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 LT |
12 | #include <linux/kernel_stat.h> |
13 | #include <linux/swap.h> | |
46017e95 | 14 | #include <linux/swapops.h> |
1da177e4 LT |
15 | #include <linux/init.h> |
16 | #include <linux/pagemap.h> | |
1da177e4 | 17 | #include <linux/backing-dev.h> |
3fb5c298 | 18 | #include <linux/blkdev.h> |
c484d410 | 19 | #include <linux/pagevec.h> |
b20a3503 | 20 | #include <linux/migrate.h> |
4b3ef9da | 21 | #include <linux/vmalloc.h> |
67afa38e | 22 | #include <linux/swap_slots.h> |
38d8b4e6 | 23 | #include <linux/huge_mm.h> |
61ef1865 | 24 | #include <linux/shmem_fs.h> |
243bce09 | 25 | #include "internal.h" |
014bb1de | 26 | #include "swap.h" |
1da177e4 LT |
27 | |
28 | /* | |
29 | * swapper_space is a fiction, retained to simplify the path through | |
7eaceacc | 30 | * vmscan's shrink_page_list. |
1da177e4 | 31 | */ |
f5e54d6e | 32 | static const struct address_space_operations swap_aops = { |
1da177e4 | 33 | .writepage = swap_writepage, |
4c4a7634 | 34 | .dirty_folio = noop_dirty_folio, |
1c93923c | 35 | #ifdef CONFIG_MIGRATION |
e965f963 | 36 | .migratepage = migrate_page, |
1c93923c | 37 | #endif |
1da177e4 LT |
38 | }; |
39 | ||
783cb68e CD |
40 | struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly; |
41 | static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly; | |
f5c754d6 | 42 | static bool enable_vma_readahead __read_mostly = true; |
ec560175 | 43 | |
ec560175 HY |
44 | #define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2) |
45 | #define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1) | |
46 | #define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK | |
47 | #define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK) | |
48 | ||
49 | #define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK) | |
50 | #define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT) | |
51 | #define SWAP_RA_ADDR(v) ((v) & PAGE_MASK) | |
52 | ||
53 | #define SWAP_RA_VAL(addr, win, hits) \ | |
54 | (((addr) & PAGE_MASK) | \ | |
55 | (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \ | |
56 | ((hits) & SWAP_RA_HITS_MASK)) | |
57 | ||
58 | /* Initial readahead hits is 4 to start up with a small window */ | |
59 | #define GET_SWAP_RA_VAL(vma) \ | |
60 | (atomic_long_read(&(vma)->swap_readahead_info) ? : 4) | |
1da177e4 | 61 | |
b96a3db2 QC |
62 | #define INC_CACHE_INFO(x) data_race(swap_cache_info.x++) |
63 | #define ADD_CACHE_INFO(x, nr) data_race(swap_cache_info.x += (nr)) | |
1da177e4 LT |
64 | |
65 | static struct { | |
66 | unsigned long add_total; | |
67 | unsigned long del_total; | |
68 | unsigned long find_success; | |
69 | unsigned long find_total; | |
1da177e4 LT |
70 | } swap_cache_info; |
71 | ||
579f8290 SL |
72 | static atomic_t swapin_readahead_hits = ATOMIC_INIT(4); |
73 | ||
1da177e4 LT |
74 | void show_swap_cache_info(void) |
75 | { | |
33806f06 | 76 | printk("%lu pages in swap cache\n", total_swapcache_pages()); |
2c97b7fc | 77 | printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n", |
1da177e4 | 78 | swap_cache_info.add_total, swap_cache_info.del_total, |
bb63be0a | 79 | swap_cache_info.find_success, swap_cache_info.find_total); |
ec8acf20 SL |
80 | printk("Free swap = %ldkB\n", |
81 | get_nr_swap_pages() << (PAGE_SHIFT - 10)); | |
1da177e4 LT |
82 | printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10)); |
83 | } | |
84 | ||
aae466b0 JK |
85 | void *get_shadow_from_swap_cache(swp_entry_t entry) |
86 | { | |
87 | struct address_space *address_space = swap_address_space(entry); | |
88 | pgoff_t idx = swp_offset(entry); | |
89 | struct page *page; | |
90 | ||
8c647dd1 | 91 | page = xa_load(&address_space->i_pages, idx); |
aae466b0 JK |
92 | if (xa_is_value(page)) |
93 | return page; | |
aae466b0 JK |
94 | return NULL; |
95 | } | |
96 | ||
1da177e4 | 97 | /* |
8d93b41c | 98 | * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space, |
1da177e4 LT |
99 | * but sets SwapCache flag and private instead of mapping and index. |
100 | */ | |
3852f676 JK |
101 | int add_to_swap_cache(struct page *page, swp_entry_t entry, |
102 | gfp_t gfp, void **shadowp) | |
1da177e4 | 103 | { |
8d93b41c | 104 | struct address_space *address_space = swap_address_space(entry); |
38d8b4e6 | 105 | pgoff_t idx = swp_offset(entry); |
8d93b41c | 106 | XA_STATE_ORDER(xas, &address_space->i_pages, idx, compound_order(page)); |
6c357848 | 107 | unsigned long i, nr = thp_nr_pages(page); |
3852f676 | 108 | void *old; |
