Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
b20a3503 | 2 | /* |
14e0f9bc | 3 | * Memory Migration functionality - linux/mm/migrate.c |
b20a3503 CL |
4 | * |
5 | * Copyright (C) 2006 Silicon Graphics, Inc., Christoph Lameter | |
6 | * | |
7 | * Page migration was first developed in the context of the memory hotplug | |
8 | * project. The main authors of the migration code are: | |
9 | * | |
10 | * IWAMOTO Toshihiro <iwamoto@valinux.co.jp> | |
11 | * Hirokazu Takahashi <taka@valinux.co.jp> | |
12 | * Dave Hansen <haveblue@us.ibm.com> | |
cde53535 | 13 | * Christoph Lameter |
b20a3503 CL |
14 | */ |
15 | ||
16 | #include <linux/migrate.h> | |
b95f1b31 | 17 | #include <linux/export.h> |
b20a3503 | 18 | #include <linux/swap.h> |
0697212a | 19 | #include <linux/swapops.h> |
b20a3503 | 20 | #include <linux/pagemap.h> |
e23ca00b | 21 | #include <linux/buffer_head.h> |
b20a3503 | 22 | #include <linux/mm_inline.h> |
b488893a | 23 | #include <linux/nsproxy.h> |
b20a3503 | 24 | #include <linux/pagevec.h> |
e9995ef9 | 25 | #include <linux/ksm.h> |
b20a3503 CL |
26 | #include <linux/rmap.h> |
27 | #include <linux/topology.h> | |
28 | #include <linux/cpu.h> | |
29 | #include <linux/cpuset.h> | |
04e62a29 | 30 | #include <linux/writeback.h> |
742755a1 CL |
31 | #include <linux/mempolicy.h> |
32 | #include <linux/vmalloc.h> | |
86c3a764 | 33 | #include <linux/security.h> |
42cb14b1 | 34 | #include <linux/backing-dev.h> |
bda807d4 | 35 | #include <linux/compaction.h> |
4f5ca265 | 36 | #include <linux/syscalls.h> |
7addf443 | 37 | #include <linux/compat.h> |
290408d4 | 38 | #include <linux/hugetlb.h> |
8e6ac7fa | 39 | #include <linux/hugetlb_cgroup.h> |
5a0e3ad6 | 40 | #include <linux/gfp.h> |
df6ad698 | 41 | #include <linux/pfn_t.h> |
a5430dda | 42 | #include <linux/memremap.h> |
8315ada7 | 43 | #include <linux/userfaultfd_k.h> |
bf6bddf1 | 44 | #include <linux/balloon_compaction.h> |
33c3fc71 | 45 | #include <linux/page_idle.h> |
d435edca | 46 | #include <linux/page_owner.h> |
6e84f315 | 47 | #include <linux/sched/mm.h> |
197e7e52 | 48 | #include <linux/ptrace.h> |
34290e2c | 49 | #include <linux/oom.h> |
884a6e5d | 50 | #include <linux/memory.h> |
ac16ec83 | 51 | #include <linux/random.h> |
c574bbe9 | 52 | #include <linux/sched/sysctl.h> |
467b171a | 53 | #include <linux/memory-tiers.h> |
b20a3503 | 54 | |
0d1836c3 MN |
55 | #include <asm/tlbflush.h> |
56 | ||
7b2a2d4a MG |
57 | #include <trace/events/migrate.h> |
58 | ||
b20a3503 CL |
59 | #include "internal.h" |
60 | ||
9e5bcd61 | 61 | int isolate_movable_page(struct page *page, isolate_mode_t mode) |
bda807d4 | 62 | { |
19979497 | 63 | struct folio *folio = folio_get_nontail_page(page); |
68f2736a | 64 | const struct movable_operations *mops; |
bda807d4 MK |
65 | |
66 | /* | |
67 | * Avoid burning cycles with pages that are yet under __free_pages(), | |
68 | * or just got freed under us. | |
69 | * | |
70 | * In case we 'win' a race for a movable page being freed under us and | |
71 | * raise its refcount preventing __free_pages() from doing its job | |
72 | * the put_page() at the end of this block will take care of | |
73 | * release this page, thus avoiding a nasty leakage. | |
74 | */ | |
19979497 | 75 | if (!folio) |
bda807d4 MK |
76 | goto out; |
77 | ||
19979497 VMO |
78 | if (unlikely(folio_test_slab(folio))) |
79 | goto out_putfolio; | |
8b881763 VB |
80 | /* Pairs with smp_wmb() in slab freeing, e.g. SLUB's __free_slab() */ |
81 | smp_rmb(); | |
bda807d4 | 82 | /* |
8b881763 VB |
83 | * Check movable flag before taking the page lock because |
84 | * we use non-atomic bitops on newly allocated page flags so | |
85 | * unconditionally grabbing the lock ruins page's owner side. | |
bda807d4 | 86 | */ |
19979497 VMO |
87 | if (unlikely(!__folio_test_movable(folio))) |
88 | goto out_putfolio; | |
8b881763 VB |
89 | /* Pairs with smp_wmb() in slab allocation, e.g. SLUB's alloc_slab_page() */ |
90 | smp_rmb(); | |
19979497 VMO |
91 | if (unlikely(folio_test_slab(folio))) |
92 | goto out_putfolio; | |
8b881763 | 93 | |
bda807d4 MK |
94 | /* |
95 | * As movable pages are not isolated from LRU lists, concurrent | |
96 | * compaction threads can race against page migration functions | |
97 | * as well as race against the releasing a page. | |
98 | * | |
99 | * In order to avoid having an already isolated movable page | |
100 | * being (wrongly) re-isolated while it is under migration, | |
101 | * or to avoid attempting to isolate pages being released, | |
102 | * lets be sure we have the page lock | |
103 | * before proceeding with the movable page isolation steps. | |
104 | */ | |
19979497 VMO |
105 | if (unlikely(!folio_trylock(folio))) |
106 | goto out_putfolio; | |
bda807d4 | 107 | |
19979497 | 108 | if (!folio_test_movable(folio) || folio_test_isolated(folio)) |
bda807d4 MK |
109 | goto out_no_isolated; |
110 | ||
19979497 VMO |
111 | mops = folio_movable_ops(folio); |
112 | VM_BUG_ON_FOLIO(!mops, folio); | |
bda807d4 | 113 | |
19979497 | 114 | if (!mops->isolate_page(&folio->page, mode)) |
bda807d4 MK |
115 | goto out_no_isolated; |
116 | ||
117 | /* Driver shouldn't use PG_isolated bit of page->flags */ | |
19979497 VMO |
118 | WARN_ON_ONCE(folio_test_isolated(folio)); |
119 | folio_set_isolated(folio); | |
120 | folio_unlock(folio); | |
bda807d4 | 121 | |
9e5bcd61 | 122 | return 0; |
bda807d4 MK |
123 | |
124 | out_no_isolated: | |
19979497 VMO |
125 | folio_unlock(folio); |
126 | out_putfolio: | |
127 | folio_put(folio); | |
bda807d4 | 128 | out: |
9e5bcd61 | 129 | return -EBUSY; |
bda807d4 MK |
130 | } |
131 | ||
280d724a | 132 | static void putback_movable_folio(struct folio *folio) |
bda807d4 | 133 | { |
280d724a | 134 | const struct movable_operations *mops = folio_movable_ops(folio); |
bda807d4 | 135 | |
280d724a VMO |
136 | mops->putback_page(&folio->page); |
137 | folio_clear_isolated(folio); | |
bda807d4 MK |
138 | } |
139 | ||
5733c7d1 RA |
140 | /* |
141 | * Put previously isolated pages back onto the appropriate lists | |
142 | * from where they were once taken off for compaction/migration. | |
143 | * | |
59c82b70 JK |
144 | * This function shall be used whenever the isolated pageset has been |
145 | * built from lru, balloon, hugetlbfs page. See isolate_migratepages_range() | |
7ce82f4c | 146 | * and isolate_hugetlb(). |
5733c7d1 RA |
147 | */ |
148 | void putback_movable_pages(struct list_head *l) | |
149 | { | |
280d724a VMO |
150 | struct folio *folio; |
151 | struct folio *folio2; | |
5733c7d1 | 152 | |
280d724a VMO |
153 | list_for_each_entry_safe(folio, folio2, l, lru) { |
154 | if (unlikely(folio_test_hugetlb(folio))) { | |
155 | folio_putback_active_hugetlb(folio); | |
31caf665 NH |
156 | continue; |
157 | } | |
280d724a | 158 | list_del(&folio->lru); |
bda807d4 | 159 | /* |
280d724a VMO |
160 | * We isolated non-lru movable folio so here we can use |
161 | * __PageMovable because LRU folio's mapping cannot have | |
bda807d4 MK |
162 | * PAGE_MAPPING_MOVABLE. |
163 | */ | |
280d724a VMO |
164 | if (unlikely(__folio_test_movable(folio))) { |
165 | VM_BUG_ON_FOLIO(!folio_test_isolated(folio), folio); | |
166 | folio_lock(folio); | |
167 | if (folio_test_movable(folio)) | |
168 | putback_movable_folio(folio); | |
bda807d4 | 169 | else |
280d724a VMO |
170 | folio_clear_isolated(folio); |
171 | folio_unlock(folio); | |
172 | folio_put(folio); | |
bda807d4 | 173 | } else { |
280d724a VMO |
174 | node_stat_mod_folio(folio, NR_ISOLATED_ANON + |
175 | folio_is_file_lru(folio), -folio_nr_pages(folio)); | |
176 | folio_putback_lru(folio); | |
bda807d4 | 177 | } |
b20a3503 | 178 | } |
b20a3503 CL |
179 | } |
180 | ||
0697212a CL |
181 | /* |
182 | * Restore a potential migration pte to a working pte entry | |
183 | */ | |
2f031c6f MWO |
184 | static bool remove_migration_pte(struct folio *folio, |
185 | struct vm_area_struct *vma, unsigned long addr, void *old) | |
0697212a | 186 | { |
4eecb8b9 | 187 | DEFINE_FOLIO_VMA_WALK(pvmw, old, vma, addr, PVMW_SYNC | PVMW_MIGRATION); |
0697212a | 188 | |
3fe87967 | 189 | while (page_vma_mapped_walk(&pvmw)) { |
6c287605 | 190 | rmap_t rmap_flags = RMAP_NONE; |
4eecb8b9 MWO |
191 | pte_t pte; |
192 | swp_entry_t entry; | |
193 | struct page *new; | |
194 | unsigned long idx = 0; | |
195 | ||
196 | /* pgoff is invalid for ksm pages, but they are never large */ | |
197 | if (folio_test_large(folio) && !folio_test_hugetlb(folio)) | |
198 | idx = linear_page_index(vma, pvmw.address) - pvmw.pgoff; | |
199 | new = folio_page(folio, idx); | |
0697212a | 200 | |
616b8371 ZY |
201 | #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION |
202 | /* PMD-mapped THP migration entry */ | |
203 | if (!pvmw.pte) { | |
4eecb8b9 MWO |
204 | VM_BUG_ON_FOLIO(folio_test_hugetlb(folio) || |
205 | !folio_test_pmd_mappable(folio), folio); | |
616b8371 ZY |
206 | remove_migration_pmd(&pvmw, new); |
207 | continue; | |
208 | } | |
209 | #endif | |
210 | ||
4eecb8b9 | 211 | folio_get(folio); |
2e346877 | 212 | pte = mk_pte(new, READ_ONCE(vma->vm_page_prot)); |
3fe87967 KS |
213 | if (pte_swp_soft_dirty(*pvmw.pte)) |
214 | pte = pte_mksoft_dirty(pte); | |
0697212a | 215 | |
3fe87967 KS |
216 | /* |
217 | * Recheck VMA as permissions can change since migration started | |
218 | */ | |
219 | entry = pte_to_swp_entry(*pvmw.pte); | |
2e346877 PX |
220 | if (!is_migration_entry_young(entry)) |
221 | pte = pte_mkold(pte); | |
222 | if (folio_test_dirty(folio) && is_migration_entry_dirty(entry)) | |
223 | pte = pte_mkdirty(pte); | |
4dd845b5 | 224 | if (is_writable_migration_entry(entry)) |
3fe87967 | 225 | pte = maybe_mkwrite(pte, vma); |
f45ec5ff PX |
226 | else if (pte_swp_uffd_wp(*pvmw.pte)) |
227 | pte = pte_mkuffd_wp(pte); | |
d3cb8bf6 | 228 | |
6c287605 DH |
229 | if (folio_test_anon(folio) && !is_readable_migration_entry(entry)) |
230 | rmap_flags |= RMAP_EXCLUSIVE; | |
231 | ||
6128763f | 232 | if (unlikely(is_device_private_page(new))) { |
4dd845b5 AP |
233 | if (pte_write(pte)) |
234 | entry = make_writable_device_private_entry( | |
235 | page_to_pfn(new)); | |
236 | else | |
237 | entry = make_readable_device_private_entry( | |
238 | page_to_pfn(new)); | |
6128763f | 239 | pte = swp_entry_to_pte(entry); |
3d321bf8 RC |
240 | if (pte_swp_soft_dirty(*pvmw.pte)) |
241 | pte = pte_swp_mksoft_dirty(pte); | |
6128763f RC |
242 | if (pte_swp_uffd_wp(*pvmw.pte)) |
243 | pte = pte_swp_mkuffd_wp(pte); | |
d2b2c6dd | 244 | } |
a5430dda | 245 | |
3ef8fd7f | 246 | #ifdef CONFIG_HUGETLB_PAGE |
4eecb8b9 | 247 | if (folio_test_hugetlb(folio)) { |
79c1c594 CL |
248 | unsigned int shift = huge_page_shift(hstate_vma(vma)); |
249 | ||
3fe87967 | 250 | pte = pte_mkhuge(pte); |
79c1c594 | 251 | pte = arch_make_huge_pte(pte, shift, vma->vm_flags); |
4eecb8b9 | 252 | if (folio_test_anon(folio)) |
28c5209d | 253 | hugepage_add_anon_rmap(new, vma, pvmw.address, |
6c287605 | 254 | rmap_flags); |
3fe87967 | 255 | else |
fb3d824d | 256 | page_dup_file_rmap(new, true); |
1eba86c0 | 257 | set_huge_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte); |
383321ab AK |
258 | } else |
259 | #endif | |
260 | { | |
4eecb8b9 | 261 | if (folio_test_anon(folio)) |
f1e2db12 | 262 | page_add_anon_rmap(new, vma, pvmw.address, |
6c287605 | 263 | rmap_flags); |
383321ab | 264 | else |
cea86fe2 | 265 | page_add_file_rmap(new, vma, false); |
1eba86c0 | 266 | set_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte); |
383321ab | 267 | } |
b7435507 | 268 | if (vma->vm_flags & VM_LOCKED) |
96f97c43 | 269 | mlock_drain_local(); |
e125fe40 | 270 | |
4cc79b33 AK |
271 | trace_remove_migration_pte(pvmw.address, pte_val(pte), |
272 | compound_order(new)); | |
273 | ||
3fe87967 KS |
274 | /* No need to invalidate - it was non-present before */ |
275 | update_mmu_cache(vma, pvmw.address, pvmw.pte); | |
276 | } | |
51afb12b | 277 | |
e4b82222 | 278 | return true; |
0697212a CL |
279 | } |
280 | ||
04e62a29 CL |
281 | /* |
282 | * Get rid of all migration entries and replace them by | |
283 | * references to the indicated page. | |
284 | */ | |
