blob: e166051566f42425c3eacccb1451bc47c88bd37b [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
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 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/mm.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090011#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/kernel_stat.h>
13#include <linux/swap.h>
Hugh Dickins46017e92008-02-04 22:28:41 -080014#include <linux/swapops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/init.h>
16#include <linux/pagemap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/backing-dev.h>
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -070018#include <linux/blkdev.h>
Hugh Dickinsc484d412006-01-06 00:10:55 -080019#include <linux/pagevec.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080020#include <linux/migrate.h>
Huang, Ying4b3ef9d2017-02-22 15:45:26 -080021#include <linux/vmalloc.h>
Tim Chen67afa382017-02-22 15:45:39 -080022#include <linux/swap_slots.h>
Huang Ying38d8b4e2017-07-06 15:37:18 -070023#include <linux/huge_mm.h>
Matthew Wilcox (Oracle)61ef1862020-10-13 16:51:17 -070024#include <linux/shmem_fs.h>
Hugh Dickins243bce02020-06-25 20:29:59 -070025#include "internal.h"
NeilBrown014bb1d2022-05-09 18:20:47 -070026#include "swap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
28/*
29 * swapper_space is a fiction, retained to simplify the path through
Jens Axboe7eaceac2011-03-10 08:52:07 +010030 * vmscan's shrink_page_list.
Linus Torvalds1da177e2005-04-16 15:20:36 -070031 */
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070032static const struct address_space_operations swap_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 .writepage = swap_writepage,
NeilBrown4c4a7632022-05-09 18:20:47 -070034 .dirty_folio = noop_dirty_folio,
Andrew Morton1c939232014-10-09 15:27:59 -070035#ifdef CONFIG_MIGRATION
Matthew Wilcox (Oracle)54184652022-06-06 10:27:41 -040036 .migrate_folio = migrate_folio,
Andrew Morton1c939232014-10-09 15:27:59 -070037#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070038};
39
Changbin Du783cb682017-11-15 17:36:06 -080040struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly;
41static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly;
Colin Ian Kingf5c754d2018-04-05 16:25:05 -070042static bool enable_vma_readahead __read_mostly = true;
Huang Yingec560172017-09-06 16:24:36 -070043
Huang Yingec560172017-09-06 16:24:36 -070044#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)
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Shaohua Li579f8292014-02-06 12:04:21 -080062static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064void show_swap_cache_info(void)
65{
Shaohua Li33806f02013-02-22 16:34:37 -080066 printk("%lu pages in swap cache\n", total_swapcache_pages());
Shaohua Liec8acf22013-02-22 16:34:38 -080067 printk("Free swap = %ldkB\n",
68 get_nr_swap_pages() << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
70}
71
Joonsoo Kimaae466b2020-08-11 18:30:50 -070072void *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);
76 struct page *page;
77
Matthew Wilcox (Oracle)8c647dd2021-02-25 17:15:33 -080078 page = xa_load(&address_space->i_pages, idx);
Joonsoo Kimaae466b2020-08-11 18:30:50 -070079 if (xa_is_value(page))
80 return page;
Joonsoo Kimaae466b2020-08-11 18:30:50 -070081 return NULL;
82}
83
Linus Torvalds1da177e2005-04-16 15:20:36 -070084/*
Matthew Wilcox (Oracle)2bb876b2022-06-01 15:13:59 -040085 * add_to_swap_cache resembles filemap_add_folio on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * but sets SwapCache flag and private instead of mapping and index.
87 */
Joonsoo Kim3852f672020-08-11 18:30:47 -070088int add_to_swap_cache(struct page *page, swp_entry_t entry,
89 gfp_t gfp, void **shadowp)
Linus Torvalds1da177e2005-04-16 15:20:36 -070090{
Matthew Wilcox8d93b412017-11-27 15:46:54 -050091 struct address_space *address_space = swap_address_space(entry);
Huang Ying38d8b4e2017-07-06 15:37:18 -070092 pgoff_t idx = swp_offset(entry);
Matthew Wilcox8d93b412017-11-27 15:46:54 -050093 XA_STATE_ORDER(xas, &address_space->i_pages, idx, compound_order(page));
Matthew Wilcox (Oracle)6c357842020-08-14 17:30:37 -070094 unsigned long i, nr = thp_nr_pages(page);
Joonsoo Kim3852f672020-08-11 18:30:47 -070095 void *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Sasha Levin309381fea2014-01-23 15:52:54 -080097 VM_BUG_ON_PAGE(!PageLocked(page), page);
98 VM_BUG_ON_PAGE(PageSwapCache(page), page);
99 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins51726b12009-01-06 14:39:25 -0800100
Huang Ying38d8b4e2017-07-06 15:37:18 -0700101 page_ref_add(page, nr);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700102 SetPageSwapCache(page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700103
Matthew Wilcox8d93b412017-11-27 15:46:54 -0500104 do {
105 xas_lock_irq(&xas);
106 xas_create_range(&xas);
107 if (xas_error(&xas))
108 goto unlock;
109 for (i = 0; i < nr; i++) {
110 VM_BUG_ON_PAGE(xas.xa_index != idx + i, page);
Joonsoo Kim3852f672020-08-11 18:30:47 -0700111 old = xas_load(&xas);
112 if (xa_is_value(old)) {
Joonsoo Kim3852f672020-08-11 18:30:47 -0700113 if (shadowp)
114 *shadowp = old;
115 }
Matthew Wilcox8d93b412017-11-27 15:46:54 -0500116 set_page_private(page + i, entry.val + i);
Matthew Wilcox (Oracle)41011962019-09-23 15:34:52 -0700117 xas_store(&xas, page);
Matthew Wilcox8d93b412017-11-27 15:46:54 -0500118 xas_next(&xas);
119 }
Huang Ying38d8b4e2017-07-06 15:37:18 -0700120 address_space->nrpages += nr;
121 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
Shakeel Buttb6038942021-02-24 12:03:55 -0800122 __mod_lruvec_page_state(page, NR_SWAPCACHE, nr);
Matthew Wilcox8d93b412017-11-27 15:46:54 -0500123unlock:
124 xas_unlock_irq(&xas);
125 } while (xas_nomem(&xas, gfp));
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700126
Matthew Wilcox8d93b412017-11-27 15:46:54 -0500127 if (!xas_error(&xas))
128 return 0;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700129
Matthew Wilcox8d93b412017-11-27 15:46:54 -0500130 ClearPageSwapCache(page);
131 page_ref_sub(page, nr);
132 return xas_error(&xas);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135/*
Matthew Wilcox (Oracle)ceff9d32022-06-17 18:50:20 +0100136 * This must be called only on folios that have
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 * been verified to be in the swap cache.
