1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/mm/swap_state.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 61da177e4SLinus Torvalds * Swap reorganised 29.12.95, Stephen Tweedie 71da177e4SLinus Torvalds * 81da177e4SLinus Torvalds * Rewritten to use page cache, (C) 1998 Stephen Tweedie 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds #include <linux/mm.h> 115a0e3ad6STejun Heo #include <linux/gfp.h> 121da177e4SLinus Torvalds #include <linux/kernel_stat.h> 131da177e4SLinus Torvalds #include <linux/swap.h> 1446017e95SHugh Dickins #include <linux/swapops.h> 151da177e4SLinus Torvalds #include <linux/init.h> 161da177e4SLinus Torvalds #include <linux/pagemap.h> 171da177e4SLinus Torvalds #include <linux/backing-dev.h> 183fb5c298SChristian Ehrhardt #include <linux/blkdev.h> 19c484d410SHugh Dickins #include <linux/pagevec.h> 20b20a3503SChristoph Lameter #include <linux/migrate.h> 214b3ef9daSHuang, Ying #include <linux/vmalloc.h> 2267afa38eSTim Chen #include <linux/swap_slots.h> 2338d8b4e6SHuang Ying #include <linux/huge_mm.h> 2461ef1865SMatthew Wilcox (Oracle) #include <linux/shmem_fs.h> 25243bce09SHugh Dickins #include "internal.h" 26014bb1deSNeilBrown #include "swap.h" 271da177e4SLinus Torvalds 281da177e4SLinus Torvalds /* 291da177e4SLinus Torvalds * swapper_space is a fiction, retained to simplify the path through 307eaceaccSJens Axboe * vmscan's shrink_page_list. 311da177e4SLinus Torvalds */ 32f5e54d6eSChristoph Hellwig static const struct address_space_operations swap_aops = { 331da177e4SLinus Torvalds .writepage = swap_writepage, 344c4a7634SNeilBrown .dirty_folio = noop_dirty_folio, 351c93923cSAndrew Morton #ifdef CONFIG_MIGRATION 36e965f963SChristoph Lameter .migratepage = migrate_page, 371c93923cSAndrew Morton #endif 381da177e4SLinus Torvalds }; 391da177e4SLinus Torvalds 40783cb68eSChangbin Du struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly; 41783cb68eSChangbin Du static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly; 42f5c754d6SColin Ian King static bool enable_vma_readahead __read_mostly = true; 43ec560175SHuang Ying 44ec560175SHuang Ying #define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2) 45ec560175SHuang Ying #define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1) 46ec560175SHuang Ying #define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK 47ec560175SHuang Ying #define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK) 48ec560175SHuang Ying 49ec560175SHuang Ying #define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK) 50ec560175SHuang Ying #define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT) 51ec560175SHuang Ying #define SWAP_RA_ADDR(v) ((v) & PAGE_MASK) 52ec560175SHuang Ying 53ec560175SHuang Ying #define SWAP_RA_VAL(addr, win, hits) \ 54ec560175SHuang Ying (((addr) & PAGE_MASK) | \ 55ec560175SHuang Ying (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \ 56ec560175SHuang Ying ((hits) & SWAP_RA_HITS_MASK)) 57ec560175SHuang Ying 58ec560175SHuang Ying /* Initial readahead hits is 4 to start up with a small window */ 59ec560175SHuang Ying #define GET_SWAP_RA_VAL(vma) \ 60ec560175SHuang Ying (atomic_long_read(&(vma)->swap_readahead_info) ? : 4) 611da177e4SLinus Torvalds 62b96a3db2SQian Cai #define INC_CACHE_INFO(x) data_race(swap_cache_info.x++) 63b96a3db2SQian Cai #define ADD_CACHE_INFO(x, nr) data_race(swap_cache_info.x += (nr)) 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds static struct { 661da177e4SLinus Torvalds unsigned long add_total; 671da177e4SLinus Torvalds unsigned long del_total; 681da177e4SLinus Torvalds unsigned long find_success; 691da177e4SLinus Torvalds unsigned long find_total; 701da177e4SLinus Torvalds } swap_cache_info; 711da177e4SLinus Torvalds 72579f8290SShaohua Li static atomic_t swapin_readahead_hits = ATOMIC_INIT(4); 73579f8290SShaohua Li 741da177e4SLinus Torvalds void show_swap_cache_info(void) 751da177e4SLinus Torvalds { 7633806f06SShaohua Li printk("%lu pages in swap cache\n", total_swapcache_pages()); 772c97b7fcSJohannes Weiner printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n", 781da177e4SLinus Torvalds swap_cache_info.add_total, swap_cache_info.del_total, 79bb63be0aSHugh Dickins swap_cache_info.find_success, swap_cache_info.find_total); 80ec8acf20SShaohua Li printk("Free swap = %ldkB\n", 81ec8acf20SShaohua Li get_nr_swap_pages() << (PAGE_SHIFT - 10)); 821da177e4SLinus Torvalds printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10)); 831da177e4SLinus Torvalds } 841da177e4SLinus Torvalds 85aae466b0SJoonsoo Kim void *get_shadow_from_swap_cache(swp_entry_t entry) 86aae466b0SJoonsoo Kim { 87aae466b0SJoonsoo Kim struct address_space *address_space = swap_address_space(entry); 88aae466b0SJoonsoo Kim pgoff_t idx = swp_offset(entry); 89aae466b0SJoonsoo Kim struct page *page; 90aae466b0SJoonsoo Kim 918c647dd1SMatthew Wilcox (Oracle) page = xa_load(&address_space->i_pages, idx); 92aae466b0SJoonsoo Kim if (xa_is_value(page)) 93aae466b0SJoonsoo Kim return page; 94aae466b0SJoonsoo Kim return NULL; 95aae466b0SJoonsoo Kim } 96aae466b0SJoonsoo Kim 971da177e4SLinus Torvalds /* 988d93b41cSMatthew Wilcox * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space, 991da177e4SLinus Torvalds * but sets SwapCache flag and private instead of mapping and index. 