Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/mm/swap_state.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie | |
7 | * | |
8 | * Rewritten to use page cache, (C) 1998 Stephen Tweedie | |
9 | */ | |
1da177e4 | 10 | #include <linux/mm.h> |
5a0e3ad6 | 11 | #include <linux/gfp.h> |
1da177e4 | 12 | #include <linux/kernel_stat.h> |
ddc1a5cb | 13 | #include <linux/mempolicy.h> |
1da177e4 | 14 | #include <linux/swap.h> |
46017e95 | 15 | #include <linux/swapops.h> |
1da177e4 LT |
16 | #include <linux/init.h> |
17 | #include <linux/pagemap.h> | |
4907e80b | 18 | #include <linux/pagevec.h> |
1da177e4 | 19 | #include <linux/backing-dev.h> |
3fb5c298 | 20 | #include <linux/blkdev.h> |
b20a3503 | 21 | #include <linux/migrate.h> |
4b3ef9da | 22 | #include <linux/vmalloc.h> |
67afa38e | 23 | #include <linux/swap_slots.h> |
38d8b4e6 | 24 | #include <linux/huge_mm.h> |
61ef1865 | 25 | #include <linux/shmem_fs.h> |
243bce09 | 26 | #include "internal.h" |
014bb1de | 27 | #include "swap.h" |
1da177e4 LT |
28 | |
29 | /* | |
30 | * swapper_space is a fiction, retained to simplify the path through | |
0ba5e806 | 31 | * vmscan's shrink_folio_list. |
1da177e4 | 32 | */ |
f5e54d6e | 33 | static const struct address_space_operations swap_aops = { |
1da177e4 | 34 | .writepage = swap_writepage, |
4c4a7634 | 35 | .dirty_folio = noop_dirty_folio, |
1c93923c | 36 | #ifdef CONFIG_MIGRATION |
54184650 | 37 | .migrate_folio = migrate_folio, |
1c93923c | 38 | #endif |
1da177e4 LT |
39 | }; |
40 | ||
783cb68e CD |
41 | struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly; |
42 | static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly; | |
f5c754d6 | 43 | static bool enable_vma_readahead __read_mostly = true; |
ec560175 | 44 | |
dce08dd2 HY |
45 | #define SWAP_RA_ORDER_CEILING 5 |
46 | ||
ec560175 HY |
47 | #define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2) |
48 | #define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1) | |
49 | #define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK | |
50 | #define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK) | |
51 | ||
52 | #define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK) | |
53 | #define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT) | |
54 | #define SWAP_RA_ADDR(v) ((v) & PAGE_MASK) | |
55 | ||
56 | #define SWAP_RA_VAL(addr, win, hits) \ | |
57 | (((addr) & PAGE_MASK) | \ | |
58 | (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \ | |
59 | ((hits) & SWAP_RA_HITS_MASK)) | |
60 | ||
61 | /* Initial readahead hits is 4 to start up with a small window */ | |
62 | #define GET_SWAP_RA_VAL(vma) \ | |
63 | (atomic_long_read(&(vma)->swap_readahead_info) ? : 4) | |
1da177e4 | 64 | |
579f8290 SL |
65 | static atomic_t swapin_readahead_hits = ATOMIC_INIT(4); |
66 | ||
1da177e4 LT |
67 | void show_swap_cache_info(void) |
68 | { | |
33806f06 | 69 | printk("%lu pages in swap cache\n", total_swapcache_pages()); |
3cb8eaa4 Z |
70 | printk("Free swap = %ldkB\n", K(get_nr_swap_pages())); |
71 | printk("Total swap = %lukB\n", K(total_swap_pages)); | |
1da177e4 LT |
72 | } |
73 | ||
aae466b0 JK |
74 | void *get_shadow_from_swap_cache(swp_entry_t entry) |
75 | { | |
76 | struct address_space *address_space = swap_address_space(entry); | |
7aad25b4 | 77 | pgoff_t idx = swap_cache_index(entry); |
4c773a44 | 78 | void *shadow; |
aae466b0 | 79 | |
4c773a44 MWO |
80 | shadow = xa_load(&address_space->i_pages, idx); |
81 | if (xa_is_value(shadow)) | |
82 | return shadow; | |
aae466b0 JK |
83 | return NULL; |
84 | } | |
85 | ||
1da177e4 | 86 | /* |
2bb876b5 | 87 | * add_to_swap_cache resembles filemap_add_folio on swapper_space, |
1da177e4 LT |
88 | * but sets SwapCache flag and private instead of mapping and index. |
89 | */ | |
a4c366f0 | 90 | int add_to_swap_cache(struct folio *folio, swp_entry_t entry, |
3852f676 | 91 | gfp_t gfp, void **shadowp) |
1da177e4 | 92 | { |
8d93b41c | 93 | struct address_space *address_space = swap_address_space(entry); |
7aad25b4 | 94 | pgoff_t idx = swap_cache_index(entry); |
a4c366f0 MWO |
95 | XA_STATE_ORDER(xas, &address_space->i_pages, idx, folio_order(folio)); |
96 | unsigned long i, nr = folio_nr_pages(folio); | |
3852f676 | 97 | void *old; |
1da177e4 | 98 | |
5649d113 YY |
99 | xas_set_update(&xas, workingset_update_node); |
100 | ||
a4c366f0 MWO |
101 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
102 | VM_BUG_ON_FOLIO(folio_test_swapcache(folio), folio); | |
103 | VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio); | |
51726b12 | 104 | |
a4c366f0 MWO |
105 | folio_ref_add(folio, nr); |
106 | folio_set_swapcache(folio); | |
3d2c9087 | 107 | folio->swap = entry; |
31a56396 | 108 | |
8d93b41c MW |
109 | do { |
110 | xas_lock_irq(&xas); | |
111 | xas_create_range(&xas); | |
112 | if (xas_error(&xas)) | |
113 | goto unlock; | |
114 | for (i = 0; i < nr; i++) { | |
