Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | #include <linux/mm.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 11 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | #include <linux/kernel_stat.h> |
| 13 | #include <linux/swap.h> |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 14 | #include <linux/swapops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 15 | #include <linux/init.h> |
| 16 | #include <linux/pagemap.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #include <linux/backing-dev.h> |
Christian Ehrhardt | 3fb5c29 | 2012-07-31 16:41:44 -0700 | [diff] [blame] | 18 | #include <linux/blkdev.h> |
Hugh Dickins | c484d41 | 2006-01-06 00:10:55 -0800 | [diff] [blame] | 19 | #include <linux/pagevec.h> |
Christoph Lameter | b20a350 | 2006-03-22 00:09:12 -0800 | [diff] [blame] | 20 | #include <linux/migrate.h> |
Huang, Ying | 4b3ef9d | 2017-02-22 15:45:26 -0800 | [diff] [blame] | 21 | #include <linux/vmalloc.h> |
Tim Chen | 67afa38 | 2017-02-22 15:45:39 -0800 | [diff] [blame] | 22 | #include <linux/swap_slots.h> |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 23 | #include <linux/huge_mm.h> |
Matthew Wilcox (Oracle) | 61ef186 | 2020-10-13 16:51:17 -0700 | [diff] [blame] | 24 | #include <linux/shmem_fs.h> |
Hugh Dickins | 243bce0 | 2020-06-25 20:29:59 -0700 | [diff] [blame] | 25 | #include "internal.h" |
NeilBrown | 014bb1d | 2022-05-09 18:20:47 -0700 | [diff] [blame] | 26 | #include "swap.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | |
| 28 | /* |
| 29 | * swapper_space is a fiction, retained to simplify the path through |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 30 | * vmscan's shrink_page_list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | */ |
Christoph Hellwig | f5e54d6 | 2006-06-28 04:26:44 -0700 | [diff] [blame] | 32 | static const struct address_space_operations swap_aops = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 33 | .writepage = swap_writepage, |
NeilBrown | 4c4a763 | 2022-05-09 18:20:47 -0700 | [diff] [blame] | 34 | .dirty_folio = noop_dirty_folio, |
Andrew Morton | 1c93923 | 2014-10-09 15:27:59 -0700 | [diff] [blame] | 35 | #ifdef CONFIG_MIGRATION |
Matthew Wilcox (Oracle) | 5418465 | 2022-06-06 10:27:41 -0400 | [diff] [blame] | 36 | .migrate_folio = migrate_folio, |
Andrew Morton | 1c93923 | 2014-10-09 15:27:59 -0700 | [diff] [blame] | 37 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 38 | }; |
| 39 | |
Changbin Du | 783cb68 | 2017-11-15 17:36:06 -0800 | [diff] [blame] | 40 | struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly; |
| 41 | static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly; |
Colin Ian King | f5c754d | 2018-04-05 16:25:05 -0700 | [diff] [blame] | 42 | static bool enable_vma_readahead __read_mostly = true; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 43 | |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 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) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 62 | static atomic_t swapin_readahead_hits = ATOMIC_INIT(4); |
| 63 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | void show_swap_cache_info(void) |
| 65 | { |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 66 | printk("%lu pages in swap cache\n", total_swapcache_pages()); |
Shaohua Li | ec8acf2 | 2013-02-22 16:34:38 -0800 | [diff] [blame] | 67 | printk("Free swap = %ldkB\n", |
| 68 | get_nr_swap_pages() << (PAGE_SHIFT - 10)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10)); |
| 70 | } |
| 71 | |
Joonsoo Kim | aae466b | 2020-08-11 18:30:50 -0700 | [diff] [blame] | 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); |
| 76 | struct page *page; |
| 77 | |
Matthew Wilcox (Oracle) | 8c647dd | 2021-02-25 17:15:33 -0800 | [diff] [blame] | 78 | page = xa_load(&address_space->i_pages, idx); |
Joonsoo Kim | aae466b | 2020-08-11 18:30:50 -0700 | [diff] [blame] | 79 | if (xa_is_value(page)) |
| 80 | return page; |
Joonsoo Kim | aae466b | 2020-08-11 18:30:50 -0700 | [diff] [blame] | 81 | return NULL; |
| 82 | } |
| 83 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | /* |
Matthew Wilcox (Oracle) | 2bb876b | 2022-06-01 15:13:59 -0400 | [diff] [blame] | 85 | * add_to_swap_cache resembles filemap_add_folio on swapper_space, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | * but sets SwapCache flag and private instead of mapping and index. |
| 87 | */ |
Joonsoo Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 88 | int add_to_swap_cache(struct page *page, swp_entry_t entry, |
| 89 | gfp_t gfp, void **shadowp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | { |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 91 | struct address_space *address_space = swap_address_space(entry); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 92 | pgoff_t idx = swp_offset(entry); |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 93 | XA_STATE_ORDER(xas, &address_space->i_pages, idx, compound_order(page)); |
Matthew Wilcox (Oracle) | 6c35784 | 2020-08-14 17:30:37 -0700 | [diff] [blame] | 94 | unsigned long i, nr = thp_nr_pages(page); |
Joonsoo Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 95 | void *old; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 96 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 97 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 98 | VM_BUG_ON_PAGE(PageSwapCache(page), page); |