1001da177e4SLinus Torvalds */ 1013852f676SJoonsoo Kim int add_to_swap_cache(struct page *page, swp_entry_t entry, 1023852f676SJoonsoo Kim gfp_t gfp, void **shadowp) 1031da177e4SLinus Torvalds { 1048d93b41cSMatthew Wilcox struct address_space *address_space = swap_address_space(entry); 10538d8b4e6SHuang Ying pgoff_t idx = swp_offset(entry); 1068d93b41cSMatthew Wilcox XA_STATE_ORDER(xas, &address_space->i_pages, idx, compound_order(page)); 1076c357848SMatthew Wilcox (Oracle) unsigned long i, nr = thp_nr_pages(page); 1083852f676SJoonsoo Kim void *old; 1091da177e4SLinus Torvalds 110309381feSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page), page); 111309381feSSasha Levin VM_BUG_ON_PAGE(PageSwapCache(page), page); 112309381feSSasha Levin VM_BUG_ON_PAGE(!PageSwapBacked(page), page); 11351726b12SHugh Dickins 11438d8b4e6SHuang Ying page_ref_add(page, nr); 1151da177e4SLinus Torvalds SetPageSwapCache(page); 116e286781dSNick Piggin 1178d93b41cSMatthew Wilcox do { 1188d93b41cSMatthew Wilcox xas_lock_irq(&xas); 1198d93b41cSMatthew Wilcox xas_create_range(&xas); 1208d93b41cSMatthew Wilcox if (xas_error(&xas)) 1218d93b41cSMatthew Wilcox goto unlock; 12238d8b4e6SHuang Ying for (i = 0; i < nr; i++) { 1238d93b41cSMatthew Wilcox VM_BUG_ON_PAGE(xas.xa_index != idx + i, page); 1243852f676SJoonsoo Kim old = xas_load(&xas); 1253852f676SJoonsoo Kim if (xa_is_value(old)) { 1263852f676SJoonsoo Kim if (shadowp) 1273852f676SJoonsoo Kim *shadowp = old; 1283852f676SJoonsoo Kim } 12938d8b4e6SHuang Ying set_page_private(page + i, entry.val + i); 1304101196bSMatthew Wilcox (Oracle) xas_store(&xas, page); 1318d93b41cSMatthew Wilcox xas_next(&xas); 1321da177e4SLinus Torvalds } 13338d8b4e6SHuang Ying address_space->nrpages += nr; 13438d8b4e6SHuang Ying __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); 135b6038942SShakeel Butt __mod_lruvec_page_state(page, NR_SWAPCACHE, nr); 13638d8b4e6SHuang Ying ADD_CACHE_INFO(add_total, nr); 1378d93b41cSMatthew Wilcox unlock: 1388d93b41cSMatthew Wilcox xas_unlock_irq(&xas); 1398d93b41cSMatthew Wilcox } while (xas_nomem(&xas, gfp)); 1408d93b41cSMatthew Wilcox 1418d93b41cSMatthew Wilcox if (!xas_error(&xas)) 1428d93b41cSMatthew Wilcox return 0; 1438d93b41cSMatthew Wilcox 14438d8b4e6SHuang Ying ClearPageSwapCache(page); 14538d8b4e6SHuang Ying page_ref_sub(page, nr); 1468d93b41cSMatthew Wilcox return xas_error(&xas); 1471da177e4SLinus Torvalds } 1481da177e4SLinus Torvalds 1491da177e4SLinus Torvalds /* 1501da177e4SLinus Torvalds * This must be called only on pages that have 1511da177e4SLinus Torvalds * been verified to be in the swap cache. 1521da177e4SLinus Torvalds */ 1533852f676SJoonsoo Kim void __delete_from_swap_cache(struct page *page, 1543852f676SJoonsoo Kim swp_entry_t entry, void *shadow) 1551da177e4SLinus Torvalds { 1564e17ec25SMatthew Wilcox struct address_space *address_space = swap_address_space(entry); 1576c357848SMatthew Wilcox (Oracle) int i, nr = thp_nr_pages(page); 1584e17ec25SMatthew Wilcox pgoff_t idx = swp_offset(entry); 1594e17ec25SMatthew Wilcox XA_STATE(xas, &address_space->i_pages, idx); 16033806f06SShaohua Li 161309381feSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page), page); 162309381feSSasha Levin VM_BUG_ON_PAGE(!PageSwapCache(page), page); 163309381feSSasha Levin VM_BUG_ON_PAGE(PageWriteback(page), page); 1641da177e4SLinus Torvalds 16538d8b4e6SHuang Ying for (i = 0; i < nr; i++) { 1663852f676SJoonsoo Kim void *entry = xas_store(&xas, shadow); 1674101196bSMatthew Wilcox (Oracle) VM_BUG_ON_PAGE(entry != page, entry); 16838d8b4e6SHuang Ying set_page_private(page + i, 0); 1694e17ec25SMatthew Wilcox xas_next(&xas); 17038d8b4e6SHuang Ying } 1711da177e4SLinus Torvalds ClearPageSwapCache(page); 17238d8b4e6SHuang Ying address_space->nrpages -= nr; 17338d8b4e6SHuang Ying __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); 174b6038942SShakeel Butt __mod_lruvec_page_state(page, NR_SWAPCACHE, -nr); 17538d8b4e6SHuang Ying ADD_CACHE_INFO(del_total, nr); 1761da177e4SLinus Torvalds } 1771da177e4SLinus Torvalds 1781da177e4SLinus Torvalds /** 17909c02e56SMatthew Wilcox (Oracle) * add_to_swap - allocate swap space for a folio 18009c02e56SMatthew Wilcox (Oracle) * @folio: folio we want to move to swap 1811da177e4SLinus Torvalds * 18209c02e56SMatthew Wilcox (Oracle) * Allocate swap space for the folio and add the folio to the 18309c02e56SMatthew Wilcox (Oracle) * swap cache. 18409c02e56SMatthew Wilcox (Oracle) * 18509c02e56SMatthew Wilcox (Oracle) * Context: Caller needs to hold the folio lock. 18609c02e56SMatthew Wilcox (Oracle) * Return: Whether the folio was added to the swap cache. 1871da177e4SLinus Torvalds */ 18809c02e56SMatthew Wilcox (Oracle) bool add_to_swap(struct folio *folio) 1891da177e4SLinus Torvalds { 1901da177e4SLinus Torvalds swp_entry_t entry; 1911da177e4SLinus Torvalds int err; 1921da177e4SLinus Torvalds 19309c02e56SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio); 19409c02e56SMatthew Wilcox (Oracle) VM_BUG_ON_FOLIO(!folio_test_uptodate(folio), folio); 1951da177e4SLinus Torvalds 196e2e3fdc7SMatthew Wilcox (Oracle) entry = folio_alloc_swap(folio); 1971da177e4SLinus Torvalds if (!entry.val) 19809c02e56SMatthew Wilcox (Oracle) return false; 1990f074658SMinchan Kim 200bd53b714SNick Piggin /* 2018d93b41cSMatthew Wilcox * XArray node allocations from PF_MEMALLOC contexts could 202bd53b714SNick Piggin * completely exhaust the page allocator. __GFP_NOMEMALLOC 203bd53b714SNick Piggin * stops emergency reserves from being allocated. 2041da177e4SLinus Torvalds * 205bd53b714SNick Piggin * TODO: this could cause a theoretical memory reclaim 206bd53b714SNick Piggin * deadlock in the swap out path. 2071da177e4SLinus Torvalds */ 2081da177e4SLinus Torvalds /* 209854e9ed0SMinchan Kim * Add it to the swap cache. 