4eecb8b9 | 285 | void remove_migration_ptes(struct folio *src, struct folio *dst, bool locked) |
04e62a29 | 286 | { |
051ac83a JK |
287 | struct rmap_walk_control rwc = { |
288 | .rmap_one = remove_migration_pte, | |
4eecb8b9 | 289 | .arg = src, |
051ac83a JK |
290 | }; |
291 | ||
e388466d | 292 | if (locked) |
2f031c6f | 293 | rmap_walk_locked(dst, &rwc); |
e388466d | 294 | else |
2f031c6f | 295 | rmap_walk(dst, &rwc); |
04e62a29 CL |
296 | } |
297 | ||
0697212a CL |
298 | /* |
299 | * Something used the pte of a page under migration. We need to | |
300 | * get to the page and wait until migration is finished. | |
301 | * When we return from this function the fault will be retried. | |
0697212a | 302 | */ |
e66f17ff | 303 | void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep, |
30dad309 | 304 | spinlock_t *ptl) |
0697212a | 305 | { |
30dad309 | 306 | pte_t pte; |
0697212a | 307 | swp_entry_t entry; |
0697212a | 308 | |
30dad309 | 309 | spin_lock(ptl); |
0697212a CL |
310 | pte = *ptep; |
311 | if (!is_swap_pte(pte)) | |
312 | goto out; | |
313 | ||
314 | entry = pte_to_swp_entry(pte); | |
315 | if (!is_migration_entry(entry)) | |
316 | goto out; | |
317 | ||
ffa65753 | 318 | migration_entry_wait_on_locked(entry, ptep, ptl); |
0697212a CL |
319 | return; |
320 | out: | |
321 | pte_unmap_unlock(ptep, ptl); | |
322 | } | |
323 | ||
30dad309 NH |
324 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, |
325 | unsigned long address) | |
326 | { | |
327 | spinlock_t *ptl = pte_lockptr(mm, pmd); | |
328 | pte_t *ptep = pte_offset_map(pmd, address); | |
329 | __migration_entry_wait(mm, ptep, ptl); | |
330 | } | |
331 | ||
ad1ac596 | 332 | #ifdef CONFIG_HUGETLB_PAGE |
fcd48540 PX |
333 | /* |
334 | * The vma read lock must be held upon entry. Holding that lock prevents either | |
335 | * the pte or the ptl from being freed. | |
336 | * | |
337 | * This function will release the vma lock before returning. | |
338 | */ | |
339 | void __migration_entry_wait_huge(struct vm_area_struct *vma, | |
340 | pte_t *ptep, spinlock_t *ptl) | |
30dad309 | 341 | { |
ad1ac596 ML |
342 | pte_t pte; |
343 | ||
fcd48540 | 344 | hugetlb_vma_assert_locked(vma); |
ad1ac596 ML |
345 | spin_lock(ptl); |
346 | pte = huge_ptep_get(ptep); | |
347 | ||
fcd48540 | 348 | if (unlikely(!is_hugetlb_entry_migration(pte))) { |
ad1ac596 | 349 | spin_unlock(ptl); |
fcd48540 PX |
350 | hugetlb_vma_unlock_read(vma); |
351 | } else { | |
352 | /* | |
353 | * If migration entry existed, safe to release vma lock | |
354 | * here because the pgtable page won't be freed without the | |
355 | * pgtable lock released. See comment right above pgtable | |
356 | * lock release in migration_entry_wait_on_locked(). | |
357 | */ | |
358 | hugetlb_vma_unlock_read(vma); | |
ad1ac596 | 359 | migration_entry_wait_on_locked(pte_to_swp_entry(pte), NULL, ptl); |
fcd48540 | 360 | } |
30dad309 NH |
361 | } |
362 | ||
ad1ac596 ML |
363 | void migration_entry_wait_huge(struct vm_area_struct *vma, pte_t *pte) |
364 | { | |
365 | spinlock_t *ptl = huge_pte_lockptr(hstate_vma(vma), vma->vm_mm, pte); | |
366 | ||
fcd48540 | 367 | __migration_entry_wait_huge(vma, pte, ptl); |
ad1ac596 ML |
368 | } |
369 | #endif | |
370 | ||
616b8371 ZY |
371 | #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION |
372 | void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd) | |
373 | { | |
374 | spinlock_t *ptl; | |
616b8371 ZY |
375 | |
376 | ptl = pmd_lock(mm, pmd); | |
377 | if (!is_pmd_migration_entry(*pmd)) | |
378 | goto unlock; | |
ffa65753 | 379 | migration_entry_wait_on_locked(pmd_to_swp_entry(*pmd), NULL, ptl); |
616b8371 ZY |
380 | return; |
381 | unlock: | |
382 | spin_unlock(ptl); | |
383 | } | |
384 | #endif | |
385 | ||
108ca835 MWO |
386 | static int folio_expected_refs(struct address_space *mapping, |
387 | struct folio *folio) | |
0b3901b3 | 388 | { |
108ca835 MWO |
389 | int refs = 1; |
390 | if (!mapping) | |
391 | return refs; | |
0b3901b3 | 392 | |
108ca835 MWO |
393 | refs += folio_nr_pages(folio); |
394 | if (folio_test_private(folio)) | |
395 | refs++; | |
396 | ||
397 | return refs; | |
0b3901b3 JK |
398 | } |
399 | ||
b20a3503 | 400 | /* |
c3fcf8a5 | 401 | * Replace the page in the mapping. |
5b5c7120 CL |
402 | * |
403 | * The number of remaining references must be: | |
404 | * 1 for anonymous pages without a mapping | |
405 | * 2 for pages with a mapping | |
266cf658 | 406 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 407 | */ |
3417013e MWO |
408 | int folio_migrate_mapping(struct address_space *mapping, |
409 | struct folio *newfolio, struct folio *folio, int extra_count) | |
b20a3503 | 410 | { |
3417013e | 411 | XA_STATE(xas, &mapping->i_pages, folio_index(folio)); |
42cb14b1 HD |
412 | struct zone *oldzone, *newzone; |
413 | int dirty; | |
108ca835 | 414 | int expected_count = folio_expected_refs(mapping, folio) + extra_count; |
3417013e | 415 | long nr = folio_nr_pages(folio); |
8763cb45 | 416 | |
6c5240ae | 417 | if (!mapping) { |
0e8c7d0f | 418 | /* Anonymous page without mapping */ |
3417013e | 419 | if (folio_ref_count(folio) != expected_count) |
6c5240ae | 420 | return -EAGAIN; |
cf4b769a HD |
421 | |
422 | /* No turning back from here */ | |
3417013e MWO |
423 | newfolio->index = folio->index; |
424 | newfolio->mapping = folio->mapping; | |
425 | if (folio_test_swapbacked(folio)) | |
426 | __folio_set_swapbacked(newfolio); | |
cf4b769a | 427 | |
78bd5209 | 428 | return MIGRATEPAGE_SUCCESS; |
6c5240ae CL |
429 | } |
430 | ||
3417013e MWO |
431 | oldzone = folio_zone(folio); |
432 | newzone = folio_zone(newfolio); | |
42cb14b1 | 433 | |
89eb946a | 434 | xas_lock_irq(&xas); |
3417013e | 435 | if (!folio_ref_freeze(folio, expected_count)) { |
89eb946a | 436 | xas_unlock_irq(&xas); |
e286781d NP |
437 | return -EAGAIN; |
438 | } | |
439 | ||
b20a3503 | 440 | /* |
3417013e | 441 | * Now we know that no one else is looking at the folio: |
cf4b769a | 442 | * no turning back from here. |
b20a3503 | 443 | */ |
3417013e MWO |
444 | newfolio->index = folio->index; |
445 | newfolio->mapping = folio->mapping; | |
446 | folio_ref_add(newfolio, nr); /* add cache reference */ | |
447 | if (folio_test_swapbacked(folio)) { | |
448 | __folio_set_swapbacked(newfolio); | |
449 | if (folio_test_swapcache(folio)) { | |
450 | folio_set_swapcache(newfolio); | |
451 | newfolio->private = folio_get_private(folio); | |
6326fec1 NP |
452 | } |
453 | } else { | |
3417013e | 454 | VM_BUG_ON_FOLIO(folio_test_swapcache(folio), folio); |
b20a3503 CL |
455 | } |
456 | ||
42cb14b1 | 457 | /* Move dirty while page refs frozen and newpage not yet exposed */ |
3417013e | 458 | dirty = folio_test_dirty(folio); |
42cb14b1 | 459 | if (dirty) { |
3417013e MWO |
460 | folio_clear_dirty(folio); |
461 | folio_set_dirty(newfolio); | |
42cb14b1 HD |
462 | } |
463 | ||
3417013e | 464 | xas_store(&xas, newfolio); |
7cf9c2c7 NP |
465 | |
466 | /* | |
937a94c9 JG |
467 | * Drop cache reference from old page by unfreezing |
468 | * to one less reference. | |
7cf9c2c7 NP |
469 | * We know this isn't the last reference. |
470 | */ | |
3417013e | 471 | folio_ref_unfreeze(folio, expected_count - nr); |
7cf9c2c7 | 472 | |
89eb946a | 473 | xas_unlock(&xas); |
42cb14b1 HD |
474 | /* Leave irq disabled to prevent preemption while updating stats */ |
475 | ||
0e8c7d0f CL |
476 | /* |
477 | * If moved to a different zone then also account | |
478 | * the page for that zone. Other VM counters will be | |
479 | * taken care of when we establish references to the | |
480 | * new page and drop references to the old page. | |
481 | * | |
482 | * Note that anonymous pages are accounted for | |
4b9d0fab | 483 | * via NR_FILE_PAGES and NR_ANON_MAPPED if they |
0e8c7d0f CL |
484 | * are mapped to swap space. |
485 | */ | |
42cb14b1 | 486 | if (newzone != oldzone) { |
0d1c2072 JW |
487 | struct lruvec *old_lruvec, *new_lruvec; |
488 | struct mem_cgroup *memcg; | |
489 | ||
3417013e | 490 | memcg = folio_memcg(folio); |
0d1c2072 JW |
491 | old_lruvec = mem_cgroup_lruvec(memcg, oldzone->zone_pgdat); |
492 | new_lruvec = mem_cgroup_lruvec(memcg, newzone->zone_pgdat); | |
493 | ||
5c447d27 SB |
494 | __mod_lruvec_state(old_lruvec, NR_FILE_PAGES, -nr); |
495 | __mod_lruvec_state(new_lruvec, NR_FILE_PAGES, nr); | |
3417013e | 496 | if (folio_test_swapbacked(folio) && !folio_test_swapcache(folio)) { |
5c447d27 SB |
497 | __mod_lruvec_state(old_lruvec, NR_SHMEM, -nr); |
498 | __mod_lruvec_state(new_lruvec, NR_SHMEM, nr); | |
42cb14b1 | 499 | } |
b6038942 | 500 | #ifdef CONFIG_SWAP |
3417013e | 501 | if (folio_test_swapcache(folio)) { |
b6038942 SB |
502 | __mod_lruvec_state(old_lruvec, NR_SWAPCACHE, -nr); |
503 | __mod_lruvec_state(new_lruvec, NR_SWAPCACHE, nr); | |
504 | } | |
505 | #endif | |
f56753ac | 506 | if (dirty && mapping_can_writeback(mapping)) { |
5c447d27 SB |
507 | __mod_lruvec_state(old_lruvec, NR_FILE_DIRTY, -nr); |
508 | __mod_zone_page_state(oldzone, NR_ZONE_WRITE_PENDING, -nr); | |
509 | __mod_lruvec_state(new_lruvec, NR_FILE_DIRTY, nr); | |
510 | __mod_zone_page_state(newzone, NR_ZONE_WRITE_PENDING, nr); | |
42cb14b1 | 511 | } |
4b02108a | 512 | } |
42cb14b1 | 513 | local_irq_enable(); |
b20a3503 | 514 | |
78bd5209 | 515 | return MIGRATEPAGE_SUCCESS; |
b20a3503 | 516 | } |
3417013e | 517 | EXPORT_SYMBOL(folio_migrate_mapping); |
b20a3503 | 518 | |
290408d4 NH |
519 | /* |
520 | * The expected number of remaining references is the same as that | |
3417013e | 521 | * of folio_migrate_mapping(). |
290408d4 NH |
522 | */ |
523 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
b890ec2a | 524 | struct folio *dst, struct folio *src) |
290408d4 | 525 | { |
b890ec2a | 526 | XA_STATE(xas, &mapping->i_pages, folio_index(src)); |
290408d4 | 527 | int expected_count; |
290408d4 | 528 | |
89eb946a | 529 | xas_lock_irq(&xas); |
b890ec2a MWO |
530 | expected_count = 2 + folio_has_private(src); |
531 | if (!folio_ref_freeze(src, expected_count)) { | |
89eb946a | 532 | xas_unlock_irq(&xas); |
290408d4 NH |
533 | return -EAGAIN; |
534 | } | |
535 | ||
b890ec2a MWO |
536 | dst->index = src->index; |
537 | dst->mapping = src->mapping; | |
6a93ca8f | 538 | |
b890ec2a | 539 | folio_get(dst); |
290408d4 | 540 | |
b890ec2a | 541 | xas_store(&xas, dst); |
290408d4 | 542 | |
b890ec2a | 543 | folio_ref_unfreeze(src, expected_count - 1); |
290408d4 | 544 | |
89eb946a | 545 | xas_unlock_irq(&xas); |
6a93ca8f | 546 | |
78bd5209 | 547 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
548 | } |
549 | ||
b20a3503 | 550 | /* |
19138349 | 551 | * Copy the flags and some other ancillary information |
b20a3503 | 552 | */ |
19138349 | 553 | void folio_migrate_flags(struct folio *newfolio, struct folio *folio) |
b20a3503 | 554 | { |
7851a45c RR |
555 | int cpupid; |
556 | ||
19138349 MWO |
557 | if (folio_test_error(folio)) |
558 | folio_set_error(newfolio); | |
559 | if (folio_test_referenced(folio)) | |
560 | folio_set_referenced(newfolio); | |
561 | if (folio_test_uptodate(folio)) | |
562 | folio_mark_uptodate(newfolio); | |
563 | if (folio_test_clear_active(folio)) { | |
564 | VM_BUG_ON_FOLIO(folio_test_unevictable(folio), folio); | |
565 | folio_set_active(newfolio); | |
566 | } else if (folio_test_clear_unevictable(folio)) | |
567 | folio_set_unevictable(newfolio); | |
568 | if (folio_test_workingset(folio)) | |
569 | folio_set_workingset(newfolio); | |
570 | if (folio_test_checked(folio)) | |
571 | folio_set_checked(newfolio); | |
6c287605 DH |
572 | /* |
573 | * PG_anon_exclusive (-> PG_mappedtodisk) is always migrated via | |
574 | * migration entries. We can still have PG_anon_exclusive set on an | |
575 | * effectively unmapped and unreferenced first sub-pages of an | |
576 | * anonymous THP: we can simply copy it here via PG_mappedtodisk. | |
577 | */ | |
19138349 MWO |
578 | if (folio_test_mappedtodisk(folio)) |
579 | folio_set_mappedtodisk(newfolio); | |
b20a3503 | 580 | |
3417013e | 581 | /* Move dirty on pages not done by folio_migrate_mapping() */ |
19138349 MWO |
582 | if (folio_test_dirty(folio)) |
583 | folio_set_dirty(newfolio); | |
