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 | { |
68f2736a | 63 | const struct movable_operations *mops; |
bda807d4 MK |
64 | |
65 | /* | |
66 | * Avoid burning cycles with pages that are yet under __free_pages(), | |
67 | * or just got freed under us. | |
68 | * | |
69 | * In case we 'win' a race for a movable page being freed under us and | |
70 | * raise its refcount preventing __free_pages() from doing its job | |
71 | * the put_page() at the end of this block will take care of | |
72 | * release this page, thus avoiding a nasty leakage. | |
73 | */ | |
74 | if (unlikely(!get_page_unless_zero(page))) | |
75 | goto out; | |
76 | ||
8b881763 VB |
77 | if (unlikely(PageSlab(page))) |
78 | goto out_putpage; | |
79 | /* Pairs with smp_wmb() in slab freeing, e.g. SLUB's __free_slab() */ | |
80 | smp_rmb(); | |
bda807d4 | 81 | /* |
8b881763 VB |
82 | * Check movable flag before taking the page lock because |
83 | * we use non-atomic bitops on newly allocated page flags so | |
84 | * unconditionally grabbing the lock ruins page's owner side. | |
bda807d4 MK |
85 | */ |
86 | if (unlikely(!__PageMovable(page))) | |
87 | goto out_putpage; | |
8b881763 VB |
88 | /* Pairs with smp_wmb() in slab allocation, e.g. SLUB's alloc_slab_page() */ |
89 | smp_rmb(); | |
90 | if (unlikely(PageSlab(page))) | |
91 | goto out_putpage; | |
92 | ||
bda807d4 MK |
93 | /* |
94 | * As movable pages are not isolated from LRU lists, concurrent | |
95 | * compaction threads can race against page migration functions | |
96 | * as well as race against the releasing a page. | |
97 | * | |
98 | * In order to avoid having an already isolated movable page | |
99 | * being (wrongly) re-isolated while it is under migration, | |
100 | * or to avoid attempting to isolate pages being released, | |
101 | * lets be sure we have the page lock | |
102 | * before proceeding with the movable page isolation steps. | |
103 | */ | |
104 | if (unlikely(!trylock_page(page))) | |
105 | goto out_putpage; | |
106 | ||
107 | if (!PageMovable(page) || PageIsolated(page)) | |
108 | goto out_no_isolated; | |
109 | ||
68f2736a MWO |
110 | mops = page_movable_ops(page); |
111 | VM_BUG_ON_PAGE(!mops, page); | |
bda807d4 | 112 | |
68f2736a | 113 | if (!mops->isolate_page(page, mode)) |
bda807d4 MK |
114 | goto out_no_isolated; |
115 | ||
116 | /* Driver shouldn't use PG_isolated bit of page->flags */ | |
117 | WARN_ON_ONCE(PageIsolated(page)); | |
356ea386 | 118 | SetPageIsolated(page); |
bda807d4 MK |
119 | unlock_page(page); |
120 | ||
9e5bcd61 | 121 | return 0; |
bda807d4 MK |
122 | |
123 | out_no_isolated: | |
124 | unlock_page(page); | |
125 | out_putpage: | |
126 | put_page(page); | |
127 | out: | |
9e5bcd61 | 128 | return -EBUSY; |
bda807d4 MK |
129 | } |
130 | ||
606a6f71 | 131 | static void putback_movable_page(struct page *page) |
bda807d4 | 132 | { |
68f2736a | 133 | const struct movable_operations *mops = page_movable_ops(page); |
bda807d4 | 134 | |
68f2736a | 135 | mops->putback_page(page); |
356ea386 | 136 | ClearPageIsolated(page); |
bda807d4 MK |
137 | } |
138 | ||
5733c7d1 RA |
139 | /* |
140 | * Put previously isolated pages back onto the appropriate lists | |
141 | * from where they were once taken off for compaction/migration. | |
142 | * | |
59c82b70 JK |
143 | * This function shall be used whenever the isolated pageset has been |
144 | * built from lru, balloon, hugetlbfs page. See isolate_migratepages_range() | |
7ce82f4c | 145 | * and isolate_hugetlb(). |
5733c7d1 RA |
146 | */ |
147 | void putback_movable_pages(struct list_head *l) | |
148 | { | |
149 | struct page *page; | |
150 | struct page *page2; | |
151 | ||
b20a3503 | 152 | list_for_each_entry_safe(page, page2, l, lru) { |
31caf665 NH |
153 | if (unlikely(PageHuge(page))) { |
154 | putback_active_hugepage(page); | |
155 | continue; | |
156 | } | |
e24f0b8f | 157 | list_del(&page->lru); |
bda807d4 MK |
158 | /* |
159 | * We isolated non-lru movable page so here we can use | |
160 | * __PageMovable because LRU page's mapping cannot have | |
161 | * PAGE_MAPPING_MOVABLE. | |
162 | */ | |
b1123ea6 | 163 | if (unlikely(__PageMovable(page))) { |
bda807d4 MK |
164 | VM_BUG_ON_PAGE(!PageIsolated(page), page); |
165 | lock_page(page); | |
166 | if (PageMovable(page)) | |
167 | putback_movable_page(page); | |
168 | else | |
356ea386 | 169 | ClearPageIsolated(page); |
bda807d4 MK |
170 | unlock_page(page); |
171 | put_page(page); | |
172 | } else { | |
e8db67eb | 173 | mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + |
6c357848 | 174 | page_is_file_lru(page), -thp_nr_pages(page)); |
fc280fe8 | 175 | putback_lru_page(page); |
bda807d4 | 176 | } |
b20a3503 | 177 | } |
b20a3503 CL |
178 | } |
179 | ||
0697212a CL |
180 | /* |
181 | * Restore a potential migration pte to a working pte entry | |
182 | */ | |
2f031c6f MWO |
183 | static bool remove_migration_pte(struct folio *folio, |
184 | struct vm_area_struct *vma, unsigned long addr, void *old) | |
0697212a | 185 | { |
4eecb8b9 | 186 | DEFINE_FOLIO_VMA_WALK(pvmw, old, vma, addr, PVMW_SYNC | PVMW_MIGRATION); |
0697212a | 187 | |
3fe87967 | 188 | while (page_vma_mapped_walk(&pvmw)) { |
6c287605 | 189 | rmap_t rmap_flags = RMAP_NONE; |
4eecb8b9 MWO |
190 | pte_t pte; |
191 | swp_entry_t entry; | |
192 | struct page *new; | |
193 | unsigned long idx = 0; | |
194 | ||
195 | /* pgoff is invalid for ksm pages, but they are never large */ | |
196 | if (folio_test_large(folio) && !folio_test_hugetlb(folio)) | |
197 | idx = linear_page_index(vma, pvmw.address) - pvmw.pgoff; | |
198 | new = folio_page(folio, idx); | |
0697212a | 199 | |
616b8371 ZY |
200 | #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION |
201 | /* PMD-mapped THP migration entry */ | |
202 | if (!pvmw.pte) { | |
4eecb8b9 MWO |
203 | VM_BUG_ON_FOLIO(folio_test_hugetlb(folio) || |
204 | !folio_test_pmd_mappable(folio), folio); | |
616b8371 ZY |
205 | remove_migration_pmd(&pvmw, new); |
206 | continue; | |
207 | } | |
208 | #endif | |
209 | ||
4eecb8b9 | 210 | folio_get(folio); |
2e346877 | 211 | pte = mk_pte(new, READ_ONCE(vma->vm_page_prot)); |
3fe87967 KS |
212 | if (pte_swp_soft_dirty(*pvmw.pte)) |
213 | pte = pte_mksoft_dirty(pte); | |
0697212a | 214 | |
3fe87967 KS |
215 | /* |
216 | * Recheck VMA as permissions can change since migration started | |
217 | */ | |
218 | entry = pte_to_swp_entry(*pvmw.pte); | |
2e346877 PX |
219 | if (!is_migration_entry_young(entry)) |
220 | pte = pte_mkold(pte); | |
221 | if (folio_test_dirty(folio) && is_migration_entry_dirty(entry)) | |
222 | pte = pte_mkdirty(pte); | |
4dd845b5 | 223 | if (is_writable_migration_entry(entry)) |
3fe87967 | 224 | pte = maybe_mkwrite(pte, vma); |
f45ec5ff PX |
225 | else if (pte_swp_uffd_wp(*pvmw.pte)) |
226 | pte = pte_mkuffd_wp(pte); | |
d3cb8bf6 | 227 | |
6c287605 DH |
228 | if (folio_test_anon(folio) && !is_readable_migration_entry(entry)) |
229 | rmap_flags |= RMAP_EXCLUSIVE; | |
230 | ||
6128763f | 231 | if (unlikely(is_device_private_page(new))) { |
4dd845b5 AP |
232 | if (pte_write(pte)) |
233 | entry = make_writable_device_private_entry( | |
234 | page_to_pfn(new)); | |
235 | else | |
236 | entry = make_readable_device_private_entry( | |
237 | page_to_pfn(new)); | |
6128763f | 238 | pte = swp_entry_to_pte(entry); |
3d321bf8 RC |
239 | if (pte_swp_soft_dirty(*pvmw.pte)) |
240 | pte = pte_swp_mksoft_dirty(pte); | |
6128763f RC |
241 | if (pte_swp_uffd_wp(*pvmw.pte)) |
242 | pte = pte_swp_mkuffd_wp(pte); | |
d2b2c6dd | 243 | } |
a5430dda | 244 | |
3ef8fd7f | 245 | #ifdef CONFIG_HUGETLB_PAGE |
4eecb8b9 | 246 | if (folio_test_hugetlb(folio)) { |
79c1c594 CL |
247 | unsigned int shift = huge_page_shift(hstate_vma(vma)); |
248 | ||
3fe87967 | 249 | pte = pte_mkhuge(pte); |
79c1c594 | 250 | pte = arch_make_huge_pte(pte, shift, vma->vm_flags); |
4eecb8b9 | 251 | if (folio_test_anon(folio)) |
28c5209d | 252 | hugepage_add_anon_rmap(new, vma, pvmw.address, |
6c287605 | 253 | rmap_flags); |
3fe87967 | 254 | else |
fb3d824d | 255 | page_dup_file_rmap(new, true); |
1eba86c0 | 256 | set_huge_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte); |
383321ab AK |
257 | } else |
258 | #endif | |
259 | { | |
4eecb8b9 | 260 | if (folio_test_anon(folio)) |
f1e2db12 | 261 | page_add_anon_rmap(new, vma, pvmw.address, |
6c287605 | 262 | rmap_flags); |
383321ab | 263 | else |
cea86fe2 | 264 | page_add_file_rmap(new, vma, false); |
1eba86c0 | 265 | set_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte); |
383321ab | 266 | } |
b7435507 | 267 | if (vma->vm_flags & VM_LOCKED) |
96f97c43 | 268 | mlock_drain_local(); |
e125fe40 | 269 | |
4cc79b33 AK |
270 | trace_remove_migration_pte(pvmw.address, pte_val(pte), |
271 | compound_order(new)); | |
272 | ||
3fe87967 KS |
273 | /* No need to invalidate - it was non-present before */ |
274 | update_mmu_cache(vma, pvmw.address, pvmw.pte); | |
275 | } | |
51afb12b | 276 | |
e4b82222 | 277 | return true; |
0697212a CL |
278 | } |
279 | ||
04e62a29 CL |
280 | /* |
281 | * Get rid of all migration entries and replace them by | |
282 | * references to the indicated page. | |
283 | */ | |
4eecb8b9 | 284 | void remove_migration_ptes(struct folio *src, struct folio *dst, bool locked) |
04e62a29 | 285 | { |
051ac83a JK |
286 | struct rmap_walk_control rwc = { |
287 | .rmap_one = remove_migration_pte, | |
4eecb8b9 | 288 | .arg = src, |
051ac83a JK |
289 | }; |
290 | ||
e388466d | 291 | if (locked) |
2f031c6f | 292 | rmap_walk_locked(dst, &rwc); |
e388466d | 293 | else |
2f031c6f | 294 | rmap_walk(dst, &rwc); |
04e62a29 CL |
295 | } |
296 | ||
0697212a CL |
297 | /* |
298 | * Something used the pte of a page under migration. We need to | |
299 | * get to the page and wait until migration is finished. | |
300 | * When we return from this function the fault will be retried. | |
0697212a | 301 | */ |
e66f17ff | 302 | void __migration_entry_wait(struct mm_struct *mm, pte_t *ptep, |
30dad309 | 303 | spinlock_t *ptl) |
0697212a | 304 | { |
30dad309 | 305 | pte_t pte; |
0697212a | 306 | swp_entry_t entry; |
0697212a | 307 | |
30dad309 | 308 | spin_lock(ptl); |
0697212a CL |
309 | pte = *ptep; |
310 | if (!is_swap_pte(pte)) | |
311 | goto out; | |
312 | ||
313 | entry = pte_to_swp_entry(pte); | |
314 | if (!is_migration_entry(entry)) | |
315 | goto out; | |
316 | ||
ffa65753 | 317 | migration_entry_wait_on_locked(entry, ptep, ptl); |
0697212a CL |
318 | return; |
319 | out: | |
320 | pte_unmap_unlock(ptep, ptl); | |
321 | } | |
322 | ||
30dad309 NH |
323 | void migration_entry_wait(struct mm_struct *mm, pmd_t *pmd, |
324 | unsigned long address) | |
325 | { | |
326 | spinlock_t *ptl = pte_lockptr(mm, pmd); | |
327 | pte_t *ptep = pte_offset_map(pmd, address); | |
328 | __migration_entry_wait(mm, ptep, ptl); | |
329 | } | |
330 | ||
ad1ac596 | 331 | #ifdef CONFIG_HUGETLB_PAGE |
fcd48540 PX |
332 | /* |
333 | * The vma read lock must be held upon entry. Holding that lock prevents either | |
334 | * the pte or the ptl from being freed. | |
335 | * | |
336 | * This function will release the vma lock before returning. | |
337 | */ | |
338 | void __migration_entry_wait_huge(struct vm_area_struct *vma, | |
339 | pte_t *ptep, spinlock_t *ptl) | |
30dad309 | 340 | { |
ad1ac596 ML |
341 | pte_t pte; |
