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> 24243bce09SHugh Dickins #include "internal.h" 251da177e4SLinus Torvalds 261da177e4SLinus Torvalds /* 271da177e4SLinus Torvalds * swapper_space is a fiction, retained to simplify the path through 287eaceaccSJens Axboe * vmscan's shrink_page_list. 291da177e4SLinus Torvalds */ 30f5e54d6eSChristoph Hellwig static const struct address_space_operations swap_aops = { 311da177e4SLinus Torvalds .writepage = swap_writepage, 3262c230bcSMel Gorman .set_page_dirty = swap_set_page_dirty, 331c93923cSAndrew Morton #ifdef CONFIG_MIGRATION 34e965f963SChristoph Lameter .migratepage = migrate_page, 351c93923cSAndrew Morton #endif 361da177e4SLinus Torvalds }; 371da177e4SLinus Torvalds 38783cb68eSChangbin Du struct address_space *swapper_spaces[MAX_SWAPFILES] __read_mostly; 39783cb68eSChangbin Du static unsigned int nr_swapper_spaces[MAX_SWAPFILES] __read_mostly; 40f5c754d6SColin Ian King static bool enable_vma_readahead __read_mostly = true; 41ec560175SHuang Ying 42ec560175SHuang Ying #define SWAP_RA_WIN_SHIFT (PAGE_SHIFT / 2) 43ec560175SHuang Ying #define SWAP_RA_HITS_MASK ((1UL << SWAP_RA_WIN_SHIFT) - 1) 44ec560175SHuang Ying #define SWAP_RA_HITS_MAX SWAP_RA_HITS_MASK 45ec560175SHuang Ying #define SWAP_RA_WIN_MASK (~PAGE_MASK & ~SWAP_RA_HITS_MASK) 46ec560175SHuang Ying 47ec560175SHuang Ying #define SWAP_RA_HITS(v) ((v) & SWAP_RA_HITS_MASK) 48ec560175SHuang Ying #define SWAP_RA_WIN(v) (((v) & SWAP_RA_WIN_MASK) >> SWAP_RA_WIN_SHIFT) 49ec560175SHuang Ying #define SWAP_RA_ADDR(v) ((v) & PAGE_MASK) 50ec560175SHuang Ying 51ec560175SHuang Ying #define SWAP_RA_VAL(addr, win, hits) \ 52ec560175SHuang Ying (((addr) & PAGE_MASK) | \ 53ec560175SHuang Ying (((win) << SWAP_RA_WIN_SHIFT) & SWAP_RA_WIN_MASK) | \ 54ec560175SHuang Ying ((hits) & SWAP_RA_HITS_MASK)) 55ec560175SHuang Ying 56ec560175SHuang Ying /* Initial readahead hits is 4 to start up with a small window */ 57ec560175SHuang Ying #define GET_SWAP_RA_VAL(vma) \ 58ec560175SHuang Ying (atomic_long_read(&(vma)->swap_readahead_info) ? : 4) 591da177e4SLinus Torvalds 601da177e4SLinus Torvalds #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0) 6138d8b4e6SHuang Ying #define ADD_CACHE_INFO(x, nr) do { swap_cache_info.x += (nr); } while (0) 621da177e4SLinus Torvalds 631da177e4SLinus Torvalds static struct { 641da177e4SLinus Torvalds unsigned long add_total; 651da177e4SLinus Torvalds unsigned long del_total; 661da177e4SLinus Torvalds unsigned long find_success; 671da177e4SLinus Torvalds unsigned long find_total; 681da177e4SLinus Torvalds } swap_cache_info; 691da177e4SLinus Torvalds 7033806f06SShaohua Li unsigned long total_swapcache_pages(void) 7133806f06SShaohua Li { 724b3ef9daSHuang, Ying unsigned int i, j, nr; 7333806f06SShaohua Li unsigned long ret = 0; 744b3ef9daSHuang, Ying struct address_space *spaces; 75054f1d1fSHuang Ying struct swap_info_struct *si; 7633806f06SShaohua Li 774b3ef9daSHuang, Ying for (i = 0; i < MAX_SWAPFILES; i++) { 78054f1d1fSHuang Ying swp_entry_t entry = swp_entry(i, 1); 79054f1d1fSHuang Ying 80054f1d1fSHuang Ying /* Avoid get_swap_device() to warn for bad swap entry */ 81054f1d1fSHuang Ying if (!swp_swap_info(entry)) 824b3ef9daSHuang, Ying continue; 83054f1d1fSHuang Ying /* Prevent swapoff to free swapper_spaces */ 84054f1d1fSHuang Ying si = get_swap_device(entry); 85054f1d1fSHuang Ying if (!si) 86054f1d1fSHuang Ying continue; 87054f1d1fSHuang Ying nr = nr_swapper_spaces[i]; 88054f1d1fSHuang Ying spaces = swapper_spaces[i]; 894b3ef9daSHuang, Ying for (j = 0; j < nr; j++) 904b3ef9daSHuang, Ying ret += spaces[j].nrpages; 91054f1d1fSHuang Ying put_swap_device(si); 924b3ef9daSHuang, Ying } 9333806f06SShaohua Li return ret; 9433806f06SShaohua Li } 9533806f06SShaohua Li 96579f8290SShaohua Li static atomic_t swapin_readahead_hits = ATOMIC_INIT(4); 97579f8290SShaohua Li 981da177e4SLinus Torvalds void show_swap_cache_info(void) 991da177e4SLinus Torvalds { 10033806f06SShaohua Li printk("%lu pages in swap cache\n", total_swapcache_pages()); 1012c97b7fcSJohannes Weiner printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n", 1021da177e4SLinus Torvalds swap_cache_info.add_total, swap_cache_info.del_total, 103bb63be0aSHugh Dickins swap_cache_info.find_success, swap_cache_info.find_total); 104ec8acf20SShaohua Li printk("Free swap = %ldkB\n", 105ec8acf20SShaohua Li get_nr_swap_pages() << (PAGE_SHIFT - 10)); 1061da177e4SLinus Torvalds printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10)); 1071da177e4SLinus Torvalds } 1081da177e4SLinus Torvalds 109*aae466b0SJoonsoo Kim void *get_shadow_from_swap_cache(swp_entry_t entry) 110*aae466b0SJoonsoo Kim { 111*aae466b0SJoonsoo Kim struct address_space *address_space = swap_address_space(entry); 112*aae466b0SJoonsoo Kim pgoff_t idx = swp_offset(entry); 113*aae466b0SJoonsoo Kim struct page *page; 114*aae466b0SJoonsoo Kim 115*aae466b0SJoonsoo Kim page = find_get_entry(address_space, idx); 116*aae466b0SJoonsoo Kim if (xa_is_value(page)) 117*aae466b0SJoonsoo Kim return page; 118*aae466b0SJoonsoo Kim if (page) 119*aae466b0SJoonsoo Kim put_page(page); 120*aae466b0SJoonsoo Kim return NULL; 121*aae466b0SJoonsoo Kim } 122*aae466b0SJoonsoo Kim 1231da177e4SLinus Torvalds /* 1248d93b41cSMatthew Wilcox * add_to_swap_cache resembles add_to_page_cache_locked on swapper_space, 1251da177e4SLinus Torvalds * but sets SwapCache flag and private instead of mapping and index. 