2101da177e4SLinus Torvalds */ 21109c02e56SMatthew Wilcox (Oracle) err = add_to_swap_cache(&folio->page, entry, 2123852f676SJoonsoo Kim __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN, NULL); 21338d8b4e6SHuang Ying if (err) 2142ca4532aSDaisuke Nishimura /* 2152ca4532aSDaisuke Nishimura * add_to_swap_cache() doesn't return -EEXIST, so we can safely 2162ca4532aSDaisuke Nishimura * clear SWAP_HAS_CACHE flag. 2172ca4532aSDaisuke Nishimura */ 2180f074658SMinchan Kim goto fail; 2199625456cSShaohua Li /* 22009c02e56SMatthew Wilcox (Oracle) * Normally the folio will be dirtied in unmap because its 22109c02e56SMatthew Wilcox (Oracle) * pte should be dirty. A special case is MADV_FREE page. The 22209c02e56SMatthew Wilcox (Oracle) * page's pte could have dirty bit cleared but the folio's 22309c02e56SMatthew Wilcox (Oracle) * SwapBacked flag is still set because clearing the dirty bit 22409c02e56SMatthew Wilcox (Oracle) * and SwapBacked flag has no lock protected. For such folio, 22509c02e56SMatthew Wilcox (Oracle) * unmap will not set dirty bit for it, so folio reclaim will 22609c02e56SMatthew Wilcox (Oracle) * not write the folio out. This can cause data corruption when 22709c02e56SMatthew Wilcox (Oracle) * the folio is swapped in later. Always setting the dirty flag 22809c02e56SMatthew Wilcox (Oracle) * for the folio solves the problem. 2299625456cSShaohua Li */ 23009c02e56SMatthew Wilcox (Oracle) folio_mark_dirty(folio); 2311da177e4SLinus Torvalds 23209c02e56SMatthew Wilcox (Oracle) return true; 23338d8b4e6SHuang Ying 23438d8b4e6SHuang Ying fail: 23509c02e56SMatthew Wilcox (Oracle) put_swap_page(&folio->page, entry); 23609c02e56SMatthew Wilcox (Oracle) return false; 23738d8b4e6SHuang Ying } 23838d8b4e6SHuang Ying 2391da177e4SLinus Torvalds /* 2401da177e4SLinus Torvalds * This must be called only on pages that have 2411da177e4SLinus Torvalds * been verified to be in the swap cache and locked. 2421da177e4SLinus Torvalds * It will never put the page into the free list, 2431da177e4SLinus Torvalds * the caller has a reference on the page. 2441da177e4SLinus Torvalds */ 2451da177e4SLinus Torvalds void delete_from_swap_cache(struct page *page) 2461da177e4SLinus Torvalds { 2474e17ec25SMatthew Wilcox swp_entry_t entry = { .val = page_private(page) }; 2484e17ec25SMatthew Wilcox struct address_space *address_space = swap_address_space(entry); 2491da177e4SLinus Torvalds 250b93b0163SMatthew Wilcox xa_lock_irq(&address_space->i_pages); 2513852f676SJoonsoo Kim __delete_from_swap_cache(page, entry, NULL); 252b93b0163SMatthew Wilcox xa_unlock_irq(&address_space->i_pages); 2531da177e4SLinus Torvalds 25475f6d6d2SMinchan Kim put_swap_page(page, entry); 2556c357848SMatthew Wilcox (Oracle) page_ref_sub(page, thp_nr_pages(page)); 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 2583852f676SJoonsoo Kim void clear_shadow_from_swap_cache(int type, unsigned long begin, 2593852f676SJoonsoo Kim unsigned long end) 2603852f676SJoonsoo Kim { 2613852f676SJoonsoo Kim unsigned long curr = begin; 2623852f676SJoonsoo Kim void *old; 2633852f676SJoonsoo Kim 2643852f676SJoonsoo Kim for (;;) { 2653852f676SJoonsoo Kim swp_entry_t entry = swp_entry(type, curr); 2663852f676SJoonsoo Kim struct address_space *address_space = swap_address_space(entry); 2673852f676SJoonsoo Kim XA_STATE(xas, &address_space->i_pages, curr); 2683852f676SJoonsoo Kim 2693852f676SJoonsoo Kim xa_lock_irq(&address_space->i_pages); 2703852f676SJoonsoo Kim xas_for_each(&xas, old, end) { 2713852f676SJoonsoo Kim if (!xa_is_value(old)) 2723852f676SJoonsoo Kim continue; 2733852f676SJoonsoo Kim xas_store(&xas, NULL); 2743852f676SJoonsoo Kim } 2753852f676SJoonsoo Kim xa_unlock_irq(&address_space->i_pages); 2763852f676SJoonsoo Kim 2773852f676SJoonsoo Kim /* search the next swapcache until we meet end */ 2783852f676SJoonsoo Kim curr >>= SWAP_ADDRESS_SPACE_SHIFT; 2793852f676SJoonsoo Kim curr++; 2803852f676SJoonsoo Kim curr <<= SWAP_ADDRESS_SPACE_SHIFT; 2813852f676SJoonsoo Kim if (curr > end) 2823852f676SJoonsoo Kim break; 2833852f676SJoonsoo Kim } 2843852f676SJoonsoo Kim } 2853852f676SJoonsoo Kim 2861da177e4SLinus Torvalds /* 2871da177e4SLinus Torvalds * If we are the only user, then try to free up the swap cache. 2881da177e4SLinus Torvalds * 2891da177e4SLinus Torvalds * Its ok to check for PageSwapCache without the page lock 2901da177e4SLinus Torvalds * here because we are going to recheck again inside 291a2c43eedSHugh Dickins * try_to_free_swap() _with_ the lock. 2921da177e4SLinus Torvalds * - Marcelo 2931da177e4SLinus Torvalds */ 294f4c4a3f4SHuang Ying void free_swap_cache(struct page *page) 2951da177e4SLinus Torvalds { 296a2c43eedSHugh Dickins if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) { 297a2c43eedSHugh Dickins try_to_free_swap(page); 2981da177e4SLinus Torvalds unlock_page(page); 2991da177e4SLinus Torvalds } 3001da177e4SLinus Torvalds } 3011da177e4SLinus Torvalds 3021da177e4SLinus Torvalds /* 3031da177e4SLinus Torvalds * Perform a free_page(), also freeing any swap cache associated with 304b8072f09SHugh Dickins * this page if it is the last user of the page. 3051da177e4SLinus Torvalds */ 3061da177e4SLinus Torvalds void free_page_and_swap_cache(struct page *page) 3071da177e4SLinus Torvalds { 3081da177e4SLinus Torvalds free_swap_cache(page); 3096fcb52a5SAaron Lu if (!is_huge_zero_page(page)) 31009cbfeafSKirill A. Shutemov put_page(page); 3111da177e4SLinus Torvalds } 3121da177e4SLinus Torvalds 3131da177e4SLinus Torvalds /* 3141da177e4SLinus Torvalds * Passed an array of pages, drop them all from swapcache and then release 3151da177e4SLinus Torvalds * them. They are removed from the LRU and freed if this is their last use. 3161da177e4SLinus Torvalds */ 3171da177e4SLinus Torvalds void free_pages_and_swap_cache(struct page **pages, int nr) 3181da177e4SLinus Torvalds { 3191da177e4SLinus Torvalds struct page **pagep = pages; 3201da177e4SLinus Torvalds int i; 3211da177e4SLinus Torvalds 322aabfb572SMichal Hocko lru_add_drain(); 323aabfb572SMichal Hocko for (i = 0; i < nr; i++) 3241da177e4SLinus Torvalds free_swap_cache(pagep[i]); 325c6f92f9fSMel Gorman release_pages(pagep, nr); 3261da177e4SLinus Torvalds } 3271da177e4SLinus Torvalds 328e9e9b7ecSMinchan Kim static inline bool swap_use_vma_readahead(void) 329e9e9b7ecSMinchan Kim { 330e9e9b7ecSMinchan Kim return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap); 331e9e9b7ecSMinchan Kim } 332e9e9b7ecSMinchan Kim 3331da177e4SLinus Torvalds /* 3341da177e4SLinus Torvalds * Lookup a swap entry in the swap cache. A found page will be returned 3351da177e4SLinus Torvalds * unlocked and with its refcount incremented - we rely on the kernel 3361da177e4SLinus Torvalds * lock getting page table operations atomic even if we drop the page 3371da177e4SLinus Torvalds * lock before returning. 