342 | ||
fcd48540 | 343 | hugetlb_vma_assert_locked(vma); |
ad1ac596 ML |
344 | spin_lock(ptl); |
345 | pte = huge_ptep_get(ptep); | |
346 | ||
fcd48540 | 347 | if (unlikely(!is_hugetlb_entry_migration(pte))) { |
ad1ac596 | 348 | spin_unlock(ptl); |
fcd48540 PX |
349 | hugetlb_vma_unlock_read(vma); |
350 | } else { | |
351 | /* | |
352 | * If migration entry existed, safe to release vma lock | |
353 | * here because the pgtable page won't be freed without the | |
354 | * pgtable lock released. See comment right above pgtable | |
355 | * lock release in migration_entry_wait_on_locked(). | |
356 | */ | |
357 | hugetlb_vma_unlock_read(vma); | |
ad1ac596 | 358 | migration_entry_wait_on_locked(pte_to_swp_entry(pte), NULL, ptl); |
fcd48540 | 359 | } |
30dad309 NH |
360 | } |
361 | ||
ad1ac596 ML |
362 | void migration_entry_wait_huge(struct vm_area_struct *vma, pte_t *pte) |
363 | { | |
364 | spinlock_t *ptl = huge_pte_lockptr(hstate_vma(vma), vma->vm_mm, pte); | |
365 | ||
fcd48540 | 366 | __migration_entry_wait_huge(vma, pte, ptl); |
ad1ac596 ML |
367 | } |
368 | #endif | |
369 | ||
616b8371 ZY |
370 | #ifdef CONFIG_ARCH_ENABLE_THP_MIGRATION |
371 | void pmd_migration_entry_wait(struct mm_struct *mm, pmd_t *pmd) | |
372 | { | |
373 | spinlock_t *ptl; | |
616b8371 ZY |
374 | |
375 | ptl = pmd_lock(mm, pmd); | |
376 | if (!is_pmd_migration_entry(*pmd)) | |
377 | goto unlock; | |
ffa65753 | 378 | migration_entry_wait_on_locked(pmd_to_swp_entry(*pmd), NULL, ptl); |
616b8371 ZY |
379 | return; |
380 | unlock: | |
381 | spin_unlock(ptl); | |
382 | } | |
383 | #endif | |
384 | ||
108ca835 MWO |
385 | static int folio_expected_refs(struct address_space *mapping, |
386 | struct folio *folio) | |
0b3901b3 | 387 | { |
108ca835 MWO |
388 | int refs = 1; |
389 | if (!mapping) | |
390 | return refs; | |
0b3901b3 | 391 | |
108ca835 MWO |
392 | refs += folio_nr_pages(folio); |
393 | if (folio_test_private(folio)) | |
394 | refs++; | |
395 | ||
396 | return refs; | |
0b3901b3 JK |
397 | } |
398 | ||
b20a3503 | 399 | /* |
c3fcf8a5 | 400 | * Replace the page in the mapping. |
5b5c7120 CL |
401 | * |
402 | * The number of remaining references must be: | |
403 | * 1 for anonymous pages without a mapping | |
404 | * 2 for pages with a mapping | |
266cf658 | 405 | * 3 for pages with a mapping and PagePrivate/PagePrivate2 set. |
b20a3503 | 406 | */ |
3417013e MWO |
407 | int folio_migrate_mapping(struct address_space *mapping, |
408 | struct folio *newfolio, struct folio *folio, int extra_count) | |
b20a3503 | 409 | { |
3417013e | 410 | XA_STATE(xas, &mapping->i_pages, folio_index(folio)); |
42cb14b1 HD |
411 | struct zone *oldzone, *newzone; |
412 | int dirty; | |
108ca835 | 413 | int expected_count = folio_expected_refs(mapping, folio) + extra_count; |
3417013e | 414 | long nr = folio_nr_pages(folio); |
8763cb45 | 415 | |
6c5240ae | 416 | if (!mapping) { |
0e8c7d0f | 417 | /* Anonymous page without mapping */ |
3417013e | 418 | if (folio_ref_count(folio) != expected_count) |
6c5240ae | 419 | return -EAGAIN; |
cf4b769a HD |
420 | |
421 | /* No turning back from here */ | |
3417013e MWO |
422 | newfolio->index = folio->index; |
423 | newfolio->mapping = folio->mapping; | |
424 | if (folio_test_swapbacked(folio)) | |
425 | __folio_set_swapbacked(newfolio); | |
cf4b769a | 426 | |
78bd5209 | 427 | return MIGRATEPAGE_SUCCESS; |
6c5240ae CL |
428 | } |
429 | ||
3417013e MWO |
430 | oldzone = folio_zone(folio); |
431 | newzone = folio_zone(newfolio); | |
42cb14b1 | 432 | |
89eb946a | 433 | xas_lock_irq(&xas); |
3417013e | 434 | if (!folio_ref_freeze(folio, expected_count)) { |
89eb946a | 435 | xas_unlock_irq(&xas); |
e286781d NP |
436 | return -EAGAIN; |
437 | } | |
438 | ||
b20a3503 | 439 | /* |
3417013e | 440 | * Now we know that no one else is looking at the folio: |
cf4b769a | 441 | * no turning back from here. |
b20a3503 | 442 | */ |
3417013e MWO |
443 | newfolio->index = folio->index; |
444 | newfolio->mapping = folio->mapping; | |
445 | folio_ref_add(newfolio, nr); /* add cache reference */ | |
446 | if (folio_test_swapbacked(folio)) { | |
447 | __folio_set_swapbacked(newfolio); | |
448 | if (folio_test_swapcache(folio)) { | |
449 | folio_set_swapcache(newfolio); | |
450 | newfolio->private = folio_get_private(folio); | |
6326fec1 NP |
451 | } |
452 | } else { | |
3417013e | 453 | VM_BUG_ON_FOLIO(folio_test_swapcache(folio), folio); |
b20a3503 CL |
454 | } |
455 | ||
42cb14b1 | 456 | /* Move dirty while page refs frozen and newpage not yet exposed */ |
3417013e | 457 | dirty = folio_test_dirty(folio); |
42cb14b1 | 458 | if (dirty) { |
3417013e MWO |
459 | folio_clear_dirty(folio); |
460 | folio_set_dirty(newfolio); | |
42cb14b1 HD |
461 | } |
462 | ||
3417013e | 463 | xas_store(&xas, newfolio); |
7cf9c2c7 NP |
464 | |
465 | /* | |
937a94c9 JG |
466 | * Drop cache reference from old page by unfreezing |
467 | * to one less reference. | |
7cf9c2c7 NP |
468 | * We know this isn't the last reference. |
469 | */ | |
3417013e | 470 | folio_ref_unfreeze(folio, expected_count - nr); |
7cf9c2c7 | 471 | |
89eb946a | 472 | xas_unlock(&xas); |
42cb14b1 HD |
473 | /* Leave irq disabled to prevent preemption while updating stats */ |
474 | ||
0e8c7d0f CL |
475 | /* |
476 | * If moved to a different zone then also account | |
477 | * the page for that zone. Other VM counters will be | |
478 | * taken care of when we establish references to the | |
479 | * new page and drop references to the old page. | |
480 | * | |
481 | * Note that anonymous pages are accounted for | |
4b9d0fab | 482 | * via NR_FILE_PAGES and NR_ANON_MAPPED if they |
0e8c7d0f CL |
483 | * are mapped to swap space. |
484 | */ | |
42cb14b1 | 485 | if (newzone != oldzone) { |
0d1c2072 JW |
486 | struct lruvec *old_lruvec, *new_lruvec; |
487 | struct mem_cgroup *memcg; | |
488 | ||
3417013e | 489 | memcg = folio_memcg(folio); |
0d1c2072 JW |
490 | old_lruvec = mem_cgroup_lruvec(memcg, oldzone->zone_pgdat); |
491 | new_lruvec = mem_cgroup_lruvec(memcg, newzone->zone_pgdat); | |
492 | ||
5c447d27 SB |
493 | __mod_lruvec_state(old_lruvec, NR_FILE_PAGES, -nr); |
494 | __mod_lruvec_state(new_lruvec, NR_FILE_PAGES, nr); | |
3417013e | 495 | if (folio_test_swapbacked(folio) && !folio_test_swapcache(folio)) { |
5c447d27 SB |
496 | __mod_lruvec_state(old_lruvec, NR_SHMEM, -nr); |
497 | __mod_lruvec_state(new_lruvec, NR_SHMEM, nr); | |
42cb14b1 | 498 | } |
b6038942 | 499 | #ifdef CONFIG_SWAP |
3417013e | 500 | if (folio_test_swapcache(folio)) { |
b6038942 SB |
501 | __mod_lruvec_state(old_lruvec, NR_SWAPCACHE, -nr); |
502 | __mod_lruvec_state(new_lruvec, NR_SWAPCACHE, nr); | |
503 | } | |
504 | #endif | |
f56753ac | 505 | if (dirty && mapping_can_writeback(mapping)) { |
5c447d27 SB |
506 | __mod_lruvec_state(old_lruvec, NR_FILE_DIRTY, -nr); |
507 | __mod_zone_page_state(oldzone, NR_ZONE_WRITE_PENDING, -nr); | |
508 | __mod_lruvec_state(new_lruvec, NR_FILE_DIRTY, nr); | |
509 | __mod_zone_page_state(newzone, NR_ZONE_WRITE_PENDING, nr); | |
42cb14b1 | 510 | } |
4b02108a | 511 | } |
42cb14b1 | 512 | local_irq_enable(); |
b20a3503 | 513 | |
78bd5209 | 514 | return MIGRATEPAGE_SUCCESS; |
b20a3503 | 515 | } |
3417013e | 516 | EXPORT_SYMBOL(folio_migrate_mapping); |
b20a3503 | 517 | |
290408d4 NH |
518 | /* |
519 | * The expected number of remaining references is the same as that | |
3417013e | 520 | * of folio_migrate_mapping(). |
290408d4 NH |
521 | */ |
522 | int migrate_huge_page_move_mapping(struct address_space *mapping, | |
b890ec2a | 523 | struct folio *dst, struct folio *src) |
290408d4 | 524 | { |
b890ec2a | 525 | XA_STATE(xas, &mapping->i_pages, folio_index(src)); |
290408d4 | 526 | int expected_count; |
290408d4 | 527 | |
89eb946a | 528 | xas_lock_irq(&xas); |
b890ec2a MWO |
529 | expected_count = 2 + folio_has_private(src); |
530 | if (!folio_ref_freeze(src, expected_count)) { | |
89eb946a | 531 | xas_unlock_irq(&xas); |
290408d4 NH |
532 | return -EAGAIN; |
533 | } | |
534 | ||
b890ec2a MWO |
535 | dst->index = src->index; |
536 | dst->mapping = src->mapping; | |
6a93ca8f | 537 | |
b890ec2a | 538 | folio_get(dst); |
290408d4 | 539 | |
b890ec2a | 540 | xas_store(&xas, dst); |
290408d4 | 541 | |
b890ec2a | 542 | folio_ref_unfreeze(src, expected_count - 1); |
290408d4 | 543 | |
89eb946a | 544 | xas_unlock_irq(&xas); |
6a93ca8f | 545 | |
78bd5209 | 546 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
547 | } |
548 | ||
b20a3503 | 549 | /* |
19138349 | 550 | * Copy the flags and some other ancillary information |
b20a3503 | 551 | */ |
19138349 | 552 | void folio_migrate_flags(struct folio *newfolio, struct folio *folio) |
b20a3503 | 553 | { |
7851a45c RR |
554 | int cpupid; |
555 | ||
19138349 MWO |
556 | if (folio_test_error(folio)) |
557 | folio_set_error(newfolio); | |
558 | if (folio_test_referenced(folio)) | |
559 | folio_set_referenced(newfolio); | |
560 | if (folio_test_uptodate(folio)) | |
561 | folio_mark_uptodate(newfolio); | |
562 | if (folio_test_clear_active(folio)) { | |
563 | VM_BUG_ON_FOLIO(folio_test_unevictable(folio), folio); | |
564 | folio_set_active(newfolio); | |
565 | } else if (folio_test_clear_unevictable(folio)) | |
566 | folio_set_unevictable(newfolio); | |
567 | if (folio_test_workingset(folio)) | |
568 | folio_set_workingset(newfolio); | |
569 | if (folio_test_checked(folio)) | |
570 | folio_set_checked(newfolio); | |
6c287605 DH |
571 | /* |
572 | * PG_anon_exclusive (-> PG_mappedtodisk) is always migrated via | |
573 | * migration entries. We can still have PG_anon_exclusive set on an | |
574 | * effectively unmapped and unreferenced first sub-pages of an | |
575 | * anonymous THP: we can simply copy it here via PG_mappedtodisk. | |
576 | */ | |
19138349 MWO |
577 | if (folio_test_mappedtodisk(folio)) |
578 | folio_set_mappedtodisk(newfolio); | |
b20a3503 | 579 | |
3417013e | 580 | /* Move dirty on pages not done by folio_migrate_mapping() */ |
19138349 MWO |
581 | if (folio_test_dirty(folio)) |
582 | folio_set_dirty(newfolio); | |
b20a3503 | 583 | |
19138349 MWO |
584 | if (folio_test_young(folio)) |
585 | folio_set_young(newfolio); | |
586 | if (folio_test_idle(folio)) | |
587 | folio_set_idle(newfolio); | |
33c3fc71 | 588 | |
7851a45c RR |
589 | /* |
590 | * Copy NUMA information to the new page, to prevent over-eager | |
591 | * future migrations of this same page. | |
592 | */ | |
19138349 | 593 | cpupid = page_cpupid_xchg_last(&folio->page, -1); |
33024536 HY |
594 | /* |
595 | * For memory tiering mode, when migrate between slow and fast | |
596 | * memory node, reset cpupid, because that is used to record | |
597 | * page access time in slow memory node. | |
598 | */ | |
599 | if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING) { | |
600 | bool f_toptier = node_is_toptier(page_to_nid(&folio->page)); | |
601 | bool t_toptier = node_is_toptier(page_to_nid(&newfolio->page)); | |
602 | ||
603 | if (f_toptier != t_toptier) | |
604 | cpupid = -1; | |
605 | } | |
19138349 | 606 | page_cpupid_xchg_last(&newfolio->page, cpupid); |
7851a45c | 607 | |
19138349 | 608 | folio_migrate_ksm(newfolio, folio); |
c8d6553b HD |
609 | /* |
610 | * Please do not reorder this without considering how mm/ksm.c's | |
611 | * get_ksm_page() depends upon ksm_migrate_page() and PageSwapCache(). | |