138 */
Matthew Wilcox (Oracle)ceff9d32022-06-17 18:50:20 +0100139void __delete_from_swap_cache(struct folio *folio,
Joonsoo Kim3852f672020-08-11 18:30:47 -0700140 swp_entry_t entry, void *shadow)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141{
Matthew Wilcox4e17ec22017-11-29 08:32:39 -0500142 struct address_space *address_space = swap_address_space(entry);
Matthew Wilcox (Oracle)ceff9d32022-06-17 18:50:20 +0100143 int i;
144 long nr = folio_nr_pages(folio);
Matthew Wilcox4e17ec22017-11-29 08:32:39 -0500145 pgoff_t idx = swp_offset(entry);
146 XA_STATE(xas, &address_space->i_pages, idx);
Shaohua Li33806f02013-02-22 16:34:37 -0800147
Matthew Wilcox (Oracle)ceff9d32022-06-17 18:50:20 +0100148 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
149 VM_BUG_ON_FOLIO(!folio_test_swapcache(folio), folio);
150 VM_BUG_ON_FOLIO(folio_test_writeback(folio), folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Huang Ying38d8b4e2017-07-06 15:37:18 -0700152 for (i = 0; i < nr; i++) {
Joonsoo Kim3852f672020-08-11 18:30:47 -0700153 void *entry = xas_store(&xas, shadow);
Matthew Wilcox (Oracle)ceff9d32022-06-17 18:50:20 +0100154 VM_BUG_ON_FOLIO(entry != folio, folio);
155 set_page_private(folio_page(folio, i), 0);
Matthew Wilcox4e17ec22017-11-29 08:32:39 -0500156 xas_next(&xas);
Huang Ying38d8b4e2017-07-06 15:37:18 -0700157 }
Matthew Wilcox (Oracle)ceff9d32022-06-17 18:50:20 +0100158 folio_clear_swapcache(folio);
Huang Ying38d8b4e2017-07-06 15:37:18 -0700159 address_space->nrpages -= nr;
Matthew Wilcox (Oracle)ceff9d32022-06-17 18:50:20 +0100160 __node_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
161 __lruvec_stat_mod_folio(folio, NR_SWAPCACHE, -nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
164/**
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700165 * add_to_swap - allocate swap space for a folio
166 * @folio: folio we want to move to swap
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 *
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700168 * Allocate swap space for the folio and add the folio to the
169 * swap cache.
170 *
171 * Context: Caller needs to hold the folio lock.
172 * Return: Whether the folio was added to the swap cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 */
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700174bool add_to_swap(struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175{
176 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 int err;
178
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700179 VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
180 VM_BUG_ON_FOLIO(!folio_test_uptodate(folio), folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
Matthew Wilcox (Oracle)e2e3fdc2022-05-12 20:23:02 -0700182 entry = folio_alloc_swap(folio);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700183 if (!entry.val)
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700184 return false;
Minchan Kim0f074652017-07-06 15:37:24 -0700185
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700186 /*
Matthew Wilcox8d93b412017-11-27 15:46:54 -0500187 * XArray node allocations from PF_MEMALLOC contexts could
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700188 * completely exhaust the page allocator. __GFP_NOMEMALLOC
189 * stops emergency reserves from being allocated.
190 *
191 * TODO: this could cause a theoretical memory reclaim
192 * deadlock in the swap out path.
193 */
194 /*
Minchan Kim854e9ed2016-01-15 16:54:53 -0800195 * Add it to the swap cache.
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700196 */
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700197 err = add_to_swap_cache(&folio->page, entry,
Joonsoo Kim3852f672020-08-11 18:30:47 -0700198 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN, NULL);
Huang Ying38d8b4e2017-07-06 15:37:18 -0700199 if (err)
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700200 /*
201 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
202 * clear SWAP_HAS_CACHE flag.
203 */
Minchan Kim0f074652017-07-06 15:37:24 -0700204 goto fail;
Shaohua Li96254562017-10-03 16:15:32 -0700205 /*
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700206 * Normally the folio will be dirtied in unmap because its
207 * pte should be dirty. A special case is MADV_FREE page. The
208 * page's pte could have dirty bit cleared but the folio's
209 * SwapBacked flag is still set because clearing the dirty bit
210 * and SwapBacked flag has no lock protected. For such folio,
211 * unmap will not set dirty bit for it, so folio reclaim will
212 * not write the folio out. This can cause data corruption when
213 * the folio is swapped in later. Always setting the dirty flag
214 * for the folio solves the problem.