1261da177e4SLinus Torvalds */ 1273852f676SJoonsoo Kim int add_to_swap_cache(struct page *page, swp_entry_t entry, 1283852f676SJoonsoo Kim gfp_t gfp, void **shadowp) 1291da177e4SLinus Torvalds { 1308d93b41cSMatthew Wilcox struct address_space *address_space = swap_address_space(entry); 13138d8b4e6SHuang Ying pgoff_t idx = swp_offset(entry); 1328d93b41cSMatthew Wilcox XA_STATE_ORDER(xas, &address_space->i_pages, idx, compound_order(page)); 133cb774451SWei Yang unsigned long i, nr = hpage_nr_pages(page); 1343852f676SJoonsoo Kim void *old; 1351da177e4SLinus Torvalds 136309381feSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page), page); 137309381feSSasha Levin VM_BUG_ON_PAGE(PageSwapCache(page), page); 138309381feSSasha Levin VM_BUG_ON_PAGE(!PageSwapBacked(page), page); 13951726b12SHugh Dickins 14038d8b4e6SHuang Ying page_ref_add(page, nr); 1411da177e4SLinus Torvalds SetPageSwapCache(page); 142e286781dSNick Piggin 1438d93b41cSMatthew Wilcox do { 1443852f676SJoonsoo Kim unsigned long nr_shadows = 0; 1453852f676SJoonsoo Kim 1468d93b41cSMatthew Wilcox xas_lock_irq(&xas); 1478d93b41cSMatthew Wilcox xas_create_range(&xas); 1488d93b41cSMatthew Wilcox if (xas_error(&xas)) 1498d93b41cSMatthew Wilcox goto unlock; 15038d8b4e6SHuang Ying for (i = 0; i < nr; i++) { 1518d93b41cSMatthew Wilcox VM_BUG_ON_PAGE(xas.xa_index != idx + i, page); 1523852f676SJoonsoo Kim old = xas_load(&xas); 1533852f676SJoonsoo Kim if (xa_is_value(old)) { 1543852f676SJoonsoo Kim nr_shadows++; 1553852f676SJoonsoo Kim if (shadowp) 1563852f676SJoonsoo Kim *shadowp = old; 1573852f676SJoonsoo Kim } 15838d8b4e6SHuang Ying set_page_private(page + i, entry.val + i); 1594101196bSMatthew Wilcox (Oracle) xas_store(&xas, page); 1608d93b41cSMatthew Wilcox xas_next(&xas); 1611da177e4SLinus Torvalds } 1623852f676SJoonsoo Kim address_space->nrexceptional -= nr_shadows; 16338d8b4e6SHuang Ying address_space->nrpages += nr; 16438d8b4e6SHuang Ying __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr); 16538d8b4e6SHuang Ying ADD_CACHE_INFO(add_total, nr); 1668d93b41cSMatthew Wilcox unlock: 1678d93b41cSMatthew Wilcox xas_unlock_irq(&xas); 1688d93b41cSMatthew Wilcox } while (xas_nomem(&xas, gfp)); 1698d93b41cSMatthew Wilcox 1708d93b41cSMatthew Wilcox if (!xas_error(&xas)) 1718d93b41cSMatthew Wilcox return 0; 1728d93b41cSMatthew Wilcox 17338d8b4e6SHuang Ying ClearPageSwapCache(page); 17438d8b4e6SHuang Ying page_ref_sub(page, nr); 1758d93b41cSMatthew Wilcox return xas_error(&xas); 1761da177e4SLinus Torvalds } 1771da177e4SLinus Torvalds 1781da177e4SLinus Torvalds /* 1791da177e4SLinus Torvalds * This must be called only on pages that have 1801da177e4SLinus Torvalds * been verified to be in the swap cache. 1811da177e4SLinus Torvalds */ 1823852f676SJoonsoo Kim void __delete_from_swap_cache(struct page *page, 1833852f676SJoonsoo Kim swp_entry_t entry, void *shadow) 1841da177e4SLinus Torvalds { 1854e17ec25SMatthew Wilcox struct address_space *address_space = swap_address_space(entry); 18638d8b4e6SHuang Ying int i, nr = hpage_nr_pages(page); 1874e17ec25SMatthew Wilcox pgoff_t idx = swp_offset(entry); 1884e17ec25SMatthew Wilcox XA_STATE(xas, &address_space->i_pages, idx); 18933806f06SShaohua Li 190309381feSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page), page); 191309381feSSasha Levin VM_BUG_ON_PAGE(!PageSwapCache(page), page); 192309381feSSasha Levin VM_BUG_ON_PAGE(PageWriteback(page), page); 1931da177e4SLinus Torvalds 19438d8b4e6SHuang Ying for (i = 0; i < nr; i++) { 1953852f676SJoonsoo Kim void *entry = xas_store(&xas, shadow); 1964101196bSMatthew Wilcox (Oracle) VM_BUG_ON_PAGE(entry != page, entry); 19738d8b4e6SHuang Ying set_page_private(page + i, 0); 1984e17ec25SMatthew Wilcox xas_next(&xas); 19938d8b4e6SHuang Ying } 2001da177e4SLinus Torvalds ClearPageSwapCache(page); 2013852f676SJoonsoo Kim if (shadow) 2023852f676SJoonsoo Kim address_space->nrexceptional += nr; 20338d8b4e6SHuang Ying address_space->nrpages -= nr; 20438d8b4e6SHuang Ying __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr); 20538d8b4e6SHuang Ying ADD_CACHE_INFO(del_total, nr); 2061da177e4SLinus Torvalds } 2071da177e4SLinus Torvalds 2081da177e4SLinus Torvalds /** 2091da177e4SLinus Torvalds * add_to_swap - allocate swap space for a page 2101da177e4SLinus Torvalds * @page: page we want to move to swap 2111da177e4SLinus Torvalds * 2121da177e4SLinus Torvalds * Allocate swap space for the page and add the page to the 2131da177e4SLinus Torvalds * swap cache. Caller needs to hold the page lock. 2141da177e4SLinus Torvalds */ 2150f074658SMinchan Kim int add_to_swap(struct page *page) 2161da177e4SLinus Torvalds { 2171da177e4SLinus Torvalds swp_entry_t entry; 2181da177e4SLinus Torvalds int err; 2191da177e4SLinus Torvalds 220309381feSSasha Levin VM_BUG_ON_PAGE(!PageLocked(page), page); 221309381feSSasha Levin VM_BUG_ON_PAGE(!PageUptodate(page), page); 2221da177e4SLinus Torvalds 22338d8b4e6SHuang Ying entry = get_swap_page(page); 2241da177e4SLinus Torvalds if (!entry.val) 2250f074658SMinchan Kim return 0; 2260f074658SMinchan Kim 227bd53b714SNick Piggin /* 2288d93b41cSMatthew Wilcox * XArray node allocations from PF_MEMALLOC contexts could 229bd53b714SNick Piggin * completely exhaust the page allocator. __GFP_NOMEMALLOC 230bd53b714SNick Piggin * stops emergency reserves from being allocated. 