612 | */ | |
19138349 MWO |
613 | if (folio_test_swapcache(folio)) |
614 | folio_clear_swapcache(folio); | |
615 | folio_clear_private(folio); | |
ad2fa371 MS |
616 | |
617 | /* page->private contains hugetlb specific flags */ | |
19138349 MWO |
618 | if (!folio_test_hugetlb(folio)) |
619 | folio->private = NULL; | |
b20a3503 CL |
620 | |
621 | /* | |
622 | * If any waiters have accumulated on the new page then | |
623 | * wake them up. | |
624 | */ | |
19138349 MWO |
625 | if (folio_test_writeback(newfolio)) |
626 | folio_end_writeback(newfolio); | |
d435edca | 627 | |
6aeff241 YS |
628 | /* |
629 | * PG_readahead shares the same bit with PG_reclaim. The above | |
630 | * end_page_writeback() may clear PG_readahead mistakenly, so set the | |
631 | * bit after that. | |
632 | */ | |
19138349 MWO |
633 | if (folio_test_readahead(folio)) |
634 | folio_set_readahead(newfolio); | |
6aeff241 | 635 | |
19138349 | 636 | folio_copy_owner(newfolio, folio); |
74485cf2 | 637 | |
19138349 | 638 | if (!folio_test_hugetlb(folio)) |
d21bba2b | 639 | mem_cgroup_migrate(folio, newfolio); |
b20a3503 | 640 | } |
19138349 | 641 | EXPORT_SYMBOL(folio_migrate_flags); |
2916ecc0 | 642 | |
715cbfd6 | 643 | void folio_migrate_copy(struct folio *newfolio, struct folio *folio) |
2916ecc0 | 644 | { |
715cbfd6 MWO |
645 | folio_copy(newfolio, folio); |
646 | folio_migrate_flags(newfolio, folio); | |
2916ecc0 | 647 | } |
715cbfd6 | 648 | EXPORT_SYMBOL(folio_migrate_copy); |
b20a3503 | 649 | |
1d8b85cc CL |
650 | /************************************************************ |
651 | * Migration functions | |
652 | ***********************************************************/ | |
653 | ||
16ce101d AP |
654 | int migrate_folio_extra(struct address_space *mapping, struct folio *dst, |
655 | struct folio *src, enum migrate_mode mode, int extra_count) | |
656 | { | |
657 | int rc; | |
658 | ||
659 | BUG_ON(folio_test_writeback(src)); /* Writeback must be complete */ | |
660 | ||
661 | rc = folio_migrate_mapping(mapping, dst, src, extra_count); | |
662 | ||
663 | if (rc != MIGRATEPAGE_SUCCESS) | |
664 | return rc; | |
665 | ||
666 | if (mode != MIGRATE_SYNC_NO_COPY) | |
667 | folio_migrate_copy(dst, src); | |
668 | else | |
669 | folio_migrate_flags(dst, src); | |
670 | return MIGRATEPAGE_SUCCESS; | |
671 | } | |
672 | ||
54184650 MWO |
673 | /** |
674 | * migrate_folio() - Simple folio migration. | |
675 | * @mapping: The address_space containing the folio. | |
676 | * @dst: The folio to migrate the data to. | |
677 | * @src: The folio containing the current data. | |
678 | * @mode: How to migrate the page. | |
679 | * | |
680 | * Common logic to directly migrate a single LRU folio suitable for | |
681 | * folios that do not use PagePrivate/PagePrivate2. | |
b20a3503 | 682 | * |
54184650 | 683 | * Folios are locked upon entry and exit. |
b20a3503 | 684 | */ |
54184650 MWO |
685 | int migrate_folio(struct address_space *mapping, struct folio *dst, |
686 | struct folio *src, enum migrate_mode mode) | |
b20a3503 | 687 | { |
16ce101d | 688 | return migrate_folio_extra(mapping, dst, src, mode, 0); |
b20a3503 | 689 | } |
54184650 | 690 | EXPORT_SYMBOL(migrate_folio); |
b20a3503 | 691 | |
9361401e | 692 | #ifdef CONFIG_BLOCK |
84ade7c1 JK |
693 | /* Returns true if all buffers are successfully locked */ |
694 | static bool buffer_migrate_lock_buffers(struct buffer_head *head, | |
695 | enum migrate_mode mode) | |
696 | { | |
697 | struct buffer_head *bh = head; | |
698 | ||
699 | /* Simple case, sync compaction */ | |
700 | if (mode != MIGRATE_ASYNC) { | |
701 | do { | |
84ade7c1 JK |
702 | lock_buffer(bh); |
703 | bh = bh->b_this_page; | |
704 | ||
705 | } while (bh != head); | |
706 | ||
707 | return true; | |
708 | } | |
709 | ||
710 | /* async case, we cannot block on lock_buffer so use trylock_buffer */ | |
711 | do { | |
84ade7c1 JK |
712 | if (!trylock_buffer(bh)) { |
713 | /* | |
714 | * We failed to lock the buffer and cannot stall in | |
715 | * async migration. Release the taken locks | |
716 | */ | |
717 | struct buffer_head *failed_bh = bh; | |
84ade7c1 JK |
718 | bh = head; |
719 | while (bh != failed_bh) { | |
720 | unlock_buffer(bh); | |
84ade7c1 JK |
721 | bh = bh->b_this_page; |
722 | } | |
723 | return false; | |
724 | } | |
725 | ||
726 | bh = bh->b_this_page; | |
727 | } while (bh != head); | |
728 | return true; | |
729 | } | |
730 | ||
67235182 MWO |
731 | static int __buffer_migrate_folio(struct address_space *mapping, |
732 | struct folio *dst, struct folio *src, enum migrate_mode mode, | |
89cb0888 | 733 | bool check_refs) |
1d8b85cc | 734 | { |
1d8b85cc CL |
735 | struct buffer_head *bh, *head; |
736 | int rc; | |
cc4f11e6 | 737 | int expected_count; |
1d8b85cc | 738 | |
67235182 MWO |
739 | head = folio_buffers(src); |
740 | if (!head) | |
54184650 | 741 | return migrate_folio(mapping, dst, src, mode); |
1d8b85cc | 742 | |
cc4f11e6 | 743 | /* Check whether page does not have extra refs before we do more work */ |
108ca835 | 744 | expected_count = folio_expected_refs(mapping, src); |
67235182 | 745 | if (folio_ref_count(src) != expected_count) |
cc4f11e6 | 746 | return -EAGAIN; |
1d8b85cc | 747 | |
cc4f11e6 JK |
748 | if (!buffer_migrate_lock_buffers(head, mode)) |
749 | return -EAGAIN; | |
1d8b85cc | 750 | |
89cb0888 JK |
751 | if (check_refs) { |
752 | bool busy; | |
753 | bool invalidated = false; | |
754 | ||
755 | recheck_buffers: | |
756 | busy = false; | |
757 | spin_lock(&mapping->private_lock); | |
758 | bh = head; | |
759 | do { | |
760 | if (atomic_read(&bh->b_count)) { | |
761 | busy = true; | |
762 | break; | |
763 | } | |
764 | bh = bh->b_this_page; | |
765 | } while (bh != head); | |
89cb0888 JK |
766 | if (busy) { |
767 | if (invalidated) { | |
768 | rc = -EAGAIN; | |
769 | goto unlock_buffers; | |
770 | } | |
ebdf4de5 | 771 | spin_unlock(&mapping->private_lock); |
89cb0888 JK |
772 | invalidate_bh_lrus(); |
773 | invalidated = true; | |
774 | goto recheck_buffers; | |
775 | } | |
776 | } | |
777 | ||
67235182 | 778 | rc = folio_migrate_mapping(mapping, dst, src, 0); |
78bd5209 | 779 | if (rc != MIGRATEPAGE_SUCCESS) |
cc4f11e6 | 780 | goto unlock_buffers; |
1d8b85cc | 781 | |
67235182 | 782 | folio_attach_private(dst, folio_detach_private(src)); |
1d8b85cc CL |
783 | |
784 | bh = head; | |
785 | do { | |
67235182 | 786 | set_bh_page(bh, &dst->page, bh_offset(bh)); |
1d8b85cc | 787 | bh = bh->b_this_page; |
1d8b85cc CL |
788 | } while (bh != head); |
789 | ||
2916ecc0 | 790 | if (mode != MIGRATE_SYNC_NO_COPY) |
67235182 | 791 | folio_migrate_copy(dst, src); |
2916ecc0 | 792 | else |
67235182 | 793 | folio_migrate_flags(dst, src); |
1d8b85cc | 794 | |
cc4f11e6 JK |
795 | rc = MIGRATEPAGE_SUCCESS; |
796 | unlock_buffers: | |
ebdf4de5 JK |
797 | if (check_refs) |
798 | spin_unlock(&mapping->private_lock); | |
1d8b85cc CL |
799 | bh = head; |
800 | do { | |
801 | unlock_buffer(bh); | |
1d8b85cc | 802 | bh = bh->b_this_page; |
1d8b85cc CL |
803 | } while (bh != head); |
804 | ||
cc4f11e6 | 805 | return rc; |
1d8b85cc | 806 | } |
89cb0888 | 807 | |
67235182 MWO |
808 | /** |
809 | * buffer_migrate_folio() - Migration function for folios with buffers. | |
810 | * @mapping: The address space containing @src. | |
811 | * @dst: The folio to migrate to. | |
812 | * @src: The folio to migrate from. | |
813 | * @mode: How to migrate the folio. | |
814 | * | |
815 | * This function can only be used if the underlying filesystem guarantees | |
816 | * that no other references to @src exist. For example attached buffer | |
817 | * heads are accessed only under the folio lock. If your filesystem cannot | |
818 | * provide this guarantee, buffer_migrate_folio_norefs() may be more | |
819 | * appropriate. | |
820 | * | |
821 | * Return: 0 on success or a negative errno on failure. | |
89cb0888 | 822 | */ |
67235182 MWO |
823 | int buffer_migrate_folio(struct address_space *mapping, |
824 | struct folio *dst, struct folio *src, enum migrate_mode mode) | |
89cb0888 | 825 | { |
67235182 | 826 | return __buffer_migrate_folio(mapping, dst, src, mode, false); |
89cb0888 | 827 | } |
67235182 MWO |
828 | EXPORT_SYMBOL(buffer_migrate_folio); |
829 | ||
830 | /** | |
831 | * buffer_migrate_folio_norefs() - Migration function for folios with buffers. | |
832 | * @mapping: The address space containing @src. | |
833 | * @dst: The folio to migrate to. | |
834 | * @src: The folio to migrate from. | |
835 | * @mode: How to migrate the folio. | |
836 | * | |
837 | * Like buffer_migrate_folio() except that this variant is more careful | |
838 | * and checks that there are also no buffer head references. This function | |
839 | * is the right one for mappings where buffer heads are directly looked | |
840 | * up and referenced (such as block device mappings). | |
841 | * | |
842 | * Return: 0 on success or a negative errno on failure. | |
89cb0888 | 843 | */ |
67235182 MWO |
844 | int buffer_migrate_folio_norefs(struct address_space *mapping, |
845 | struct folio *dst, struct folio *src, enum migrate_mode mode) | |
89cb0888 | 846 | { |
67235182 | 847 | return __buffer_migrate_folio(mapping, dst, src, mode, true); |
89cb0888 | 848 | } |
e26355e2 | 849 | EXPORT_SYMBOL_GPL(buffer_migrate_folio_norefs); |
9361401e | 850 | #endif |
1d8b85cc | 851 | |
2ec810d5 MWO |
852 | int filemap_migrate_folio(struct address_space *mapping, |
853 | struct folio *dst, struct folio *src, enum migrate_mode mode) | |
854 | { | |
855 | int ret; | |
856 | ||
857 | ret = folio_migrate_mapping(mapping, dst, src, 0); | |
858 | if (ret != MIGRATEPAGE_SUCCESS) | |
859 | return ret; | |
860 | ||
861 | if (folio_get_private(src)) | |
862 | folio_attach_private(dst, folio_detach_private(src)); | |
863 | ||
864 | if (mode != MIGRATE_SYNC_NO_COPY) | |
865 | folio_migrate_copy(dst, src); | |
866 | else | |
867 | folio_migrate_flags(dst, src); | |
868 | return MIGRATEPAGE_SUCCESS; | |
869 | } | |
870 | EXPORT_SYMBOL_GPL(filemap_migrate_folio); | |
871 | ||
04e62a29 | 872 | /* |
2be7fa10 | 873 | * Writeback a folio to clean the dirty state |
04e62a29 | 874 | */ |
2be7fa10 | 875 | static int writeout(struct address_space *mapping, struct folio *folio) |
8351a6e4 | 876 | { |
04e62a29 CL |
877 | struct writeback_control wbc = { |
878 | .sync_mode = WB_SYNC_NONE, | |
879 | .nr_to_write = 1, | |
880 | .range_start = 0, | |
881 | .range_end = LLONG_MAX, | |
04e62a29 CL |
882 | .for_reclaim = 1 |
883 | }; | |
884 | int rc; | |
885 | ||
886 | if (!mapping->a_ops->writepage) | |
887 | /* No write method for the address space */ | |
888 | return -EINVAL; | |
889 | ||
2be7fa10 | 890 | if (!folio_clear_dirty_for_io(folio)) |
04e62a29 CL |
891 | /* Someone else already triggered a write */ |
892 | return -EAGAIN; | |
893 | ||
8351a6e4 | 894 | /* |
2be7fa10 MWO |
895 | * A dirty folio may imply that the underlying filesystem has |
896 | * the folio on some queue. So the folio must be clean for | |
897 | * migration. Writeout may mean we lose the lock and the | |
898 | * folio state is no longer what we checked for earlier. | |
04e62a29 CL |
899 | * At this point we know that the migration attempt cannot |
900 | * be successful. | |
8351a6e4 | 901 | */ |