3381da177e4SLinus Torvalds */ 339ec560175SHuang Ying struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma, 340ec560175SHuang Ying unsigned long addr) 3411da177e4SLinus Torvalds { 3421da177e4SLinus Torvalds struct page *page; 343eb085574SHuang Ying struct swap_info_struct *si; 3441da177e4SLinus Torvalds 345eb085574SHuang Ying si = get_swap_device(entry); 346eb085574SHuang Ying if (!si) 347eb085574SHuang Ying return NULL; 348f6ab1f7fSHuang Ying page = find_get_page(swap_address_space(entry), swp_offset(entry)); 349eb085574SHuang Ying put_swap_device(si); 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds INC_CACHE_INFO(find_total); 352ec560175SHuang Ying if (page) { 353eaf649ebSMinchan Kim bool vma_ra = swap_use_vma_readahead(); 354eaf649ebSMinchan Kim bool readahead; 355eaf649ebSMinchan Kim 356ec560175SHuang Ying INC_CACHE_INFO(find_success); 357eaf649ebSMinchan Kim /* 358eaf649ebSMinchan Kim * At the moment, we don't support PG_readahead for anon THP 359eaf649ebSMinchan Kim * so let's bail out rather than confusing the readahead stat. 360eaf649ebSMinchan Kim */ 361ec560175SHuang Ying if (unlikely(PageTransCompound(page))) 362ec560175SHuang Ying return page; 363eaf649ebSMinchan Kim 364ec560175SHuang Ying readahead = TestClearPageReadahead(page); 365eaf649ebSMinchan Kim if (vma && vma_ra) { 366eaf649ebSMinchan Kim unsigned long ra_val; 367eaf649ebSMinchan Kim int win, hits; 368eaf649ebSMinchan Kim 369eaf649ebSMinchan Kim ra_val = GET_SWAP_RA_VAL(vma); 370eaf649ebSMinchan Kim win = SWAP_RA_WIN(ra_val); 371eaf649ebSMinchan Kim hits = SWAP_RA_HITS(ra_val); 372ec560175SHuang Ying if (readahead) 373ec560175SHuang Ying hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX); 374ec560175SHuang Ying atomic_long_set(&vma->swap_readahead_info, 375ec560175SHuang Ying SWAP_RA_VAL(addr, win, hits)); 376ec560175SHuang Ying } 377eaf649ebSMinchan Kim 378ec560175SHuang Ying if (readahead) { 379ec560175SHuang Ying count_vm_event(SWAP_RA_HIT); 380eaf649ebSMinchan Kim if (!vma || !vma_ra) 381ec560175SHuang Ying atomic_inc(&swapin_readahead_hits); 382ec560175SHuang Ying } 383ec560175SHuang Ying } 384eaf649ebSMinchan Kim 3851da177e4SLinus Torvalds return page; 3861da177e4SLinus Torvalds } 3871da177e4SLinus Torvalds 38861ef1865SMatthew Wilcox (Oracle) /** 38961ef1865SMatthew Wilcox (Oracle) * find_get_incore_page - Find and get a page from the page or swap caches. 39061ef1865SMatthew Wilcox (Oracle) * @mapping: The address_space to search. 39161ef1865SMatthew Wilcox (Oracle) * @index: The page cache index. 39261ef1865SMatthew Wilcox (Oracle) * 39361ef1865SMatthew Wilcox (Oracle) * This differs from find_get_page() in that it will also look for the 39461ef1865SMatthew Wilcox (Oracle) * page in the swap cache. 39561ef1865SMatthew Wilcox (Oracle) * 39661ef1865SMatthew Wilcox (Oracle) * Return: The found page or %NULL. 39761ef1865SMatthew Wilcox (Oracle) */ 39861ef1865SMatthew Wilcox (Oracle) struct page *find_get_incore_page(struct address_space *mapping, pgoff_t index) 39961ef1865SMatthew Wilcox (Oracle) { 40061ef1865SMatthew Wilcox (Oracle) swp_entry_t swp; 40161ef1865SMatthew Wilcox (Oracle) struct swap_info_struct *si; 40244835d20SMatthew Wilcox (Oracle) struct page *page = pagecache_get_page(mapping, index, 40344835d20SMatthew Wilcox (Oracle) FGP_ENTRY | FGP_HEAD, 0); 40461ef1865SMatthew Wilcox (Oracle) 405a6de4b48SMatthew Wilcox (Oracle) if (!page) 40661ef1865SMatthew Wilcox (Oracle) return page; 407a6de4b48SMatthew Wilcox (Oracle) if (!xa_is_value(page)) 408a6de4b48SMatthew Wilcox (Oracle) return find_subpage(page, index); 40961ef1865SMatthew Wilcox (Oracle) if (!shmem_mapping(mapping)) 41061ef1865SMatthew Wilcox (Oracle) return NULL; 41161ef1865SMatthew Wilcox (Oracle) 41261ef1865SMatthew Wilcox (Oracle) swp = radix_to_swp_entry(page); 41361ef1865SMatthew Wilcox (Oracle) /* Prevent swapoff from happening to us */ 41461ef1865SMatthew Wilcox (Oracle) si = get_swap_device(swp); 41561ef1865SMatthew Wilcox (Oracle) if (!si) 41661ef1865SMatthew Wilcox (Oracle) return NULL; 41761ef1865SMatthew Wilcox (Oracle) page = find_get_page(swap_address_space(swp), swp_offset(swp)); 41861ef1865SMatthew Wilcox (Oracle) put_swap_device(si); 41961ef1865SMatthew Wilcox (Oracle) return page; 42061ef1865SMatthew Wilcox (Oracle) } 42161ef1865SMatthew Wilcox (Oracle) 4225b999aadSDmitry Safonov struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, 4235b999aadSDmitry Safonov struct vm_area_struct *vma, unsigned long addr, 4245b999aadSDmitry Safonov bool *new_page_allocated) 4251da177e4SLinus Torvalds { 426eb085574SHuang Ying struct swap_info_struct *si; 4274c6355b2SJohannes Weiner struct page *page; 428aae466b0SJoonsoo Kim void *shadow = NULL; 4294c6355b2SJohannes Weiner 4305b999aadSDmitry Safonov *new_page_allocated = false; 4311da177e4SLinus Torvalds 4324c6355b2SJohannes Weiner for (;;) { 4334c6355b2SJohannes Weiner int err; 4341da177e4SLinus Torvalds /* 4351da177e4SLinus Torvalds * First check the swap cache. Since this is normally 4361da177e4SLinus Torvalds * called after lookup_swap_cache() failed, re-calling 4371da177e4SLinus Torvalds * that would confuse statistics. 