Shaohua Li96254562017-10-03 16:15:32 -0700215 */
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700216 folio_mark_dirty(folio);
Huang Ying38d8b4e2017-07-06 15:37:18 -0700217
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700218 return true;
Huang Ying38d8b4e2017-07-06 15:37:18 -0700219
Huang Ying38d8b4e2017-07-06 15:37:18 -0700220fail:
Matthew Wilcox (Oracle)09c02e52022-05-12 20:23:02 -0700221 put_swap_page(&folio->page, entry);
222 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
224
225/*
Matthew Wilcox (Oracle)75fa68a2022-06-17 18:50:19 +0100226 * This must be called only on folios that have
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 * been verified to be in the swap cache and locked.
Matthew Wilcox (Oracle)75fa68a2022-06-17 18:50:19 +0100228 * It will never put the folio into the free list,
229 * the caller has a reference on the folio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 */
Matthew Wilcox (Oracle)75fa68a2022-06-17 18:50:19 +0100231void delete_from_swap_cache(struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232{
Matthew Wilcox (Oracle)75fa68a2022-06-17 18:50:19 +0100233 swp_entry_t entry = folio_swap_entry(folio);
Matthew Wilcox4e17ec22017-11-29 08:32:39 -0500234 struct address_space *address_space = swap_address_space(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700236 xa_lock_irq(&address_space->i_pages);
Matthew Wilcox (Oracle)ceff9d32022-06-17 18:50:20 +0100237 __delete_from_swap_cache(folio, entry, NULL);
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700238 xa_unlock_irq(&address_space->i_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Matthew Wilcox (Oracle)75fa68a2022-06-17 18:50:19 +0100240 put_swap_page(&folio->page, entry);
241 folio_ref_sub(folio, folio_nr_pages(folio));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Joonsoo Kim3852f672020-08-11 18:30:47 -0700244void clear_shadow_from_swap_cache(int type, unsigned long begin,
245 unsigned long end)
246{
247 unsigned long curr = begin;
248 void *old;
249
250 for (;;) {
Joonsoo Kim3852f672020-08-11 18:30:47 -0700251 swp_entry_t entry = swp_entry(type, curr);
252 struct address_space *address_space = swap_address_space(entry);
253 XA_STATE(xas, &address_space->i_pages, curr);
254
255 xa_lock_irq(&address_space->i_pages);
256 xas_for_each(&xas, old, end) {
257 if (!xa_is_value(old))
258 continue;
259 xas_store(&xas, NULL);
Joonsoo Kim3852f672020-08-11 18:30:47 -0700260 }
Joonsoo Kim3852f672020-08-11 18:30:47 -0700261 xa_unlock_irq(&address_space->i_pages);
262
263 /* search the next swapcache until we meet end */
264 curr >>= SWAP_ADDRESS_SPACE_SHIFT;
265 curr++;
266 curr <<= SWAP_ADDRESS_SPACE_SHIFT;
267 if (curr > end)
268 break;
269 }
270}
271
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272/*
273 * If we are the only user, then try to free up the swap cache.
274 *
275 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800276 * here because we are going to recheck again inside
277 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 * - Marcelo
279 */
Huang Yingf4c4a3f2021-06-28 19:37:12 -0700280void free_swap_cache(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800282 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
283 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 unlock_page(page);
285 }
286}
287
288/*
289 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700290 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 */
292void free_page_and_swap_cache(struct page *page)
293{
294 free_swap_cache(page);
Aaron Lu6fcb52a2016-10-07 17:00:08 -0700295 if (!is_huge_zero_page(page))
Gerald Schaefer770a5372016-06-08 15:33:50 -0700296 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297}
298
299/*
300 * Passed an array of pages, drop them all from swapcache and then release
301 * them. They are removed from the LRU and freed if this is their last use.
302 */
303void free_pages_and_swap_cache(struct page **pages, int nr)
304{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 struct page **pagep = pages;
Michal Hockoaabfb572014-10-09 15:28:52 -0700306 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307
308 lru_add_drain();
Michal Hockoaabfb572014-10-09 15:28:52 -0700309 for (i = 0; i < nr; i++)
310 free_swap_cache(pagep[i]);
Mel Gormanc6f92f92017-11-15 17:37:55 -0800311 release_pages(pagep, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312}
313
Minchan Kime9e9b7e2018-04-05 16:23:42 -0700314static inline bool swap_use_vma_readahead(void)
315{
316 return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap);
317}
318
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319/*
320 * Lookup a swap entry in the swap cache. A found page will be returned
321 * unlocked and with its refcount incremented - we rely on the kernel
322 * lock getting page table operations atomic even if we drop the page
323 * lock before returning.
324 */
Huang Yingec560172017-09-06 16:24:36 -0700325struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma,
326 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327{
328 struct page *page;
Huang Yingeb085572019-07-11 20:55:33 -0700329 struct swap_info_struct *si;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
Huang Yingeb085572019-07-11 20:55:33 -0700331 si = get_swap_device(entry);
332 if (!si)
333 return NULL;
Huang Yingf6ab1f72016-10-07 17:00:21 -0700334 page = find_get_page(swap_address_space(entry), swp_offset(entry));
Huang Yingeb085572019-07-11 20:55:33 -0700335 put_swap_device(si);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
Huang Yingec560172017-09-06 16:24:36 -0700337 if (page) {
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700338 bool vma_ra = swap_use_vma_readahead();
339 bool readahead;
340
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700341 /*
342 * At the moment, we don't support PG_readahead for anon THP
343 * so let's bail out rather than confusing the readahead stat.