4eecb8b9 | 902 | remove_migration_ptes(folio, folio, false); |
8351a6e4 | 903 | |
2be7fa10 | 904 | rc = mapping->a_ops->writepage(&folio->page, &wbc); |
8351a6e4 | 905 | |
04e62a29 CL |
906 | if (rc != AOP_WRITEPAGE_ACTIVATE) |
907 | /* unlocked. Relock */ | |
2be7fa10 | 908 | folio_lock(folio); |
04e62a29 | 909 | |
bda8550d | 910 | return (rc < 0) ? -EIO : -EAGAIN; |
04e62a29 CL |
911 | } |
912 | ||
913 | /* | |
914 | * Default handling if a filesystem does not provide a migration function. | |
915 | */ | |
8faa8ef5 MWO |
916 | static int fallback_migrate_folio(struct address_space *mapping, |
917 | struct folio *dst, struct folio *src, enum migrate_mode mode) | |
04e62a29 | 918 | { |
8faa8ef5 MWO |
919 | if (folio_test_dirty(src)) { |
920 | /* Only writeback folios in full synchronous migration */ | |
2916ecc0 JG |
921 | switch (mode) { |
922 | case MIGRATE_SYNC: | |
923 | case MIGRATE_SYNC_NO_COPY: | |
924 | break; | |
925 | default: | |
b969c4ab | 926 | return -EBUSY; |
2916ecc0 | 927 | } |
2be7fa10 | 928 | return writeout(mapping, src); |
b969c4ab | 929 | } |
8351a6e4 CL |
930 | |
931 | /* | |
932 | * Buffers may be managed in a filesystem specific way. | |
933 | * We must have no buffers or drop them. | |
934 | */ | |
8faa8ef5 MWO |
935 | if (folio_test_private(src) && |
936 | !filemap_release_folio(src, GFP_KERNEL)) | |
806031bb | 937 | return mode == MIGRATE_SYNC ? -EAGAIN : -EBUSY; |
8351a6e4 | 938 | |
54184650 | 939 | return migrate_folio(mapping, dst, src, mode); |
8351a6e4 CL |
940 | } |
941 | ||
e24f0b8f CL |
942 | /* |
943 | * Move a page to a newly allocated page | |
944 | * The page is locked and all ptes have been successfully removed. | |
945 | * | |
946 | * The new page will have replaced the old page if this function | |
947 | * is successful. | |
894bc310 LS |
948 | * |
949 | * Return value: | |
950 | * < 0 - error code | |
78bd5209 | 951 | * MIGRATEPAGE_SUCCESS - success |
e24f0b8f | 952 | */ |
e7e3ffeb | 953 | static int move_to_new_folio(struct folio *dst, struct folio *src, |
5c3f9a67 | 954 | enum migrate_mode mode) |
e24f0b8f | 955 | { |
bda807d4 | 956 | int rc = -EAGAIN; |
e7e3ffeb | 957 | bool is_lru = !__PageMovable(&src->page); |
e24f0b8f | 958 | |
e7e3ffeb MWO |
959 | VM_BUG_ON_FOLIO(!folio_test_locked(src), src); |
960 | VM_BUG_ON_FOLIO(!folio_test_locked(dst), dst); | |
e24f0b8f | 961 | |
bda807d4 | 962 | if (likely(is_lru)) { |
68f2736a MWO |
963 | struct address_space *mapping = folio_mapping(src); |
964 | ||
bda807d4 | 965 | if (!mapping) |
54184650 | 966 | rc = migrate_folio(mapping, dst, src, mode); |
5490da4f | 967 | else if (mapping->a_ops->migrate_folio) |
bda807d4 | 968 | /* |
5490da4f MWO |
969 | * Most folios have a mapping and most filesystems |
970 | * provide a migrate_folio callback. Anonymous folios | |
bda807d4 | 971 | * are part of swap space which also has its own |
5490da4f | 972 | * migrate_folio callback. This is the most common path |
bda807d4 MK |
973 | * for page migration. |
974 | */ | |
5490da4f MWO |
975 | rc = mapping->a_ops->migrate_folio(mapping, dst, src, |
976 | mode); | |
bda807d4 | 977 | else |
8faa8ef5 | 978 | rc = fallback_migrate_folio(mapping, dst, src, mode); |
bda807d4 | 979 | } else { |
68f2736a MWO |
980 | const struct movable_operations *mops; |
981 | ||
e24f0b8f | 982 | /* |
bda807d4 MK |
983 | * In case of non-lru page, it could be released after |
984 | * isolation step. In that case, we shouldn't try migration. | |
e24f0b8f | 985 | */ |
e7e3ffeb MWO |
986 | VM_BUG_ON_FOLIO(!folio_test_isolated(src), src); |
987 | if (!folio_test_movable(src)) { | |
bda807d4 | 988 | rc = MIGRATEPAGE_SUCCESS; |
e7e3ffeb | 989 | folio_clear_isolated(src); |
bda807d4 MK |
990 | goto out; |
991 | } | |
992 | ||
68f2736a MWO |
993 | mops = page_movable_ops(&src->page); |
994 | rc = mops->migrate_page(&dst->page, &src->page, mode); | |
bda807d4 | 995 | WARN_ON_ONCE(rc == MIGRATEPAGE_SUCCESS && |
e7e3ffeb | 996 | !folio_test_isolated(src)); |
bda807d4 | 997 | } |
e24f0b8f | 998 | |
5c3f9a67 | 999 | /* |
e7e3ffeb MWO |
1000 | * When successful, old pagecache src->mapping must be cleared before |
1001 | * src is freed; but stats require that PageAnon be left as PageAnon. | |
5c3f9a67 HD |
1002 | */ |
1003 | if (rc == MIGRATEPAGE_SUCCESS) { | |
e7e3ffeb MWO |
1004 | if (__PageMovable(&src->page)) { |
1005 | VM_BUG_ON_FOLIO(!folio_test_isolated(src), src); | |
bda807d4 MK |
1006 | |
1007 | /* | |
1008 | * We clear PG_movable under page_lock so any compactor | |
1009 | * cannot try to migrate this page. | |
1010 | */ | |
e7e3ffeb | 1011 | folio_clear_isolated(src); |
bda807d4 MK |
1012 | } |
1013 | ||
1014 | /* | |
e7e3ffeb | 1015 | * Anonymous and movable src->mapping will be cleared by |
bda807d4 MK |
1016 | * free_pages_prepare so don't reset it here for keeping |
1017 | * the type to work PageAnon, for example. | |
1018 | */ | |
e7e3ffeb MWO |
1019 | if (!folio_mapping_flags(src)) |
1020 | src->mapping = NULL; | |
d2b2c6dd | 1021 | |
e7e3ffeb MWO |
1022 | if (likely(!folio_is_zone_device(dst))) |
1023 | flush_dcache_folio(dst); | |
3fe2011f | 1024 | } |
bda807d4 | 1025 | out: |
e24f0b8f CL |
1026 | return rc; |
1027 | } | |
1028 | ||
682a71a1 | 1029 | static int __unmap_and_move(struct folio *src, struct folio *dst, |
9c620e2b | 1030 | int force, enum migrate_mode mode) |
e24f0b8f | 1031 | { |
0dabec93 | 1032 | int rc = -EAGAIN; |
213ecb31 | 1033 | bool page_was_mapped = false; |
3f6c8272 | 1034 | struct anon_vma *anon_vma = NULL; |
682a71a1 | 1035 | bool is_lru = !__PageMovable(&src->page); |
95a402c3 | 1036 | |
682a71a1 | 1037 | if (!folio_trylock(src)) { |
a6bc32b8 | 1038 | if (!force || mode == MIGRATE_ASYNC) |
0dabec93 | 1039 | goto out; |
3e7d3449 MG |
1040 | |
1041 | /* | |
1042 | * It's not safe for direct compaction to call lock_page. | |
1043 | * For example, during page readahead pages are added locked | |
1044 | * to the LRU. Later, when the IO completes the pages are | |
1045 | * marked uptodate and unlocked. However, the queueing | |
1046 | * could be merging multiple pages for one bio (e.g. | |
d4388340 | 1047 | * mpage_readahead). If an allocation happens for the |
3e7d3449 MG |
1048 | * second or third page, the process can end up locking |
1049 | * the same page twice and deadlocking. Rather than | |
1050 | * trying to be clever about what pages can be locked, | |
1051 | * avoid the use of lock_page for direct compaction | |
1052 | * altogether. | |
1053 | */ | |
1054 | if (current->flags & PF_MEMALLOC) | |
0dabec93 | 1055 | goto out; |
3e7d3449 | 1056 | |
682a71a1 | 1057 | folio_lock(src); |
e24f0b8f CL |
1058 | } |
1059 | ||
682a71a1 | 1060 | if (folio_test_writeback(src)) { |
11bc82d6 | 1061 | /* |
fed5b64a | 1062 | * Only in the case of a full synchronous migration is it |
a6bc32b8 MG |
1063 | * necessary to wait for PageWriteback. In the async case, |
1064 | * the retry loop is too short and in the sync-light case, | |
1065 | * the overhead of stalling is too much | |
11bc82d6 | 1066 | */ |
2916ecc0 JG |
1067 | switch (mode) { |
1068 | case MIGRATE_SYNC: | |
1069 | case MIGRATE_SYNC_NO_COPY: | |
1070 | break; | |
1071 | default: | |
11bc82d6 | 1072 | rc = -EBUSY; |
0a31bc97 | 1073 | goto out_unlock; |
11bc82d6 AA |
1074 | } |
1075 | if (!force) | |
0a31bc97 | 1076 | goto out_unlock; |
682a71a1 | 1077 | folio_wait_writeback(src); |
e24f0b8f | 1078 | } |
03f15c86 | 1079 | |
e24f0b8f | 1080 | /* |
682a71a1 MWO |
1081 | * By try_to_migrate(), src->mapcount goes down to 0 here. In this case, |
1082 | * we cannot notice that anon_vma is freed while we migrate a page. | |
1ce82b69 | 1083 | * This get_anon_vma() delays freeing anon_vma pointer until the end |
dc386d4d | 1084 | * of migration. File cache pages are no problem because of page_lock() |
989f89c5 KH |
1085 | * File Caches may use write_page() or lock_page() in migration, then, |
1086 | * just care Anon page here. | |
03f15c86 | 1087 | * |
29eea9b5 | 1088 | * Only folio_get_anon_vma() understands the subtleties of |
03f15c86 HD |
1089 | * getting a hold on an anon_vma from outside one of its mms. |
1090 | * But if we cannot get anon_vma, then we won't need it anyway, | |
1091 | * because that implies that the anon page is no longer mapped | |
1092 | * (and cannot be remapped so long as we hold the page lock). | |
dc386d4d | 1093 | */ |
682a71a1 | 1094 | if (folio_test_anon(src) && !folio_test_ksm(src)) |
29eea9b5 | 1095 | anon_vma = folio_get_anon_vma(src); |
62e1c553 | 1096 | |
7db7671f HD |
1097 | /* |
1098 | * Block others from accessing the new page when we get around to | |
1099 | * establishing additional references. We are usually the only one | |
682a71a1 MWO |
1100 | * holding a reference to dst at this point. We used to have a BUG |
1101 | * here if folio_trylock(dst) fails, but would like to allow for | |
1102 | * cases where there might be a race with the previous use of dst. | |
7db7671f HD |
1103 | * This is much like races on refcount of oldpage: just don't BUG(). |
1104 | */ | |
682a71a1 | 1105 | if (unlikely(!folio_trylock(dst))) |
7db7671f HD |
1106 | goto out_unlock; |
1107 | ||
bda807d4 | 1108 | if (unlikely(!is_lru)) { |
682a71a1 | 1109 | rc = move_to_new_folio(dst, src, mode); |
bda807d4 MK |
1110 | goto out_unlock_both; |
1111 | } | |
1112 | ||
dc386d4d | 1113 | /* |
62e1c553 SL |
1114 | * Corner case handling: |
1115 | * 1. When a new swap-cache page is read into, it is added to the LRU | |
1116 | * and treated as swapcache but it has no rmap yet. | |
682a71a1 | 1117 | * Calling try_to_unmap() against a src->mapping==NULL page will |
62e1c553 | 1118 | * trigger a BUG. So handle it here. |
d12b8951 | 1119 | * 2. An orphaned page (see truncate_cleanup_page) might have |
62e1c553 SL |
1120 | * fs-private metadata. The page can be picked up due to memory |
1121 | * offlining. Everywhere else except page reclaim, the page is | |
1122 | * invisible to the vm, so the page can not be migrated. So try to | |
1123 | * free the metadata, so the page can be freed. | |
e24f0b8f | 1124 | */ |
682a71a1 MWO |
1125 | if (!src->mapping) { |
1126 | if (folio_test_private(src)) { | |
1127 | try_to_free_buffers(src); | |
7db7671f | 1128 | goto out_unlock_both; |
62e1c553 | 1129 | } |
682a71a1 | 1130 | } else if (folio_mapped(src)) { |
7db7671f | 1131 | /* Establish migration ptes */ |
682a71a1 MWO |
1132 | VM_BUG_ON_FOLIO(folio_test_anon(src) && |
1133 | !folio_test_ksm(src) && !anon_vma, src); | |
1134 | try_to_migrate(src, 0); | |
213ecb31 | 1135 | page_was_mapped = true; |
2ebba6b7 | 1136 | } |
dc386d4d | 1137 | |
682a71a1 MWO |
1138 | if (!folio_mapped(src)) |
1139 | rc = move_to_new_folio(dst, src, mode); | |
e24f0b8f | 1140 | |
c3096e67 | 1141 | /* |
682a71a1 | 1142 | * When successful, push dst to LRU immediately: so that if it |
c3096e67 | 1143 | * turns out to be an mlocked page, remove_migration_ptes() will |
682a71a1 | 1144 | * automatically build up the correct dst->mlock_count for it. |
c3096e67 HD |
1145 | * |
1146 | * We would like to do something similar for the old page, when | |
1147 | * unsuccessful, and other cases when a page has been temporarily | |
1148 | * isolated from the unevictable LRU: but this case is the easiest. | |
1149 | */ | |
1150 | if (rc == MIGRATEPAGE_SUCCESS) { | |
