800948
Christoph Lametere24f0b82006-06-23 02:03:51 -0700949 lock_page(page);
950 }
951
952 if (PageWriteback(page)) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700953 /*
Jianguo Wufed5b642013-04-29 15:07:58 -0700954 * Only in the case of a full synchronous migration is it
Mel Gormana6bc32b2012-01-12 17:19:43 -0800955 * necessary to wait for PageWriteback. In the async case,
956 * the retry loop is too short and in the sync-light case,
957 * the overhead of stalling is too much
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700958 */
Mel Gormana6bc32b2012-01-12 17:19:43 -0800959 if (mode != MIGRATE_SYNC) {
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700960 rc = -EBUSY;
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700961 goto out_unlock;
Andrea Arcangeli11bc82d2011-03-22 16:33:11 -0700962 }
963 if (!force)
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700964 goto out_unlock;
Christoph Lametere24f0b82006-06-23 02:03:51 -0700965 wait_on_page_writeback(page);
966 }
Hugh Dickins03f15c82015-11-05 18:49:56 -0800967
Christoph Lametere24f0b82006-06-23 02:03:51 -0700968 /*
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700969 * By try_to_unmap(), page->mapcount goes down to 0 here. In this case,
970 * we cannot notice that anon_vma is freed while we migrates a page.
Hugh Dickins1ce82b62011-01-13 15:47:30 -0800971 * This get_anon_vma() delays freeing anon_vma pointer until the end
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -0700972 * of migration. File cache pages are no problem because of page_lock()
KAMEZAWA Hiroyuki989f89c2007-08-30 23:56:21 -0700973 * File Caches may use write_page() or lock_page() in migration, then,
974 * just care Anon page here.
Hugh Dickins03f15c82015-11-05 18:49:56 -0800975 *
976 * Only page_get_anon_vma() understands the subtleties of
977 * getting a hold on an anon_vma from outside one of its mms.
978 * But if we cannot get anon_vma, then we won't need it anyway,
979 * because that implies that the anon page is no longer mapped
980 * (and cannot be remapped so long as we hold the page lock).
Christoph Lametere24f0b82006-06-23 02:03:51 -0700981 */
Hugh Dickins03f15c82015-11-05 18:49:56 -0800982 if (PageAnon(page) && !PageKsm(page))
Peter Zijlstra746b18d2011-05-24 17:12:10 -0700983 anon_vma = page_get_anon_vma(page);
Shaohua Li62e1c552008-02-04 22:29:33 -0800984
Hugh Dickins7db76712015-11-05 18:49:49 -0800985 /*
986 * Block others from accessing the new page when we get around to
987 * establishing additional references. We are usually the only one
988 * holding a reference to newpage at this point. We used to have a BUG
989 * here if trylock_page(newpage) fails, but would like to allow for
990 * cases where there might be a race with the previous use of newpage.
991 * This is much like races on refcount of oldpage: just don't BUG().
992 */
993 if (unlikely(!trylock_page(newpage)))
994 goto out_unlock;
995
Minchan Kimbda807d2016-07-26 15:23:05 -0700996 if (unlikely(!is_lru)) {
997 rc = move_to_new_page(newpage, page, mode);
998 goto out_unlock_both;
999 }
1000
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001001 /*
Shaohua Li62e1c552008-02-04 22:29:33 -08001002 * Corner case handling:
1003 * 1. When a new swap-cache page is read into, it is added to the LRU
1004 * and treated as swapcache but it has no rmap yet.
1005 * Calling try_to_unmap() against a page->mapping==NULL page will
1006 * trigger a BUG. So handle it here.
1007 * 2. An orphaned page (see truncate_complete_page) might have
1008 * fs-private metadata. The page can be picked up due to memory
1009 * offlining. Everywhere else except page reclaim, the page is
1010 * invisible to the vm, so the page can not be migrated. So try to
1011 * free the metadata, so the page can be freed.
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001012 */
Shaohua Li62e1c552008-02-04 22:29:33 -08001013 if (!page->mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -08001014 VM_BUG_ON_PAGE(PageAnon(page), page);
Hugh Dickins1ce82b62011-01-13 15:47:30 -08001015 if (page_has_private(page)) {
Shaohua Li62e1c552008-02-04 22:29:33 -08001016 try_to_free_buffers(page);
Hugh Dickins7db76712015-11-05 18:49:49 -08001017 goto out_unlock_both;
Shaohua Li62e1c552008-02-04 22:29:33 -08001018 }
Hugh Dickins7db76712015-11-05 18:49:49 -08001019 } else if (page_mapped(page)) {
1020 /* Establish migration ptes */
Hugh Dickins03f15c82015-11-05 18:49:56 -08001021 VM_BUG_ON_PAGE(PageAnon(page) && !PageKsm(page) && !anon_vma,
1022 page);
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001023 try_to_unmap(page,
Wanpeng Lida1b13c2015-09-08 15:03:27 -07001024 TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001025 page_was_mapped = 1;
1026 }
KAMEZAWA Hiroyukidc386d42007-07-26 10:41:07 -07001027
Christoph Lametere6a15302006-06-25 05:46:49 -07001028 if (!page_mapped(page))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001029 rc = move_to_new_page(newpage, page, mode);
Christoph Lametere24f0b82006-06-23 02:03:51 -07001030
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001031 if (page_was_mapped)
1032 remove_migration_ptes(page,
Kirill A. Shutemove3884662016-03-17 14:20:07 -07001033 rc == MIGRATEPAGE_SUCCESS ? newpage : page, false);
Mel Gorman3f6c8272010-05-24 14:32:17 -07001034
Hugh Dickins7db76712015-11-05 18:49:49 -08001035out_unlock_both:
1036 unlock_page(newpage);
1037out_unlock:
Mel Gorman3f6c8272010-05-24 14:32:17 -07001038 /* Drop an anon_vma reference if we took one */
Rik van Riel76545062010-08-09 17:18:41 -07001039 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -07001040 put_anon_vma(anon_vma);
Christoph Lametere24f0b82006-06-23 02:03:51 -07001041 unlock_page(page);
Minchan Kim0dabec92011-10-31 17:06:57 -07001042out:
Minchan Kimc6c919e2016-07-26 15:23:02 -07001043 /*
1044 * If migration is successful, decrease refcount of the newpage
1045 * which will not free the page because new page owner increased
1046 * refcounter. As well, if it is LRU page, add the page to LRU
David Hildenbrandd2de58e2019-02-01 14:21:19 -08001047 * list in here. Use the old state of the isolated source page to
1048 * determine if we migrated a LRU page. newpage was already unlocked
1049 * and possibly modified by its owner - don't rely on the page
1050 * state.
Minchan Kimc6c919e2016-07-26 15:23:02 -07001051 */
1052 if (rc == MIGRATEPAGE_SUCCESS) {
David Hildenbrandd2de58e2019-02-01 14:21:19 -08001053 if (unlikely(!is_lru))
Minchan Kimc6c919e2016-07-26 15:23:02 -07001054 put_page(newpage);
1055 else
1056 putback_lru_page(newpage);
1057 }
1058
Minchan Kim0dabec92011-10-31 17:06:57 -07001059 return rc;
1060}
Christoph Lameter95a402c2006-06-23 02:03:53 -07001061
Minchan Kim0dabec92011-10-31 17:06:57 -07001062/*
Geert Uytterhoevenef2a5152015-04-14 15:44:22 -07001063 * gcc 4.7 and 4.8 on arm get an ICEs when inlining unmap_and_move(). Work
1064 * around it.
1065 */
1066#if (GCC_VERSION >= 40700 && GCC_VERSION < 40900) && defined(CONFIG_ARM)
1067#define ICE_noinline noinline
1068#else
1069#define ICE_noinline
1070#endif
1071
1072/*
Minchan Kim0dabec92011-10-31 17:06:57 -07001073 * Obtain the lock on page, remove all ptes and migrate the page
1074 * to the newly allocated page in newpage.