4381da177e4SLinus Torvalds */ 439eb085574SHuang Ying si = get_swap_device(entry); 440eb085574SHuang Ying if (!si) 4414c6355b2SJohannes Weiner return NULL; 4424c6355b2SJohannes Weiner page = find_get_page(swap_address_space(entry), 443eb085574SHuang Ying swp_offset(entry)); 444eb085574SHuang Ying put_swap_device(si); 4454c6355b2SJohannes Weiner if (page) 4464c6355b2SJohannes Weiner return page; 4471da177e4SLinus Torvalds 448ba81f838SHuang Ying /* 449ba81f838SHuang Ying * Just skip read ahead for unused swap slot. 450ba81f838SHuang Ying * During swap_off when swap_slot_cache is disabled, 451ba81f838SHuang Ying * we have to handle the race between putting 452ba81f838SHuang Ying * swap entry in swap cache and marking swap slot 453ba81f838SHuang Ying * as SWAP_HAS_CACHE. That's done in later part of code or 454ba81f838SHuang Ying * else swap_off will be aborted if we return NULL. 455ba81f838SHuang Ying */ 456ba81f838SHuang Ying if (!__swp_swapcount(entry) && swap_slot_cache_enabled) 4574c6355b2SJohannes Weiner return NULL; 458e8c26ab6STim Chen 4591da177e4SLinus Torvalds /* 4604c6355b2SJohannes Weiner * Get a new page to read into from swap. Allocate it now, 4614c6355b2SJohannes Weiner * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will 4624c6355b2SJohannes Weiner * cause any racers to loop around until we add it to cache. 4631da177e4SLinus Torvalds */ 4644c6355b2SJohannes Weiner page = alloc_page_vma(gfp_mask, vma, addr); 4654c6355b2SJohannes Weiner if (!page) 4664c6355b2SJohannes Weiner return NULL; 4671da177e4SLinus Torvalds 4681da177e4SLinus Torvalds /* 469f000944dSHugh Dickins * Swap entry may have been freed since our caller observed it. 470f000944dSHugh Dickins */ 471355cfa73SKAMEZAWA Hiroyuki err = swapcache_prepare(entry); 4724c6355b2SJohannes Weiner if (!err) 473f000944dSHugh Dickins break; 474f000944dSHugh Dickins 4754c6355b2SJohannes Weiner put_page(page); 4764c6355b2SJohannes Weiner if (err != -EEXIST) 4774c6355b2SJohannes Weiner return NULL; 4781da177e4SLinus Torvalds 4794c6355b2SJohannes Weiner /* 4804c6355b2SJohannes Weiner * We might race against __delete_from_swap_cache(), and 4814c6355b2SJohannes Weiner * stumble across a swap_map entry whose SWAP_HAS_CACHE 4824c6355b2SJohannes Weiner * has not yet been cleared. Or race against another 4834c6355b2SJohannes Weiner * __read_swap_cache_async(), which has set SWAP_HAS_CACHE 4844c6355b2SJohannes Weiner * in swap_map, but not yet added its page to swap cache. 4854c6355b2SJohannes Weiner */ 486029c4628SGuo Ziliang schedule_timeout_uninterruptible(1); 4874c6355b2SJohannes Weiner } 4884c6355b2SJohannes Weiner 4894c6355b2SJohannes Weiner /* 4904c6355b2SJohannes Weiner * The swap entry is ours to swap in. Prepare the new page. 4914c6355b2SJohannes Weiner */ 4924c6355b2SJohannes Weiner 4934c6355b2SJohannes Weiner __SetPageLocked(page); 4944c6355b2SJohannes Weiner __SetPageSwapBacked(page); 4954c6355b2SJohannes Weiner 4960add0c77SShakeel Butt if (mem_cgroup_swapin_charge_page(page, NULL, gfp_mask, entry)) 4974c6355b2SJohannes Weiner goto fail_unlock; 4984c6355b2SJohannes Weiner 4990add0c77SShakeel Butt /* May fail (-ENOMEM) if XArray node allocation failed. */ 5000add0c77SShakeel Butt if (add_to_swap_cache(page, entry, gfp_mask & GFP_RECLAIM_MASK, &shadow)) 5014c6355b2SJohannes Weiner goto fail_unlock; 5020add0c77SShakeel Butt 5030add0c77SShakeel Butt mem_cgroup_swapin_uncharge_swap(entry); 5044c6355b2SJohannes Weiner 505aae466b0SJoonsoo Kim if (shadow) 5060995d7e5SMatthew Wilcox (Oracle) workingset_refault(page_folio(page), shadow); 507314b57fbSJohannes Weiner 5084c6355b2SJohannes Weiner /* Caller will initiate read into locked page */ 5096058eaecSJohannes Weiner lru_cache_add(page); 5104c6355b2SJohannes Weiner *new_page_allocated = true; 5114c6355b2SJohannes Weiner return page; 5124c6355b2SJohannes Weiner 5134c6355b2SJohannes Weiner fail_unlock: 5140add0c77SShakeel Butt put_swap_page(page, entry); 5154c6355b2SJohannes Weiner unlock_page(page); 5164c6355b2SJohannes Weiner put_page(page); 5174c6355b2SJohannes Weiner return NULL; 5181da177e4SLinus Torvalds } 51946017e95SHugh Dickins 5205b999aadSDmitry Safonov /* 5215b999aadSDmitry Safonov * Locate a page of swap in physical memory, reserving swap cache space 5225b999aadSDmitry Safonov * and reading the disk if it is not already cached. 5235b999aadSDmitry Safonov * A failure return means that either the page allocation failed or that 5245b999aadSDmitry Safonov * the swap entry is no longer in use. 5255b999aadSDmitry Safonov */ 5265b999aadSDmitry Safonov struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, 5275169b844SNeilBrown struct vm_area_struct *vma, 5285169b844SNeilBrown unsigned long addr, bool do_poll, 5295169b844SNeilBrown struct swap_iocb **plug) 5305b999aadSDmitry Safonov { 5315b999aadSDmitry Safonov bool page_was_allocated; 5325b999aadSDmitry Safonov struct page *retpage = __read_swap_cache_async(entry, gfp_mask, 5335b999aadSDmitry Safonov vma, addr, &page_was_allocated); 5345b999aadSDmitry Safonov 5355b999aadSDmitry Safonov if (page_was_allocated) 5365169b844SNeilBrown swap_readpage(retpage, do_poll, plug); 5375b999aadSDmitry Safonov 5385b999aadSDmitry Safonov return retpage; 5395b999aadSDmitry Safonov } 5405b999aadSDmitry Safonov 541ec560175SHuang Ying static unsigned int __swapin_nr_pages(unsigned long prev_offset, 542ec560175SHuang Ying unsigned long offset, 543ec560175SHuang Ying int hits, 544ec560175SHuang Ying int max_pages, 545ec560175SHuang Ying int prev_win) 546579f8290SShaohua Li { 547ec560175SHuang Ying unsigned int pages, last_ra; 548579f8290SShaohua Li 549579f8290SShaohua Li /* 550579f8290SShaohua Li * This heuristic has been found to work well on both sequential and 551579f8290SShaohua Li * random loads, swapping to hard disk or to SSD: please don't ask 552579f8290SShaohua Li * what the "+ 2" means, it just happens to work well, that's all. 553579f8290SShaohua Li */ 554ec560175SHuang Ying pages = hits + 2; 555579f8290SShaohua Li if (pages == 2) { 556579f8290SShaohua Li /* 557579f8290SShaohua Li * We can have no readahead hits to judge by: but must not get 558579f8290SShaohua Li * stuck here forever, so check for an adjacent offset instead 559579f8290SShaohua Li * (and don't even bother to check whether swap type is same). 