]>
Commit | Line | Data |
---|---|---|
b20a3503 | 1 | /* |
14e0f9bc | 2 | * Memory Migration functionality - linux/mm/migrate.c |
b20a3503 CL |
3 | * |
4 | * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter | |
5 | * | |
6 | * Page migration was first developed in the context of the memory hotplug | |
7 | * project. The main authors of the migration code are: | |
8 | * | |
9 | * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> | |
10 | * Hirokazu Takahashi <taka@valinux.co.jp> | |
11 | * Dave Hansen <haveblue@us.ibm.com> | |
cde53535 | 12 | * Christoph Lameter |
b20a3503 CL |
13 | */ |
14 | ||
15 | #include <linux/migrate.h> | |
b95f1b31 | 16 | #include <linux/export.h> |
b20a3503 | 17 | #include <linux/swap.h> |
0697212a | 18 | #include <linux/swapops.h> |
b20a3503 | 19 | #include <linux/pagemap.h> |
e23ca00b | 20 | #include <linux/buffer_head.h> |
b20a3503 | 21 | #include <linux/mm_inline.h> |
b488893a | 22 | #include <linux/nsproxy.h> |
b20a3503 | 23 | #include <linux/pagevec.h> |
e9995ef9 | 24 | #include <linux/ksm.h> |
b20a3503 CL |
25 | #include <linux/rmap.h> |
26 | #include <linux/topology.h> | |
27 | #include <linux/cpu.h> | |
28 | #include <linux/cpuset.h> | |
04e62a29 | 29 | #include <linux/writeback.h> |
742755a1 CL |
30 | #include <linux/mempolicy.h> |
31 | #include <linux/vmalloc.h> | |
86c3a764 | 32 | #include <linux/security.h> |
42cb14b1 | 33 | #include <linux/backing-dev.h> |
bda807d4 | 34 | #include <linux/compaction.h> |
4f5ca265 | 35 | #include <linux/syscalls.h> |
290408d4 | 36 | #include <linux/hugetlb.h> |
8e6ac7fa | 37 | #include <linux/hugetlb_cgroup.h> |
5a0e3ad6 | 38 | #include <linux/gfp.h> |
bf6bddf1 | 39 | #include <linux/balloon_compaction.h> |
f714f4f2 | 40 | #include <linux/mmu_notifier.h> |
33c3fc71 | 41 | #include <linux/page_idle.h> |
d435edca | 42 | #include <linux/page_owner.h> |
6e84f315 | 43 | #include <linux/sched/mm.h> |
197e7e52 | 44 | #include <linux/ptrace.h> |
b20a3503 | 45 | |
0d1836c3 MN |
46 | #include <asm/tlbflush.h> |
47 | ||
7b2a2d4a MG |
48 | #define CREATE_TRACE_POINTS |
49 | #include <trace/events/migrate.h> | |
50 | ||
b20a3503 CL |
51 | #include "internal.h" |
52 | ||
b20a3503 | 53 | /* |
742755a1 | 54 | * migrate_prep() needs to be called before we start compiling a list of pages |
748446bb MG |
55 | * to be migrated using isolate_lru_page(). If scheduling work on other CPUs is |
56 | * undesirable, use migrate_prep_local() | |
b20a3503 CL |
57 | */ |
58 | int migrate_prep(void) | |
59 | { | |
b20a3503 CL |
60 | /* |
61 | * Clear the LRU lists so pages can be isolated. | |
62 | * Note that pages may be moved off the LRU after we have | |
63 | * drained them. Those pages will fail to migrate like other | |
64 | * pages that may be busy. | |
65 | */ | |
66 | lru_add_drain_all(); | |
67 | ||
68 | return 0; | |
69 | } | |
70 | ||
748446bb MG |
71 | /* Do the necessary work of migrate_prep but not if it involves other CPUs */ |
72 | int migrate_prep_local(void) | |
73 | { | |
74 | lru_add_drain(); | |
75 | ||
76 | return 0; | |
77 | } | |
78 | ||
9e5bcd61 | 79 | int isolate_movable_page(struct page *page, isolate_mode_t mode) |
bda807d4 MK |
80 | { |
81 | struct address_space *mapping; | |
82 | ||
83 | /* | |
84 | * Avoid burning cycles with pages that are yet under __free_pages(), | |
85 | * or just got freed under us. | |
86 | * | |
87 | * In case we 'win' a race for a movable page being freed under us and | |
88 | * raise its refcount preventing __free_pages() from doing its job | |
89 | * the put_page() at the end of this block will take care of | |
90 | * release this page, thus avoiding a nasty leakage. | |
91 | */ | |
92 | if (unlikely(!get_page_unless_zero(page))) | |
93 | goto out; | |
94 | ||
95 | /* | |
96 | * Check PageMovable before holding a PG_lock because page's owner | |
97 | * assumes anybody doesn't touch PG_lock of newly allocated page | |
98 | * so unconditionally grapping the lock ruins page's owner side. | |
99 | */ | |
100 | if (unlikely(!__PageMovable(page))) | |
101 | goto out_putpage; | |
102 | /* | |
103 | * As movable pages are not isolated from LRU lists, concurrent | |
104 | * compaction threads can race against page migration functions | |
105 | * as well as race against the releasing a page. | |
106 | * | |
107 | * In order to avoid having an already isolated movable page | |
108 | * being (wrongly) re-isolated while it is under migration, | |
109 | * or to avoid attempting to isolate pages being released, | |
110 | * lets be sure we have the page lock | |
111 | * before proceeding with the movable page isolation steps. | |
112 | */ | |
113 | if (unlikely(!trylock_page(page))) | |
114 | goto out_putpage; | |
115 | ||
116 | if (!PageMovable(page) || PageIsolated(page)) | |
117 | goto out_no_isolated; | |
118 | ||
119 | mapping = page_mapping(page); | |
120 | VM_BUG_ON_PAGE(!mapping, page); | |
121 | ||
122 | if (!mapping->a_ops->isolate_page(page, mode)) | |
123 | goto out_no_isolated; | |
124 | ||
125 | /* Driver shouldn't use PG_isolated bit of page->flags */ | |
126 | WARN_ON_ONCE(PageIsolated(page)); | |
127 | __SetPageIsolated(page); | |
128 | unlock_page(page); | |
129 | ||
9e5bcd61 | 130 | return 0; |
bda807d4 MK |
131 | |
132 | out_no_isolated: | |
133 | unlock_page(page); | |
134 | out_putpage: | |
135 | put_page(page); | |
136 | out: | |
9e5bcd61 | 137 | return -EBUSY; |
bda807d4 MK |
138 | } |
139 | ||
140 | /* It should be called on page which is PG_movable */ | |
141 | void putback_movable_page(struct page *page) | |
142 | { | |
143 | struct address_space *mapping; | |
144 | ||
145 | VM_BUG_ON_PAGE(!PageLocked(page), page); | |
146 | VM_BUG_ON_PAGE(!PageMovable(page), page); | |
147 | VM_BUG_ON_PAGE(!PageIsolated(page), page); | |
148 | ||
149 | mapping = page_mapping(page); | |
150 | mapping->a_ops->putback_page(page); | |
151 | __ClearPageIsolated(page); | |
152 | } | |
153 | ||
5733c7d1 RA |
154 | /* |
155 | * Put previously isolated pages back onto the appropriate lists | |
156 | * from where they were once taken off for compaction/migration. | |
157 | * | |
59c82b70 JK |
158 | * This function shall be used whenever the isolated pageset has been |
159 | * built from lru, balloon, hugetlbfs page. See isolate_migratepages_range() | |
160 | * and isolate_huge_page(). | |
5733c7d1 RA |
161 | */ |
162 | void putback_movable_pages(struct list_head *l) | |
163 | { | |
164 | struct page *page; | |
165 | struct page *page2; | |
166 | ||
b20a3503 | 167 | list_for_each_entry_safe(page, page2, l, lru) { |
31caf665 NH |
168 | if (unlikely(PageHuge(page))) { |
169 | putback_active_hugepage(page); | |
170 | continue; | |
171 | } | |
e24f0b8f | 172 | list_del(&page->lru); |
bda807d4 MK |
173 | /* |
174 | * We isolated non-lru movable page so here we can use | |
175 | * __PageMovable because LRU page's mapping cannot have | |
176 | * PAGE_MAPPING_MOVABLE. | |
177 | */ | |
b1123ea6 | 178 | if (unlikely(__PageMovable(page))) { |
bda807d4 MK |
179 | VM_BUG_ON_PAGE(!PageIsolated(page), page); |
180 | lock_page(page); | |
181 | if (PageMovable(page)) | |
182 | putback_movable_page(page); | |
183 | else | |
184 | __ClearPageIsolated(page); | |
185 | unlock_page(page); | |
186 | put_page(page); | |
187 | } else { | |
e8db67eb NH |
188 | mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + |
189 | page_is_file_cache(page), -hpage_nr_pages(page)); | |
fc280fe8 | 190 | putback_lru_page(page); |
bda807d4 | 191 | } |
b20a3503 | 192 | } |
b20a3503 CL |
193 | } |
194 | ||
0697212a CL |
195 | /* |
196 | * Restore a potential migration pte to a working pte entry | |
197 | */ | |
e4b82222 | 198 | static bool remove_migration_pte(struct page *page, struct vm_area_struct *vma, |
e9995ef9 | 199 | unsigned long addr, void *old) |
0697212a | 200 | { |
3fe87967 KS |
201 | struct page_vma_mapped_walk pvmw = { |
202 | .page = old, | |
203 | .vma = vma, | |
204 | .address = addr, | |
205 | .flags = PVMW_SYNC | PVMW_MIGRATION, | |
206 | }; | |
207 | struct page *new; | |
208 | pte_t pte; | |
0697212a | 209 | swp_entry_t entry; |
0697212a | 210 | |
3fe87967 KS |
211 | VM_BUG_ON_PAGE(PageTail(page), page); |
212 | while (page_vma_mapped_walk(&pvmw)) { | |
4b0ece6f NH |
213 | if (PageKsm(page)) |
214 | new = page; | |
215 | else | |
216 | new = page - pvmw.page->index + | |
217 | linear_page_index(vma, pvmw.address); | |
0697212a | 218 | |
616b8371 ZY |
219 | #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION |
220 | /* PMD-mapped THP migration entry */ | |
221 | if (!pvmw.pte) { | |
222 | VM_BUG_ON_PAGE(PageHuge(page) || !PageTransCompound(page), page); | |
223 | remove_migration_pmd(&pvmw, new); | |
224 | continue; | |
225 | } | |
226 | #endif | |
227 | ||
3fe87967 KS |
228 | get_page(new); |
229 | pte = pte_mkold(mk_pte(new, READ_ONCE(vma->vm_page_prot))); | |
230 | if (pte_swp_soft_dirty(*pvmw.pte)) | |
231 | pte = pte_mksoft_dirty(pte); | |
0697212a | 232 | |
3fe87967 KS |
233 | /* |
234 | * Recheck VMA as permissions can change since migration started | |
235 | */ | |
236 | entry = pte_to_swp_entry(*pvmw.pte); | |
237 | if (is_write_migration_entry(entry)) | |
238 | pte = maybe_mkwrite(pte, vma); | |
d3cb8bf6 | 239 | |
383321ab | 240 | flush_dcache_page(new); |
3ef8fd7f | 241 | #ifdef CONFIG_HUGETLB_PAGE |
3fe87967 KS |
242 | if (PageHuge(new)) { |
243 | pte = pte_mkhuge(pte); | |
244 | pte = arch_make_huge_pte(pte, vma, new, 0); | |
383321ab | 245 | set_huge_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte); |
3fe87967 KS |
246 | if (PageAnon(new)) |
247 | hugepage_add_anon_rmap(new, vma, pvmw.address); | |
248 | else | |
249 | page_dup_rmap(new, true); | |
383321ab AK |
250 | } else |
251 | #endif | |
252 | { | |
253 | set_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte); | |
04e62a29 | 254 | |
383321ab AK |
255 | if (PageAnon(new)) |
256 | page_add_anon_rmap(new, vma, pvmw.address, false); | |
257 | else | |
258 | page_add_file_rmap(new, false); | |
259 | } | |
3fe87967 KS |
260 | if (vma->vm_flags & VM_LOCKED && !PageTransCompound(new)) |
261 | mlock_vma_page(new); | |
262 | ||
263 | /* No need to invalidate - it was non-present before */ | |