b20a3503 | 584 | |
19138349 MWO |
585 | if (folio_test_young(folio)) |
586 | folio_set_young(newfolio); | |
587 | if (folio_test_idle(folio)) | |
588 | folio_set_idle(newfolio); | |
33c3fc71 | 589 | |
7851a45c RR |
590 | /* |
591 | * Copy NUMA information to the new page, to prevent over-eager | |
592 | * future migrations of this same page. | |
593 | */ | |
19138349 | 594 | cpupid = page_cpupid_xchg_last(&folio->page, -1); |
33024536 HY |
595 | /* |
596 | * For memory tiering mode, when migrate between slow and fast | |
597 | * memory node, reset cpupid, because that is used to record | |
598 | * page access time in slow memory node. | |
599 | */ | |
600 | if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) { | |
601 | bool f_toptier = node_is_toptier(page_to_nid(&folio->page)); | |
602 | bool t_toptier = node_is_toptier(page_to_nid(&newfolio->page)); | |
603 | ||
604 | if (f_toptier != t_toptier) | |
605 | cpupid = -1; | |
606 | } | |
19138349 | 607 | page_cpupid_xchg_last(&newfolio->page, cpupid); |
7851a45c | 608 | |
19138349 | 609 | folio_migrate_ksm(newfolio, folio); |
c8d6553b HD |
610 | /* |
611 | * Please do not reorder this without considering how mm/ksm.c's | |
612 | * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache(). | |
613 | */ | |
19138349 MWO |
614 | if (folio_test_swapcache(folio)) |
615 | folio_clear_swapcache(folio); | |
616 | folio_clear_private(folio); | |
ad2fa371 MS |
617 | |
618 | /* page->private contains hugetlb specific flags */ | |
19138349 MWO |
619 | if (!folio_test_hugetlb(folio)) |
620 | folio->private = NULL; | |
b20a3503 CL |
621 | |
622 | /* | |
623 | * If any waiters have accumulated on the new page then | |
624 | * wake them up. | |
625 | */ | |
19138349 MWO |
626 | if (folio_test_writeback(newfolio)) |
627 | folio_end_writeback(newfolio); | |
d435edca | 628 | |
6aeff241 YS |
629 | /* |
630 | * PG_readahead shares the same bit with PG_reclaim. The above | |
631 | * end_page_writeback() may clear PG_readahead mistakenly, so set the | |
632 | * bit after that. | |
633 | */ | |
19138349 MWO |
634 | if (folio_test_readahead(folio)) |
635 | folio_set_readahead(newfolio); | |
6aeff241 | 636 | |
19138349 | 637 | folio_copy_owner(newfolio, folio); |
74485cf2 | 638 | |
19138349 | 639 | if (!folio_test_hugetlb(folio)) |
d21bba2b | 640 | mem_cgroup_migrate(folio, newfolio); |
b20a3503 | 641 | } |
19138349 | 642 | EXPORT_SYMBOL(folio_migrate_flags); |
2916ecc0 | 643 | |
715cbfd6 | 644 | void folio_migrate_copy(struct folio *newfolio, struct folio *folio) |
2916ecc0 | 645 | { |
715cbfd6 MWO |
646 | folio_copy(newfolio, folio); |
647 | folio_migrate_flags(newfolio, folio); | |
2916ecc0 | 648 | } |
715cbfd6 | 649 | EXPORT_SYMBOL(folio_migrate_copy); |
b20a3503 | 650 | |
1d8b85cc CL |
651 | /************************************************************ |
652 | * Migration functions | |
653 | ***********************************************************/ | |
654 | ||
16ce101d AP |
655 | int migrate_folio_extra(struct address_space *mapping, struct folio *dst, |
656 | struct folio *src, enum migrate_mode mode, int extra_count) | |
657 | { | |
658 | int rc; | |
659 | ||
660 | BUG_ON(folio_test_writeback(src)); /* Writeback must be complete */ | |
661 | ||
662 | rc = folio_migrate_mapping(mapping, dst, src, extra_count); | |
663 | ||
664 | if (rc != MIGRATEPAGE_SUCCESS) | |
665 | return rc; | |
666 | ||
667 | if (mode != MIGRATE_SYNC_NO_COPY) | |
668 | folio_migrate_copy(dst, src); | |
669 | else | |
670 | folio_migrate_flags(dst, src); | |
671 | return MIGRATEPAGE_SUCCESS; | |
672 | } | |
673 | ||
54184650 MWO |
674 | /** |
675 | * migrate_folio() - Simple folio migration. | |
676 | * @mapping: The address_space containing the folio. | |
677 | * @dst: The folio to migrate the data to. | |
678 | * @src: The folio containing the current data. | |
679 | * @mode: How to migrate the page. | |
680 | * | |
681 | * Common logic to directly migrate a single LRU folio suitable for | |
682 | * folios that do not use PagePrivate/PagePrivate2. | |
b20a3503 | 683 | * |
54184650 | 684 | * Folios are locked upon entry and exit. |
b20a3503 | 685 | */ |
54184650 MWO |
686 | int migrate_folio(struct address_space *mapping, struct folio *dst, |
687 | struct folio *src, enum migrate_mode mode) | |
b20a3503 | 688 | { |
16ce101d | 689 | return migrate_folio_extra(mapping, dst, src, mode, 0); |
b20a3503 | 690 | } |
54184650 | 691 | EXPORT_SYMBOL(migrate_folio); |
b20a3503 | 692 | |
9361401e | 693 | #ifdef CONFIG_BLOCK |
84ade7c1 JK |
694 | /* Returns true if all buffers are successfully locked */ |
695 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, | |
696 | enum migrate_mode mode) | |
697 | { | |
698 | struct buffer_head *bh = head; | |
699 | ||
700 | /* Simple case, sync compaction */ | |
701 | if (mode != MIGRATE_ASYNC) { | |
702 | do { | |
84ade7c1 JK |
703 | lock_buffer(bh); |
704 | bh = bh->b_this_page; | |
705 | ||
706 | } while (bh != head); | |
707 | ||
708 | return true; | |
709 | } | |
710 | ||
711 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ | |
712 | do { | |
84ade7c1 JK |
713 | if (!trylock_buffer(bh)) { |
714 | /* | |
715 | * We failed to lock the buffer and cannot stall in | |
716 | * async migration. Release the taken locks | |
717 | */ | |
718 | struct buffer_head *failed_bh = bh; | |
84ade7c1 JK |
719 | bh = head; |
720 | while (bh != failed_bh) { | |
721 | unlock_buffer(bh); | |
84ade7c1 JK |
722 | bh = bh->b_this_page; |
723 | } | |
724 | return false; | |
725 | } | |
726 | ||
727 | bh = bh->b_this_page; | |
728 | } while (bh != head); | |
729 | return true; | |
730 | } | |
731 | ||
67235182 MWO |
732 | static int __buffer_migrate_folio(struct address_space *mapping, |
733 | struct folio *dst, struct folio *src, enum migrate_mode mode, | |
89cb0888 | 734 | bool check_refs) |
1d8b85cc | 735 | { |
1d8b85cc CL |
736 | struct buffer_head *bh, *head; |
737 | int rc; | |
cc4f11e6 | 738 | int expected_count; |
1d8b85cc | 739 | |
67235182 MWO |
740 | head = folio_buffers(src); |
741 | if (!head) | |
54184650 | 742 | return migrate_folio(mapping, dst, src, mode); |
1d8b85cc | 743 | |
cc4f11e6 | 744 | /* Check whether page does not have extra refs before we do more work */ |
108ca835 | 745 | expected_count = folio_expected_refs(mapping, src); |
67235182 | 746 | if (folio_ref_count(src) != expected_count) |
cc4f11e6 | 747 | return -EAGAIN; |
1d8b85cc | 748 | |
cc4f11e6 JK |
749 | if (!buffer_migrate_lock_buffers(head, mode)) |
750 | return -EAGAIN; | |
1d8b85cc | 751 | |
89cb0888 JK |
752 | if (check_refs) { |
753 | bool busy; | |
754 | bool invalidated = false; | |
755 | ||
756 | recheck_buffers: | |
757 | busy = false; | |
758 | spin_lock(&mapping->private_lock); | |
759 | bh = head; | |
760 | do { | |
761 | if (atomic_read(&bh->b_count)) { | |
762 | busy = true; | |
763 | break; | |
764 | } | |
765 | bh = bh->b_this_page; | |
766 | } while (bh != head); | |
89cb0888 JK |
767 | if (busy) { |
768 | if (invalidated) { | |
769 | rc = -EAGAIN; | |
770 | goto unlock_buffers; | |
771 | } | |
ebdf4de5 | 772 | spin_unlock(&mapping->private_lock); |
89cb0888 JK |
773 | invalidate_bh_lrus(); |
774 | invalidated = true; | |
775 | goto recheck_buffers; | |
776 | } | |
777 | } | |
778 | ||
67235182 | 779 | rc = folio_migrate_mapping(mapping, dst, src, 0); |
78bd5209 | 780 | if (rc != MIGRATEPAGE_SUCCESS) |
cc4f11e6 | 781 | goto unlock_buffers; |
1d8b85cc | 782 | |
67235182 | 783 | folio_attach_private(dst, folio_detach_private(src)); |
1d8b85cc CL |
784 | |
785 | bh = head; | |
786 | do { | |
67235182 | 787 | set_bh_page(bh, &dst->page, bh_offset(bh)); |
1d8b85cc | 788 | bh = bh->b_this_page; |
1d8b85cc CL |
789 | } while (bh != head); |
790 | ||
2916ecc0 | 791 | if (mode != MIGRATE_SYNC_NO_COPY) |
67235182 | 792 | folio_migrate_copy(dst, src); |
2916ecc0 | 793 | else |
67235182 | 794 | folio_migrate_flags(dst, src); |
1d8b85cc | 795 | |
cc4f11e6 JK |
796 | rc = MIGRATEPAGE_SUCCESS; |
797 | unlock_buffers: | |
ebdf4de5 JK |
798 | if (check_refs) |
799 | spin_unlock(&mapping->private_lock); | |
1d8b85cc CL |
800 | bh = head; |
801 | do { | |
802 | unlock_buffer(bh); | |
1d8b85cc | 803 | bh = bh->b_this_page; |
1d8b85cc CL |
804 | } while (bh != head); |
805 | ||
cc4f11e6 | 806 | return rc; |
1d8b85cc | 807 | } |
89cb0888 | 808 | |
67235182 MWO |
809 | /** |
810 | * buffer_migrate_folio() - Migration function for folios with buffers. | |
811 | * @mapping: The address space containing @src. | |
812 | * @dst: The folio to migrate to. | |
813 | * @src: The folio to migrate from. | |
814 | * @mode: How to migrate the folio. | |
815 | * | |
816 | * This function can only be used if the underlying filesystem guarantees | |
817 | * that no other references to @src exist. For example attached buffer | |
818 | * heads are accessed only under the folio lock. If your filesystem cannot | |
819 | * provide this guarantee, buffer_migrate_folio_norefs() may be more | |
820 | * appropriate. | |
821 | * | |
822 | * Return: 0 on success or a negative errno on failure. | |
89cb0888 | 823 | */ |
67235182 MWO |
824 | int buffer_migrate_folio(struct address_space *mapping, |
825 | struct folio *dst, struct folio *src, enum migrate_mode mode) | |
89cb0888 | 826 | { |
67235182 | 827 | return __buffer_migrate_folio(mapping, dst, src, mode, false); |
89cb0888 | 828 | } |
67235182 MWO |
829 | EXPORT_SYMBOL(buffer_migrate_folio); |
830 | ||
831 | /** | |
832 | * buffer_migrate_folio_norefs() - Migration function for folios with buffers. | |
833 | * @mapping: The address space containing @src. | |
834 | * @dst: The folio to migrate to. | |
835 | * @src: The folio to migrate from. | |
836 | * @mode: How to migrate the folio. | |
837 | * | |
838 | * Like buffer_migrate_folio() except that this variant is more careful | |
839 | * and checks that there are also no buffer head references. This function | |
840 | * is the right one for mappings where buffer heads are directly looked | |
841 | * up and referenced (such as block device mappings). | |
842 | * | |
843 | * Return: 0 on success or a negative errno on failure. | |
89cb0888 | 844 | */ |
67235182 MWO |
845 | int buffer_migrate_folio_norefs(struct address_space *mapping, |
846 | struct folio *dst, struct folio *src, enum migrate_mode mode) | |
89cb0888 | 847 | { |
67235182 | 848 | return __buffer_migrate_folio(mapping, dst, src, mode, true); |
89cb0888 | 849 | } |
e26355e2 | 850 | EXPORT_SYMBOL_GPL(buffer_migrate_folio_norefs); |
9361401e | 851 | #endif |
1d8b85cc | 852 | |
2ec810d5 MWO |
853 | int filemap_migrate_folio(struct address_space *mapping, |
854 | struct folio *dst, struct folio *src, enum migrate_mode mode) | |
855 | { | |
856 | int ret; | |
857 | ||
858 | ret = folio_migrate_mapping(mapping, dst, src, 0); | |
859 | if (ret != MIGRATEPAGE_SUCCESS) | |
860 | return ret; | |
861 | ||
862 | if (folio_get_private(src)) | |
863 | folio_attach_private(dst, folio_detach_private(src)); | |
864 | ||
865 | if (mode != MIGRATE_SYNC_NO_COPY) | |
866 | folio_migrate_copy(dst, src); | |
867 | else | |
868 | folio_migrate_flags(dst, src); | |
869 | return MIGRATEPAGE_SUCCESS; | |
870 | } | |
871 | EXPORT_SYMBOL_GPL(filemap_migrate_folio); | |
872 | ||
04e62a29 | 873 | /* |
2be7fa10 | 874 | * Writeback a folio to clean the dirty state |
04e62a29 | 875 | */ |
2be7fa10 | 876 | static int writeout(struct address_space *mapping, struct folio *folio) |
8351a6e4 | 877 | { |
04e62a29 CL |
878 | struct writeback_control wbc = { |
879 | .sync_mode = WB_SYNC_NONE, | |
880 | .nr_to_write = 1, | |
881 | .range_start = 0, | |
882 | .range_end = LLONG_MAX, | |
04e62a29 CL |
883 | .for_reclaim = 1 |
884 | }; | |
885 | int rc; | |
886 | ||
887 | if (!mapping->a_ops->writepage) | |
888 | /* No write method for the address space */ | |
889 | return -EINVAL; | |
890 | ||
2be7fa10 | 891 | if (!folio_clear_dirty_for_io(folio)) |
04e62a29 CL |
892 | /* Someone else already triggered a write */ |
893 | return -EAGAIN; | |
894 | ||
8351a6e4 | 895 | /* |
2be7fa10 MWO |
896 | * A dirty folio may imply that the underlying filesystem has |