2311da177e4SLinus Torvalds * 232bd53b714SNick Piggin * TODO: this could cause a theoretical memory reclaim 233bd53b714SNick Piggin * deadlock in the swap out path. 2341da177e4SLinus Torvalds */ 2351da177e4SLinus Torvalds /* 236854e9ed0SMinchan Kim * Add it to the swap cache. 2371da177e4SLinus Torvalds */ 238f000944dSHugh Dickins err = add_to_swap_cache(page, entry, 2393852f676SJoonsoo Kim __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN, NULL); 24038d8b4e6SHuang Ying if (err) 2412ca4532aSDaisuke Nishimura /* 2422ca4532aSDaisuke Nishimura * add_to_swap_cache() doesn't return -EEXIST, so we can safely 2432ca4532aSDaisuke Nishimura * clear SWAP_HAS_CACHE flag. 2442ca4532aSDaisuke Nishimura */ 2450f074658SMinchan Kim goto fail; 2469625456cSShaohua Li /* 2479625456cSShaohua Li * Normally the page will be dirtied in unmap because its pte should be 2489625456cSShaohua Li * dirty. A special case is MADV_FREE page. The page'e pte could have 2499625456cSShaohua Li * dirty bit cleared but the page's SwapBacked bit is still set because 2509625456cSShaohua Li * clearing the dirty bit and SwapBacked bit has no lock protected. For 2519625456cSShaohua Li * such page, unmap will not set dirty bit for it, so page reclaim will 2529625456cSShaohua Li * not write the page out. This can cause data corruption when the page 2539625456cSShaohua Li * is swap in later. Always setting the dirty bit for the page solves 2549625456cSShaohua Li * the problem. 2559625456cSShaohua Li */ 2569625456cSShaohua Li set_page_dirty(page); 2571da177e4SLinus Torvalds 25838d8b4e6SHuang Ying return 1; 25938d8b4e6SHuang Ying 26038d8b4e6SHuang Ying fail: 2610f074658SMinchan Kim put_swap_page(page, entry); 26238d8b4e6SHuang Ying return 0; 26338d8b4e6SHuang Ying } 26438d8b4e6SHuang Ying 2651da177e4SLinus Torvalds /* 2661da177e4SLinus Torvalds * This must be called only on pages that have 2671da177e4SLinus Torvalds * been verified to be in the swap cache and locked. 2681da177e4SLinus Torvalds * It will never put the page into the free list, 2691da177e4SLinus Torvalds * the caller has a reference on the page. 2701da177e4SLinus Torvalds */ 2711da177e4SLinus Torvalds void delete_from_swap_cache(struct page *page) 2721da177e4SLinus Torvalds { 2734e17ec25SMatthew Wilcox swp_entry_t entry = { .val = page_private(page) }; 2744e17ec25SMatthew Wilcox struct address_space *address_space = swap_address_space(entry); 2751da177e4SLinus Torvalds 276b93b0163SMatthew Wilcox xa_lock_irq(&address_space->i_pages); 2773852f676SJoonsoo Kim __delete_from_swap_cache(page, entry, NULL); 278b93b0163SMatthew Wilcox xa_unlock_irq(&address_space->i_pages); 2791da177e4SLinus Torvalds 28075f6d6d2SMinchan Kim put_swap_page(page, entry); 28138d8b4e6SHuang Ying page_ref_sub(page, hpage_nr_pages(page)); 2821da177e4SLinus Torvalds } 2831da177e4SLinus Torvalds 2843852f676SJoonsoo Kim void clear_shadow_from_swap_cache(int type, unsigned long begin, 2853852f676SJoonsoo Kim unsigned long end) 2863852f676SJoonsoo Kim { 2873852f676SJoonsoo Kim unsigned long curr = begin; 2883852f676SJoonsoo Kim void *old; 2893852f676SJoonsoo Kim 2903852f676SJoonsoo Kim for (;;) { 2913852f676SJoonsoo Kim unsigned long nr_shadows = 0; 2923852f676SJoonsoo Kim swp_entry_t entry = swp_entry(type, curr); 2933852f676SJoonsoo Kim struct address_space *address_space = swap_address_space(entry); 2943852f676SJoonsoo Kim XA_STATE(xas, &address_space->i_pages, curr); 2953852f676SJoonsoo Kim 2963852f676SJoonsoo Kim xa_lock_irq(&address_space->i_pages); 2973852f676SJoonsoo Kim xas_for_each(&xas, old, end) { 2983852f676SJoonsoo Kim if (!xa_is_value(old)) 2993852f676SJoonsoo Kim continue; 3003852f676SJoonsoo Kim xas_store(&xas, NULL); 3013852f676SJoonsoo Kim nr_shadows++; 3023852f676SJoonsoo Kim } 3033852f676SJoonsoo Kim address_space->nrexceptional -= nr_shadows; 3043852f676SJoonsoo Kim xa_unlock_irq(&address_space->i_pages); 3053852f676SJoonsoo Kim 3063852f676SJoonsoo Kim /* search the next swapcache until we meet end */ 3073852f676SJoonsoo Kim curr >>= SWAP_ADDRESS_SPACE_SHIFT; 3083852f676SJoonsoo Kim curr++; 3093852f676SJoonsoo Kim curr <<= SWAP_ADDRESS_SPACE_SHIFT; 3103852f676SJoonsoo Kim if (curr > end) 3113852f676SJoonsoo Kim break; 3123852f676SJoonsoo Kim } 3133852f676SJoonsoo Kim } 3143852f676SJoonsoo Kim 3151da177e4SLinus Torvalds /* 3161da177e4SLinus Torvalds * If we are the only user, then try to free up the swap cache. 