1075 */
Geert Uytterhoevenef2a5152015-04-14 15:44:22 -07001076static ICE_noinline int unmap_and_move(new_page_t get_new_page,
1077 free_page_t put_new_page,
1078 unsigned long private, struct page *page,
Naoya Horiguchiadd05ce2015-06-24 16:56:50 -07001079 int force, enum migrate_mode mode,
1080 enum migrate_reason reason)
Minchan Kim0dabec92011-10-31 17:06:57 -07001081{
Hugh Dickins2def7422015-11-05 18:49:46 -08001082 int rc = MIGRATEPAGE_SUCCESS;
Minchan Kim0dabec92011-10-31 17:06:57 -07001083 int *result = NULL;
Hugh Dickins2def7422015-11-05 18:49:46 -08001084 struct page *newpage;
Minchan Kim0dabec92011-10-31 17:06:57 -07001085
Hugh Dickins2def7422015-11-05 18:49:46 -08001086 newpage = get_new_page(page, private, &result);
Minchan Kim0dabec92011-10-31 17:06:57 -07001087 if (!newpage)
1088 return -ENOMEM;
1089
1090 if (page_count(page) == 1) {
1091 /* page was freed from under us. So we are done. */
Minchan Kimc6c919e2016-07-26 15:23:02 -07001092 ClearPageActive(page);
1093 ClearPageUnevictable(page);
Minchan Kimbda807d2016-07-26 15:23:05 -07001094 if (unlikely(__PageMovable(page))) {
1095 lock_page(page);
1096 if (!PageMovable(page))
1097 __ClearPageIsolated(page);
1098 unlock_page(page);
1099 }
Minchan Kimc6c919e2016-07-26 15:23:02 -07001100 if (put_new_page)
1101 put_new_page(newpage, private);
1102 else
1103 put_page(newpage);
Minchan Kim0dabec92011-10-31 17:06:57 -07001104 goto out;
1105 }
1106
Kirill A. Shutemov4d2fa962016-01-15 16:54:00 -08001107 if (unlikely(PageTransHuge(page))) {
1108 lock_page(page);
1109 rc = split_huge_page(page);
1110 unlock_page(page);
1111 if (rc)
Minchan Kim0dabec92011-10-31 17:06:57 -07001112 goto out;
Kirill A. Shutemov4d2fa962016-01-15 16:54:00 -08001113 }
Minchan Kim0dabec92011-10-31 17:06:57 -07001114
Hugh Dickins9c620e22013-02-22 16:35:14 -08001115 rc = __unmap_and_move(page, newpage, force, mode);
Minchan Kimc6c919e2016-07-26 15:23:02 -07001116 if (rc == MIGRATEPAGE_SUCCESS)
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001117 set_page_owner_migrate_reason(newpage, reason);
Rafael Aquinibf6bddf2012-12-11 16:02:42 -08001118
Minchan Kim0dabec92011-10-31 17:06:57 -07001119out:
Christoph Lametere24f0b82006-06-23 02:03:51 -07001120 if (rc != -EAGAIN) {
Minchan Kim0dabec92011-10-31 17:06:57 -07001121 /*
1122 * A page that has been migrated has all references
1123 * removed and will be freed. A page that has not been
1124 * migrated will have kepts its references and be
1125 * restored.
1126 */
1127 list_del(&page->lru);
Ming Ling5d7d3622016-12-12 16:42:26 -08001128
1129 /*
1130 * Compaction can migrate also non-LRU pages which are
1131 * not accounted to NR_ISOLATED_*. They can be recognized
1132 * as __PageMovable
1133 */
1134 if (likely(!__PageMovable(page)))
1135 dec_node_page_state(page, NR_ISOLATED_ANON +
1136 page_is_file_cache(page));
Christoph Lametere24f0b82006-06-23 02:03:51 -07001137 }
David Rientjes68711a72014-06-04 16:08:25 -07001138
Christoph Lameter95a402c2006-06-23 02:03:53 -07001139 /*
Minchan Kimc6c919e2016-07-26 15:23:02 -07001140 * If migration is successful, releases reference grabbed during
1141 * isolation. Otherwise, restore the page to right list unless
1142 * we want to retry.
Christoph Lameter95a402c2006-06-23 02:03:53 -07001143 */
Minchan Kimc6c919e2016-07-26 15:23:02 -07001144 if (rc == MIGRATEPAGE_SUCCESS) {
1145 put_page(page);
1146 if (reason == MR_MEMORY_FAILURE) {
1147 /*
1148 * Set PG_HWPoison on just freed page
1149 * intentionally. Although it's rather weird,
1150 * it's how HWPoison flag works at the moment.
1151 */
1152 if (!test_set_page_hwpoison(page))
1153 num_poisoned_pages_inc();
1154 }
1155 } else {
Minchan Kimbda807d2016-07-26 15:23:05 -07001156 if (rc != -EAGAIN) {
1157 if (likely(!__PageMovable(page))) {
1158 putback_lru_page(page);
1159 goto put_new;
1160 }
1161
1162 lock_page(page);
1163 if (PageMovable(page))
1164 putback_movable_page(page);
1165 else
1166 __ClearPageIsolated(page);
1167 unlock_page(page);
1168 put_page(page);
1169 }
1170put_new:
Minchan Kimc6c919e2016-07-26 15:23:02 -07001171 if (put_new_page)
1172 put_new_page(newpage, private);
1173 else
1174 put_page(newpage);
1175 }
David Rientjes68711a72014-06-04 16:08:25 -07001176
Christoph Lameter742755a2006-06-23 02:03:55 -07001177 if (result) {
1178 if (rc)
1179 *result = rc;
1180 else
1181 *result = page_to_nid(newpage);
1182 }
Christoph Lametere24f0b82006-06-23 02:03:51 -07001183 return rc;
1184}
1185
1186/*
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001187 * Counterpart of unmap_and_move_page() for hugepage migration.
1188 *
1189 * This function doesn't wait the completion of hugepage I/O
1190 * because there is no race between I/O and migration for hugepage.
1191 * Note that currently hugepage I/O occurs only in direct I/O
1192 * where no lock is held and PG_writeback is irrelevant,
1193 * and writeback status of all subpages are counted in the reference
1194 * count of the head page (i.e. if all subpages of a 2MB hugepage are
1195 * under direct I/O, the reference of the head page is 512 and a bit more.)
1196 * This means that when we try to migrate hugepage whose subpages are
1197 * doing direct I/O, some references remain after try_to_unmap() and
1198 * hugepage migration fails without data corruption.
1199 *
1200 * There is also no race when direct I/O is issued on the page under migration,
1201 * because then pte is replaced with migration swap entry and direct I/O code
1202 * will wait in the page fault for migration to complete.
1203 */
1204static int unmap_and_move_huge_page(new_page_t get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001205 free_page_t put_new_page, unsigned long private,
1206 struct page *hpage, int force,
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001207 enum migrate_mode mode, int reason)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001208{
Hugh Dickins2def7422015-11-05 18:49:46 -08001209 int rc = -EAGAIN;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001210 int *result = NULL;
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001211 int page_was_mapped = 0;
Joonsoo Kim32665f22014-01-21 15:51:15 -08001212 struct page *new_hpage;
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001213 struct anon_vma *anon_vma = NULL;
1214
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001215 /*
1216 * Movability of hugepages depends on architectures and hugepage size.
1217 * This check is necessary because some callers of hugepage migration
1218 * like soft offline and memory hotremove don't walk through page
1219 * tables or check whether the hugepage is pmd-based or not before
1220 * kicking migration.