344 */
Huang Yingec560172017-09-06 16:24:36 -0700345 if (unlikely(PageTransCompound(page)))
346 return page;
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700347
Huang Yingec560172017-09-06 16:24:36 -0700348 readahead = TestClearPageReadahead(page);
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700349 if (vma && vma_ra) {
350 unsigned long ra_val;
351 int win, hits;
352
353 ra_val = GET_SWAP_RA_VAL(vma);
354 win = SWAP_RA_WIN(ra_val);
355 hits = SWAP_RA_HITS(ra_val);
Huang Yingec560172017-09-06 16:24:36 -0700356 if (readahead)
357 hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX);
358 atomic_long_set(&vma->swap_readahead_info,
359 SWAP_RA_VAL(addr, win, hits));
360 }
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700361
Huang Yingec560172017-09-06 16:24:36 -0700362 if (readahead) {
Huang Yingcbc65df2017-09-06 16:24:29 -0700363 count_vm_event(SWAP_RA_HIT);
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700364 if (!vma || !vma_ra)
Huang Yingec560172017-09-06 16:24:36 -0700365 atomic_inc(&swapin_readahead_hits);
Huang Yingcbc65df2017-09-06 16:24:29 -0700366 }
Shaohua Li579f8292014-02-06 12:04:21 -0800367 }
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700368
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 return page;
370}
371
Matthew Wilcox (Oracle)61ef1862020-10-13 16:51:17 -0700372/**
373 * find_get_incore_page - Find and get a page from the page or swap caches.
374 * @mapping: The address_space to search.
375 * @index: The page cache index.
376 *
377 * This differs from find_get_page() in that it will also look for the
378 * page in the swap cache.
379 *
380 * Return: The found page or %NULL.
381 */
382struct page *find_get_incore_page(struct address_space *mapping, pgoff_t index)
383{
384 swp_entry_t swp;
385 struct swap_info_struct *si;
Matthew Wilcox (Oracle)44835d22021-02-25 17:15:36 -0800386 struct page *page = pagecache_get_page(mapping, index,
387 FGP_ENTRY | FGP_HEAD, 0);
Matthew Wilcox (Oracle)61ef1862020-10-13 16:51:17 -0700388
Matthew Wilcox (Oracle)a6de4b42020-10-13 16:51:34 -0700389 if (!page)
Matthew Wilcox (Oracle)61ef1862020-10-13 16:51:17 -0700390 return page;
Matthew Wilcox (Oracle)a6de4b42020-10-13 16:51:34 -0700391 if (!xa_is_value(page))
392 return find_subpage(page, index);
Matthew Wilcox (Oracle)61ef1862020-10-13 16:51:17 -0700393 if (!shmem_mapping(mapping))
394 return NULL;
395
396 swp = radix_to_swp_entry(page);
Miaohe Linba6851b2022-05-19 20:50:30 +0800397 /* There might be swapin error entries in shmem mapping. */
398 if (non_swap_entry(swp))
399 return NULL;
Matthew Wilcox (Oracle)61ef1862020-10-13 16:51:17 -0700400 /* Prevent swapoff from happening to us */
401 si = get_swap_device(swp);
402 if (!si)
403 return NULL;
404 page = find_get_page(swap_address_space(swp), swp_offset(swp));
405 put_swap_device(si);
406 return page;
407}
408
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700409struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
410 struct vm_area_struct *vma, unsigned long addr,
411 bool *new_page_allocated)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412{
Huang Yingeb085572019-07-11 20:55:33 -0700413 struct swap_info_struct *si;
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700414 struct page *page;
Joonsoo Kimaae466b2020-08-11 18:30:50 -0700415 void *shadow = NULL;
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700416
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700417 *new_page_allocated = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700419 for (;;) {
420 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 /*
422 * First check the swap cache. Since this is normally
423 * called after lookup_swap_cache() failed, re-calling
424 * that would confuse statistics.
425 */
Huang Yingeb085572019-07-11 20:55:33 -0700426 si = get_swap_device(entry);
427 if (!si)
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700428 return NULL;
429 page = find_get_page(swap_address_space(entry),
430 swp_offset(entry));
Huang Yingeb085572019-07-11 20:55:33 -0700431 put_swap_device(si);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700432 if (page)
433 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434
Huang Yingba81f832017-02-22 15:45:46 -0800435 /*
436 * Just skip read ahead for unused swap slot.
437 * During swap_off when swap_slot_cache is disabled,
438 * we have to handle the race between putting
439 * swap entry in swap cache and marking swap slot
440 * as SWAP_HAS_CACHE. That's done in later part of code or
441 * else swap_off will be aborted if we return NULL.
442 */
443 if (!__swp_swapcount(entry) && swap_slot_cache_enabled)
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700444 return NULL;
Tim Chene8c26ab2017-02-22 15:45:29 -0800445
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 /*
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700447 * Get a new page to read into from swap. Allocate it now,
448 * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will
449 * cause any racers to loop around until we add it to cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 */
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700451 page = alloc_page_vma(gfp_mask, vma, addr);
452 if (!page)
453 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
455 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800456 * Swap entry may have been freed since our caller observed it.
457 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700458 err = swapcache_prepare(entry);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700459 if (!err)
Hugh Dickinsf0009442008-02-04 22:28:49 -0800460 break;
461
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700462 put_page(page);
463 if (err != -EEXIST)
464 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700466 /*
467 * We might race against __delete_from_swap_cache(), and
468 * stumble across a swap_map entry whose SWAP_HAS_CACHE
469 * has not yet been cleared. Or race against another
470 * __read_swap_cache_async(), which has set SWAP_HAS_CACHE
471 * in swap_map, but not yet added its page to swap cache.