897 | * the folio on some queue. So the folio must be clean for | |
898 | * migration. Writeout may mean we lose the lock and the | |
899 | * folio state is no longer what we checked for earlier. | |
04e62a29 CL |
900 | * At this point we know that the migration attempt cannot |
901 | * be successful. | |
8351a6e4 | 902 | */ |
4eecb8b9 | 903 | remove_migration_ptes(folio, folio, false); |
8351a6e4 | 904 | |
2be7fa10 | 905 | rc = mapping->a_ops->writepage(&folio->page, &wbc); |
8351a6e4 | 906 | |
04e62a29 CL |
907 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
908 | /* unlocked. Relock */ | |
2be7fa10 | 909 | folio_lock(folio); |
04e62a29 | 910 | |
bda8550d | 911 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
912 | } |
913 | ||
914 | /* | |
915 | * Default handling if a filesystem does not provide a migration function. | |
916 | */ | |
8faa8ef5 MWO |
917 | static int fallback_migrate_folio(struct address_space *mapping, |
918 | struct folio *dst, struct folio *src, enum migrate_mode mode) | |
04e62a29 | 919 | { |
8faa8ef5 MWO |
920 | if (folio_test_dirty(src)) { |
921 | /* Only writeback folios in full synchronous migration */ | |
2916ecc0 JG |
922 | switch (mode) { |
923 | case MIGRATE_SYNC: | |
924 | case MIGRATE_SYNC_NO_COPY: | |
925 | break; | |
926 | default: | |
b969c4ab | 927 | return -EBUSY; |
2916ecc0 | 928 | } |
2be7fa10 | 929 | return writeout(mapping, src); |
b969c4ab | 930 | } |
8351a6e4 CL |
931 | |
932 | /* | |
933 | * Buffers may be managed in a filesystem specific way. | |
934 | * We must have no buffers or drop them. | |
935 | */ | |
8faa8ef5 MWO |
936 | if (folio_test_private(src) && |
937 | !filemap_release_folio(src, GFP_KERNEL)) | |
806031bb | 938 | return mode == MIGRATE_SYNC ? -EAGAIN : -EBUSY; |
8351a6e4 | 939 | |
54184650 | 940 | return migrate_folio(mapping, dst, src, mode); |
8351a6e4 CL |
941 | } |
942 | ||
e24f0b8f CL |
943 | /* |
944 | * Move a page to a newly allocated page | |
945 | * The page is locked and all ptes have been successfully removed. | |
946 | * | |
947 | * The new page will have replaced the old page if this function | |
948 | * is successful. | |
894bc310 LS |
949 | * |
950 | * Return value: | |
951 | * < 0 - error code | |
78bd5209 | 952 | * MIGRATEPAGE_SUCCESS - success |
e24f0b8f | 953 | */ |
e7e3ffeb | 954 | static int move_to_new_folio(struct folio *dst, struct folio *src, |
5c3f9a67 | 955 | enum migrate_mode mode) |
e24f0b8f | 956 | { |
bda807d4 | 957 | int rc = -EAGAIN; |
e7e3ffeb | 958 | bool is_lru = !__PageMovable(&src->page); |
e24f0b8f | 959 | |
e7e3ffeb MWO |
960 | VM_BUG_ON_FOLIO(!folio_test_locked(src), src); |
961 | VM_BUG_ON_FOLIO(!folio_test_locked(dst), dst); | |
e24f0b8f | 962 | |
bda807d4 | 963 | if (likely(is_lru)) { |
68f2736a MWO |
964 | struct address_space *mapping = folio_mapping(src); |
965 | ||
bda807d4 | 966 | if (!mapping) |
54184650 | 967 | rc = migrate_folio(mapping, dst, src, mode); |
5490da4f | 968 | else if (mapping->a_ops->migrate_folio) |
bda807d4 | 969 | /* |
5490da4f MWO |
970 | * Most folios have a mapping and most filesystems |
971 | * provide a migrate_folio callback. Anonymous folios | |
bda807d4 | 972 | * are part of swap space which also has its own |
5490da4f | 973 | * migrate_folio callback. This is the most common path |
bda807d4 MK |
974 | * for page migration. |
975 | */ | |
5490da4f MWO |
976 | rc = mapping->a_ops->migrate_folio(mapping, dst, src, |
977 | mode); | |
bda807d4 | 978 | else |
8faa8ef5 | 979 | rc = fallback_migrate_folio(mapping, dst, src, mode); |
bda807d4 | 980 | } else { |
68f2736a MWO |
981 | const struct movable_operations *mops; |
982 | ||
e24f0b8f | 983 | /* |
bda807d4 MK |
984 | * In case of non-lru page, it could be released after |
985 | * isolation step. In that case, we shouldn't try migration. | |
e24f0b8f | 986 | */ |
e7e3ffeb MWO |
987 | VM_BUG_ON_FOLIO(!folio_test_isolated(src), src); |
988 | if (!folio_test_movable(src)) { | |
bda807d4 | 989 | rc = MIGRATEPAGE_SUCCESS; |
e7e3ffeb | 990 | folio_clear_isolated(src); |
bda807d4 MK |
991 | goto out; |
992 | } | |
993 | ||
da707a6d | 994 | mops = folio_movable_ops(src); |
68f2736a | 995 | rc = mops->migrate_page(&dst->page, &src->page, mode); |
bda807d4 | 996 | WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS && |
e7e3ffeb | 997 | !folio_test_isolated(src)); |
bda807d4 | 998 | } |
e24f0b8f | 999 | |
5c3f9a67 | 1000 | /* |
e7e3ffeb MWO |
1001 | * When successful, old pagecache src->mapping must be cleared before |
1002 | * src is freed; but stats require that PageAnon be left as PageAnon. | |
5c3f9a67 HD |
1003 | */ |
1004 | if (rc == MIGRATEPAGE_SUCCESS) { | |
e7e3ffeb MWO |
1005 | if (__PageMovable(&src->page)) { |
1006 | VM_BUG_ON_FOLIO(!folio_test_isolated(src), src); | |
bda807d4 MK |
1007 | |
1008 | /* | |
1009 | * We clear PG_movable under page_lock so any compactor | |
1010 | * cannot try to migrate this page. | |
1011 | */ | |
e7e3ffeb | 1012 | folio_clear_isolated(src); |
bda807d4 MK |
1013 | } |
1014 | ||
1015 | /* | |
e7e3ffeb | 1016 | * Anonymous and movable src->mapping will be cleared by |
bda807d4 MK |
1017 | * free_pages_prepare so don't reset it here for keeping |
1018 | * the type to work PageAnon, for example. | |
1019 | */ | |
e7e3ffeb MWO |
1020 | if (!folio_mapping_flags(src)) |
1021 | src->mapping = NULL; | |
d2b2c6dd | 1022 | |
e7e3ffeb MWO |
1023 | if (likely(!folio_is_zone_device(dst))) |
1024 | flush_dcache_folio(dst); | |
3fe2011f | 1025 | } |
bda807d4 | 1026 | out: |
e24f0b8f CL |
1027 | return rc; |
1028 | } | |
1029 | ||
682a71a1 | 1030 | static int __unmap_and_move(struct folio *src, struct folio *dst, |
9c620e2b | 1031 | int force, enum migrate_mode mode) |
e24f0b8f | 1032 | { |
0dabec93 | 1033 | int rc = -EAGAIN; |
213ecb31 | 1034 | bool page_was_mapped = false; |
3f6c8272 | 1035 | struct anon_vma *anon_vma = NULL; |
682a71a1 | 1036 | bool is_lru = !__PageMovable(&src->page); |
95a402c3 | 1037 | |
682a71a1 | 1038 | if (!folio_trylock(src)) { |
a6bc32b8 | 1039 | if (!force || mode == MIGRATE_ASYNC) |
0dabec93 | 1040 | goto out; |
3e7d3449 MG |
1041 | |
1042 | /* | |
1043 | * It's not safe for direct compaction to call lock_page. | |
1044 | * For example, during page readahead pages are added locked | |
1045 | * to the LRU. Later, when the IO completes the pages are | |
1046 | * marked uptodate and unlocked. However, the queueing | |
1047 | * could be merging multiple pages for one bio (e.g. | |
d4388340 | 1048 | * mpage_readahead). If an allocation happens for the |
3e7d3449 MG |
1049 | * second or third page, the process can end up locking |
1050 | * the same page twice and deadlocking. Rather than | |
1051 | * trying to be clever about what pages can be locked, | |
1052 | * avoid the use of lock_page for direct compaction | |
1053 | * altogether. | |
1054 | */ | |
1055 | if (current->flags & PF_MEMALLOC) | |
0dabec93 | 1056 | goto out; |
3e7d3449 | 1057 | |
682a71a1 | 1058 | folio_lock(src); |
e24f0b8f CL |
1059 | } |
1060 | ||
682a71a1 | 1061 | if (folio_test_writeback(src)) { |
11bc82d6 | 1062 | /* |
fed5b64a | 1063 | * Only in the case of a full synchronous migration is it |
a6bc32b8 MG |
1064 | * necessary to wait for PageWriteback. In the async case, |
1065 | * the retry loop is too short and in the sync-light case, | |
1066 | * the overhead of stalling is too much | |
11bc82d6 | 1067 | */ |
2916ecc0 JG |
1068 | switch (mode) { |
1069 | case MIGRATE_SYNC: | |
1070 | case MIGRATE_SYNC_NO_COPY: | |
1071 | break; | |
1072 | default: | |
11bc82d6 | 1073 | rc = -EBUSY; |
0a31bc97 | 1074 | goto out_unlock; |
11bc82d6 AA |
1075 | } |
1076 | if (!force) | |
0a31bc97 | 1077 | goto out_unlock; |
682a71a1 | 1078 | folio_wait_writeback(src); |
e24f0b8f | 1079 | } |
03f15c86 | 1080 | |
e24f0b8f | 1081 | /* |
682a71a1 MWO |
1082 | * By try_to_migrate(), src->mapcount goes down to 0 here. In this case, |
1083 | * we cannot notice that anon_vma is freed while we migrate a page. | |
1ce82b69 | 1084 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
dc386d4d | 1085 | * of migration. File cache pages are no problem because of page_lock() |
989f89c5 KH |
1086 | * File Caches may use write_page() or lock_page() in migration, then, |
1087 | * just care Anon page here. | |
03f15c86 | 1088 | * |
29eea9b5 | 1089 | * Only folio_get_anon_vma() understands the subtleties of |
03f15c86 HD |
1090 | * getting a hold on an anon_vma from outside one of its mms. |
1091 | * But if we cannot get anon_vma, then we won't need it anyway, | |
1092 | * because that implies that the anon page is no longer mapped | |
1093 | * (and cannot be remapped so long as we hold the page lock). | |
dc386d4d | 1094 | */ |
682a71a1 | 1095 | if (folio_test_anon(src) && !folio_test_ksm(src)) |
29eea9b5 | 1096 | anon_vma = folio_get_anon_vma(src); |
62e1c553 | 1097 | |
7db7671f HD |
1098 | /* |
1099 | * Block others from accessing the new page when we get around to | |
1100 | * establishing additional references. We are usually the only one | |
682a71a1 MWO |
1101 | * holding a reference to dst at this point. We used to have a BUG |
1102 | * here if folio_trylock(dst) fails, but would like to allow for | |
1103 | * cases where there might be a race with the previous use of dst. | |
7db7671f HD |
1104 | * This is much like races on refcount of oldpage: just don't BUG(). |
1105 | */ | |
682a71a1 | 1106 | if (unlikely(!folio_trylock(dst))) |
7db7671f HD |
1107 | goto out_unlock; |
1108 | ||
bda807d4 | 1109 | if (unlikely(!is_lru)) { |
682a71a1 | 1110 | rc = move_to_new_folio(dst, src, mode); |
bda807d4 MK |
1111 | goto out_unlock_both; |
1112 | } | |
1113 | ||
dc386d4d | 1114 | /* |
62e1c553 SL |
1115 | * Corner case handling: |
1116 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
1117 | * and treated as swapcache but it has no rmap yet. | |
682a71a1 | 1118 | * Calling try_to_unmap() against a src->mapping==NULL page will |
62e1c553 | 1119 | * trigger a BUG. So handle it here. |
d12b8951 | 1120 | * 2. An orphaned page (see truncate_cleanup_page) might have |
62e1c553 SL |
1121 | * fs-private metadata. The page can be picked up due to memory |
1122 | * offlining. Everywhere else except page reclaim, the page is | |
1123 | * invisible to the vm, so the page can not be migrated. So try to | |
1124 | * free the metadata, so the page can be freed. | |
e24f0b8f | 1125 | */ |
682a71a1 MWO |
1126 | if (!src->mapping) { |
1127 | if (folio_test_private(src)) { | |
1128 | try_to_free_buffers(src); | |
7db7671f | 1129 | goto out_unlock_both; |
62e1c553 | 1130 | } |
682a71a1 | 1131 | } else if (folio_mapped(src)) { |
7db7671f | 1132 | /* Establish migration ptes */ |
682a71a1 MWO |
1133 | VM_BUG_ON_FOLIO(folio_test_anon(src) && |
1134 | !folio_test_ksm(src) && !anon_vma, src); | |
1135 | try_to_migrate(src, 0); | |
213ecb31 | 1136 | page_was_mapped = true; |
2ebba6b7 | 1137 | } |
dc386d4d | 1138 | |
682a71a1 MWO |
1139 | if (!folio_mapped(src)) |
1140 | rc = move_to_new_folio(dst, src, mode); | |
e24f0b8f | 1141 | |
c3096e67 | 1142 | /* |
682a71a1 | 1143 | * When successful, push dst to LRU immediately: so that if it |
c3096e67 | 1144 | * turns out to be an mlocked page, remove_migration_ptes() will |
682a71a1 | 1145 | * automatically build up the correct dst->mlock_count for it. |
c3096e67 HD |
1146 | * |
1147 | * We would like to do something similar for the old page, when | |
1148 | * unsuccessful, and other cases when a page has been temporarily | |
1149 | * isolated from the unevictable LRU: but this case is the easiest. | |
1150 | */ | |
1151 | if (rc == MIGRATEPAGE_SUCCESS) { | |
682a71a1 | 1152 | folio_add_lru(dst); |
c3096e67 HD |
1153 | if (page_was_mapped) |
1154 | lru_add_drain(); | |
1155 | } | |
1156 | ||
5c3f9a67 | 1157 | if (page_was_mapped) |
682a71a1 MWO |
1158 | remove_migration_ptes(src, |
1159 | rc == MIGRATEPAGE_SUCCESS ? dst : src, false); | |
3f6c8272 | 1160 | |
7db7671f | 1161 | out_unlock_both: |
682a71a1 | 1162 | folio_unlock(dst); |
7db7671f | 1163 | out_unlock: |