a4c366f0 | 115 | VM_BUG_ON_FOLIO(xas.xa_index != idx + i, folio); |
e5b306a0 KS |
116 | if (shadowp) { |
117 | old = xas_load(&xas); | |
118 | if (xa_is_value(old)) | |
3852f676 JK |
119 | *shadowp = old; |
120 | } | |
a4c366f0 | 121 | xas_store(&xas, folio); |
8d93b41c MW |
122 | xas_next(&xas); |
123 | } | |
38d8b4e6 | 124 | address_space->nrpages += nr; |
a4c366f0 MWO |
125 | __node_stat_mod_folio(folio, NR_FILE_PAGES, nr); |
126 | __lruvec_stat_mod_folio(folio, NR_SWAPCACHE, nr); | |
8d93b41c MW |
127 | unlock: |
128 | xas_unlock_irq(&xas); | |
129 | } while (xas_nomem(&xas, gfp)); | |
31a56396 | 130 | |
8d93b41c MW |
131 | if (!xas_error(&xas)) |
132 | return 0; | |
31a56396 | 133 | |
a4c366f0 MWO |
134 | folio_clear_swapcache(folio); |
135 | folio_ref_sub(folio, nr); | |
8d93b41c | 136 | return xas_error(&xas); |
1da177e4 LT |
137 | } |
138 | ||
1da177e4 | 139 | /* |
ceff9d33 | 140 | * This must be called only on folios that have |
1da177e4 LT |
141 | * been verified to be in the swap cache. |
142 | */ | |
ceff9d33 | 143 | void __delete_from_swap_cache(struct folio *folio, |
3852f676 | 144 | swp_entry_t entry, void *shadow) |
1da177e4 | 145 | { |
4e17ec25 | 146 | struct address_space *address_space = swap_address_space(entry); |
ceff9d33 MWO |
147 | int i; |
148 | long nr = folio_nr_pages(folio); | |
7aad25b4 | 149 | pgoff_t idx = swap_cache_index(entry); |
4e17ec25 | 150 | XA_STATE(xas, &address_space->i_pages, idx); |
33806f06 | 151 | |
5649d113 YY |
152 | xas_set_update(&xas, workingset_update_node); |
153 | ||
ceff9d33 MWO |
154 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
155 | VM_BUG_ON_FOLIO(!folio_test_swapcache(folio), folio); | |
156 | VM_BUG_ON_FOLIO(folio_test_writeback(folio), folio); | |
1da177e4 | 157 | |
38d8b4e6 | 158 | for (i = 0; i < nr; i++) { |
3852f676 | 159 | void *entry = xas_store(&xas, shadow); |
b9eb7776 | 160 | VM_BUG_ON_PAGE(entry != folio, entry); |
4e17ec25 | 161 | xas_next(&xas); |
38d8b4e6 | 162 | } |
3d2c9087 | 163 | folio->swap.val = 0; |
ceff9d33 | 164 | folio_clear_swapcache(folio); |
38d8b4e6 | 165 | address_space->nrpages -= nr; |
ceff9d33 MWO |
166 | __node_stat_mod_folio(folio, NR_FILE_PAGES, -nr); |
167 | __lruvec_stat_mod_folio(folio, NR_SWAPCACHE, -nr); | |
1da177e4 LT |
168 | } |
169 | ||
170 | /** | |
09c02e56 MWO |
171 | * add_to_swap - allocate swap space for a folio |
172 | * @folio: folio we want to move to swap | |
1da177e4 | 173 | * |
09c02e56 MWO |
174 | * Allocate swap space for the folio and add the folio to the |
175 | * swap cache. | |
176 | * | |
177 | * Context: Caller needs to hold the folio lock. | |
178 | * Return: Whether the folio was added to the swap cache. | |
1da177e4 | 179 | */ |
09c02e56 | 180 | bool add_to_swap(struct folio *folio) |
1da177e4 LT |
181 | { |
182 | swp_entry_t entry; | |
1da177e4 LT |
183 | int err; |
184 | ||
09c02e56 MWO |
185 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
186 | VM_BUG_ON_FOLIO(!folio_test_uptodate(folio), folio); | |
1da177e4 | 187 | |
e2e3fdc7 | 188 | entry = folio_alloc_swap(folio); |
2ca4532a | 189 | if (!entry.val) |
09c02e56 | 190 | return false; |
0f074658 | 191 | |
2ca4532a | 192 | /* |
8d93b41c | 193 | * XArray node allocations from PF_MEMALLOC contexts could |
2ca4532a DN |
194 | * completely exhaust the page allocator. __GFP_NOMEMALLOC |
195 | * stops emergency reserves from being allocated. | |
196 | * | |
197 | * TODO: this could cause a theoretical memory reclaim | |
198 | * deadlock in the swap out path. | |
199 | */ | |
200 | /* | |
854e9ed0 | 201 | * Add it to the swap cache. |
2ca4532a | 202 | */ |
a4c366f0 | 203 | err = add_to_swap_cache(folio, entry, |
3852f676 | 204 | __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN, NULL); |
38d8b4e6 | 205 | if (err) |
bd53b714 | 206 | /* |
2ca4532a DN |
207 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely |
208 | * clear SWAP_HAS_CACHE flag. | |
1da177e4 | 209 | */ |
0f074658 | 210 | goto fail; |
9625456c | 211 | /* |
09c02e56 MWO |
212 | * Normally the folio will be dirtied in unmap because its |
213 | * pte should be dirty. A special case is MADV_FREE page. The | |
214 | * page's pte could have dirty bit cleared but the folio's | |
215 | * SwapBacked flag is still set because clearing the dirty bit | |
216 | * and SwapBacked flag has no lock protected. For such folio, | |
217 | * unmap will not set dirty bit for it, so folio reclaim will | |
218 | * not write the folio out. This can cause data corruption when | |
219 | * the folio is swapped in later. Always setting the dirty flag | |
220 | * for the folio solves the problem. | |
9625456c | 221 | */ |
09c02e56 | 222 | folio_mark_dirty(folio); |
38d8b4e6 | 223 | |
09c02e56 | 224 | return true; |
38d8b4e6 | 225 | |
38d8b4e6 | 226 | fail: |
4081f744 | 227 | put_swap_folio(folio, entry); |
09c02e56 | 228 | return false; |
1da177e4 LT |
229 | } |
230 | ||
231 | /* | |