| 99 | VM_BUG_ON_PAGE(!PageSwapBacked(page), page); |
Hugh Dickins | 51726b1 | 2009-01-06 14:39:25 -0800 | [diff] [blame] | 100 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 101 | page_ref_add(page, nr); |
Daisuke Nishimura | 31a5639 | 2009-09-21 17:02:50 -0700 | [diff] [blame] | 102 | SetPageSwapCache(page); |
Daisuke Nishimura | 31a5639 | 2009-09-21 17:02:50 -0700 | [diff] [blame] | 103 | |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 104 | 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 Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 111 | old = xas_load(&xas); |
| 112 | if (xa_is_value(old)) { |
Joonsoo Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 113 | if (shadowp) |
| 114 | *shadowp = old; |
| 115 | } |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 116 | set_page_private(page + i, entry.val + i); |
Matthew Wilcox (Oracle) | 4101196 | 2019-09-23 15:34:52 -0700 | [diff] [blame] | 117 | xas_store(&xas, page); |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 118 | xas_next(&xas); |
| 119 | } |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 120 | address_space->nrpages += nr; |
| 121 | __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); |
Shakeel Butt | b603894 | 2021-02-24 12:03:55 -0800 | [diff] [blame] | 122 | __mod_lruvec_page_state(page, NR_SWAPCACHE, nr); |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 123 | unlock: |
| 124 | xas_unlock_irq(&xas); |
| 125 | } while (xas_nomem(&xas, gfp)); |
Daisuke Nishimura | 31a5639 | 2009-09-21 17:02:50 -0700 | [diff] [blame] | 126 | |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 127 | if (!xas_error(&xas)) |
| 128 | return 0; |
Daisuke Nishimura | 31a5639 | 2009-09-21 17:02:50 -0700 | [diff] [blame] | 129 | |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 130 | ClearPageSwapCache(page); |
| 131 | page_ref_sub(page, nr); |
| 132 | return xas_error(&xas); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | } |
| 134 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | /* |
Matthew Wilcox (Oracle) | ceff9d3 | 2022-06-17 18:50:20 +0100 | [diff] [blame] | 136 | * This must be called only on folios that have |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | * been verified to be in the swap cache. |
| 138 | */ |
Matthew Wilcox (Oracle) | ceff9d3 | 2022-06-17 18:50:20 +0100 | [diff] [blame] | 139 | void __delete_from_swap_cache(struct folio *folio, |
Joonsoo Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 140 | swp_entry_t entry, void *shadow) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | { |
Matthew Wilcox | 4e17ec2 | 2017-11-29 08:32:39 -0500 | [diff] [blame] | 142 | struct address_space *address_space = swap_address_space(entry); |
Matthew Wilcox (Oracle) | ceff9d3 | 2022-06-17 18:50:20 +0100 | [diff] [blame] | 143 | int i; |
| 144 | long nr = folio_nr_pages(folio); |
Matthew Wilcox | 4e17ec2 | 2017-11-29 08:32:39 -0500 | [diff] [blame] | 145 | pgoff_t idx = swp_offset(entry); |
| 146 | XA_STATE(xas, &address_space->i_pages, idx); |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 147 | |
Matthew Wilcox (Oracle) | ceff9d3 | 2022-06-17 18:50:20 +0100 | [diff] [blame] | 148 | 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 152 | for (i = 0; i < nr; i++) { |
Joonsoo Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 153 | void *entry = xas_store(&xas, shadow); |
Matthew Wilcox (Oracle) | ceff9d3 | 2022-06-17 18:50:20 +0100 | [diff] [blame] | 154 | VM_BUG_ON_FOLIO(entry != folio, folio); |
| 155 | set_page_private(folio_page(folio, i), 0); |
Matthew Wilcox | 4e17ec2 | 2017-11-29 08:32:39 -0500 | [diff] [blame] | 156 | xas_next(&xas); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 157 | } |
Matthew Wilcox (Oracle) | ceff9d3 | 2022-06-17 18:50:20 +0100 | [diff] [blame] | 158 | folio_clear_swapcache(folio); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 159 | address_space->nrpages -= nr; |
Matthew Wilcox (Oracle) | ceff9d3 | 2022-06-17 18:50:20 +0100 | [diff] [blame] | 160 | __node_stat_mod_folio(folio, NR_FILE_PAGES, -nr); |
| 161 | __lruvec_stat_mod_folio(folio, NR_SWAPCACHE, -nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | } |
| 163 | |
| 164 | /** |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 165 | * add_to_swap - allocate swap space for a folio |
| 166 | * @folio: folio we want to move to swap |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | * |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 168 | * 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | */ |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 174 | bool add_to_swap(struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | { |
| 176 | swp_entry_t entry; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | int err; |
| 178 | |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 179 | VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); |
| 180 | VM_BUG_ON_FOLIO(!folio_test_uptodate(folio), folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | |
Matthew Wilcox (Oracle) | e2e3fdc | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 182 | entry = folio_alloc_swap(folio); |
Daisuke Nishimura | 2ca4532 | 2009-09-21 17:02:52 -0700 | [diff] [blame] | 183 | if (!entry.val) |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 184 | return false; |