3f6c8272 | 1164 | /* Drop an anon_vma reference if we took one */ |
76545066 | 1165 | if (anon_vma) |
9e60109f | 1166 | put_anon_vma(anon_vma); |
682a71a1 | 1167 | folio_unlock(src); |
0dabec93 | 1168 | out: |
c6c919eb | 1169 | /* |
682a71a1 | 1170 | * If migration is successful, decrease refcount of dst, |
c6c919eb | 1171 | * which will not free the page because new page owner increased |
c3096e67 | 1172 | * refcounter. |
c6c919eb | 1173 | */ |
c3096e67 | 1174 | if (rc == MIGRATEPAGE_SUCCESS) |
682a71a1 | 1175 | folio_put(dst); |
c6c919eb | 1176 | |
0dabec93 MK |
1177 | return rc; |
1178 | } | |
95a402c3 | 1179 | |
0dabec93 | 1180 | /* |
49f51859 HY |
1181 | * Obtain the lock on folio, remove all ptes and migrate the folio |
1182 | * to the newly allocated folio in dst. | |
0dabec93 | 1183 | */ |
6ec4476a | 1184 | static int unmap_and_move(new_page_t get_new_page, |
ef2a5153 | 1185 | free_page_t put_new_page, |
49f51859 | 1186 | unsigned long private, struct folio *src, |
add05cec | 1187 | int force, enum migrate_mode mode, |
dd4ae78a YS |
1188 | enum migrate_reason reason, |
1189 | struct list_head *ret) | |
0dabec93 | 1190 | { |
49f51859 | 1191 | struct folio *dst; |
2def7424 | 1192 | int rc = MIGRATEPAGE_SUCCESS; |
74d4a579 | 1193 | struct page *newpage = NULL; |
0dabec93 | 1194 | |
49f51859 | 1195 | if (!thp_migration_supported() && folio_test_transhuge(src)) |
d532e2e5 | 1196 | return -ENOSYS; |
94723aaf | 1197 | |
49f51859 HY |
1198 | if (folio_ref_count(src) == 1) { |
1199 | /* Folio was freed from under us. So we are done. */ | |
1200 | folio_clear_active(src); | |
1201 | folio_clear_unevictable(src); | |
160088b3 | 1202 | /* free_pages_prepare() will clear PG_isolated. */ |
0dabec93 MK |
1203 | goto out; |
1204 | } | |
1205 | ||
49f51859 | 1206 | newpage = get_new_page(&src->page, private); |
74d4a579 YS |
1207 | if (!newpage) |
1208 | return -ENOMEM; | |
682a71a1 | 1209 | dst = page_folio(newpage); |
74d4a579 | 1210 | |
4c74b65f | 1211 | dst->private = NULL; |
682a71a1 | 1212 | rc = __unmap_and_move(src, dst, force, mode); |
c6c919eb | 1213 | if (rc == MIGRATEPAGE_SUCCESS) |
49f51859 | 1214 | set_page_owner_migrate_reason(&dst->page, reason); |
bf6bddf1 | 1215 | |
0dabec93 | 1216 | out: |
e24f0b8f | 1217 | if (rc != -EAGAIN) { |
0dabec93 | 1218 | /* |
49f51859 HY |
1219 | * A folio that has been migrated has all references |
1220 | * removed and will be freed. A folio that has not been | |
c23a0c99 | 1221 | * migrated will have kept its references and be restored. |
0dabec93 | 1222 | */ |
49f51859 | 1223 | list_del(&src->lru); |
dd4ae78a | 1224 | } |
6afcf8ef | 1225 | |
dd4ae78a YS |
1226 | /* |
1227 | * If migration is successful, releases reference grabbed during | |
49f51859 | 1228 | * isolation. Otherwise, restore the folio to right list unless |
dd4ae78a YS |
1229 | * we want to retry. |
1230 | */ | |
1231 | if (rc == MIGRATEPAGE_SUCCESS) { | |
6afcf8ef | 1232 | /* |
49f51859 | 1233 | * Compaction can migrate also non-LRU folios which are |
6afcf8ef | 1234 | * not accounted to NR_ISOLATED_*. They can be recognized |
49f51859 | 1235 | * as __folio_test_movable |
6afcf8ef | 1236 | */ |
49f51859 HY |
1237 | if (likely(!__folio_test_movable(src))) |
1238 | mod_node_page_state(folio_pgdat(src), NR_ISOLATED_ANON + | |
1239 | folio_is_file_lru(src), -folio_nr_pages(src)); | |
c6c919eb | 1240 | |
79f5f8fa | 1241 | if (reason != MR_MEMORY_FAILURE) |
d7e69488 | 1242 | /* |
49f51859 | 1243 | * We release the folio in page_handle_poison. |
d7e69488 | 1244 | */ |
49f51859 | 1245 | folio_put(src); |
c6c919eb | 1246 | } else { |
dd4ae78a | 1247 | if (rc != -EAGAIN) |
49f51859 | 1248 | list_add_tail(&src->lru, ret); |
bda807d4 | 1249 | |
c6c919eb | 1250 | if (put_new_page) |
49f51859 | 1251 | put_new_page(&dst->page, private); |
c6c919eb | 1252 | else |
49f51859 | 1253 | folio_put(dst); |
e24f0b8f | 1254 | } |
68711a74 | 1255 | |
e24f0b8f CL |
1256 | return rc; |
1257 | } | |
1258 | ||
290408d4 NH |
1259 | /* |
1260 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
1261 | * | |
1262 | * This function doesn't wait the completion of hugepage I/O | |
1263 | * because there is no race between I/O and migration for hugepage. | |
1264 | * Note that currently hugepage I/O occurs only in direct I/O | |
1265 | * where no lock is held and PG_writeback is irrelevant, | |
1266 | * and writeback status of all subpages are counted in the reference | |
1267 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
1268 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
1269 | * This means that when we try to migrate hugepage whose subpages are | |
1270 | * doing direct I/O, some references remain after try_to_unmap() and | |
1271 | * hugepage migration fails without data corruption. | |
1272 | * | |
1273 | * There is also no race when direct I/O is issued on the page under migration, | |
1274 | * because then pte is replaced with migration swap entry and direct I/O code | |
1275 | * will wait in the page fault for migration to complete. | |
1276 | */ | |
1277 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
68711a74 DR |
1278 | free_page_t put_new_page, unsigned long private, |
1279 | struct page *hpage, int force, | |
dd4ae78a YS |
1280 | enum migrate_mode mode, int reason, |
1281 | struct list_head *ret) | |
290408d4 | 1282 | { |
4eecb8b9 | 1283 | struct folio *dst, *src = page_folio(hpage); |
2def7424 | 1284 | int rc = -EAGAIN; |
2ebba6b7 | 1285 | int page_was_mapped = 0; |
32665f2b | 1286 | struct page *new_hpage; |
290408d4 | 1287 | struct anon_vma *anon_vma = NULL; |
c0d0381a | 1288 | struct address_space *mapping = NULL; |
290408d4 | 1289 | |
83467efb | 1290 | /* |
7ed2c31d | 1291 | * Migratability of hugepages depends on architectures and their size. |
83467efb NH |
1292 | * This check is necessary because some callers of hugepage migration |
1293 | * like soft offline and memory hotremove don't walk through page | |
1294 | * tables or check whether the hugepage is pmd-based or not before | |
1295 | * kicking migration. | |
1296 | */ | |
577be05c | 1297 | if (!hugepage_migration_supported(page_hstate(hpage))) |
83467efb NH |
1298 | return -ENOSYS; |
1299 | ||
c33db292 | 1300 | if (folio_ref_count(src) == 1) { |
71a64f61 | 1301 | /* page was freed from under us. So we are done. */ |
ea8e72f4 | 1302 | folio_putback_active_hugetlb(src); |
71a64f61 MS |
1303 | return MIGRATEPAGE_SUCCESS; |
1304 | } | |
1305 | ||
666feb21 | 1306 | new_hpage = get_new_page(hpage, private); |
290408d4 NH |
1307 | if (!new_hpage) |
1308 | return -ENOMEM; | |
4eecb8b9 | 1309 | dst = page_folio(new_hpage); |
290408d4 | 1310 | |
c33db292 | 1311 | if (!folio_trylock(src)) { |
2916ecc0 | 1312 | if (!force) |
290408d4 | 1313 | goto out; |
2916ecc0 JG |
1314 | switch (mode) { |
1315 | case MIGRATE_SYNC: | |
1316 | case MIGRATE_SYNC_NO_COPY: | |
1317 | break; | |
1318 | default: | |
1319 | goto out; | |
1320 | } | |
c33db292 | 1321 | folio_lock(src); |
290408d4 NH |
1322 | } |
1323 | ||
cb6acd01 MK |
1324 | /* |
1325 | * Check for pages which are in the process of being freed. Without | |
c33db292 | 1326 | * folio_mapping() set, hugetlbfs specific move page routine will not |
cb6acd01 MK |
1327 | * be called and we could leak usage counts for subpools. |
1328 | */ | |
345c62d1 | 1329 | if (hugetlb_folio_subpool(src) && !folio_mapping(src)) { |
cb6acd01 MK |
1330 | rc = -EBUSY; |
1331 | goto out_unlock; | |
1332 | } | |
1333 | ||
c33db292 | 1334 | if (folio_test_anon(src)) |
29eea9b5 | 1335 | anon_vma = folio_get_anon_vma(src); |
290408d4 | 1336 | |
c33db292 | 1337 | if (unlikely(!folio_trylock(dst))) |
7db7671f HD |
1338 | goto put_anon; |
1339 | ||
c33db292 | 1340 | if (folio_mapped(src)) { |
a98a2f0c | 1341 | enum ttu_flags ttu = 0; |
336bf30e | 1342 | |
c33db292 | 1343 | if (!folio_test_anon(src)) { |
336bf30e MK |
1344 | /* |
1345 | * In shared mappings, try_to_unmap could potentially | |
1346 | * call huge_pmd_unshare. Because of this, take | |
1347 | * semaphore in write mode here and set TTU_RMAP_LOCKED | |
1348 | * to let lower levels know we have taken the lock. | |
1349 | */ | |
1350 | mapping = hugetlb_page_mapping_lock_write(hpage); | |
1351 | if (unlikely(!mapping)) | |
1352 | goto unlock_put_anon; | |
1353 | ||
5202978b | 1354 | ttu = TTU_RMAP_LOCKED; |
336bf30e | 1355 | } |
c0d0381a | 1356 | |
4b8554c5 | 1357 | try_to_migrate(src, ttu); |
2ebba6b7 | 1358 | page_was_mapped = 1; |
336bf30e | 1359 | |
5202978b | 1360 | if (ttu & TTU_RMAP_LOCKED) |
336bf30e | 1361 | i_mmap_unlock_write(mapping); |
2ebba6b7 | 1362 | } |
290408d4 | 1363 | |
c33db292 | 1364 | if (!folio_mapped(src)) |
e7e3ffeb | 1365 | rc = move_to_new_folio(dst, src, mode); |
290408d4 | 1366 | |
336bf30e | 1367 | if (page_was_mapped) |
4eecb8b9 MWO |
1368 | remove_migration_ptes(src, |
1369 | rc == MIGRATEPAGE_SUCCESS ? dst : src, false); | |
290408d4 | 1370 | |
c0d0381a | 1371 | unlock_put_anon: |
c33db292 | 1372 | folio_unlock(dst); |
7db7671f HD |
1373 | |
1374 | put_anon: | |
fd4a4663 | 1375 | if (anon_vma) |
9e60109f | 1376 | put_anon_vma(anon_vma); |
8e6ac7fa | 1377 | |
2def7424 | 1378 | if (rc == MIGRATEPAGE_SUCCESS) { |
345c62d1 | 1379 | move_hugetlb_state(src, dst, reason); |
2def7424 HD |
1380 | put_new_page = NULL; |
1381 | } | |
8e6ac7fa | 1382 | |
cb6acd01 | 1383 | out_unlock: |
c33db292 | 1384 | folio_unlock(src); |
09761333 | 1385 | out: |
dd4ae78a | 1386 | if (rc == MIGRATEPAGE_SUCCESS) |
ea8e72f4 | 1387 | folio_putback_active_hugetlb(src); |
a04840c6 | 1388 | else if (rc != -EAGAIN) |
c33db292 | 1389 | list_move_tail(&src->lru, ret); |
68711a74 DR |
1390 | |
1391 | /* | |
1392 | * If migration was not successful and there's a freeing callback, use | |
1393 | * it. Otherwise, put_page() will drop the reference grabbed during | |
1394 | * isolation. | |
1395 | */ | |
2def7424 | 1396 | if (put_new_page) |
68711a74 DR |
1397 | put_new_page(new_hpage, private); |
1398 | else | |
ea8e72f4 | 1399 | folio_putback_active_hugetlb(dst); |
68711a74 | 1400 | |
290408d4 NH |
1401 | return rc; |
1402 | } | |
1403 | ||
eaec4e63 | 1404 | static inline int try_split_folio(struct folio *folio, struct list_head *split_folios) |
d532e2e5 | 1405 | { |
9c62ff00 | 1406 | int rc; |
d532e2e5 | 1407 | |
eaec4e63 HY |
1408 | folio_lock(folio); |
1409 | rc = split_folio_to_list(folio, split_folios); | |
1410 | folio_unlock(folio); | |
e6fa8a79 | 1411 | if (!rc) |
eaec4e63 | 1412 | list_move_tail(&folio->lru, split_folios); |
d532e2e5 YS |
1413 | |
1414 | return rc; | |
1415 | } | |
1416 | ||
e5bfff8b HY |
1417 | #define NR_MAX_MIGRATE_PAGES_RETRY 10 |
1418 | ||
5b855937 HY |
1419 | struct migrate_pages_stats { |
1420 | int nr_succeeded; /* Normal and large folios migrated successfully, in | |
1421 | units of base pages */ | |
1422 | int nr_failed_pages; /* Normal and large folios failed to be migrated, in | |
1423 | units of base pages. Untried folios aren't counted */ | |
1424 | int nr_thp_succeeded; /* THP migrated successfully */ | |
1425 | int nr_thp_failed; /* THP failed to be migrated */ | |
1426 | int nr_thp_split; /* THP split before migrating */ | |
1427 | }; | |
1428 | ||
e5bfff8b HY |
1429 | /* |
1430 | * Returns the number of hugetlb folios that were not migrated, or an error code | |
1431 | * after NR_MAX_MIGRATE_PAGES_RETRY attempts or if no hugetlb folios are movable | |
1432 | * any more because the list has become empty or no retryable hugetlb folios | |
1433 | * exist any more. It is caller's responsibility to call putback_movable_pages() | |
1434 | * only if ret != 0. | |
1435 | */ | |
1436 | static int migrate_hugetlbs(struct list_head *from, new_page_t get_new_page, | |
1437 | free_page_t put_new_page, unsigned long private, | |
1438 | enum migrate_mode mode, int reason, | |
1439 | struct migrate_pages_stats *stats, | |
1440 | struct list_head *ret_folios) | |
1441 | { | |
1442 | int retry = 1; | |
1443 | int nr_failed = 0; | |
1444 | int nr_retry_pages = 0; | |
1445 | int pass = 0; | |
1446 | struct folio *folio, *folio2; | |
1447 | int rc, nr_pages; | |
1448 | ||