1da177e4 | 109 | |
309381fe SL |
110 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
111 | VM_BUG_ON_PAGE(PageSwapCache(page), page); | |
112 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); | |
51726b12 | 113 | |
38d8b4e6 | 114 | page_ref_add(page, nr); |
31a56396 | 115 | SetPageSwapCache(page); |
31a56396 | 116 | |
8d93b41c MW |
117 | do { |
118 | xas_lock_irq(&xas); | |
119 | xas_create_range(&xas); | |
120 | if (xas_error(&xas)) | |
121 | goto unlock; | |
122 | for (i = 0; i < nr; i++) { | |
123 | VM_BUG_ON_PAGE(xas.xa_index != idx + i, page); | |
3852f676 JK |
124 | old = xas_load(&xas); |
125 | if (xa_is_value(old)) { | |
3852f676 JK |
126 | if (shadowp) |
127 | *shadowp = old; | |
128 | } | |
8d93b41c | 129 | set_page_private(page + i, entry.val + i); |
4101196b | 130 | xas_store(&xas, page); |
8d93b41c MW |
131 | xas_next(&xas); |
132 | } | |
38d8b4e6 HY |
133 | address_space->nrpages += nr; |
134 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); | |
b6038942 | 135 | __mod_lruvec_page_state(page, NR_SWAPCACHE, nr); |
38d8b4e6 | 136 | ADD_CACHE_INFO(add_total, nr); |
8d93b41c MW |
137 | unlock: |
138 | xas_unlock_irq(&xas); | |
139 | } while (xas_nomem(&xas, gfp)); | |
31a56396 | 140 | |
8d93b41c MW |
141 | if (!xas_error(&xas)) |
142 | return 0; | |
31a56396 | 143 | |
8d93b41c MW |
144 | ClearPageSwapCache(page); |
145 | page_ref_sub(page, nr); | |
146 | return xas_error(&xas); | |
1da177e4 LT |
147 | } |
148 | ||
1da177e4 LT |
149 | /* |
150 | * This must be called only on pages that have | |
151 | * been verified to be in the swap cache. | |
152 | */ | |
3852f676 JK |
153 | void __delete_from_swap_cache(struct page *page, |
154 | swp_entry_t entry, void *shadow) | |
1da177e4 | 155 | { |
4e17ec25 | 156 | struct address_space *address_space = swap_address_space(entry); |
6c357848 | 157 | int i, nr = thp_nr_pages(page); |
4e17ec25 MW |
158 | pgoff_t idx = swp_offset(entry); |
159 | XA_STATE(xas, &address_space->i_pages, idx); | |
33806f06 | 160 | |
309381fe SL |
161 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
162 | VM_BUG_ON_PAGE(!PageSwapCache(page), page); | |
163 | VM_BUG_ON_PAGE(PageWriteback(page), page); | |
1da177e4 | 164 | |
38d8b4e6 | 165 | for (i = 0; i < nr; i++) { |
3852f676 | 166 | void *entry = xas_store(&xas, shadow); |
4101196b | 167 | VM_BUG_ON_PAGE(entry != page, entry); |
38d8b4e6 | 168 | set_page_private(page + i, 0); |
4e17ec25 | 169 | xas_next(&xas); |
38d8b4e6 | 170 | } |
1da177e4 | 171 | ClearPageSwapCache(page); |
38d8b4e6 HY |
172 | address_space->nrpages -= nr; |
173 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); | |
b6038942 | 174 | __mod_lruvec_page_state(page, NR_SWAPCACHE, -nr); |
38d8b4e6 | 175 | ADD_CACHE_INFO(del_total, nr); |
1da177e4 LT |
176 | } |
177 | ||
178 | /** | |
179 | * add_to_swap - allocate swap space for a page | |
180 | * @page: page we want to move to swap | |
181 | * | |
182 | * Allocate swap space for the page and add the page to the | |
183 | * swap cache. Caller needs to hold the page lock. | |
184 | */ | |
0f074658 | 185 | int add_to_swap(struct page *page) |
1da177e4 LT |
186 | { |
187 | swp_entry_t entry; | |
1da177e4 LT |
188 | int err; |
189 | ||
309381fe SL |
190 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
191 | VM_BUG_ON_PAGE(!PageUptodate(page), page); | |
1da177e4 | 192 | |
38d8b4e6 | 193 | entry = get_swap_page(page); |
2ca4532a | 194 | if (!entry.val) |
0f074658 MK |
195 | return 0; |
196 | ||
2ca4532a | 197 | /* |
8d93b41c | 198 | * XArray node allocations from PF_MEMALLOC contexts could |
2ca4532a DN |
199 | * completely exhaust the page allocator. __GFP_NOMEMALLOC |
200 | * stops emergency reserves from being allocated. | |
201 | * | |
202 | * TODO: this could cause a theoretical memory reclaim | |
203 | * deadlock in the swap out path. | |
204 | */ | |
205 | /* | |
854e9ed0 | 206 | * Add it to the swap cache. |
2ca4532a DN |
207 | */ |
208 | err = add_to_swap_cache(page, entry, | |
3852f676 | 209 | __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN, NULL); |
38d8b4e6 | 210 | if (err) |
bd53b714 | 211 | /* |
2ca4532a DN |
212 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely |
213 | * clear SWAP_HAS_CACHE flag. | |
1da177e4 | 214 | */ |
0f074658 | 215 | goto fail; |
9625456c SL |
216 | /* |
217 | * Normally the page will be dirtied in unmap because its pte should be | |
0e9aa675 | 218 | * dirty. A special case is MADV_FREE page. The page's pte could have |
