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 | |
7eaceacc | 31 | * vmscan's shrink_page_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 | |
ec560175 HY |
45 | #define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2) |
46 | #define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1) | |
47 | #define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK | |
48 | #define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK) | |
49 | ||
50 | #define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK) | |
51 | #define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT) | |
52 | #define SWAP_RA_ADDR(v) ((v) & PAGE_MASK) | |
53 | ||
54 | #define SWAP_RA_VAL(addr, win, hits) \ | |
55 | (((addr) & PAGE_MASK) | \ | |
56 | (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \ | |
57 | ((hits) & SWAP_RA_HITS_MASK)) | |
58 | ||
59 | /* Initial readahead hits is 4 to start up with a small window */ | |
60 | #define GET_SWAP_RA_VAL(vma) \ | |
61 | (atomic_long_read(&(vma)->swap_readahead_info) ? : 4) | |
1da177e4 | 62 | |
579f8290 SL |
63 | static atomic_t swapin_readahead_hits = ATOMIC_INIT(4); |
64 | ||
1da177e4 LT |
65 | void show_swap_cache_info(void) |
66 | { | |
33806f06 | 67 | printk("%lu pages in swap cache\n", total_swapcache_pages()); |
3cb8eaa4 Z |
68 | printk("Free swap = %ldkB\n", K(get_nr_swap_pages())); |
69 | printk("Total swap = %lukB\n", K(total_swap_pages)); | |
1da177e4 LT |
70 | } |
71 | ||
aae466b0 JK |
72 | void *get_shadow_from_swap_cache(swp_entry_t entry) |
73 | { | |
74 | struct address_space *address_space = swap_address_space(entry); | |
75 | pgoff_t idx = swp_offset(entry); | |
4c773a44 | 76 | void *shadow; |
aae466b0 | 77 | |
4c773a44 MWO |
78 | shadow = xa_load(&address_space->i_pages, idx); |
79 | if (xa_is_value(shadow)) | |
80 | return shadow; | |
aae466b0 JK |
81 | return NULL; |
82 | } | |
83 | ||
1da177e4 | 84 | /* |
2bb876b5 | 85 | * add_to_swap_cache resembles filemap_add_folio on swapper_space, |
1da177e4 LT |
86 | * but sets SwapCache flag and private instead of mapping and index. |
87 | */ | |
a4c366f0 | 88 | int add_to_swap_cache(struct folio *folio, swp_entry_t entry, |
3852f676 | 89 | gfp_t gfp, void **shadowp) |
1da177e4 | 90 | { |
8d93b41c | 91 | struct address_space *address_space = swap_address_space(entry); |
38d8b4e6 | 92 | pgoff_t idx = swp_offset(entry); |
a4c366f0 MWO |
93 | XA_STATE_ORDER(xas, &address_space->i_pages, idx, folio_order(folio)); |
94 | unsigned long i, nr = folio_nr_pages(folio); | |
3852f676 | 95 | void *old; |
1da177e4 | 96 | |
5649d113 YY |
97 | xas_set_update(&xas, workingset_update_node); |
98 | ||
a4c366f0 MWO |
99 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
100 | VM_BUG_ON_FOLIO(folio_test_swapcache(folio), folio); | |
101 | VM_BUG_ON_FOLIO(!folio_test_swapbacked(folio), folio); | |
51726b12 | 102 | |
a4c366f0 MWO |
103 | folio_ref_add(folio, nr); |
104 | folio_set_swapcache(folio); | |
3d2c9087 | 105 | folio->swap = entry; |
31a56396 | 106 | |
8d93b41c MW |
107 | do { |
108 | xas_lock_irq(&xas); | |
109 | xas_create_range(&xas); | |
110 | if (xas_error(&xas)) | |
111 | goto unlock; | |
112 | for (i = 0; i < nr; i++) { | |
a4c366f0 | 113 | VM_BUG_ON_FOLIO(xas.xa_index != idx + i, folio); |
e5b306a0 KS |
114 | if (shadowp) { |
115 | old = xas_load(&xas); | |
116 | if (xa_is_value(old)) | |
3852f676 JK |
117 | *shadowp = old; |
118 | } | |
a4c366f0 | 119 | xas_store(&xas, folio); |
8d93b41c MW |
120 | xas_next(&xas); |
121 | } | |
38d8b4e6 | 122 | address_space->nrpages += nr; |
a4c366f0 MWO |
123 | __node_stat_mod_folio(folio, NR_FILE_PAGES, nr); |
124 | __lruvec_stat_mod_folio(folio, NR_SWAPCACHE, nr); | |
8d93b41c MW |
125 | unlock: |
126 | xas_unlock_irq(&xas); | |
127 | } while (xas_nomem(&xas, gfp)); | |
31a56396 | 128 | |
8d93b41c MW |
129 | if (!xas_error(&xas)) |
130 | return 0; | |
31a56396 | 131 | |
a4c366f0 MWO |
132 | folio_clear_swapcache(folio); |
133 | folio_ref_sub(folio, nr); | |
8d93b41c | 134 | return xas_error(&xas); |
1da177e4 LT |
135 | } |
136 | ||
1da177e4 | 137 | /* |
ceff9d33 | 138 | * This must be called only on folios that have |
1da177e4 LT |
139 | * been verified to be in the swap cache. |
140 | */ | |
ceff9d33 | 141 | void __delete_from_swap_cache(struct folio *folio, |
3852f676 | 142 | swp_entry_t entry, void *shadow) |
1da177e4 | 143 | { |
4e17ec25 | 144 | struct address_space *address_space = swap_address_space(entry); |
ceff9d33 MWO |
145 | int i; |
146 | long nr = folio_nr_pages(folio); | |
4e17ec25 MW |
147 | pgoff_t idx = swp_offset(entry); |
148 | XA_STATE(xas, &address_space->i_pages, idx); | |
33806f06 | 149 | |
5649d113 YY |
150 | xas_set_update(&xas, workingset_update_node); |
151 | ||
ceff9d33 MWO |