75fa68a5 | 232 | * This must be called only on folios that have |
1da177e4 | 233 | * been verified to be in the swap cache and locked. |
75fa68a5 MWO |
234 | * It will never put the folio into the free list, |
235 | * the caller has a reference on the folio. | |
1da177e4 | 236 | */ |
75fa68a5 | 237 | void delete_from_swap_cache(struct folio *folio) |
1da177e4 | 238 | { |
3d2c9087 | 239 | swp_entry_t entry = folio->swap; |
4e17ec25 | 240 | struct address_space *address_space = swap_address_space(entry); |
1da177e4 | 241 | |
b93b0163 | 242 | xa_lock_irq(&address_space->i_pages); |
ceff9d33 | 243 | __delete_from_swap_cache(folio, entry, NULL); |
b93b0163 | 244 | xa_unlock_irq(&address_space->i_pages); |
1da177e4 | 245 | |
4081f744 | 246 | put_swap_folio(folio, entry); |
75fa68a5 | 247 | folio_ref_sub(folio, folio_nr_pages(folio)); |
1da177e4 LT |
248 | } |
249 | ||
3852f676 JK |
250 | void clear_shadow_from_swap_cache(int type, unsigned long begin, |
251 | unsigned long end) | |
252 | { | |
253 | unsigned long curr = begin; | |
254 | void *old; | |
255 | ||
256 | for (;;) { | |
3852f676 | 257 | swp_entry_t entry = swp_entry(type, curr); |
7aad25b4 | 258 | unsigned long index = curr & SWAP_ADDRESS_SPACE_MASK; |
3852f676 | 259 | struct address_space *address_space = swap_address_space(entry); |
7aad25b4 | 260 | XA_STATE(xas, &address_space->i_pages, index); |
3852f676 | 261 | |
5649d113 YY |
262 | xas_set_update(&xas, workingset_update_node); |
263 | ||
3852f676 | 264 | xa_lock_irq(&address_space->i_pages); |
7aad25b4 | 265 | xas_for_each(&xas, old, min(index + (end - curr), SWAP_ADDRESS_SPACE_PAGES)) { |
3852f676 JK |
266 | if (!xa_is_value(old)) |
267 | continue; | |
268 | xas_store(&xas, NULL); | |
3852f676 | 269 | } |
3852f676 JK |
270 | xa_unlock_irq(&address_space->i_pages); |
271 | ||
272 | /* search the next swapcache until we meet end */ | |
273 | curr >>= SWAP_ADDRESS_SPACE_SHIFT; | |
274 | curr++; | |
275 | curr <<= SWAP_ADDRESS_SPACE_SHIFT; | |
276 | if (curr > end) | |
277 | break; | |
278 | } | |
279 | } | |
280 | ||
c33c7948 RR |
281 | /* |
282 | * If we are the only user, then try to free up the swap cache. | |
283 | * | |
aedd74d4 | 284 | * Its ok to check the swapcache flag without the folio lock |
a2c43eed | 285 | * here because we are going to recheck again inside |
aedd74d4 | 286 | * folio_free_swap() _with_ the lock. |
1da177e4 LT |
287 | * - Marcelo |
288 | */ | |
63b77499 | 289 | void free_swap_cache(struct folio *folio) |
1da177e4 | 290 | { |
aedd74d4 MWO |
291 | if (folio_test_swapcache(folio) && !folio_mapped(folio) && |
292 | folio_trylock(folio)) { | |
293 | folio_free_swap(folio); | |
294 | folio_unlock(folio); | |
1da177e4 LT |
295 | } |
296 | } | |
297 | ||
c33c7948 | 298 | /* |
1da177e4 | 299 | * Perform a free_page(), also freeing any swap cache associated with |
b8072f09 | 300 | * this page if it is the last user of the page. |
1da177e4 LT |
301 | */ |
302 | void free_page_and_swap_cache(struct page *page) | |
303 | { | |
63b77499 MWO |
304 | struct folio *folio = page_folio(page); |
305 | ||
306 | free_swap_cache(folio); | |
5beaee54 | 307 | if (!is_huge_zero_folio(folio)) |
63b77499 | 308 | folio_put(folio); |
1da177e4 LT |
309 | } |
310 | ||
311 | /* | |
312 | * Passed an array of pages, drop them all from swapcache and then release | |
313 | * them. They are removed from the LRU and freed if this is their last use. | |
314 | */ | |
7cc8f9c7 | 315 | void free_pages_and_swap_cache(struct encoded_page **pages, int nr) |
1da177e4 | 316 | { |
4907e80b MWO |
317 | struct folio_batch folios; |
318 | unsigned int refs[PAGEVEC_SIZE]; | |
319 | ||
1da177e4 | 320 | lru_add_drain(); |
4907e80b | 321 | folio_batch_init(&folios); |
d7f861b9 | 322 | for (int i = 0; i < nr; i++) { |
4907e80b | 323 | struct folio *folio = page_folio(encoded_page_ptr(pages[i])); |
d7f861b9 | 324 | |
63b77499 | 325 | free_swap_cache(folio); |
4907e80b | 326 | refs[folios.nr] = 1; |
d7f861b9 DH |
327 | if (unlikely(encoded_page_flags(pages[i]) & |
328 | ENCODED_PAGE_BIT_NR_PAGES_NEXT)) | |
4907e80b | 329 | refs[folios.nr] = encoded_nr_pages(pages[++i]); |
d7f861b9 | 330 | |
4907e80b MWO |
331 | if (folio_batch_add(&folios, folio) == 0) |
332 | folios_put_refs(&folios, refs); | |
d7f861b9 | 333 | } |
4907e80b MWO |
334 | if (folios.nr) |
335 | folios_put_refs(&folios, refs); | |
1da177e4 LT |
336 | } |
337 | ||
e9e9b7ec MK |
338 | static inline bool swap_use_vma_readahead(void) |
339 | { | |
340 | return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap); | |
341 | } | |
342 | ||
1da177e4 | 343 | /* |
c9edc242 | 344 | * Lookup a swap entry in the swap cache. A found folio will be returned |
1da177e4 | 345 | * unlocked and with its refcount incremented - we rely on the kernel |
c9edc242 | 346 | * lock getting page table operations atomic even if we drop the folio |
1da177e4 | 347 | * lock before returning. |
cbc2bd98 KS |