560579f8290SShaohua Li */ 561579f8290SShaohua Li if (offset != prev_offset + 1 && offset != prev_offset - 1) 562579f8290SShaohua Li pages = 1; 563579f8290SShaohua Li } else { 564579f8290SShaohua Li unsigned int roundup = 4; 565579f8290SShaohua Li while (roundup < pages) 566579f8290SShaohua Li roundup <<= 1; 567579f8290SShaohua Li pages = roundup; 568579f8290SShaohua Li } 569579f8290SShaohua Li 570579f8290SShaohua Li if (pages > max_pages) 571579f8290SShaohua Li pages = max_pages; 572579f8290SShaohua Li 573579f8290SShaohua Li /* Don't shrink readahead too fast */ 574ec560175SHuang Ying last_ra = prev_win / 2; 575579f8290SShaohua Li if (pages < last_ra) 576579f8290SShaohua Li pages = last_ra; 577ec560175SHuang Ying 578ec560175SHuang Ying return pages; 579ec560175SHuang Ying } 580ec560175SHuang Ying 581ec560175SHuang Ying static unsigned long swapin_nr_pages(unsigned long offset) 582ec560175SHuang Ying { 583ec560175SHuang Ying static unsigned long prev_offset; 584ec560175SHuang Ying unsigned int hits, pages, max_pages; 585ec560175SHuang Ying static atomic_t last_readahead_pages; 586ec560175SHuang Ying 587ec560175SHuang Ying max_pages = 1 << READ_ONCE(page_cluster); 588ec560175SHuang Ying if (max_pages <= 1) 589ec560175SHuang Ying return 1; 590ec560175SHuang Ying 591ec560175SHuang Ying hits = atomic_xchg(&swapin_readahead_hits, 0); 592d6c1f098SQian Cai pages = __swapin_nr_pages(READ_ONCE(prev_offset), offset, hits, 593d6c1f098SQian Cai max_pages, 594ec560175SHuang Ying atomic_read(&last_readahead_pages)); 595ec560175SHuang Ying if (!hits) 596d6c1f098SQian Cai WRITE_ONCE(prev_offset, offset); 597579f8290SShaohua Li atomic_set(&last_readahead_pages, pages); 598579f8290SShaohua Li 599579f8290SShaohua Li return pages; 600579f8290SShaohua Li } 601579f8290SShaohua Li 60246017e95SHugh Dickins /** 603e9e9b7ecSMinchan Kim * swap_cluster_readahead - swap in pages in hope we need them soon 60446017e95SHugh Dickins * @entry: swap entry of this memory 6057682486bSRandy Dunlap * @gfp_mask: memory allocation flags 606e9e9b7ecSMinchan Kim * @vmf: fault information 60746017e95SHugh Dickins * 60846017e95SHugh Dickins * Returns the struct page for entry and addr, after queueing swapin. 60946017e95SHugh Dickins * 61046017e95SHugh Dickins * Primitive swap readahead code. We simply read an aligned block of 61146017e95SHugh Dickins * (1 << page_cluster) entries in the swap area. This method is chosen 61246017e95SHugh Dickins * because it doesn't cost us any seek time. We also make sure to queue 61346017e95SHugh Dickins * the 'original' request together with the readahead ones... 61446017e95SHugh Dickins * 61546017e95SHugh Dickins * This has been extended to use the NUMA policies from the mm triggering 61646017e95SHugh Dickins * the readahead. 61746017e95SHugh Dickins * 618c1e8d7c6SMichel Lespinasse * Caller must hold read mmap_lock if vmf->vma is not NULL. 61946017e95SHugh Dickins */ 620e9e9b7ecSMinchan Kim struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask, 621e9e9b7ecSMinchan Kim struct vm_fault *vmf) 62246017e95SHugh Dickins { 62346017e95SHugh Dickins struct page *page; 624579f8290SShaohua Li unsigned long entry_offset = swp_offset(entry); 625579f8290SShaohua Li unsigned long offset = entry_offset; 62667f96aa2SRik van Riel unsigned long start_offset, end_offset; 627579f8290SShaohua Li unsigned long mask; 628e9a6effaSHuang Ying struct swap_info_struct *si = swp_swap_info(entry); 6293fb5c298SChristian Ehrhardt struct blk_plug plug; 6305169b844SNeilBrown struct swap_iocb *splug = NULL; 631c4fa6309SHuang Ying bool do_poll = true, page_allocated; 632e9e9b7ecSMinchan Kim struct vm_area_struct *vma = vmf->vma; 633e9e9b7ecSMinchan Kim unsigned long addr = vmf->address; 63446017e95SHugh Dickins 635579f8290SShaohua Li mask = swapin_nr_pages(offset) - 1; 636579f8290SShaohua Li if (!mask) 637579f8290SShaohua Li goto skip; 638579f8290SShaohua Li 63923955622SShaohua Li do_poll = false; 64067f96aa2SRik van Riel /* Read a page_cluster sized and aligned cluster around offset. */ 64167f96aa2SRik van Riel start_offset = offset & ~mask; 64267f96aa2SRik van Riel end_offset = offset | mask; 64367f96aa2SRik van Riel if (!start_offset) /* First page is swap header. */ 64467f96aa2SRik van Riel start_offset++; 645e9a6effaSHuang Ying if (end_offset >= si->max) 646e9a6effaSHuang Ying end_offset = si->max - 1; 64767f96aa2SRik van Riel 6483fb5c298SChristian Ehrhardt blk_start_plug(&plug); 64967f96aa2SRik van Riel for (offset = start_offset; offset <= end_offset ; offset++) { 65046017e95SHugh Dickins /* Ok, do the async read-ahead now */ 651c4fa6309SHuang Ying page = __read_swap_cache_async( 652c4fa6309SHuang Ying swp_entry(swp_type(entry), offset), 653c4fa6309SHuang Ying gfp_mask, vma, addr, &page_allocated); 65446017e95SHugh Dickins if (!page) 65567f96aa2SRik van Riel continue; 656c4fa6309SHuang Ying if (page_allocated) { 6575169b844SNeilBrown swap_readpage(page, false, &splug); 658eaf649ebSMinchan Kim if (offset != entry_offset) { 659579f8290SShaohua Li SetPageReadahead(page); 660cbc65df2SHuang Ying count_vm_event(SWAP_RA); 661cbc65df2SHuang Ying } 662c4fa6309SHuang Ying } 66309cbfeafSKirill A. Shutemov put_page(page); 66446017e95SHugh Dickins } 6653fb5c298SChristian Ehrhardt blk_finish_plug(&plug); 6665169b844SNeilBrown swap_read_unplug(splug); 6673fb5c298SChristian Ehrhardt 66846017e95SHugh Dickins lru_add_drain(); /* Push any new pages onto the LRU now */ 669579f8290SShaohua Li skip: 6705169b844SNeilBrown /* The page was likely read above, so no need for plugging here */ 6715169b844SNeilBrown return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll, NULL); 67246017e95SHugh Dickins } 6734b3ef9daSHuang, Ying 6744b3ef9daSHuang, Ying int init_swap_address_space(unsigned int type, unsigned long nr_pages) 