152 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
153 | VM_BUG_ON_FOLIO(!folio_test_swapcache(folio), folio); | |
154 | VM_BUG_ON_FOLIO(folio_test_writeback(folio), folio); | |
1da177e4 | 155 | |
38d8b4e6 | 156 | for (i = 0; i < nr; i++) { |
3852f676 | 157 | void *entry = xas_store(&xas, shadow); |
b9eb7776 | 158 | VM_BUG_ON_PAGE(entry != folio, entry); |
4e17ec25 | 159 | xas_next(&xas); |
38d8b4e6 | 160 | } |
3d2c9087 | 161 | folio->swap.val = 0; |
ceff9d33 | 162 | folio_clear_swapcache(folio); |
38d8b4e6 | 163 | address_space->nrpages -= nr; |
ceff9d33 MWO |
164 | __node_stat_mod_folio(folio, NR_FILE_PAGES, -nr); |
165 | __lruvec_stat_mod_folio(folio, NR_SWAPCACHE, -nr); | |
1da177e4 LT |
166 | } |
167 | ||
168 | /** | |
09c02e56 MWO |
169 | * add_to_swap - allocate swap space for a folio |
170 | * @folio: folio we want to move to swap | |
1da177e4 | 171 | * |
09c02e56 MWO |
172 | * Allocate swap space for the folio and add the folio to the |
173 | * swap cache. | |
174 | * | |
175 | * Context: Caller needs to hold the folio lock. | |
176 | * Return: Whether the folio was added to the swap cache. | |
1da177e4 | 177 | */ |
09c02e56 | 178 | bool add_to_swap(struct folio *folio) |
1da177e4 LT |
179 | { |
180 | swp_entry_t entry; | |
1da177e4 LT |
181 | int err; |
182 | ||
09c02e56 MWO |
183 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
184 | VM_BUG_ON_FOLIO(!folio_test_uptodate(folio), folio); | |
1da177e4 | 185 | |
e2e3fdc7 | 186 | entry = folio_alloc_swap(folio); |
2ca4532a | 187 | if (!entry.val) |
09c02e56 | 188 | return false; |
0f074658 | 189 | |
2ca4532a | 190 | /* |
8d93b41c | 191 | * XArray node allocations from PF_MEMALLOC contexts could |
2ca4532a DN |
192 | * completely exhaust the page allocator. __GFP_NOMEMALLOC |
193 | * stops emergency reserves from being allocated. | |
194 | * | |
195 | * TODO: this could cause a theoretical memory reclaim | |
196 | * deadlock in the swap out path. | |
197 | */ | |
198 | /* | |
854e9ed0 | 199 | * Add it to the swap cache. |
2ca4532a | 200 | */ |
a4c366f0 | 201 | err = add_to_swap_cache(folio, entry, |
3852f676 | 202 | __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN, NULL); |
38d8b4e6 | 203 | if (err) |
bd53b714 | 204 | /* |
2ca4532a DN |
205 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely |
206 | * clear SWAP_HAS_CACHE flag. | |
1da177e4 | 207 | */ |
0f074658 | 208 | goto fail; |
9625456c | 209 | /* |
09c02e56 MWO |
210 | * Normally the folio will be dirtied in unmap because its |
211 | * pte should be dirty. A special case is MADV_FREE page. The | |
212 | * page's pte could have dirty bit cleared but the folio's | |
213 | * SwapBacked flag is still set because clearing the dirty bit | |
214 | * and SwapBacked flag has no lock protected. For such folio, | |
215 | * unmap will not set dirty bit for it, so folio reclaim will | |
216 | * not write the folio out. This can cause data corruption when | |
217 | * the folio is swapped in later. Always setting the dirty flag | |
218 | * for the folio solves the problem. | |
9625456c | 219 | */ |
09c02e56 | 220 | folio_mark_dirty(folio); |
38d8b4e6 | 221 | |
09c02e56 | 222 | return true; |
38d8b4e6 | 223 | |
38d8b4e6 | 224 | fail: |
4081f744 | 225 | put_swap_folio(folio, entry); |
09c02e56 | 226 | return false; |
1da177e4 LT |
227 | } |
228 | ||
229 | /* | |
75fa68a5 | 230 | * This must be called only on folios that have |
1da177e4 | 231 | * been verified to be in the swap cache and locked. |
75fa68a5 MWO |
232 | * It will never put the folio into the free list, |
233 | * the caller has a reference on the folio. | |
1da177e4 | 234 | */ |
75fa68a5 | 235 | void delete_from_swap_cache(struct folio *folio) |
1da177e4 | 236 | { |
3d2c9087 | 237 | swp_entry_t entry = folio->swap; |
4e17ec25 | 238 | struct address_space *address_space = swap_address_space(entry); |
1da177e4 | 239 | |
b93b0163 | 240 | xa_lock_irq(&address_space->i_pages); |
ceff9d33 | 241 | __delete_from_swap_cache(folio, entry, NULL); |
b93b0163 | 242 | xa_unlock_irq(&address_space->i_pages); |
1da177e4 | 243 | |
4081f744 | 244 | put_swap_folio(folio, entry); |
75fa68a5 | 245 | folio_ref_sub(folio, folio_nr_pages(folio)); |
1da177e4 LT |
246 | } |
247 | ||
3852f676 JK |
248 | void clear_shadow_from_swap_cache(int type, unsigned long begin, |
249 | unsigned long end) | |
250 | { | |
251 | unsigned long curr = begin; | |
252 | void *old; | |
253 | ||
254 | for (;;) { | |
3852f676 JK |
255 | swp_entry_t entry = swp_entry(type, curr); |
256 | struct address_space *address_space = swap_address_space(entry); | |
257 | XA_STATE(xas, &address_space->i_pages, curr); | |
258 | ||
5649d113 YY |
259 | xas_set_update(&xas, workingset_update_node); |
260 | ||
3852f676 JK |
261 | xa_lock_irq(&address_space->i_pages); |