Minchan Kim | 0f07465 | 2017-07-06 15:37:24 -0700 | [diff] [blame] | 185 | |
Daisuke Nishimura | 2ca4532 | 2009-09-21 17:02:52 -0700 | [diff] [blame] | 186 | /* |
Matthew Wilcox | 8d93b41 | 2017-11-27 15:46:54 -0500 | [diff] [blame] | 187 | * XArray node allocations from PF_MEMALLOC contexts could |
Daisuke Nishimura | 2ca4532 | 2009-09-21 17:02:52 -0700 | [diff] [blame] | 188 | * 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 Kim | 854e9ed | 2016-01-15 16:54:53 -0800 | [diff] [blame] | 195 | * Add it to the swap cache. |
Daisuke Nishimura | 2ca4532 | 2009-09-21 17:02:52 -0700 | [diff] [blame] | 196 | */ |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 197 | err = add_to_swap_cache(&folio->page, entry, |
Joonsoo Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 198 | __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN, NULL); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 199 | if (err) |
Daisuke Nishimura | 2ca4532 | 2009-09-21 17:02:52 -0700 | [diff] [blame] | 200 | /* |
| 201 | * add_to_swap_cache() doesn't return -EEXIST, so we can safely |
| 202 | * clear SWAP_HAS_CACHE flag. |
| 203 | */ |
Minchan Kim | 0f07465 | 2017-07-06 15:37:24 -0700 | [diff] [blame] | 204 | goto fail; |
Shaohua Li | 9625456 | 2017-10-03 16:15:32 -0700 | [diff] [blame] | 205 | /* |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 206 | * 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 Li | 9625456 | 2017-10-03 16:15:32 -0700 | [diff] [blame] | 215 | */ |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 216 | folio_mark_dirty(folio); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 217 | |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 218 | return true; |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 219 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 220 | fail: |
Matthew Wilcox (Oracle) | 09c02e5 | 2022-05-12 20:23:02 -0700 | [diff] [blame] | 221 | put_swap_page(&folio->page, entry); |
| 222 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | } |
| 224 | |
| 225 | /* |
Matthew Wilcox (Oracle) | 75fa68a | 2022-06-17 18:50:19 +0100 | [diff] [blame] | 226 | * This must be called only on folios that have |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | * been verified to be in the swap cache and locked. |
Matthew Wilcox (Oracle) | 75fa68a | 2022-06-17 18:50:19 +0100 | [diff] [blame] | 228 | * It will never put the folio into the free list, |
| 229 | * the caller has a reference on the folio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 230 | */ |
Matthew Wilcox (Oracle) | 75fa68a | 2022-06-17 18:50:19 +0100 | [diff] [blame] | 231 | void delete_from_swap_cache(struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | { |
Matthew Wilcox (Oracle) | 75fa68a | 2022-06-17 18:50:19 +0100 | [diff] [blame] | 233 | swp_entry_t entry = folio_swap_entry(folio); |
Matthew Wilcox | 4e17ec2 | 2017-11-29 08:32:39 -0500 | [diff] [blame] | 234 | struct address_space *address_space = swap_address_space(entry); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 236 | xa_lock_irq(&address_space->i_pages); |
Matthew Wilcox (Oracle) | ceff9d3 | 2022-06-17 18:50:20 +0100 | [diff] [blame] | 237 | __delete_from_swap_cache(folio, entry, NULL); |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 238 | xa_unlock_irq(&address_space->i_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | |
Matthew Wilcox (Oracle) | 75fa68a | 2022-06-17 18:50:19 +0100 | [diff] [blame] | 240 | put_swap_page(&folio->page, entry); |
| 241 | folio_ref_sub(folio, folio_nr_pages(folio)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 242 | } |
| 243 | |
Joonsoo Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 244 | void 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 Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 251 | 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 Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 260 | } |
Joonsoo Kim | 3852f67 | 2020-08-11 18:30:47 -0700 | [diff] [blame] | 261 | 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 272 | /* |
| 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 Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 276 | * here because we are going to recheck again inside |
| 277 | * try_to_free_swap() _with_ the lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 278 | * - Marcelo |
| 279 | */ |
Huang Ying | f4c4a3f | 2021-06-28 19:37:12 -0700 | [diff] [blame] | 280 | void free_swap_cache(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 281 | { |
Hugh Dickins | a2c43ee | 2009-01-06 14:39:36 -0800 | [diff] [blame] | 282 | if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) { |
| 283 | try_to_free_swap(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 284 | unlock_page(page); |
| 285 | } |
| 286 | } |
| 287 | |
| 288 | /* |
| 289 | * Perform a free_page(), also freeing any swap cache associated with |
Hugh Dickins | b8072f0 | 2005-10-29 18:16:41 -0700 | [diff] [blame] | 290 | * this page if it is the last user of the page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | */ |
| 292 | void free_page_and_swap_cache(struct page *page) |
| 293 | { |
| 294 | free_swap_cache(page); |
Aaron Lu | 6fcb52a | 2016-10-07 17:00:08 -0700 | [diff] [blame] | 295 | if (!is_huge_zero_page(page)) |