682a71a1 | 1151 | folio_add_lru(dst); |
c3096e67 HD |
1152 | if (page_was_mapped) |
1153 | lru_add_drain(); | |
1154 | } | |
1155 | ||
5c3f9a67 | 1156 | if (page_was_mapped) |
682a71a1 MWO |
1157 | remove_migration_ptes(src, |
1158 | rc == MIGRATEPAGE_SUCCESS ? dst : src, false); | |
3f6c8272 | 1159 | |
7db7671f | 1160 | out_unlock_both: |
682a71a1 | 1161 | folio_unlock(dst); |
7db7671f | 1162 | out_unlock: |
3f6c8272 | 1163 | /* Drop an anon_vma reference if we took one */ |
76545066 | 1164 | if (anon_vma) |
9e60109f | 1165 | put_anon_vma(anon_vma); |
682a71a1 | 1166 | folio_unlock(src); |
0dabec93 | 1167 | out: |
c6c919eb | 1168 | /* |
682a71a1 | 1169 | * If migration is successful, decrease refcount of dst, |
c6c919eb | 1170 | * which will not free the page because new page owner increased |
c3096e67 | 1171 | * refcounter. |
c6c919eb | 1172 | */ |
c3096e67 | 1173 | if (rc == MIGRATEPAGE_SUCCESS) |
682a71a1 | 1174 | folio_put(dst); |
c6c919eb | 1175 | |
0dabec93 MK |
1176 | return rc; |
1177 | } | |
95a402c3 | 1178 | |
0dabec93 | 1179 | /* |
49f51859 HY |
1180 | * Obtain the lock on folio, remove all ptes and migrate the folio |
1181 | * to the newly allocated folio in dst. | |
0dabec93 | 1182 | */ |
6ec4476a | 1183 | static int unmap_and_move(new_page_t get_new_page, |
ef2a5153 | 1184 | free_page_t put_new_page, |
49f51859 | 1185 | unsigned long private, struct folio *src, |
add05cec | 1186 | int force, enum migrate_mode mode, |
dd4ae78a YS |
1187 | enum migrate_reason reason, |
1188 | struct list_head *ret) | |
0dabec93 | 1189 | { |
49f51859 | 1190 | struct folio *dst; |
2def7424 | 1191 | int rc = MIGRATEPAGE_SUCCESS; |
74d4a579 | 1192 | struct page *newpage = NULL; |
0dabec93 | 1193 | |
49f51859 | 1194 | if (!thp_migration_supported() && folio_test_transhuge(src)) |
d532e2e5 | 1195 | return -ENOSYS; |
94723aaf | 1196 | |
49f51859 HY |
1197 | if (folio_ref_count(src) == 1) { |
1198 | /* Folio was freed from under us. So we are done. */ | |
1199 | folio_clear_active(src); | |
1200 | folio_clear_unevictable(src); | |
160088b3 | 1201 | /* free_pages_prepare() will clear PG_isolated. */ |
0dabec93 MK |
1202 | goto out; |
1203 | } | |
1204 | ||
49f51859 | 1205 | newpage = get_new_page(&src->page, private); |
74d4a579 YS |
1206 | if (!newpage) |
1207 | return -ENOMEM; | |
682a71a1 | 1208 | dst = page_folio(newpage); |
74d4a579 | 1209 | |
4c74b65f | 1210 | dst->private = NULL; |
682a71a1 | 1211 | rc = __unmap_and_move(src, dst, force, mode); |
c6c919eb | 1212 | if (rc == MIGRATEPAGE_SUCCESS) |
49f51859 | 1213 | set_page_owner_migrate_reason(&dst->page, reason); |
bf6bddf1 | 1214 | |
0dabec93 | 1215 | out: |
e24f0b8f | 1216 | if (rc != -EAGAIN) { |
0dabec93 | 1217 | /* |
49f51859 HY |
1218 | * A folio that has been migrated has all references |
1219 | * removed and will be freed. A folio that has not been | |
c23a0c99 | 1220 | * migrated will have kept its references and be restored. |
0dabec93 | 1221 | */ |
49f51859 | 1222 | list_del(&src->lru); |
dd4ae78a | 1223 | } |
6afcf8ef | 1224 | |
dd4ae78a YS |
1225 | /* |
1226 | * If migration is successful, releases reference grabbed during | |
49f51859 | 1227 | * isolation. Otherwise, restore the folio to right list unless |
dd4ae78a YS |
1228 | * we want to retry. |
1229 | */ | |
1230 | if (rc == MIGRATEPAGE_SUCCESS) { | |
6afcf8ef | 1231 | /* |
49f51859 | 1232 | * Compaction can migrate also non-LRU folios which are |
6afcf8ef | 1233 | * not accounted to NR_ISOLATED_*. They can be recognized |
49f51859 | 1234 | * as __folio_test_movable |
6afcf8ef | 1235 | */ |
49f51859 HY |
1236 | if (likely(!__folio_test_movable(src))) |
1237 | mod_node_page_state(folio_pgdat(src), NR_ISOLATED_ANON + | |
1238 | folio_is_file_lru(src), -folio_nr_pages(src)); | |
c6c919eb | 1239 | |
79f5f8fa | 1240 | if (reason != MR_MEMORY_FAILURE) |
d7e69488 | 1241 | /* |
49f51859 | 1242 | * We release the folio in page_handle_poison. |
d7e69488 | 1243 | */ |
49f51859 | 1244 | folio_put(src); |
c6c919eb | 1245 | } else { |
dd4ae78a | 1246 | if (rc != -EAGAIN) |
49f51859 | 1247 | list_add_tail(&src->lru, ret); |
bda807d4 | 1248 | |
c6c919eb | 1249 | if (put_new_page) |
49f51859 | 1250 | put_new_page(&dst->page, private); |
c6c919eb | 1251 | else |
49f51859 | 1252 | folio_put(dst); |
e24f0b8f | 1253 | } |
68711a74 | 1254 | |
e24f0b8f CL |
1255 | return rc; |
1256 | } | |
1257 | ||
290408d4 NH |
1258 | /* |
1259 | * Counterpart of unmap_and_move_page() for hugepage migration. | |
1260 | * | |
1261 | * This function doesn't wait the completion of hugepage I/O | |
1262 | * because there is no race between I/O and migration for hugepage. | |
1263 | * Note that currently hugepage I/O occurs only in direct I/O | |
1264 | * where no lock is held and PG_writeback is irrelevant, | |
1265 | * and writeback status of all subpages are counted in the reference | |
1266 | * count of the head page (i.e. if all subpages of a 2MB hugepage are | |
1267 | * under direct I/O, the reference of the head page is 512 and a bit more.) | |
1268 | * This means that when we try to migrate hugepage whose subpages are | |
1269 | * doing direct I/O, some references remain after try_to_unmap() and | |
1270 | * hugepage migration fails without data corruption. | |
1271 | * | |
1272 | * There is also no race when direct I/O is issued on the page under migration, | |
1273 | * because then pte is replaced with migration swap entry and direct I/O code | |
1274 | * will wait in the page fault for migration to complete. | |
1275 | */ | |
1276 | static int unmap_and_move_huge_page(new_page_t get_new_page, | |
68711a74 DR |
1277 | free_page_t put_new_page, unsigned long private, |
1278 | struct page *hpage, int force, | |
dd4ae78a YS |
1279 | enum migrate_mode mode, int reason, |
1280 | struct list_head *ret) | |
290408d4 | 1281 | { |
4eecb8b9 | 1282 | struct folio *dst, *src = page_folio(hpage); |
2def7424 | 1283 | int rc = -EAGAIN; |
2ebba6b7 | 1284 | int page_was_mapped = 0; |
32665f2b | 1285 | struct page *new_hpage; |
290408d4 | 1286 | struct anon_vma *anon_vma = NULL; |
c0d0381a | 1287 | struct address_space *mapping = NULL; |
290408d4 | 1288 | |
83467efb | 1289 | /* |
7ed2c31d | 1290 | * Migratability of hugepages depends on architectures and their size. |
83467efb NH |
1291 | * This check is necessary because some callers of hugepage migration |
1292 | * like soft offline and memory hotremove don't walk through page | |
1293 | * tables or check whether the hugepage is pmd-based or not before | |
1294 | * kicking migration. | |
1295 | */ | |
577be05c | 1296 | if (!hugepage_migration_supported(page_hstate(hpage))) |
83467efb NH |
1297 | return -ENOSYS; |
1298 | ||
c33db292 | 1299 | if (folio_ref_count(src) == 1) { |
71a64f61 MS |
1300 | /* page was freed from under us. So we are done. */ |
1301 | putback_active_hugepage(hpage); | |
1302 | return MIGRATEPAGE_SUCCESS; | |
1303 | } | |
1304 | ||
666feb21 | 1305 | new_hpage = get_new_page(hpage, private); |
290408d4 NH |
1306 | if (!new_hpage) |
1307 | return -ENOMEM; | |
4eecb8b9 | 1308 | dst = page_folio(new_hpage); |
290408d4 | 1309 | |
c33db292 | 1310 | if (!folio_trylock(src)) { |
2916ecc0 | 1311 | if (!force) |
290408d4 | 1312 | goto out; |
2916ecc0 JG |
1313 | switch (mode) { |
1314 | case MIGRATE_SYNC: | |
1315 | case MIGRATE_SYNC_NO_COPY: | |
1316 | break; | |
1317 | default: | |
1318 | goto out; | |
1319 | } | |
c33db292 | 1320 | folio_lock(src); |
290408d4 NH |
1321 | } |
1322 | ||
cb6acd01 MK |
1323 | /* |
1324 | * Check for pages which are in the process of being freed. Without | |
c33db292 | 1325 | * folio_mapping() set, hugetlbfs specific move page routine will not |
cb6acd01 MK |
1326 | * be called and we could leak usage counts for subpools. |
1327 | */ | |
345c62d1 | 1328 | if (hugetlb_folio_subpool(src) && !folio_mapping(src)) { |
cb6acd01 MK |
1329 | rc = -EBUSY; |
1330 | goto out_unlock; | |
1331 | } | |
1332 | ||
c33db292 | 1333 | if (folio_test_anon(src)) |
29eea9b5 | 1334 | anon_vma = folio_get_anon_vma(src); |
290408d4 | 1335 | |
c33db292 | 1336 | if (unlikely(!folio_trylock(dst))) |
7db7671f HD |
1337 | goto put_anon; |
1338 | ||
c33db292 | 1339 | if (folio_mapped(src)) { |
a98a2f0c | 1340 | enum ttu_flags ttu = 0; |
336bf30e | 1341 | |
c33db292 | 1342 | if (!folio_test_anon(src)) { |
336bf30e MK |
1343 | /* |
1344 | * In shared mappings, try_to_unmap could potentially | |
1345 | * call huge_pmd_unshare. Because of this, take | |
1346 | * semaphore in write mode here and set TTU_RMAP_LOCKED | |
1347 | * to let lower levels know we have taken the lock. | |
1348 | */ | |
1349 | mapping = hugetlb_page_mapping_lock_write(hpage); | |
1350 | if (unlikely(!mapping)) | |
1351 | goto unlock_put_anon; | |
1352 | ||
5202978b | 1353 | ttu = TTU_RMAP_LOCKED; |
336bf30e | 1354 | } |
c0d0381a | 1355 | |
4b8554c5 | 1356 | try_to_migrate(src, ttu); |
2ebba6b7 | 1357 | page_was_mapped = 1; |
336bf30e | 1358 | |
5202978b | 1359 | if (ttu & TTU_RMAP_LOCKED) |
336bf30e | 1360 | i_mmap_unlock_write(mapping); |
2ebba6b7 | 1361 | } |
290408d4 | 1362 | |
c33db292 | 1363 | if (!folio_mapped(src)) |
e7e3ffeb | 1364 | rc = move_to_new_folio(dst, src, mode); |
290408d4 | 1365 | |
336bf30e | 1366 | if (page_was_mapped) |
4eecb8b9 MWO |
1367 | remove_migration_ptes(src, |
1368 | rc == MIGRATEPAGE_SUCCESS ? dst : src, false); | |
290408d4 | 1369 | |
c0d0381a | 1370 | unlock_put_anon: |
c33db292 | 1371 | folio_unlock(dst); |
7db7671f HD |
1372 | |
1373 | put_anon: | |
fd4a4663 | 1374 | if (anon_vma) |
9e60109f | 1375 | put_anon_vma(anon_vma); |
8e6ac7fa | 1376 | |
2def7424 | 1377 | if (rc == MIGRATEPAGE_SUCCESS) { |
345c62d1 | 1378 | move_hugetlb_state(src, dst, reason); |
2def7424 HD |
1379 | put_new_page = NULL; |
1380 | } | |
8e6ac7fa | 1381 | |
cb6acd01 | 1382 | out_unlock: |
c33db292 | 1383 | folio_unlock(src); |
09761333 | 1384 | out: |
dd4ae78a | 1385 | if (rc == MIGRATEPAGE_SUCCESS) |
b8ec1cee | 1386 | putback_active_hugepage(hpage); |
a04840c6 | 1387 | else if (rc != -EAGAIN) |
c33db292 | 1388 | list_move_tail(&src->lru, ret); |
68711a74 DR |
1389 | |
1390 | /* | |
1391 | * If migration was not successful and there's a freeing callback, use | |
1392 | * it. Otherwise, put_page() will drop the reference grabbed during | |
1393 | * isolation. | |
1394 | */ | |
2def7424 | 1395 | if (put_new_page) |
68711a74 DR |
1396 | put_new_page(new_hpage, private); |
1397 | else | |
3aaa76e1 | 1398 | putback_active_hugepage(new_hpage); |
68711a74 | 1399 | |
290408d4 NH |
1400 | return rc; |
1401 | } | |
1402 | ||
eaec4e63 | 1403 | static inline int try_split_folio(struct folio *folio, struct list_head *split_folios) |
d532e2e5 | 1404 | { |
9c62ff00 | 1405 | int rc; |
d532e2e5 | 1406 | |
eaec4e63 HY |
1407 | folio_lock(folio); |
1408 | rc = split_folio_to_list(folio, split_folios); | |
1409 | folio_unlock(folio); | |
e6fa8a79 | 1410 | if (!rc) |
eaec4e63 | 1411 | list_move_tail(&folio->lru, split_folios); |
d532e2e5 YS |
1412 | |
1413 | return rc; | |
1414 | } | |
1415 | ||
b20a3503 | 1416 | /* |
eaec4e63 | 1417 | * migrate_pages - migrate the folios specified in a list, to the free folios |
c73e5c9c | 1418 | * supplied as the target for the page migration |
b20a3503 | 1419 | * |
eaec4e63 HY |