6754b3ef9daSHuang, Ying { 6764b3ef9daSHuang, Ying struct address_space *spaces, *space; 6774b3ef9daSHuang, Ying unsigned int i, nr; 6784b3ef9daSHuang, Ying 6794b3ef9daSHuang, Ying nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES); 680778e1cddSKees Cook spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL); 6814b3ef9daSHuang, Ying if (!spaces) 6824b3ef9daSHuang, Ying return -ENOMEM; 6834b3ef9daSHuang, Ying for (i = 0; i < nr; i++) { 6844b3ef9daSHuang, Ying space = spaces + i; 685a2833486SMatthew Wilcox xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ); 6864b3ef9daSHuang, Ying atomic_set(&space->i_mmap_writable, 0); 6874b3ef9daSHuang, Ying space->a_ops = &swap_aops; 6884b3ef9daSHuang, Ying /* swap cache doesn't use writeback related tags */ 6894b3ef9daSHuang, Ying mapping_set_no_writeback_tags(space); 6904b3ef9daSHuang, Ying } 6914b3ef9daSHuang, Ying nr_swapper_spaces[type] = nr; 692054f1d1fSHuang Ying swapper_spaces[type] = spaces; 6934b3ef9daSHuang, Ying 6944b3ef9daSHuang, Ying return 0; 6954b3ef9daSHuang, Ying } 6964b3ef9daSHuang, Ying 6974b3ef9daSHuang, Ying void exit_swap_address_space(unsigned int type) 6984b3ef9daSHuang, Ying { 699eea4a501SHuang Ying int i; 700eea4a501SHuang Ying struct address_space *spaces = swapper_spaces[type]; 701eea4a501SHuang Ying 702eea4a501SHuang Ying for (i = 0; i < nr_swapper_spaces[type]; i++) 703eea4a501SHuang Ying VM_WARN_ON_ONCE(!mapping_empty(&spaces[i])); 704eea4a501SHuang Ying kvfree(spaces); 7054b3ef9daSHuang, Ying nr_swapper_spaces[type] = 0; 706054f1d1fSHuang Ying swapper_spaces[type] = NULL; 7074b3ef9daSHuang, Ying } 708ec560175SHuang Ying 709ec560175SHuang Ying static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma, 710ec560175SHuang Ying unsigned long faddr, 711ec560175SHuang Ying unsigned long lpfn, 712ec560175SHuang Ying unsigned long rpfn, 713ec560175SHuang Ying unsigned long *start, 714ec560175SHuang Ying unsigned long *end) 715ec560175SHuang Ying { 716ec560175SHuang Ying *start = max3(lpfn, PFN_DOWN(vma->vm_start), 717ec560175SHuang Ying PFN_DOWN(faddr & PMD_MASK)); 718ec560175SHuang Ying *end = min3(rpfn, PFN_DOWN(vma->vm_end), 719ec560175SHuang Ying PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE)); 720ec560175SHuang Ying } 721ec560175SHuang Ying 722eaf649ebSMinchan Kim static void swap_ra_info(struct vm_fault *vmf, 723eaf649ebSMinchan Kim struct vma_swap_readahead *ra_info) 724ec560175SHuang Ying { 725ec560175SHuang Ying struct vm_area_struct *vma = vmf->vma; 726eaf649ebSMinchan Kim unsigned long ra_val; 727ec560175SHuang Ying unsigned long faddr, pfn, fpfn; 728ec560175SHuang Ying unsigned long start, end; 729eaf649ebSMinchan Kim pte_t *pte, *orig_pte; 730ec560175SHuang Ying unsigned int max_win, hits, prev_win, win, left; 731ec560175SHuang Ying #ifndef CONFIG_64BIT 732ec560175SHuang Ying pte_t *tpte; 733ec560175SHuang Ying #endif 734ec560175SHuang Ying 73561b63972SHuang Ying max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster), 73661b63972SHuang Ying SWAP_RA_ORDER_CEILING); 73761b63972SHuang Ying if (max_win == 1) { 738eaf649ebSMinchan Kim ra_info->win = 1; 739eaf649ebSMinchan Kim return; 74061b63972SHuang Ying } 74161b63972SHuang Ying 742ec560175SHuang Ying faddr = vmf->address; 743eaf649ebSMinchan Kim orig_pte = pte = pte_offset_map(vmf->pmd, faddr); 744ec560175SHuang Ying 745ec560175SHuang Ying fpfn = PFN_DOWN(faddr); 746eaf649ebSMinchan Kim ra_val = GET_SWAP_RA_VAL(vma); 747eaf649ebSMinchan Kim pfn = PFN_DOWN(SWAP_RA_ADDR(ra_val)); 748eaf649ebSMinchan Kim prev_win = SWAP_RA_WIN(ra_val); 749eaf649ebSMinchan Kim hits = SWAP_RA_HITS(ra_val); 750eaf649ebSMinchan Kim ra_info->win = win = __swapin_nr_pages(pfn, fpfn, hits, 751ec560175SHuang Ying max_win, prev_win); 752ec560175SHuang Ying atomic_long_set(&vma->swap_readahead_info, 753ec560175SHuang Ying SWAP_RA_VAL(faddr, win, 0)); 754ec560175SHuang Ying 755eaf649ebSMinchan Kim if (win == 1) { 756eaf649ebSMinchan Kim pte_unmap(orig_pte); 757eaf649ebSMinchan Kim return; 758eaf649ebSMinchan Kim } 759ec560175SHuang Ying 760ec560175SHuang Ying /* Copy the PTEs because the page table may be unmapped */ 761ec560175SHuang Ying if (fpfn == pfn + 1) 762ec560175SHuang Ying swap_ra_clamp_pfn(vma, faddr, fpfn, fpfn + win, &start, &end); 763ec560175SHuang Ying else if (pfn == fpfn + 1) 764ec560175SHuang Ying swap_ra_clamp_pfn(vma, faddr, fpfn - win + 1, fpfn + 1, 765ec560175SHuang Ying &start, &end); 766ec560175SHuang Ying else { 767ec560175SHuang Ying left = (win - 1) / 2; 768ec560175SHuang Ying swap_ra_clamp_pfn(vma, faddr, fpfn - left, fpfn + win - left, 769ec560175SHuang Ying &start, &end); 770ec560175SHuang Ying } 771eaf649ebSMinchan Kim ra_info->nr_pte = end - start; 772eaf649ebSMinchan Kim ra_info->offset = fpfn - start; 773eaf649ebSMinchan Kim pte -= ra_info->offset; 774ec560175SHuang Ying #ifdef CONFIG_64BIT 775eaf649ebSMinchan Kim ra_info->ptes = pte; 776ec560175SHuang Ying #else 777eaf649ebSMinchan Kim tpte = ra_info->ptes; 778ec560175SHuang Ying for (pfn = start; pfn != end; pfn++) 779ec560175SHuang Ying *tpte++ = *pte++; 780ec560175SHuang Ying #endif 781eaf649ebSMinchan Kim pte_unmap(orig_pte); 782ec560175SHuang Ying } 783ec560175SHuang Ying 784e9f59873SYang Shi /** 785e9f59873SYang Shi * swap_vma_readahead - swap in pages in hope we need them soon 78627ec4878SKrzysztof Kozlowski * @fentry: swap entry of this memory 787e9f59873SYang Shi * @gfp_mask: memory allocation flags 788e9f59873SYang Shi * @vmf: fault information 789e9f59873SYang Shi * 790e9f59873SYang Shi * Returns the struct page for entry and addr, after queueing swapin. 