348 | * |
349 | * Caller must lock the swap device or hold a reference to keep it valid. | |
1da177e4 | 350 | */ |
c9edc242 MWO |
351 | struct folio *swap_cache_get_folio(swp_entry_t entry, |
352 | struct vm_area_struct *vma, unsigned long addr) | |
1da177e4 | 353 | { |
c9edc242 | 354 | struct folio *folio; |
1da177e4 | 355 | |
7aad25b4 | 356 | folio = filemap_get_folio(swap_address_space(entry), swap_cache_index(entry)); |
66dabbb6 | 357 | if (!IS_ERR(folio)) { |
eaf649eb MK |
358 | bool vma_ra = swap_use_vma_readahead(); |
359 | bool readahead; | |
360 | ||
eaf649eb MK |
361 | /* |
362 | * At the moment, we don't support PG_readahead for anon THP | |
363 | * so let's bail out rather than confusing the readahead stat. | |
364 | */ | |
c9edc242 MWO |
365 | if (unlikely(folio_test_large(folio))) |
366 | return folio; | |
eaf649eb | 367 | |
c9edc242 | 368 | readahead = folio_test_clear_readahead(folio); |
eaf649eb MK |
369 | if (vma && vma_ra) { |
370 | unsigned long ra_val; | |
371 | int win, hits; | |
372 | ||
373 | ra_val = GET_SWAP_RA_VAL(vma); | |
374 | win = SWAP_RA_WIN(ra_val); | |
375 | hits = SWAP_RA_HITS(ra_val); | |
ec560175 HY |
376 | if (readahead) |
377 | hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX); | |
378 | atomic_long_set(&vma->swap_readahead_info, | |
379 | SWAP_RA_VAL(addr, win, hits)); | |
380 | } | |
eaf649eb | 381 | |
ec560175 | 382 | if (readahead) { |
cbc65df2 | 383 | count_vm_event(SWAP_RA_HIT); |
eaf649eb | 384 | if (!vma || !vma_ra) |
ec560175 | 385 | atomic_inc(&swapin_readahead_hits); |
cbc65df2 | 386 | } |
66dabbb6 CH |
387 | } else { |
388 | folio = NULL; | |
579f8290 | 389 | } |
eaf649eb | 390 | |
c9edc242 MWO |
391 | return folio; |
392 | } | |
393 | ||
61ef1865 | 394 | /** |
524984ff | 395 | * filemap_get_incore_folio - Find and get a folio from the page or swap caches. |
61ef1865 MWO |
396 | * @mapping: The address_space to search. |
397 | * @index: The page cache index. | |
398 | * | |
524984ff MWO |
399 | * This differs from filemap_get_folio() in that it will also look for the |
400 | * folio in the swap cache. | |
61ef1865 | 401 | * |
524984ff | 402 | * Return: The found folio or %NULL. |
61ef1865 | 403 | */ |
524984ff MWO |
404 | struct folio *filemap_get_incore_folio(struct address_space *mapping, |
405 | pgoff_t index) | |
61ef1865 MWO |
406 | { |
407 | swp_entry_t swp; | |
408 | struct swap_info_struct *si; | |
097b3e59 | 409 | struct folio *folio = filemap_get_entry(mapping, index); |
61ef1865 | 410 | |
66dabbb6 CH |
411 | if (!folio) |
412 | return ERR_PTR(-ENOENT); | |
dd8095b1 | 413 | if (!xa_is_value(folio)) |
66dabbb6 | 414 | return folio; |
61ef1865 | 415 | if (!shmem_mapping(mapping)) |
66dabbb6 | 416 | return ERR_PTR(-ENOENT); |
61ef1865 | 417 | |
dd8095b1 | 418 | swp = radix_to_swp_entry(folio); |
ba6851b4 ML |
419 | /* There might be swapin error entries in shmem mapping. */ |
420 | if (non_swap_entry(swp)) | |
66dabbb6 | 421 | return ERR_PTR(-ENOENT); |
61ef1865 MWO |
422 | /* Prevent swapoff from happening to us */ |
423 | si = get_swap_device(swp); | |
424 | if (!si) | |
66dabbb6 | 425 | return ERR_PTR(-ENOENT); |
7aad25b4 | 426 | index = swap_cache_index(swp); |
dd8095b1 | 427 | folio = filemap_get_folio(swap_address_space(swp), index); |
61ef1865 | 428 | put_swap_device(si); |
524984ff | 429 | return folio; |
61ef1865 MWO |
430 | } |
431 | ||
96c7b0b4 MWO |
432 | struct folio *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
433 | struct mempolicy *mpol, pgoff_t ilx, bool *new_page_allocated, | |
434 | bool skip_if_exists) | |
1da177e4 | 435 | { |
eb085574 | 436 | struct swap_info_struct *si; |
a0d3374b | 437 | struct folio *folio; |
aae466b0 | 438 | void *shadow = NULL; |
4c6355b2 | 439 | |
5b999aad | 440 | *new_page_allocated = false; |
46a774d3 HY |
441 | si = get_swap_device(entry); |
442 | if (!si) | |
443 | return NULL; | |
1da177e4 | 444 | |
4c6355b2 JW |
445 | for (;;) { |
446 | int err; | |
1da177e4 LT |
447 | /* |
448 | * First check the swap cache. Since this is normally | |
cb691e2f | 449 | * called after swap_cache_get_folio() failed, re-calling |
1da177e4 LT |
450 | * that would confuse statistics. |
451 | */ | |
a0d3374b | 452 | folio = filemap_get_folio(swap_address_space(entry), |
7aad25b4 | 453 | swap_cache_index(entry)); |
96c7b0b4 MWO |
454 | if (!IS_ERR(folio)) |
455 | goto got_folio; | |
1da177e4 | 456 | |
ba81f838 HY |
457 | /* |
458 | * Just skip read ahead for unused swap slot. | |
459 | * During swap_off when swap_slot_cache is disabled, | |
460 | * we have to handle the race between putting | |
461 | * swap entry in swap cache and marking swap slot | |
462 | * as SWAP_HAS_CACHE. That's done in later part of code or | |
463 | * else swap_off will be aborted if we return NULL. | |
464 | */ | |
3ecdeb0f | 465 | if (!swap_swapcount(si, entry) && swap_slot_cache_enabled) |
46a774d3 | 466 | goto fail_put_swap; |