1420 | * @from: The list of folios to be migrated. |
1421 | * @get_new_page: The function used to allocate free folios to be used | |
1422 | * as the target of the folio migration. | |
1423 | * @put_new_page: The function used to free target folios if migration | |
68711a74 | 1424 | * fails, or NULL if no special handling is necessary. |
c73e5c9c SB |
1425 | * @private: Private data to be passed on to get_new_page() |
1426 | * @mode: The migration mode that specifies the constraints for | |
eaec4e63 HY |
1427 | * folio migration, if any. |
1428 | * @reason: The reason for folio migration. | |
1429 | * @ret_succeeded: Set to the number of folios migrated successfully if | |
5ac95884 | 1430 | * the caller passes a non-NULL pointer. |
b20a3503 | 1431 | * |
eaec4e63 HY |
1432 | * The function returns after 10 attempts or if no folios are movable any more |
1433 | * because the list has become empty or no retryable folios exist any more. | |
1434 | * It is caller's responsibility to call putback_movable_pages() to return folios | |
dd4ae78a | 1435 | * to the LRU or free list only if ret != 0. |
b20a3503 | 1436 | * |
eaec4e63 HY |
1437 | * Returns the number of {normal folio, large folio, hugetlb} that were not |
1438 | * migrated, or an error code. The number of large folio splits will be | |
1439 | * considered as the number of non-migrated large folio, no matter how many | |
1440 | * split folios of the large folio are migrated successfully. | |
b20a3503 | 1441 | */ |
9c620e2b | 1442 | int migrate_pages(struct list_head *from, new_page_t get_new_page, |
68711a74 | 1443 | free_page_t put_new_page, unsigned long private, |
5ac95884 | 1444 | enum migrate_mode mode, int reason, unsigned int *ret_succeeded) |
b20a3503 | 1445 | { |
e24f0b8f | 1446 | int retry = 1; |
eaec4e63 | 1447 | int large_retry = 1; |
1a5bae25 | 1448 | int thp_retry = 1; |
b20a3503 | 1449 | int nr_failed = 0; |
b5bade97 | 1450 | int nr_failed_pages = 0; |
077309bc | 1451 | int nr_retry_pages = 0; |
5647bc29 | 1452 | int nr_succeeded = 0; |
1a5bae25 | 1453 | int nr_thp_succeeded = 0; |
eaec4e63 | 1454 | int nr_large_failed = 0; |
1a5bae25 AK |
1455 | int nr_thp_failed = 0; |
1456 | int nr_thp_split = 0; | |
b20a3503 | 1457 | int pass = 0; |
eaec4e63 | 1458 | bool is_large = false; |
1a5bae25 | 1459 | bool is_thp = false; |
eaec4e63 HY |
1460 | struct folio *folio, *folio2; |
1461 | int rc, nr_pages; | |
1462 | LIST_HEAD(ret_folios); | |
1463 | LIST_HEAD(split_folios); | |
b0b515bf | 1464 | bool nosplit = (reason == MR_NUMA_MISPLACED); |
eaec4e63 | 1465 | bool no_split_folio_counting = false; |
b20a3503 | 1466 | |
7bc1aec5 LM |
1467 | trace_mm_migrate_pages_start(mode, reason); |
1468 | ||
eaec4e63 HY |
1469 | split_folio_migration: |
1470 | for (pass = 0; pass < 10 && (retry || large_retry); pass++) { | |
e24f0b8f | 1471 | retry = 0; |
eaec4e63 | 1472 | large_retry = 0; |
1a5bae25 | 1473 | thp_retry = 0; |
077309bc | 1474 | nr_retry_pages = 0; |
b20a3503 | 1475 | |
eaec4e63 | 1476 | list_for_each_entry_safe(folio, folio2, from, lru) { |
1a5bae25 | 1477 | /* |
eaec4e63 HY |
1478 | * Large folio statistics is based on the source large |
1479 | * folio. Capture required information that might get | |
1480 | * lost during migration. | |
1a5bae25 | 1481 | */ |
eaec4e63 HY |
1482 | is_large = folio_test_large(folio) && !folio_test_hugetlb(folio); |
1483 | is_thp = is_large && folio_test_pmd_mappable(folio); | |
1484 | nr_pages = folio_nr_pages(folio); | |
e24f0b8f | 1485 | cond_resched(); |
2d1db3b1 | 1486 | |
eaec4e63 | 1487 | if (folio_test_hugetlb(folio)) |
31caf665 | 1488 | rc = unmap_and_move_huge_page(get_new_page, |
eaec4e63 HY |
1489 | put_new_page, private, |
1490 | &folio->page, pass > 2, mode, | |
1491 | reason, | |
1492 | &ret_folios); | |
31caf665 | 1493 | else |
68711a74 | 1494 | rc = unmap_and_move(get_new_page, put_new_page, |
eaec4e63 HY |
1495 | private, folio, pass > 2, mode, |
1496 | reason, &ret_folios); | |
dd4ae78a YS |
1497 | /* |
1498 | * The rules are: | |
eaec4e63 HY |
1499 | * Success: non hugetlb folio will be freed, hugetlb |
1500 | * folio will be put back | |
dd4ae78a YS |
1501 | * -EAGAIN: stay on the from list |
1502 | * -ENOMEM: stay on the from list | |
577be05c | 1503 | * -ENOSYS: stay on the from list |
eaec4e63 | 1504 | * Other errno: put on ret_folios list then splice to |
dd4ae78a YS |
1505 | * from list |
1506 | */ | |
e24f0b8f | 1507 | switch(rc) { |
d532e2e5 | 1508 | /* |
eaec4e63 HY |
1509 | * Large folio migration might be unsupported or |
1510 | * the allocation could've failed so we should retry | |
1511 | * on the same folio with the large folio split | |
1512 | * to normal folios. | |
d532e2e5 | 1513 | * |
eaec4e63 | 1514 | * Split folios are put in split_folios, and |
e6fa8a79 HY |
1515 | * we will migrate them after the rest of the |
1516 | * list is processed. | |
d532e2e5 YS |
1517 | */ |
1518 | case -ENOSYS: | |
eaec4e63 HY |
1519 | /* Large folio migration is unsupported */ |
1520 | if (is_large) { | |
1521 | nr_large_failed++; | |
1522 | nr_thp_failed += is_thp; | |
1523 | if (!try_split_folio(folio, &split_folios)) { | |
1524 | nr_thp_split += is_thp; | |
e6fa8a79 | 1525 | break; |
d532e2e5 | 1526 | } |
d532e2e5 | 1527 | /* Hugetlb migration is unsupported */ |
eaec4e63 | 1528 | } else if (!no_split_folio_counting) { |
b5bade97 | 1529 | nr_failed++; |
f430893b ML |
1530 | } |
1531 | ||
eaec4e63 HY |
1532 | nr_failed_pages += nr_pages; |
1533 | list_move_tail(&folio->lru, &ret_folios); | |
d532e2e5 | 1534 | break; |
95a402c3 | 1535 | case -ENOMEM: |
94723aaf | 1536 | /* |
d532e2e5 | 1537 | * When memory is low, don't bother to try to migrate |
eaec4e63 | 1538 | * other folios, just exit. |
94723aaf | 1539 | */ |
eaec4e63 HY |
1540 | if (is_large) { |
1541 | nr_large_failed++; | |
1542 | nr_thp_failed += is_thp; | |
1543 | /* Large folio NUMA faulting doesn't split to retry. */ | |
fd4a7ac3 | 1544 | if (!nosplit) { |
eaec4e63 | 1545 | int ret = try_split_folio(folio, &split_folios); |
fd4a7ac3 BW |
1546 | |
1547 | if (!ret) { | |
eaec4e63 | 1548 | nr_thp_split += is_thp; |
fd4a7ac3 BW |
1549 | break; |
1550 | } else if (reason == MR_LONGTERM_PIN && | |
1551 | ret == -EAGAIN) { | |
1552 | /* | |
eaec4e63 HY |
1553 | * Try again to split large folio to |
1554 | * mitigate the failure of longterm pinning. | |
fd4a7ac3 | 1555 | */ |
eaec4e63 HY |
1556 | large_retry++; |
1557 | thp_retry += is_thp; | |
1558 | nr_retry_pages += nr_pages; | |
fd4a7ac3 BW |
1559 | break; |
1560 | } | |
94723aaf | 1561 | } |
eaec4e63 | 1562 | } else if (!no_split_folio_counting) { |
f430893b | 1563 | nr_failed++; |
1a5bae25 | 1564 | } |
b5bade97 | 1565 | |
eaec4e63 | 1566 | nr_failed_pages += nr_pages + nr_retry_pages; |
69a041ff | 1567 | /* |
eaec4e63 HY |
1568 | * There might be some split folios of fail-to-migrate large |
1569 | * folios left in split_folios list. Move them back to migration | |
69a041ff | 1570 | * list so that they could be put back to the right list by |
eaec4e63 | 1571 | * the caller otherwise the folio refcnt will be leaked. |
69a041ff | 1572 | */ |
eaec4e63 | 1573 | list_splice_init(&split_folios, from); |
fbed53b4 | 1574 | /* nr_failed isn't updated for not used */ |
eaec4e63 | 1575 | nr_large_failed += large_retry; |
69a041ff | 1576 | nr_thp_failed += thp_retry; |
95a402c3 | 1577 | goto out; |
e24f0b8f | 1578 | case -EAGAIN: |
eaec4e63 HY |
1579 | if (is_large) { |
1580 | large_retry++; | |
1581 | thp_retry += is_thp; | |
1582 | } else if (!no_split_folio_counting) { | |
f430893b | 1583 | retry++; |
eaec4e63 HY |
1584 | } |
1585 | nr_retry_pages += nr_pages; | |
e24f0b8f | 1586 | break; |
78bd5209 | 1587 | case MIGRATEPAGE_SUCCESS: |
eaec4e63 HY |
1588 | nr_succeeded += nr_pages; |
1589 | nr_thp_succeeded += is_thp; | |
e24f0b8f CL |
1590 | break; |
1591 | default: | |
354a3363 | 1592 | /* |
d532e2e5 | 1593 | * Permanent failure (-EBUSY, etc.): |
eaec4e63 HY |
1594 | * unlike -EAGAIN case, the failed folio is |
1595 | * removed from migration folio list and not | |
354a3363 NH |
1596 | * retried in the next outer loop. |
1597 | */ | |
eaec4e63 HY |
1598 | if (is_large) { |
1599 | nr_large_failed++; | |
1600 | nr_thp_failed += is_thp; | |
1601 | } else if (!no_split_folio_counting) { | |
b5bade97 | 1602 | nr_failed++; |
eaec4e63 | 1603 | } |
f430893b | 1604 | |
eaec4e63 | 1605 | nr_failed_pages += nr_pages; |
e24f0b8f | 1606 | break; |
2d1db3b1 | 1607 | } |
b20a3503 CL |
1608 | } |
1609 | } | |
7047b5a4 | 1610 | nr_failed += retry; |
eaec4e63 | 1611 | nr_large_failed += large_retry; |
1a5bae25 | 1612 | nr_thp_failed += thp_retry; |
077309bc | 1613 | nr_failed_pages += nr_retry_pages; |
b5bade97 | 1614 | /* |
eaec4e63 HY |
1615 | * Try to migrate split folios of fail-to-migrate large folios, no |
1616 | * nr_failed counting in this round, since all split folios of a | |
1617 | * large folio is counted as 1 failure in the first round. | |
b5bade97 | 1618 | */ |
eaec4e63 | 1619 | if (!list_empty(&split_folios)) { |
b5bade97 | 1620 | /* |
eaec4e63 | 1621 | * Move non-migrated folios (after 10 retries) to ret_folios |
b5bade97 BW |
1622 | * to avoid migrating them again. |
1623 | */ | |
eaec4e63 HY |
1624 | list_splice_init(from, &ret_folios); |
1625 | list_splice_init(&split_folios, from); | |
1626 | no_split_folio_counting = true; | |
b5bade97 | 1627 | retry = 1; |
eaec4e63 | 1628 | goto split_folio_migration; |
b5bade97 BW |
1629 | } |
1630 | ||
eaec4e63 | 1631 | rc = nr_failed + nr_large_failed; |
95a402c3 | 1632 | out: |
dd4ae78a | 1633 | /* |
eaec4e63 | 1634 | * Put the permanent failure folio back to migration list, they |
dd4ae78a YS |
1635 | * will be put back to the right list by the caller. |
1636 | */ | |
eaec4e63 | 1637 | list_splice(&ret_folios, from); |
dd4ae78a | 1638 | |
03e5f82e | 1639 | /* |
eaec4e63 HY |
1640 | * Return 0 in case all split folios of fail-to-migrate large folios |
1641 | * are migrated successfully. | |
03e5f82e BW |
1642 | */ |
1643 | if (list_empty(from)) | |
1644 | rc = 0; | |
1645 | ||
1a5bae25 | 1646 | count_vm_events(PGMIGRATE_SUCCESS, nr_succeeded); |
b5bade97 | 1647 | count_vm_events(PGMIGRATE_FAIL, nr_failed_pages); |
1a5bae25 AK |
1648 | count_vm_events(THP_MIGRATION_SUCCESS, nr_thp_succeeded); |
1649 | count_vm_events(THP_MIGRATION_FAIL, nr_thp_failed); | |
1650 | count_vm_events(THP_MIGRATION_SPLIT, nr_thp_split); | |
b5bade97 | 1651 | trace_mm_migrate_pages(nr_succeeded, nr_failed_pages, nr_thp_succeeded, |
1a5bae25 | 1652 | nr_thp_failed, nr_thp_split, mode, reason); |
7b2a2d4a | 1653 | |
5ac95884 YS |
1654 | if (ret_succeeded) |
1655 | *ret_succeeded = nr_succeeded; | |
1656 | ||
78bd5209 | 1657 | return rc; |
b20a3503 | 1658 | } |
95a402c3 | 1659 | |
19fc7bed | 1660 | struct page *alloc_migration_target(struct page *page, unsigned long private) |
b4b38223 | 1661 | { |
ffe06786 | 1662 | struct folio *folio = page_folio(page); |
19fc7bed JK |
1663 | struct migration_target_control *mtc; |
1664 | gfp_t gfp_mask; | |
b4b38223 | 1665 | unsigned int order = 0; |
e37d3e83 | 1666 | struct folio *hugetlb_folio = NULL; |
ffe06786 | 1667 | struct folio *new_folio = NULL; |
19fc7bed JK |
1668 | int nid; |
1669 | int zidx; | |
1670 | ||
1671 | mtc = (struct migration_target_control *)private; | |