1221 */
Naoya Horiguchi100873d2014-06-04 16:10:56 -07001222 if (!hugepage_migration_supported(page_hstate(hpage))) {
Joonsoo Kim32665f22014-01-21 15:51:15 -08001223 putback_active_hugepage(hpage);
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001224 return -ENOSYS;
Joonsoo Kim32665f22014-01-21 15:51:15 -08001225 }
Naoya Horiguchi83467ef2013-09-11 14:22:11 -07001226
Joonsoo Kim32665f22014-01-21 15:51:15 -08001227 new_hpage = get_new_page(hpage, private, &result);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001228 if (!new_hpage)
1229 return -ENOMEM;
1230
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001231 if (!trylock_page(hpage)) {
Mel Gormana6bc32b2012-01-12 17:19:43 -08001232 if (!force || mode != MIGRATE_SYNC)
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001233 goto out;
1234 lock_page(hpage);
1235 }
1236
Mike Kravetzb010e032019-02-28 16:22:02 -08001237 /*
1238 * Check for pages which are in the process of being freed. Without
1239 * page_mapping() set, hugetlbfs specific move page routine will not
1240 * be called and we could leak usage counts for subpools.
1241 */
1242 if (page_private(hpage) && !page_mapping(hpage)) {
1243 rc = -EBUSY;
1244 goto out_unlock;
1245 }
1246
Peter Zijlstra746b18d2011-05-24 17:12:10 -07001247 if (PageAnon(hpage))
1248 anon_vma = page_get_anon_vma(hpage);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001249
Hugh Dickins7db76712015-11-05 18:49:49 -08001250 if (unlikely(!trylock_page(new_hpage)))
1251 goto put_anon;
1252
Hugh Dickins2ebba6b2014-12-12 16:56:19 -08001253 if (page_mapped(hpage)) {
1254 try_to_unmap(hpage,
1255 TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
1256 page_was_mapped = 1;
1257 }
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001258
1259 if (!page_mapped(hpage))
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001260 rc = move_to_new_page(new_hpage, hpage, mode);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001261
Hugh Dickins5c3f9a62015-11-05 18:49:53 -08001262 if (page_was_mapped)
1263 remove_migration_ptes(hpage,
Kirill A. Shutemove3884662016-03-17 14:20:07 -07001264 rc == MIGRATEPAGE_SUCCESS ? new_hpage : hpage, false);
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001265
Hugh Dickins7db76712015-11-05 18:49:49 -08001266 unlock_page(new_hpage);
1267
1268put_anon:
Hugh Dickinsfd4a4662011-01-13 15:47:31 -08001269 if (anon_vma)
Peter Zijlstra9e601092011-03-22 16:32:46 -07001270 put_anon_vma(anon_vma);
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001271
Hugh Dickins2def7422015-11-05 18:49:46 -08001272 if (rc == MIGRATEPAGE_SUCCESS) {
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001273 hugetlb_cgroup_migrate(hpage, new_hpage);
Hugh Dickins2def7422015-11-05 18:49:46 -08001274 put_new_page = NULL;
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001275 set_page_owner_migrate_reason(new_hpage, reason);
Hugh Dickins2def7422015-11-05 18:49:46 -08001276 }
Aneesh Kumar K.V8e6ac7f2012-07-31 16:42:27 -07001277
Mike Kravetzb010e032019-02-28 16:22:02 -08001278out_unlock:
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001279 unlock_page(hpage);
Hillf Danton09761332011-12-08 14:34:20 -08001280out:
Naoya Horiguchib8ec1ce2013-09-11 14:22:01 -07001281 if (rc != -EAGAIN)
1282 putback_active_hugepage(hpage);
David Rientjes68711a72014-06-04 16:08:25 -07001283
1284 /*
1285 * If migration was not successful and there's a freeing callback, use
1286 * it. Otherwise, put_page() will drop the reference grabbed during
1287 * isolation.
1288 */
Hugh Dickins2def7422015-11-05 18:49:46 -08001289 if (put_new_page)
David Rientjes68711a72014-06-04 16:08:25 -07001290 put_new_page(new_hpage, private);
1291 else
Naoya Horiguchi3aaa76e2015-09-22 14:59:14 -07001292 putback_active_hugepage(new_hpage);
David Rientjes68711a72014-06-04 16:08:25 -07001293
Naoya Horiguchi290408d2010-09-08 10:19:35 +09001294 if (result) {
1295 if (rc)
1296 *result = rc;
1297 else
1298 *result = page_to_nid(new_hpage);
1299 }
1300 return rc;
1301}
1302
1303/*
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001304 * migrate_pages - migrate the pages specified in a list, to the free pages
1305 * supplied as the target for the page migration
Christoph Lameterb20a3502006-03-22 00:09:12 -08001306 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001307 * @from: The list of pages to be migrated.
1308 * @get_new_page: The function used to allocate free pages to be used
1309 * as the target of the page migration.
David Rientjes68711a72014-06-04 16:08:25 -07001310 * @put_new_page: The function used to free target pages if migration
1311 * fails, or NULL if no special handling is necessary.
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001312 * @private: Private data to be passed on to get_new_page()
1313 * @mode: The migration mode that specifies the constraints for
1314 * page migration, if any.
1315 * @reason: The reason for page migration.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001316 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001317 * The function returns after 10 attempts or if no pages are movable any more
1318 * because the list has become empty or no retryable pages exist any more.
Hugh Dickins14e0f9b2015-11-05 18:49:43 -08001319 * The caller should call putback_movable_pages() to return pages to the LRU
Minchan Kim28bd6572011-01-25 15:07:26 -08001320 * or free list only if ret != 0.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001321 *
Srivatsa S. Bhatc73e5c92013-04-29 15:08:16 -07001322 * Returns the number of pages that were not migrated, or an error code.
Christoph Lameterb20a3502006-03-22 00:09:12 -08001323 */
Hugh Dickins9c620e22013-02-22 16:35:14 -08001324int migrate_pages(struct list_head *from, new_page_t get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001325 free_page_t put_new_page, unsigned long private,
1326 enum migrate_mode mode, int reason)
Christoph Lameterb20a3502006-03-22 00:09:12 -08001327{
Christoph Lametere24f0b82006-06-23 02:03:51 -07001328 int retry = 1;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001329 int nr_failed = 0;
Mel Gorman5647bc22012-10-19 10:46:20 +01001330 int nr_succeeded = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001331 int pass = 0;
1332 struct page *page;
1333 struct page *page2;
1334 int swapwrite = current->flags & PF_SWAPWRITE;
1335 int rc;
1336
1337 if (!swapwrite)
1338 current->flags |= PF_SWAPWRITE;
1339
Christoph Lametere24f0b82006-06-23 02:03:51 -07001340 for(pass = 0; pass < 10 && retry; pass++) {
1341 retry = 0;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001342
Christoph Lametere24f0b82006-06-23 02:03:51 -07001343 list_for_each_entry_safe(page, page2, from, lru) {
Christoph Lametere24f0b82006-06-23 02:03:51 -07001344 cond_resched();
Christoph Lameterb20a3502006-03-22 00:09:12 -08001345
Naoya Horiguchi31caf662013-09-11 14:21:59 -07001346 if (PageHuge(page))
1347 rc = unmap_and_move_huge_page(get_new_page,
David Rientjes68711a72014-06-04 16:08:25 -07001348 put_new_page, private, page,
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07001349 pass > 2, mode, reason);
Naoya Horiguchi31caf662013-09-11 14:21:59 -07001350 else
David Rientjes68711a72014-06-04 16:08:25 -07001351 rc = unmap_and_move(get_new_page, put_new_page,
Naoya Horiguchiadd05ce2015-06-24 16:56:50 -07001352 private, page, pass > 2, mode,
1353 reason);
Christoph Lameterb20a3502006-03-22 00:09:12 -08001354
Christoph Lametere24f0b82006-06-23 02:03:51 -07001355 switch(rc) {
Christoph Lameter95a402c2006-06-23 02:03:53 -07001356 case -ENOMEM:
David Rientjesdfef2ef2016-05-20 16:59:05 -07001357 nr_failed++;
Christoph Lameter95a402c2006-06-23 02:03:53 -07001358 goto out;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001359 case -EAGAIN:
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001360 retry++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001361 break;
Rafael Aquini78bd5202012-12-11 16:02:31 -08001362 case MIGRATEPAGE_SUCCESS:
Mel Gorman5647bc22012-10-19 10:46:20 +01001363 nr_succeeded++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001364 break;
1365 default:
Naoya Horiguchi354a3362014-01-21 15:51:14 -08001366 /*
1367 * Permanent failure (-EBUSY, -ENOSYS, etc.):
1368 * unlike -EAGAIN case, the failed page is
1369 * removed from migration page list and not
1370 * retried in the next outer loop.