262 | xas_for_each(&xas, old, end) { | |
263 | if (!xa_is_value(old)) | |
264 | continue; | |
265 | xas_store(&xas, NULL); | |
3852f676 | 266 | } |
3852f676 JK |
267 | xa_unlock_irq(&address_space->i_pages); |
268 | ||
269 | /* search the next swapcache until we meet end */ | |
270 | curr >>= SWAP_ADDRESS_SPACE_SHIFT; | |
271 | curr++; | |
272 | curr <<= SWAP_ADDRESS_SPACE_SHIFT; | |
273 | if (curr > end) | |
274 | break; | |
275 | } | |
276 | } | |
277 | ||
c33c7948 RR |
278 | /* |
279 | * If we are the only user, then try to free up the swap cache. | |
280 | * | |
aedd74d4 | 281 | * Its ok to check the swapcache flag without the folio lock |
a2c43eed | 282 | * here because we are going to recheck again inside |
aedd74d4 | 283 | * folio_free_swap() _with_ the lock. |
1da177e4 LT |
284 | * - Marcelo |
285 | */ | |
63b77499 | 286 | void free_swap_cache(struct folio *folio) |
1da177e4 | 287 | { |
aedd74d4 MWO |
288 | if (folio_test_swapcache(folio) && !folio_mapped(folio) && |
289 | folio_trylock(folio)) { | |
290 | folio_free_swap(folio); | |
291 | folio_unlock(folio); | |
1da177e4 LT |
292 | } |
293 | } | |
294 | ||
c33c7948 | 295 | /* |
1da177e4 | 296 | * Perform a free_page(), also freeing any swap cache associated with |
b8072f09 | 297 | * this page if it is the last user of the page. |
1da177e4 LT |
298 | */ |
299 | void free_page_and_swap_cache(struct page *page) | |
300 | { | |
63b77499 MWO |
301 | struct folio *folio = page_folio(page); |
302 | ||
303 | free_swap_cache(folio); | |
5beaee54 | 304 | if (!is_huge_zero_folio(folio)) |
63b77499 | 305 | folio_put(folio); |
1da177e4 LT |
306 | } |
307 | ||
308 | /* | |
309 | * Passed an array of pages, drop them all from swapcache and then release | |
310 | * them. They are removed from the LRU and freed if this is their last use. | |
311 | */ | |
7cc8f9c7 | 312 | void free_pages_and_swap_cache(struct encoded_page **pages, int nr) |
1da177e4 | 313 | { |
4907e80b MWO |
314 | struct folio_batch folios; |
315 | unsigned int refs[PAGEVEC_SIZE]; | |
316 | ||
1da177e4 | 317 | lru_add_drain(); |
4907e80b | 318 | folio_batch_init(&folios); |
d7f861b9 | 319 | for (int i = 0; i < nr; i++) { |
4907e80b | 320 | struct folio *folio = page_folio(encoded_page_ptr(pages[i])); |
d7f861b9 | 321 | |
63b77499 | 322 | free_swap_cache(folio); |
4907e80b | 323 | refs[folios.nr] = 1; |
d7f861b9 DH |
324 | if (unlikely(encoded_page_flags(pages[i]) & |
325 | ENCODED_PAGE_BIT_NR_PAGES_NEXT)) | |
4907e80b | 326 | refs[folios.nr] = encoded_nr_pages(pages[++i]); |
d7f861b9 | 327 | |
4907e80b MWO |
328 | if (folio_batch_add(&folios, folio) == 0) |
329 | folios_put_refs(&folios, refs); | |
d7f861b9 | 330 | } |
4907e80b MWO |
331 | if (folios.nr) |
332 | folios_put_refs(&folios, refs); | |
1da177e4 LT |
333 | } |
334 | ||
e9e9b7ec MK |
335 | static inline bool swap_use_vma_readahead(void) |
336 | { | |
337 | return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap); | |
338 | } | |
339 | ||
1da177e4 | 340 | /* |
c9edc242 | 341 | * Lookup a swap entry in the swap cache. A found folio will be returned |
1da177e4 | 342 | * unlocked and with its refcount incremented - we rely on the kernel |
c9edc242 | 343 | * lock getting page table operations atomic even if we drop the folio |
1da177e4 | 344 | * lock before returning. |
cbc2bd98 KS |
345 | * |
346 | * Caller must lock the swap device or hold a reference to keep it valid. | |
1da177e4 | 347 | */ |
c9edc242 MWO |
348 | struct folio *swap_cache_get_folio(swp_entry_t entry, |
349 | struct vm_area_struct *vma, unsigned long addr) | |
1da177e4 | 350 | { |
c9edc242 | 351 | struct folio *folio; |
1da177e4 | 352 | |
c9edc242 | 353 | folio = filemap_get_folio(swap_address_space(entry), swp_offset(entry)); |
66dabbb6 | 354 | if (!IS_ERR(folio)) { |
eaf649eb MK |
355 | bool vma_ra = swap_use_vma_readahead(); |
356 | bool readahead; | |
357 | ||
eaf649eb MK |
358 | /* |
359 | * At the moment, we don't support PG_readahead for anon THP | |
360 | * so let's bail out rather than confusing the readahead stat. | |
361 | */ | |
c9edc242 MWO |
362 | if (unlikely(folio_test_large(folio))) |
363 | return folio; | |
eaf649eb | 364 | |
c9edc242 | 365 | readahead = folio_test_clear_readahead(folio); |
eaf649eb MK |
366 | if (vma && vma_ra) { |
367 | unsigned long ra_val; | |
368 | int win, hits; | |
369 | ||
370 | ra_val = GET_SWAP_RA_VAL(vma); | |
371 | win = SWAP_RA_WIN(ra_val); | |
372 | hits = SWAP_RA_HITS(ra_val); | |
ec560175 HY |
373 | if (readahead) |
374 | hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX); | |
375 | atomic_long_set(&vma->swap_readahead_info, | |
376 | SWAP_RA_VAL(addr, win, hits)); | |
377 | } | |
eaf649eb | 378 | |
ec560175 | 379 | if (readahead) { |