264 | update_mmu_cache(vma, pvmw.address, pvmw.pte); | |
265 | } | |
51afb12b | 266 | |
e4b82222 | 267 | return true; |
0697212a CL |
268 | } |
269 | ||
04e62a29 CL |
270 | /* |
271 | * Get rid of all migration entries and replace them by | |
272 | * references to the indicated page. | |
273 | */ | |
e388466d | 274 | void remove_migration_ptes(struct page *old, struct page *new, bool locked) |
04e62a29 | 275 | { |
051ac83a JK |
276 | struct rmap_walk_control rwc = { |
277 | .rmap_one = remove_migration_pte, | |
278 | .arg = old, | |
279 | }; | |
280 | ||
e388466d KS |
281 | if (locked) |
282 | rmap_walk_locked(new, &rwc); | |
283 | else | |
284 | rmap_walk(new, &rwc); | |
04e62a29 CL |
285 | } |
286 | ||
0697212a CL |
287 | /* |
288 | * Something used the pte of a page under migration. We need to | |
289 | * get to the page and wait until migration is finished. | |
290 | * When we return from this function the fault will be retried. | |
0697212a | 291 | */ |
e66f17ff | 292 | void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep, |
30dad309 | 293 | spinlock_t *ptl) |
0697212a | 294 | { |
30dad309 | 295 | pte_t pte; |
0697212a CL |
296 | swp_entry_t entry; |
297 | struct page *page; | |
298 | ||
30dad309 | 299 | spin_lock(ptl); |
0697212a CL |
300 | pte = *ptep; |
301 | if (!is_swap_pte(pte)) | |
302 | goto out; | |
303 | ||
304 | entry = pte_to_swp_entry(pte); | |
305 | if (!is_migration_entry(entry)) | |
306 | goto out; | |
307 | ||
308 | page = migration_entry_to_page(entry); | |
309 | ||
e286781d NP |
310 | /* |
311 | * Once radix-tree replacement of page migration started, page_count | |
312 | * *must* be zero. And, we don't want to call wait_on_page_locked() | |
313 | * against a page without get_page(). | |
314 | * So, we use get_page_unless_zero(), here. Even failed, page fault | |
315 | * will occur again. | |
316 | */ | |
317 | if (!get_page_unless_zero(page)) | |
318 | goto out; | |
0697212a CL |
319 | pte_unmap_unlock(ptep, ptl); |
320 | wait_on_page_locked(page); | |
321 | put_page(page); | |
322 | return; | |
323 | out: | |
324 | pte_unmap_unlock(ptep, ptl); | |
325 | } | |
326 | ||
30dad309 NH |
327 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, |
328 | unsigned long address) | |
329 | { | |
330 | spinlock_t *ptl = pte_lockptr(mm, pmd); | |
331 | pte_t *ptep = pte_offset_map(pmd, address); | |
332 | __migration_entry_wait(mm, ptep, ptl); | |
333 | } | |
334 | ||
cb900f41 KS |
335 | void migration_entry_wait_huge(struct vm_area_struct *vma, |
336 | struct mm_struct *mm, pte_t *pte) | |
30dad309 | 337 | { |
cb900f41 | 338 | spinlock_t *ptl = huge_pte_lockptr(hstate_vma(vma), mm, pte); |
30dad309 NH |
339 | __migration_entry_wait(mm, pte, ptl); |
340 | } | |
341 | ||
616b8371 ZY |
342 | #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION |
343 | void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd) | |
344 | { | |
345 | spinlock_t *ptl; | |
346 | struct page *page; | |
347 | ||
348 | ptl = pmd_lock(mm, pmd); | |
349 | if (!is_pmd_migration_entry(*pmd)) | |
350 | goto unlock; | |
351 | page = migration_entry_to_page(pmd_to_swp_entry(*pmd)); | |
352 | if (!get_page_unless_zero(page)) | |
353 | goto unlock; | |
354 | spin_unlock(ptl); | |
355 | wait_on_page_locked(page); | |
356 | put_page(page); | |
357 | return; | |
358 | unlock: | |
359 | spin_unlock(ptl); | |
360 | } | |
361 | #endif | |
362 | ||
b969c4ab MG |
363 | #ifdef CONFIG_BLOCK |
364 | /* Returns true if all buffers are successfully locked */ | |
a6bc32b8 MG |
365 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, |
366 | enum migrate_mode mode) | |
b969c4ab MG |
367 | { |
368 | struct buffer_head *bh = head; | |
369 | ||
370 | /* Simple case, sync compaction */ | |
a6bc32b8 | 371 | if (mode != MIGRATE_ASYNC) { |
b969c4ab MG |
372 | do { |
373 | get_bh(bh); | |
374 | lock_buffer(bh); | |
375 | bh = bh->b_this_page; | |
376 | ||
377 | } while (bh != head); | |
378 | ||
379 | return true; | |
380 | } | |
381 | ||
382 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ | |
383 | do { | |
384 | get_bh(bh); | |
385 | if (!trylock_buffer(bh)) { | |
386 | /* | |
387 | * We failed to lock the buffer and cannot stall in | |
388 | * async migration. Release the taken locks | |
389 | */ | |
390 | struct buffer_head *failed_bh = bh; | |
391 | put_bh(failed_bh); | |
392 | bh = head; | |
393 | while (bh != failed_bh) { | |
394 | unlock_buffer(bh); | |
395 | put_bh(bh); | |
396 | bh = bh->b_this_page; | |
397 | } | |
398 | return false; | |
399 | } | |
400 | ||
401 | bh = bh->b_this_page; | |
402 | } while (bh != head); | |
403 | return true; | |
404 | } | |
405 | #else | |
406 | static inline bool buffer_migrate_lock_buffers(struct buffer_head *head, | |
a6bc32b8 | 407 | enum migrate_mode mode) |
b969c4ab MG |
408 | { |
409 | return true; | |
410 | } | |
411 | #endif /* CONFIG_BLOCK */ | |
412 | ||
b20a3503 | 413 | /* |
c3fcf8a5 | 414 | * Replace the page in the mapping. |
5b5c7120 CL |
415 | * |
416 | * The number of remaining references must be: | |
417 | * 1 for anonymous pages without a mapping | |
418 | * 2 for pages with a mapping | |
266cf658 | 419 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 420 | */ |
36bc08cc | 421 | int migrate_page_move_mapping(struct address_space *mapping, |
b969c4ab | 422 | struct page *newpage, struct page *page, |
8e321fef BL |
423 | struct buffer_head *head, enum migrate_mode mode, |
424 | int extra_count) | |
b20a3503 | 425 | { |
42cb14b1 HD |
426 | struct zone *oldzone, *newzone; |
427 | int dirty; | |
8e321fef | 428 | int expected_count = 1 + extra_count; |
7cf9c2c7 | 429 | void **pslot; |
b20a3503 | 430 | |
6c5240ae | 431 | if (!mapping) { |
0e8c7d0f | 432 | /* Anonymous page without mapping */ |
8e321fef | 433 | if (page_count(page) != expected_count) |
6c5240ae | 434 | return -EAGAIN; |
cf4b769a HD |
435 | |
436 | /* No turning back from here */ | |
cf4b769a HD |
437 | newpage->index = page->index; |
438 | newpage->mapping = page->mapping; | |
439 | if (PageSwapBacked(page)) | |
fa9949da | 440 | __SetPageSwapBacked(newpage); |
cf4b769a | 441 | |
78bd5209 | 442 | return MIGRATEPAGE_SUCCESS; |
6c5240ae CL |
443 | } |
444 | ||
42cb14b1 HD |
445 | oldzone = page_zone(page); |
446 | newzone = page_zone(newpage); | |
447 | ||
19fd6231 | 448 | spin_lock_irq(&mapping->tree_lock); |
b20a3503 | 449 | |
7cf9c2c7 NP |
450 | pslot = radix_tree_lookup_slot(&mapping->page_tree, |
451 | page_index(page)); | |
b20a3503 | 452 | |
8e321fef | 453 | expected_count += 1 + page_has_private(page); |
e286781d | 454 | if (page_count(page) != expected_count || |
29c1f677 | 455 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
19fd6231 | 456 | spin_unlock_irq(&mapping->tree_lock); |
e23ca00b | 457 | return -EAGAIN; |
b20a3503 CL |
458 | } |
459 | ||
fe896d18 | 460 | if (!page_ref_freeze(page, expected_count)) { |
19fd6231 | 461 | spin_unlock_irq(&mapping->tree_lock); |
e286781d NP |
462 | return -EAGAIN; |
463 | } | |
464 | ||
b969c4ab MG |
465 | /* |
466 | * In the async migration case of moving a page with buffers, lock the | |
467 | * buffers using trylock before the mapping is moved. If the mapping | |
468 | * was moved, we later failed to lock the buffers and could not move | |
469 | * the mapping back due to an elevated page count, we would have to | |
470 | * block waiting on other references to be dropped. | |
471 | */ | |
a6bc32b8 MG |
472 | if (mode == MIGRATE_ASYNC && head && |
473 | !buffer_migrate_lock_buffers(head, mode)) { | |
fe896d18 | 474 | page_ref_unfreeze(page, expected_count); |
b969c4ab MG |
475 | spin_unlock_irq(&mapping->tree_lock); |
476 | return -EAGAIN; | |
477 | } | |
478 | ||
b20a3503 | 479 | /* |
cf4b769a HD |
480 | * Now we know that no one else is looking at the page: |
481 | * no turning back from here. | |
b20a3503 | 482 | */ |
cf4b769a HD |
483 | newpage->index = page->index; |
484 | newpage->mapping = page->mapping; | |
7cf9c2c7 | 485 | get_page(newpage); /* add cache reference */ |
6326fec1 NP |
486 | if (PageSwapBacked(page)) { |
487 | __SetPageSwapBacked(newpage); | |
488 | if (PageSwapCache(page)) { | |
489 | SetPageSwapCache(newpage); | |
490 | set_page_private(newpage, page_private(page)); | |
491 | } | |
492 | } else { | |
493 | VM_BUG_ON_PAGE(PageSwapCache(page), page); | |
b20a3503 CL |
494 | } |
495 | ||
42cb14b1 HD |
496 | /* Move dirty while page refs frozen and newpage not yet exposed */ |
497 | dirty = PageDirty(page); | |
498 | if (dirty) { | |
499 | ClearPageDirty(page); | |
500 | SetPageDirty(newpage); | |
501 | } | |
502 | ||
6d75f366 | 503 | radix_tree_replace_slot(&mapping->page_tree, pslot, newpage); |
7cf9c2c7 NP |
504 | |
505 | /* | |
937a94c9 JG |
506 | * Drop cache reference from old page by unfreezing |
507 | * to one less reference. | |
7cf9c2c7 NP |
508 | * We know this isn't the last reference. |
509 | */ | |
fe896d18 | 510 | page_ref_unfreeze(page, expected_count - 1); |
7cf9c2c7 | 511 | |
42cb14b1 HD |
512 | spin_unlock(&mapping->tree_lock); |
513 | /* Leave irq disabled to prevent preemption while updating stats */ | |
514 | ||
0e8c7d0f CL |
515 | /* |
516 | * If moved to a different zone then also account | |
517 | * the page for that zone. Other VM counters will be | |
518 | * taken care of when we establish references to the | |
519 | * new page and drop references to the old page. | |
520 | * | |
521 | * Note that anonymous pages are accounted for | |
4b9d0fab | 522 | * via NR_FILE_PAGES and NR_ANON_MAPPED if they |
0e8c7d0f CL |
523 | * are mapped to swap space. |
524 | */ | |
42cb14b1 | 525 | if (newzone != oldzone) { |
11fb9989 MG |
526 | __dec_node_state(oldzone->zone_pgdat, NR_FILE_PAGES); |
527 | __inc_node_state(newzone->zone_pgdat, NR_FILE_PAGES); | |
42cb14b1 | 528 | if (PageSwapBacked(page) && !PageSwapCache(page)) { |
11fb9989 MG |
529 | __dec_node_state(oldzone->zone_pgdat, NR_SHMEM); |
530 | __inc_node_state(newzone->zone_pgdat, NR_SHMEM); | |
42cb14b1 HD |
531 | } |
532 | if (dirty && mapping_cap_account_dirty(mapping)) { | |
11fb9989 | 533 | __dec_node_state(oldzone->zone_pgdat, NR_FILE_DIRTY); |
5a1c84b4 | 534 | __dec_zone_state(oldzone, NR_ZONE_WRITE_PENDING); |