Gerald Schaefer | 770a537 | 2016-06-08 15:33:50 -0700 | [diff] [blame] | 296 | put_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | } |
| 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 | */ |
| 303 | void free_pages_and_swap_cache(struct page **pages, int nr) |
| 304 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | struct page **pagep = pages; |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 306 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | |
| 308 | lru_add_drain(); |
Michal Hocko | aabfb57 | 2014-10-09 15:28:52 -0700 | [diff] [blame] | 309 | for (i = 0; i < nr; i++) |
| 310 | free_swap_cache(pagep[i]); |
Mel Gorman | c6f92f9 | 2017-11-15 17:37:55 -0800 | [diff] [blame] | 311 | release_pages(pagep, nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | } |
| 313 | |
Minchan Kim | e9e9b7e | 2018-04-05 16:23:42 -0700 | [diff] [blame] | 314 | static inline bool swap_use_vma_readahead(void) |
| 315 | { |
| 316 | return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap); |
| 317 | } |
| 318 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | /* |
| 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 Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 325 | struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma, |
| 326 | unsigned long addr) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 327 | { |
| 328 | struct page *page; |
Huang Ying | eb08557 | 2019-07-11 20:55:33 -0700 | [diff] [blame] | 329 | struct swap_info_struct *si; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | |
Huang Ying | eb08557 | 2019-07-11 20:55:33 -0700 | [diff] [blame] | 331 | si = get_swap_device(entry); |
| 332 | if (!si) |
| 333 | return NULL; |
Huang Ying | f6ab1f7 | 2016-10-07 17:00:21 -0700 | [diff] [blame] | 334 | page = find_get_page(swap_address_space(entry), swp_offset(entry)); |
Huang Ying | eb08557 | 2019-07-11 20:55:33 -0700 | [diff] [blame] | 335 | put_swap_device(si); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 337 | if (page) { |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 338 | bool vma_ra = swap_use_vma_readahead(); |
| 339 | bool readahead; |
| 340 | |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 341 | /* |
| 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 Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 345 | if (unlikely(PageTransCompound(page))) |
| 346 | return page; |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 347 | |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 348 | readahead = TestClearPageReadahead(page); |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 349 | 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 Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 356 | 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 Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 361 | |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 362 | if (readahead) { |
Huang Ying | cbc65df | 2017-09-06 16:24:29 -0700 | [diff] [blame] | 363 | count_vm_event(SWAP_RA_HIT); |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 364 | if (!vma || !vma_ra) |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 365 | atomic_inc(&swapin_readahead_hits); |
Huang Ying | cbc65df | 2017-09-06 16:24:29 -0700 | [diff] [blame] | 366 | } |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 367 | } |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 368 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | return page; |
| 370 | } |
| 371 | |
Matthew Wilcox (Oracle) | 61ef186 | 2020-10-13 16:51:17 -0700 | [diff] [blame] | 372 | /** |
| 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 | */ |
| 382 | struct 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) | 44835d2 | 2021-02-25 17:15:36 -0800 | [diff] [blame] | 386 | struct page *page = pagecache_get_page(mapping, index, |
| 387 | FGP_ENTRY | FGP_HEAD, 0); |
Matthew Wilcox (Oracle) | 61ef186 | 2020-10-13 16:51:17 -0700 | [diff] [blame] | 388 | |
Matthew Wilcox (Oracle) | a6de4b4 | 2020-10-13 16:51:34 -0700 | [diff] [blame] | 389 | if (!page) |
Matthew Wilcox (Oracle) | 61ef186 | 2020-10-13 16:51:17 -0700 | [diff] [blame] | 390 | return page; |
Matthew Wilcox (Oracle) | a6de4b4 | 2020-10-13 16:51:34 -0700 | [diff] [blame] | 391 | if (!xa_is_value(page)) |
| 392 | return find_subpage(page, index); |
Matthew Wilcox (Oracle) | 61ef186 | 2020-10-13 16:51:17 -0700 | [diff] [blame] | 393 | if (!shmem_mapping(mapping)) |
| 394 | return NULL; |
| 395 | |
| 396 | swp = radix_to_swp_entry(page); |
Miaohe Lin | ba6851b | 2022-05-19 20:50:30 +0800 | [diff] [blame] | 397 | /* There might be swapin error entries in shmem mapping. */ |
| 398 | if (non_swap_entry(swp)) |
| 399 | return NULL; |
Matthew Wilcox (Oracle) | 61ef186 | 2020-10-13 16:51:17 -0700 | [diff] [blame] | 400 | /* 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 Safonov | 5b999aa | 2015-09-08 15:05:00 -0700 | [diff] [blame] | 409 | struct 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | { |
Huang Ying | eb08557 | 2019-07-11 20:55:33 -0700 | [diff] [blame] | 413 | struct swap_info_struct *si; |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 414 | struct page *page; |
Joonsoo Kim | aae466b | 2020-08-11 18:30:50 -0700 | [diff] [blame] | 415 | void *shadow = NULL; |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 416 | |