3171da177e4SLinus Torvalds * 3181da177e4SLinus Torvalds * Its ok to check for PageSwapCache without the page lock 3191da177e4SLinus Torvalds * here because we are going to recheck again inside 320a2c43eedSHugh Dickins * try_to_free_swap() _with_ the lock. 3211da177e4SLinus Torvalds * - Marcelo 3221da177e4SLinus Torvalds */ 3231da177e4SLinus Torvalds static inline void free_swap_cache(struct page *page) 3241da177e4SLinus Torvalds { 325a2c43eedSHugh Dickins if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) { 326a2c43eedSHugh Dickins try_to_free_swap(page); 3271da177e4SLinus Torvalds unlock_page(page); 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds 3311da177e4SLinus Torvalds /* 3321da177e4SLinus Torvalds * Perform a free_page(), also freeing any swap cache associated with 333b8072f09SHugh Dickins * this page if it is the last user of the page. 3341da177e4SLinus Torvalds */ 3351da177e4SLinus Torvalds void free_page_and_swap_cache(struct page *page) 3361da177e4SLinus Torvalds { 3371da177e4SLinus Torvalds free_swap_cache(page); 3386fcb52a5SAaron Lu if (!is_huge_zero_page(page)) 33909cbfeafSKirill A. Shutemov put_page(page); 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds /* 3431da177e4SLinus Torvalds * Passed an array of pages, drop them all from swapcache and then release 3441da177e4SLinus Torvalds * them. They are removed from the LRU and freed if this is their last use. 3451da177e4SLinus Torvalds */ 3461da177e4SLinus Torvalds void free_pages_and_swap_cache(struct page **pages, int nr) 3471da177e4SLinus Torvalds { 3481da177e4SLinus Torvalds struct page **pagep = pages; 3491da177e4SLinus Torvalds int i; 3501da177e4SLinus Torvalds 351aabfb572SMichal Hocko lru_add_drain(); 352aabfb572SMichal Hocko for (i = 0; i < nr; i++) 3531da177e4SLinus Torvalds free_swap_cache(pagep[i]); 354c6f92f9fSMel Gorman release_pages(pagep, nr); 3551da177e4SLinus Torvalds } 3561da177e4SLinus Torvalds 357e9e9b7ecSMinchan Kim static inline bool swap_use_vma_readahead(void) 358e9e9b7ecSMinchan Kim { 359e9e9b7ecSMinchan Kim return READ_ONCE(enable_vma_readahead) && !atomic_read(&nr_rotate_swap); 360e9e9b7ecSMinchan Kim } 361e9e9b7ecSMinchan Kim 3621da177e4SLinus Torvalds /* 3631da177e4SLinus Torvalds * Lookup a swap entry in the swap cache. A found page will be returned 3641da177e4SLinus Torvalds * unlocked and with its refcount incremented - we rely on the kernel 3651da177e4SLinus Torvalds * lock getting page table operations atomic even if we drop the page 3661da177e4SLinus Torvalds * lock before returning. 3671da177e4SLinus Torvalds */ 368ec560175SHuang Ying struct page *lookup_swap_cache(swp_entry_t entry, struct vm_area_struct *vma, 369ec560175SHuang Ying unsigned long addr) 3701da177e4SLinus Torvalds { 3711da177e4SLinus Torvalds struct page *page; 372eb085574SHuang Ying struct swap_info_struct *si; 3731da177e4SLinus Torvalds 374eb085574SHuang Ying si = get_swap_device(entry); 375eb085574SHuang Ying if (!si) 376eb085574SHuang Ying return NULL; 377f6ab1f7fSHuang Ying page = find_get_page(swap_address_space(entry), swp_offset(entry)); 378eb085574SHuang Ying put_swap_device(si); 3791da177e4SLinus Torvalds 3801da177e4SLinus Torvalds INC_CACHE_INFO(find_total); 381ec560175SHuang Ying if (page) { 382eaf649ebSMinchan Kim bool vma_ra = swap_use_vma_readahead(); 383eaf649ebSMinchan Kim bool readahead; 384eaf649ebSMinchan Kim 385ec560175SHuang Ying INC_CACHE_INFO(find_success); 386eaf649ebSMinchan Kim /* 387eaf649ebSMinchan Kim * At the moment, we don't support PG_readahead for anon THP 388eaf649ebSMinchan Kim * so let's bail out rather than confusing the readahead stat. 389eaf649ebSMinchan Kim */ 390ec560175SHuang Ying if (unlikely(PageTransCompound(page))) 391ec560175SHuang Ying return page; 392eaf649ebSMinchan Kim 393ec560175SHuang Ying readahead = TestClearPageReadahead(page); 394eaf649ebSMinchan Kim if (vma && vma_ra) { 395eaf649ebSMinchan Kim unsigned long ra_val; 396eaf649ebSMinchan Kim int win, hits; 397eaf649ebSMinchan Kim 398eaf649ebSMinchan Kim ra_val = GET_SWAP_RA_VAL(vma); 399eaf649ebSMinchan Kim win = SWAP_RA_WIN(ra_val); 400eaf649ebSMinchan Kim hits = SWAP_RA_HITS(ra_val); 401ec560175SHuang Ying if (readahead) 402ec560175SHuang Ying hits = min_t(int, hits + 1, SWAP_RA_HITS_MAX); 403ec560175SHuang Ying atomic_long_set(&vma->swap_readahead_info, 404ec560175SHuang Ying SWAP_RA_VAL(addr, win, hits)); 405ec560175SHuang Ying } 406eaf649ebSMinchan Kim 407ec560175SHuang Ying if (readahead) { 408ec560175SHuang Ying count_vm_event(SWAP_RA_HIT); 409eaf649ebSMinchan Kim if (!vma || !vma_ra) 410ec560175SHuang Ying atomic_inc(&swapin_readahead_hits); 411ec560175SHuang Ying } 412ec560175SHuang Ying } 413eaf649ebSMinchan Kim 4141da177e4SLinus Torvalds return page; 4151da177e4SLinus Torvalds } 4161da177e4SLinus Torvalds 4175b999aadSDmitry Safonov struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, 4185b999aadSDmitry Safonov struct vm_area_struct *vma, unsigned long addr, 4195b999aadSDmitry Safonov bool *new_page_allocated) 4201da177e4SLinus Torvalds { 421eb085574SHuang Ying struct swap_info_struct *si; 4224c6355b2SJohannes Weiner struct page *page; 423*aae466b0SJoonsoo Kim void *shadow = NULL; 4244c6355b2SJohannes Weiner 4255b999aadSDmitry Safonov *new_page_allocated = false; 4261da177e4SLinus Torvalds 4274c6355b2SJohannes Weiner for (;;) { 4284c6355b2SJohannes Weiner int err; 4291da177e4SLinus Torvalds /* 4301da177e4SLinus Torvalds * First check the swap cache. Since this is normally 4311da177e4SLinus Torvalds * called after lookup_swap_cache() failed, re-calling 4321da177e4SLinus Torvalds * that would confuse statistics. 