1371 */
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001372 nr_failed++;
Christoph Lametere24f0b82006-06-23 02:03:51 -07001373 break;
Christoph Lameter2d1db3b2006-06-23 02:03:33 -07001374 }
Christoph Lameterb20a3502006-03-22 00:09:12 -08001375 }
1376 }
Vlastimil Babkaf2f81fb2015-11-05 18:47:03 -08001377 nr_failed += retry;
1378 rc = nr_failed;
Christoph Lameter95a402c2006-06-23 02:03:53 -07001379out:
Mel Gorman5647bc22012-10-19 10:46:20 +01001380 if (nr_succeeded)
1381 count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded);
1382 if (nr_failed)
1383 count_vm_events(PGMIGRATE_FAIL, nr_failed);
Mel Gorman7b2a2d42012-10-19 14:07:31 +01001384 trace_mm_migrate_pages(nr_succeeded, nr_failed, mode, reason);
1385
Christoph Lameterb20a3502006-03-22 00:09:12 -08001386 if (!swapwrite)
1387 current->flags &= ~PF_SWAPWRITE;
1388
Rafael Aquini78bd5202012-12-11 16:02:31 -08001389 return rc;
Christoph Lameterb20a3502006-03-22 00:09:12 -08001390}
1391
Christoph Lameter742755a2006-06-23 02:03:55 -07001392#ifdef CONFIG_NUMA
1393/*
1394 * Move a list of individual pages
1395 */
1396struct page_to_node {
1397 unsigned long addr;
1398 struct page *page;
1399 int node;
1400 int status;
1401};
1402
1403static struct page *new_page_node(struct page *p, unsigned long private,
1404 int **result)
1405{
1406 struct page_to_node *pm = (struct page_to_node *)private;
1407
1408 while (pm->node != MAX_NUMNODES && pm->page != p)
1409 pm++;
1410
1411 if (pm->node == MAX_NUMNODES)
1412 return NULL;
1413
1414 *result = &pm->status;
1415
Naoya Horiguchie632a932013-09-11 14:22:04 -07001416 if (PageHuge(p))
1417 return alloc_huge_page_node(page_hstate(compound_head(p)),
1418 pm->node);
1419 else
Vlastimil Babka96db8002015-09-08 15:03:50 -07001420 return __alloc_pages_node(pm->node,
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001421 GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, 0);
Christoph Lameter742755a2006-06-23 02:03:55 -07001422}
1423
1424/*
1425 * Move a set of pages as indicated in the pm array. The addr
1426 * field must be set to the virtual address of the page to be moved
1427 * and the node number must contain a valid target node.
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001428 * The pm array ends with node = MAX_NUMNODES.
Christoph Lameter742755a2006-06-23 02:03:55 -07001429 */
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001430static int do_move_page_to_node_array(struct mm_struct *mm,
1431 struct page_to_node *pm,
1432 int migrate_all)
Christoph Lameter742755a2006-06-23 02:03:55 -07001433{
1434 int err;
1435 struct page_to_node *pp;
1436 LIST_HEAD(pagelist);
1437
1438 down_read(&mm->mmap_sem);
1439
1440 /*
1441 * Build a list of pages to migrate
1442 */
Christoph Lameter742755a2006-06-23 02:03:55 -07001443 for (pp = pm; pp->node != MAX_NUMNODES; pp++) {
1444 struct vm_area_struct *vma;
1445 struct page *page;
1446
Christoph Lameter742755a2006-06-23 02:03:55 -07001447 err = -EFAULT;
1448 vma = find_vma(mm, pp->addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001449 if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma))
Christoph Lameter742755a2006-06-23 02:03:55 -07001450 goto set_status;
1451
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001452 /* FOLL_DUMP to ignore special (like zero) pages */
1453 page = follow_page(vma, pp->addr,
1454 FOLL_GET | FOLL_SPLIT | FOLL_DUMP);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001455
1456 err = PTR_ERR(page);
1457 if (IS_ERR(page))
1458 goto set_status;
1459
Christoph Lameter742755a2006-06-23 02:03:55 -07001460 err = -ENOENT;
1461 if (!page)
1462 goto set_status;
1463
Christoph Lameter742755a2006-06-23 02:03:55 -07001464 pp->page = page;
1465 err = page_to_nid(page);
1466
1467 if (err == pp->node)
1468 /*
1469 * Node already in the right place
1470 */
1471 goto put_and_set;
1472
1473 err = -EACCES;
1474 if (page_mapcount(page) > 1 &&
1475 !migrate_all)
1476 goto put_and_set;
1477
Naoya Horiguchie632a932013-09-11 14:22:04 -07001478 if (PageHuge(page)) {
Naoya Horiguchie66f17f2015-02-11 15:25:22 -08001479 if (PageHead(page))
1480 isolate_huge_page(page, &pagelist);
Naoya Horiguchie632a932013-09-11 14:22:04 -07001481 goto put_and_set;
1482 }
1483
Nick Piggin62695a82008-10-18 20:26:09 -07001484 err = isolate_lru_page(page);
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001485 if (!err) {
Nick Piggin62695a82008-10-18 20:26:09 -07001486 list_add_tail(&page->lru, &pagelist);
Mel Gorman599d0c92016-07-28 15:45:31 -07001487 inc_node_page_state(page, NR_ISOLATED_ANON +
KOSAKI Motohiro6d9c2852009-12-14 17:58:11 -08001488 page_is_file_cache(page));
1489 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001490put_and_set:
1491 /*
1492 * Either remove the duplicate refcount from
1493 * isolate_lru_page() or drop the page ref if it was
1494 * not isolated.
1495 */
1496 put_page(page);
1497set_status:
1498 pp->status = err;
1499 }
1500
Brice Gogline78bbfa2008-10-18 20:27:15 -07001501 err = 0;
Minchan Kimcf608ac2010-10-26 14:21:29 -07001502 if (!list_empty(&pagelist)) {
David Rientjes68711a72014-06-04 16:08:25 -07001503 err = migrate_pages(&pagelist, new_page_node, NULL,
Hugh Dickins9c620e22013-02-22 16:35:14 -08001504 (unsigned long)pm, MIGRATE_SYNC, MR_SYSCALL);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001505 if (err)
Naoya Horiguchie632a932013-09-11 14:22:04 -07001506 putback_movable_pages(&pagelist);
Minchan Kimcf608ac2010-10-26 14:21:29 -07001507 }
Christoph Lameter742755a2006-06-23 02:03:55 -07001508
1509 up_read(&mm->mmap_sem);
1510 return err;
1511}
1512
1513/*
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001514 * Migrate an array of page address onto an array of nodes and fill
1515 * the corresponding array of status.