cbc65df2 | 380 | count_vm_event(SWAP_RA_HIT); |
eaf649eb | 381 | if (!vma || !vma_ra) |
ec560175 | 382 | atomic_inc(&swapin_readahead_hits); |
cbc65df2 | 383 | } |
66dabbb6 CH |
384 | } else { |
385 | folio = NULL; | |
579f8290 | 386 | } |
eaf649eb | 387 | |
c9edc242 MWO |
388 | return folio; |
389 | } | |
390 | ||
61ef1865 | 391 | /** |
524984ff | 392 | * filemap_get_incore_folio - Find and get a folio from the page or swap caches. |
61ef1865 MWO |
393 | * @mapping: The address_space to search. |
394 | * @index: The page cache index. | |
395 | * | |
524984ff MWO |
396 | * This differs from filemap_get_folio() in that it will also look for the |
397 | * folio in the swap cache. | |
61ef1865 | 398 | * |
524984ff | 399 | * Return: The found folio or %NULL. |
61ef1865 | 400 | */ |
524984ff MWO |
401 | struct folio *filemap_get_incore_folio(struct address_space *mapping, |
402 | pgoff_t index) | |
61ef1865 MWO |
403 | { |
404 | swp_entry_t swp; | |
405 | struct swap_info_struct *si; | |
097b3e59 | 406 | struct folio *folio = filemap_get_entry(mapping, index); |
61ef1865 | 407 | |
66dabbb6 CH |
408 | if (!folio) |
409 | return ERR_PTR(-ENOENT); | |
dd8095b1 | 410 | if (!xa_is_value(folio)) |
66dabbb6 | 411 | return folio; |
61ef1865 | 412 | if (!shmem_mapping(mapping)) |
66dabbb6 | 413 | return ERR_PTR(-ENOENT); |
61ef1865 | 414 | |
dd8095b1 | 415 | swp = radix_to_swp_entry(folio); |
ba6851b4 ML |
416 | /* There might be swapin error entries in shmem mapping. */ |
417 | if (non_swap_entry(swp)) | |
66dabbb6 | 418 | return ERR_PTR(-ENOENT); |
61ef1865 MWO |
419 | /* Prevent swapoff from happening to us */ |
420 | si = get_swap_device(swp); | |
421 | if (!si) | |
66dabbb6 | 422 | return ERR_PTR(-ENOENT); |
dd8095b1 MWO |
423 | index = swp_offset(swp); |
424 | folio = filemap_get_folio(swap_address_space(swp), index); | |
61ef1865 | 425 | put_swap_device(si); |
524984ff | 426 | return folio; |
61ef1865 MWO |
427 | } |
428 | ||
96c7b0b4 MWO |
429 | struct folio *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
430 | struct mempolicy *mpol, pgoff_t ilx, bool *new_page_allocated, | |
431 | bool skip_if_exists) | |
1da177e4 | 432 | { |
eb085574 | 433 | struct swap_info_struct *si; |
a0d3374b | 434 | struct folio *folio; |
aae466b0 | 435 | void *shadow = NULL; |
4c6355b2 | 436 | |
5b999aad | 437 | *new_page_allocated = false; |
46a774d3 HY |
438 | si = get_swap_device(entry); |
439 | if (!si) | |
440 | return NULL; | |
1da177e4 | 441 | |
4c6355b2 JW |
442 | for (;;) { |
443 | int err; | |
1da177e4 LT |
444 | /* |
445 | * First check the swap cache. Since this is normally | |
cb691e2f | 446 | * called after swap_cache_get_folio() failed, re-calling |
1da177e4 LT |
447 | * that would confuse statistics. |
448 | */ | |
a0d3374b MWO |
449 | folio = filemap_get_folio(swap_address_space(entry), |
450 | swp_offset(entry)); | |
96c7b0b4 MWO |
451 | if (!IS_ERR(folio)) |
452 | goto got_folio; | |
1da177e4 | 453 | |
ba81f838 HY |
454 | /* |
455 | * Just skip read ahead for unused swap slot. | |
456 | * During swap_off when swap_slot_cache is disabled, | |
457 | * we have to handle the race between putting | |
458 | * swap entry in swap cache and marking swap slot | |
459 | * as SWAP_HAS_CACHE. That's done in later part of code or | |
460 | * else swap_off will be aborted if we return NULL. | |
461 | */ | |
3ecdeb0f | 462 | if (!swap_swapcount(si, entry) && swap_slot_cache_enabled) |
46a774d3 | 463 | goto fail_put_swap; |
e8c26ab6 | 464 | |
1da177e4 | 465 | /* |
96c7b0b4 | 466 | * Get a new folio to read into from swap. Allocate it now, |
4c6355b2 JW |
467 | * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will |
468 | * cause any racers to loop around until we add it to cache. | |
1da177e4 | 469 | */ |
ddc1a5cb HD |
470 | folio = (struct folio *)alloc_pages_mpol(gfp_mask, 0, |
471 | mpol, ilx, numa_node_id()); | |
a0d3374b | 472 | if (!folio) |
46a774d3 | 473 | goto fail_put_swap; |
1da177e4 | 474 | |
f000944d HD |
475 | /* |
476 | * Swap entry may have been freed since our caller observed it. | |
477 | */ | |
355cfa73 | 478 | err = swapcache_prepare(entry); |
4c6355b2 | 479 | if (!err) |
f000944d HD |
480 | break; |
481 | ||
a0d3374b | 482 | folio_put(folio); |
4c6355b2 | 483 | if (err != -EEXIST) |
46a774d3 | 484 | goto fail_put_swap; |
4c6355b2 | 485 | |
a65b0e76 DC |
486 | /* |
487 | * Protect against a recursive call to __read_swap_cache_async() | |
488 | * on the same entry waiting forever here because SWAP_HAS_CACHE | |
489 | * is set but the folio is not the swap cache yet. This can | |
490 | * happen today if mem_cgroup_swapin_charge_folio() below | |
491 | * triggers reclaim through zswap, which may call | |
492 | * __read_swap_cache_async() in the writeback path. | |
493 | */ | |