791e9f59873SYang Shi * 792cb152a1aSShijie Luo * Primitive swap readahead code. We simply read in a few pages whose 793e9f59873SYang Shi * virtual addresses are around the fault address in the same vma. 794e9f59873SYang Shi * 795c1e8d7c6SMichel Lespinasse * Caller must hold read mmap_lock if vmf->vma is not NULL. 796e9f59873SYang Shi * 797e9f59873SYang Shi */ 798f5c754d6SColin Ian King static struct page *swap_vma_readahead(swp_entry_t fentry, gfp_t gfp_mask, 799eaf649ebSMinchan Kim struct vm_fault *vmf) 800ec560175SHuang Ying { 801ec560175SHuang Ying struct blk_plug plug; 8025169b844SNeilBrown struct swap_iocb *splug = NULL; 803ec560175SHuang Ying struct vm_area_struct *vma = vmf->vma; 804ec560175SHuang Ying struct page *page; 805ec560175SHuang Ying pte_t *pte, pentry; 806ec560175SHuang Ying swp_entry_t entry; 807ec560175SHuang Ying unsigned int i; 808ec560175SHuang Ying bool page_allocated; 809e97af699SMiaohe Lin struct vma_swap_readahead ra_info = { 810e97af699SMiaohe Lin .win = 1, 811e97af699SMiaohe Lin }; 812ec560175SHuang Ying 813eaf649ebSMinchan Kim swap_ra_info(vmf, &ra_info); 814eaf649ebSMinchan Kim if (ra_info.win == 1) 815ec560175SHuang Ying goto skip; 816ec560175SHuang Ying 817ec560175SHuang Ying blk_start_plug(&plug); 818eaf649ebSMinchan Kim for (i = 0, pte = ra_info.ptes; i < ra_info.nr_pte; 819ec560175SHuang Ying i++, pte++) { 820ec560175SHuang Ying pentry = *pte; 82192bafb20SMiaohe Lin if (!is_swap_pte(pentry)) 822ec560175SHuang Ying continue; 823ec560175SHuang Ying entry = pte_to_swp_entry(pentry); 824ec560175SHuang Ying if (unlikely(non_swap_entry(entry))) 825ec560175SHuang Ying continue; 826ec560175SHuang Ying page = __read_swap_cache_async(entry, gfp_mask, vma, 827ec560175SHuang Ying vmf->address, &page_allocated); 828ec560175SHuang Ying if (!page) 829ec560175SHuang Ying continue; 830ec560175SHuang Ying if (page_allocated) { 8315169b844SNeilBrown swap_readpage(page, false, &splug); 832eaf649ebSMinchan Kim if (i != ra_info.offset) { 833ec560175SHuang Ying SetPageReadahead(page); 834ec560175SHuang Ying count_vm_event(SWAP_RA); 835ec560175SHuang Ying } 836ec560175SHuang Ying } 837ec560175SHuang Ying put_page(page); 838ec560175SHuang Ying } 839ec560175SHuang Ying blk_finish_plug(&plug); 8405169b844SNeilBrown swap_read_unplug(splug); 841ec560175SHuang Ying lru_add_drain(); 842ec560175SHuang Ying skip: 8435169b844SNeilBrown /* The page was likely read above, so no need for plugging here */ 844ec560175SHuang Ying return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address, 8455169b844SNeilBrown ra_info.win == 1, NULL); 846ec560175SHuang Ying } 847d9bfcfdcSHuang Ying 848e9e9b7ecSMinchan Kim /** 849e9e9b7ecSMinchan Kim * swapin_readahead - swap in pages in hope we need them soon 850e9e9b7ecSMinchan Kim * @entry: swap entry of this memory 851e9e9b7ecSMinchan Kim * @gfp_mask: memory allocation flags 852e9e9b7ecSMinchan Kim * @vmf: fault information 853e9e9b7ecSMinchan Kim * 854e9e9b7ecSMinchan Kim * Returns the struct page for entry and addr, after queueing swapin. 855e9e9b7ecSMinchan Kim * 856e9e9b7ecSMinchan Kim * It's a main entry function for swap readahead. By the configuration, 857e9e9b7ecSMinchan Kim * it will read ahead blocks by cluster-based(ie, physical disk based) 858e9e9b7ecSMinchan Kim * or vma-based(ie, virtual address based on faulty address) readahead. 859e9e9b7ecSMinchan Kim */ 860e9e9b7ecSMinchan Kim struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask, 861e9e9b7ecSMinchan Kim struct vm_fault *vmf) 862e9e9b7ecSMinchan Kim { 863e9e9b7ecSMinchan Kim return swap_use_vma_readahead() ? 864e9e9b7ecSMinchan Kim swap_vma_readahead(entry, gfp_mask, vmf) : 865e9e9b7ecSMinchan Kim swap_cluster_readahead(entry, gfp_mask, vmf); 866e9e9b7ecSMinchan Kim } 867e9e9b7ecSMinchan Kim 868d9bfcfdcSHuang Ying #ifdef CONFIG_SYSFS 869d9bfcfdcSHuang Ying static ssize_t vma_ra_enabled_show(struct kobject *kobj, 870d9bfcfdcSHuang Ying struct kobj_attribute *attr, char *buf) 871d9bfcfdcSHuang Ying { 872ae7a927dSJoe Perches return sysfs_emit(buf, "%s\n", 873ae7a927dSJoe Perches enable_vma_readahead ? "true" : "false"); 874d9bfcfdcSHuang Ying } 875d9bfcfdcSHuang Ying static ssize_t vma_ra_enabled_store(struct kobject *kobj, 876d9bfcfdcSHuang Ying struct kobj_attribute *attr, 877d9bfcfdcSHuang Ying const char *buf, size_t count) 878d9bfcfdcSHuang Ying { 879717aeab4SJagdish Gediya ssize_t ret; 880717aeab4SJagdish Gediya 881717aeab4SJagdish Gediya ret = kstrtobool(buf, &enable_vma_readahead); 882717aeab4SJagdish Gediya if (ret) 883717aeab4SJagdish Gediya return ret; 884d9bfcfdcSHuang Ying 885d9bfcfdcSHuang Ying return count; 886d9bfcfdcSHuang Ying } 887*6106b93eSMiaohe Lin static struct kobj_attribute vma_ra_enabled_attr = __ATTR_RW(vma_ra_enabled); 888d9bfcfdcSHuang Ying 889d9bfcfdcSHuang Ying static struct attribute *swap_attrs[] = { 890d9bfcfdcSHuang Ying &vma_ra_enabled_attr.attr, 891d9bfcfdcSHuang Ying NULL, 892d9bfcfdcSHuang Ying }; 893d9bfcfdcSHuang Ying 894e48333b6SRikard Falkeborn static const struct attribute_group swap_attr_group = { 895d9bfcfdcSHuang Ying .attrs = swap_attrs, 896d9bfcfdcSHuang Ying }; 897d9bfcfdcSHuang Ying 898d9bfcfdcSHuang Ying static int __init swap_init_sysfs(void) 899d9bfcfdcSHuang Ying { 900d9bfcfdcSHuang Ying int err; 901d9bfcfdcSHuang Ying struct kobject *swap_kobj; 902d9bfcfdcSHuang Ying 903d9bfcfdcSHuang Ying swap_kobj = kobject_create_and_add("swap", mm_kobj); 904d9bfcfdcSHuang Ying if (!swap_kobj) { 905d9bfcfdcSHuang Ying pr_err("failed to create swap kobject\n"); 906d9bfcfdcSHuang Ying return -ENOMEM; 907d9bfcfdcSHuang Ying } 908d9bfcfdcSHuang Ying err = sysfs_create_group(swap_kobj, &swap_attr_group); 909d9bfcfdcSHuang Ying if (err) { 910d9bfcfdcSHuang Ying pr_err("failed to register swap group\n"); 911d9bfcfdcSHuang Ying goto delete_obj; 912d9bfcfdcSHuang Ying } 913d9bfcfdcSHuang Ying return 0; 914d9bfcfdcSHuang Ying 915d9bfcfdcSHuang Ying delete_obj: 916d9bfcfdcSHuang Ying kobject_put(swap_kobj); 917d9bfcfdcSHuang Ying return err; 918d9bfcfdcSHuang Ying } 919d9bfcfdcSHuang Ying subsys_initcall(swap_init_sysfs); 920d9bfcfdcSHuang Ying #endif 921