472 */
Guo Ziliang029c4622022-03-16 16:15:03 -0700473 schedule_timeout_uninterruptible(1);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700474 }
475
476 /*
477 * The swap entry is ours to swap in. Prepare the new page.
478 */
479
480 __SetPageLocked(page);
481 __SetPageSwapBacked(page);
482
Shakeel Butt0add0c72021-04-29 22:56:36 -0700483 if (mem_cgroup_swapin_charge_page(page, NULL, gfp_mask, entry))
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700484 goto fail_unlock;
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700485
Shakeel Butt0add0c72021-04-29 22:56:36 -0700486 /* May fail (-ENOMEM) if XArray node allocation failed. */
487 if (add_to_swap_cache(page, entry, gfp_mask & GFP_RECLAIM_MASK, &shadow))
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700488 goto fail_unlock;
Shakeel Butt0add0c72021-04-29 22:56:36 -0700489
490 mem_cgroup_swapin_uncharge_swap(entry);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700491
Joonsoo Kimaae466b2020-08-11 18:30:50 -0700492 if (shadow)
Matthew Wilcox (Oracle)0995d7e2021-04-29 10:27:16 -0400493 workingset_refault(page_folio(page), shadow);
Johannes Weiner314b57f2020-06-03 16:03:03 -0700494
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700495 /* Caller will initiate read into locked page */
Johannes Weiner6058eae2020-06-03 16:02:40 -0700496 lru_cache_add(page);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700497 *new_page_allocated = true;
498 return page;
499
500fail_unlock:
Shakeel Butt0add0c72021-04-29 22:56:36 -0700501 put_swap_page(page, entry);
Johannes Weiner4c6355b2020-06-03 16:02:17 -0700502 unlock_page(page);
503 put_page(page);
504 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
Hugh Dickins46017e92008-02-04 22:28:41 -0800506
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700507/*
508 * Locate a page of swap in physical memory, reserving swap cache space
509 * and reading the disk if it is not already cached.
510 * A failure return means that either the page allocation failed or that
511 * the swap entry is no longer in use.
512 */
513struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
NeilBrown5169b842022-05-09 18:20:49 -0700514 struct vm_area_struct *vma,
515 unsigned long addr, bool do_poll,
516 struct swap_iocb **plug)
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700517{
518 bool page_was_allocated;
519 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
520 vma, addr, &page_was_allocated);
521
522 if (page_was_allocated)
NeilBrown5169b842022-05-09 18:20:49 -0700523 swap_readpage(retpage, do_poll, plug);
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700524
525 return retpage;
526}
527
Huang Yingec560172017-09-06 16:24:36 -0700528static unsigned int __swapin_nr_pages(unsigned long prev_offset,
529 unsigned long offset,
530 int hits,
531 int max_pages,
532 int prev_win)
Shaohua Li579f8292014-02-06 12:04:21 -0800533{
Huang Yingec560172017-09-06 16:24:36 -0700534 unsigned int pages, last_ra;
Shaohua Li579f8292014-02-06 12:04:21 -0800535
536 /*
537 * This heuristic has been found to work well on both sequential and
538 * random loads, swapping to hard disk or to SSD: please don't ask
539 * what the "+ 2" means, it just happens to work well, that's all.
540 */
Huang Yingec560172017-09-06 16:24:36 -0700541 pages = hits + 2;
Shaohua Li579f8292014-02-06 12:04:21 -0800542 if (pages == 2) {
543 /*
544 * We can have no readahead hits to judge by: but must not get
545 * stuck here forever, so check for an adjacent offset instead
546 * (and don't even bother to check whether swap type is same).
547 */
548 if (offset != prev_offset + 1 && offset != prev_offset - 1)
549 pages = 1;
Shaohua Li579f8292014-02-06 12:04:21 -0800550 } else {
551 unsigned int roundup = 4;
552 while (roundup < pages)
553 roundup <<= 1;
554 pages = roundup;
555 }
556
557 if (pages > max_pages)
558 pages = max_pages;
559
560 /* Don't shrink readahead too fast */
Huang Yingec560172017-09-06 16:24:36 -0700561 last_ra = prev_win / 2;
Shaohua Li579f8292014-02-06 12:04:21 -0800562 if (pages < last_ra)
563 pages = last_ra;
Huang Yingec560172017-09-06 16:24:36 -0700564
565 return pages;
566}
567
568static unsigned long swapin_nr_pages(unsigned long offset)
569{
570 static unsigned long prev_offset;
571 unsigned int hits, pages, max_pages;
572 static atomic_t last_readahead_pages;
573
574 max_pages = 1 << READ_ONCE(page_cluster);
575 if (max_pages <= 1)
576 return 1;
577
578 hits = atomic_xchg(&swapin_readahead_hits, 0);
Qian Caid6c1f092020-06-01 21:48:40 -0700579 pages = __swapin_nr_pages(READ_ONCE(prev_offset), offset, hits,
580 max_pages,
Huang Yingec560172017-09-06 16:24:36 -0700581 atomic_read(&last_readahead_pages));
582 if (!hits)
Qian Caid6c1f092020-06-01 21:48:40 -0700583 WRITE_ONCE(prev_offset, offset);
Shaohua Li579f8292014-02-06 12:04:21 -0800584 atomic_set(&last_readahead_pages, pages);
585
586 return pages;
587}
588
Hugh Dickins46017e92008-02-04 22:28:41 -0800589/**
Minchan Kime9e9b7e2018-04-05 16:23:42 -0700590 * swap_cluster_readahead - swap in pages in hope we need them soon
Hugh Dickins46017e92008-02-04 22:28:41 -0800591 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700592 * @gfp_mask: memory allocation flags
Minchan Kime9e9b7e2018-04-05 16:23:42 -0700593 * @vmf: fault information
Hugh Dickins46017e92008-02-04 22:28:41 -0800594 *
595 * Returns the struct page for entry and addr, after queueing swapin.