9625456c SL |
219 | * dirty bit cleared but the page's SwapBacked bit is still set because |
220 | * clearing the dirty bit and SwapBacked bit has no lock protected. For | |
221 | * such page, unmap will not set dirty bit for it, so page reclaim will | |
222 | * not write the page out. This can cause data corruption when the page | |
223 | * is swap in later. Always setting the dirty bit for the page solves | |
224 | * the problem. | |
225 | */ | |
226 | set_page_dirty(page); | |
38d8b4e6 HY |
227 | |
228 | return 1; | |
229 | ||
38d8b4e6 | 230 | fail: |
0f074658 | 231 | put_swap_page(page, entry); |
38d8b4e6 | 232 | return 0; |
1da177e4 LT |
233 | } |
234 | ||
235 | /* | |
236 | * This must be called only on pages that have | |
237 | * been verified to be in the swap cache and locked. | |
238 | * It will never put the page into the free list, | |
239 | * the caller has a reference on the page. | |
240 | */ | |
241 | void delete_from_swap_cache(struct page *page) | |
242 | { | |
4e17ec25 MW |
243 | swp_entry_t entry = { .val = page_private(page) }; |
244 | struct address_space *address_space = swap_address_space(entry); | |
1da177e4 | 245 | |
b93b0163 | 246 | xa_lock_irq(&address_space->i_pages); |
3852f676 | 247 | __delete_from_swap_cache(page, entry, NULL); |
b93b0163 | 248 | xa_unlock_irq(&address_space->i_pages); |
1da177e4 | 249 | |
75f6d6d2 | 250 | put_swap_page(page, entry); |
6c357848 | 251 | page_ref_sub(page, thp_nr_pages(page)); |
1da177e4 LT |
252 | } |
253 | ||
3852f676 JK |
254 | void clear_shadow_from_swap_cache(int type, unsigned long begin, |
255 | unsigned long end) | |
256 | { | |
257 | unsigned long curr = begin; | |
258 | void *old; | |
259 | ||
260 | for (;;) { | |
3852f676 JK |
261 | swp_entry_t entry = swp_entry(type, curr); |
262 | struct address_space *address_space = swap_address_space(entry); | |
263 | XA_STATE(xas, &address_space->i_pages, curr); | |
264 | ||
265 | xa_lock_irq(&address_space->i_pages); | |
266 | xas_for_each(&xas, old, end) { | |
267 | if (!xa_is_value(old)) | |
268 | continue; | |
269 | xas_store(&xas, NULL); | |
3852f676 | 270 | } |
3852f676 JK |
271 | xa_unlock_irq(&address_space->i_pages); |
272 | ||
273 | /* search the next swapcache until we meet end */ | |
274 | curr >>= SWAP_ADDRESS_SPACE_SHIFT; | |
275 | curr++; | |
276 | curr <<= SWAP_ADDRESS_SPACE_SHIFT; | |
277 | if (curr > end) | |
278 | break; | |
279 | } | |
280 | } | |
281 | ||
1da177e4 LT |
282 | /* |
283 | * If we are the only user, then try to free up the swap cache. | |
284 | * | |
285 | * Its ok to check for PageSwapCache without the page lock | |
a2c43eed HD |
286 | * here because we are going to recheck again inside |
287 | * try_to_free_swap() _with_ the lock. | |
1da177e4 LT |
288 | * - Marcelo |
289 | */ | |
f4c4a3f4 | 290 | void free_swap_cache(struct page *page) |
1da177e4 | 291 | { |
a2c43eed HD |
292 | if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) { |
293 | try_to_free_swap(page); | |
1da177e4 LT |
294 | unlock_page(page); |
295 | } | |
296 | } | |
297 | ||
298 | /* | |
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 | { | |
304 | free_swap_cache(page); | |
6fcb52a5 | 305 | if (!is_huge_zero_page(page)) |
770a5370 | 306 | put_page(page); |
1da177e4 LT |
307 | } |
308 | ||
309 | /* | |
310 | * Passed an array of pages, drop them all from swapcache and then release | |
311 | * them. They are removed from the LRU and freed if this is their last use. | |
312 | */ | |
313 | void free_pages_and_swap_cache(struct page **pages, int nr) | |
314 | { | |
1da177e4 | 315 | struct page **pagep = pages; |
aabfb572 | 316 | int i; |
1da177e4 LT |
317 | |
318 | lru_add_drain(); | |
aabfb572 MH |
319 | for (i = 0; i < nr; i++) |
320 | free_swap_cache(pagep[i]); | |
c6f92f9f | 321 | release_pages(pagep, nr); |
1da177e4 LT |
322 | } |
323 | ||
e9e9b7ec MK |
324 | static inline bool swap_use_vma_readahead(void) |
325 | { | |
326 | return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap); | |
327 | } | |
328 | ||
1da177e4 LT |
329 | /* |
330 | * Lookup a swap entry in the swap cache. A found page will be returned | |
331 | * unlocked and with its refcount incremented - we rely on the kernel | |
332 | * lock getting page table operations atomic even if we drop the page | |
333 | * lock before returning. | |
334 | */ | |
ec560175 HY |
335 | struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma, |
336 | unsigned long addr) | |
1da177e4 LT |
337 | { |
338 | struct page *page; | |