1449 | for (pass = 0; pass < NR_MAX_MIGRATE_PAGES_RETRY && retry; pass++) { | |
1450 | retry = 0; | |
1451 | nr_retry_pages = 0; | |
1452 | ||
1453 | list_for_each_entry_safe(folio, folio2, from, lru) { | |
1454 | if (!folio_test_hugetlb(folio)) | |
1455 | continue; | |
1456 | ||
1457 | nr_pages = folio_nr_pages(folio); | |
1458 | ||
1459 | cond_resched(); | |
1460 | ||
1461 | rc = unmap_and_move_huge_page(get_new_page, | |
1462 | put_new_page, private, | |
1463 | &folio->page, pass > 2, mode, | |
1464 | reason, ret_folios); | |
1465 | /* | |
1466 | * The rules are: | |
1467 | * Success: hugetlb folio will be put back | |
1468 | * -EAGAIN: stay on the from list | |
1469 | * -ENOMEM: stay on the from list | |
1470 | * -ENOSYS: stay on the from list | |
1471 | * Other errno: put on ret_folios list | |
1472 | */ | |
1473 | switch(rc) { | |
1474 | case -ENOSYS: | |
1475 | /* Hugetlb migration is unsupported */ | |
1476 | nr_failed++; | |
1477 | stats->nr_failed_pages += nr_pages; | |
1478 | list_move_tail(&folio->lru, ret_folios); | |
1479 | break; | |
1480 | case -ENOMEM: | |
1481 | /* | |
1482 | * When memory is low, don't bother to try to migrate | |
1483 | * other folios, just exit. | |
1484 | */ | |
1485 | stats->nr_failed_pages += nr_pages + nr_retry_pages; | |
1486 | return -ENOMEM; | |
1487 | case -EAGAIN: | |
1488 | retry++; | |
1489 | nr_retry_pages += nr_pages; | |
1490 | break; | |
1491 | case MIGRATEPAGE_SUCCESS: | |
1492 | stats->nr_succeeded += nr_pages; | |
1493 | break; | |
1494 | default: | |
1495 | /* | |
1496 | * Permanent failure (-EBUSY, etc.): | |
1497 | * unlike -EAGAIN case, the failed folio is | |
1498 | * removed from migration folio list and not | |
1499 | * retried in the next outer loop. | |
1500 | */ | |
1501 | nr_failed++; | |
1502 | stats->nr_failed_pages += nr_pages; | |
1503 | break; | |
1504 | } | |
1505 | } | |
1506 | } | |
1507 | /* | |
1508 | * nr_failed is number of hugetlb folios failed to be migrated. After | |
1509 | * NR_MAX_MIGRATE_PAGES_RETRY attempts, give up and count retried hugetlb | |
1510 | * folios as failed. | |
1511 | */ | |
1512 | nr_failed += retry; | |
1513 | stats->nr_failed_pages += nr_retry_pages; | |
1514 | ||
1515 | return nr_failed; | |
1516 | } | |
1517 | ||
b20a3503 | 1518 | /* |
eaec4e63 | 1519 | * migrate_pages - migrate the folios specified in a list, to the free folios |
c73e5c9c | 1520 | * supplied as the target for the page migration |
b20a3503 | 1521 | * |
eaec4e63 HY |
1522 | * @from: The list of folios to be migrated. |
1523 | * @get_new_page: The function used to allocate free folios to be used | |
1524 | * as the target of the folio migration. | |
1525 | * @put_new_page: The function used to free target folios if migration | |
68711a74 | 1526 | * fails, or NULL if no special handling is necessary. |
c73e5c9c SB |
1527 | * @private: Private data to be passed on to get_new_page() |
1528 | * @mode: The migration mode that specifies the constraints for | |
eaec4e63 HY |
1529 | * folio migration, if any. |
1530 | * @reason: The reason for folio migration. | |
1531 | * @ret_succeeded: Set to the number of folios migrated successfully if | |
5ac95884 | 1532 | * the caller passes a non-NULL pointer. |
b20a3503 | 1533 | * |
e5bfff8b HY |
1534 | * The function returns after NR_MAX_MIGRATE_PAGES_RETRY attempts or if no folios |
1535 | * are movable any more because the list has become empty or no retryable folios | |
1536 | * exist any more. It is caller's responsibility to call putback_movable_pages() | |
1537 | * only if ret != 0. | |
b20a3503 | 1538 | * |
eaec4e63 HY |
1539 | * Returns the number of {normal folio, large folio, hugetlb} that were not |
1540 | * migrated, or an error code. The number of large folio splits will be | |
1541 | * considered as the number of non-migrated large folio, no matter how many | |
1542 | * split folios of the large folio are migrated successfully. | |
b20a3503 | 1543 | */ |
9c620e2b | 1544 | int migrate_pages(struct list_head *from, new_page_t get_new_page, |
68711a74 | 1545 | free_page_t put_new_page, unsigned long private, |
5ac95884 | 1546 | enum migrate_mode mode, int reason, unsigned int *ret_succeeded) |
b20a3503 | 1547 | { |
e24f0b8f | 1548 | int retry = 1; |
eaec4e63 | 1549 | int large_retry = 1; |
1a5bae25 | 1550 | int thp_retry = 1; |
e5bfff8b | 1551 | int nr_failed; |
077309bc | 1552 | int nr_retry_pages = 0; |
eaec4e63 | 1553 | int nr_large_failed = 0; |
b20a3503 | 1554 | int pass = 0; |
eaec4e63 | 1555 | bool is_large = false; |
1a5bae25 | 1556 | bool is_thp = false; |
eaec4e63 HY |
1557 | struct folio *folio, *folio2; |
1558 | int rc, nr_pages; | |
1559 | LIST_HEAD(ret_folios); | |
1560 | LIST_HEAD(split_folios); | |
b0b515bf | 1561 | bool nosplit = (reason == MR_NUMA_MISPLACED); |
eaec4e63 | 1562 | bool no_split_folio_counting = false; |
5b855937 | 1563 | struct migrate_pages_stats stats; |
b20a3503 | 1564 | |
7bc1aec5 LM |
1565 | trace_mm_migrate_pages_start(mode, reason); |
1566 | ||
5b855937 | 1567 | memset(&stats, 0, sizeof(stats)); |
e5bfff8b HY |
1568 | rc = migrate_hugetlbs(from, get_new_page, put_new_page, private, mode, reason, |
1569 | &stats, &ret_folios); | |
1570 | if (rc < 0) | |
1571 | goto out; | |
1572 | nr_failed = rc; | |
1573 | ||
eaec4e63 | 1574 | split_folio_migration: |
e5bfff8b HY |
1575 | for (pass = 0; |
1576 | pass < NR_MAX_MIGRATE_PAGES_RETRY && (retry || large_retry); | |
1577 | pass++) { | |
e24f0b8f | 1578 | retry = 0; |
eaec4e63 | 1579 | large_retry = 0; |
1a5bae25 | 1580 | thp_retry = 0; |
077309bc | 1581 | nr_retry_pages = 0; |
b20a3503 | 1582 | |
eaec4e63 | 1583 | list_for_each_entry_safe(folio, folio2, from, lru) { |
e5bfff8b HY |
1584 | /* Retried hugetlb folios will be kept in list */ |
1585 | if (folio_test_hugetlb(folio)) { | |
1586 | list_move_tail(&folio->lru, &ret_folios); | |
1587 | continue; | |
1588 | } | |
1589 | ||
1a5bae25 | 1590 | /* |
eaec4e63 HY |
1591 | * Large folio statistics is based on the source large |
1592 | * folio. Capture required information that might get | |
1593 | * lost during migration. | |
1a5bae25 | 1594 | */ |
e5bfff8b | 1595 | is_large = folio_test_large(folio); |
eaec4e63 HY |
1596 | is_thp = is_large && folio_test_pmd_mappable(folio); |
1597 | nr_pages = folio_nr_pages(folio); | |
e5bfff8b | 1598 | |
e24f0b8f | 1599 | cond_resched(); |
2d1db3b1 | 1600 | |
e5bfff8b HY |
1601 | rc = unmap_and_move(get_new_page, put_new_page, |
1602 | private, folio, pass > 2, mode, | |
1603 | reason, &ret_folios); | |
dd4ae78a YS |
1604 | /* |
1605 | * The rules are: | |
e5bfff8b | 1606 | * Success: folio will be freed |
dd4ae78a YS |
1607 | * -EAGAIN: stay on the from list |
1608 | * -ENOMEM: stay on the from list | |
577be05c | 1609 | * -ENOSYS: stay on the from list |
eaec4e63 | 1610 | * Other errno: put on ret_folios list then splice to |
dd4ae78a YS |
1611 | * from list |
1612 | */ | |
e24f0b8f | 1613 | switch(rc) { |
d532e2e5 | 1614 | /* |
eaec4e63 HY |
1615 | * Large folio migration might be unsupported or |
1616 | * the allocation could've failed so we should retry | |
1617 | * on the same folio with the large folio split | |
1618 | * to normal folios. | |
d532e2e5 | 1619 | * |
eaec4e63 | 1620 | * Split folios are put in split_folios, and |
e6fa8a79 HY |
1621 | * we will migrate them after the rest of the |
1622 | * list is processed. | |
d532e2e5 YS |
1623 | */ |
1624 | case -ENOSYS: | |
eaec4e63 HY |
1625 | /* Large folio migration is unsupported */ |
1626 | if (is_large) { | |
1627 | nr_large_failed++; | |
5b855937 | 1628 | stats.nr_thp_failed += is_thp; |
eaec4e63 | 1629 | if (!try_split_folio(folio, &split_folios)) { |
5b855937 | 1630 | stats.nr_thp_split += is_thp; |
e6fa8a79 | 1631 | break; |
d532e2e5 | 1632 | } |
eaec4e63 | 1633 | } else if (!no_split_folio_counting) { |
b5bade97 | 1634 | nr_failed++; |
f430893b ML |
1635 | } |
1636 | ||
5b855937 | 1637 | stats.nr_failed_pages += nr_pages; |
eaec4e63 | 1638 | list_move_tail(&folio->lru, &ret_folios); |
d532e2e5 | 1639 | break; |
95a402c3 | 1640 | case -ENOMEM: |
94723aaf | 1641 | /* |
d532e2e5 | 1642 | * When memory is low, don't bother to try to migrate |
eaec4e63 | 1643 | * other folios, just exit. |
94723aaf | 1644 | */ |
eaec4e63 HY |
1645 | if (is_large) { |
1646 | nr_large_failed++; | |
5b855937 | 1647 | stats.nr_thp_failed += is_thp; |
eaec4e63 | 1648 | /* Large folio NUMA faulting doesn't split to retry. */ |
fd4a7ac3 | 1649 | if (!nosplit) { |
eaec4e63 | 1650 | int ret = try_split_folio(folio, &split_folios); |
fd4a7ac3 BW |
1651 | |
1652 | if (!ret) { | |
5b855937 | 1653 | stats.nr_thp_split += is_thp; |
fd4a7ac3 BW |
1654 | break; |
1655 | } else if (reason == MR_LONGTERM_PIN && | |
1656 | ret == -EAGAIN) { | |
1657 | /* | |
eaec4e63 HY |
1658 | * Try again to split large folio to |
1659 | * mitigate the failure of longterm pinning. | |
fd4a7ac3 | 1660 | */ |
eaec4e63 HY |
1661 | large_retry++; |
1662 | thp_retry += is_thp; | |
1663 | nr_retry_pages += nr_pages; | |
fd4a7ac3 BW |
1664 | break; |
1665 | } | |
94723aaf | 1666 | } |
eaec4e63 | 1667 | } else if (!no_split_folio_counting) { |
f430893b | 1668 | nr_failed++; |
1a5bae25 | 1669 | } |
b5bade97 | 1670 | |
5b855937 | 1671 | stats.nr_failed_pages += nr_pages + nr_retry_pages; |
69a041ff | 1672 | /* |
eaec4e63 HY |
1673 | * There might be some split folios of fail-to-migrate large |
1674 | * folios left in split_folios list. Move them back to migration | |
69a041ff | 1675 | * list so that they could be put back to the right list by |
eaec4e63 | 1676 | * the caller otherwise the folio refcnt will be leaked. |
69a041ff | 1677 | */ |
eaec4e63 | 1678 | list_splice_init(&split_folios, from); |
fbed53b4 | 1679 | /* nr_failed isn't updated for not used */ |
eaec4e63 | 1680 | nr_large_failed += large_retry; |
5b855937 | 1681 | stats.nr_thp_failed += thp_retry; |
95a402c3 | 1682 | goto out; |
e24f0b8f | 1683 | case -EAGAIN: |
eaec4e63 HY |
1684 | if (is_large) { |
1685 | large_retry++; | |
1686 | thp_retry += is_thp; | |
1687 | } else if (!no_split_folio_counting) { | |
f430893b | 1688 | retry++; |
eaec4e63 HY |
1689 | } |
1690 | nr_retry_pages += nr_pages; | |
e24f0b8f | 1691 | break; |
78bd5209 | 1692 | case MIGRATEPAGE_SUCCESS: |
5b855937 HY |
1693 | stats.nr_succeeded += nr_pages; |
1694 | stats.nr_thp_succeeded += is_thp; | |
e24f0b8f CL |
1695 | break; |
1696 | default: | |
354a3363 | 1697 | /* |
d532e2e5 | 1698 | * Permanent failure (-EBUSY, etc.): |
eaec4e63 HY |
1699 | * unlike -EAGAIN case, the failed folio is |
1700 | * removed from migration folio list and not | |
354a3363 NH |
1701 | * retried in the next outer loop. |
1702 | */ | |
eaec4e63 HY |
1703 | if (is_large) { |
1704 | nr_large_failed++; | |
5b855937 | 1705 | stats.nr_thp_failed += is_thp; |
eaec4e63 | 1706 | } else if (!no_split_folio_counting) { |
b5bade97 | 1707 | nr_failed++; |
eaec4e63 | 1708 | } |
f430893b | 1709 | |
5b855937 | 1710 | stats.nr_failed_pages += nr_pages; |
e24f0b8f | 1711 | break; |
2d1db3b1 | 1712 | } |
b20a3503 CL |
1713 | } |
1714 | } | |
7047b5a4 | 1715 | nr_failed += retry; |
eaec4e63 | 1716 | nr_large_failed += large_retry; |
5b855937 HY |
1717 | stats.nr_thp_failed += thp_retry; |
1718 | stats.nr_failed_pages += nr_retry_pages; | |
b5bade97 | 1719 | /* |
eaec4e63 HY |
1720 | * Try to migrate split folios of fail-to-migrate large folios, no |
1721 | * nr_failed counting in this round, since all split folios of a | |
1722 | * large folio is counted as 1 failure in the first round. | |
b5bade97 | 1723 | */ |
eaec4e63 | 1724 | if (!list_empty(&split_folios)) { |
b5bade97 | 1725 | /* |
e5bfff8b HY |
1726 | * Move non-migrated folios (after NR_MAX_MIGRATE_PAGES_RETRY |
1727 | * retries) to ret_folios to avoid migrating them again. | |
b5bade97 | 1728 | */ |
eaec4e63 HY |
1729 | list_splice_init(from, &ret_folios); |
1730 | list_splice_init(&split_folios, from); | |
1731 | no_split_folio_counting = true; | |
b5bade97 | 1732 | retry = 1; |
eaec4e63 | 1733 | goto split_folio_migration; |
b5bade97 BW |
1734 | } |
1735 | ||
eaec4e63 | 1736 | rc = nr_failed + nr_large_failed; |