596 *
597 * Primitive swap readahead code. We simply read an aligned block of
598 * (1 << page_cluster) entries in the swap area. This method is chosen
599 * because it doesn't cost us any seek time. We also make sure to queue
600 * the 'original' request together with the readahead ones...
601 *
602 * This has been extended to use the NUMA policies from the mm triggering
603 * the readahead.
604 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700605 * Caller must hold read mmap_lock if vmf->vma is not NULL.
Hugh Dickins46017e92008-02-04 22:28:41 -0800606 */
Minchan Kime9e9b7e2018-04-05 16:23:42 -0700607struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask,
608 struct vm_fault *vmf)
Hugh Dickins46017e92008-02-04 22:28:41 -0800609{
Hugh Dickins46017e92008-02-04 22:28:41 -0800610 struct page *page;
Shaohua Li579f8292014-02-06 12:04:21 -0800611 unsigned long entry_offset = swp_offset(entry);
612 unsigned long offset = entry_offset;
Rik van Riel67f96aa2012-03-21 16:33:50 -0700613 unsigned long start_offset, end_offset;
Shaohua Li579f8292014-02-06 12:04:21 -0800614 unsigned long mask;
Huang Yinge9a6eff2017-11-15 17:33:15 -0800615 struct swap_info_struct *si = swp_swap_info(entry);
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700616 struct blk_plug plug;
NeilBrown5169b842022-05-09 18:20:49 -0700617 struct swap_iocb *splug = NULL;
Huang Yingc4fa6302017-09-06 16:24:33 -0700618 bool do_poll = true, page_allocated;
Minchan Kime9e9b7e2018-04-05 16:23:42 -0700619 struct vm_area_struct *vma = vmf->vma;
620 unsigned long addr = vmf->address;
Hugh Dickins46017e92008-02-04 22:28:41 -0800621
Shaohua Li579f8292014-02-06 12:04:21 -0800622 mask = swapin_nr_pages(offset) - 1;
623 if (!mask)
624 goto skip;
625
Shaohua Li23955622017-07-10 15:47:11 -0700626 do_poll = false;
Rik van Riel67f96aa2012-03-21 16:33:50 -0700627 /* Read a page_cluster sized and aligned cluster around offset. */
628 start_offset = offset & ~mask;
629 end_offset = offset | mask;
630 if (!start_offset) /* First page is swap header. */
631 start_offset++;
Huang Yinge9a6eff2017-11-15 17:33:15 -0800632 if (end_offset >= si->max)
633 end_offset = si->max - 1;
Rik van Riel67f96aa2012-03-21 16:33:50 -0700634
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700635 blk_start_plug(&plug);
Rik van Riel67f96aa2012-03-21 16:33:50 -0700636 for (offset = start_offset; offset <= end_offset ; offset++) {
Hugh Dickins46017e92008-02-04 22:28:41 -0800637 /* Ok, do the async read-ahead now */
Huang Yingc4fa6302017-09-06 16:24:33 -0700638 page = __read_swap_cache_async(
639 swp_entry(swp_type(entry), offset),
640 gfp_mask, vma, addr, &page_allocated);
Hugh Dickins46017e92008-02-04 22:28:41 -0800641 if (!page)
Rik van Riel67f96aa2012-03-21 16:33:50 -0700642 continue;
Huang Yingc4fa6302017-09-06 16:24:33 -0700643 if (page_allocated) {
NeilBrown5169b842022-05-09 18:20:49 -0700644 swap_readpage(page, false, &splug);
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700645 if (offset != entry_offset) {
Huang Yingc4fa6302017-09-06 16:24:33 -0700646 SetPageReadahead(page);
647 count_vm_event(SWAP_RA);
648 }
Huang Yingcbc65df2017-09-06 16:24:29 -0700649 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300650 put_page(page);
Hugh Dickins46017e92008-02-04 22:28:41 -0800651 }
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700652 blk_finish_plug(&plug);
NeilBrown5169b842022-05-09 18:20:49 -0700653 swap_read_unplug(splug);
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700654
Hugh Dickins46017e92008-02-04 22:28:41 -0800655 lru_add_drain(); /* Push any new pages onto the LRU now */
Shaohua Li579f8292014-02-06 12:04:21 -0800656skip:
NeilBrown5169b842022-05-09 18:20:49 -0700657 /* The page was likely read above, so no need for plugging here */
658 return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll, NULL);
Hugh Dickins46017e92008-02-04 22:28:41 -0800659}
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800660
661int init_swap_address_space(unsigned int type, unsigned long nr_pages)
662{
663 struct address_space *spaces, *space;
664 unsigned int i, nr;
665
666 nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES);
Kees Cook778e1cd2018-06-12 14:04:48 -0700667 spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL);
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800668 if (!spaces)
669 return -ENOMEM;
670 for (i = 0; i < nr; i++) {
671 space = spaces + i;
Matthew Wilcoxa2833482017-12-05 19:04:20 -0500672 xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ);
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800673 atomic_set(&space->i_mmap_writable, 0);
674 space->a_ops = &swap_aops;
675 /* swap cache doesn't use writeback related tags */
676 mapping_set_no_writeback_tags(space);
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800677 }