494 | if (skip_if_exists) | |
495 | goto fail_put_swap; | |
496 | ||
2ca4532a | 497 | /* |
4c6355b2 JW |
498 | * We might race against __delete_from_swap_cache(), and |
499 | * stumble across a swap_map entry whose SWAP_HAS_CACHE | |
500 | * has not yet been cleared. Or race against another | |
501 | * __read_swap_cache_async(), which has set SWAP_HAS_CACHE | |
96c7b0b4 | 502 | * in swap_map, but not yet added its folio to swap cache. |
2ca4532a | 503 | */ |
029c4628 | 504 | schedule_timeout_uninterruptible(1); |
4c6355b2 JW |
505 | } |
506 | ||
507 | /* | |
96c7b0b4 | 508 | * The swap entry is ours to swap in. Prepare the new folio. |
4c6355b2 JW |
509 | */ |
510 | ||
a0d3374b MWO |
511 | __folio_set_locked(folio); |
512 | __folio_set_swapbacked(folio); | |
4c6355b2 | 513 | |
65995918 | 514 | if (mem_cgroup_swapin_charge_folio(folio, NULL, gfp_mask, entry)) |
4c6355b2 | 515 | goto fail_unlock; |
4c6355b2 | 516 | |
0add0c77 | 517 | /* May fail (-ENOMEM) if XArray node allocation failed. */ |
a4c366f0 | 518 | if (add_to_swap_cache(folio, entry, gfp_mask & GFP_RECLAIM_MASK, &shadow)) |
4c6355b2 | 519 | goto fail_unlock; |
0add0c77 SB |
520 | |
521 | mem_cgroup_swapin_uncharge_swap(entry); | |
4c6355b2 | 522 | |
aae466b0 | 523 | if (shadow) |
a0d3374b | 524 | workingset_refault(folio, shadow); |
314b57fb | 525 | |
a0d3374b MWO |
526 | /* Caller will initiate read into locked folio */ |
527 | folio_add_lru(folio); | |
4c6355b2 | 528 | *new_page_allocated = true; |
96c7b0b4 | 529 | got_folio: |
46a774d3 | 530 | put_swap_device(si); |
96c7b0b4 | 531 | return folio; |
1da177e4 | 532 | |
4c6355b2 | 533 | fail_unlock: |
4081f744 | 534 | put_swap_folio(folio, entry); |
a0d3374b MWO |
535 | folio_unlock(folio); |
536 | folio_put(folio); | |
46a774d3 HY |
537 | fail_put_swap: |
538 | put_swap_device(si); | |
4c6355b2 | 539 | return NULL; |
1da177e4 | 540 | } |
46017e95 | 541 | |
5b999aad DS |
542 | /* |
543 | * Locate a page of swap in physical memory, reserving swap cache space | |
544 | * and reading the disk if it is not already cached. | |
545 | * A failure return means that either the page allocation failed or that | |
546 | * the swap entry is no longer in use. | |
46a774d3 HY |
547 | * |
548 | * get/put_swap_device() aren't needed to call this function, because | |
c9bdf768 | 549 | * __read_swap_cache_async() call them and swap_read_folio() holds the |
46a774d3 | 550 | * swap cache folio lock. |
5b999aad | 551 | */ |
6e03492e MWO |
552 | struct folio *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
553 | struct vm_area_struct *vma, unsigned long addr, | |
554 | struct swap_iocb **plug) | |
5b999aad | 555 | { |
ddc1a5cb HD |
556 | bool page_allocated; |
557 | struct mempolicy *mpol; | |
558 | pgoff_t ilx; | |
96c7b0b4 | 559 | struct folio *folio; |
5b999aad | 560 | |
ddc1a5cb | 561 | mpol = get_vma_policy(vma, addr, 0, &ilx); |
96c7b0b4 | 562 | folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx, |
a65b0e76 | 563 | &page_allocated, false); |
ddc1a5cb | 564 | mpol_cond_put(mpol); |
5b999aad | 565 | |
ddc1a5cb | 566 | if (page_allocated) |
c9bdf768 | 567 | swap_read_folio(folio, false, plug); |
6e03492e | 568 | return folio; |
5b999aad DS |
569 | } |
570 | ||
ec560175 HY |
571 | static unsigned int __swapin_nr_pages(unsigned long prev_offset, |
572 | unsigned long offset, | |
573 | int hits, | |
574 | int max_pages, | |
575 | int prev_win) | |
579f8290 | 576 | { |
ec560175 | 577 | unsigned int pages, last_ra; |
579f8290 SL |
578 | |
579 | /* | |
580 | * This heuristic has been found to work well on both sequential and | |
581 | * random loads, swapping to hard disk or to SSD: please don't ask | |
582 | * what the "+ 2" means, it just happens to work well, that's all. | |
583 | */ | |
ec560175 | 584 | pages = hits + 2; |
579f8290 SL |
585 | if (pages == 2) { |
586 | /* | |
587 | * We can have no readahead hits to judge by: but must not get | |
588 | * stuck here forever, so check for an adjacent offset instead | |
589 | * (and don't even bother to check whether swap type is same). | |
590 | */ | |
591 | if (offset != prev_offset + 1 && offset != prev_offset - 1) | |
592 | pages = 1; | |
579f8290 SL |
593 | } else { |
594 | unsigned int roundup = 4; | |
595 | while (roundup < pages) | |
596 | roundup <<= 1; | |
597 | pages = roundup; | |
598 | } | |
599 | ||
600 | if (pages > max_pages) | |
601 | pages = max_pages; | |
602 | ||
603 | /* Don't shrink readahead too fast */ | |
ec560175 | 604 | last_ra = prev_win / 2; |
579f8290 SL |
605 | if (pages < last_ra) |
606 | pages = last_ra; | |
ec560175 HY |
607 | |
608 | return pages; | |
609 | } | |
610 | ||
611 | static unsigned long swapin_nr_pages(unsigned long offset) | |
612 | { | |
613 | static unsigned long prev_offset; | |