Dmitry Safonov | 5b999aa | 2015-09-08 15:05:00 -0700 | [diff] [blame] | 417 | *new_page_allocated = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 419 | for (;;) { |
| 420 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | /* |
| 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 Ying | eb08557 | 2019-07-11 20:55:33 -0700 | [diff] [blame] | 426 | si = get_swap_device(entry); |
| 427 | if (!si) |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 428 | return NULL; |
| 429 | page = find_get_page(swap_address_space(entry), |
| 430 | swp_offset(entry)); |
Huang Ying | eb08557 | 2019-07-11 20:55:33 -0700 | [diff] [blame] | 431 | put_swap_device(si); |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 432 | if (page) |
| 433 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | |
Huang Ying | ba81f83 | 2017-02-22 15:45:46 -0800 | [diff] [blame] | 435 | /* |
| 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 Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 444 | return NULL; |
Tim Chen | e8c26ab | 2017-02-22 15:45:29 -0800 | [diff] [blame] | 445 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 446 | /* |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 447 | * 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 450 | */ |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 451 | page = alloc_page_vma(gfp_mask, vma, addr); |
| 452 | if (!page) |
| 453 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 454 | |
| 455 | /* |
Hugh Dickins | f000944 | 2008-02-04 22:28:49 -0800 | [diff] [blame] | 456 | * Swap entry may have been freed since our caller observed it. |
| 457 | */ |
KAMEZAWA Hiroyuki | 355cfa7 | 2009-06-16 15:32:53 -0700 | [diff] [blame] | 458 | err = swapcache_prepare(entry); |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 459 | if (!err) |
Hugh Dickins | f000944 | 2008-02-04 22:28:49 -0800 | [diff] [blame] | 460 | break; |
| 461 | |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 462 | put_page(page); |
| 463 | if (err != -EEXIST) |
| 464 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 465 | |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 466 | /* |
| 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 Ziliang | 029c462 | 2022-03-16 16:15:03 -0700 | [diff] [blame] | 473 | schedule_timeout_uninterruptible(1); |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 474 | } |
| 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 Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame] | 483 | if (mem_cgroup_swapin_charge_page(page, NULL, gfp_mask, entry)) |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 484 | goto fail_unlock; |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 485 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame] | 486 | /* May fail (-ENOMEM) if XArray node allocation failed. */ |
| 487 | if (add_to_swap_cache(page, entry, gfp_mask & GFP_RECLAIM_MASK, &shadow)) |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 488 | goto fail_unlock; |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame] | 489 | |
| 490 | mem_cgroup_swapin_uncharge_swap(entry); |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 491 | |
Joonsoo Kim | aae466b | 2020-08-11 18:30:50 -0700 | [diff] [blame] | 492 | if (shadow) |
Matthew Wilcox (Oracle) | 0995d7e | 2021-04-29 10:27:16 -0400 | [diff] [blame] | 493 | workingset_refault(page_folio(page), shadow); |
Johannes Weiner | 314b57f | 2020-06-03 16:03:03 -0700 | [diff] [blame] | 494 | |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 495 | /* Caller will initiate read into locked page */ |
Johannes Weiner | 6058eae | 2020-06-03 16:02:40 -0700 | [diff] [blame] | 496 | lru_cache_add(page); |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 497 | *new_page_allocated = true; |
| 498 | return page; |
| 499 | |
| 500 | fail_unlock: |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame] | 501 | put_swap_page(page, entry); |
Johannes Weiner | 4c6355b | 2020-06-03 16:02:17 -0700 | [diff] [blame] | 502 | unlock_page(page); |
| 503 | put_page(page); |
| 504 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 505 | } |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 506 | |
Dmitry Safonov | 5b999aa | 2015-09-08 15:05:00 -0700 | [diff] [blame] | 507 | /* |
| 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 | */ |
| 513 | struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 514 | struct vm_area_struct *vma, |
| 515 | unsigned long addr, bool do_poll, |
| 516 | struct swap_iocb **plug) |
Dmitry Safonov | 5b999aa | 2015-09-08 15:05:00 -0700 | [diff] [blame] | 517 | { |
| 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) |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 523 | swap_readpage(retpage, do_poll, plug); |
Dmitry Safonov | 5b999aa | 2015-09-08 15:05:00 -0700 | [diff] [blame] | 524 | |
| 525 | return retpage; |
| 526 | } |
| 527 | |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 528 | static 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 Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 533 | { |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 534 | unsigned int pages, last_ra; |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 535 | |