678 nr_swapper_spaces[type] = nr;
Huang Ying054f1d12019-07-11 20:55:37 -0700679 swapper_spaces[type] = spaces;
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800680
681 return 0;
682}
683
684void exit_swap_address_space(unsigned int type)
685{
Huang Yingeea4a502021-06-28 19:37:16 -0700686 int i;
687 struct address_space *spaces = swapper_spaces[type];
688
689 for (i = 0; i < nr_swapper_spaces[type]; i++)
690 VM_WARN_ON_ONCE(!mapping_empty(&spaces[i]));
691 kvfree(spaces);
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800692 nr_swapper_spaces[type] = 0;
Huang Ying054f1d12019-07-11 20:55:37 -0700693 swapper_spaces[type] = NULL;
Huang, Ying4b3ef9d2017-02-22 15:45:26 -0800694}
Huang Yingec560172017-09-06 16:24:36 -0700695
696static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma,
697 unsigned long faddr,
698 unsigned long lpfn,
699 unsigned long rpfn,
700 unsigned long *start,
701 unsigned long *end)
702{
703 *start = max3(lpfn, PFN_DOWN(vma->vm_start),
704 PFN_DOWN(faddr & PMD_MASK));
705 *end = min3(rpfn, PFN_DOWN(vma->vm_end),
706 PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE));
707}
708
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700709static void swap_ra_info(struct vm_fault *vmf,
710 struct vma_swap_readahead *ra_info)
Huang Yingec560172017-09-06 16:24:36 -0700711{
712 struct vm_area_struct *vma = vmf->vma;
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700713 unsigned long ra_val;
Huang Yingec560172017-09-06 16:24:36 -0700714 unsigned long faddr, pfn, fpfn;
715 unsigned long start, end;
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700716 pte_t *pte, *orig_pte;
Huang Yingec560172017-09-06 16:24:36 -0700717 unsigned int max_win, hits, prev_win, win, left;
718#ifndef CONFIG_64BIT
719 pte_t *tpte;
720#endif
721
Huang Ying61b63972017-10-13 15:58:29 -0700722 max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster),
723 SWAP_RA_ORDER_CEILING);
724 if (max_win == 1) {
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700725 ra_info->win = 1;
726 return;
Huang Ying61b63972017-10-13 15:58:29 -0700727 }
728
Huang Yingec560172017-09-06 16:24:36 -0700729 faddr = vmf->address;
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700730 orig_pte = pte = pte_offset_map(vmf->pmd, faddr);
Huang Yingec560172017-09-06 16:24:36 -0700731
Huang Yingec560172017-09-06 16:24:36 -0700732 fpfn = PFN_DOWN(faddr);
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700733 ra_val = GET_SWAP_RA_VAL(vma);
734 pfn = PFN_DOWN(SWAP_RA_ADDR(ra_val));
735 prev_win = SWAP_RA_WIN(ra_val);
736 hits = SWAP_RA_HITS(ra_val);
737 ra_info->win = win = __swapin_nr_pages(pfn, fpfn, hits,
Huang Yingec560172017-09-06 16:24:36 -0700738 max_win, prev_win);
739 atomic_long_set(&vma->swap_readahead_info,
740 SWAP_RA_VAL(faddr, win, 0));
741
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700742 if (win == 1) {
743 pte_unmap(orig_pte);
744 return;
745 }
Huang Yingec560172017-09-06 16:24:36 -0700746
747 /* Copy the PTEs because the page table may be unmapped */
748 if (fpfn == pfn + 1)
749 swap_ra_clamp_pfn(vma, faddr, fpfn, fpfn + win, &start, &end);
750 else if (pfn == fpfn + 1)
751 swap_ra_clamp_pfn(vma, faddr, fpfn - win + 1, fpfn + 1,
752 &start, &end);
753 else {
754 left = (win - 1) / 2;
755 swap_ra_clamp_pfn(vma, faddr, fpfn - left, fpfn + win - left,
756 &start, &end);
757 }
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700758 ra_info->nr_pte = end - start;
759 ra_info->offset = fpfn - start;
760 pte -= ra_info->offset;
Huang Yingec560172017-09-06 16:24:36 -0700761#ifdef CONFIG_64BIT
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700762 ra_info->ptes = pte;
Huang Yingec560172017-09-06 16:24:36 -0700763#else
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700764 tpte = ra_info->ptes;
Huang Yingec560172017-09-06 16:24:36 -0700765 for (pfn = start; pfn != end; pfn++)
766 *tpte++ = *pte++;
767#endif
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700768 pte_unmap(orig_pte);
Huang Yingec560172017-09-06 16:24:36 -0700769}
770
Yang Shie9f59872019-03-05 15:44:15 -0800771/**
772 * swap_vma_readahead - swap in pages in hope we need them soon
Krzysztof Kozlowski27ec4872020-08-06 23:20:14 -0700773 * @fentry: swap entry of this memory
Yang Shie9f59872019-03-05 15:44:15 -0800774 * @gfp_mask: memory allocation flags
775 * @vmf: fault information
776 *
777 * Returns the struct page for entry and addr, after queueing swapin.
778 *
Shijie Luocb152a12021-05-06 18:05:51 -0700779 * Primitive swap readahead code. We simply read in a few pages whose
Yang Shie9f59872019-03-05 15:44:15 -0800780 * virtual addresses are around the fault address in the same vma.
781 *
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700782 * Caller must hold read mmap_lock if vmf->vma is not NULL.