11fb9989 | 535 | __inc_node_state(newzone->zone_pgdat, NR_FILE_DIRTY); |
5a1c84b4 | 536 | __inc_zone_state(newzone, NR_ZONE_WRITE_PENDING); |
42cb14b1 | 537 | } |
4b02108a | 538 | } |
42cb14b1 | 539 | local_irq_enable(); |
b20a3503 | 540 | |
78bd5209 | 541 | return MIGRATEPAGE_SUCCESS; |
b20a3503 | 542 | } |
1118dce7 | 543 | EXPORT_SYMBOL(migrate_page_move_mapping); |
b20a3503 | 544 | |
290408d4 NH |
545 | /* |
546 | * The expected number of remaining references is the same as that | |
547 | * of migrate_page_move_mapping(). | |
548 | */ | |
549 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
550 | struct page *newpage, struct page *page) | |
551 | { | |
552 | int expected_count; | |
553 | void **pslot; | |
554 | ||
290408d4 NH |
555 | spin_lock_irq(&mapping->tree_lock); |
556 | ||
557 | pslot = radix_tree_lookup_slot(&mapping->page_tree, | |
558 | page_index(page)); | |
559 | ||
560 | expected_count = 2 + page_has_private(page); | |
561 | if (page_count(page) != expected_count || | |
29c1f677 | 562 | radix_tree_deref_slot_protected(pslot, &mapping->tree_lock) != page) { |
290408d4 NH |
563 | spin_unlock_irq(&mapping->tree_lock); |
564 | return -EAGAIN; | |
565 | } | |
566 | ||
fe896d18 | 567 | if (!page_ref_freeze(page, expected_count)) { |
290408d4 NH |
568 | spin_unlock_irq(&mapping->tree_lock); |
569 | return -EAGAIN; | |
570 | } | |
571 | ||
cf4b769a HD |
572 | newpage->index = page->index; |
573 | newpage->mapping = page->mapping; | |
6a93ca8f | 574 | |
290408d4 NH |
575 | get_page(newpage); |
576 | ||
6d75f366 | 577 | radix_tree_replace_slot(&mapping->page_tree, pslot, newpage); |
290408d4 | 578 | |
fe896d18 | 579 | page_ref_unfreeze(page, expected_count - 1); |
290408d4 NH |
580 | |
581 | spin_unlock_irq(&mapping->tree_lock); | |
6a93ca8f | 582 | |
78bd5209 | 583 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
584 | } |
585 | ||
30b0a105 DH |
586 | /* |
587 | * Gigantic pages are so large that we do not guarantee that page++ pointer | |
588 | * arithmetic will work across the entire page. We need something more | |
589 | * specialized. | |
590 | */ | |
591 | static void __copy_gigantic_page(struct page *dst, struct page *src, | |
592 | int nr_pages) | |
593 | { | |
594 | int i; | |
595 | struct page *dst_base = dst; | |
596 | struct page *src_base = src; | |
597 | ||
598 | for (i = 0; i < nr_pages; ) { | |
599 | cond_resched(); | |
600 | copy_highpage(dst, src); | |
601 | ||
602 | i++; | |
603 | dst = mem_map_next(dst, dst_base, i); | |
604 | src = mem_map_next(src, src_base, i); | |
605 | } | |
606 | } | |
607 | ||
608 | static void copy_huge_page(struct page *dst, struct page *src) | |
609 | { | |
610 | int i; | |
611 | int nr_pages; | |
612 | ||
613 | if (PageHuge(src)) { | |
614 | /* hugetlbfs page */ | |
615 | struct hstate *h = page_hstate(src); | |
616 | nr_pages = pages_per_huge_page(h); | |
617 | ||
618 | if (unlikely(nr_pages > MAX_ORDER_NR_PAGES)) { | |
619 | __copy_gigantic_page(dst, src, nr_pages); | |
620 | return; | |
621 | } | |
622 | } else { | |
623 | /* thp page */ | |
624 | BUG_ON(!PageTransHuge(src)); | |
625 | nr_pages = hpage_nr_pages(src); | |
626 | } | |
627 | ||
628 | for (i = 0; i < nr_pages; i++) { | |
629 | cond_resched(); | |
630 | copy_highpage(dst + i, src + i); | |
631 | } | |
632 | } | |
633 | ||
b20a3503 CL |
634 | /* |
635 | * Copy the page to its new location | |
636 | */ | |
2916ecc0 | 637 | void migrate_page_states(struct page *newpage, struct page *page) |
b20a3503 | 638 | { |
7851a45c RR |
639 | int cpupid; |
640 | ||
b20a3503 CL |
641 | if (PageError(page)) |
642 | SetPageError(newpage); | |
643 | if (PageReferenced(page)) | |
644 | SetPageReferenced(newpage); | |
645 | if (PageUptodate(page)) | |
646 | SetPageUptodate(newpage); | |
894bc310 | 647 | if (TestClearPageActive(page)) { |
309381fe | 648 | VM_BUG_ON_PAGE(PageUnevictable(page), page); |
b20a3503 | 649 | SetPageActive(newpage); |
418b27ef LS |
650 | } else if (TestClearPageUnevictable(page)) |
651 | SetPageUnevictable(newpage); | |
b20a3503 CL |
652 | if (PageChecked(page)) |
653 | SetPageChecked(newpage); | |
654 | if (PageMappedToDisk(page)) | |
655 | SetPageMappedToDisk(newpage); | |
656 | ||
42cb14b1 HD |
657 | /* Move dirty on pages not done by migrate_page_move_mapping() */ |
658 | if (PageDirty(page)) | |
659 | SetPageDirty(newpage); | |
b20a3503 | 660 | |
33c3fc71 VD |
661 | if (page_is_young(page)) |
662 | set_page_young(newpage); | |
663 | if (page_is_idle(page)) | |
664 | set_page_idle(newpage); | |
665 | ||
7851a45c RR |
666 | /* |
667 | * Copy NUMA information to the new page, to prevent over-eager | |
668 | * future migrations of this same page. | |
669 | */ | |
670 | cpupid = page_cpupid_xchg_last(page, -1); | |
671 | page_cpupid_xchg_last(newpage, cpupid); | |
672 | ||
e9995ef9 | 673 | ksm_migrate_page(newpage, page); |
c8d6553b HD |
674 | /* |
675 | * Please do not reorder this without considering how mm/ksm.c's | |
676 | * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache(). | |
677 | */ | |
b3b3a99c NH |
678 | if (PageSwapCache(page)) |
679 | ClearPageSwapCache(page); | |
b20a3503 CL |
680 | ClearPagePrivate(page); |
681 | set_page_private(page, 0); | |
b20a3503 CL |
682 | |
683 | /* | |
684 | * If any waiters have accumulated on the new page then | |
685 | * wake them up. | |
686 | */ | |
687 | if (PageWriteback(newpage)) | |
688 | end_page_writeback(newpage); | |
d435edca VB |
689 | |
690 | copy_page_owner(page, newpage); | |
74485cf2 JW |
691 | |
692 | mem_cgroup_migrate(page, newpage); | |
b20a3503 | 693 | } |
2916ecc0 JG |
694 | EXPORT_SYMBOL(migrate_page_states); |
695 | ||
696 | void migrate_page_copy(struct page *newpage, struct page *page) | |
697 | { | |
698 | if (PageHuge(page) || PageTransHuge(page)) | |
699 | copy_huge_page(newpage, page); | |
700 | else | |
701 | copy_highpage(newpage, page); | |
702 | ||
703 | migrate_page_states(newpage, page); | |
704 | } | |
1118dce7 | 705 | EXPORT_SYMBOL(migrate_page_copy); |
b20a3503 | 706 | |
1d8b85cc CL |
707 | /************************************************************ |
708 | * Migration functions | |
709 | ***********************************************************/ | |
710 | ||
b20a3503 | 711 | /* |
bda807d4 | 712 | * Common logic to directly migrate a single LRU page suitable for |
266cf658 | 713 | * pages that do not use PagePrivate/PagePrivate2. |
b20a3503 CL |
714 | * |
715 | * Pages are locked upon entry and exit. | |
716 | */ | |
2d1db3b1 | 717 | int migrate_page(struct address_space *mapping, |
a6bc32b8 MG |
718 | struct page *newpage, struct page *page, |
719 | enum migrate_mode mode) | |
b20a3503 CL |
720 | { |
721 | int rc; | |
722 | ||
723 | BUG_ON(PageWriteback(page)); /* Writeback must be complete */ | |
724 | ||
8e321fef | 725 | rc = migrate_page_move_mapping(mapping, newpage, page, NULL, mode, 0); |
b20a3503 | 726 | |
78bd5209 | 727 | if (rc != MIGRATEPAGE_SUCCESS) |
b20a3503 CL |
728 | return rc; |
729 | ||
2916ecc0 JG |
730 | if (mode != MIGRATE_SYNC_NO_COPY) |
731 | migrate_page_copy(newpage, page); | |
732 | else | |
733 | migrate_page_states(newpage, page); | |
78bd5209 | 734 | return MIGRATEPAGE_SUCCESS; |
b20a3503 CL |
735 | } |
736 | EXPORT_SYMBOL(migrate_page); | |
737 | ||
9361401e | 738 | #ifdef CONFIG_BLOCK |
1d8b85cc CL |
739 | /* |
740 | * Migration function for pages with buffers. This function can only be used | |
741 | * if the underlying filesystem guarantees that no other references to "page" | |
742 | * exist. | |
743 | */ | |
2d1db3b1 | 744 | int buffer_migrate_page(struct address_space *mapping, |
a6bc32b8 | 745 | struct page *newpage, struct page *page, enum migrate_mode mode) |
1d8b85cc | 746 | { |
1d8b85cc CL |
747 | struct buffer_head *bh, *head; |
748 | int rc; | |
749 | ||
1d8b85cc | 750 | if (!page_has_buffers(page)) |
a6bc32b8 | 751 | return migrate_page(mapping, newpage, page, mode); |
1d8b85cc CL |
752 | |
753 | head = page_buffers(page); | |
754 | ||
8e321fef | 755 | rc = migrate_page_move_mapping(mapping, newpage, page, head, mode, 0); |
1d8b85cc | 756 | |
78bd5209 | 757 | if (rc != MIGRATEPAGE_SUCCESS) |
1d8b85cc CL |
758 | return rc; |
759 | ||
b969c4ab MG |
760 | /* |
761 | * In the async case, migrate_page_move_mapping locked the buffers | |
762 | * with an IRQ-safe spinlock held. In the sync case, the buffers | |
763 | * need to be locked now | |
764 | */ | |
a6bc32b8 MG |
765 | if (mode != MIGRATE_ASYNC) |
766 | BUG_ON(!buffer_migrate_lock_buffers(head, mode)); | |
1d8b85cc CL |
767 | |
768 | ClearPagePrivate(page); | |
769 | set_page_private(newpage, page_private(page)); | |
770 | set_page_private(page, 0); | |
771 | put_page(page); | |
772 | get_page(newpage); | |
773 | ||
774 | bh = head; | |
775 | do { | |
776 | set_bh_page(bh, newpage, bh_offset(bh)); | |
777 | bh = bh->b_this_page; | |
778 | ||
779 | } while (bh != head); | |
780 | ||
781 | SetPagePrivate(newpage); | |
782 | ||
2916ecc0 JG |
783 | if (mode != MIGRATE_SYNC_NO_COPY) |
784 | migrate_page_copy(newpage, page); | |
785 | else | |
786 | migrate_page_states(newpage, page); | |
1d8b85cc CL |
787 | |
788 | bh = head; | |
789 | do { | |
790 | unlock_buffer(bh); | |
2916ecc0 | 791 | put_bh(bh); |
1d8b85cc CL |
792 | bh = bh->b_this_page; |
793 | ||
794 | } while (bh != head); | |
795 | ||
78bd5209 | 796 | return MIGRATEPAGE_SUCCESS; |
1d8b85cc CL |
797 | } |
798 | EXPORT_SYMBOL(buffer_migrate_page); | |
9361401e | 799 | #endif |
1d8b85cc | 800 | |
04e62a29 CL |
801 | /* |
802 | * Writeback a page to clean the dirty state | |
803 | */ | |
804 | static int writeout(struct address_space *mapping, struct page *page) | |
8351a6e4 | 805 | { |
04e62a29 CL |
806 | struct writeback_control wbc = { |
807 | .sync_mode = WB_SYNC_NONE, | |
808 | .nr_to_write = 1, | |
809 | .range_start = 0, | |
810 | .range_end = LLONG_MAX, | |
04e62a29 CL |
811 | .for_reclaim = 1 |
812 | }; | |
813 | int rc; | |
814 | ||
815 | if (!mapping->a_ops->writepage) | |
816 | /* No write method for the address space */ | |
817 | return -EINVAL; | |
818 | ||
819 | if (!clear_page_dirty_for_io(page)) | |
820 | /* Someone else already triggered a write */ | |
821 | return -EAGAIN; | |
822 | ||