| 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 Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 541 | pages = hits + 2; |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 542 | 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 Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 550 | } 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 Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 561 | last_ra = prev_win / 2; |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 562 | if (pages < last_ra) |
| 563 | pages = last_ra; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 564 | |
| 565 | return pages; |
| 566 | } |
| 567 | |
| 568 | static 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 Cai | d6c1f09 | 2020-06-01 21:48:40 -0700 | [diff] [blame] | 579 | pages = __swapin_nr_pages(READ_ONCE(prev_offset), offset, hits, |
| 580 | max_pages, |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 581 | atomic_read(&last_readahead_pages)); |
| 582 | if (!hits) |
Qian Cai | d6c1f09 | 2020-06-01 21:48:40 -0700 | [diff] [blame] | 583 | WRITE_ONCE(prev_offset, offset); |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 584 | atomic_set(&last_readahead_pages, pages); |
| 585 | |
| 586 | return pages; |
| 587 | } |
| 588 | |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 589 | /** |
Minchan Kim | e9e9b7e | 2018-04-05 16:23:42 -0700 | [diff] [blame] | 590 | * swap_cluster_readahead - swap in pages in hope we need them soon |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 591 | * @entry: swap entry of this memory |
Randy Dunlap | 7682486 | 2008-03-19 17:00:40 -0700 | [diff] [blame] | 592 | * @gfp_mask: memory allocation flags |
Minchan Kim | e9e9b7e | 2018-04-05 16:23:42 -0700 | [diff] [blame] | 593 | * @vmf: fault information |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 594 | * |
| 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 Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 605 | * Caller must hold read mmap_lock if vmf->vma is not NULL. |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 606 | */ |
Minchan Kim | e9e9b7e | 2018-04-05 16:23:42 -0700 | [diff] [blame] | 607 | struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask, |
| 608 | struct vm_fault *vmf) |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 609 | { |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 610 | struct page *page; |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 611 | unsigned long entry_offset = swp_offset(entry); |
| 612 | unsigned long offset = entry_offset; |
Rik van Riel | 67f96aa | 2012-03-21 16:33:50 -0700 | [diff] [blame] | 613 | unsigned long start_offset, end_offset; |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 614 | unsigned long mask; |
Huang Ying | e9a6eff | 2017-11-15 17:33:15 -0800 | [diff] [blame] | 615 | struct swap_info_struct *si = swp_swap_info(entry); |
Christian Ehrhardt | 3fb5c29 | 2012-07-31 16:41:44 -0700 | [diff] [blame] | 616 | struct blk_plug plug; |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 617 | struct swap_iocb *splug = NULL; |
Huang Ying | c4fa630 | 2017-09-06 16:24:33 -0700 | [diff] [blame] | 618 | bool do_poll = true, page_allocated; |
Minchan Kim | e9e9b7e | 2018-04-05 16:23:42 -0700 | [diff] [blame] | 619 | struct vm_area_struct *vma = vmf->vma; |
| 620 | unsigned long addr = vmf->address; |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 621 | |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 622 | mask = swapin_nr_pages(offset) - 1; |
| 623 | if (!mask) |
| 624 | goto skip; |
| 625 | |
Shaohua Li | 2395562 | 2017-07-10 15:47:11 -0700 | [diff] [blame] | 626 | do_poll = false; |
Rik van Riel | 67f96aa | 2012-03-21 16:33:50 -0700 | [diff] [blame] | 627 | /* 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 Ying | e9a6eff | 2017-11-15 17:33:15 -0800 | [diff] [blame] | 632 | if (end_offset >= si->max) |
| 633 | end_offset = si->max - 1; |
Rik van Riel | 67f96aa | 2012-03-21 16:33:50 -0700 | [diff] [blame] | 634 | |
Christian Ehrhardt | 3fb5c29 | 2012-07-31 16:41:44 -0700 | [diff] [blame] | 635 | blk_start_plug(&plug); |
Rik van Riel | 67f96aa | 2012-03-21 16:33:50 -0700 | [diff] [blame] | 636 | for (offset = start_offset; offset <= end_offset ; offset++) { |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 637 | /* Ok, do the async read-ahead now */ |
Huang Ying | c4fa630 | 2017-09-06 16:24:33 -0700 | [diff] [blame] | 638 | page = __read_swap_cache_async( |
| 639 | swp_entry(swp_type(entry), offset), |
| 640 | gfp_mask, vma, addr, &page_allocated); |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 641 | if (!page) |
Rik van Riel | 67f96aa | 2012-03-21 16:33:50 -0700 | [diff] [blame] | 642 | continue; |
Huang Ying | c4fa630 | 2017-09-06 16:24:33 -0700 | [diff] [blame] | 643 | if (page_allocated) { |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 644 | swap_readpage(page, false, &splug); |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 645 | if (offset != entry_offset) { |
Huang Ying | c4fa630 | 2017-09-06 16:24:33 -0700 | [diff] [blame] | 646 | SetPageReadahead(page); |
| 647 | count_vm_event(SWAP_RA); |
| 648 | } |
Huang Ying | cbc65df | 2017-09-06 16:24:29 -0700 | [diff] [blame] | 649 | } |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 650 | put_page(page); |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 651 | } |