e8c26ab6 | 467 | |
1da177e4 | 468 | /* |
96c7b0b4 | 469 | * Get a new folio to read into from swap. Allocate it now, |
4c6355b2 JW |
470 | * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will |
471 | * cause any racers to loop around until we add it to cache. | |
1da177e4 | 472 | */ |
2ef52d5b | 473 | folio = folio_alloc_mpol(gfp_mask, 0, mpol, ilx, numa_node_id()); |
a0d3374b | 474 | if (!folio) |
46a774d3 | 475 | goto fail_put_swap; |
1da177e4 | 476 | |
f000944d HD |
477 | /* |
478 | * Swap entry may have been freed since our caller observed it. | |
479 | */ | |
355cfa73 | 480 | err = swapcache_prepare(entry); |
4c6355b2 | 481 | if (!err) |
f000944d HD |
482 | break; |
483 | ||
a0d3374b | 484 | folio_put(folio); |
4c6355b2 | 485 | if (err != -EEXIST) |
46a774d3 | 486 | goto fail_put_swap; |
4c6355b2 | 487 | |
a65b0e76 DC |
488 | /* |
489 | * Protect against a recursive call to __read_swap_cache_async() | |
490 | * on the same entry waiting forever here because SWAP_HAS_CACHE | |
491 | * is set but the folio is not the swap cache yet. This can | |
492 | * happen today if mem_cgroup_swapin_charge_folio() below | |
493 | * triggers reclaim through zswap, which may call | |
494 | * __read_swap_cache_async() in the writeback path. | |
495 | */ | |
496 | if (skip_if_exists) | |
497 | goto fail_put_swap; | |
498 | ||
2ca4532a | 499 | /* |
4c6355b2 JW |
500 | * We might race against __delete_from_swap_cache(), and |
501 | * stumble across a swap_map entry whose SWAP_HAS_CACHE | |
502 | * has not yet been cleared. Or race against another | |
503 | * __read_swap_cache_async(), which has set SWAP_HAS_CACHE | |
96c7b0b4 | 504 | * in swap_map, but not yet added its folio to swap cache. |
2ca4532a | 505 | */ |
029c4628 | 506 | schedule_timeout_uninterruptible(1); |
4c6355b2 JW |
507 | } |
508 | ||
509 | /* | |
96c7b0b4 | 510 | * The swap entry is ours to swap in. Prepare the new folio. |
4c6355b2 JW |
511 | */ |
512 | ||
a0d3374b MWO |
513 | __folio_set_locked(folio); |
514 | __folio_set_swapbacked(folio); | |
4c6355b2 | 515 | |
65995918 | 516 | if (mem_cgroup_swapin_charge_folio(folio, NULL, gfp_mask, entry)) |
4c6355b2 | 517 | goto fail_unlock; |
4c6355b2 | 518 | |
0add0c77 | 519 | /* May fail (-ENOMEM) if XArray node allocation failed. */ |
a4c366f0 | 520 | if (add_to_swap_cache(folio, entry, gfp_mask & GFP_RECLAIM_MASK, &shadow)) |
4c6355b2 | 521 | goto fail_unlock; |
0add0c77 SB |
522 | |
523 | mem_cgroup_swapin_uncharge_swap(entry); | |
4c6355b2 | 524 | |
aae466b0 | 525 | if (shadow) |
a0d3374b | 526 | workingset_refault(folio, shadow); |
314b57fb | 527 | |
a0d3374b MWO |
528 | /* Caller will initiate read into locked folio */ |
529 | folio_add_lru(folio); | |
4c6355b2 | 530 | *new_page_allocated = true; |
96c7b0b4 | 531 | got_folio: |
46a774d3 | 532 | put_swap_device(si); |
96c7b0b4 | 533 | return folio; |
1da177e4 | 534 | |
4c6355b2 | 535 | fail_unlock: |
4081f744 | 536 | put_swap_folio(folio, entry); |
a0d3374b MWO |
537 | folio_unlock(folio); |
538 | folio_put(folio); | |
46a774d3 HY |
539 | fail_put_swap: |
540 | put_swap_device(si); | |
4c6355b2 | 541 | return NULL; |
1da177e4 | 542 | } |
46017e95 | 543 | |
5b999aad DS |
544 | /* |
545 | * Locate a page of swap in physical memory, reserving swap cache space | |
546 | * and reading the disk if it is not already cached. | |
547 | * A failure return means that either the page allocation failed or that | |
548 | * the swap entry is no longer in use. | |
46a774d3 HY |
549 | * |
550 | * get/put_swap_device() aren't needed to call this function, because | |
c9bdf768 | 551 | * __read_swap_cache_async() call them and swap_read_folio() holds the |
46a774d3 | 552 | * swap cache folio lock. |
5b999aad | 553 | */ |
6e03492e MWO |
554 | struct folio *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
555 | struct vm_area_struct *vma, unsigned long addr, | |
556 | struct swap_iocb **plug) | |
5b999aad | 557 | { |
ddc1a5cb HD |
558 | bool page_allocated; |
559 | struct mempolicy *mpol; | |
560 | pgoff_t ilx; | |
96c7b0b4 | 561 | struct folio *folio; |
5b999aad | 562 | |
ddc1a5cb | 563 | mpol = get_vma_policy(vma, addr, 0, &ilx); |
96c7b0b4 | 564 | folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx, |
a65b0e76 | 565 | &page_allocated, false); |
ddc1a5cb | 566 | mpol_cond_put(mpol); |
5b999aad | 567 | |
ddc1a5cb | 568 | if (page_allocated) |
b2d1f38b | 569 | swap_read_folio(folio, plug); |
6e03492e | 570 | return folio; |
5b999aad DS |
571 | } |
572 | ||
ec560175 HY |
573 | static unsigned int __swapin_nr_pages(unsigned long prev_offset, |
574 | unsigned long offset, | |
575 | int hits, | |
576 | int max_pages, | |
577 | int prev_win) | |
579f8290 | 578 | { |
ec560175 | 579 | unsigned int pages, last_ra; |
579f8290 SL |
580 | |
581 | /* | |
582 | * This heuristic has been found to work well on both sequential and | |
583 | * random loads, swapping to hard disk or to SSD: please don't ask | |