614 | unsigned int hits, pages, max_pages; | |
615 | static atomic_t last_readahead_pages; | |
616 | ||
617 | max_pages = 1 << READ_ONCE(page_cluster); | |
618 | if (max_pages <= 1) | |
619 | return 1; | |
620 | ||
621 | hits = atomic_xchg(&swapin_readahead_hits, 0); | |
d6c1f098 QC |
622 | pages = __swapin_nr_pages(READ_ONCE(prev_offset), offset, hits, |
623 | max_pages, | |
ec560175 HY |
624 | atomic_read(&last_readahead_pages)); |
625 | if (!hits) | |
d6c1f098 | 626 | WRITE_ONCE(prev_offset, offset); |
579f8290 SL |
627 | atomic_set(&last_readahead_pages, pages); |
628 | ||
629 | return pages; | |
630 | } | |
631 | ||
46017e95 | 632 | /** |
e9e9b7ec | 633 | * swap_cluster_readahead - swap in pages in hope we need them soon |
46017e95 | 634 | * @entry: swap entry of this memory |
7682486b | 635 | * @gfp_mask: memory allocation flags |
ddc1a5cb HD |
636 | * @mpol: NUMA memory allocation policy to be applied |
637 | * @ilx: NUMA interleave index, for use only when MPOL_INTERLEAVE | |
46017e95 | 638 | * |
a4575c41 | 639 | * Returns the struct folio for entry and addr, after queueing swapin. |
46017e95 HD |
640 | * |
641 | * Primitive swap readahead code. We simply read an aligned block of | |
642 | * (1 << page_cluster) entries in the swap area. This method is chosen | |
643 | * because it doesn't cost us any seek time. We also make sure to queue | |
644 | * the 'original' request together with the readahead ones... | |
645 | * | |
ddc1a5cb HD |
646 | * Note: it is intentional that the same NUMA policy and interleave index |
647 | * are used for every page of the readahead: neighbouring pages on swap | |
648 | * are fairly likely to have been swapped out from the same node. | |
46017e95 | 649 | */ |
a4575c41 | 650 | struct folio *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask, |
ddc1a5cb | 651 | struct mempolicy *mpol, pgoff_t ilx) |
46017e95 | 652 | { |
96c7b0b4 | 653 | struct folio *folio; |
579f8290 SL |
654 | unsigned long entry_offset = swp_offset(entry); |
655 | unsigned long offset = entry_offset; | |
67f96aa2 | 656 | unsigned long start_offset, end_offset; |
579f8290 | 657 | unsigned long mask; |
e9a6effa | 658 | struct swap_info_struct *si = swp_swap_info(entry); |
3fb5c298 | 659 | struct blk_plug plug; |
5169b844 | 660 | struct swap_iocb *splug = NULL; |
b243dcbf | 661 | bool page_allocated; |
46017e95 | 662 | |
579f8290 SL |
663 | mask = swapin_nr_pages(offset) - 1; |
664 | if (!mask) | |
665 | goto skip; | |
666 | ||
67f96aa2 RR |
667 | /* Read a page_cluster sized and aligned cluster around offset. */ |
668 | start_offset = offset & ~mask; | |
669 | end_offset = offset | mask; | |
670 | if (!start_offset) /* First page is swap header. */ | |
671 | start_offset++; | |
e9a6effa HY |
672 | if (end_offset >= si->max) |
673 | end_offset = si->max - 1; | |
67f96aa2 | 674 | |
3fb5c298 | 675 | blk_start_plug(&plug); |
67f96aa2 | 676 | for (offset = start_offset; offset <= end_offset ; offset++) { |
46017e95 | 677 | /* Ok, do the async read-ahead now */ |
96c7b0b4 | 678 | folio = __read_swap_cache_async( |
ddc1a5cb | 679 | swp_entry(swp_type(entry), offset), |
a65b0e76 | 680 | gfp_mask, mpol, ilx, &page_allocated, false); |
96c7b0b4 | 681 | if (!folio) |
67f96aa2 | 682 | continue; |
c4fa6309 | 683 | if (page_allocated) { |
c9bdf768 | 684 | swap_read_folio(folio, false, &splug); |
eaf649eb | 685 | if (offset != entry_offset) { |
96c7b0b4 | 686 | folio_set_readahead(folio); |
c4fa6309 HY |
687 | count_vm_event(SWAP_RA); |
688 | } | |
cbc65df2 | 689 | } |
96c7b0b4 | 690 | folio_put(folio); |
46017e95 | 691 | } |
3fb5c298 | 692 | blk_finish_plug(&plug); |
5169b844 | 693 | swap_read_unplug(splug); |
46017e95 | 694 | lru_add_drain(); /* Push any new pages onto the LRU now */ |
579f8290 | 695 | skip: |
5169b844 | 696 | /* The page was likely read above, so no need for plugging here */ |
96c7b0b4 | 697 | folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx, |
a65b0e76 | 698 | &page_allocated, false); |
16e96ba5 NP |
699 | if (unlikely(page_allocated)) { |
700 | zswap_folio_swapin(folio); | |
c9bdf768 | 701 | swap_read_folio(folio, false, NULL); |
16e96ba5 | 702 | } |
a4575c41 | 703 | return folio; |
46017e95 | 704 | } |
4b3ef9da HY |
705 | |
706 | int init_swap_address_space(unsigned int type, unsigned long nr_pages) | |
707 | { | |
708 | struct address_space *spaces, *space; | |
709 | unsigned int i, nr; | |
710 | ||
711 | nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES); | |
778e1cdd | 712 | spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL); |
4b3ef9da HY |
713 | if (!spaces) |
714 | return -ENOMEM; | |
715 | for (i = 0; i < nr; i++) { | |
716 | space = spaces + i; | |
a2833486 | 717 | xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ); |