4331da177e4SLinus Torvalds */ 434eb085574SHuang Ying si = get_swap_device(entry); 435eb085574SHuang Ying if (!si) 4364c6355b2SJohannes Weiner return NULL; 4374c6355b2SJohannes Weiner page = find_get_page(swap_address_space(entry), 438eb085574SHuang Ying swp_offset(entry)); 439eb085574SHuang Ying put_swap_device(si); 4404c6355b2SJohannes Weiner if (page) 4414c6355b2SJohannes Weiner return page; 4421da177e4SLinus Torvalds 443ba81f838SHuang Ying /* 444ba81f838SHuang Ying * Just skip read ahead for unused swap slot. 445ba81f838SHuang Ying * During swap_off when swap_slot_cache is disabled, 446ba81f838SHuang Ying * we have to handle the race between putting 447ba81f838SHuang Ying * swap entry in swap cache and marking swap slot 448ba81f838SHuang Ying * as SWAP_HAS_CACHE. That's done in later part of code or 449ba81f838SHuang Ying * else swap_off will be aborted if we return NULL. 450ba81f838SHuang Ying */ 451ba81f838SHuang Ying if (!__swp_swapcount(entry) && swap_slot_cache_enabled) 4524c6355b2SJohannes Weiner return NULL; 453e8c26ab6STim Chen 4541da177e4SLinus Torvalds /* 4554c6355b2SJohannes Weiner * Get a new page to read into from swap. Allocate it now, 4564c6355b2SJohannes Weiner * before marking swap_map SWAP_HAS_CACHE, when -EEXIST will 4574c6355b2SJohannes Weiner * cause any racers to loop around until we add it to cache. 4581da177e4SLinus Torvalds */ 4594c6355b2SJohannes Weiner page = alloc_page_vma(gfp_mask, vma, addr); 4604c6355b2SJohannes Weiner if (!page) 4614c6355b2SJohannes Weiner return NULL; 4621da177e4SLinus Torvalds 4631da177e4SLinus Torvalds /* 464f000944dSHugh Dickins * Swap entry may have been freed since our caller observed it. 465f000944dSHugh Dickins */ 466355cfa73SKAMEZAWA Hiroyuki err = swapcache_prepare(entry); 4674c6355b2SJohannes Weiner if (!err) 468f000944dSHugh Dickins break; 469f000944dSHugh Dickins 4704c6355b2SJohannes Weiner put_page(page); 4714c6355b2SJohannes Weiner if (err != -EEXIST) 4724c6355b2SJohannes Weiner return NULL; 4731da177e4SLinus Torvalds 4744c6355b2SJohannes Weiner /* 4754c6355b2SJohannes Weiner * We might race against __delete_from_swap_cache(), and 4764c6355b2SJohannes Weiner * stumble across a swap_map entry whose SWAP_HAS_CACHE 4774c6355b2SJohannes Weiner * has not yet been cleared. Or race against another 4784c6355b2SJohannes Weiner * __read_swap_cache_async(), which has set SWAP_HAS_CACHE 4794c6355b2SJohannes Weiner * in swap_map, but not yet added its page to swap cache. 4804c6355b2SJohannes Weiner */ 4814c6355b2SJohannes Weiner cond_resched(); 4824c6355b2SJohannes Weiner } 4834c6355b2SJohannes Weiner 4844c6355b2SJohannes Weiner /* 4854c6355b2SJohannes Weiner * The swap entry is ours to swap in. Prepare the new page. 4864c6355b2SJohannes Weiner */ 4874c6355b2SJohannes Weiner 4884c6355b2SJohannes Weiner __SetPageLocked(page); 4894c6355b2SJohannes Weiner __SetPageSwapBacked(page); 4904c6355b2SJohannes Weiner 4914c6355b2SJohannes Weiner /* May fail (-ENOMEM) if XArray node allocation failed. */ 492*aae466b0SJoonsoo Kim if (add_to_swap_cache(page, entry, gfp_mask & GFP_RECLAIM_MASK, &shadow)) { 4934c6355b2SJohannes Weiner put_swap_page(page, entry); 4944c6355b2SJohannes Weiner goto fail_unlock; 4954c6355b2SJohannes Weiner } 4964c6355b2SJohannes Weiner 497d9eb1ea2SJohannes Weiner if (mem_cgroup_charge(page, NULL, gfp_mask)) { 4984c6355b2SJohannes Weiner delete_from_swap_cache(page); 4994c6355b2SJohannes Weiner goto fail_unlock; 5004c6355b2SJohannes Weiner } 5014c6355b2SJohannes Weiner 502*aae466b0SJoonsoo Kim if (shadow) 503*aae466b0SJoonsoo Kim workingset_refault(page, shadow); 504314b57fbSJohannes Weiner 5054c6355b2SJohannes Weiner /* Caller will initiate read into locked page */ 5064c6355b2SJohannes Weiner SetPageWorkingset(page); 5076058eaecSJohannes Weiner lru_cache_add(page); 5084c6355b2SJohannes Weiner *new_page_allocated = true; 5094c6355b2SJohannes Weiner return page; 5104c6355b2SJohannes Weiner 5114c6355b2SJohannes Weiner fail_unlock: 5124c6355b2SJohannes Weiner unlock_page(page); 5134c6355b2SJohannes Weiner put_page(page); 5144c6355b2SJohannes Weiner return NULL; 5151da177e4SLinus Torvalds } 51646017e95SHugh Dickins 5175b999aadSDmitry Safonov /* 5185b999aadSDmitry Safonov * Locate a page of swap in physical memory, reserving swap cache space 5195b999aadSDmitry Safonov * and reading the disk if it is not already cached. 5205b999aadSDmitry Safonov * A failure return means that either the page allocation failed or that 5215b999aadSDmitry Safonov * the swap entry is no longer in use. 5225b999aadSDmitry Safonov */ 5235b999aadSDmitry Safonov struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask, 52423955622SShaohua Li struct vm_area_struct *vma, unsigned long addr, bool do_poll) 5255b999aadSDmitry Safonov { 5265b999aadSDmitry Safonov bool page_was_allocated; 5275b999aadSDmitry Safonov struct page *retpage = __read_swap_cache_async(entry, gfp_mask, 5285b999aadSDmitry Safonov vma, addr, &page_was_allocated); 5295b999aadSDmitry Safonov 5305b999aadSDmitry Safonov if (page_was_allocated) 53123955622SShaohua Li swap_readpage(retpage, do_poll); 5325b999aadSDmitry Safonov 5335b999aadSDmitry Safonov return retpage; 5345b999aadSDmitry Safonov } 5355b999aadSDmitry Safonov 536ec560175SHuang Ying static unsigned int __swapin_nr_pages(unsigned long prev_offset, 537ec560175SHuang Ying unsigned long offset, 538ec560175SHuang Ying int hits, 539ec560175SHuang Ying int max_pages, 540ec560175SHuang Ying int prev_win) 541579f8290SShaohua Li { 542ec560175SHuang Ying unsigned int pages, last_ra; 543579f8290SShaohua Li 544579f8290SShaohua Li /* 545579f8290SShaohua Li * This heuristic has been found to work well on both sequential and 546579f8290SShaohua Li * random loads, swapping to hard disk or to SSD: please don't ask 547579f8290SShaohua Li * what the "+ 2" means, it just happens to work well, that's all. 548579f8290SShaohua Li */ 549ec560175SHuang Ying pages = hits + 2; 550579f8290SShaohua Li if (pages == 2) { 551579f8290SShaohua Li /* 552579f8290SShaohua Li * We can have no readahead hits to judge by: but must not get 553579f8290SShaohua Li * stuck here forever, so check for an adjacent offset instead 554579f8290SShaohua Li * (and don't even bother to check whether swap type is same). 