eb085574 | 339 | struct swap_info_struct *si; |
1da177e4 | 340 | |
eb085574 HY |
341 | si = get_swap_device(entry); |
342 | if (!si) | |
343 | return NULL; | |
f6ab1f7f | 344 | page = find_get_page(swap_address_space(entry), swp_offset(entry)); |
eb085574 | 345 | put_swap_device(si); |
1da177e4 | 346 | |
ec560175 HY |
347 | INC_CACHE_INFO(find_total); |
348 | if (page) { | |
eaf649eb MK |
349 | bool vma_ra = swap_use_vma_readahead(); |
350 | bool readahead; | |
351 | ||
1da177e4 | 352 | INC_CACHE_INFO(find_success); |
eaf649eb MK |
353 | /* |
354 | * At the moment, we don't support PG_readahead for anon THP | |
355 | * so let's bail out rather than confusing the readahead stat. | |
356 | */ | |
ec560175 HY |
357 | if (unlikely(PageTransCompound(page))) |
358 | return page; | |
eaf649eb | 359 | |
ec560175 | 360 | readahead = TestClearPageReadahead(page); |
eaf649eb MK |
361 | if (vma && vma_ra) { |
362 | unsigned long ra_val; | |
363 | int win, hits; | |
364 | ||
365 | ra_val = GET_SWAP_RA_VAL(vma); | |
366 | win = SWAP_RA_WIN(ra_val); | |
367 | hits = SWAP_RA_HITS(ra_val); | |
ec560175 HY |
368 | if (readahead) |
369 | hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX); | |
370 | atomic_long_set(&vma->swap_readahead_info, | |
371 | SWAP_RA_VAL(addr, win, hits)); | |
372 | } | |
eaf649eb | 373 | |
ec560175 | 374 | if (readahead) { |
cbc65df2 | 375 | count_vm_event(SWAP_RA_HIT); |
eaf649eb | 376 | if (!vma || !vma_ra) |
ec560175 | 377 | atomic_inc(&swapin_readahead_hits); |
cbc65df2 | 378 | } |
579f8290 | 379 | } |
eaf649eb | 380 | |
1da177e4 LT |
381 | return page; |
382 | } | |
383 | ||
61ef1865 MWO |
384 | /** |
385 | * find_get_incore_page - Find and get a page from the page or swap caches. | |
386 | * @mapping: The address_space to search. | |
387 | * @index: The page cache index. | |
388 | * | |
389 | * This differs from find_get_page() in that it will also look for the | |
390 | * page in the swap cache. | |
391 | * | |
392 | * Return: The found page or %NULL. | |
393 | */ | |
394 | struct page *find_get_incore_page(struct address_space *mapping, pgoff_t index) | |
395 | { | |
396 | swp_entry_t swp; | |
397 | struct swap_info_struct *si; | |
44835d20 MWO |
398 | struct page *page = pagecache_get_page(mapping, index, |
399 | FGP_ENTRY | FGP_HEAD, 0); | |
61ef1865 | 400 | |
a6de4b48 | 401 | if (!page) |
61ef1865 | 402 | return page; |
a6de4b48 MWO |
403 | if (!xa_is_value(page)) |
404 | return find_subpage(page, index); | |
61ef1865 MWO |
405 | if (!shmem_mapping(mapping)) |
406 | return NULL; | |
407 | ||
408 | swp = radix_to_swp_entry(page); | |
409 | /* Prevent swapoff from happening to us */ | |
410 | si = get_swap_device(swp); | |
411 | if (!si) | |
412 | return NULL; | |
413 | page = find_get_page(swap_address_space(swp), swp_offset(swp)); | |
414 | put_swap_device(si); | |
415 | return page; | |
416 | } | |
417 | ||
5b999aad DS |
418 | struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
419 | struct vm_area_struct *vma, unsigned long addr, | |
420 | bool *new_page_allocated) | |
1da177e4 | 421 | { |
eb085574 | 422 | struct swap_info_struct *si; |
4c6355b2 | 423 | struct page *page; |
aae466b0 | 424 | void *shadow = NULL; |
4c6355b2 | 425 | |
5b999aad | 426 | *new_page_allocated = false; |
1da177e4 | 427 | |
4c6355b2 JW |
428 | for (;;) { |
429 | int err; | |
1da177e4 LT |
430 | /* |
431 | * First check the swap cache. Since this is normally | |
432 | * called after lookup_swap_cache() failed, re-calling | |
433 | * that would confuse statistics. | |
434 | */ | |
eb085574 HY |
435 | si = get_swap_device(entry); |
436 | if (!si) | |
4c6355b2 JW |
437 | return NULL; |
438 | page = find_get_page(swap_address_space(entry), | |
439 | swp_offset(entry)); | |
eb085574 | 440 | put_swap_device(si); |
4c6355b2 JW |
441 | if (page) |
442 | return page; | |
1da177e4 | 443 | |
ba81f838 HY |
444 | /* |
445 | * Just skip read ahead for unused swap slot. | |
446 | * During swap_off when swap_slot_cache is disabled, | |
447 | * we have to handle the race between putting | |
448 | * swap entry in swap cache and marking swap slot | |
449 | * as SWAP_HAS_CACHE. That's done in later part of code or | |
450 | * else swap_off will be aborted if we return NULL. | |
451 | */ | |
452 | if (!__swp_swapcount(entry) && swap_slot_cache_enabled) | |
4c6355b2 | 453 | return NULL; |
e8c26ab6 | 454 | |
1da177e4 | 455 | /* |
4c6355b2 JW |
456 | * Get a new page to read into from swap. Allocate it now, |