584 | * what the "+ 2" means, it just happens to work well, that's all. | |
585 | */ | |
ec560175 | 586 | pages = hits + 2; |
579f8290 SL |
587 | if (pages == 2) { |
588 | /* | |
589 | * We can have no readahead hits to judge by: but must not get | |
590 | * stuck here forever, so check for an adjacent offset instead | |
591 | * (and don't even bother to check whether swap type is same). | |
592 | */ | |
593 | if (offset != prev_offset + 1 && offset != prev_offset - 1) | |
594 | pages = 1; | |
579f8290 SL |
595 | } else { |
596 | unsigned int roundup = 4; | |
597 | while (roundup < pages) | |
598 | roundup <<= 1; | |
599 | pages = roundup; | |
600 | } | |
601 | ||
602 | if (pages > max_pages) | |
603 | pages = max_pages; | |
604 | ||
605 | /* Don't shrink readahead too fast */ | |
ec560175 | 606 | last_ra = prev_win / 2; |
579f8290 SL |
607 | if (pages < last_ra) |
608 | pages = last_ra; | |
ec560175 HY |
609 | |
610 | return pages; | |
611 | } | |
612 | ||
613 | static unsigned long swapin_nr_pages(unsigned long offset) | |
614 | { | |
615 | static unsigned long prev_offset; | |
616 | unsigned int hits, pages, max_pages; | |
617 | static atomic_t last_readahead_pages; | |
618 | ||
619 | max_pages = 1 << READ_ONCE(page_cluster); | |
620 | if (max_pages <= 1) | |
621 | return 1; | |
622 | ||
623 | hits = atomic_xchg(&swapin_readahead_hits, 0); | |
d6c1f098 QC |
624 | pages = __swapin_nr_pages(READ_ONCE(prev_offset), offset, hits, |
625 | max_pages, | |
ec560175 HY |
626 | atomic_read(&last_readahead_pages)); |
627 | if (!hits) | |
d6c1f098 | 628 | WRITE_ONCE(prev_offset, offset); |
579f8290 SL |
629 | atomic_set(&last_readahead_pages, pages); |
630 | ||
631 | return pages; | |
632 | } | |
633 | ||
46017e95 | 634 | /** |
e9e9b7ec | 635 | * swap_cluster_readahead - swap in pages in hope we need them soon |
46017e95 | 636 | * @entry: swap entry of this memory |
7682486b | 637 | * @gfp_mask: memory allocation flags |
ddc1a5cb HD |
638 | * @mpol: NUMA memory allocation policy to be applied |
639 | * @ilx: NUMA interleave index, for use only when MPOL_INTERLEAVE | |
46017e95 | 640 | * |
a4575c41 | 641 | * Returns the struct folio for entry and addr, after queueing swapin. |
46017e95 HD |
642 | * |
643 | * Primitive swap readahead code. We simply read an aligned block of | |
644 | * (1 << page_cluster) entries in the swap area. This method is chosen | |
645 | * because it doesn't cost us any seek time. We also make sure to queue | |
646 | * the 'original' request together with the readahead ones... | |
647 | * | |
ddc1a5cb HD |
648 | * Note: it is intentional that the same NUMA policy and interleave index |
649 | * are used for every page of the readahead: neighbouring pages on swap | |
650 | * are fairly likely to have been swapped out from the same node. | |
46017e95 | 651 | */ |
a4575c41 | 652 | struct folio *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask, |
ddc1a5cb | 653 | struct mempolicy *mpol, pgoff_t ilx) |
46017e95 | 654 | { |
96c7b0b4 | 655 | struct folio *folio; |
579f8290 SL |
656 | unsigned long entry_offset = swp_offset(entry); |
657 | unsigned long offset = entry_offset; | |
67f96aa2 | 658 | unsigned long start_offset, end_offset; |
579f8290 | 659 | unsigned long mask; |
e9a6effa | 660 | struct swap_info_struct *si = swp_swap_info(entry); |
3fb5c298 | 661 | struct blk_plug plug; |
5169b844 | 662 | struct swap_iocb *splug = NULL; |
b243dcbf | 663 | bool page_allocated; |
46017e95 | 664 | |
579f8290 SL |
665 | mask = swapin_nr_pages(offset) - 1; |
666 | if (!mask) | |
667 | goto skip; | |
668 | ||
67f96aa2 RR |
669 | /* Read a page_cluster sized and aligned cluster around offset. */ |
670 | start_offset = offset & ~mask; | |
671 | end_offset = offset | mask; | |
672 | if (!start_offset) /* First page is swap header. */ | |
673 | start_offset++; | |
e9a6effa HY |
674 | if (end_offset >= si->max) |
675 | end_offset = si->max - 1; | |
67f96aa2 | 676 | |
3fb5c298 | 677 | blk_start_plug(&plug); |
67f96aa2 | 678 | for (offset = start_offset; offset <= end_offset ; offset++) { |
46017e95 | 679 | /* Ok, do the async read-ahead now */ |
96c7b0b4 | 680 | folio = __read_swap_cache_async( |
ddc1a5cb | 681 | swp_entry(swp_type(entry), offset), |
a65b0e76 | 682 | gfp_mask, mpol, ilx, &page_allocated, false); |
96c7b0b4 | 683 | if (!folio) |
67f96aa2 | 684 | continue; |
c4fa6309 | 685 | if (page_allocated) { |
b2d1f38b | 686 | swap_read_folio(folio, &splug); |
eaf649eb | 687 | if (offset != entry_offset) { |
96c7b0b4 | 688 | folio_set_readahead(folio); |
c4fa6309 HY |
689 | count_vm_event(SWAP_RA); |
690 | } | |
cbc65df2 | 691 | } |
96c7b0b4 | 692 | folio_put(folio); |
46017e95 | 693 | } |
3fb5c298 | 694 | blk_finish_plug(&plug); |
5169b844 | 695 | swap_read_unplug(splug); |
46017e95 | 696 | lru_add_drain(); /* Push any new pages onto the LRU now */ |
579f8290 | 697 | skip: |