8351a6e4 | 823 | /* |
04e62a29 CL |
824 | * A dirty page may imply that the underlying filesystem has |
825 | * the page on some queue. So the page must be clean for | |
826 | * migration. Writeout may mean we loose the lock and the | |
827 | * page state is no longer what we checked for earlier. | |
828 | * At this point we know that the migration attempt cannot | |
829 | * be successful. | |
8351a6e4 | 830 | */ |
e388466d | 831 | remove_migration_ptes(page, page, false); |
8351a6e4 | 832 | |
04e62a29 | 833 | rc = mapping->a_ops->writepage(page, &wbc); |
8351a6e4 | 834 | |
04e62a29 CL |
835 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
836 | /* unlocked. Relock */ | |
837 | lock_page(page); | |
838 | ||
bda8550d | 839 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
840 | } |
841 | ||
842 | /* | |
843 | * Default handling if a filesystem does not provide a migration function. | |
844 | */ | |
845 | static int fallback_migrate_page(struct address_space *mapping, | |
a6bc32b8 | 846 | struct page *newpage, struct page *page, enum migrate_mode mode) |
04e62a29 | 847 | { |
b969c4ab | 848 | if (PageDirty(page)) { |
a6bc32b8 | 849 | /* Only writeback pages in full synchronous migration */ |
2916ecc0 JG |
850 | switch (mode) { |
851 | case MIGRATE_SYNC: | |
852 | case MIGRATE_SYNC_NO_COPY: | |
853 | break; | |
854 | default: | |
b969c4ab | 855 | return -EBUSY; |
2916ecc0 | 856 | } |
04e62a29 | 857 | return writeout(mapping, page); |
b969c4ab | 858 | } |
8351a6e4 CL |
859 | |
860 | /* | |
861 | * Buffers may be managed in a filesystem specific way. | |
862 | * We must have no buffers or drop them. | |
863 | */ | |
266cf658 | 864 | if (page_has_private(page) && |
8351a6e4 CL |
865 | !try_to_release_page(page, GFP_KERNEL)) |
866 | return -EAGAIN; | |
867 | ||
a6bc32b8 | 868 | return migrate_page(mapping, newpage, page, mode); |
8351a6e4 CL |
869 | } |
870 | ||
e24f0b8f CL |
871 | /* |
872 | * Move a page to a newly allocated page | |
873 | * The page is locked and all ptes have been successfully removed. | |
874 | * | |
875 | * The new page will have replaced the old page if this function | |
876 | * is successful. | |
894bc310 LS |
877 | * |
878 | * Return value: | |
879 | * < 0 - error code | |
78bd5209 | 880 | * MIGRATEPAGE_SUCCESS - success |
e24f0b8f | 881 | */ |
3fe2011f | 882 | static int move_to_new_page(struct page *newpage, struct page *page, |
5c3f9a67 | 883 | enum migrate_mode mode) |
e24f0b8f CL |
884 | { |
885 | struct address_space *mapping; | |
bda807d4 MK |
886 | int rc = -EAGAIN; |
887 | bool is_lru = !__PageMovable(page); | |
e24f0b8f | 888 | |
7db7671f HD |
889 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
890 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); | |
e24f0b8f | 891 | |
e24f0b8f | 892 | mapping = page_mapping(page); |
bda807d4 MK |
893 | |
894 | if (likely(is_lru)) { | |
895 | if (!mapping) | |
896 | rc = migrate_page(mapping, newpage, page, mode); | |
897 | else if (mapping->a_ops->migratepage) | |
898 | /* | |
899 | * Most pages have a mapping and most filesystems | |
900 | * provide a migratepage callback. Anonymous pages | |
901 | * are part of swap space which also has its own | |
902 | * migratepage callback. This is the most common path | |
903 | * for page migration. | |
904 | */ | |
905 | rc = mapping->a_ops->migratepage(mapping, newpage, | |
906 | page, mode); | |
907 | else | |
908 | rc = fallback_migrate_page(mapping, newpage, | |
909 | page, mode); | |
910 | } else { | |
e24f0b8f | 911 | /* |
bda807d4 MK |
912 | * In case of non-lru page, it could be released after |
913 | * isolation step. In that case, we shouldn't try migration. | |
e24f0b8f | 914 | */ |
bda807d4 MK |
915 | VM_BUG_ON_PAGE(!PageIsolated(page), page); |
916 | if (!PageMovable(page)) { | |
917 | rc = MIGRATEPAGE_SUCCESS; | |
918 | __ClearPageIsolated(page); | |
919 | goto out; | |
920 | } | |
921 | ||
922 | rc = mapping->a_ops->migratepage(mapping, newpage, | |
923 | page, mode); | |
924 | WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS && | |
925 | !PageIsolated(page)); | |
926 | } | |
e24f0b8f | 927 | |
5c3f9a67 HD |
928 | /* |
929 | * When successful, old pagecache page->mapping must be cleared before | |
930 | * page is freed; but stats require that PageAnon be left as PageAnon. | |
931 | */ | |
932 | if (rc == MIGRATEPAGE_SUCCESS) { | |
bda807d4 MK |
933 | if (__PageMovable(page)) { |
934 | VM_BUG_ON_PAGE(!PageIsolated(page), page); | |
935 | ||
936 | /* | |
937 | * We clear PG_movable under page_lock so any compactor | |
938 | * cannot try to migrate this page. | |
939 | */ | |
940 | __ClearPageIsolated(page); | |
941 | } | |
942 | ||
943 | /* | |
944 | * Anonymous and movable page->mapping will be cleard by | |
945 | * free_pages_prepare so don't reset it here for keeping | |
946 | * the type to work PageAnon, for example. | |
947 | */ | |
948 | if (!PageMappingFlags(page)) | |
5c3f9a67 | 949 | page->mapping = NULL; |
3fe2011f | 950 | } |
bda807d4 | 951 | out: |
e24f0b8f CL |
952 | return rc; |
953 | } | |
954 | ||
0dabec93 | 955 | static int __unmap_and_move(struct page *page, struct page *newpage, |
9c620e2b | 956 | int force, enum migrate_mode mode) |
e24f0b8f | 957 | { |
0dabec93 | 958 | int rc = -EAGAIN; |
2ebba6b7 | 959 | int page_was_mapped = 0; |
3f6c8272 | 960 | struct anon_vma *anon_vma = NULL; |
bda807d4 | 961 | bool is_lru = !__PageMovable(page); |
95a402c3 | 962 | |
529ae9aa | 963 | if (!trylock_page(page)) { |
a6bc32b8 | 964 | if (!force || mode == MIGRATE_ASYNC) |
0dabec93 | 965 | goto out; |
3e7d3449 MG |
966 | |
967 | /* | |
968 | * It's not safe for direct compaction to call lock_page. | |
969 | * For example, during page readahead pages are added locked | |
970 | * to the LRU. Later, when the IO completes the pages are | |
971 | * marked uptodate and unlocked. However, the queueing | |
972 | * could be merging multiple pages for one bio (e.g. | |
973 | * mpage_readpages). If an allocation happens for the | |
974 | * second or third page, the process can end up locking | |
975 | * the same page twice and deadlocking. Rather than | |
976 | * trying to be clever about what pages can be locked, | |
977 | * avoid the use of lock_page for direct compaction | |
978 | * altogether. | |
979 | */ | |
980 | if (current->flags & PF_MEMALLOC) | |
0dabec93 | 981 | goto out; |
3e7d3449 | 982 | |
e24f0b8f CL |
983 | lock_page(page); |
984 | } | |
985 | ||
986 | if (PageWriteback(page)) { | |
11bc82d6 | 987 | /* |
fed5b64a | 988 | * Only in the case of a full synchronous migration is it |
a6bc32b8 MG |
989 | * necessary to wait for PageWriteback. In the async case, |
990 | * the retry loop is too short and in the sync-light case, | |
991 | * the overhead of stalling is too much | |
11bc82d6 | 992 | */ |
2916ecc0 JG |
993 | switch (mode) { |
994 | case MIGRATE_SYNC: | |
995 | case MIGRATE_SYNC_NO_COPY: | |
996 | break; | |
997 | default: | |
11bc82d6 | 998 | rc = -EBUSY; |
0a31bc97 | 999 | goto out_unlock; |
11bc82d6 AA |
1000 | } |
1001 | if (!force) | |
0a31bc97 | 1002 | goto out_unlock; |
e24f0b8f CL |
1003 | wait_on_page_writeback(page); |
1004 | } | |
03f15c86 | 1005 | |
e24f0b8f | 1006 | /* |
dc386d4d KH |
1007 | * By try_to_unmap(), page->mapcount goes down to 0 here. In this case, |
1008 | * we cannot notice that anon_vma is freed while we migrates a page. | |
1ce82b69 | 1009 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
dc386d4d | 1010 | * of migration. File cache pages are no problem because of page_lock() |
989f89c5 KH |
1011 | * File Caches may use write_page() or lock_page() in migration, then, |
1012 | * just care Anon page here. | |
03f15c86 HD |
1013 | * |
1014 | * Only page_get_anon_vma() understands the subtleties of | |
1015 | * getting a hold on an anon_vma from outside one of its mms. | |
1016 | * But if we cannot get anon_vma, then we won't need it anyway, | |
1017 | * because that implies that the anon page is no longer mapped | |
1018 | * (and cannot be remapped so long as we hold the page lock). | |
dc386d4d | 1019 | */ |
03f15c86 | 1020 | if (PageAnon(page) && !PageKsm(page)) |
746b18d4 | 1021 | anon_vma = page_get_anon_vma(page); |
62e1c553 | 1022 | |
7db7671f HD |
1023 | /* |
1024 | * Block others from accessing the new page when we get around to | |
1025 | * establishing additional references. We are usually the only one | |
1026 | * holding a reference to newpage at this point. We used to have a BUG | |
1027 | * here if trylock_page(newpage) fails, but would like to allow for | |
1028 | * cases where there might be a race with the previous use of newpage. | |
1029 | * This is much like races on refcount of oldpage: just don't BUG(). | |
1030 | */ | |
1031 | if (unlikely(!trylock_page(newpage))) | |
1032 | goto out_unlock; | |
1033 | ||
bda807d4 MK |
1034 | if (unlikely(!is_lru)) { |
1035 | rc = move_to_new_page(newpage, page, mode); | |
1036 | goto out_unlock_both; | |
1037 | } | |
1038 | ||
dc386d4d | 1039 | /* |
62e1c553 SL |
1040 | * Corner case handling: |
1041 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
1042 | * and treated as swapcache but it has no rmap yet. | |
1043 | * Calling try_to_unmap() against a page->mapping==NULL page will | |
1044 | * trigger a BUG. So handle it here. | |
1045 | * 2. An orphaned page (see truncate_complete_page) might have | |
1046 | * fs-private metadata. The page can be picked up due to memory | |
1047 | * offlining. Everywhere else except page reclaim, the page is | |
1048 | * invisible to the vm, so the page can not be migrated. So try to | |
1049 | * free the metadata, so the page can be freed. | |
e24f0b8f | 1050 | */ |
62e1c553 | 1051 | if (!page->mapping) { |
309381fe | 1052 | VM_BUG_ON_PAGE(PageAnon(page), page); |
1ce82b69 | 1053 | if (page_has_private(page)) { |
62e1c553 | 1054 | try_to_free_buffers(page); |
7db7671f | 1055 | goto out_unlock_both; |
62e1c553 | 1056 | } |
7db7671f HD |
1057 | } else if (page_mapped(page)) { |
1058 | /* Establish migration ptes */ | |
03f15c86 HD |
1059 | VM_BUG_ON_PAGE(PageAnon(page) && !PageKsm(page) && !anon_vma, |
1060 | page); | |
2ebba6b7 | 1061 | try_to_unmap(page, |