457 | * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will | |
458 | * cause any racers to loop around until we add it to cache. | |
1da177e4 | 459 | */ |
4c6355b2 JW |
460 | page = alloc_page_vma(gfp_mask, vma, addr); |
461 | if (!page) | |
462 | return NULL; | |
1da177e4 | 463 | |
f000944d HD |
464 | /* |
465 | * Swap entry may have been freed since our caller observed it. | |
466 | */ | |
355cfa73 | 467 | err = swapcache_prepare(entry); |
4c6355b2 | 468 | if (!err) |
f000944d HD |
469 | break; |
470 | ||
4c6355b2 JW |
471 | put_page(page); |
472 | if (err != -EEXIST) | |
473 | return NULL; | |
474 | ||
2ca4532a | 475 | /* |
4c6355b2 JW |
476 | * We might race against __delete_from_swap_cache(), and |
477 | * stumble across a swap_map entry whose SWAP_HAS_CACHE | |
478 | * has not yet been cleared. Or race against another | |
479 | * __read_swap_cache_async(), which has set SWAP_HAS_CACHE | |
480 | * in swap_map, but not yet added its page to swap cache. | |
2ca4532a | 481 | */ |
029c4628 | 482 | schedule_timeout_uninterruptible(1); |
4c6355b2 JW |
483 | } |
484 | ||
485 | /* | |
486 | * The swap entry is ours to swap in. Prepare the new page. | |
487 | */ | |
488 | ||
489 | __SetPageLocked(page); | |
490 | __SetPageSwapBacked(page); | |
491 | ||
0add0c77 | 492 | if (mem_cgroup_swapin_charge_page(page, NULL, gfp_mask, entry)) |
4c6355b2 | 493 | goto fail_unlock; |
4c6355b2 | 494 | |
0add0c77 SB |
495 | /* May fail (-ENOMEM) if XArray node allocation failed. */ |
496 | if (add_to_swap_cache(page, entry, gfp_mask & GFP_RECLAIM_MASK, &shadow)) | |
4c6355b2 | 497 | goto fail_unlock; |
0add0c77 SB |
498 | |
499 | mem_cgroup_swapin_uncharge_swap(entry); | |
4c6355b2 | 500 | |
aae466b0 | 501 | if (shadow) |
0995d7e5 | 502 | workingset_refault(page_folio(page), shadow); |
314b57fb | 503 | |
4c6355b2 | 504 | /* Caller will initiate read into locked page */ |
6058eaec | 505 | lru_cache_add(page); |
4c6355b2 JW |
506 | *new_page_allocated = true; |
507 | return page; | |
1da177e4 | 508 | |
4c6355b2 | 509 | fail_unlock: |
0add0c77 | 510 | put_swap_page(page, entry); |
4c6355b2 JW |
511 | unlock_page(page); |
512 | put_page(page); | |
513 | return NULL; | |
1da177e4 | 514 | } |
46017e95 | 515 | |
5b999aad DS |
516 | /* |
517 | * Locate a page of swap in physical memory, reserving swap cache space | |
518 | * and reading the disk if it is not already cached. | |
519 | * A failure return means that either the page allocation failed or that | |
520 | * the swap entry is no longer in use. | |
521 | */ | |
522 | struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, | |
5169b844 N |
523 | struct vm_area_struct *vma, |
524 | unsigned long addr, bool do_poll, | |
525 | struct swap_iocb **plug) | |
5b999aad DS |
526 | { |
527 | bool page_was_allocated; | |
528 | struct page *retpage = __read_swap_cache_async(entry, gfp_mask, | |
529 | vma, addr, &page_was_allocated); | |
530 | ||
531 | if (page_was_allocated) | |
5169b844 | 532 | swap_readpage(retpage, do_poll, plug); |
5b999aad DS |
533 | |
534 | return retpage; | |
535 | } | |
536 | ||
ec560175 HY |
537 | static unsigned int __swapin_nr_pages(unsigned long prev_offset, |
538 | unsigned long offset, | |
539 | int hits, | |
540 | int max_pages, | |
541 | int prev_win) | |
579f8290 | 542 | { |
ec560175 | 543 | unsigned int pages, last_ra; |
579f8290 SL |
544 | |
545 | /* | |
546 | * This heuristic has been found to work well on both sequential and | |
547 | * random loads, swapping to hard disk or to SSD: please don't ask | |
548 | * what the "+ 2" means, it just happens to work well, that's all. | |
549 | */ | |
ec560175 | 550 | pages = hits + 2; |
579f8290 SL |
551 | if (pages == 2) { |
552 | /* | |
553 | * We can have no readahead hits to judge by: but must not get | |
554 | * stuck here forever, so check for an adjacent offset instead | |
555 | * (and don't even bother to check whether swap type is same). | |
556 | */ | |
557 | if (offset != prev_offset + 1 && offset != prev_offset - 1) | |
558 | pages = 1; | |
579f8290 SL |
559 | } else { |
560 | unsigned int roundup = 4; | |
561 | while (roundup < pages) | |
562 | roundup <<= 1; | |
563 | pages = roundup; | |
564 | } | |
565 | ||
566 | if (pages > max_pages) | |
567 | pages = max_pages; | |
568 | ||
569 | /* Don't shrink readahead too fast */ | |
ec560175 | 570 | last_ra = prev_win / 2; |
579f8290 SL |
571 | if (pages < last_ra) |
572 | pages = last_ra; | |
ec560175 HY |
573 | |