1672 | gfp_mask = mtc->gfp_mask; | |
1673 | nid = mtc->nid; | |
1674 | if (nid == NUMA_NO_NODE) | |
ffe06786 | 1675 | nid = folio_nid(folio); |
b4b38223 | 1676 | |
ffe06786 | 1677 | if (folio_test_hugetlb(folio)) { |
e51da3a9 | 1678 | struct hstate *h = folio_hstate(folio); |
d92bbc27 | 1679 | |
19fc7bed | 1680 | gfp_mask = htlb_modify_alloc_mask(h, gfp_mask); |
e37d3e83 SK |
1681 | hugetlb_folio = alloc_hugetlb_folio_nodemask(h, nid, |
1682 | mtc->nmask, gfp_mask); | |
1683 | return &hugetlb_folio->page; | |
d92bbc27 | 1684 | } |
b4b38223 | 1685 | |
ffe06786 | 1686 | if (folio_test_large(folio)) { |
9933a0c8 JK |
1687 | /* |
1688 | * clear __GFP_RECLAIM to make the migration callback | |
1689 | * consistent with regular THP allocations. | |
1690 | */ | |
1691 | gfp_mask &= ~__GFP_RECLAIM; | |
b4b38223 | 1692 | gfp_mask |= GFP_TRANSHUGE; |
ffe06786 | 1693 | order = folio_order(folio); |
b4b38223 | 1694 | } |
ffe06786 | 1695 | zidx = zone_idx(folio_zone(folio)); |
19fc7bed | 1696 | if (is_highmem_idx(zidx) || zidx == ZONE_MOVABLE) |
b4b38223 JK |
1697 | gfp_mask |= __GFP_HIGHMEM; |
1698 | ||
ffe06786 | 1699 | new_folio = __folio_alloc(gfp_mask, order, nid, mtc->nmask); |
b4b38223 | 1700 | |
ffe06786 | 1701 | return &new_folio->page; |
b4b38223 JK |
1702 | } |
1703 | ||
742755a1 | 1704 | #ifdef CONFIG_NUMA |
742755a1 | 1705 | |
a49bd4d7 | 1706 | static int store_status(int __user *status, int start, int value, int nr) |
742755a1 | 1707 | { |
a49bd4d7 MH |
1708 | while (nr-- > 0) { |
1709 | if (put_user(value, status + start)) | |
1710 | return -EFAULT; | |
1711 | start++; | |
1712 | } | |
1713 | ||
1714 | return 0; | |
1715 | } | |
1716 | ||
1717 | static int do_move_pages_to_node(struct mm_struct *mm, | |
1718 | struct list_head *pagelist, int node) | |
1719 | { | |
1720 | int err; | |
a0976311 JK |
1721 | struct migration_target_control mtc = { |
1722 | .nid = node, | |
1723 | .gfp_mask = GFP_HIGHUSER_MOVABLE | __GFP_THISNODE, | |
1724 | }; | |
a49bd4d7 | 1725 | |
a0976311 | 1726 | err = migrate_pages(pagelist, alloc_migration_target, NULL, |
5ac95884 | 1727 | (unsigned long)&mtc, MIGRATE_SYNC, MR_SYSCALL, NULL); |
a49bd4d7 MH |
1728 | if (err) |
1729 | putback_movable_pages(pagelist); | |
1730 | return err; | |
742755a1 CL |
1731 | } |
1732 | ||
1733 | /* | |
a49bd4d7 MH |
1734 | * Resolves the given address to a struct page, isolates it from the LRU and |
1735 | * puts it to the given pagelist. | |
e0153fc2 YS |
1736 | * Returns: |
1737 | * errno - if the page cannot be found/isolated | |
1738 | * 0 - when it doesn't have to be migrated because it is already on the | |
1739 | * target node | |
1740 | * 1 - when it has been queued | |
742755a1 | 1741 | */ |
a49bd4d7 MH |
1742 | static int add_page_for_migration(struct mm_struct *mm, unsigned long addr, |
1743 | int node, struct list_head *pagelist, bool migrate_all) | |
742755a1 | 1744 | { |
a49bd4d7 MH |
1745 | struct vm_area_struct *vma; |
1746 | struct page *page; | |
742755a1 | 1747 | int err; |
742755a1 | 1748 | |
d8ed45c5 | 1749 | mmap_read_lock(mm); |
a49bd4d7 | 1750 | err = -EFAULT; |
cb1c37b1 ML |
1751 | vma = vma_lookup(mm, addr); |
1752 | if (!vma || !vma_migratable(vma)) | |
a49bd4d7 | 1753 | goto out; |
742755a1 | 1754 | |
a49bd4d7 | 1755 | /* FOLL_DUMP to ignore special (like zero) pages */ |
87d2762e | 1756 | page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP); |
89f5b7da | 1757 | |
a49bd4d7 MH |
1758 | err = PTR_ERR(page); |
1759 | if (IS_ERR(page)) | |
1760 | goto out; | |
89f5b7da | 1761 | |
a49bd4d7 | 1762 | err = -ENOENT; |
f7091ed6 | 1763 | if (!page) |
a49bd4d7 | 1764 | goto out; |
742755a1 | 1765 | |
f7091ed6 HW |
1766 | if (is_zone_device_page(page)) |
1767 | goto out_putpage; | |
1768 | ||
a49bd4d7 MH |
1769 | err = 0; |
1770 | if (page_to_nid(page) == node) | |
1771 | goto out_putpage; | |
742755a1 | 1772 | |
a49bd4d7 MH |
1773 | err = -EACCES; |
1774 | if (page_mapcount(page) > 1 && !migrate_all) | |
1775 | goto out_putpage; | |
742755a1 | 1776 | |
a49bd4d7 MH |
1777 | if (PageHuge(page)) { |
1778 | if (PageHead(page)) { | |
6aa3a920 | 1779 | err = isolate_hugetlb(page_folio(page), pagelist); |
7ce82f4c ML |
1780 | if (!err) |
1781 | err = 1; | |
e632a938 | 1782 | } |
a49bd4d7 MH |
1783 | } else { |
1784 | struct page *head; | |
e632a938 | 1785 | |
e8db67eb NH |
1786 | head = compound_head(page); |
1787 | err = isolate_lru_page(head); | |
cf608ac1 | 1788 | if (err) |
a49bd4d7 | 1789 | goto out_putpage; |
742755a1 | 1790 | |
e0153fc2 | 1791 | err = 1; |
a49bd4d7 MH |
1792 | list_add_tail(&head->lru, pagelist); |
1793 | mod_node_page_state(page_pgdat(head), | |
9de4f22a | 1794 | NR_ISOLATED_ANON + page_is_file_lru(head), |
6c357848 | 1795 | thp_nr_pages(head)); |
a49bd4d7 MH |
1796 | } |
1797 | out_putpage: | |
1798 | /* | |
1799 | * Either remove the duplicate refcount from | |
1800 | * isolate_lru_page() or drop the page ref if it was | |
1801 | * not isolated. | |
1802 | */ | |
1803 | put_page(page); | |
1804 | out: | |
d8ed45c5 | 1805 | mmap_read_unlock(mm); |
742755a1 CL |
1806 | return err; |
1807 | } | |
1808 | ||
7ca8783a WY |
1809 | static int move_pages_and_store_status(struct mm_struct *mm, int node, |
1810 | struct list_head *pagelist, int __user *status, | |
1811 | int start, int i, unsigned long nr_pages) | |
1812 | { | |
1813 | int err; | |
1814 | ||
5d7ae891 WY |
1815 | if (list_empty(pagelist)) |
1816 | return 0; | |
1817 | ||
7ca8783a WY |
1818 | err = do_move_pages_to_node(mm, pagelist, node); |
1819 | if (err) { | |
1820 | /* | |
1821 | * Positive err means the number of failed | |
1822 | * pages to migrate. Since we are going to | |
1823 | * abort and return the number of non-migrated | |
ab9dd4f8 | 1824 | * pages, so need to include the rest of the |
7ca8783a WY |
1825 | * nr_pages that have not been attempted as |
1826 | * well. | |
1827 | */ | |
1828 | if (err > 0) | |
a7504ed1 | 1829 | err += nr_pages - i; |
7ca8783a WY |
1830 | return err; |
1831 | } | |
1832 | return store_status(status, start, node, i - start); | |
1833 | } | |
1834 | ||
5e9a0f02 BG |
1835 | /* |
1836 | * Migrate an array of page address onto an array of nodes and fill | |
1837 | * the corresponding array of status. | |
1838 | */ | |
3268c63e | 1839 | static int do_pages_move(struct mm_struct *mm, nodemask_t task_nodes, |
5e9a0f02 BG |
1840 | unsigned long nr_pages, |
1841 | const void __user * __user *pages, | |
1842 | const int __user *nodes, | |
1843 | int __user *status, int flags) | |
1844 | { | |
a49bd4d7 MH |
1845 | int current_node = NUMA_NO_NODE; |
1846 | LIST_HEAD(pagelist); | |
1847 | int start, i; | |
1848 | int err = 0, err1; | |
35282a2d | 1849 | |
361a2a22 | 1850 | lru_cache_disable(); |
35282a2d | 1851 | |
a49bd4d7 MH |
1852 | for (i = start = 0; i < nr_pages; i++) { |
1853 | const void __user *p; | |
1854 | unsigned long addr; | |
1855 | int node; | |
3140a227 | 1856 | |
a49bd4d7 MH |
1857 | err = -EFAULT; |
1858 | if (get_user(p, pages + i)) | |
1859 | goto out_flush; | |
1860 | if (get_user(node, nodes + i)) | |
1861 | goto out_flush; | |
057d3389 | 1862 | addr = (unsigned long)untagged_addr(p); |
a49bd4d7 MH |
1863 | |
1864 | err = -ENODEV; | |
1865 | if (node < 0 || node >= MAX_NUMNODES) | |
1866 | goto out_flush; | |
1867 | if (!node_state(node, N_MEMORY)) | |
1868 | goto out_flush; | |
5e9a0f02 | 1869 | |
a49bd4d7 MH |
1870 | err = -EACCES; |
1871 | if (!node_isset(node, task_nodes)) | |
1872 | goto out_flush; | |
1873 | ||
1874 | if (current_node == NUMA_NO_NODE) { | |
1875 | current_node = node; | |
1876 | start = i; | |
1877 | } else if (node != current_node) { | |
7ca8783a WY |
1878 | err = move_pages_and_store_status(mm, current_node, |
1879 | &pagelist, status, start, i, nr_pages); | |
a49bd4d7 MH |
1880 | if (err) |
1881 | goto out; | |
1882 | start = i; | |
1883 | current_node = node; | |
3140a227 BG |
1884 | } |
1885 | ||
a49bd4d7 MH |
1886 | /* |
1887 | * Errors in the page lookup or isolation are not fatal and we simply | |
1888 | * report them via status | |
1889 | */ | |
1890 | err = add_page_for_migration(mm, addr, current_node, | |
1891 | &pagelist, flags & MPOL_MF_MOVE_ALL); | |
e0153fc2 | 1892 | |
d08221a0 | 1893 | if (err > 0) { |
e0153fc2 YS |
1894 | /* The page is successfully queued for migration */ |
1895 | continue; | |
1896 | } | |
3140a227 | 1897 | |
65462462 JH |
1898 | /* |
1899 | * The move_pages() man page does not have an -EEXIST choice, so | |
1900 | * use -EFAULT instead. | |
1901 | */ | |
1902 | if (err == -EEXIST) | |
1903 | err = -EFAULT; | |
1904 | ||
d08221a0 WY |
1905 | /* |
1906 | * If the page is already on the target node (!err), store the | |
1907 | * node, otherwise, store the err. | |
1908 | */ | |
1909 | err = store_status(status, i, err ? : current_node, 1); | |
a49bd4d7 MH |
1910 | if (err) |
1911 | goto out_flush; | |
5e9a0f02 | 1912 | |
7ca8783a WY |
1913 | err = move_pages_and_store_status(mm, current_node, &pagelist, |
1914 | status, start, i, nr_pages); | |
a7504ed1 HY |
1915 | if (err) { |
1916 | /* We have accounted for page i */ | |
1917 | if (err > 0) | |
1918 | err--; | |
4afdacec | 1919 | goto out; |
a7504ed1 | 1920 | } |
a49bd4d7 | 1921 | current_node = NUMA_NO_NODE; |
3140a227 | 1922 | } |
a49bd4d7 MH |
1923 | out_flush: |
1924 | /* Make sure we do not overwrite the existing error */ | |
7ca8783a WY |
1925 | err1 = move_pages_and_store_status(mm, current_node, &pagelist, |
1926 | status, start, i, nr_pages); | |
dfe9aa23 | 1927 | if (err >= 0) |
a49bd4d7 | 1928 | err = err1; |
5e9a0f02 | 1929 | out: |
361a2a22 | 1930 | lru_cache_enable(); |
5e9a0f02 BG |
1931 | return err; |
1932 | } | |
1933 | ||
742755a1 | 1934 | /* |
2f007e74 | 1935 | * Determine the nodes of an array of pages and store it in an array of status. |
742755a1 | 1936 | */ |
80bba129 BG |
1937 | static void do_pages_stat_array(struct mm_struct *mm, unsigned long nr_pages, |
1938 | const void __user **pages, int *status) | |
742755a1 | 1939 | { |
2f007e74 | 1940 | unsigned long i; |
2f007e74 | 1941 | |
d8ed45c5 | 1942 | mmap_read_lock(mm); |
742755a1 | 1943 | |
2f007e74 | 1944 | for (i = 0; i < nr_pages; i++) { |
80bba129 | 1945 | unsigned long addr = (unsigned long)(*pages); |
742755a1 CL |
1946 | struct vm_area_struct *vma; |
1947 | struct page *page; | |
c095adbc | 1948 | int err = -EFAULT; |
2f007e74 | 1949 | |
059b8b48 LH |
1950 | vma = vma_lookup(mm, addr); |
1951 | if (!vma) | |
742755a1 CL |
1952 | goto set_status; |
1953 | ||
d899844e | 1954 | /* FOLL_DUMP to ignore special (like zero) pages */ |
16fd6b31 | 1955 | page = follow_page(vma, addr, FOLL_GET | FOLL_DUMP); |
89f5b7da LT |
1956 | |
1957 | err = PTR_ERR(page); | |
1958 | if (IS_ERR(page)) | |
1959 | goto set_status; | |
1960 | ||
f7091ed6 HW |
1961 | err = -ENOENT; |
1962 | if (!page) | |
1963 | goto set_status; | |
1964 | ||
1965 | if (!is_zone_device_page(page)) | |
4cd61484 | 1966 | err = page_to_nid(page); |
f7091ed6 | 1967 | |
16fd6b31 | 1968 | put_page(page); |
742755a1 | 1969 | set_status: |
80bba129 BG |
1970 | *status = err; |
1971 | ||
1972 | pages++; | |
1973 | status++; | |
1974 | } | |
1975 | ||
d8ed45c5 | 1976 | mmap_read_unlock(mm); |
80bba129 BG |
1977 | } |
1978 | ||
5b1b561b AB |