95a402c3 | 1737 | out: |
dd4ae78a | 1738 | /* |
eaec4e63 | 1739 | * Put the permanent failure folio back to migration list, they |
dd4ae78a YS |
1740 | * will be put back to the right list by the caller. |
1741 | */ | |
eaec4e63 | 1742 | list_splice(&ret_folios, from); |
dd4ae78a | 1743 | |
03e5f82e | 1744 | /* |
eaec4e63 HY |
1745 | * Return 0 in case all split folios of fail-to-migrate large folios |
1746 | * are migrated successfully. | |
03e5f82e BW |
1747 | */ |
1748 | if (list_empty(from)) | |
1749 | rc = 0; | |
1750 | ||
5b855937 HY |
1751 | count_vm_events(PGMIGRATE_SUCCESS, stats.nr_succeeded); |
1752 | count_vm_events(PGMIGRATE_FAIL, stats.nr_failed_pages); | |
1753 | count_vm_events(THP_MIGRATION_SUCCESS, stats.nr_thp_succeeded); | |
1754 | count_vm_events(THP_MIGRATION_FAIL, stats.nr_thp_failed); | |
1755 | count_vm_events(THP_MIGRATION_SPLIT, stats.nr_thp_split); | |
1756 | trace_mm_migrate_pages(stats.nr_succeeded, stats.nr_failed_pages, | |
1757 | stats.nr_thp_succeeded, stats.nr_thp_failed, | |
1758 | stats.nr_thp_split, mode, reason); | |
7b2a2d4a | 1759 | |
5ac95884 | 1760 | if (ret_succeeded) |
5b855937 | 1761 | *ret_succeeded = stats.nr_succeeded; |
5ac95884 | 1762 | |
78bd5209 | 1763 | return rc; |
b20a3503 | 1764 | } |
95a402c3 | 1765 | |
19fc7bed | 1766 | struct page *alloc_migration_target(struct page *page, unsigned long private) |
b4b38223 | 1767 | { |
ffe06786 | 1768 | struct folio *folio = page_folio(page); |
19fc7bed JK |
1769 | struct migration_target_control *mtc; |
1770 | gfp_t gfp_mask; | |
b4b38223 | 1771 | unsigned int order = 0; |
e37d3e83 | 1772 | struct folio *hugetlb_folio = NULL; |
ffe06786 | 1773 | struct folio *new_folio = NULL; |
19fc7bed JK |
1774 | int nid; |
1775 | int zidx; | |
1776 | ||
1777 | mtc = (struct migration_target_control *)private; | |
1778 | gfp_mask = mtc->gfp_mask; | |
1779 | nid = mtc->nid; | |
1780 | if (nid == NUMA_NO_NODE) | |
ffe06786 | 1781 | nid = folio_nid(folio); |
b4b38223 | 1782 | |
ffe06786 | 1783 | if (folio_test_hugetlb(folio)) { |
e51da3a9 | 1784 | struct hstate *h = folio_hstate(folio); |
d92bbc27 | 1785 | |
19fc7bed | 1786 | gfp_mask = htlb_modify_alloc_mask(h, gfp_mask); |
e37d3e83 SK |
1787 | hugetlb_folio = alloc_hugetlb_folio_nodemask(h, nid, |
1788 | mtc->nmask, gfp_mask); | |
1789 | return &hugetlb_folio->page; | |
d92bbc27 | 1790 | } |
b4b38223 | 1791 | |
ffe06786 | 1792 | if (folio_test_large(folio)) { |
9933a0c8 JK |
1793 | /* |
1794 | * clear __GFP_RECLAIM to make the migration callback | |
1795 | * consistent with regular THP allocations. | |
1796 | */ | |
1797 | gfp_mask &= ~__GFP_RECLAIM; | |
b4b38223 | 1798 | gfp_mask |= GFP_TRANSHUGE; |
ffe06786 | 1799 | order = folio_order(folio); |
b4b38223 | 1800 | } |
ffe06786 | 1801 | zidx = zone_idx(folio_zone(folio)); |
19fc7bed | 1802 | if (is_highmem_idx(zidx) || zidx == ZONE_MOVABLE) |
b4b38223 JK |
1803 | gfp_mask |= __GFP_HIGHMEM; |
1804 | ||
ffe06786 | 1805 | new_folio = __folio_alloc(gfp_mask, order, nid, mtc->nmask); |
b4b38223 | 1806 | |
ffe06786 | 1807 | return &new_folio->page; |
b4b38223 JK |
1808 | } |
1809 | ||
742755a1 | 1810 | #ifdef CONFIG_NUMA |
742755a1 | 1811 | |
a49bd4d7 | 1812 | static int store_status(int __user *status, int start, int value, int nr) |
742755a1 | 1813 | { |
a49bd4d7 MH |
1814 | while (nr-- > 0) { |
1815 | if (put_user(value, status + start)) | |
1816 | return -EFAULT; | |
1817 | start++; | |
1818 | } | |
1819 | ||
1820 | return 0; | |
1821 | } | |
1822 | ||
1823 | static int do_move_pages_to_node(struct mm_struct *mm, | |
1824 | struct list_head *pagelist, int node) | |
1825 | { | |
1826 | int err; | |
a0976311 JK |
1827 | struct migration_target_control mtc = { |
1828 | .nid = node, | |
1829 | .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, | |
1830 | }; | |
a49bd4d7 | 1831 | |
a0976311 | 1832 | err = migrate_pages(pagelist, alloc_migration_target, NULL, |
5ac95884 | 1833 | (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL); |
a49bd4d7 MH |
1834 | if (err) |
1835 | putback_movable_pages(pagelist); | |
1836 | return err; | |
742755a1 CL |
1837 | } |
1838 | ||
1839 | /* | |
a49bd4d7 MH |
1840 | * Resolves the given address to a struct page, isolates it from the LRU and |
1841 | * puts it to the given pagelist. | |
e0153fc2 YS |
1842 | * Returns: |
1843 | * errno - if the page cannot be found/isolated | |
1844 | * 0 - when it doesn't have to be migrated because it is already on the | |
1845 | * target node | |
1846 | * 1 - when it has been queued | |
742755a1 | 1847 | */ |
a49bd4d7 MH |
1848 | static int add_page_for_migration(struct mm_struct *mm, unsigned long addr, |
1849 | int node, struct list_head *pagelist, bool migrate_all) | |
742755a1 | 1850 | { |
a49bd4d7 MH |
1851 | struct vm_area_struct *vma; |
1852 | struct page *page; | |
742755a1 | 1853 | int err; |
742755a1 | 1854 | |
d8ed45c5 | 1855 | mmap_read_lock(mm); |
a49bd4d7 | 1856 | err = -EFAULT; |
cb1c37b1 ML |
1857 | vma = vma_lookup(mm, addr); |
1858 | if (!vma || !vma_migratable(vma)) | |
a49bd4d7 | 1859 | goto out; |
742755a1 | 1860 | |
a49bd4d7 | 1861 | /* FOLL_DUMP to ignore special (like zero) pages */ |
87d2762e | 1862 | page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP); |
89f5b7da | 1863 | |
a49bd4d7 MH |
1864 | err = PTR_ERR(page); |
1865 | if (IS_ERR(page)) | |
1866 | goto out; | |
89f5b7da | 1867 | |
a49bd4d7 | 1868 | err = -ENOENT; |
f7091ed6 | 1869 | if (!page) |
a49bd4d7 | 1870 | goto out; |
742755a1 | 1871 | |
f7091ed6 HW |
1872 | if (is_zone_device_page(page)) |
1873 | goto out_putpage; | |
1874 | ||
a49bd4d7 MH |
1875 | err = 0; |
1876 | if (page_to_nid(page) == node) | |
1877 | goto out_putpage; | |
742755a1 | 1878 | |
a49bd4d7 MH |
1879 | err = -EACCES; |
1880 | if (page_mapcount(page) > 1 && !migrate_all) | |
1881 | goto out_putpage; | |
742755a1 | 1882 | |
a49bd4d7 MH |
1883 | if (PageHuge(page)) { |
1884 | if (PageHead(page)) { | |
6aa3a920 | 1885 | err = isolate_hugetlb(page_folio(page), pagelist); |
7ce82f4c ML |
1886 | if (!err) |
1887 | err = 1; | |
e632a938 | 1888 | } |
a49bd4d7 MH |
1889 | } else { |
1890 | struct page *head; | |
e632a938 | 1891 | |
e8db67eb NH |
1892 | head = compound_head(page); |
1893 | err = isolate_lru_page(head); | |
cf608ac1 | 1894 | if (err) |
a49bd4d7 | 1895 | goto out_putpage; |
742755a1 | 1896 | |
e0153fc2 | 1897 | err = 1; |
a49bd4d7 MH |
1898 | list_add_tail(&head->lru, pagelist); |
1899 | mod_node_page_state(page_pgdat(head), | |
9de4f22a | 1900 | NR_ISOLATED_ANON + page_is_file_lru(head), |
6c357848 | 1901 | thp_nr_pages(head)); |
a49bd4d7 MH |
1902 | } |
1903 | out_putpage: | |
1904 | /* | |
1905 | * Either remove the duplicate refcount from | |
1906 | * isolate_lru_page() or drop the page ref if it was | |
1907 | * not isolated. | |
1908 | */ | |
1909 | put_page(page); | |
1910 | out: | |
d8ed45c5 | 1911 | mmap_read_unlock(mm); |
742755a1 CL |
1912 | return err; |
1913 | } | |
1914 | ||
7ca8783a WY |
1915 | static int move_pages_and_store_status(struct mm_struct *mm, int node, |
1916 | struct list_head *pagelist, int __user *status, | |
1917 | int start, int i, unsigned long nr_pages) | |
1918 | { | |
1919 | int err; | |
1920 | ||
5d7ae891 WY |
1921 | if (list_empty(pagelist)) |
1922 | return 0; | |
1923 | ||
7ca8783a WY |
1924 | err = do_move_pages_to_node(mm, pagelist, node); |
1925 | if (err) { | |
1926 | /* | |
1927 | * Positive err means the number of failed | |
1928 | * pages to migrate. Since we are going to | |
1929 | * abort and return the number of non-migrated | |
ab9dd4f8 | 1930 | * pages, so need to include the rest of the |
7ca8783a WY |
1931 | * nr_pages that have not been attempted as |
1932 | * well. | |
1933 | */ | |
1934 | if (err > 0) | |
a7504ed1 | 1935 | err += nr_pages - i; |
7ca8783a WY |
1936 | return err; |
1937 | } | |
1938 | return store_status(status, start, node, i - start); | |
1939 | } | |
1940 | ||
5e9a0f02 BG |
1941 | /* |
1942 | * Migrate an array of page address onto an array of nodes and fill | |
1943 | * the corresponding array of status. | |
1944 | */ | |
3268c63e | 1945 | static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, |
5e9a0f02 BG |
1946 | unsigned long nr_pages, |
1947 | const void __user * __user *pages, | |
1948 | const int __user *nodes, | |
1949 | int __user *status, int flags) | |
1950 | { | |
a49bd4d7 MH |
1951 | int current_node = NUMA_NO_NODE; |
1952 | LIST_HEAD(pagelist); | |
1953 | int start, i; | |
1954 | int err = 0, err1; | |
35282a2d | 1955 | |
361a2a22 | 1956 | lru_cache_disable(); |
35282a2d | 1957 | |
a49bd4d7 MH |
1958 | for (i = start = 0; i < nr_pages; i++) { |
1959 | const void __user *p; | |
1960 | unsigned long addr; | |
1961 | int node; | |
3140a227 | 1962 | |
a49bd4d7 MH |
1963 | err = -EFAULT; |
1964 | if (get_user(p, pages + i)) | |
1965 | goto out_flush; | |
1966 | if (get_user(node, nodes + i)) | |
1967 | goto out_flush; | |
057d3389 | 1968 | addr = (unsigned long)untagged_addr(p); |
a49bd4d7 MH |
1969 | |
1970 | err = -ENODEV; | |
1971 | if (node < 0 || node >= MAX_NUMNODES) | |
1972 | goto out_flush; | |
1973 | if (!node_state(node, N_MEMORY)) | |
1974 | goto out_flush; | |
5e9a0f02 | 1975 | |
a49bd4d7 MH |
1976 | err = -EACCES; |
1977 | if (!node_isset(node, task_nodes)) | |
1978 | goto out_flush; | |
1979 | ||
1980 | if (current_node == NUMA_NO_NODE) { | |
1981 | current_node = node; | |
1982 | start = i; | |
1983 | } else if (node != current_node) { | |
7ca8783a WY |
1984 | err = move_pages_and_store_status(mm, current_node, |
1985 | &pagelist, status, start, i, nr_pages); | |
a49bd4d7 MH |
1986 | if (err) |
1987 | goto out; | |
1988 | start = i; | |
1989 | current_node = node; | |
3140a227 BG |
1990 | } |
1991 | ||
a49bd4d7 MH |
1992 | /* |
1993 | * Errors in the page lookup or isolation are not fatal and we simply | |
1994 | * report them via status | |
1995 | */ | |
1996 | err = add_page_for_migration(mm, addr, current_node, | |
1997 | &pagelist, flags & MPOL_MF_MOVE_ALL); | |
e0153fc2 | 1998 | |
d08221a0 | 1999 | if (err > 0) { |
e0153fc2 YS |
2000 | /* The page is successfully queued for migration */ |
2001 | continue; | |
2002 | } | |
3140a227 | 2003 | |
65462462 JH |
2004 | /* |
2005 | * The move_pages() man page does not have an -EEXIST choice, so | |
2006 | * use -EFAULT instead. | |
2007 | */ | |
2008 | if (err == -EEXIST) | |
2009 | err = -EFAULT; | |
2010 | ||
d08221a0 WY |
2011 | /* |
2012 | * If the page is already on the target node (!err), store the | |
2013 | * node, otherwise, store the err. | |
2014 | */ | |
2015 | err = store_status(status, i, err ? : current_node, 1); | |
a49bd4d7 MH |
2016 | if (err) |
2017 | goto out_flush; | |
5e9a0f02 | 2018 | |
7ca8783a WY |
2019 | err = move_pages_and_store_status(mm, current_node, &pagelist, |
2020 | status, start, i, nr_pages); | |
a7504ed1 HY |
2021 | if (err) { |
2022 | /* We have accounted for page i */ | |
2023 | if (err > 0) | |
2024 | err--; | |
4afdacec | 2025 | goto out; |
a7504ed1 | 2026 | } |
a49bd4d7 | 2027 | current_node = NUMA_NO_NODE; |
3140a227 | 2028 | } |
a49bd4d7 MH |
2029 | out_flush: |
2030 | /* Make sure we do not overwrite the existing error */ | |
7ca8783a WY |
2031 | err1 = move_pages_and_store_status(mm, current_node, &pagelist, |
2032 | status, start, i, nr_pages); | |
dfe9aa23 | 2033 | if (err >= 0) |
a49bd4d7 | 2034 | err = err1; |
5e9a0f02 | 2035 | out: |
361a2a22 | 2036 | lru_cache_enable(); |
5e9a0f02 BG |
2037 | return err; |
2038 | } | |
2039 | ||
742755a1 | 2040 | /* |
2f007e74 | 2041 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 2042 | */ |
80bba129 BG |
2043 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
2044 | const void __user **pages, int *status) | |
742755a1 | 2045 | { |
2f007e74 | 2046 | unsigned long i; |
2f007e74 | 2047 | |
d8ed45c5 | 2048 | mmap_read_lock(mm); |
742755a1 | 2049 | |
2f007e74 | 2050 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 2051 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
2052 | struct vm_area_struct *vma; |
2053 | struct page *page; | |