574 | return pages; | |
575 | } | |
576 | ||
577 | static unsigned long swapin_nr_pages(unsigned long offset) | |
578 | { | |
579 | static unsigned long prev_offset; | |
580 | unsigned int hits, pages, max_pages; | |
581 | static atomic_t last_readahead_pages; | |
582 | ||
583 | max_pages = 1 << READ_ONCE(page_cluster); | |
584 | if (max_pages <= 1) | |
585 | return 1; | |
586 | ||
587 | hits = atomic_xchg(&swapin_readahead_hits, 0); | |
d6c1f098 QC |
588 | pages = __swapin_nr_pages(READ_ONCE(prev_offset), offset, hits, |
589 | max_pages, | |
ec560175 HY |
590 | atomic_read(&last_readahead_pages)); |
591 | if (!hits) | |
d6c1f098 | 592 | WRITE_ONCE(prev_offset, offset); |
579f8290 SL |
593 | atomic_set(&last_readahead_pages, pages); |
594 | ||
595 | return pages; | |
596 | } | |
597 | ||
46017e95 | 598 | /** |
e9e9b7ec | 599 | * swap_cluster_readahead - swap in pages in hope we need them soon |
46017e95 | 600 | * @entry: swap entry of this memory |
7682486b | 601 | * @gfp_mask: memory allocation flags |
e9e9b7ec | 602 | * @vmf: fault information |
46017e95 HD |
603 | * |
604 | * Returns the struct page for entry and addr, after queueing swapin. | |
605 | * | |
606 | * Primitive swap readahead code. We simply read an aligned block of | |
607 | * (1 << page_cluster) entries in the swap area. This method is chosen | |
608 | * because it doesn't cost us any seek time. We also make sure to queue | |
609 | * the 'original' request together with the readahead ones... | |
610 | * | |
611 | * This has been extended to use the NUMA policies from the mm triggering | |
612 | * the readahead. | |
613 | * | |
c1e8d7c6 | 614 | * Caller must hold read mmap_lock if vmf->vma is not NULL. |
46017e95 | 615 | */ |
e9e9b7ec MK |
616 | struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask, |
617 | struct vm_fault *vmf) | |
46017e95 | 618 | { |
46017e95 | 619 | struct page *page; |
579f8290 SL |
620 | unsigned long entry_offset = swp_offset(entry); |
621 | unsigned long offset = entry_offset; | |
67f96aa2 | 622 | unsigned long start_offset, end_offset; |
579f8290 | 623 | unsigned long mask; |
e9a6effa | 624 | struct swap_info_struct *si = swp_swap_info(entry); |
3fb5c298 | 625 | struct blk_plug plug; |
5169b844 | 626 | struct swap_iocb *splug = NULL; |
c4fa6309 | 627 | bool do_poll = true, page_allocated; |
e9e9b7ec MK |
628 | struct vm_area_struct *vma = vmf->vma; |
629 | unsigned long addr = vmf->address; | |
46017e95 | 630 | |
579f8290 SL |
631 | mask = swapin_nr_pages(offset) - 1; |
632 | if (!mask) | |
633 | goto skip; | |
634 | ||
23955622 | 635 | do_poll = false; |
67f96aa2 RR |
636 | /* Read a page_cluster sized and aligned cluster around offset. */ |
637 | start_offset = offset & ~mask; | |
638 | end_offset = offset | mask; | |
639 | if (!start_offset) /* First page is swap header. */ | |
640 | start_offset++; | |
e9a6effa HY |
641 | if (end_offset >= si->max) |
642 | end_offset = si->max - 1; | |
67f96aa2 | 643 | |
3fb5c298 | 644 | blk_start_plug(&plug); |
67f96aa2 | 645 | for (offset = start_offset; offset <= end_offset ; offset++) { |
46017e95 | 646 | /* Ok, do the async read-ahead now */ |
c4fa6309 HY |
647 | page = __read_swap_cache_async( |
648 | swp_entry(swp_type(entry), offset), | |
649 | gfp_mask, vma, addr, &page_allocated); | |
46017e95 | 650 | if (!page) |
67f96aa2 | 651 | continue; |
c4fa6309 | 652 | if (page_allocated) { |
5169b844 | 653 | swap_readpage(page, false, &splug); |
eaf649eb | 654 | if (offset != entry_offset) { |
c4fa6309 HY |
655 | SetPageReadahead(page); |
656 | count_vm_event(SWAP_RA); | |
657 | } | |
cbc65df2 | 658 | } |
09cbfeaf | 659 | put_page(page); |
46017e95 | 660 | } |
3fb5c298 | 661 | blk_finish_plug(&plug); |
5169b844 | 662 | swap_read_unplug(splug); |
3fb5c298 | 663 | |
46017e95 | 664 | lru_add_drain(); /* Push any new pages onto the LRU now */ |
579f8290 | 665 | skip: |
5169b844 N |
666 | /* The page was likely read above, so no need for plugging here */ |
667 | return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll, NULL); | |
46017e95 | 668 | } |
4b3ef9da HY |
669 | |
670 | int init_swap_address_space(unsigned int type, unsigned long nr_pages) | |
671 | { | |
672 | struct address_space *spaces, *space; | |
673 | unsigned int i, nr; | |
674 | ||
675 | nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES); | |
778e1cdd | 676 | spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL); |
4b3ef9da HY |
677 | if (!spaces) |
678 | return -ENOMEM; | |