Yang Shie9f59872019-03-05 15:44:15 -0800783 *
784 */
Colin Ian Kingf5c754d2018-04-05 16:25:05 -0700785static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask,
786 struct vm_fault *vmf)
Huang Yingec560172017-09-06 16:24:36 -0700787{
788 struct blk_plug plug;
NeilBrown5169b842022-05-09 18:20:49 -0700789 struct swap_iocb *splug = NULL;
Huang Yingec560172017-09-06 16:24:36 -0700790 struct vm_area_struct *vma = vmf->vma;
791 struct page *page;
792 pte_t *pte, pentry;
793 swp_entry_t entry;
794 unsigned int i;
795 bool page_allocated;
Miaohe Line97af692020-12-14 19:06:01 -0800796 struct vma_swap_readahead ra_info = {
797 .win = 1,
798 };
Huang Yingec560172017-09-06 16:24:36 -0700799
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700800 swap_ra_info(vmf, &ra_info);
801 if (ra_info.win == 1)
Huang Yingec560172017-09-06 16:24:36 -0700802 goto skip;
803
804 blk_start_plug(&plug);
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700805 for (i = 0, pte = ra_info.ptes; i < ra_info.nr_pte;
Huang Yingec560172017-09-06 16:24:36 -0700806 i++, pte++) {
807 pentry = *pte;
Miaohe Lin92bafb22022-05-19 14:08:50 -0700808 if (!is_swap_pte(pentry))
Huang Yingec560172017-09-06 16:24:36 -0700809 continue;
810 entry = pte_to_swp_entry(pentry);
811 if (unlikely(non_swap_entry(entry)))
812 continue;
813 page = __read_swap_cache_async(entry, gfp_mask, vma,
814 vmf->address, &page_allocated);
815 if (!page)
816 continue;
817 if (page_allocated) {
NeilBrown5169b842022-05-09 18:20:49 -0700818 swap_readpage(page, false, &splug);
Minchan Kimeaf649eb2018-04-05 16:23:39 -0700819 if (i != ra_info.offset) {
Huang Yingec560172017-09-06 16:24:36 -0700820 SetPageReadahead(page);
821 count_vm_event(SWAP_RA);
822 }
823 }
824 put_page(page);
825 }
826 blk_finish_plug(&plug);
NeilBrown5169b842022-05-09 18:20:49 -0700827 swap_read_unplug(splug);
Huang Yingec560172017-09-06 16:24:36 -0700828 lru_add_drain();
829skip:
NeilBrown5169b842022-05-09 18:20:49 -0700830 /* The page was likely read above, so no need for plugging here */
Huang Yingec560172017-09-06 16:24:36 -0700831 return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address,
NeilBrown5169b842022-05-09 18:20:49 -0700832 ra_info.win == 1, NULL);
Huang Yingec560172017-09-06 16:24:36 -0700833}
Huang Yingd9bfcfd2017-09-06 16:24:40 -0700834
Minchan Kime9e9b7e2018-04-05 16:23:42 -0700835/**
836 * swapin_readahead - swap in pages in hope we need them soon
837 * @entry: swap entry of this memory
838 * @gfp_mask: memory allocation flags
839 * @vmf: fault information
840 *
841 * Returns the struct page for entry and addr, after queueing swapin.
842 *
843 * It's a main entry function for swap readahead. By the configuration,
844 * it will read ahead blocks by cluster-based(ie, physical disk based)
845 * or vma-based(ie, virtual address based on faulty address) readahead.
846 */
847struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
848 struct vm_fault *vmf)
849{
850 return swap_use_vma_readahead() ?
851 swap_vma_readahead(entry, gfp_mask, vmf) :
852 swap_cluster_readahead(entry, gfp_mask, vmf);
853}
854
Huang Yingd9bfcfd2017-09-06 16:24:40 -0700855#ifdef CONFIG_SYSFS
856static ssize_t vma_ra_enabled_show(struct kobject *kobj,
857 struct kobj_attribute *attr, char *buf)
858{
Joe Perchesae7a9272020-12-14 19:14:42 -0800859 return sysfs_emit(buf, "%s\n",
860 enable_vma_readahead ? "true" : "false");
Huang Yingd9bfcfd2017-09-06 16:24:40 -0700861}
862static ssize_t vma_ra_enabled_store(struct kobject *kobj,
863 struct kobj_attribute *attr,
864 const char *buf, size_t count)
865{
Jagdish Gediya717aeab2022-05-12 20:22:59 -0700866 ssize_t ret;
867
868 ret = kstrtobool(buf, &enable_vma_readahead);
869 if (ret)
870 return ret;
Huang Yingd9bfcfd2017-09-06 16:24:40 -0700871
872 return count;
873}
Miaohe Lin6106b932022-05-19 14:08:50 -0700874static struct kobj_attribute vma_ra_enabled_attr = __ATTR_RW(vma_ra_enabled);
Huang Yingd9bfcfd2017-09-06 16:24:40 -0700875
Huang Yingd9bfcfd2017-09-06 16:24:40 -0700876static struct attribute *swap_attrs[] = {
877 &vma_ra_enabled_attr.attr,
Huang Yingd9bfcfd2017-09-06 16:24:40 -0700878 NULL,
879};
880
Rikard Falkeborne48333b2021-02-24 12:03:05 -0800881static const struct attribute_group swap_attr_group = {
Huang Yingd9bfcfd2017-09-06 16:24:40 -0700882 .attrs = swap_attrs,
883};
884
885static int __init swap_init_sysfs(void)
886{
887 int err;
888 struct kobject *swap_kobj;
889
890 swap_kobj = kobject_create_and_add("swap", mm_kobj);
891 if (!swap_kobj) {
892 pr_err("failed to create swap kobject\n");
893 return -ENOMEM;
894 }
895 err = sysfs_create_group(swap_kobj, &swap_attr_group);
896 if (err) {
897 pr_err("failed to register swap group\n");
898 goto delete_obj;
899 }
900 return 0;
901
902delete_obj:
903 kobject_put(swap_kobj);
904 return err;
905}
906subsys_initcall(swap_init_sysfs);
907#endif