da1b13cc | 1062 | TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); |
2ebba6b7 HD |
1063 | page_was_mapped = 1; |
1064 | } | |
dc386d4d | 1065 | |
e6a1530d | 1066 | if (!page_mapped(page)) |
5c3f9a67 | 1067 | rc = move_to_new_page(newpage, page, mode); |
e24f0b8f | 1068 | |
5c3f9a67 HD |
1069 | if (page_was_mapped) |
1070 | remove_migration_ptes(page, | |
e388466d | 1071 | rc == MIGRATEPAGE_SUCCESS ? newpage : page, false); |
3f6c8272 | 1072 | |
7db7671f HD |
1073 | out_unlock_both: |
1074 | unlock_page(newpage); | |
1075 | out_unlock: | |
3f6c8272 | 1076 | /* Drop an anon_vma reference if we took one */ |
76545066 | 1077 | if (anon_vma) |
9e60109f | 1078 | put_anon_vma(anon_vma); |
e24f0b8f | 1079 | unlock_page(page); |
0dabec93 | 1080 | out: |
c6c919eb MK |
1081 | /* |
1082 | * If migration is successful, decrease refcount of the newpage | |
1083 | * which will not free the page because new page owner increased | |
1084 | * refcounter. As well, if it is LRU page, add the page to LRU | |
1085 | * list in here. | |
1086 | */ | |
1087 | if (rc == MIGRATEPAGE_SUCCESS) { | |
b1123ea6 | 1088 | if (unlikely(__PageMovable(newpage))) |
c6c919eb MK |
1089 | put_page(newpage); |
1090 | else | |
1091 | putback_lru_page(newpage); | |
1092 | } | |
1093 | ||
0dabec93 MK |
1094 | return rc; |
1095 | } | |
95a402c3 | 1096 | |
ef2a5153 GU |
1097 | /* |
1098 | * gcc 4.7 and 4.8 on arm get an ICEs when inlining unmap_and_move(). Work | |
1099 | * around it. | |
1100 | */ | |
1101 | #if (GCC_VERSION >= 40700 && GCC_VERSION < 40900) && defined(CONFIG_ARM) | |
1102 | #define ICE_noinline noinline | |
1103 | #else | |
1104 | #define ICE_noinline | |
1105 | #endif | |
1106 | ||
0dabec93 MK |
1107 | /* |
1108 | * Obtain the lock on page, remove all ptes and migrate the page | |
1109 | * to the newly allocated page in newpage. | |
1110 | */ | |
ef2a5153 GU |
1111 | static ICE_noinline int unmap_and_move(new_page_t get_new_page, |
1112 | free_page_t put_new_page, | |
1113 | unsigned long private, struct page *page, | |
add05cec NH |
1114 | int force, enum migrate_mode mode, |
1115 | enum migrate_reason reason) | |
0dabec93 | 1116 | { |
2def7424 | 1117 | int rc = MIGRATEPAGE_SUCCESS; |
0dabec93 | 1118 | int *result = NULL; |
2def7424 | 1119 | struct page *newpage; |
0dabec93 | 1120 | |
2def7424 | 1121 | newpage = get_new_page(page, private, &result); |
0dabec93 MK |
1122 | if (!newpage) |
1123 | return -ENOMEM; | |
1124 | ||
1125 | if (page_count(page) == 1) { | |
1126 | /* page was freed from under us. So we are done. */ | |
c6c919eb MK |
1127 | ClearPageActive(page); |
1128 | ClearPageUnevictable(page); | |
bda807d4 MK |
1129 | if (unlikely(__PageMovable(page))) { |
1130 | lock_page(page); | |
1131 | if (!PageMovable(page)) | |
1132 | __ClearPageIsolated(page); | |
1133 | unlock_page(page); | |
1134 | } | |
c6c919eb MK |
1135 | if (put_new_page) |
1136 | put_new_page(newpage, private); | |
1137 | else | |
1138 | put_page(newpage); | |
0dabec93 MK |
1139 | goto out; |
1140 | } | |
1141 | ||
616b8371 | 1142 | if (unlikely(PageTransHuge(page) && !PageTransHuge(newpage))) { |
4d2fa965 KS |
1143 | lock_page(page); |
1144 | rc = split_huge_page(page); | |
1145 | unlock_page(page); | |
1146 | if (rc) | |
0dabec93 | 1147 | goto out; |
4d2fa965 | 1148 | } |
0dabec93 | 1149 | |
9c620e2b | 1150 | rc = __unmap_and_move(page, newpage, force, mode); |
c6c919eb | 1151 | if (rc == MIGRATEPAGE_SUCCESS) |
7cd12b4a | 1152 | set_page_owner_migrate_reason(newpage, reason); |
bf6bddf1 | 1153 | |
0dabec93 | 1154 | out: |
e24f0b8f | 1155 | if (rc != -EAGAIN) { |
0dabec93 MK |
1156 | /* |
1157 | * A page that has been migrated has all references | |
1158 | * removed and will be freed. A page that has not been | |
1159 | * migrated will have kepts its references and be | |
1160 | * restored. | |
1161 | */ | |
1162 | list_del(&page->lru); | |
6afcf8ef ML |
1163 | |
1164 | /* | |
1165 | * Compaction can migrate also non-LRU pages which are | |
1166 | * not accounted to NR_ISOLATED_*. They can be recognized | |
1167 | * as __PageMovable | |
1168 | */ | |
1169 | if (likely(!__PageMovable(page))) | |
e8db67eb NH |
1170 | mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + |
1171 | page_is_file_cache(page), -hpage_nr_pages(page)); | |
c6c919eb MK |
1172 | } |
1173 | ||
1174 | /* | |
1175 | * If migration is successful, releases reference grabbed during | |
1176 | * isolation. Otherwise, restore the page to right list unless | |
1177 | * we want to retry. | |
1178 | */ | |
1179 | if (rc == MIGRATEPAGE_SUCCESS) { | |
1180 | put_page(page); | |
1181 | if (reason == MR_MEMORY_FAILURE) { | |
d7e69488 | 1182 | /* |
c6c919eb MK |
1183 | * Set PG_HWPoison on just freed page |
1184 | * intentionally. Although it's rather weird, | |
1185 | * it's how HWPoison flag works at the moment. | |
d7e69488 | 1186 | */ |
da1b13cc WL |
1187 | if (!test_set_page_hwpoison(page)) |
1188 | num_poisoned_pages_inc(); | |
c6c919eb MK |
1189 | } |
1190 | } else { | |
bda807d4 MK |
1191 | if (rc != -EAGAIN) { |
1192 | if (likely(!__PageMovable(page))) { | |
1193 | putback_lru_page(page); | |
1194 | goto put_new; | |
1195 | } | |
1196 | ||
1197 | lock_page(page); | |
1198 | if (PageMovable(page)) | |
1199 | putback_movable_page(page); | |
1200 | else | |
1201 | __ClearPageIsolated(page); | |
1202 | unlock_page(page); | |
1203 | put_page(page); | |
1204 | } | |
1205 | put_new: | |
c6c919eb MK |
1206 | if (put_new_page) |
1207 | put_new_page(newpage, private); | |
1208 | else | |
1209 | put_page(newpage); | |
e24f0b8f | 1210 | } |
68711a74 | 1211 | |
742755a1 CL |
1212 | if (result) { |
1213 | if (rc) | |
1214 | *result = rc; | |
1215 | else | |
1216 | *result = page_to_nid(newpage); | |
1217 | } | |
e24f0b8f CL |
1218 | return rc; |
1219 | } | |
1220 | ||
290408d4 NH |
1221 | /* |
1222 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
1223 | * | |
1224 | * This function doesn't wait the completion of hugepage I/O | |
1225 | * because there is no race between I/O and migration for hugepage. | |
1226 | * Note that currently hugepage I/O occurs only in direct I/O | |
1227 | * where no lock is held and PG_writeback is irrelevant, | |
1228 | * and writeback status of all subpages are counted in the reference | |
1229 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
1230 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
1231 | * This means that when we try to migrate hugepage whose subpages are | |
1232 | * doing direct I/O, some references remain after try_to_unmap() and | |
1233 | * hugepage migration fails without data corruption. | |
1234 | * | |
1235 | * There is also no race when direct I/O is issued on the page under migration, | |
1236 | * because then pte is replaced with migration swap entry and direct I/O code | |
1237 | * will wait in the page fault for migration to complete. | |
1238 | */ | |
1239 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
68711a74 DR |
1240 | free_page_t put_new_page, unsigned long private, |
1241 | struct page *hpage, int force, | |
7cd12b4a | 1242 | enum migrate_mode mode, int reason) |
290408d4 | 1243 | { |
2def7424 | 1244 | int rc = -EAGAIN; |
290408d4 | 1245 | int *result = NULL; |
2ebba6b7 | 1246 | int page_was_mapped = 0; |
32665f2b | 1247 | struct page *new_hpage; |
290408d4 NH |
1248 | struct anon_vma *anon_vma = NULL; |
1249 | ||
83467efb NH |
1250 | /* |
1251 | * Movability of hugepages depends on architectures and hugepage size. | |
1252 | * This check is necessary because some callers of hugepage migration | |
1253 | * like soft offline and memory hotremove don't walk through page | |
1254 | * tables or check whether the hugepage is pmd-based or not before | |
1255 | * kicking migration. | |
1256 | */ | |
100873d7 | 1257 | if (!hugepage_migration_supported(page_hstate(hpage))) { |
32665f2b | 1258 | putback_active_hugepage(hpage); |
83467efb | 1259 | return -ENOSYS; |
32665f2b | 1260 | } |
83467efb | 1261 | |
32665f2b | 1262 | new_hpage = get_new_page(hpage, private, &result); |
290408d4 NH |
1263 | if (!new_hpage) |
1264 | return -ENOMEM; | |
1265 | ||
290408d4 | 1266 | if (!trylock_page(hpage)) { |
2916ecc0 | 1267 | if (!force) |
290408d4 | 1268 | goto out; |
2916ecc0 JG |
1269 | switch (mode) { |
1270 | case MIGRATE_SYNC: | |
1271 | case MIGRATE_SYNC_NO_COPY: | |
1272 | break; | |
1273 | default: | |
1274 | goto out; | |
1275 | } | |
290408d4 NH |
1276 | lock_page(hpage); |
1277 | } | |
1278 | ||
746b18d4 PZ |
1279 | if (PageAnon(hpage)) |
1280 | anon_vma = page_get_anon_vma(hpage); | |
290408d4 | 1281 | |
7db7671f HD |
1282 | if (unlikely(!trylock_page(new_hpage))) |
1283 | goto put_anon; | |
1284 | ||
2ebba6b7 HD |
1285 | if (page_mapped(hpage)) { |
1286 | try_to_unmap(hpage, | |
1287 | TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS); | |
1288 | page_was_mapped = 1; | |
1289 | } | |
290408d4 NH |
1290 | |
1291 | if (!page_mapped(hpage)) | |
5c3f9a67 | 1292 | rc = move_to_new_page(new_hpage, hpage, mode); |
290408d4 | 1293 | |
5c3f9a67 HD |
1294 | if (page_was_mapped) |
1295 | remove_migration_ptes(hpage, | |
e388466d | 1296 | rc == MIGRATEPAGE_SUCCESS ? new_hpage : hpage, false); |
290408d4 | 1297 | |
7db7671f HD |
1298 | unlock_page(new_hpage); |
1299 | ||
1300 | put_anon: | |
fd4a4663 | 1301 | if (anon_vma) |
9e60109f | 1302 | put_anon_vma(anon_vma); |
8e6ac7fa | 1303 | |
2def7424 | 1304 | if (rc == MIGRATEPAGE_SUCCESS) { |
8e6ac7fa | 1305 | hugetlb_cgroup_migrate(hpage, new_hpage); |
2def7424 | 1306 | put_new_page = NULL; |
7cd12b4a | 1307 | set_page_owner_migrate_reason(new_hpage, reason); |
2def7424 | 1308 | } |
8e6ac7fa | 1309 | |
290408d4 | 1310 | unlock_page(hpage); |
09761333 | 1311 | out: |
b8ec1cee NH |
1312 | if (rc != -EAGAIN) |
1313 | putback_active_hugepage(hpage); | |
c3114a84 AK |
1314 | if (reason == MR_MEMORY_FAILURE && !test_set_page_hwpoison(hpage)) |
1315 | num_poisoned_pages_inc(); | |
68711a74 DR |
1316 | |
1317 | /* | |
1318 | * If migration was not successful and there's a freeing callback, use | |
1319 | * it. Otherwise, put_page() will drop the reference grabbed during | |
1320 | * isolation. | |
1321 | */ | |
2def7424 | 1322 | if (put_new_page) |