Christian Ehrhardt | 3fb5c29 | 2012-07-31 16:41:44 -0700 | [diff] [blame] | 652 | blk_finish_plug(&plug); |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 653 | swap_read_unplug(splug); |
Christian Ehrhardt | 3fb5c29 | 2012-07-31 16:41:44 -0700 | [diff] [blame] | 654 | |
Hugh Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 655 | lru_add_drain(); /* Push any new pages onto the LRU now */ |
Shaohua Li | 579f829 | 2014-02-06 12:04:21 -0800 | [diff] [blame] | 656 | skip: |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 657 | /* 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 Dickins | 46017e9 | 2008-02-04 22:28:41 -0800 | [diff] [blame] | 659 | } |
Huang, Ying | 4b3ef9d | 2017-02-22 15:45:26 -0800 | [diff] [blame] | 660 | |
| 661 | int 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 Cook | 778e1cd | 2018-06-12 14:04:48 -0700 | [diff] [blame] | 667 | spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL); |
Huang, Ying | 4b3ef9d | 2017-02-22 15:45:26 -0800 | [diff] [blame] | 668 | if (!spaces) |
| 669 | return -ENOMEM; |
| 670 | for (i = 0; i < nr; i++) { |
| 671 | space = spaces + i; |
Matthew Wilcox | a283348 | 2017-12-05 19:04:20 -0500 | [diff] [blame] | 672 | xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ); |
Huang, Ying | 4b3ef9d | 2017-02-22 15:45:26 -0800 | [diff] [blame] | 673 | 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, Ying | 4b3ef9d | 2017-02-22 15:45:26 -0800 | [diff] [blame] | 677 | } |
| 678 | nr_swapper_spaces[type] = nr; |
Huang Ying | 054f1d1 | 2019-07-11 20:55:37 -0700 | [diff] [blame] | 679 | swapper_spaces[type] = spaces; |
Huang, Ying | 4b3ef9d | 2017-02-22 15:45:26 -0800 | [diff] [blame] | 680 | |
| 681 | return 0; |
| 682 | } |
| 683 | |
| 684 | void exit_swap_address_space(unsigned int type) |
| 685 | { |
Huang Ying | eea4a50 | 2021-06-28 19:37:16 -0700 | [diff] [blame] | 686 | 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, Ying | 4b3ef9d | 2017-02-22 15:45:26 -0800 | [diff] [blame] | 692 | nr_swapper_spaces[type] = 0; |
Huang Ying | 054f1d1 | 2019-07-11 20:55:37 -0700 | [diff] [blame] | 693 | swapper_spaces[type] = NULL; |
Huang, Ying | 4b3ef9d | 2017-02-22 15:45:26 -0800 | [diff] [blame] | 694 | } |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 695 | |
| 696 | static 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 Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 709 | static void swap_ra_info(struct vm_fault *vmf, |
| 710 | struct vma_swap_readahead *ra_info) |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 711 | { |
| 712 | struct vm_area_struct *vma = vmf->vma; |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 713 | unsigned long ra_val; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 714 | unsigned long faddr, pfn, fpfn; |
| 715 | unsigned long start, end; |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 716 | pte_t *pte, *orig_pte; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 717 | unsigned int max_win, hits, prev_win, win, left; |
| 718 | #ifndef CONFIG_64BIT |
| 719 | pte_t *tpte; |
| 720 | #endif |
| 721 | |
Huang Ying | 61b6397 | 2017-10-13 15:58:29 -0700 | [diff] [blame] | 722 | max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster), |
| 723 | SWAP_RA_ORDER_CEILING); |
| 724 | if (max_win == 1) { |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 725 | ra_info->win = 1; |
| 726 | return; |
Huang Ying | 61b6397 | 2017-10-13 15:58:29 -0700 | [diff] [blame] | 727 | } |
| 728 | |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 729 | faddr = vmf->address; |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 730 | orig_pte = pte = pte_offset_map(vmf->pmd, faddr); |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 731 | |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 732 | fpfn = PFN_DOWN(faddr); |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 733 | 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 Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 738 | max_win, prev_win); |
| 739 | atomic_long_set(&vma->swap_readahead_info, |
| 740 | SWAP_RA_VAL(faddr, win, 0)); |
| 741 | |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 742 | if (win == 1) { |
| 743 | pte_unmap(orig_pte); |
| 744 | return; |
| 745 | } |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 746 | |
| 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 Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 758 | ra_info->nr_pte = end - start; |
| 759 | ra_info->offset = fpfn - start; |
| 760 | pte -= ra_info->offset; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 761 | #ifdef CONFIG_64BIT |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 762 | ra_info->ptes = pte; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 763 | #else |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 764 | tpte = ra_info->ptes; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 765 | for (pfn = start; pfn != end; pfn++) |
| 766 | *tpte++ = *pte++; |
| 767 | #endif |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 768 | pte_unmap(orig_pte); |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 769 | } |
| 770 | |
Yang Shi | e9f5987 | 2019-03-05 15:44:15 -0800 | [diff] [blame] | 771 | /** |
| 772 | * swap_vma_readahead - swap in pages in hope we need them soon |
Krzysztof Kozlowski | 27ec487 | 2020-08-06 23:20:14 -0700 | [diff] [blame] | 773 | * @fentry: swap entry of this memory |