5169b844 | 698 | /* The page was likely read above, so no need for plugging here */ |
96c7b0b4 | 699 | folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx, |
a65b0e76 | 700 | &page_allocated, false); |
16e96ba5 NP |
701 | if (unlikely(page_allocated)) { |
702 | zswap_folio_swapin(folio); | |
b2d1f38b | 703 | swap_read_folio(folio, NULL); |
16e96ba5 | 704 | } |
a4575c41 | 705 | return folio; |
46017e95 | 706 | } |
4b3ef9da HY |
707 | |
708 | int init_swap_address_space(unsigned int type, unsigned long nr_pages) | |
709 | { | |
710 | struct address_space *spaces, *space; | |
711 | unsigned int i, nr; | |
712 | ||
713 | nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES); | |
778e1cdd | 714 | spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL); |
4b3ef9da HY |
715 | if (!spaces) |
716 | return -ENOMEM; | |
717 | for (i = 0; i < nr; i++) { | |
718 | space = spaces + i; | |
a2833486 | 719 | xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ); |
4b3ef9da HY |
720 | atomic_set(&space->i_mmap_writable, 0); |
721 | space->a_ops = &swap_aops; | |
722 | /* swap cache doesn't use writeback related tags */ | |
723 | mapping_set_no_writeback_tags(space); | |
4b3ef9da HY |
724 | } |
725 | nr_swapper_spaces[type] = nr; | |
054f1d1f | 726 | swapper_spaces[type] = spaces; |
4b3ef9da HY |
727 | |
728 | return 0; | |
729 | } | |
730 | ||
731 | void exit_swap_address_space(unsigned int type) | |
732 | { | |
eea4a501 HY |
733 | int i; |
734 | struct address_space *spaces = swapper_spaces[type]; | |
735 | ||
736 | for (i = 0; i < nr_swapper_spaces[type]; i++) | |
737 | VM_WARN_ON_ONCE(!mapping_empty(&spaces[i])); | |
738 | kvfree(spaces); | |
4b3ef9da | 739 | nr_swapper_spaces[type] = 0; |
054f1d1f | 740 | swapper_spaces[type] = NULL; |
4b3ef9da | 741 | } |
ec560175 | 742 | |
ba518f4d HY |
743 | static int swap_vma_ra_win(struct vm_fault *vmf, unsigned long *start, |
744 | unsigned long *end) | |
ec560175 HY |
745 | { |
746 | struct vm_area_struct *vma = vmf->vma; | |
eaf649eb | 747 | unsigned long ra_val; |
ba518f4d | 748 | unsigned long faddr, prev_faddr, left, right; |
16ba391e | 749 | unsigned int max_win, hits, prev_win, win; |
ec560175 | 750 | |
ba518f4d | 751 | max_win = 1 << min(READ_ONCE(page_cluster), SWAP_RA_ORDER_CEILING); |
dce08dd2 HY |
752 | if (max_win == 1) |
753 | return 1; | |
61b63972 | 754 | |
ec560175 | 755 | faddr = vmf->address; |
eaf649eb | 756 | ra_val = GET_SWAP_RA_VAL(vma); |
ba518f4d | 757 | prev_faddr = SWAP_RA_ADDR(ra_val); |
eaf649eb MK |
758 | prev_win = SWAP_RA_WIN(ra_val); |
759 | hits = SWAP_RA_HITS(ra_val); | |
ba518f4d HY |
760 | win = __swapin_nr_pages(PFN_DOWN(prev_faddr), PFN_DOWN(faddr), hits, |
761 | max_win, prev_win); | |
762 | atomic_long_set(&vma->swap_readahead_info, SWAP_RA_VAL(faddr, win, 0)); | |
18ad72f5 | 763 | if (win == 1) |
dce08dd2 | 764 | return 1; |
ec560175 | 765 | |
ba518f4d HY |
766 | if (faddr == prev_faddr + PAGE_SIZE) |
767 | left = faddr; | |
768 | else if (prev_faddr == faddr + PAGE_SIZE) | |
769 | left = faddr - (win << PAGE_SHIFT) + PAGE_SIZE; | |
770 | else | |
771 | left = faddr - (((win - 1) / 2) << PAGE_SHIFT); | |
772 | right = left + (win << PAGE_SHIFT); | |
773 | if ((long)left < 0) | |
774 | left = 0; | |
775 | *start = max3(left, vma->vm_start, faddr & PMD_MASK); | |
776 | *end = min3(right, vma->vm_end, (faddr & PMD_MASK) + PMD_SIZE); | |
dce08dd2 HY |
777 | |
778 | return win; | |
ec560175 HY |
779 | } |
780 | ||
e9f59873 YS |
781 | /** |
782 | * swap_vma_readahead - swap in pages in hope we need them soon | |
ddc1a5cb | 783 | * @targ_entry: swap entry of the targeted memory |
e9f59873 | 784 | * @gfp_mask: memory allocation flags |
ddc1a5cb HD |
785 | * @mpol: NUMA memory allocation policy to be applied |
786 | * @targ_ilx: NUMA interleave index, for use only when MPOL_INTERLEAVE | |
e9f59873 YS |
787 | * @vmf: fault information |
788 | * | |
a4575c41 | 789 | * Returns the struct folio for entry and addr, after queueing swapin. |
e9f59873 | 790 | * |
cb152a1a | 791 | * Primitive swap readahead code. We simply read in a few pages whose |
e9f59873 YS |
792 | * virtual addresses are around the fault address in the same vma. |
793 | * | |
c1e8d7c6 | 794 | * Caller must hold read mmap_lock if vmf->vma is not NULL. |
e9f59873 YS |
795 | * |
796 | */ | |
a4575c41 MWO |
797 | static struct folio *swap_vma_readahead(swp_entry_t targ_entry, gfp_t gfp_mask, |
798 | struct mempolicy *mpol, pgoff_t targ_ilx, struct vm_fault *vmf) | |
ec560175 HY |
799 | { |
800 | struct blk_plug plug; | |
5169b844 | 801 | struct swap_iocb *splug = NULL; |
96c7b0b4 | 802 | struct folio *folio; |
4f8fcf4c | 803 | pte_t *pte = NULL, pentry; |
ba518f4d HY |
804 | int win; |
805 | unsigned long start, end, addr; | |
ec560175 | 806 | swp_entry_t entry; |
ddc1a5cb | 807 | pgoff_t ilx; |
ec560175 HY |
808 | bool page_allocated; |
809 | ||