68711a74 DR |
1323 | put_new_page(new_hpage, private); |
1324 | else | |
3aaa76e1 | 1325 | putback_active_hugepage(new_hpage); |
68711a74 | 1326 | |
290408d4 NH |
1327 | if (result) { |
1328 | if (rc) | |
1329 | *result = rc; | |
1330 | else | |
1331 | *result = page_to_nid(new_hpage); | |
1332 | } | |
1333 | return rc; | |
1334 | } | |
1335 | ||
b20a3503 | 1336 | /* |
c73e5c9c SB |
1337 | * migrate_pages - migrate the pages specified in a list, to the free pages |
1338 | * supplied as the target for the page migration | |
b20a3503 | 1339 | * |
c73e5c9c SB |
1340 | * @from: The list of pages to be migrated. |
1341 | * @get_new_page: The function used to allocate free pages to be used | |
1342 | * as the target of the page migration. | |
68711a74 DR |
1343 | * @put_new_page: The function used to free target pages if migration |
1344 | * fails, or NULL if no special handling is necessary. | |
c73e5c9c SB |
1345 | * @private: Private data to be passed on to get_new_page() |
1346 | * @mode: The migration mode that specifies the constraints for | |
1347 | * page migration, if any. | |
1348 | * @reason: The reason for page migration. | |
b20a3503 | 1349 | * |
c73e5c9c SB |
1350 | * The function returns after 10 attempts or if no pages are movable any more |
1351 | * because the list has become empty or no retryable pages exist any more. | |
14e0f9bc | 1352 | * The caller should call putback_movable_pages() to return pages to the LRU |
28bd6578 | 1353 | * or free list only if ret != 0. |
b20a3503 | 1354 | * |
c73e5c9c | 1355 | * Returns the number of pages that were not migrated, or an error code. |
b20a3503 | 1356 | */ |
9c620e2b | 1357 | int migrate_pages(struct list_head *from, new_page_t get_new_page, |
68711a74 DR |
1358 | free_page_t put_new_page, unsigned long private, |
1359 | enum migrate_mode mode, int reason) | |
b20a3503 | 1360 | { |
e24f0b8f | 1361 | int retry = 1; |
b20a3503 | 1362 | int nr_failed = 0; |
5647bc29 | 1363 | int nr_succeeded = 0; |
b20a3503 CL |
1364 | int pass = 0; |
1365 | struct page *page; | |
1366 | struct page *page2; | |
1367 | int swapwrite = current->flags & PF_SWAPWRITE; | |
1368 | int rc; | |
1369 | ||
1370 | if (!swapwrite) | |
1371 | current->flags |= PF_SWAPWRITE; | |
1372 | ||
e24f0b8f CL |
1373 | for(pass = 0; pass < 10 && retry; pass++) { |
1374 | retry = 0; | |
b20a3503 | 1375 | |
e24f0b8f | 1376 | list_for_each_entry_safe(page, page2, from, lru) { |
e24f0b8f | 1377 | cond_resched(); |
2d1db3b1 | 1378 | |
31caf665 NH |
1379 | if (PageHuge(page)) |
1380 | rc = unmap_and_move_huge_page(get_new_page, | |
68711a74 | 1381 | put_new_page, private, page, |
7cd12b4a | 1382 | pass > 2, mode, reason); |
31caf665 | 1383 | else |
68711a74 | 1384 | rc = unmap_and_move(get_new_page, put_new_page, |
add05cec NH |
1385 | private, page, pass > 2, mode, |
1386 | reason); | |
2d1db3b1 | 1387 | |
e24f0b8f | 1388 | switch(rc) { |
95a402c3 | 1389 | case -ENOMEM: |
dfef2ef4 | 1390 | nr_failed++; |
95a402c3 | 1391 | goto out; |
e24f0b8f | 1392 | case -EAGAIN: |
2d1db3b1 | 1393 | retry++; |
e24f0b8f | 1394 | break; |
78bd5209 | 1395 | case MIGRATEPAGE_SUCCESS: |
5647bc29 | 1396 | nr_succeeded++; |
e24f0b8f CL |
1397 | break; |
1398 | default: | |
354a3363 NH |
1399 | /* |
1400 | * Permanent failure (-EBUSY, -ENOSYS, etc.): | |
1401 | * unlike -EAGAIN case, the failed page is | |
1402 | * removed from migration page list and not | |
1403 | * retried in the next outer loop. | |
1404 | */ | |
2d1db3b1 | 1405 | nr_failed++; |
e24f0b8f | 1406 | break; |
2d1db3b1 | 1407 | } |
b20a3503 CL |
1408 | } |
1409 | } | |
f2f81fb2 VB |
1410 | nr_failed += retry; |
1411 | rc = nr_failed; | |
95a402c3 | 1412 | out: |
5647bc29 MG |
1413 | if (nr_succeeded) |
1414 | count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded); | |
1415 | if (nr_failed) | |
1416 | count_vm_events(PGMIGRATE_FAIL, nr_failed); | |
7b2a2d4a MG |
1417 | trace_mm_migrate_pages(nr_succeeded, nr_failed, mode, reason); |
1418 | ||
b20a3503 CL |
1419 | if (!swapwrite) |
1420 | current->flags &= ~PF_SWAPWRITE; | |
1421 | ||
78bd5209 | 1422 | return rc; |
b20a3503 | 1423 | } |
95a402c3 | 1424 | |
742755a1 CL |
1425 | #ifdef CONFIG_NUMA |
1426 | /* | |
1427 | * Move a list of individual pages | |
1428 | */ | |
1429 | struct page_to_node { | |
1430 | unsigned long addr; | |
1431 | struct page *page; | |
1432 | int node; | |
1433 | int status; | |
1434 | }; | |
1435 | ||
1436 | static struct page *new_page_node(struct page *p, unsigned long private, | |
1437 | int **result) | |
1438 | { | |
1439 | struct page_to_node *pm = (struct page_to_node *)private; | |
1440 | ||
1441 | while (pm->node != MAX_NUMNODES && pm->page != p) | |
1442 | pm++; | |
1443 | ||
1444 | if (pm->node == MAX_NUMNODES) | |
1445 | return NULL; | |
1446 | ||
1447 | *result = &pm->status; | |
1448 | ||
e632a938 NH |
1449 | if (PageHuge(p)) |
1450 | return alloc_huge_page_node(page_hstate(compound_head(p)), | |
1451 | pm->node); | |
e8db67eb NH |
1452 | else if (thp_migration_supported() && PageTransHuge(p)) { |
1453 | struct page *thp; | |
1454 | ||
1455 | thp = alloc_pages_node(pm->node, | |
1456 | (GFP_TRANSHUGE | __GFP_THISNODE) & ~__GFP_RECLAIM, | |
1457 | HPAGE_PMD_ORDER); | |
1458 | if (!thp) | |
1459 | return NULL; | |
1460 | prep_transhuge_page(thp); | |
1461 | return thp; | |
1462 | } else | |
96db800f | 1463 | return __alloc_pages_node(pm->node, |
e97ca8e5 | 1464 | GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, 0); |
742755a1 CL |
1465 | } |
1466 | ||
1467 | /* | |
1468 | * Move a set of pages as indicated in the pm array. The addr | |
1469 | * field must be set to the virtual address of the page to be moved | |
1470 | * and the node number must contain a valid target node. | |
5e9a0f02 | 1471 | * The pm array ends with node = MAX_NUMNODES. |
742755a1 | 1472 | */ |
5e9a0f02 BG |
1473 | static int do_move_page_to_node_array(struct mm_struct *mm, |
1474 | struct page_to_node *pm, | |
1475 | int migrate_all) | |
742755a1 CL |
1476 | { |
1477 | int err; | |
1478 | struct page_to_node *pp; | |
1479 | LIST_HEAD(pagelist); | |
1480 | ||
1481 | down_read(&mm->mmap_sem); | |
1482 | ||
1483 | /* | |
1484 | * Build a list of pages to migrate | |
1485 | */ | |
742755a1 CL |
1486 | for (pp = pm; pp->node != MAX_NUMNODES; pp++) { |
1487 | struct vm_area_struct *vma; | |
1488 | struct page *page; | |
e8db67eb NH |
1489 | struct page *head; |
1490 | unsigned int follflags; | |
742755a1 | 1491 | |
742755a1 CL |
1492 | err = -EFAULT; |
1493 | vma = find_vma(mm, pp->addr); | |
70384dc6 | 1494 | if (!vma || pp->addr < vma->vm_start || !vma_migratable(vma)) |
742755a1 CL |
1495 | goto set_status; |
1496 | ||
d899844e | 1497 | /* FOLL_DUMP to ignore special (like zero) pages */ |
e8db67eb NH |
1498 | follflags = FOLL_GET | FOLL_DUMP; |
1499 | if (!thp_migration_supported()) | |
1500 | follflags |= FOLL_SPLIT; | |
1501 | page = follow_page(vma, pp->addr, follflags); | |
89f5b7da LT |
1502 | |
1503 | err = PTR_ERR(page); | |
1504 | if (IS_ERR(page)) | |
1505 | goto set_status; | |
1506 | ||
742755a1 CL |
1507 | err = -ENOENT; |
1508 | if (!page) | |
1509 | goto set_status; | |
1510 | ||
742755a1 CL |
1511 | err = page_to_nid(page); |
1512 | ||
1513 | if (err == pp->node) | |
1514 | /* | |
1515 | * Node already in the right place | |
1516 | */ | |
1517 | goto put_and_set; | |
1518 | ||
1519 | err = -EACCES; | |
1520 | if (page_mapcount(page) > 1 && | |
1521 | !migrate_all) | |
1522 | goto put_and_set; | |
1523 | ||
e632a938 | 1524 | if (PageHuge(page)) { |
e8db67eb | 1525 | if (PageHead(page)) { |
e66f17ff | 1526 | isolate_huge_page(page, &pagelist); |
e8db67eb NH |
1527 | err = 0; |
1528 | pp->page = page; | |
1529 | } | |
e632a938 NH |
1530 | goto put_and_set; |
1531 | } | |
1532 | ||
e8db67eb NH |
1533 | pp->page = compound_head(page); |
1534 | head = compound_head(page); | |
1535 | err = isolate_lru_page(head); | |
6d9c285a | 1536 | if (!err) { |
e8db67eb NH |
1537 | list_add_tail(&head->lru, &pagelist); |
1538 | mod_node_page_state(page_pgdat(head), | |
1539 | NR_ISOLATED_ANON + page_is_file_cache(head), | |
1540 | hpage_nr_pages(head)); | |
6d9c285a | 1541 | } |
742755a1 CL |
1542 | put_and_set: |
1543 | /* | |
1544 | * Either remove the duplicate refcount from | |
1545 | * isolate_lru_page() or drop the page ref if it was | |
1546 | * not isolated. | |
1547 | */ | |
1548 | put_page(page); | |
1549 | set_status: | |
1550 | pp->status = err; | |
1551 | } | |
1552 | ||
e78bbfa8 | 1553 | err = 0; |
cf608ac1 | 1554 | if (!list_empty(&pagelist)) { |
68711a74 | 1555 | err = migrate_pages(&pagelist, new_page_node, NULL, |
9c620e2b | 1556 | (unsigned long)pm, MIGRATE_SYNC, MR_SYSCALL); |
cf608ac1 | 1557 | if (err) |
e632a938 | 1558 | putback_movable_pages(&pagelist); |
cf608ac1 | 1559 | } |
742755a1 CL |
1560 | |
1561 | up_read(&mm->mmap_sem); | |
1562 | return err; | |
1563 | } | |
1564 | ||
5e9a0f02 BG |
1565 | /* |
1566 | * Migrate an array of page address onto an array of nodes and fill | |
1567 | * the corresponding array of status. | |
1568 | */ | |
3268c63e | 1569 | static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, |
5e9a0f02 BG |
1570 | unsigned long nr_pages, |
1571 | const void __user * __user *pages, | |
1572 | const int __user *nodes, | |
1573 | int __user *status, int flags) | |
1574 | { | |
3140a227 | 1575 | struct page_to_node *pm; |
3140a227 BG |
1576 | unsigned long chunk_nr_pages; |
1577 | unsigned long chunk_start; | |
1578 | int err; | |
5e9a0f02 | 1579 | |
3140a227 BG |
1580 | err = -ENOMEM; |
1581 | pm = (struct page_to_node *)__get_free_page(GFP_KERNEL); | |
1582 | if (!pm) | |
5e9a0f02 | 1583 | goto out; |
35282a2d BG |
1584 | |
1585 | migrate_prep(); | |
1586 | ||
5e9a0f02 | 1587 | /* |
3140a227 BG |
1588 | * Store a chunk of page_to_node array in a page, |
1589 | * but keep the last one as a marker | |
5e9a0f02 | 1590 | */ |
3140a227 | 1591 | chunk_nr_pages = (PAGE_SIZE / sizeof(struct page_to_node)) - 1; |
5e9a0f02 | 1592 | |
3140a227 BG |
1593 | for (chunk_start = 0; |
1594 | chunk_start < nr_pages; | |
1595 | chunk_start += chunk_nr_pages) { | |
1596 | int j; | |
5e9a0f02 | 1597 | |
3140a227 BG |