1516 */
Christoph Lameter3268c632012-03-21 16:34:06 -07001517static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes,
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001518 unsigned long nr_pages,
1519 const void __user * __user *pages,
1520 const int __user *nodes,
1521 int __user *status, int flags)
1522{
Brice Goglin3140a222009-01-06 14:38:57 -08001523 struct page_to_node *pm;
Brice Goglin3140a222009-01-06 14:38:57 -08001524 unsigned long chunk_nr_pages;
1525 unsigned long chunk_start;
1526 int err;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001527
Brice Goglin3140a222009-01-06 14:38:57 -08001528 err = -ENOMEM;
1529 pm = (struct page_to_node *)__get_free_page(GFP_KERNEL);
1530 if (!pm)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001531 goto out;
Brice Goglin35282a22009-06-16 15:32:43 -07001532
1533 migrate_prep();
1534
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001535 /*
Brice Goglin3140a222009-01-06 14:38:57 -08001536 * Store a chunk of page_to_node array in a page,
1537 * but keep the last one as a marker
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001538 */
Brice Goglin3140a222009-01-06 14:38:57 -08001539 chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001540
Brice Goglin3140a222009-01-06 14:38:57 -08001541 for (chunk_start = 0;
1542 chunk_start < nr_pages;
1543 chunk_start += chunk_nr_pages) {
1544 int j;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001545
Brice Goglin3140a222009-01-06 14:38:57 -08001546 if (chunk_start + chunk_nr_pages > nr_pages)
1547 chunk_nr_pages = nr_pages - chunk_start;
1548
1549 /* fill the chunk pm with addrs and nodes from user-space */
1550 for (j = 0; j < chunk_nr_pages; j++) {
1551 const void __user *p;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001552 int node;
1553
Brice Goglin3140a222009-01-06 14:38:57 -08001554 err = -EFAULT;
1555 if (get_user(p, pages + j + chunk_start))
1556 goto out_pm;
1557 pm[j].addr = (unsigned long) p;
1558
1559 if (get_user(node, nodes + j + chunk_start))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001560 goto out_pm;
1561
1562 err = -ENODEV;
Linus Torvalds6f5a55f2010-02-05 16:16:50 -08001563 if (node < 0 || node >= MAX_NUMNODES)
1564 goto out_pm;
1565
Lai Jiangshan389162c2012-12-12 13:51:30 -08001566 if (!node_state(node, N_MEMORY))
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001567 goto out_pm;
1568
1569 err = -EACCES;
1570 if (!node_isset(node, task_nodes))
1571 goto out_pm;
1572
Brice Goglin3140a222009-01-06 14:38:57 -08001573 pm[j].node = node;
1574 }
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001575
Brice Goglin3140a222009-01-06 14:38:57 -08001576 /* End marker for this chunk */
1577 pm[chunk_nr_pages].node = MAX_NUMNODES;
1578
1579 /* Migrate this chunk */
1580 err = do_move_page_to_node_array(mm, pm,
1581 flags & MPOL_MF_MOVE_ALL);
1582 if (err < 0)
1583 goto out_pm;
1584
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001585 /* Return status information */
Brice Goglin3140a222009-01-06 14:38:57 -08001586 for (j = 0; j < chunk_nr_pages; j++)
1587 if (put_user(pm[j].status, status + j + chunk_start)) {
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001588 err = -EFAULT;
Brice Goglin3140a222009-01-06 14:38:57 -08001589 goto out_pm;
1590 }
1591 }
1592 err = 0;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001593
1594out_pm:
Brice Goglin3140a222009-01-06 14:38:57 -08001595 free_page((unsigned long)pm);
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001596out:
1597 return err;
1598}
1599
1600/*
Brice Goglin2f007e72008-10-18 20:27:16 -07001601 * Determine the nodes of an array of pages and store it in an array of status.
Christoph Lameter742755a2006-06-23 02:03:55 -07001602 */
Brice Goglin80bba122008-12-09 13:14:23 -08001603static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages,
1604 const void __user **pages, int *status)
Christoph Lameter742755a2006-06-23 02:03:55 -07001605{
Brice Goglin2f007e72008-10-18 20:27:16 -07001606 unsigned long i;
Brice Goglin2f007e72008-10-18 20:27:16 -07001607
Christoph Lameter742755a2006-06-23 02:03:55 -07001608 down_read(&mm->mmap_sem);
1609
Brice Goglin2f007e72008-10-18 20:27:16 -07001610 for (i = 0; i < nr_pages; i++) {
Brice Goglin80bba122008-12-09 13:14:23 -08001611 unsigned long addr = (unsigned long)(*pages);
Christoph Lameter742755a2006-06-23 02:03:55 -07001612 struct vm_area_struct *vma;
1613 struct page *page;
KOSAKI Motohiroc095adb2008-12-16 16:06:43 +09001614 int err = -EFAULT;
Brice Goglin2f007e72008-10-18 20:27:16 -07001615
1616 vma = find_vma(mm, addr);
Gleb Natapov70384dc2010-10-26 14:22:07 -07001617 if (!vma || addr < vma->vm_start)
Christoph Lameter742755a2006-06-23 02:03:55 -07001618 goto set_status;
1619
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001620 /* FOLL_DUMP to ignore special (like zero) pages */
1621 page = follow_page(vma, addr, FOLL_DUMP);
Linus Torvalds89f5b7d2008-06-20 11:18:25 -07001622
1623 err = PTR_ERR(page);
1624 if (IS_ERR(page))
1625 goto set_status;
1626
Kirill A. Shutemovd8998442015-09-04 15:47:53 -07001627 err = page ? page_to_nid(page) : -ENOENT;
Christoph Lameter742755a2006-06-23 02:03:55 -07001628set_status:
Brice Goglin80bba122008-12-09 13:14:23 -08001629 *status = err;
1630
1631 pages++;
1632 status++;
1633 }
1634
1635 up_read(&mm->mmap_sem);
1636}
1637
1638/*
1639 * Determine the nodes of a user array of pages and store it in
1640 * a user array of status.
1641 */
1642static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages,
1643 const void __user * __user *pages,
1644 int __user *status)
1645{
1646#define DO_PAGES_STAT_CHUNK_NR 16
1647 const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR];
1648 int chunk_status[DO_PAGES_STAT_CHUNK_NR];
Brice Goglin80bba122008-12-09 13:14:23 -08001649
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001650 while (nr_pages) {
1651 unsigned long chunk_nr;
Brice Goglin80bba122008-12-09 13:14:23 -08001652
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001653 chunk_nr = nr_pages;
1654 if (chunk_nr > DO_PAGES_STAT_CHUNK_NR)
1655 chunk_nr = DO_PAGES_STAT_CHUNK_NR;
1656
1657 if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages)))
1658 break;
Brice Goglin80bba122008-12-09 13:14:23 -08001659
1660 do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status);
1661
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001662 if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status)))
1663 break;
Christoph Lameter742755a2006-06-23 02:03:55 -07001664
H. Peter Anvin87b8d1a2010-02-18 16:13:40 -08001665 pages += chunk_nr;
1666 status += chunk_nr;
1667 nr_pages -= chunk_nr;
1668 }
1669 return nr_pages ? -EFAULT : 0;
Christoph Lameter742755a2006-06-23 02:03:55 -07001670}
1671
1672/*
1673 * Move a list of pages in the address space of the currently executing
1674 * process.