4b3ef9da HY |
718 | atomic_set(&space->i_mmap_writable, 0); |
719 | space->a_ops = &swap_aops; | |
720 | /* swap cache doesn't use writeback related tags */ | |
721 | mapping_set_no_writeback_tags(space); | |
4b3ef9da HY |
722 | } |
723 | nr_swapper_spaces[type] = nr; | |
054f1d1f | 724 | swapper_spaces[type] = spaces; |
4b3ef9da HY |
725 | |
726 | return 0; | |
727 | } | |
728 | ||
729 | void exit_swap_address_space(unsigned int type) | |
730 | { | |
eea4a501 HY |
731 | int i; |
732 | struct address_space *spaces = swapper_spaces[type]; | |
733 | ||
734 | for (i = 0; i < nr_swapper_spaces[type]; i++) | |
735 | VM_WARN_ON_ONCE(!mapping_empty(&spaces[i])); | |
736 | kvfree(spaces); | |
4b3ef9da | 737 | nr_swapper_spaces[type] = 0; |
054f1d1f | 738 | swapper_spaces[type] = NULL; |
4b3ef9da | 739 | } |
ec560175 | 740 | |
4f8fcf4c HD |
741 | #define SWAP_RA_ORDER_CEILING 5 |
742 | ||
743 | struct vma_swap_readahead { | |
744 | unsigned short win; | |
745 | unsigned short offset; | |
746 | unsigned short nr_pte; | |
747 | }; | |
748 | ||
eaf649eb | 749 | static void swap_ra_info(struct vm_fault *vmf, |
16ba391e | 750 | struct vma_swap_readahead *ra_info) |
ec560175 HY |
751 | { |
752 | struct vm_area_struct *vma = vmf->vma; | |
eaf649eb | 753 | unsigned long ra_val; |
16ba391e | 754 | unsigned long faddr, pfn, fpfn, lpfn, rpfn; |
ec560175 | 755 | unsigned long start, end; |
16ba391e | 756 | unsigned int max_win, hits, prev_win, win; |
ec560175 | 757 | |
61b63972 HY |
758 | max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster), |
759 | SWAP_RA_ORDER_CEILING); | |
760 | if (max_win == 1) { | |
eaf649eb MK |
761 | ra_info->win = 1; |
762 | return; | |
61b63972 HY |
763 | } |
764 | ||
ec560175 | 765 | faddr = vmf->address; |
ec560175 | 766 | fpfn = PFN_DOWN(faddr); |
eaf649eb MK |
767 | ra_val = GET_SWAP_RA_VAL(vma); |
768 | pfn = PFN_DOWN(SWAP_RA_ADDR(ra_val)); | |
769 | prev_win = SWAP_RA_WIN(ra_val); | |
770 | hits = SWAP_RA_HITS(ra_val); | |
771 | ra_info->win = win = __swapin_nr_pages(pfn, fpfn, hits, | |
ec560175 HY |
772 | max_win, prev_win); |
773 | atomic_long_set(&vma->swap_readahead_info, | |
774 | SWAP_RA_VAL(faddr, win, 0)); | |
18ad72f5 | 775 | if (win == 1) |
eaf649eb | 776 | return; |
ec560175 | 777 | |
16ba391e KS |
778 | if (fpfn == pfn + 1) { |
779 | lpfn = fpfn; | |
780 | rpfn = fpfn + win; | |
781 | } else if (pfn == fpfn + 1) { | |
782 | lpfn = fpfn - win + 1; | |
783 | rpfn = fpfn + 1; | |
784 | } else { | |
785 | unsigned int left = (win - 1) / 2; | |
786 | ||
787 | lpfn = fpfn - left; | |
788 | rpfn = fpfn + win - left; | |
ec560175 | 789 | } |
16ba391e KS |
790 | start = max3(lpfn, PFN_DOWN(vma->vm_start), |
791 | PFN_DOWN(faddr & PMD_MASK)); | |
792 | end = min3(rpfn, PFN_DOWN(vma->vm_end), | |
793 | PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE)); | |
794 | ||
eaf649eb MK |
795 | ra_info->nr_pte = end - start; |
796 | ra_info->offset = fpfn - start; | |
ec560175 HY |
797 | } |
798 | ||
e9f59873 YS |
799 | /** |
800 | * swap_vma_readahead - swap in pages in hope we need them soon | |
ddc1a5cb | 801 | * @targ_entry: swap entry of the targeted memory |
e9f59873 | 802 | * @gfp_mask: memory allocation flags |
ddc1a5cb HD |
803 | * @mpol: NUMA memory allocation policy to be applied |
804 | * @targ_ilx: NUMA interleave index, for use only when MPOL_INTERLEAVE | |
e9f59873 YS |
805 | * @vmf: fault information |
806 | * | |
a4575c41 | 807 | * Returns the struct folio for entry and addr, after queueing swapin. |
e9f59873 | 808 | * |
cb152a1a | 809 | * Primitive swap readahead code. We simply read in a few pages whose |
e9f59873 YS |
810 | * virtual addresses are around the fault address in the same vma. |
811 | * | |
c1e8d7c6 | 812 | * Caller must hold read mmap_lock if vmf->vma is not NULL. |
e9f59873 YS |
813 | * |
814 | */ | |
a4575c41 MWO |
815 | static struct folio *swap_vma_readahead(swp_entry_t targ_entry, gfp_t gfp_mask, |
816 | struct mempolicy *mpol, pgoff_t targ_ilx, struct vm_fault *vmf) | |
ec560175 HY |
817 | { |
818 | struct blk_plug plug; | |
5169b844 | 819 | struct swap_iocb *splug = NULL; |
96c7b0b4 | 820 | struct folio *folio; |
4f8fcf4c HD |
821 | pte_t *pte = NULL, pentry; |
822 | unsigned long addr; | |
ec560175 | 823 | swp_entry_t entry; |
ddc1a5cb | 824 | pgoff_t ilx; |
ec560175 HY |
825 | unsigned int i; |
826 | bool page_allocated; | |
e97af699 ML |
827 | struct vma_swap_readahead ra_info = { |
828 | .win = 1, | |
829 | }; | |
ec560175 | 830 | |
eaf649eb MK |
831 | swap_ra_info(vmf, &ra_info); |
832 | if (ra_info.win == 1) | |
ec560175 HY |
833 | goto skip; |
834 | ||
4f8fcf4c | 835 | addr = vmf->address - (ra_info.offset * PAGE_SIZE); |
ddc1a5cb | 836 | ilx = targ_ilx - ra_info.offset; |