Yang Shi | e9f5987 | 2019-03-05 15:44:15 -0800 | [diff] [blame] | 774 | * @gfp_mask: memory allocation flags |
| 775 | * @vmf: fault information |
| 776 | * |
| 777 | * Returns the struct page for entry and addr, after queueing swapin. |
| 778 | * |
Shijie Luo | cb152a1 | 2021-05-06 18:05:51 -0700 | [diff] [blame] | 779 | * Primitive swap readahead code. We simply read in a few pages whose |
Yang Shi | e9f5987 | 2019-03-05 15:44:15 -0800 | [diff] [blame] | 780 | * virtual addresses are around the fault address in the same vma. |
| 781 | * |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 782 | * Caller must hold read mmap_lock if vmf->vma is not NULL. |
Yang Shi | e9f5987 | 2019-03-05 15:44:15 -0800 | [diff] [blame] | 783 | * |
| 784 | */ |
Colin Ian King | f5c754d | 2018-04-05 16:25:05 -0700 | [diff] [blame] | 785 | static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask, |
| 786 | struct vm_fault *vmf) |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 787 | { |
| 788 | struct blk_plug plug; |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 789 | struct swap_iocb *splug = NULL; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 790 | 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 Lin | e97af69 | 2020-12-14 19:06:01 -0800 | [diff] [blame] | 796 | struct vma_swap_readahead ra_info = { |
| 797 | .win = 1, |
| 798 | }; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 799 | |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 800 | swap_ra_info(vmf, &ra_info); |
| 801 | if (ra_info.win == 1) |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 802 | goto skip; |
| 803 | |
| 804 | blk_start_plug(&plug); |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 805 | for (i = 0, pte = ra_info.ptes; i < ra_info.nr_pte; |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 806 | i++, pte++) { |
| 807 | pentry = *pte; |
Miaohe Lin | 92bafb2 | 2022-05-19 14:08:50 -0700 | [diff] [blame] | 808 | if (!is_swap_pte(pentry)) |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 809 | 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) { |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 818 | swap_readpage(page, false, &splug); |
Minchan Kim | eaf649eb | 2018-04-05 16:23:39 -0700 | [diff] [blame] | 819 | if (i != ra_info.offset) { |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 820 | SetPageReadahead(page); |
| 821 | count_vm_event(SWAP_RA); |
| 822 | } |
| 823 | } |
| 824 | put_page(page); |
| 825 | } |
| 826 | blk_finish_plug(&plug); |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 827 | swap_read_unplug(splug); |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 828 | lru_add_drain(); |
| 829 | skip: |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 830 | /* The page was likely read above, so no need for plugging here */ |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 831 | return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address, |
NeilBrown | 5169b84 | 2022-05-09 18:20:49 -0700 | [diff] [blame] | 832 | ra_info.win == 1, NULL); |
Huang Ying | ec56017 | 2017-09-06 16:24:36 -0700 | [diff] [blame] | 833 | } |
Huang Ying | d9bfcfd | 2017-09-06 16:24:40 -0700 | [diff] [blame] | 834 | |
Minchan Kim | e9e9b7e | 2018-04-05 16:23:42 -0700 | [diff] [blame] | 835 | /** |
| 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 | */ |
| 847 | struct 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 Ying | d9bfcfd | 2017-09-06 16:24:40 -0700 | [diff] [blame] | 855 | #ifdef CONFIG_SYSFS |
| 856 | static ssize_t vma_ra_enabled_show(struct kobject *kobj, |
| 857 | struct kobj_attribute *attr, char *buf) |
| 858 | { |
Joe Perches | ae7a927 | 2020-12-14 19:14:42 -0800 | [diff] [blame] | 859 | return sysfs_emit(buf, "%s\n", |
| 860 | enable_vma_readahead ? "true" : "false"); |
Huang Ying | d9bfcfd | 2017-09-06 16:24:40 -0700 | [diff] [blame] | 861 | } |
| 862 | static ssize_t vma_ra_enabled_store(struct kobject *kobj, |
| 863 | struct kobj_attribute *attr, |
| 864 | const char *buf, size_t count) |
| 865 | { |
Jagdish Gediya | 717aeab | 2022-05-12 20:22:59 -0700 | [diff] [blame] | 866 | ssize_t ret; |
| 867 | |
| 868 | ret = kstrtobool(buf, &enable_vma_readahead); |
| 869 | if (ret) |
| 870 | return ret; |
Huang Ying | d9bfcfd | 2017-09-06 16:24:40 -0700 | [diff] [blame] | 871 | |
| 872 | return count; |
| 873 | } |
Miaohe Lin | 6106b93 | 2022-05-19 14:08:50 -0700 | [diff] [blame] | 874 | static struct kobj_attribute vma_ra_enabled_attr = __ATTR_RW(vma_ra_enabled); |
Huang Ying | d9bfcfd | 2017-09-06 16:24:40 -0700 | [diff] [blame] | 875 | |
Huang Ying | d9bfcfd | 2017-09-06 16:24:40 -0700 | [diff] [blame] | 876 | static struct attribute *swap_attrs[] = { |
| 877 | &vma_ra_enabled_attr.attr, |
Huang Ying | d9bfcfd | 2017-09-06 16:24:40 -0700 | [diff] [blame] | 878 | NULL, |
| 879 | }; |
| 880 | |
Rikard Falkeborn | e48333b | 2021-02-24 12:03:05 -0800 | [diff] [blame] | 881 | static const struct attribute_group swap_attr_group = { |
Huang Ying | d9bfcfd | 2017-09-06 16:24:40 -0700 | [diff] [blame] | 882 | .attrs = swap_attrs, |
| 883 | }; |
| 884 | |
| 885 | static 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 | |
| 902 | delete_obj: |
| 903 | kobject_put(swap_kobj); |
| 904 | return err; |
| 905 | } |
| 906 | subsys_initcall(swap_init_sysfs); |
| 907 | #endif |