555579f8290SShaohua Li */ 556579f8290SShaohua Li if (offset != prev_offset + 1 && offset != prev_offset - 1) 557579f8290SShaohua Li pages = 1; 558579f8290SShaohua Li } else { 559579f8290SShaohua Li unsigned int roundup = 4; 560579f8290SShaohua Li while (roundup < pages) 561579f8290SShaohua Li roundup <<= 1; 562579f8290SShaohua Li pages = roundup; 563579f8290SShaohua Li } 564579f8290SShaohua Li 565579f8290SShaohua Li if (pages > max_pages) 566579f8290SShaohua Li pages = max_pages; 567579f8290SShaohua Li 568579f8290SShaohua Li /* Don't shrink readahead too fast */ 569ec560175SHuang Ying last_ra = prev_win / 2; 570579f8290SShaohua Li if (pages < last_ra) 571579f8290SShaohua Li pages = last_ra; 572ec560175SHuang Ying 573ec560175SHuang Ying return pages; 574ec560175SHuang Ying } 575ec560175SHuang Ying 576ec560175SHuang Ying static unsigned long swapin_nr_pages(unsigned long offset) 577ec560175SHuang Ying { 578ec560175SHuang Ying static unsigned long prev_offset; 579ec560175SHuang Ying unsigned int hits, pages, max_pages; 580ec560175SHuang Ying static atomic_t last_readahead_pages; 581ec560175SHuang Ying 582ec560175SHuang Ying max_pages = 1 << READ_ONCE(page_cluster); 583ec560175SHuang Ying if (max_pages <= 1) 584ec560175SHuang Ying return 1; 585ec560175SHuang Ying 586ec560175SHuang Ying hits = atomic_xchg(&swapin_readahead_hits, 0); 587d6c1f098SQian Cai pages = __swapin_nr_pages(READ_ONCE(prev_offset), offset, hits, 588d6c1f098SQian Cai max_pages, 589ec560175SHuang Ying atomic_read(&last_readahead_pages)); 590ec560175SHuang Ying if (!hits) 591d6c1f098SQian Cai WRITE_ONCE(prev_offset, offset); 592579f8290SShaohua Li atomic_set(&last_readahead_pages, pages); 593579f8290SShaohua Li 594579f8290SShaohua Li return pages; 595579f8290SShaohua Li } 596579f8290SShaohua Li 59746017e95SHugh Dickins /** 598e9e9b7ecSMinchan Kim * swap_cluster_readahead - swap in pages in hope we need them soon 59946017e95SHugh Dickins * @entry: swap entry of this memory 6007682486bSRandy Dunlap * @gfp_mask: memory allocation flags 601e9e9b7ecSMinchan Kim * @vmf: fault information 60246017e95SHugh Dickins * 60346017e95SHugh Dickins * Returns the struct page for entry and addr, after queueing swapin. 60446017e95SHugh Dickins * 60546017e95SHugh Dickins * Primitive swap readahead code. We simply read an aligned block of 60646017e95SHugh Dickins * (1 << page_cluster) entries in the swap area. This method is chosen 60746017e95SHugh Dickins * because it doesn't cost us any seek time. We also make sure to queue 60846017e95SHugh Dickins * the 'original' request together with the readahead ones... 60946017e95SHugh Dickins * 61046017e95SHugh Dickins * This has been extended to use the NUMA policies from the mm triggering 61146017e95SHugh Dickins * the readahead. 61246017e95SHugh Dickins * 613c1e8d7c6SMichel Lespinasse * Caller must hold read mmap_lock if vmf->vma is not NULL. 61446017e95SHugh Dickins */ 615e9e9b7ecSMinchan Kim struct page *swap_cluster_readahead(swp_entry_t entry, gfp_t gfp_mask, 616e9e9b7ecSMinchan Kim struct vm_fault *vmf) 61746017e95SHugh Dickins { 61846017e95SHugh Dickins struct page *page; 619579f8290SShaohua Li unsigned long entry_offset = swp_offset(entry); 620579f8290SShaohua Li unsigned long offset = entry_offset; 62167f96aa2SRik van Riel unsigned long start_offset, end_offset; 622579f8290SShaohua Li unsigned long mask; 623e9a6effaSHuang Ying struct swap_info_struct *si = swp_swap_info(entry); 6243fb5c298SChristian Ehrhardt struct blk_plug plug; 625c4fa6309SHuang Ying bool do_poll = true, page_allocated; 626e9e9b7ecSMinchan Kim struct vm_area_struct *vma = vmf->vma; 627e9e9b7ecSMinchan Kim unsigned long addr = vmf->address; 62846017e95SHugh Dickins 629579f8290SShaohua Li mask = swapin_nr_pages(offset) - 1; 630579f8290SShaohua Li if (!mask) 631579f8290SShaohua Li goto skip; 632579f8290SShaohua Li 6338fd2e0b5SYang Shi /* Test swap type to make sure the dereference is safe */ 6348fd2e0b5SYang Shi if (likely(si->flags & (SWP_BLKDEV | SWP_FS))) { 6358fd2e0b5SYang Shi struct inode *inode = si->swap_file->f_mapping->host; 6368fd2e0b5SYang Shi if (inode_read_congested(inode)) 6378fd2e0b5SYang Shi goto skip; 6388fd2e0b5SYang Shi } 6398fd2e0b5SYang Shi 64023955622SShaohua Li do_poll = false; 64167f96aa2SRik van Riel /* Read a page_cluster sized and aligned cluster around offset. */ 64267f96aa2SRik van Riel start_offset = offset & ~mask; 64367f96aa2SRik van Riel end_offset = offset | mask; 64467f96aa2SRik van Riel if (!start_offset) /* First