679 | for (i = 0; i < nr; i++) { | |
680 | space = spaces + i; | |
a2833486 | 681 | xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ); |
4b3ef9da HY |
682 | atomic_set(&space->i_mmap_writable, 0); |
683 | space->a_ops = &swap_aops; | |
684 | /* swap cache doesn't use writeback related tags */ | |
685 | mapping_set_no_writeback_tags(space); | |
4b3ef9da HY |
686 | } |
687 | nr_swapper_spaces[type] = nr; | |
054f1d1f | 688 | swapper_spaces[type] = spaces; |
4b3ef9da HY |
689 | |
690 | return 0; | |
691 | } | |
692 | ||
693 | void exit_swap_address_space(unsigned int type) | |
694 | { | |
eea4a501 HY |
695 | int i; |
696 | struct address_space *spaces = swapper_spaces[type]; | |
697 | ||
698 | for (i = 0; i < nr_swapper_spaces[type]; i++) | |
699 | VM_WARN_ON_ONCE(!mapping_empty(&spaces[i])); | |
700 | kvfree(spaces); | |
4b3ef9da | 701 | nr_swapper_spaces[type] = 0; |
054f1d1f | 702 | swapper_spaces[type] = NULL; |
4b3ef9da | 703 | } |
ec560175 HY |
704 | |
705 | static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma, | |
706 | unsigned long faddr, | |
707 | unsigned long lpfn, | |
708 | unsigned long rpfn, | |
709 | unsigned long *start, | |
710 | unsigned long *end) | |
711 | { | |
712 | *start = max3(lpfn, PFN_DOWN(vma->vm_start), | |
713 | PFN_DOWN(faddr & PMD_MASK)); | |
714 | *end = min3(rpfn, PFN_DOWN(vma->vm_end), | |
715 | PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE)); | |
716 | } | |
717 | ||
eaf649eb MK |
718 | static void swap_ra_info(struct vm_fault *vmf, |
719 | struct vma_swap_readahead *ra_info) | |
ec560175 HY |
720 | { |
721 | struct vm_area_struct *vma = vmf->vma; | |
eaf649eb | 722 | unsigned long ra_val; |
ec560175 HY |
723 | unsigned long faddr, pfn, fpfn; |
724 | unsigned long start, end; | |
eaf649eb | 725 | pte_t *pte, *orig_pte; |
ec560175 HY |
726 | unsigned int max_win, hits, prev_win, win, left; |
727 | #ifndef CONFIG_64BIT | |
728 | pte_t *tpte; | |
729 | #endif | |
730 | ||
61b63972 HY |
731 | max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster), |
732 | SWAP_RA_ORDER_CEILING); | |
733 | if (max_win == 1) { | |
eaf649eb MK |
734 | ra_info->win = 1; |
735 | return; | |
61b63972 HY |
736 | } |
737 | ||
ec560175 | 738 | faddr = vmf->address; |
eaf649eb | 739 | orig_pte = pte = pte_offset_map(vmf->pmd, faddr); |
ec560175 | 740 | |
ec560175 | 741 | fpfn = PFN_DOWN(faddr); |
eaf649eb MK |
742 | ra_val = GET_SWAP_RA_VAL(vma); |
743 | pfn = PFN_DOWN(SWAP_RA_ADDR(ra_val)); | |
744 | prev_win = SWAP_RA_WIN(ra_val); | |
745 | hits = SWAP_RA_HITS(ra_val); | |
746 | ra_info->win = win = __swapin_nr_pages(pfn, fpfn, hits, | |
ec560175 HY |
747 | max_win, prev_win); |
748 | atomic_long_set(&vma->swap_readahead_info, | |
749 | SWAP_RA_VAL(faddr, win, 0)); | |
750 | ||
eaf649eb MK |
751 | if (win == 1) { |
752 | pte_unmap(orig_pte); | |
753 | return; | |
754 | } | |
ec560175 HY |
755 | |
756 | /* Copy the PTEs because the page table may be unmapped */ | |
757 | if (fpfn == pfn + 1) | |
758 | swap_ra_clamp_pfn(vma, faddr, fpfn, fpfn + win, &start, &end); | |
759 | else if (pfn == fpfn + 1) | |
760 | swap_ra_clamp_pfn(vma, faddr, fpfn - win + 1, fpfn + 1, | |
761 | &start, &end); | |
762 | else { | |
763 | left = (win - 1) / 2; | |
764 | swap_ra_clamp_pfn(vma, faddr, fpfn - left, fpfn + win - left, | |
765 | &start, &end); | |
766 | } | |
eaf649eb MK |
767 | ra_info->nr_pte = end - start; |
768 | ra_info->offset = fpfn - start; | |
769 | pte -= ra_info->offset; | |
ec560175 | 770 | #ifdef CONFIG_64BIT |
eaf649eb | 771 | ra_info->ptes = pte; |
ec560175 | 772 | #else |
eaf649eb | 773 | tpte = ra_info->ptes; |
ec560175 HY |
774 | for (pfn = start; pfn != end; pfn++) |
775 | *tpte++ = *pte++; | |
776 | #endif | |
eaf649eb | 777 | pte_unmap(orig_pte); |
ec560175 HY |
778 | } |
779 | ||
e9f59873 YS |
780 | /** |
781 | * swap_vma_readahead - swap in pages in hope we need them soon | |
27ec4878 | 782 | * @fentry: swap entry of this memory |
e9f59873 YS |
783 | * @gfp_mask: memory allocation flags |
784 | * @vmf: fault information | |
785 | * | |
786 | * Returns the struct page for entry and addr, after queueing swapin. | |
787 | * | |
cb152a1a | 788 | * Primitive swap readahead code. We simply read in a few pages whose |
e9f59873 YS |
789 | * virtual addresses are around the fault address in the same vma. |
790 | * | |
c1e8d7c6 | 791 | * Caller must hold read mmap_lock if vmf->vma is not NULL. |
e9f59873 YS |
792 | * |
793 | */ | |
f5c754d6 CIK |