ba518f4d | 810 | win = swap_vma_ra_win(vmf, &start, &end); |
dce08dd2 | 811 | if (win == 1) |
ec560175 HY |
812 | goto skip; |
813 | ||
ba518f4d | 814 | ilx = targ_ilx - PFN_DOWN(vmf->address - start); |
4f8fcf4c | 815 | |
ec560175 | 816 | blk_start_plug(&plug); |
ba518f4d | 817 | for (addr = start; addr < end; ilx++, addr += PAGE_SIZE) { |
4f8fcf4c HD |
818 | if (!pte++) { |
819 | pte = pte_offset_map(vmf->pmd, addr); | |
820 | if (!pte) | |
821 | break; | |
822 | } | |
823 | pentry = ptep_get_lockless(pte); | |
92bafb20 | 824 | if (!is_swap_pte(pentry)) |
ec560175 HY |
825 | continue; |
826 | entry = pte_to_swp_entry(pentry); | |
827 | if (unlikely(non_swap_entry(entry))) | |
828 | continue; | |
4f8fcf4c HD |
829 | pte_unmap(pte); |
830 | pte = NULL; | |
96c7b0b4 | 831 | folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx, |
a65b0e76 | 832 | &page_allocated, false); |
96c7b0b4 | 833 | if (!folio) |
ec560175 HY |
834 | continue; |
835 | if (page_allocated) { | |
b2d1f38b | 836 | swap_read_folio(folio, &splug); |
ba518f4d | 837 | if (addr != vmf->address) { |
96c7b0b4 | 838 | folio_set_readahead(folio); |
ec560175 HY |
839 | count_vm_event(SWAP_RA); |
840 | } | |
841 | } | |
96c7b0b4 | 842 | folio_put(folio); |
ec560175 | 843 | } |
4f8fcf4c HD |
844 | if (pte) |
845 | pte_unmap(pte); | |
ec560175 | 846 | blk_finish_plug(&plug); |
5169b844 | 847 | swap_read_unplug(splug); |
ec560175 HY |
848 | lru_add_drain(); |
849 | skip: | |
96c7b0b4 MWO |
850 | /* The folio was likely read above, so no need for plugging here */ |
851 | folio = __read_swap_cache_async(targ_entry, gfp_mask, mpol, targ_ilx, | |
a65b0e76 | 852 | &page_allocated, false); |
16e96ba5 NP |
853 | if (unlikely(page_allocated)) { |
854 | zswap_folio_swapin(folio); | |
b2d1f38b | 855 | swap_read_folio(folio, NULL); |
16e96ba5 | 856 | } |
a4575c41 | 857 | return folio; |
ec560175 | 858 | } |
d9bfcfdc | 859 | |
e9e9b7ec MK |
860 | /** |
861 | * swapin_readahead - swap in pages in hope we need them soon | |
862 | * @entry: swap entry of this memory | |
863 | * @gfp_mask: memory allocation flags | |
864 | * @vmf: fault information | |
865 | * | |
866 | * Returns the struct page for entry and addr, after queueing swapin. | |
867 | * | |
868 | * It's a main entry function for swap readahead. By the configuration, | |
869 | * it will read ahead blocks by cluster-based(ie, physical disk based) | |
870 | * or vma-based(ie, virtual address based on faulty address) readahead. | |
871 | */ | |
872 | struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask, | |
873 | struct vm_fault *vmf) | |
874 | { | |
ddc1a5cb HD |
875 | struct mempolicy *mpol; |
876 | pgoff_t ilx; | |
a4575c41 | 877 | struct folio *folio; |
ddc1a5cb HD |
878 | |
879 | mpol = get_vma_policy(vmf->vma, vmf->address, 0, &ilx); | |
a4575c41 | 880 | folio = swap_use_vma_readahead() ? |
ddc1a5cb HD |
881 | swap_vma_readahead(entry, gfp_mask, mpol, ilx, vmf) : |
882 | swap_cluster_readahead(entry, gfp_mask, mpol, ilx); | |
883 | mpol_cond_put(mpol); | |
a4575c41 MWO |
884 | |
885 | if (!folio) | |
886 | return NULL; | |
887 | return folio_file_page(folio, swp_offset(entry)); | |
e9e9b7ec MK |
888 | } |
889 | ||
d9bfcfdc HY |
890 | #ifdef CONFIG_SYSFS |
891 | static ssize_t vma_ra_enabled_show(struct kobject *kobj, | |
892 | struct kobj_attribute *attr, char *buf) | |
893 | { | |
ae7a927d JP |
894 | return sysfs_emit(buf, "%s\n", |
895 | enable_vma_readahead ? "true" : "false"); | |
d9bfcfdc HY |
896 | } |
897 | static ssize_t vma_ra_enabled_store(struct kobject *kobj, | |
898 | struct kobj_attribute *attr, | |
899 | const char *buf, size_t count) | |
900 | { | |
717aeab4 JG |
901 | ssize_t ret; |
902 | ||
903 | ret = kstrtobool(buf, &enable_vma_readahead); | |
904 | if (ret) | |
905 | return ret; | |
d9bfcfdc HY |
906 | |
907 | return count; | |
908 | } | |
6106b93e | 909 | static struct kobj_attribute vma_ra_enabled_attr = __ATTR_RW(vma_ra_enabled); |
d9bfcfdc | 910 | |
d9bfcfdc HY |
911 | static struct attribute *swap_attrs[] = { |
912 | &vma_ra_enabled_attr.attr, | |
d9bfcfdc HY |
913 | NULL, |
914 | }; | |
915 | ||
e48333b6 | 916 | static const struct attribute_group swap_attr_group = { |
d9bfcfdc HY |
917 | .attrs = swap_attrs, |
918 | }; | |
919 | ||
920 | static int __init swap_init_sysfs(void) | |
921 | { | |
922 | int err; | |
923 | struct kobject *swap_kobj; | |
924 | ||
925 | swap_kobj = kobject_create_and_add("swap", mm_kobj); | |
926 | if (!swap_kobj) { | |
927 | pr_err("failed to create swap kobject\n"); | |
928 | return -ENOMEM; | |
929 | } | |
930 | err = sysfs_create_group(swap_kobj, &swap_attr_group); | |
931 | if (err) { | |
932 | pr_err("failed to register swap group\n"); | |
933 | goto delete_obj; | |
934 | } | |
935 | return 0; | |
936 | ||
937 | delete_obj: | |
938 | kobject_put(swap_kobj); | |
939 | return err; | |
940 | } | |
941 | subsys_initcall(swap_init_sysfs); | |
942 | #endif |