4f8fcf4c | 837 | |
ec560175 | 838 | blk_start_plug(&plug); |
ddc1a5cb | 839 | for (i = 0; i < ra_info.nr_pte; i++, ilx++, addr += PAGE_SIZE) { |
4f8fcf4c HD |
840 | if (!pte++) { |
841 | pte = pte_offset_map(vmf->pmd, addr); | |
842 | if (!pte) | |
843 | break; | |
844 | } | |
845 | pentry = ptep_get_lockless(pte); | |
92bafb20 | 846 | if (!is_swap_pte(pentry)) |
ec560175 HY |
847 | continue; |
848 | entry = pte_to_swp_entry(pentry); | |
849 | if (unlikely(non_swap_entry(entry))) | |
850 | continue; | |
4f8fcf4c HD |
851 | pte_unmap(pte); |
852 | pte = NULL; | |
96c7b0b4 | 853 | folio = __read_swap_cache_async(entry, gfp_mask, mpol, ilx, |
a65b0e76 | 854 | &page_allocated, false); |
96c7b0b4 | 855 | if (!folio) |
ec560175 HY |
856 | continue; |
857 | if (page_allocated) { | |
c9bdf768 | 858 | swap_read_folio(folio, false, &splug); |
eaf649eb | 859 | if (i != ra_info.offset) { |
96c7b0b4 | 860 | folio_set_readahead(folio); |
ec560175 HY |
861 | count_vm_event(SWAP_RA); |
862 | } | |
863 | } | |
96c7b0b4 | 864 | folio_put(folio); |
ec560175 | 865 | } |
4f8fcf4c HD |
866 | if (pte) |
867 | pte_unmap(pte); | |
ec560175 | 868 | blk_finish_plug(&plug); |
5169b844 | 869 | swap_read_unplug(splug); |
ec560175 HY |
870 | lru_add_drain(); |
871 | skip: | |
96c7b0b4 MWO |
872 | /* The folio was likely read above, so no need for plugging here */ |
873 | folio = __read_swap_cache_async(targ_entry, gfp_mask, mpol, targ_ilx, | |
a65b0e76 | 874 | &page_allocated, false); |
16e96ba5 NP |
875 | if (unlikely(page_allocated)) { |
876 | zswap_folio_swapin(folio); | |
c9bdf768 | 877 | swap_read_folio(folio, false, NULL); |
16e96ba5 | 878 | } |
a4575c41 | 879 | return folio; |
ec560175 | 880 | } |
d9bfcfdc | 881 | |
e9e9b7ec MK |
882 | /** |
883 | * swapin_readahead - swap in pages in hope we need them soon | |
884 | * @entry: swap entry of this memory | |
885 | * @gfp_mask: memory allocation flags | |
886 | * @vmf: fault information | |
887 | * | |
888 | * Returns the struct page for entry and addr, after queueing swapin. | |
889 | * | |
890 | * It's a main entry function for swap readahead. By the configuration, | |
891 | * it will read ahead blocks by cluster-based(ie, physical disk based) | |
892 | * or vma-based(ie, virtual address based on faulty address) readahead. | |
893 | */ | |
894 | struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask, | |
895 | struct vm_fault *vmf) | |
896 | { | |
ddc1a5cb HD |
897 | struct mempolicy *mpol; |
898 | pgoff_t ilx; | |
a4575c41 | 899 | struct folio *folio; |
ddc1a5cb HD |
900 | |
901 | mpol = get_vma_policy(vmf->vma, vmf->address, 0, &ilx); | |
a4575c41 | 902 | folio = swap_use_vma_readahead() ? |
ddc1a5cb HD |
903 | swap_vma_readahead(entry, gfp_mask, mpol, ilx, vmf) : |
904 | swap_cluster_readahead(entry, gfp_mask, mpol, ilx); | |
905 | mpol_cond_put(mpol); | |
a4575c41 MWO |
906 | |
907 | if (!folio) | |
908 | return NULL; | |
909 | return folio_file_page(folio, swp_offset(entry)); | |
e9e9b7ec MK |
910 | } |
911 | ||
d9bfcfdc HY |
912 | #ifdef CONFIG_SYSFS |
913 | static ssize_t vma_ra_enabled_show(struct kobject *kobj, | |
914 | struct kobj_attribute *attr, char *buf) | |
915 | { | |
ae7a927d JP |
916 | return sysfs_emit(buf, "%s\n", |
917 | enable_vma_readahead ? "true" : "false"); | |
d9bfcfdc HY |
918 | } |
919 | static ssize_t vma_ra_enabled_store(struct kobject *kobj, | |
920 | struct kobj_attribute *attr, | |
921 | const char *buf, size_t count) | |
922 | { | |
717aeab4 JG |
923 | ssize_t ret; |
924 | ||
925 | ret = kstrtobool(buf, &enable_vma_readahead); | |
926 | if (ret) | |
927 | return ret; | |
d9bfcfdc HY |
928 | |
929 | return count; | |
930 | } | |
6106b93e | 931 | static struct kobj_attribute vma_ra_enabled_attr = __ATTR_RW(vma_ra_enabled); |
d9bfcfdc | 932 | |
d9bfcfdc HY |
933 | static struct attribute *swap_attrs[] = { |
934 | &vma_ra_enabled_attr.attr, | |
d9bfcfdc HY |
935 | NULL, |
936 | }; | |
937 | ||
e48333b6 | 938 | static const struct attribute_group swap_attr_group = { |
d9bfcfdc HY |
939 | .attrs = swap_attrs, |
940 | }; | |
941 | ||
942 | static int __init swap_init_sysfs(void) | |
943 | { | |
944 | int err; | |
945 | struct kobject *swap_kobj; | |
946 | ||
947 | swap_kobj = kobject_create_and_add("swap", mm_kobj); | |
948 | if (!swap_kobj) { | |
949 | pr_err("failed to create swap kobject\n"); | |
950 | return -ENOMEM; | |
951 | } | |
952 | err = sysfs_create_group(swap_kobj, &swap_attr_group); | |
953 | if (err) { | |
954 | pr_err("failed to register swap group\n"); | |
955 | goto delete_obj; | |
956 | } | |
957 | return 0; | |
958 | ||
959 | delete_obj: | |
960 | kobject_put(swap_kobj); | |
961 | return err; | |
962 | } | |
963 | subsys_initcall(swap_init_sysfs); | |
964 | #endif |