page is swap header. */ 64567f96aa2SRik van Riel start_offset++; 646e9a6effaSHuang Ying if (end_offset >= si->max) 647e9a6effaSHuang Ying end_offset = si->max - 1; 64867f96aa2SRik van Riel 6493fb5c298SChristian Ehrhardt blk_start_plug(&plug); 65067f96aa2SRik van Riel for (offset = start_offset; offset <= end_offset ; offset++) { 65146017e95SHugh Dickins /* Ok, do the async read-ahead now */ 652c4fa6309SHuang Ying page = __read_swap_cache_async( 653c4fa6309SHuang Ying swp_entry(swp_type(entry), offset), 654c4fa6309SHuang Ying gfp_mask, vma, addr, &page_allocated); 65546017e95SHugh Dickins if (!page) 65667f96aa2SRik van Riel continue; 657c4fa6309SHuang Ying if (page_allocated) { 658c4fa6309SHuang Ying swap_readpage(page, false); 659eaf649ebSMinchan Kim if (offset != entry_offset) { 660579f8290SShaohua Li SetPageReadahead(page); 661cbc65df2SHuang Ying count_vm_event(SWAP_RA); 662cbc65df2SHuang Ying } 663c4fa6309SHuang Ying } 66409cbfeafSKirill A. Shutemov put_page(page); 66546017e95SHugh Dickins } 6663fb5c298SChristian Ehrhardt blk_finish_plug(&plug); 6673fb5c298SChristian Ehrhardt 66846017e95SHugh Dickins lru_add_drain(); /* Push any new pages onto the LRU now */ 669579f8290SShaohua Li skip: 67023955622SShaohua Li return read_swap_cache_async(entry, gfp_mask, vma, addr, do_poll); 67146017e95SHugh Dickins } 6724b3ef9daSHuang, Ying 6734b3ef9daSHuang, Ying int init_swap_address_space(unsigned int type, unsigned long nr_pages) 6744b3ef9daSHuang, Ying { 6754b3ef9daSHuang, Ying struct address_space *spaces, *space; 6764b3ef9daSHuang, Ying unsigned int i, nr; 6774b3ef9daSHuang, Ying 6784b3ef9daSHuang, Ying nr = DIV_ROUND_UP(nr_pages, SWAP_ADDRESS_SPACE_PAGES); 679778e1cddSKees Cook spaces = kvcalloc(nr, sizeof(struct address_space), GFP_KERNEL); 6804b3ef9daSHuang, Ying if (!spaces) 6814b3ef9daSHuang, Ying return -ENOMEM; 6824b3ef9daSHuang, Ying for (i = 0; i < nr; i++) { 6834b3ef9daSHuang, Ying space = spaces + i; 684a2833486SMatthew Wilcox xa_init_flags(&space->i_pages, XA_FLAGS_LOCK_IRQ); 6854b3ef9daSHuang, Ying atomic_set(&space->i_mmap_writable, 0); 6864b3ef9daSHuang, Ying space->a_ops = &swap_aops; 6874b3ef9daSHuang, Ying /* swap cache doesn't use writeback related tags */ 6884b3ef9daSHuang, Ying mapping_set_no_writeback_tags(space); 6894b3ef9daSHuang, Ying } 6904b3ef9daSHuang, Ying nr_swapper_spaces[type] = nr; 691054f1d1fSHuang Ying swapper_spaces[type] = spaces; 6924b3ef9daSHuang, Ying 6934b3ef9daSHuang, Ying return 0; 6944b3ef9daSHuang, Ying } 6954b3ef9daSHuang, Ying 6964b3ef9daSHuang, Ying void exit_swap_address_space(unsigned int type) 6974b3ef9daSHuang, Ying { 698054f1d1fSHuang Ying kvfree(swapper_spaces[type]); 6994b3ef9daSHuang, Ying nr_swapper_spaces[type] = 0; 700054f1d1fSHuang Ying swapper_spaces[type] = NULL; 7014b3ef9daSHuang, Ying } 702ec560175SHuang Ying 703ec560175SHuang Ying static inline void swap_ra_clamp_pfn(struct vm_area_struct *vma, 704ec560175SHuang Ying unsigned long faddr, 705ec560175SHuang Ying unsigned long lpfn, 706ec560175SHuang Ying unsigned long rpfn, 707ec560175SHuang Ying unsigned long *start, 708ec560175SHuang Ying unsigned long *end) 709ec560175SHuang Ying { 710ec560175SHuang Ying *start = max3(lpfn, PFN_DOWN(vma->vm_start), 711ec560175SHuang Ying PFN_DOWN(faddr & PMD_MASK)); 712ec560175SHuang Ying *end = min3(rpfn, PFN_DOWN(vma->vm_end), 713ec560175SHuang Ying PFN_DOWN((faddr & PMD_MASK) + PMD_SIZE)); 714ec560175SHuang Ying } 715ec560175SHuang Ying 716eaf649ebSMinchan Kim static void swap_ra_info(struct vm_fault *vmf, 717eaf649ebSMinchan Kim struct vma_swap_readahead *ra_info) 718ec560175SHuang Ying { 719ec560175SHuang Ying struct vm_area_struct *vma = vmf->vma; 720eaf649ebSMinchan Kim unsigned long ra_val; 721ec560175SHuang Ying swp_entry_t entry; 722ec560175SHuang Ying unsigned long faddr, pfn, fpfn; 723ec560175SHuang Ying unsigned long start, end; 724eaf649ebSMinchan Kim pte_t *pte, *orig_pte; 725ec560175SHuang Ying unsigned int max_win, hits, prev_win, win, left; 726ec560175SHuang Ying #ifndef CONFIG_64BIT 727ec560175SHuang Ying pte_t *tpte; 728ec560175SHuang Ying #endif 729ec560175SHuang Ying 73061b63972SHuang Ying max_win = 1 << min_t(unsigned int, READ_ONCE(page_cluster), 73161b63972SHuang Ying SWAP_RA_ORDER_CEILING); 73261b63972SHuang Ying if (max_win == 1) { 733eaf649ebSMinchan Kim ra_info->win = 1; 734eaf649ebSMinchan Kim return; 73561b63972SHuang Ying } 73661b63972SHuang Ying 737ec560175SHuang Ying faddr = vmf->address; 738eaf649ebSMinchan Kim orig_pte = pte = pte_offset_map(vmf->pmd, faddr); 739eaf649ebSMinchan Kim entry = pte_to_swp_entry(*pte); 740eaf649ebSMinchan Kim if ((unlikely(non_swap_entry(entry)))) { 741eaf649ebSMinchan Kim pte_unmap(orig_pte); 742eaf649ebSMinchan Kim return; 743eaf649ebSMinchan Kim } 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 * 792e9f59873SYang Shi * Primitive swap readahead code. We simply read in a few pages whoes 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; 802ec560175SHuang Ying struct vm_area_struct *vma = vmf->vma; 803ec560175SHuang Ying struct page *page; 804ec560175SHuang Ying pte_t *pte, pentry; 805ec560175SHuang Ying swp_entry_t entry; 806ec560175SHuang Ying unsigned int i; 807ec560175SHuang Ying bool page_allocated; 808eaf649ebSMinchan