1675 */
Heiko Carstens938bb9f2009-01-14 14:14:30 +01001676SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages,
1677 const void __user * __user *, pages,
1678 const int __user *, nodes,
1679 int __user *, status, int, flags)
Christoph Lameter742755a2006-06-23 02:03:55 -07001680{
Christoph Lameter742755a2006-06-23 02:03:55 -07001681 struct task_struct *task;
Christoph Lameter742755a2006-06-23 02:03:55 -07001682 struct mm_struct *mm;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001683 int err;
Christoph Lameter3268c632012-03-21 16:34:06 -07001684 nodemask_t task_nodes;
Christoph Lameter742755a2006-06-23 02:03:55 -07001685
1686 /* Check flags */
1687 if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL))
1688 return -EINVAL;
1689
1690 if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE))
1691 return -EPERM;
1692
1693 /* Find the mm_struct */
Greg Thelena879bf52011-02-25 14:44:13 -08001694 rcu_read_lock();
Pavel Emelyanov228ebcb2007-10-18 23:40:16 -07001695 task = pid ? find_task_by_vpid(pid) : current;
Christoph Lameter742755a2006-06-23 02:03:55 -07001696 if (!task) {
Greg Thelena879bf52011-02-25 14:44:13 -08001697 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001698 return -ESRCH;
1699 }
Christoph Lameter3268c632012-03-21 16:34:06 -07001700 get_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001701
1702 /*
1703 * Check if this process has the right to modify the specified
Linus Torvalds61332dc2017-08-20 13:26:27 -07001704 * process. Use the regular "ptrace_may_access()" checks.
Christoph Lameter742755a2006-06-23 02:03:55 -07001705 */
Linus Torvalds61332dc2017-08-20 13:26:27 -07001706 if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) {
David Howellsc69e8d92008-11-14 10:39:19 +11001707 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001708 err = -EPERM;
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001709 goto out;
Christoph Lameter742755a2006-06-23 02:03:55 -07001710 }
David Howellsc69e8d92008-11-14 10:39:19 +11001711 rcu_read_unlock();
Christoph Lameter742755a2006-06-23 02:03:55 -07001712
David Quigley86c3a762006-06-23 02:04:02 -07001713 err = security_task_movememory(task);
1714 if (err)
Brice Goglin5e9a0f02008-10-18 20:27:17 -07001715 goto out;
David Quigley86c3a762006-06-23 02:04:02 -07001716
Christoph Lameter3268c632012-03-21 16:34:06 -07001717 task_nodes = cpuset_mems_allowed(task);
1718 mm = get_task_mm(task);
1719 put_task_struct(task);
1720
Sasha Levin6e8b09e2012-04-25 16:01:53 -07001721 if (!mm)
1722 return -EINVAL;
1723
1724 if (nodes)
1725 err = do_pages_move(mm, task_nodes, nr_pages, pages,
1726 nodes, status, flags);
1727 else
1728 err = do_pages_stat(mm, nr_pages, pages, status);
Christoph Lameter3268c632012-03-21 16:34:06 -07001729
1730 mmput(mm);
1731 return err;
David Quigley86c3a762006-06-23 02:04:02 -07001732
Christoph Lameter742755a2006-06-23 02:03:55 -07001733out:
Christoph Lameter3268c632012-03-21 16:34:06 -07001734 put_task_struct(task);
Christoph Lameter742755a2006-06-23 02:03:55 -07001735 return err;
1736}
Christoph Lameter742755a2006-06-23 02:03:55 -07001737
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001738#ifdef CONFIG_NUMA_BALANCING
1739/*
1740 * Returns true if this is a safe migration target node for misplaced NUMA
1741 * pages. Currently it only checks the watermarks which crude
1742 */
1743static bool migrate_balanced_pgdat(struct pglist_data *pgdat,
Mel Gorman3abef4e2013-02-22 16:34:27 -08001744 unsigned long nr_migrate_pages)
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001745{
1746 int z;
Mel Gorman599d0c92016-07-28 15:45:31 -07001747
1748 if (!pgdat_reclaimable(pgdat))
1749 return false;
1750
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001751 for (z = pgdat->nr_zones - 1; z >= 0; z--) {
1752 struct zone *zone = pgdat->node_zones + z;
1753
1754 if (!populated_zone(zone))
1755 continue;
1756
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001757 /* Avoid waking kswapd by allocating pages_to_migrate pages. */
1758 if (!zone_watermark_ok(zone, 0,
1759 high_wmark_pages(zone) +
1760 nr_migrate_pages,
1761 0, 0))
1762 continue;
1763 return true;
1764 }
1765 return false;
1766}
1767
1768static struct page *alloc_misplaced_dst_page(struct page *page,
1769 unsigned long data,
1770 int **result)
1771{
1772 int nid = (int) data;
1773 struct page *newpage;
1774
Vlastimil Babka96db8002015-09-08 15:03:50 -07001775 newpage = __alloc_pages_node(nid,
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001776 (GFP_HIGHUSER_MOVABLE |
1777 __GFP_THISNODE | __GFP_NOMEMALLOC |
1778 __GFP_NORETRY | __GFP_NOWARN) &
Mel Gorman8479eba2016-02-26 15:19:31 -08001779 ~__GFP_RECLAIM, 0);
Hillf Dantonbac03822012-11-27 14:46:24 +00001780
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001781 return newpage;
1782}
1783
1784/*
Mel Gormana8f60772012-11-14 21:41:46 +00001785 * page migration rate limiting control.
1786 * Do not migrate more than @pages_to_migrate in a @migrate_interval_millisecs
1787 * window of time. Default here says do not migrate more than 1280M per second.
1788 */
1789static unsigned int migrate_interval_millisecs __read_mostly = 100;
1790static unsigned int ratelimit_pages __read_mostly = 128 << (20 - PAGE_SHIFT);
1791
Mel Gormanb32967f2012-11-19 12:35:47 +00001792/* Returns true if the node is migrate rate-limited after the update */
Mel Gorman1c30e012014-01-21 15:50:58 -08001793static bool numamigrate_update_ratelimit(pg_data_t *pgdat,
1794 unsigned long nr_pages)
Mel Gormanb32967f2012-11-19 12:35:47 +00001795{
Mel Gormanb32967f2012-11-19 12:35:47 +00001796 /*
1797 * Rate-limit the amount of data that is being migrated to a node.
1798 * Optimal placement is no good if the memory bus is saturated and
1799 * all the time is being spent migrating!
1800 */
Mel Gormanb32967f2012-11-19 12:35:47 +00001801 if (time_after(jiffies, pgdat->numabalancing_migrate_next_window)) {
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001802 spin_lock(&pgdat->numabalancing_migrate_lock);
Mel Gormanb32967f2012-11-19 12:35:47 +00001803 pgdat->numabalancing_migrate_nr_pages = 0;
1804 pgdat->numabalancing_migrate_next_window = jiffies +
1805 msecs_to_jiffies(migrate_interval_millisecs);
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001806 spin_unlock(&pgdat->numabalancing_migrate_lock);
Mel Gormanb32967f2012-11-19 12:35:47 +00001807 }
Mel Gormanaf1839d2014-01-21 15:51:01 -08001808 if (pgdat->numabalancing_migrate_nr_pages > ratelimit_pages) {
1809 trace_mm_numa_migrate_ratelimit(current, pgdat->node_id,
1810 nr_pages);
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001811 return true;
Mel Gormanaf1839d2014-01-21 15:51:01 -08001812 }
Mel Gorman1c5e9c22014-01-21 15:50:59 -08001813
1814 /*
1815 * This is an unlocked non-atomic update so errors are possible.
1816 * The consequences are failing to migrate when we potentiall should
1817 * have which is not severe enough to warrant locking. If it is ever
1818 * a problem, it can be converted to a per-cpu counter.