1598 | if (chunk_start + chunk_nr_pages > nr_pages) |
1599 | chunk_nr_pages = nr_pages - chunk_start; | |
1600 | ||
1601 | /* fill the chunk pm with addrs and nodes from user-space */ | |
1602 | for (j = 0; j < chunk_nr_pages; j++) { | |
1603 | const void __user *p; | |
5e9a0f02 BG |
1604 | int node; |
1605 | ||
3140a227 BG |
1606 | err = -EFAULT; |
1607 | if (get_user(p, pages + j + chunk_start)) | |
1608 | goto out_pm; | |
1609 | pm[j].addr = (unsigned long) p; | |
1610 | ||
1611 | if (get_user(node, nodes + j + chunk_start)) | |
5e9a0f02 BG |
1612 | goto out_pm; |
1613 | ||
1614 | err = -ENODEV; | |
6f5a55f1 LT |
1615 | if (node < 0 || node >= MAX_NUMNODES) |
1616 | goto out_pm; | |
1617 | ||
389162c2 | 1618 | if (!node_state(node, N_MEMORY)) |
5e9a0f02 BG |
1619 | goto out_pm; |
1620 | ||
1621 | err = -EACCES; | |
1622 | if (!node_isset(node, task_nodes)) | |
1623 | goto out_pm; | |
1624 | ||
3140a227 BG |
1625 | pm[j].node = node; |
1626 | } | |
1627 | ||
1628 | /* End marker for this chunk */ | |
1629 | pm[chunk_nr_pages].node = MAX_NUMNODES; | |
1630 | ||
1631 | /* Migrate this chunk */ | |
1632 | err = do_move_page_to_node_array(mm, pm, | |
1633 | flags & MPOL_MF_MOVE_ALL); | |
1634 | if (err < 0) | |
1635 | goto out_pm; | |
5e9a0f02 | 1636 | |
5e9a0f02 | 1637 | /* Return status information */ |
3140a227 BG |
1638 | for (j = 0; j < chunk_nr_pages; j++) |
1639 | if (put_user(pm[j].status, status + j + chunk_start)) { | |
5e9a0f02 | 1640 | err = -EFAULT; |
3140a227 BG |
1641 | goto out_pm; |
1642 | } | |
1643 | } | |
1644 | err = 0; | |
5e9a0f02 BG |
1645 | |
1646 | out_pm: | |
3140a227 | 1647 | free_page((unsigned long)pm); |
5e9a0f02 BG |
1648 | out: |
1649 | return err; | |
1650 | } | |
1651 | ||
742755a1 | 1652 | /* |
2f007e74 | 1653 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 1654 | */ |
80bba129 BG |
1655 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
1656 | const void __user **pages, int *status) | |
742755a1 | 1657 | { |
2f007e74 | 1658 | unsigned long i; |
2f007e74 | 1659 | |
742755a1 CL |
1660 | down_read(&mm->mmap_sem); |
1661 | ||
2f007e74 | 1662 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 1663 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
1664 | struct vm_area_struct *vma; |
1665 | struct page *page; | |
c095adbc | 1666 | int err = -EFAULT; |
2f007e74 BG |
1667 | |
1668 | vma = find_vma(mm, addr); | |
70384dc6 | 1669 | if (!vma || addr < vma->vm_start) |
742755a1 CL |
1670 | goto set_status; |
1671 | ||
d899844e KS |
1672 | /* FOLL_DUMP to ignore special (like zero) pages */ |
1673 | page = follow_page(vma, addr, FOLL_DUMP); | |
89f5b7da LT |
1674 | |
1675 | err = PTR_ERR(page); | |
1676 | if (IS_ERR(page)) | |
1677 | goto set_status; | |
1678 | ||
d899844e | 1679 | err = page ? page_to_nid(page) : -ENOENT; |
742755a1 | 1680 | set_status: |
80bba129 BG |
1681 | *status = err; |
1682 | ||
1683 | pages++; | |
1684 | status++; | |
1685 | } | |
1686 | ||
1687 | up_read(&mm->mmap_sem); | |
1688 | } | |
1689 | ||
1690 | /* | |
1691 | * Determine the nodes of a user array of pages and store it in | |
1692 | * a user array of status. | |
1693 | */ | |
1694 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
1695 | const void __user * __user *pages, | |
1696 | int __user *status) | |
1697 | { | |
1698 | #define DO_PAGES_STAT_CHUNK_NR 16 | |
1699 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; | |
1700 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
80bba129 | 1701 | |
87b8d1ad PA |
1702 | while (nr_pages) { |
1703 | unsigned long chunk_nr; | |
80bba129 | 1704 | |
87b8d1ad PA |
1705 | chunk_nr = nr_pages; |
1706 | if (chunk_nr > DO_PAGES_STAT_CHUNK_NR) | |
1707 | chunk_nr = DO_PAGES_STAT_CHUNK_NR; | |
1708 | ||
1709 | if (copy_from_user(chunk_pages, pages, chunk_nr * sizeof(*chunk_pages))) | |
1710 | break; | |
80bba129 BG |
1711 | |
1712 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
1713 | ||
87b8d1ad PA |
1714 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
1715 | break; | |
742755a1 | 1716 | |
87b8d1ad PA |
1717 | pages += chunk_nr; |
1718 | status += chunk_nr; | |
1719 | nr_pages -= chunk_nr; | |
1720 | } | |
1721 | return nr_pages ? -EFAULT : 0; | |
742755a1 CL |
1722 | } |
1723 | ||
1724 | /* | |
1725 | * Move a list of pages in the address space of the currently executing | |
1726 | * process. | |
1727 | */ | |
938bb9f5 HC |
1728 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
1729 | const void __user * __user *, pages, | |
1730 | const int __user *, nodes, | |
1731 | int __user *, status, int, flags) | |
742755a1 | 1732 | { |
742755a1 | 1733 | struct task_struct *task; |
742755a1 | 1734 | struct mm_struct *mm; |
5e9a0f02 | 1735 | int err; |
3268c63e | 1736 | nodemask_t task_nodes; |
742755a1 CL |
1737 | |
1738 | /* Check flags */ | |
1739 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
1740 | return -EINVAL; | |
1741 | ||
1742 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) | |
1743 | return -EPERM; | |
1744 | ||
1745 | /* Find the mm_struct */ | |
a879bf58 | 1746 | rcu_read_lock(); |
228ebcbe | 1747 | task = pid ? find_task_by_vpid(pid) : current; |
742755a1 | 1748 | if (!task) { |
a879bf58 | 1749 | rcu_read_unlock(); |
742755a1 CL |
1750 | return -ESRCH; |
1751 | } | |
3268c63e | 1752 | get_task_struct(task); |
742755a1 CL |
1753 | |
1754 | /* | |
1755 | * Check if this process has the right to modify the specified | |
197e7e52 | 1756 | * process. Use the regular "ptrace_may_access()" checks. |
742755a1 | 1757 | */ |
197e7e52 | 1758 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 1759 | rcu_read_unlock(); |
742755a1 | 1760 | err = -EPERM; |
5e9a0f02 | 1761 | goto out; |
742755a1 | 1762 | } |
c69e8d9c | 1763 | rcu_read_unlock(); |
742755a1 | 1764 | |
86c3a764 DQ |
1765 | err = security_task_movememory(task); |
1766 | if (err) | |
5e9a0f02 | 1767 | goto out; |
86c3a764 | 1768 | |
3268c63e CL |
1769 | task_nodes = cpuset_mems_allowed(task); |
1770 | mm = get_task_mm(task); | |
1771 | put_task_struct(task); | |
1772 | ||
6e8b09ea SL |
1773 | if (!mm) |
1774 | return -EINVAL; | |
1775 | ||
1776 | if (nodes) | |
1777 | err = do_pages_move(mm, task_nodes, nr_pages, pages, | |
1778 | nodes, status, flags); | |
1779 | else | |
1780 | err = do_pages_stat(mm, nr_pages, pages, status); | |
742755a1 | 1781 | |
742755a1 CL |
1782 | mmput(mm); |
1783 | return err; | |
3268c63e CL |
1784 | |
1785 | out: | |
1786 | put_task_struct(task); | |
1787 | return err; | |
742755a1 | 1788 | } |
742755a1 | 1789 | |
7039e1db PZ |
1790 | #ifdef CONFIG_NUMA_BALANCING |
1791 | /* | |
1792 | * Returns true if this is a safe migration target node for misplaced NUMA | |
1793 | * pages. Currently it only checks the watermarks which crude | |
1794 | */ | |
1795 | static bool migrate_balanced_pgdat(struct pglist_data *pgdat, | |
3abef4e6 | 1796 | unsigned long nr_migrate_pages) |
7039e1db PZ |
1797 | { |
1798 | int z; | |
599d0c95 | 1799 | |
7039e1db PZ |
1800 | for (z = pgdat->nr_zones - 1; z >= 0; z--) { |
1801 | struct zone *zone = pgdat->node_zones + z; | |
1802 | ||
1803 | if (!populated_zone(zone)) | |
1804 | continue; | |
1805 | ||
7039e1db PZ |
1806 | /* Avoid waking kswapd by allocating pages_to_migrate pages. */ |
1807 | if (!zone_watermark_ok(zone, 0, | |
1808 | high_wmark_pages(zone) + | |
1809 | nr_migrate_pages, | |
1810 | 0, 0)) | |
1811 | continue; | |
1812 | return true; | |
1813 | } | |
1814 | return false; | |
1815 | } | |
1816 | ||
1817 | static struct page *alloc_misplaced_dst_page(struct page *page, | |
1818 | unsigned long data, | |
1819 | int **result) | |
1820 | { | |
1821 | int nid = (int) data; | |
1822 | struct page *newpage; | |
1823 | ||
96db800f | 1824 | newpage = __alloc_pages_node(nid, |
e97ca8e5 JW |
1825 | (GFP_HIGHUSER_MOVABLE | |
1826 | __GFP_THISNODE | __GFP_NOMEMALLOC | | |
1827 | __GFP_NORETRY | __GFP_NOWARN) & | |
8479eba7 | 1828 | ~__GFP_RECLAIM, 0); |
bac0382c | 1829 | |
7039e1db PZ |
1830 | return newpage; |
1831 | } | |
1832 | ||
a8f60772 MG |
1833 | /* |
1834 | * page migration rate limiting control. | |
1835 | * Do not migrate more than @pages_to_migrate in a @migrate_interval_millisecs | |
1836 | * window of time. Default here says do not migrate more than 1280M per second. | |
1837 | */ | |
1838 | static unsigned int migrate_interval_millisecs __read_mostly = 100; | |
1839 | static unsigned int ratelimit_pages __read_mostly = 128 << (20 - PAGE_SHIFT); | |
1840 | ||
b32967ff | 1841 | /* Returns true if the node is migrate rate-limited after the update */ |
1c30e017 MG |
1842 | static bool numamigrate_update_ratelimit(pg_data_t *pgdat, |
1843 | unsigned long nr_pages) | |
7039e1db | 1844 | { |
a8f60772 MG |
1845 | /* |
1846 | * Rate-limit the amount of data that is being migrated to a node. | |
1847 | * Optimal placement is no good if the memory bus is saturated and | |
1848 | * all the time is being spent migrating! | |
1849 | */ | |
a8f60772 | 1850 | if (time_after(jiffies, pgdat->numabalancing_migrate_next_window)) { |
1c5e9c27 | 1851 | spin_lock(&pgdat->numabalancing_migrate_lock); |
a8f60772 MG |
1852 | pgdat->numabalancing_migrate_nr_pages = 0; |
1853 | pgdat->numabalancing_migrate_next_window = jiffies + | |
1854 | msecs_to_jiffies(migrate_interval_millisecs); | |
1c5e9c27 | 1855 | spin_unlock(&pgdat->numabalancing_migrate_lock); |
a8f60772 | 1856 | } |
af1839d7 MG |
1857 | if (pgdat->numabalancing_migrate_nr_pages > ratelimit_pages) { |
1858 | trace_mm_numa_migrate_ratelimit(current, pgdat->node_id, | |
1859 | nr_pages); | |
1c5e9c27 | 1860 | return true; |
af1839d7 | 1861 | } |
1c5e9c27 MG |
1862 | |
1863 | /* | |
1864 | * This is an unlocked non-atomic update so errors are possible. | |
1865 | * The consequences are failing to migrate when we potentiall should | |
1866 | * have which is not severe enough to warrant locking. If it is ever | |
1867 | * a problem, it can be converted to a per-cpu counter. | |
1868 | */ | |
1869 | pgdat->numabalancing_migrate_nr_pages += nr_pages; | |
1870 | return false; | |
b32967ff MG |
1871 | } |
1872 | ||
1c30e017 | 1873 | static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page) |
b32967ff | 1874 | { |