794 | static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask, |
795 | struct vm_fault *vmf) | |
ec560175 HY |
796 | { |
797 | struct blk_plug plug; | |
5169b844 | 798 | struct swap_iocb *splug = NULL; |
ec560175 HY |
799 | struct vm_area_struct *vma = vmf->vma; |
800 | struct page *page; | |
801 | pte_t *pte, pentry; | |
802 | swp_entry_t entry; | |
803 | unsigned int i; | |
804 | bool page_allocated; | |
e97af699 ML |
805 | struct vma_swap_readahead ra_info = { |
806 | .win = 1, | |
807 | }; | |
ec560175 | 808 | |
eaf649eb MK |
809 | swap_ra_info(vmf, &ra_info); |
810 | if (ra_info.win == 1) | |
ec560175 HY |
811 | goto skip; |
812 | ||
813 | blk_start_plug(&plug); | |
eaf649eb | 814 | for (i = 0, pte = ra_info.ptes; i < ra_info.nr_pte; |
ec560175 HY |
815 | i++, pte++) { |
816 | pentry = *pte; | |
817 | if (pte_none(pentry)) | |
818 | continue; | |
819 | if (pte_present(pentry)) | |
820 | continue; | |
821 | entry = pte_to_swp_entry(pentry); | |
822 | if (unlikely(non_swap_entry(entry))) | |
823 | continue; | |
824 | page = __read_swap_cache_async(entry, gfp_mask, vma, | |
825 | vmf->address, &page_allocated); | |
826 | if (!page) | |
827 | continue; | |
828 | if (page_allocated) { | |
5169b844 | 829 | swap_readpage(page, false, &splug); |
eaf649eb | 830 | if (i != ra_info.offset) { |
ec560175 HY |
831 | SetPageReadahead(page); |
832 | count_vm_event(SWAP_RA); | |
833 | } | |
834 | } | |
835 | put_page(page); | |
836 | } | |
837 | blk_finish_plug(&plug); | |
5169b844 | 838 | swap_read_unplug(splug); |
ec560175 HY |
839 | lru_add_drain(); |
840 | skip: | |
5169b844 | 841 | /* The page was likely read above, so no need for plugging here */ |
ec560175 | 842 | return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address, |
5169b844 | 843 | ra_info.win == 1, NULL); |
ec560175 | 844 | } |
d9bfcfdc | 845 | |
e9e9b7ec MK |
846 | /** |
847 | * swapin_readahead - swap in pages in hope we need them soon | |
848 | * @entry: swap entry of this memory | |
849 | * @gfp_mask: memory allocation flags | |
850 | * @vmf: fault information | |
851 | * | |
852 | * Returns the struct page for entry and addr, after queueing swapin. | |
853 | * | |
854 | * It's a main entry function for swap readahead. By the configuration, | |
855 | * it will read ahead blocks by cluster-based(ie, physical disk based) | |
856 | * or vma-based(ie, virtual address based on faulty address) readahead. | |
857 | */ | |
858 | struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask, | |
859 | struct vm_fault *vmf) | |
860 | { | |
861 | return swap_use_vma_readahead() ? | |
862 | swap_vma_readahead(entry, gfp_mask, vmf) : | |
863 | swap_cluster_readahead(entry, gfp_mask, vmf); | |
864 | } | |
865 | ||
d9bfcfdc HY |
866 | #ifdef CONFIG_SYSFS |
867 | static ssize_t vma_ra_enabled_show(struct kobject *kobj, | |
868 | struct kobj_attribute *attr, char *buf) | |
869 | { | |
ae7a927d JP |
870 | return sysfs_emit(buf, "%s\n", |
871 | enable_vma_readahead ? "true" : "false"); | |
d9bfcfdc HY |
872 | } |
873 | static ssize_t vma_ra_enabled_store(struct kobject *kobj, | |
874 | struct kobj_attribute *attr, | |
875 | const char *buf, size_t count) | |
876 | { | |
717aeab4 JG |
877 | ssize_t ret; |
878 | ||
879 | ret = kstrtobool(buf, &enable_vma_readahead); | |
880 | if (ret) | |
881 | return ret; | |
d9bfcfdc HY |
882 | |
883 | return count; | |
884 | } | |
885 | static struct kobj_attribute vma_ra_enabled_attr = | |
886 | __ATTR(vma_ra_enabled, 0644, vma_ra_enabled_show, | |
887 | vma_ra_enabled_store); | |
888 | ||
d9bfcfdc HY |
889 | static struct attribute *swap_attrs[] = { |
890 | &vma_ra_enabled_attr.attr, | |
d9bfcfdc HY |
891 | NULL, |
892 | }; | |
893 | ||
e48333b6 | 894 | static const struct attribute_group swap_attr_group = { |
d9bfcfdc HY |
895 | .attrs = swap_attrs, |
896 | }; | |
897 | ||
898 | static int __init swap_init_sysfs(void) | |
899 | { | |
900 | int err; | |
901 | struct kobject *swap_kobj; | |
902 | ||
903 | swap_kobj = kobject_create_and_add("swap", mm_kobj); | |
904 | if (!swap_kobj) { | |
905 | pr_err("failed to create swap kobject\n"); | |
906 | return -ENOMEM; | |
907 | } | |
908 | err = sysfs_create_group(swap_kobj, &swap_attr_group); | |
909 | if (err) { | |
910 | pr_err("failed to register swap group\n"); | |
911 | goto delete_obj; | |
912 | } | |
913 | return 0; | |
914 | ||
915 | delete_obj: | |
916 | kobject_put(swap_kobj); | |
917 | return err; | |
918 | } | |
919 | subsys_initcall(swap_init_sysfs); | |
920 | #endif |