1819 */
1820 pgdat->numabalancing_migrate_nr_pages += nr_pages;
1821 return false;
Mel Gormanb32967f2012-11-19 12:35:47 +00001822}
1823
Mel Gorman1c30e012014-01-21 15:50:58 -08001824static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page)
Mel Gormanb32967f2012-11-19 12:35:47 +00001825{
Hugh Dickins340ef392013-02-22 16:34:33 -08001826 int page_lru;
Mel Gormanb32967f2012-11-19 12:35:47 +00001827
Sasha Levin309381fea2014-01-23 15:52:54 -08001828 VM_BUG_ON_PAGE(compound_order(page) && !PageTransHuge(page), page);
Mel Gorman3abef4e2013-02-22 16:34:27 -08001829
Mel Gormanb32967f2012-11-19 12:35:47 +00001830 /* Avoid migrating to a node that is nearly full */
Hugh Dickins340ef392013-02-22 16:34:33 -08001831 if (!migrate_balanced_pgdat(pgdat, 1UL << compound_order(page)))
1832 return 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001833
Hugh Dickins340ef392013-02-22 16:34:33 -08001834 if (isolate_lru_page(page))
1835 return 0;
Mel Gormanb32967f2012-11-19 12:35:47 +00001836
1837 /*
Hugh Dickins340ef392013-02-22 16:34:33 -08001838 * migrate_misplaced_transhuge_page() skips page migration's usual
1839 * check on page_count(), so we must do it here, now that the page
1840 * has been isolated: a GUP pin, or any other pin, prevents migration.
1841 * The expected page count is 3: 1 for page's mapcount and 1 for the
1842 * caller's pin and 1 for the reference taken by isolate_lru_page().
1843 */
1844 if (PageTransHuge(page) && page_count(page) != 3) {
1845 putback_lru_page(page);
1846 return 0;
1847 }
1848
1849 page_lru = page_is_file_cache(page);
Mel Gorman599d0c92016-07-28 15:45:31 -07001850 mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + page_lru,
Hugh Dickins340ef392013-02-22 16:34:33 -08001851 hpage_nr_pages(page));
1852
1853 /*
1854 * Isolating the page has taken another reference, so the
1855 * caller's reference can be safely dropped without the page
1856 * disappearing underneath us during migration.
Mel Gormanb32967f2012-11-19 12:35:47 +00001857 */
1858 put_page(page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001859 return 1;
Mel Gormanb32967f2012-11-19 12:35:47 +00001860}
1861
Mel Gormande466bd2013-12-18 17:08:42 -08001862bool pmd_trans_migrating(pmd_t pmd)
1863{
1864 struct page *page = pmd_page(pmd);
1865 return PageLocked(page);
1866}
1867
Mel Gormana8f60772012-11-14 21:41:46 +00001868/*
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001869 * Attempt to migrate a misplaced page to the specified destination
1870 * node. Caller is expected to have an elevated reference count on
1871 * the page that will be dropped by this function before returning.
1872 */
Mel Gorman1bc115d2013-10-07 11:29:05 +01001873int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma,
1874 int node)
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001875{
Mel Gormana8f60772012-11-14 21:41:46 +00001876 pg_data_t *pgdat = NODE_DATA(node);
Hugh Dickins340ef392013-02-22 16:34:33 -08001877 int isolated;
Mel Gormanb32967f2012-11-19 12:35:47 +00001878 int nr_remaining;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001879 LIST_HEAD(migratepages);
1880
1881 /*
Mel Gorman1bc115d2013-10-07 11:29:05 +01001882 * Don't migrate file pages that are mapped in multiple processes
1883 * with execute permissions as they are probably shared libraries.
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001884 */
Mel Gorman1bc115d2013-10-07 11:29:05 +01001885 if (page_mapcount(page) != 1 && page_is_file_cache(page) &&
1886 (vma->vm_flags & VM_EXEC))
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001887 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001888
Mel Gormana8f60772012-11-14 21:41:46 +00001889 /*
1890 * Rate-limit the amount of data that is being migrated to a node.
1891 * Optimal placement is no good if the memory bus is saturated and
1892 * all the time is being spent migrating!
1893 */
Hugh Dickins340ef392013-02-22 16:34:33 -08001894 if (numamigrate_update_ratelimit(pgdat, 1))
Mel Gormana8f60772012-11-14 21:41:46 +00001895 goto out;
Mel Gormana8f60772012-11-14 21:41:46 +00001896
Mel Gormanb32967f2012-11-19 12:35:47 +00001897 isolated = numamigrate_isolate_page(pgdat, page);
1898 if (!isolated)
1899 goto out;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001900
Mel Gormanb32967f2012-11-19 12:35:47 +00001901 list_add(&page->lru, &migratepages);
Hugh Dickins9c620e22013-02-22 16:35:14 -08001902 nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page,
David Rientjes68711a72014-06-04 16:08:25 -07001903 NULL, node, MIGRATE_ASYNC,
1904 MR_NUMA_MISPLACED);
Mel Gormanb32967f2012-11-19 12:35:47 +00001905 if (nr_remaining) {
Joonsoo Kim59c82b72014-01-21 15:51:17 -08001906 if (!list_empty(&migratepages)) {
1907 list_del(&page->lru);
Mel Gorman599d0c92016-07-28 15:45:31 -07001908 dec_node_page_state(page, NR_ISOLATED_ANON +
Joonsoo Kim59c82b72014-01-21 15:51:17 -08001909 page_is_file_cache(page));
1910 putback_lru_page(page);
1911 }
Mel Gormanb32967f2012-11-19 12:35:47 +00001912 isolated = 0;
1913 } else
1914 count_vm_numa_event(NUMA_PAGE_MIGRATE);
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001915 BUG_ON(!list_empty(&migratepages));
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001916 return isolated;
Hugh Dickins340ef392013-02-22 16:34:33 -08001917
1918out:
1919 put_page(page);
1920 return 0;
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02001921}
Mel Gorman220018d2012-12-05 09:32:56 +00001922#endif /* CONFIG_NUMA_BALANCING */
Mel Gormanb32967f2012-11-19 12:35:47 +00001923
Mel Gorman220018d2012-12-05 09:32:56 +00001924#if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE)
Hugh Dickins340ef392013-02-22 16:34:33 -08001925/*
1926 * Migrates a THP to a given target node. page must be locked and is unlocked
1927 * before returning.
1928 */
Mel Gormanb32967f2012-11-19 12:35:47 +00001929int migrate_misplaced_transhuge_page(struct mm_struct *mm,
1930 struct vm_area_struct *vma,
1931 pmd_t *pmd, pmd_t entry,
1932 unsigned long address,
1933 struct page *page, int node)
1934{
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001935 spinlock_t *ptl;
Mel Gormanb32967f2012-11-19 12:35:47 +00001936 pg_data_t *pgdat = NODE_DATA(node);
1937 int isolated = 0;
1938 struct page *new_page = NULL;
Mel Gormanb32967f2012-11-19 12:35:47 +00001939 int page_lru = page_is_file_cache(page);
Mel Gormanf714f4f2013-12-18 17:08:33 -08001940 unsigned long mmun_start = address & HPAGE_PMD_MASK;
1941 unsigned long mmun_end = mmun_start + HPAGE_PMD_SIZE;
Mel Gorman2b4847e2013-12-18 17:08:32 -08001942 pmd_t orig_entry;
Mel Gormanb32967f2012-11-19 12:35:47 +00001943
1944 /*
Mel Gormanb32967f2012-11-19 12:35:47 +00001945 * Rate-limit the amount of data that is being migrated to a node.
1946 * Optimal placement is no good if the memory bus is saturated and
1947 * all the time is being spent migrating!