1979 | static int get_compat_pages_array(const void __user *chunk_pages[], |
1980 | const void __user * __user *pages, | |
1981 | unsigned long chunk_nr) | |
1982 | { | |
1983 | compat_uptr_t __user *pages32 = (compat_uptr_t __user *)pages; | |
1984 | compat_uptr_t p; | |
1985 | int i; | |
1986 | ||
1987 | for (i = 0; i < chunk_nr; i++) { | |
1988 | if (get_user(p, pages32 + i)) | |
1989 | return -EFAULT; | |
1990 | chunk_pages[i] = compat_ptr(p); | |
1991 | } | |
1992 | ||
1993 | return 0; | |
1994 | } | |
1995 | ||
80bba129 BG |
1996 | /* |
1997 | * Determine the nodes of a user array of pages and store it in | |
1998 | * a user array of status. | |
1999 | */ | |
2000 | static int do_pages_stat(struct mm_struct *mm, unsigned long nr_pages, | |
2001 | const void __user * __user *pages, | |
2002 | int __user *status) | |
2003 | { | |
3eefb826 | 2004 | #define DO_PAGES_STAT_CHUNK_NR 16UL |
80bba129 BG |
2005 | const void __user *chunk_pages[DO_PAGES_STAT_CHUNK_NR]; |
2006 | int chunk_status[DO_PAGES_STAT_CHUNK_NR]; | |
80bba129 | 2007 | |
87b8d1ad | 2008 | while (nr_pages) { |
3eefb826 | 2009 | unsigned long chunk_nr = min(nr_pages, DO_PAGES_STAT_CHUNK_NR); |
87b8d1ad | 2010 | |
5b1b561b AB |
2011 | if (in_compat_syscall()) { |
2012 | if (get_compat_pages_array(chunk_pages, pages, | |
2013 | chunk_nr)) | |
2014 | break; | |
2015 | } else { | |
2016 | if (copy_from_user(chunk_pages, pages, | |
2017 | chunk_nr * sizeof(*chunk_pages))) | |
2018 | break; | |
2019 | } | |
80bba129 BG |
2020 | |
2021 | do_pages_stat_array(mm, chunk_nr, chunk_pages, chunk_status); | |
2022 | ||
87b8d1ad PA |
2023 | if (copy_to_user(status, chunk_status, chunk_nr * sizeof(*status))) |
2024 | break; | |
742755a1 | 2025 | |
87b8d1ad PA |
2026 | pages += chunk_nr; |
2027 | status += chunk_nr; | |
2028 | nr_pages -= chunk_nr; | |
2029 | } | |
2030 | return nr_pages ? -EFAULT : 0; | |
742755a1 CL |
2031 | } |
2032 | ||
4dc200ce | 2033 | static struct mm_struct *find_mm_struct(pid_t pid, nodemask_t *mem_nodes) |
742755a1 | 2034 | { |
742755a1 | 2035 | struct task_struct *task; |
742755a1 | 2036 | struct mm_struct *mm; |
742755a1 | 2037 | |
4dc200ce ML |
2038 | /* |
2039 | * There is no need to check if current process has the right to modify | |
2040 | * the specified process when they are same. | |
2041 | */ | |
2042 | if (!pid) { | |
2043 | mmget(current->mm); | |
2044 | *mem_nodes = cpuset_mems_allowed(current); | |
2045 | return current->mm; | |
2046 | } | |
742755a1 CL |
2047 | |
2048 | /* Find the mm_struct */ | |
a879bf58 | 2049 | rcu_read_lock(); |
4dc200ce | 2050 | task = find_task_by_vpid(pid); |
742755a1 | 2051 | if (!task) { |
a879bf58 | 2052 | rcu_read_unlock(); |
4dc200ce | 2053 | return ERR_PTR(-ESRCH); |
742755a1 | 2054 | } |
3268c63e | 2055 | get_task_struct(task); |
742755a1 CL |
2056 | |
2057 | /* | |
2058 | * Check if this process has the right to modify the specified | |
197e7e52 | 2059 | * process. Use the regular "ptrace_may_access()" checks. |
742755a1 | 2060 | */ |
197e7e52 | 2061 | if (!ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS)) { |
c69e8d9c | 2062 | rcu_read_unlock(); |
4dc200ce | 2063 | mm = ERR_PTR(-EPERM); |
5e9a0f02 | 2064 | goto out; |
742755a1 | 2065 | } |
c69e8d9c | 2066 | rcu_read_unlock(); |
742755a1 | 2067 | |
4dc200ce ML |
2068 | mm = ERR_PTR(security_task_movememory(task)); |
2069 | if (IS_ERR(mm)) | |
5e9a0f02 | 2070 | goto out; |
4dc200ce | 2071 | *mem_nodes = cpuset_mems_allowed(task); |
3268c63e | 2072 | mm = get_task_mm(task); |
4dc200ce | 2073 | out: |
3268c63e | 2074 | put_task_struct(task); |
6e8b09ea | 2075 | if (!mm) |
4dc200ce ML |
2076 | mm = ERR_PTR(-EINVAL); |
2077 | return mm; | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * Move a list of pages in the address space of the currently executing | |
2082 | * process. | |
2083 | */ | |
2084 | static int kernel_move_pages(pid_t pid, unsigned long nr_pages, | |
2085 | const void __user * __user *pages, | |
2086 | const int __user *nodes, | |
2087 | int __user *status, int flags) | |
2088 | { | |
2089 | struct mm_struct *mm; | |
2090 | int err; | |
2091 | nodemask_t task_nodes; | |
2092 | ||
2093 | /* Check flags */ | |
2094 | if (flags & ~(MPOL_MF_MOVE|MPOL_MF_MOVE_ALL)) | |
6e8b09ea SL |
2095 | return -EINVAL; |
2096 | ||
4dc200ce ML |
2097 | if ((flags & MPOL_MF_MOVE_ALL) && !capable(CAP_SYS_NICE)) |
2098 | return -EPERM; | |
2099 | ||
2100 | mm = find_mm_struct(pid, &task_nodes); | |
2101 | if (IS_ERR(mm)) | |
2102 | return PTR_ERR(mm); | |
2103 | ||
6e8b09ea SL |
2104 | if (nodes) |
2105 | err = do_pages_move(mm, task_nodes, nr_pages, pages, | |
2106 | nodes, status, flags); | |
2107 | else | |
2108 | err = do_pages_stat(mm, nr_pages, pages, status); | |
742755a1 | 2109 | |
742755a1 CL |
2110 | mmput(mm); |
2111 | return err; | |
2112 | } | |
742755a1 | 2113 | |
7addf443 DB |
2114 | SYSCALL_DEFINE6(move_pages, pid_t, pid, unsigned long, nr_pages, |
2115 | const void __user * __user *, pages, | |
2116 | const int __user *, nodes, | |
2117 | int __user *, status, int, flags) | |
2118 | { | |
2119 | return kernel_move_pages(pid, nr_pages, pages, nodes, status, flags); | |
2120 | } | |
2121 | ||
7039e1db PZ |
2122 | #ifdef CONFIG_NUMA_BALANCING |
2123 | /* | |
2124 | * Returns true if this is a safe migration target node for misplaced NUMA | |
bc53008e | 2125 | * pages. Currently it only checks the watermarks which is crude. |
7039e1db PZ |
2126 | */ |
2127 | static bool migrate_balanced_pgdat(struct pglist_data *pgdat, | |
3abef4e6 | 2128 | unsigned long nr_migrate_pages) |
7039e1db PZ |
2129 | { |
2130 | int z; | |
599d0c95 | 2131 | |
7039e1db PZ |
2132 | for (z = pgdat->nr_zones - 1; z >= 0; z--) { |
2133 | struct zone *zone = pgdat->node_zones + z; | |
2134 | ||
bc53008e | 2135 | if (!managed_zone(zone)) |
7039e1db PZ |
2136 | continue; |
2137 | ||
7039e1db PZ |
2138 | /* Avoid waking kswapd by allocating pages_to_migrate pages. */ |
2139 | if (!zone_watermark_ok(zone, 0, | |
2140 | high_wmark_pages(zone) + | |
2141 | nr_migrate_pages, | |
bfe9d006 | 2142 | ZONE_MOVABLE, 0)) |
7039e1db PZ |
2143 | continue; |
2144 | return true; | |
2145 | } | |
2146 | return false; | |
2147 | } | |
2148 | ||
2149 | static struct page *alloc_misplaced_dst_page(struct page *page, | |
666feb21 | 2150 | unsigned long data) |
7039e1db PZ |
2151 | { |
2152 | int nid = (int) data; | |
c185e494 MWO |
2153 | int order = compound_order(page); |
2154 | gfp_t gfp = __GFP_THISNODE; | |
2155 | struct folio *new; | |
2156 | ||
2157 | if (order > 0) | |
2158 | gfp |= GFP_TRANSHUGE_LIGHT; | |
2159 | else { | |
2160 | gfp |= GFP_HIGHUSER_MOVABLE | __GFP_NOMEMALLOC | __GFP_NORETRY | | |
2161 | __GFP_NOWARN; | |
2162 | gfp &= ~__GFP_RECLAIM; | |
2163 | } | |
2164 | new = __folio_alloc_node(gfp, order, nid); | |
c5b5a3dd | 2165 | |
c185e494 | 2166 | return &new->page; |
c5b5a3dd YS |
2167 | } |
2168 | ||
1c30e017 | 2169 | static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page) |
b32967ff | 2170 | { |
2b9b624f | 2171 | int nr_pages = thp_nr_pages(page); |
c574bbe9 | 2172 | int order = compound_order(page); |
a8f60772 | 2173 | |
c574bbe9 | 2174 | VM_BUG_ON_PAGE(order && !PageTransHuge(page), page); |
3abef4e6 | 2175 | |
662aeea7 YS |
2176 | /* Do not migrate THP mapped by multiple processes */ |
2177 | if (PageTransHuge(page) && total_mapcount(page) > 1) | |
2178 | return 0; | |
2179 | ||
7039e1db | 2180 | /* Avoid migrating to a node that is nearly full */ |
c574bbe9 HY |
2181 | if (!migrate_balanced_pgdat(pgdat, nr_pages)) { |
2182 | int z; | |
2183 | ||
2184 | if (!(sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING)) | |
2185 | return 0; | |
2186 | for (z = pgdat->nr_zones - 1; z >= 0; z--) { | |
bc53008e | 2187 | if (managed_zone(pgdat->node_zones + z)) |
c574bbe9 HY |
2188 | break; |
2189 | } | |
2190 | wakeup_kswapd(pgdat->node_zones + z, 0, order, ZONE_MOVABLE); | |
340ef390 | 2191 | return 0; |
c574bbe9 | 2192 | } |
7039e1db | 2193 | |
340ef390 HD |
2194 | if (isolate_lru_page(page)) |
2195 | return 0; | |
7039e1db | 2196 | |
b75454e1 | 2197 | mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + page_is_file_lru(page), |
2b9b624f | 2198 | nr_pages); |
340ef390 | 2199 | |
149c33e1 | 2200 | /* |
340ef390 HD |
2201 | * Isolating the page has taken another reference, so the |
2202 | * caller's reference can be safely dropped without the page | |
2203 | * disappearing underneath us during migration. | |
149c33e1 MG |
2204 | */ |
2205 | put_page(page); | |
340ef390 | 2206 | return 1; |
b32967ff MG |
2207 | } |
2208 | ||
2209 | /* | |
2210 | * Attempt to migrate a misplaced page to the specified destination | |
2211 | * node. Caller is expected to have an elevated reference count on | |
2212 | * the page that will be dropped by this function before returning. | |
2213 | */ | |
1bc115d8 MG |
2214 | int migrate_misplaced_page(struct page *page, struct vm_area_struct *vma, |
2215 | int node) | |
b32967ff MG |
2216 | { |
2217 | pg_data_t *pgdat = NODE_DATA(node); | |
340ef390 | 2218 | int isolated; |
b32967ff | 2219 | int nr_remaining; |
e39bb6be | 2220 | unsigned int nr_succeeded; |
b32967ff | 2221 | LIST_HEAD(migratepages); |
b5916c02 | 2222 | int nr_pages = thp_nr_pages(page); |
c5b5a3dd | 2223 | |
b32967ff | 2224 | /* |
1bc115d8 MG |
2225 | * Don't migrate file pages that are mapped in multiple processes |
2226 | * with execute permissions as they are probably shared libraries. | |
b32967ff | 2227 | */ |
7ee820ee ML |
2228 | if (page_mapcount(page) != 1 && page_is_file_lru(page) && |
2229 | (vma->vm_flags & VM_EXEC)) | |
b32967ff | 2230 | goto out; |
b32967ff | 2231 | |
09a913a7 MG |
2232 | /* |
2233 | * Also do not migrate dirty pages as not all filesystems can move | |
2234 | * dirty pages in MIGRATE_ASYNC mode which is a waste of cycles. | |
2235 | */ | |
9de4f22a | 2236 | if (page_is_file_lru(page) && PageDirty(page)) |
09a913a7 MG |
2237 | goto out; |
2238 | ||
b32967ff MG |
2239 | isolated = numamigrate_isolate_page(pgdat, page); |
2240 | if (!isolated) | |
2241 | goto out; | |
2242 | ||
2243 | list_add(&page->lru, &migratepages); | |
c185e494 MWO |
2244 | nr_remaining = migrate_pages(&migratepages, alloc_misplaced_dst_page, |
2245 | NULL, node, MIGRATE_ASYNC, | |
2246 | MR_NUMA_MISPLACED, &nr_succeeded); | |
b32967ff | 2247 | if (nr_remaining) { |
59c82b70 JK |
2248 | if (!list_empty(&migratepages)) { |
2249 | list_del(&page->lru); | |
c5fc5c3a YS |
2250 | mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + |
2251 | page_is_file_lru(page), -nr_pages); | |
59c82b70 JK |
2252 | putback_lru_page(page); |
2253 | } | |
b32967ff | 2254 | isolated = 0; |
e39bb6be HY |
2255 | } |
2256 | if (nr_succeeded) { | |
2257 | count_vm_numa_events(NUMA_PAGE_MIGRATE, nr_succeeded); | |
2258 | if (!node_is_toptier(page_to_nid(page)) && node_is_toptier(node)) | |
2259 | mod_node_page_state(pgdat, PGPROMOTE_SUCCESS, | |
2260 | nr_succeeded); | |
2261 | } | |
7039e1db | 2262 | BUG_ON(!list_empty(&migratepages)); |
7039e1db | 2263 | return isolated; |
340ef390 HD |
2264 | |
2265 | out: | |
2266 | put_page(page); | |
2267 | return 0; | |
7039e1db | 2268 | } |
220018d3 | 2269 | #endif /* CONFIG_NUMA_BALANCING */ |
91952440 | 2270 | #endif /* CONFIG_NUMA */ |