Kim struct vma_swap_readahead ra_info = {0,}; 809ec560175SHuang Ying 810eaf649ebSMinchan Kim swap_ra_info(vmf, &ra_info); 811eaf649ebSMinchan Kim if (ra_info.win == 1) 812ec560175SHuang Ying goto skip; 813ec560175SHuang Ying 814ec560175SHuang Ying blk_start_plug(&plug); 815eaf649ebSMinchan Kim for (i = 0, pte = ra_info.ptes; i < ra_info.nr_pte; 816ec560175SHuang Ying i++, pte++) { 817ec560175SHuang Ying pentry = *pte; 818ec560175SHuang Ying if (pte_none(pentry)) 819ec560175SHuang Ying continue; 820ec560175SHuang Ying if (pte_present(pentry)) 821ec560175SHuang Ying continue; 822ec560175SHuang Ying entry = pte_to_swp_entry(pentry); 823ec560175SHuang Ying if (unlikely(non_swap_entry(entry))) 824ec560175SHuang Ying continue; 825ec560175SHuang Ying page = __read_swap_cache_async(entry, gfp_mask, vma, 826ec560175SHuang Ying vmf->address, &page_allocated); 827ec560175SHuang Ying if (!page) 828ec560175SHuang Ying continue; 829ec560175SHuang Ying if (page_allocated) { 830ec560175SHuang Ying swap_readpage(page, false); 831eaf649ebSMinchan Kim if (i != ra_info.offset) { 832ec560175SHuang Ying SetPageReadahead(page); 833ec560175SHuang Ying count_vm_event(SWAP_RA); 834ec560175SHuang Ying } 835ec560175SHuang Ying } 836ec560175SHuang Ying put_page(page); 837ec560175SHuang Ying } 838ec560175SHuang Ying blk_finish_plug(&plug); 839ec560175SHuang Ying lru_add_drain(); 840ec560175SHuang Ying skip: 841ec560175SHuang Ying return read_swap_cache_async(fentry, gfp_mask, vma, vmf->address, 842eaf649ebSMinchan Kim ra_info.win == 1); 843ec560175SHuang Ying } 844d9bfcfdcSHuang Ying 845e9e9b7ecSMinchan Kim /** 846e9e9b7ecSMinchan Kim * swapin_readahead - swap in pages in hope we need them soon 847e9e9b7ecSMinchan Kim * @entry: swap entry of this memory 848e9e9b7ecSMinchan Kim * @gfp_mask: memory allocation flags 849e9e9b7ecSMinchan Kim * @vmf: fault information 850e9e9b7ecSMinchan Kim * 851e9e9b7ecSMinchan Kim * Returns the struct page for entry and addr, after queueing swapin. 852e9e9b7ecSMinchan Kim * 853e9e9b7ecSMinchan Kim * It's a main entry function for swap readahead. By the configuration, 854e9e9b7ecSMinchan Kim * it will read ahead blocks by cluster-based(ie, physical disk based) 855e9e9b7ecSMinchan Kim * or vma-based(ie, virtual address based on faulty address) readahead. 856e9e9b7ecSMinchan Kim */ 857e9e9b7ecSMinchan Kim struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask, 858e9e9b7ecSMinchan Kim struct vm_fault *vmf) 859e9e9b7ecSMinchan Kim { 860e9e9b7ecSMinchan Kim return swap_use_vma_readahead() ? 861e9e9b7ecSMinchan Kim swap_vma_readahead(entry, gfp_mask, vmf) : 862e9e9b7ecSMinchan Kim swap_cluster_readahead(entry, gfp_mask, vmf); 863e9e9b7ecSMinchan Kim } 864e9e9b7ecSMinchan Kim 865d9bfcfdcSHuang Ying #ifdef CONFIG_SYSFS 866d9bfcfdcSHuang Ying static ssize_t vma_ra_enabled_show(struct kobject *kobj, 867d9bfcfdcSHuang Ying struct kobj_attribute *attr, char *buf) 868d9bfcfdcSHuang Ying { 869e9e9b7ecSMinchan Kim return sprintf(buf, "%s\n", enable_vma_readahead ? "true" : "false"); 870d9bfcfdcSHuang Ying } 871d9bfcfdcSHuang Ying static ssize_t vma_ra_enabled_store(struct kobject *kobj, 872d9bfcfdcSHuang Ying struct kobj_attribute *attr, 873d9bfcfdcSHuang Ying const char *buf, size_t count) 874d9bfcfdcSHuang Ying { 875d9bfcfdcSHuang Ying if (!strncmp(buf, "true", 4) || !strncmp(buf, "1", 1)) 876e9e9b7ecSMinchan Kim enable_vma_readahead = true; 877d9bfcfdcSHuang Ying else if (!strncmp(buf, "false", 5) || !strncmp(buf, "0", 1)) 878e9e9b7ecSMinchan Kim enable_vma_readahead = false; 879d9bfcfdcSHuang Ying else 880d9bfcfdcSHuang Ying return -EINVAL; 881d9bfcfdcSHuang Ying 882d9bfcfdcSHuang Ying return count; 883d9bfcfdcSHuang Ying } 884d9bfcfdcSHuang Ying static struct kobj_attribute vma_ra_enabled_attr = 885d9bfcfdcSHuang Ying __ATTR(vma_ra_enabled, 0644, vma_ra_enabled_show, 886d9bfcfdcSHuang Ying vma_ra_enabled_store); 887d9bfcfdcSHuang Ying 888d9bfcfdcSHuang Ying static struct attribute *swap_attrs[] = { 889d9bfcfdcSHuang Ying &vma_ra_enabled_attr.attr, 890d9bfcfdcSHuang Ying NULL, 891d9bfcfdcSHuang Ying }; 892d9bfcfdcSHuang Ying 893d9bfcfdcSHuang Ying static struct attribute_group swap_attr_group = { 894d9bfcfdcSHuang Ying .attrs = swap_attrs, 895d9bfcfdcSHuang Ying }; 896d9bfcfdcSHuang Ying 897d9bfcfdcSHuang Ying static int __init swap_init_sysfs(void) 898d9bfcfdcSHuang Ying { 899d9bfcfdcSHuang Ying int err; 900d9bfcfdcSHuang Ying struct kobject *swap_kobj; 901d9bfcfdcSHuang Ying 902d9bfcfdcSHuang Ying swap_kobj = kobject_create_and_add("swap", mm_kobj); 903d9bfcfdcSHuang Ying if (!swap_kobj) { 904d9bfcfdcSHuang Ying pr_err("failed to create swap kobject\n"); 905d9bfcfdcSHuang Ying return -ENOMEM; 906d9bfcfdcSHuang Ying } 907d9bfcfdcSHuang Ying err = sysfs_create_group(swap_kobj, &swap_attr_group); 908d9bfcfdcSHuang Ying if (err) { 909d9bfcfdcSHuang Ying pr_err("failed to register swap group\n"); 910d9bfcfdcSHuang Ying goto delete_obj; 911d9bfcfdcSHuang Ying } 912d9bfcfdcSHuang Ying return 0; 913d9bfcfdcSHuang Ying 914d9bfcfdcSHuang Ying delete_obj: 915d9bfcfdcSHuang Ying kobject_put(swap_kobj); 916d9bfcfdcSHuang Ying return err; 917d9bfcfdcSHuang Ying } 918d9bfcfdcSHuang Ying subsys_initcall(swap_init_sysfs); 919d9bfcfdcSHuang Ying #endif 920