1948 */
Mel Gormand28d43352012-11-29 09:24:36 +00001949 if (numamigrate_update_ratelimit(pgdat, HPAGE_PMD_NR))
Mel Gormanb32967f2012-11-19 12:35:47 +00001950 goto out_dropref;
1951
1952 new_page = alloc_pages_node(node,
Vlastimil Babka25160352016-07-28 15:49:25 -07001953 (GFP_TRANSHUGE_LIGHT | __GFP_THISNODE),
Johannes Weinere97ca8e2014-03-10 15:49:43 -07001954 HPAGE_PMD_ORDER);
Hugh Dickins340ef392013-02-22 16:34:33 -08001955 if (!new_page)
1956 goto out_fail;
Kirill A. Shutemov9a982252016-01-15 16:54:17 -08001957 prep_transhuge_page(new_page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001958
Mel Gormanb32967f2012-11-19 12:35:47 +00001959 isolated = numamigrate_isolate_page(pgdat, page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001960 if (!isolated) {
Mel Gormanb32967f2012-11-19 12:35:47 +00001961 put_page(new_page);
Hugh Dickins340ef392013-02-22 16:34:33 -08001962 goto out_fail;
Mel Gormanb32967f2012-11-19 12:35:47 +00001963 }
Aneesh Kumar K.V458aa762016-03-17 14:18:56 -07001964 /*
1965 * We are not sure a pending tlb flush here is for a huge page
1966 * mapping or not. Hence use the tlb range variant
1967 */
Mel Gormanb0943d62013-12-18 17:08:46 -08001968 if (mm_tlb_flush_pending(mm))
1969 flush_tlb_range(vma, mmun_start, mmun_end);
1970
Mel Gormanb32967f2012-11-19 12:35:47 +00001971 /* Prepare a page as a migration target */
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -08001972 __SetPageLocked(new_page);
Hugh Dickinsfa9949d2016-05-19 17:12:41 -07001973 __SetPageSwapBacked(new_page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001974
1975 /* anon mapping, we can simply copy page->mapping to the new page: */
1976 new_page->mapping = page->mapping;
1977 new_page->index = page->index;
1978 migrate_page_copy(new_page, page);
1979 WARN_ON(PageLRU(new_page));
1980
1981 /* Recheck the target PMD */
Mel Gormanf714f4f2013-12-18 17:08:33 -08001982 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001983 ptl = pmd_lock(mm, pmd);
Mel Gorman2b4847e2013-12-18 17:08:32 -08001984 if (unlikely(!pmd_same(*pmd, entry) || page_count(page) != 2)) {
1985fail_putback:
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08001986 spin_unlock(ptl);
Mel Gormanf714f4f2013-12-18 17:08:33 -08001987 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Mel Gormanb32967f2012-11-19 12:35:47 +00001988
1989 /* Reverse changes made by migrate_page_copy() */
1990 if (TestClearPageActive(new_page))
1991 SetPageActive(page);
1992 if (TestClearPageUnevictable(new_page))
1993 SetPageUnevictable(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00001994
1995 unlock_page(new_page);
1996 put_page(new_page); /* Free it */
1997
Mel Gormana54a4072013-10-07 11:28:46 +01001998 /* Retake the callers reference and putback on LRU */
1999 get_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00002000 putback_lru_page(page);
Mel Gorman599d0c92016-07-28 15:45:31 -07002001 mod_node_page_state(page_pgdat(page),
Mel Gormana54a4072013-10-07 11:28:46 +01002002 NR_ISOLATED_ANON + page_lru, -HPAGE_PMD_NR);
Mel Gormaneb4489f62013-12-18 17:08:39 -08002003
2004 goto out_unlock;
Mel Gormanb32967f2012-11-19 12:35:47 +00002005 }
2006
Mel Gorman2b4847e2013-12-18 17:08:32 -08002007 orig_entry = *pmd;
Kirill A. Shutemov10102452016-07-26 15:25:29 -07002008 entry = mk_huge_pmd(new_page, vma->vm_page_prot);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002009 entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
Mel Gormanb32967f2012-11-19 12:35:47 +00002010
Mel Gorman2b4847e2013-12-18 17:08:32 -08002011 /*
2012 * Clear the old entry under pagetable lock and establish the new PTE.
2013 * Any parallel GUP will either observe the old page blocking on the
2014 * page lock, block on the page table lock or observe the new page.
2015 * The SetPageUptodate on the new page and page_add_new_anon_rmap
2016 * guarantee the copy is visible before the pagetable update.
2017 */
Mel Gormanf714f4f2013-12-18 17:08:33 -08002018 flush_cache_range(vma, mmun_start, mmun_end);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002019 page_add_anon_rmap(new_page, vma, mmun_start, true);
Aneesh Kumar K.V8809aa22015-06-24 16:57:44 -07002020 pmdp_huge_clear_flush_notify(vma, mmun_start, pmd);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002021 set_pmd_at(mm, mmun_start, pmd, entry);
Stephen Rothwellce4a9cc2012-12-10 19:50:57 +11002022 update_mmu_cache_pmd(vma, address, &entry);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002023
2024 if (page_count(page) != 2) {
Mel Gormanf714f4f2013-12-18 17:08:33 -08002025 set_pmd_at(mm, mmun_start, pmd, orig_entry);
Aneesh Kumar K.V458aa762016-03-17 14:18:56 -07002026 flush_pmd_tlb_range(vma, mmun_start, mmun_end);
Joerg Roedel34ee6452014-11-13 13:46:09 +11002027 mmu_notifier_invalidate_range(mm, mmun_start, mmun_end);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002028 update_mmu_cache_pmd(vma, address, &entry);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002029 page_remove_rmap(new_page, true);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002030 goto fail_putback;
2031 }
2032
Hugh Dickins51afb122015-11-05 18:49:37 -08002033 mlock_migrate_page(new_page, page);
Kirill A. Shutemovd281ee62016-01-15 16:52:16 -08002034 page_remove_rmap(page, true);
Vlastimil Babka7cd12b42016-03-15 14:56:18 -07002035 set_page_owner_migrate_reason(new_page, MR_NUMA_MISPLACED);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002036
Kirill A. Shutemovc4088eb2013-11-14 14:31:04 -08002037 spin_unlock(ptl);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002038 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
Mel Gormanb32967f2012-11-19 12:35:47 +00002039
Mel Gorman11de9922014-06-04 16:07:41 -07002040 /* Take an "isolate" reference and put new page on the LRU. */
2041 get_page(new_page);
2042 putback_lru_page(new_page);
2043
Mel Gormanb32967f2012-11-19 12:35:47 +00002044 unlock_page(new_page);
2045 unlock_page(page);
2046 put_page(page); /* Drop the rmap reference */
2047 put_page(page); /* Drop the LRU isolation reference */
2048
2049 count_vm_events(PGMIGRATE_SUCCESS, HPAGE_PMD_NR);
2050 count_vm_numa_events(NUMA_PAGE_MIGRATE, HPAGE_PMD_NR);
2051
Mel Gorman599d0c92016-07-28 15:45:31 -07002052 mod_node_page_state(page_pgdat(page),
Mel Gormanb32967f2012-11-19 12:35:47 +00002053 NR_ISOLATED_ANON + page_lru,
2054 -HPAGE_PMD_NR);
2055 return isolated;
2056
Hugh Dickins340ef392013-02-22 16:34:33 -08002057out_fail:
2058 count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR);
Mel Gormanb32967f2012-11-19 12:35:47 +00002059out_dropref:
Mel Gorman2b4847e2013-12-18 17:08:32 -08002060 ptl = pmd_lock(mm, pmd);
2061 if (pmd_same(*pmd, entry)) {
Mel Gorman4d942462015-02-12 14:58:28 -08002062 entry = pmd_modify(entry, vma->vm_page_prot);
Mel Gormanf714f4f2013-12-18 17:08:33 -08002063 set_pmd_at(mm, mmun_start, pmd, entry);
Mel Gorman2b4847e2013-12-18 17:08:32 -08002064 update_mmu_cache_pmd(vma, address, &entry);
2065 }
2066 spin_unlock(ptl);
Mel Gormana54a4072013-10-07 11:28:46 +01002067
Mel Gormaneb4489f62013-12-18 17:08:39 -08002068out_unlock:
Hugh Dickins340ef392013-02-22 16:34:33 -08002069 unlock_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00002070 put_page(page);
Mel Gormanb32967f2012-11-19 12:35:47 +00002071 return 0;
2072}
Peter Zijlstra7039e1d2012-10-25 14:16:34 +02002073#endif /* CONFIG_NUMA_BALANCING */
2074
2075#endif /* CONFIG_NUMA */