c095adbc | 2054 | int err = -EFAULT; |
2f007e74 | 2055 | |
059b8b48 LH |
2056 | vma = vma_lookup(mm, addr); |
2057 | if (!vma) | |
742755a1 CL |
2058 | goto set_status; |
2059 | ||
d899844e | 2060 | /* FOLL_DUMP to ignore special (like zero) pages */ |
16fd6b31 | 2061 | page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP); |
89f5b7da LT |
2062 | |
2063 | err = PTR_ERR(page); | |
2064 | if (IS_ERR(page)) | |
2065 | goto set_status; | |
2066 | ||
f7091ed6 HW |
2067 | err = -ENOENT; |
2068 | if (!page) | |
2069 | goto set_status; | |
2070 | ||
2071 | if (!is_zone_device_page(page)) | |
4cd61484 | 2072 | err = page_to_nid(page); |
f7091ed6 | 2073 | |
16fd6b31 | 2074 | put_page(page); |
742755a1 | 2075 | set_status: |
80bba129 BG |
2076 | *status = err; |
2077 | ||
2078 | pages++; | |
2079 | status++; | |
2080 | } | |
2081 | ||
d8ed45c5 | 2082 | mmap_read_unlock(mm); |
80bba129 BG |
2083 | } |
2084 | ||
5b1b561b AB |
2085 | static int get_compat_pages_array(const void __user *chunk_pages[], |
2086 | const void __user * __user *pages, | |
2087 | unsigned long chunk_nr) | |
2088 | { | |
2089 | compat_uptr_t __user *pages32 = (compat_uptr_t __user *)pages; | |
2090 | compat_uptr_t p; | |
2091 | int i; | |
2092 | ||
2093 | for (i = 0; i < chunk_nr; i++) { | |
2094 | if (get_user(p, pages32 + i)) | |
2095 | return -EFAULT; | |
2096 | chunk_pages[i] = compat_ptr(p); | |
2097 | } | |
2098 | ||
2099 | return 0; | |
2100 | } | |
2101 | ||
80bba129 BG |
2102 | /* |
2103 | * Determine the nodes of a user array of pages and store it in | |
2104 | * a user array of status. | |
2105 | */ | |
2106 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
2107 | const void __user * __user *pages, | |
2108 | int __user *status) | |
2109 | { | |
3eefb826 | 2110 | #define DO_PAGES_STAT_CHUNK_NR 16UL |
80bba129 BG |
2111 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; |
2112 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
80bba129 | 2113 | |
87b8d1ad | 2114 | while (nr_pages) { |
3eefb826 | 2115 | unsigned long chunk_nr = min(nr_pages, DO_PAGES_STAT_CHUNK_NR); |
87b8d1ad | 2116 | |
5b1b561b AB |
2117 | if (in_compat_syscall()) { |
2118 | if (get_compat_pages_array(chunk_pages, pages, | |
2119 | chunk_nr)) | |
2120 | break; | |
2121 | } else { | |
2122 | if (copy_from_user(chunk_pages, pages, | |
2123 | chunk_nr * sizeof(*chunk_pages))) | |
2124 | break; | |
2125 | } | |
80bba129 BG |
2126 | |
2127 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
2128 | ||
87b8d1ad PA |
2129 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
2130 | break; | |
742755a1 | 2131 | |
87b8d1ad PA |
2132 | pages += chunk_nr; |
2133 | status += chunk_nr; | |
2134 | nr_pages -= chunk_nr; | |
2135 | } | |
2136 | return nr_pages ? -EFAULT : 0; | |
742755a1 CL |
2137 | } |
2138 | ||
4dc200ce | 2139 | static struct mm_struct *find_mm_struct(pid_t pid, nodemask_t *mem_nodes) |
742755a1 | 2140 | { |
742755a1 | 2141 | struct task_struct *task; |
742755a1 | 2142 | struct mm_struct *mm; |
742755a1 | 2143 | |
4dc200ce ML |
2144 | /* |
2145 | * There is no need to check if current process has the right to modify | |
2146 | * the specified process when they are same. | |
2147 | */ | |
2148 | if (!pid) { | |
2149 | mmget(current->mm); | |
2150 | *mem_nodes = cpuset_mems_allowed(current); | |
2151 | return current->mm; | |
2152 | } | |
742755a1 CL |
2153 | |
2154 | /* Find the mm_struct */ | |
a879bf58 | 2155 | rcu_read_lock(); |
4dc200ce | 2156 | task = find_task_by_vpid(pid); |
742755a1 | 2157 | if (!task) { |
a879bf58 | 2158 | rcu_read_unlock(); |
4dc200ce | 2159 | return ERR_PTR(-ESRCH); |
742755a1 | 2160 | } |
3268c63e | 2161 | get_task_struct(task); |
742755a1 CL |
2162 | |
2163 | /* | |
2164 | * Check if this process has the right to modify the specified | |
197e7e52 | 2165 | * process. Use the regular "ptrace_may_access()" checks. |
742755a1 | 2166 | */ |
197e7e52 | 2167 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 2168 | rcu_read_unlock(); |
4dc200ce | 2169 | mm = ERR_PTR(-EPERM); |
5e9a0f02 | 2170 | goto out; |
742755a1 | 2171 | } |
c69e8d9c | 2172 | rcu_read_unlock(); |
742755a1 | 2173 | |
4dc200ce ML |
2174 | mm = ERR_PTR(security_task_movememory(task)); |
2175 | if (IS_ERR(mm)) | |
5e9a0f02 | 2176 | goto out; |
4dc200ce | 2177 | *mem_nodes = cpuset_mems_allowed(task); |
3268c63e | 2178 | mm = get_task_mm(task); |
4dc200ce | 2179 | out: |
3268c63e | 2180 | put_task_struct(task); |
6e8b09ea | 2181 | if (!mm) |
4dc200ce ML |
2182 | mm = ERR_PTR(-EINVAL); |
2183 | return mm; | |
2184 | } | |
2185 | ||
2186 | /* | |
2187 | * Move a list of pages in the address space of the currently executing | |
2188 | * process. | |
2189 | */ | |
2190 | static int kernel_move_pages(pid_t pid, unsigned long nr_pages, | |
2191 | const void __user * __user *pages, | |
2192 | const int __user *nodes, | |
2193 | int __user *status, int flags) | |
2194 | { | |
2195 | struct mm_struct *mm; | |
2196 | int err; | |
2197 | nodemask_t task_nodes; | |
2198 | ||
2199 | /* Check flags */ | |
2200 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
6e8b09ea SL |
2201 | return -EINVAL; |
2202 | ||
4dc200ce ML |
2203 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
2204 | return -EPERM; | |
2205 | ||
2206 | mm = find_mm_struct(pid, &task_nodes); | |
2207 | if (IS_ERR(mm)) | |
2208 | return PTR_ERR(mm); | |
2209 | ||
6e8b09ea SL |
2210 | if (nodes) |
2211 | err = do_pages_move(mm, task_nodes, nr_pages, pages, | |
2212 | nodes, status, flags); | |
2213 | else | |
2214 | err = do_pages_stat(mm, nr_pages, pages, status); | |
742755a1 | 2215 | |
742755a1 CL |
2216 | mmput(mm); |
2217 | return err; | |
2218 | } | |
742755a1 | 2219 | |
7addf443 DB |
2220 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
2221 | const void __user * __user *, pages, | |
2222 | const int __user *, nodes, | |
2223 | int __user *, status, int, flags) | |
2224 | { | |
2225 | return kernel_move_pages(pid, nr_pages, pages, nodes, status, flags); | |
2226 | } | |
2227 | ||
7039e1db PZ |
2228 | #ifdef CONFIG_NUMA_BALANCING |
2229 | /* | |
2230 | * Returns true if this is a safe migration target node for misplaced NUMA | |
bc53008e | 2231 | * pages. Currently it only checks the watermarks which is crude. |
7039e1db PZ |
2232 | */ |
2233 | static bool migrate_balanced_pgdat(struct pglist_data *pgdat, | |
3abef4e6 | 2234 | unsigned long nr_migrate_pages) |
7039e1db PZ |
2235 | { |
2236 | int z; | |
599d0c95 | 2237 | |
7039e1db PZ |
2238 | for (z = pgdat->nr_zones - 1; z >= 0; z--) { |
2239 | struct zone *zone = pgdat->node_zones + z; | |
2240 | ||
bc53008e | 2241 | if (!managed_zone(zone)) |
7039e1db PZ |
2242 | continue; |
2243 | ||
7039e1db PZ |
2244 | /* Avoid waking kswapd by allocating pages_to_migrate pages. */ |
2245 | if (!zone_watermark_ok(zone, 0, | |
2246 | high_wmark_pages(zone) + | |
2247 | nr_migrate_pages, | |
bfe9d006 | 2248 | ZONE_MOVABLE, 0)) |
7039e1db PZ |
2249 | continue; |
2250 | return true; | |
2251 | } | |
2252 | return false; | |
2253 | } | |
2254 | ||
2255 | static struct page *alloc_misplaced_dst_page(struct page *page, | |
666feb21 | 2256 | unsigned long data) |
7039e1db PZ |
2257 | { |
2258 | int nid = (int) data; | |
c185e494 MWO |
2259 | int order = compound_order(page); |
2260 | gfp_t gfp = __GFP_THISNODE; | |
2261 | struct folio *new; | |
2262 | ||
2263 | if (order > 0) | |
2264 | gfp |= GFP_TRANSHUGE_LIGHT; | |
2265 | else { | |
2266 | gfp |= GFP_HIGHUSER_MOVABLE | __GFP_NOMEMALLOC | __GFP_NORETRY | | |
2267 | __GFP_NOWARN; | |
2268 | gfp &= ~__GFP_RECLAIM; | |
2269 | } | |
2270 | new = __folio_alloc_node(gfp, order, nid); | |
c5b5a3dd | 2271 | |
c185e494 | 2272 | return &new->page; |
c5b5a3dd YS |
2273 | } |
2274 | ||
1c30e017 | 2275 | static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page) |
b32967ff | 2276 | { |
2b9b624f | 2277 | int nr_pages = thp_nr_pages(page); |
c574bbe9 | 2278 | int order = compound_order(page); |
a8f60772 | 2279 | |
c574bbe9 | 2280 | VM_BUG_ON_PAGE(order && !PageTransHuge(page), page); |
3abef4e6 | 2281 | |
662aeea7 YS |
2282 | /* Do not migrate THP mapped by multiple processes */ |
2283 | if (PageTransHuge(page) && total_mapcount(page) > 1) | |
2284 | return 0; | |
2285 | ||
7039e1db | 2286 | /* Avoid migrating to a node that is nearly full */ |
c574bbe9 HY |
2287 | if (!migrate_balanced_pgdat(pgdat, nr_pages)) { |
2288 | int z; | |
2289 | ||
2290 | if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING)) | |
2291 | return 0; | |
2292 | for (z = pgdat->nr_zones - 1; z >= 0; z--) { | |
bc53008e | 2293 | if (managed_zone(pgdat->node_zones + z)) |
c574bbe9 HY |
2294 | break; |
2295 | } | |
2296 | wakeup_kswapd(pgdat->node_zones + z, 0, order, ZONE_MOVABLE); | |
340ef390 | 2297 | return 0; |
c574bbe9 | 2298 | } |
7039e1db | 2299 | |
340ef390 HD |
2300 | if (isolate_lru_page(page)) |
2301 | return 0; | |
7039e1db | 2302 | |
b75454e1 | 2303 | mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + page_is_file_lru(page), |
2b9b624f | 2304 | nr_pages); |
340ef390 | 2305 | |
149c33e1 | 2306 | /* |
340ef390 HD |
2307 | * Isolating the page has taken another reference, so the |
2308 | * caller's reference can be safely dropped without the page | |
2309 | * disappearing underneath us during migration. | |
149c33e1 MG |
2310 | */ |
2311 | put_page(page); | |
340ef390 | 2312 | return 1; |
b32967ff MG |
2313 | } |
2314 | ||
2315 | /* | |
2316 | * Attempt to migrate a misplaced page to the specified destination | |
2317 | * node. Caller is expected to have an elevated reference count on | |
2318 | * the page that will be dropped by this function before returning. | |
2319 | */ | |
1bc115d8 MG |
2320 | int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma, |
2321 | int node) | |
b32967ff MG |
2322 | { |
2323 | pg_data_t *pgdat = NODE_DATA(node); | |
340ef390 | 2324 | int isolated; |
b32967ff | 2325 | int nr_remaining; |
e39bb6be | 2326 | unsigned int nr_succeeded; |
b32967ff | 2327 | LIST_HEAD(migratepages); |
b5916c02 | 2328 | int nr_pages = thp_nr_pages(page); |
c5b5a3dd | 2329 | |
b32967ff | 2330 | /* |
1bc115d8 MG |
2331 | * Don't migrate file pages that are mapped in multiple processes |
2332 | * with execute permissions as they are probably shared libraries. | |
b32967ff | 2333 | */ |
7ee820ee ML |
2334 | if (page_mapcount(page) != 1 && page_is_file_lru(page) && |
2335 | (vma->vm_flags & VM_EXEC)) | |
b32967ff | 2336 | goto out; |
b32967ff | 2337 | |
09a913a7 MG |
2338 | /* |
2339 | * Also do not migrate dirty pages as not all filesystems can move | |
2340 | * dirty pages in MIGRATE_ASYNC mode which is a waste of cycles. | |
2341 | */ | |
9de4f22a | 2342 | if (page_is_file_lru(page) && PageDirty(page)) |
09a913a7 MG |
2343 | goto out; |
2344 | ||
b32967ff MG |
2345 | isolated = numamigrate_isolate_page(pgdat, page); |
2346 | if (!isolated) | |
2347 | goto out; | |
2348 | ||
2349 | list_add(&page->lru, &migratepages); | |
c185e494 MWO |
2350 | nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page, |
2351 | NULL, node, MIGRATE_ASYNC, | |
2352 | MR_NUMA_MISPLACED, &nr_succeeded); | |
b32967ff | 2353 | if (nr_remaining) { |
59c82b70 JK |
2354 | if (!list_empty(&migratepages)) { |
2355 | list_del(&page->lru); | |
c5fc5c3a YS |
2356 | mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + |
2357 | page_is_file_lru(page), -nr_pages); | |
59c82b70 JK |
2358 | putback_lru_page(page); |
2359 | } | |
b32967ff | 2360 | isolated = 0; |
e39bb6be HY |
2361 | } |
2362 | if (nr_succeeded) { | |
2363 | count_vm_numa_events(NUMA_PAGE_MIGRATE, nr_succeeded); | |
2364 | if (!node_is_toptier(page_to_nid(page)) && node_is_toptier(node)) | |
2365 | mod_node_page_state(pgdat, PGPROMOTE_SUCCESS, | |
2366 | nr_succeeded); | |
2367 | } | |
7039e1db | 2368 | BUG_ON(!list_empty(&migratepages)); |
7039e1db | 2369 | return isolated; |
340ef390 HD |
2370 | |
2371 | out: | |
2372 | put_page(page); | |
2373 | return 0; | |
7039e1db | 2374 | } |
220018d3 | 2375 | #endif /* CONFIG_NUMA_BALANCING */ |
91952440 | 2376 | #endif /* CONFIG_NUMA */ |