340ef390 | 1875 | int page_lru; |
a8f60772 | 1876 | |
309381fe | 1877 | VM_BUG_ON_PAGE(compound_order(page) && !PageTransHuge(page), page); |
3abef4e6 | 1878 | |
7039e1db | 1879 | /* Avoid migrating to a node that is nearly full */ |
340ef390 HD |
1880 | if (!migrate_balanced_pgdat(pgdat, 1UL << compound_order(page))) |
1881 | return 0; | |
7039e1db | 1882 | |
340ef390 HD |
1883 | if (isolate_lru_page(page)) |
1884 | return 0; | |
7039e1db | 1885 | |
340ef390 HD |
1886 | /* |
1887 | * migrate_misplaced_transhuge_page() skips page migration's usual | |
1888 | * check on page_count(), so we must do it here, now that the page | |
1889 | * has been isolated: a GUP pin, or any other pin, prevents migration. | |
1890 | * The expected page count is 3: 1 for page's mapcount and 1 for the | |
1891 | * caller's pin and 1 for the reference taken by isolate_lru_page(). | |
1892 | */ | |
1893 | if (PageTransHuge(page) && page_count(page) != 3) { | |
1894 | putback_lru_page(page); | |
1895 | return 0; | |
7039e1db PZ |
1896 | } |
1897 | ||
340ef390 | 1898 | page_lru = page_is_file_cache(page); |
599d0c95 | 1899 | mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + page_lru, |
340ef390 HD |
1900 | hpage_nr_pages(page)); |
1901 | ||
149c33e1 | 1902 | /* |
340ef390 HD |
1903 | * Isolating the page has taken another reference, so the |
1904 | * caller's reference can be safely dropped without the page | |
1905 | * disappearing underneath us during migration. | |
149c33e1 MG |
1906 | */ |
1907 | put_page(page); | |
340ef390 | 1908 | return 1; |
b32967ff MG |
1909 | } |
1910 | ||
de466bd6 MG |
1911 | bool pmd_trans_migrating(pmd_t pmd) |
1912 | { | |
1913 | struct page *page = pmd_page(pmd); | |
1914 | return PageLocked(page); | |
1915 | } | |
1916 | ||
b32967ff MG |
1917 | /* |
1918 | * Attempt to migrate a misplaced page to the specified destination | |
1919 | * node. Caller is expected to have an elevated reference count on | |
1920 | * the page that will be dropped by this function before returning. | |
1921 | */ | |
1bc115d8 MG |
1922 | int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma, |
1923 | int node) | |
b32967ff MG |
1924 | { |
1925 | pg_data_t *pgdat = NODE_DATA(node); | |
340ef390 | 1926 | int isolated; |
b32967ff MG |
1927 | int nr_remaining; |
1928 | LIST_HEAD(migratepages); | |
1929 | ||
1930 | /* | |
1bc115d8 MG |
1931 | * Don't migrate file pages that are mapped in multiple processes |
1932 | * with execute permissions as they are probably shared libraries. | |
b32967ff | 1933 | */ |
1bc115d8 MG |
1934 | if (page_mapcount(page) != 1 && page_is_file_cache(page) && |
1935 | (vma->vm_flags & VM_EXEC)) | |
b32967ff | 1936 | goto out; |
b32967ff MG |
1937 | |
1938 | /* | |
1939 | * Rate-limit the amount of data that is being migrated to a node. | |
1940 | * Optimal placement is no good if the memory bus is saturated and | |
1941 | * all the time is being spent migrating! | |
1942 | */ | |
340ef390 | 1943 | if (numamigrate_update_ratelimit(pgdat, 1)) |
b32967ff | 1944 | goto out; |
b32967ff MG |
1945 | |
1946 | isolated = numamigrate_isolate_page(pgdat, page); | |
1947 | if (!isolated) | |
1948 | goto out; | |
1949 | ||
1950 | list_add(&page->lru, &migratepages); | |
9c620e2b | 1951 | nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page, |
68711a74 DR |
1952 | NULL, node, MIGRATE_ASYNC, |
1953 | MR_NUMA_MISPLACED); | |
b32967ff | 1954 | if (nr_remaining) { |
59c82b70 JK |
1955 | if (!list_empty(&migratepages)) { |
1956 | list_del(&page->lru); | |
599d0c95 | 1957 | dec_node_page_state(page, NR_ISOLATED_ANON + |
59c82b70 JK |
1958 | page_is_file_cache(page)); |
1959 | putback_lru_page(page); | |
1960 | } | |
b32967ff MG |
1961 | isolated = 0; |
1962 | } else | |
1963 | count_vm_numa_event(NUMA_PAGE_MIGRATE); | |
7039e1db | 1964 | BUG_ON(!list_empty(&migratepages)); |
7039e1db | 1965 | return isolated; |
340ef390 HD |
1966 | |
1967 | out: | |
1968 | put_page(page); | |
1969 | return 0; | |
7039e1db | 1970 | } |
220018d3 | 1971 | #endif /* CONFIG_NUMA_BALANCING */ |
b32967ff | 1972 | |
220018d3 | 1973 | #if defined(CONFIG_NUMA_BALANCING) && defined(CONFIG_TRANSPARENT_HUGEPAGE) |
340ef390 HD |
1974 | /* |
1975 | * Migrates a THP to a given target node. page must be locked and is unlocked | |
1976 | * before returning. | |
1977 | */ | |
b32967ff MG |
1978 | int migrate_misplaced_transhuge_page(struct mm_struct *mm, |
1979 | struct vm_area_struct *vma, | |
1980 | pmd_t *pmd, pmd_t entry, | |
1981 | unsigned long address, | |
1982 | struct page *page, int node) | |
1983 | { | |
c4088ebd | 1984 | spinlock_t *ptl; |
b32967ff MG |
1985 | pg_data_t *pgdat = NODE_DATA(node); |
1986 | int isolated = 0; | |
1987 | struct page *new_page = NULL; | |
b32967ff | 1988 | int page_lru = page_is_file_cache(page); |
f714f4f2 MG |
1989 | unsigned long mmun_start = address & HPAGE_PMD_MASK; |
1990 | unsigned long mmun_end = mmun_start + HPAGE_PMD_SIZE; | |
b32967ff | 1991 | |
b32967ff MG |
1992 | /* |
1993 | * Rate-limit the amount of data that is being migrated to a node. | |
1994 | * Optimal placement is no good if the memory bus is saturated and | |
1995 | * all the time is being spent migrating! | |
1996 | */ | |
d28d4335 | 1997 | if (numamigrate_update_ratelimit(pgdat, HPAGE_PMD_NR)) |
b32967ff MG |
1998 | goto out_dropref; |
1999 | ||
2000 | new_page = alloc_pages_node(node, | |
25160354 | 2001 | (GFP_TRANSHUGE_LIGHT | __GFP_THISNODE), |
e97ca8e5 | 2002 | HPAGE_PMD_ORDER); |
340ef390 HD |
2003 | if (!new_page) |
2004 | goto out_fail; | |
9a982250 | 2005 | prep_transhuge_page(new_page); |
340ef390 | 2006 | |
b32967ff | 2007 | isolated = numamigrate_isolate_page(pgdat, page); |
340ef390 | 2008 | if (!isolated) { |
b32967ff | 2009 | put_page(new_page); |
340ef390 | 2010 | goto out_fail; |
b32967ff | 2011 | } |
b0943d61 | 2012 | |
b32967ff | 2013 | /* Prepare a page as a migration target */ |
48c935ad | 2014 | __SetPageLocked(new_page); |
d44d363f SL |
2015 | if (PageSwapBacked(page)) |
2016 | __SetPageSwapBacked(new_page); | |
b32967ff MG |
2017 | |
2018 | /* anon mapping, we can simply copy page->mapping to the new page: */ | |
2019 | new_page->mapping = page->mapping; | |
2020 | new_page->index = page->index; | |
2021 | migrate_page_copy(new_page, page); | |
2022 | WARN_ON(PageLRU(new_page)); | |
2023 | ||
2024 | /* Recheck the target PMD */ | |
f714f4f2 | 2025 | mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end); |
c4088ebd | 2026 | ptl = pmd_lock(mm, pmd); |
f4e177d1 | 2027 | if (unlikely(!pmd_same(*pmd, entry) || !page_ref_freeze(page, 2))) { |
c4088ebd | 2028 | spin_unlock(ptl); |
f714f4f2 | 2029 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
b32967ff MG |
2030 | |
2031 | /* Reverse changes made by migrate_page_copy() */ | |
2032 | if (TestClearPageActive(new_page)) | |
2033 | SetPageActive(page); | |
2034 | if (TestClearPageUnevictable(new_page)) | |
2035 | SetPageUnevictable(page); | |
b32967ff MG |
2036 | |
2037 | unlock_page(new_page); | |
2038 | put_page(new_page); /* Free it */ | |
2039 | ||
a54a407f MG |
2040 | /* Retake the callers reference and putback on LRU */ |
2041 | get_page(page); | |
b32967ff | 2042 | putback_lru_page(page); |
599d0c95 | 2043 | mod_node_page_state(page_pgdat(page), |
a54a407f | 2044 | NR_ISOLATED_ANON + page_lru, -HPAGE_PMD_NR); |
eb4489f6 MG |
2045 | |
2046 | goto out_unlock; | |
b32967ff MG |
2047 | } |
2048 | ||
10102459 | 2049 | entry = mk_huge_pmd(new_page, vma->vm_page_prot); |
2b4847e7 | 2050 | entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); |
b32967ff | 2051 | |
2b4847e7 MG |
2052 | /* |
2053 | * Clear the old entry under pagetable lock and establish the new PTE. | |
2054 | * Any parallel GUP will either observe the old page blocking on the | |
2055 | * page lock, block on the page table lock or observe the new page. | |
2056 | * The SetPageUptodate on the new page and page_add_new_anon_rmap | |
2057 | * guarantee the copy is visible before the pagetable update. | |
2058 | */ | |
f714f4f2 | 2059 | flush_cache_range(vma, mmun_start, mmun_end); |
d281ee61 | 2060 | page_add_anon_rmap(new_page, vma, mmun_start, true); |
8809aa2d | 2061 | pmdp_huge_clear_flush_notify(vma, mmun_start, pmd); |
f714f4f2 | 2062 | set_pmd_at(mm, mmun_start, pmd, entry); |
ce4a9cc5 | 2063 | update_mmu_cache_pmd(vma, address, &entry); |
2b4847e7 | 2064 | |
f4e177d1 | 2065 | page_ref_unfreeze(page, 2); |
51afb12b | 2066 | mlock_migrate_page(new_page, page); |
d281ee61 | 2067 | page_remove_rmap(page, true); |
7cd12b4a | 2068 | set_page_owner_migrate_reason(new_page, MR_NUMA_MISPLACED); |
2b4847e7 | 2069 | |
c4088ebd | 2070 | spin_unlock(ptl); |
f714f4f2 | 2071 | mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end); |
b32967ff | 2072 | |
11de9927 MG |
2073 | /* Take an "isolate" reference and put new page on the LRU. */ |
2074 | get_page(new_page); | |
2075 | putback_lru_page(new_page); | |
2076 | ||
b32967ff MG |
2077 | unlock_page(new_page); |
2078 | unlock_page(page); | |
2079 | put_page(page); /* Drop the rmap reference */ | |
2080 | put_page(page); /* Drop the LRU isolation reference */ | |
2081 | ||
2082 | count_vm_events(PGMIGRATE_SUCCESS, HPAGE_PMD_NR); | |
2083 | count_vm_numa_events(NUMA_PAGE_MIGRATE, HPAGE_PMD_NR); | |
2084 | ||
599d0c95 | 2085 | mod_node_page_state(page_pgdat(page), |
b32967ff MG |
2086 | NR_ISOLATED_ANON + page_lru, |
2087 | -HPAGE_PMD_NR); | |
2088 | return isolated; | |
2089 | ||
340ef390 HD |
2090 | out_fail: |
2091 | count_vm_events(PGMIGRATE_FAIL, HPAGE_PMD_NR); | |
b32967ff | 2092 | out_dropref: |
2b4847e7 MG |
2093 | ptl = pmd_lock(mm, pmd); |
2094 | if (pmd_same(*pmd, entry)) { | |
4d942466 | 2095 | entry = pmd_modify(entry, vma->vm_page_prot); |
f714f4f2 | 2096 | set_pmd_at(mm, mmun_start, pmd, entry); |
2b4847e7 MG |
2097 | update_mmu_cache_pmd(vma, address, &entry); |
2098 | } | |
2099 | spin_unlock(ptl); | |
a54a407f | 2100 | |
eb4489f6 | 2101 | out_unlock: |
340ef390 | 2102 | unlock_page(page); |
b32967ff | 2103 | put_page(page); |
b32967ff MG |
2104 | return 0; |
2105 | } | |
7039e1db PZ |
2106 | #endif /* CONFIG_NUMA_BALANCING */ |
2107 | ||
2108 | #endif /* CONFIG_NUMA */ |