blob: 5ac584665c8241fdf3f7f260ae13a37ee58a4c9b [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/swap_state.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
6 *
7 * Rewritten to use page cache, (C) 1998 Stephen Tweedie
8 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07009#include <linux/mm.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090010#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/kernel_stat.h>
12#include <linux/swap.h>
Hugh Dickins46017e92008-02-04 22:28:41 -080013#include <linux/swapops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/init.h>
15#include <linux/pagemap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/backing-dev.h>
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -070017#include <linux/blkdev.h>
Hugh Dickinsc484d412006-01-06 00:10:55 -080018#include <linux/pagevec.h>
Christoph Lameterb20a3502006-03-22 00:09:12 -080019#include <linux/migrate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
21#include <asm/pgtable.h>
22
23/*
24 * swapper_space is a fiction, retained to simplify the path through
Jens Axboe7eaceac2011-03-10 08:52:07 +010025 * vmscan's shrink_page_list.
Linus Torvalds1da177e2005-04-16 15:20:36 -070026 */
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070027static const struct address_space_operations swap_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070028 .writepage = swap_writepage,
Mel Gorman62c230b2012-07-31 16:44:55 -070029 .set_page_dirty = swap_set_page_dirty,
Andrew Morton1c939232014-10-09 15:27:59 -070030#ifdef CONFIG_MIGRATION
Christoph Lametere965f962006-02-01 03:05:41 -080031 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -070032#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070033};
34
Shaohua Li33806f02013-02-22 16:34:37 -080035struct address_space swapper_spaces[MAX_SWAPFILES] = {
36 [0 ... MAX_SWAPFILES - 1] = {
37 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
David Herrmann4bb5f5d2014-08-08 14:25:25 -070038 .i_mmap_writable = ATOMIC_INIT(0),
Shaohua Li33806f02013-02-22 16:34:37 -080039 .a_ops = &swap_aops,
Huang Ying371a0962016-10-07 16:59:30 -070040 /* swap cache doesn't use writeback related tags */
41 .flags = 1 << AS_NO_WRITEBACK_TAGS,
Shaohua Li33806f02013-02-22 16:34:37 -080042 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070043};
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
46
47static struct {
48 unsigned long add_total;
49 unsigned long del_total;
50 unsigned long find_success;
51 unsigned long find_total;
Linus Torvalds1da177e2005-04-16 15:20:36 -070052} swap_cache_info;
53
Shaohua Li33806f02013-02-22 16:34:37 -080054unsigned long total_swapcache_pages(void)
55{
56 int i;
57 unsigned long ret = 0;
58
59 for (i = 0; i < MAX_SWAPFILES; i++)
60 ret += swapper_spaces[i].nrpages;
61 return ret;
62}
63
Shaohua Li579f8292014-02-06 12:04:21 -080064static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066void show_swap_cache_info(void)
67{
Shaohua Li33806f02013-02-22 16:34:37 -080068 printk("%lu pages in swap cache\n", total_swapcache_pages());
Johannes Weiner2c97b7f2008-07-25 19:46:01 -070069 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 swap_cache_info.add_total, swap_cache_info.del_total,
Hugh Dickinsbb63be02008-02-04 22:28:49 -080071 swap_cache_info.find_success, swap_cache_info.find_total);
Shaohua Liec8acf22013-02-22 16:34:38 -080072 printk("Free swap = %ldkB\n",
73 get_nr_swap_pages() << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
75}
76
77/*
Daisuke Nishimura31a56392009-09-21 17:02:50 -070078 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * but sets SwapCache flag and private instead of mapping and index.
80 */
Seth Jennings2f772e62013-04-29 15:08:34 -070081int __add_to_swap_cache(struct page *page, swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
83 int error;
Shaohua Li33806f02013-02-22 16:34:37 -080084 struct address_space *address_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Sasha Levin309381fea2014-01-23 15:52:54 -080086 VM_BUG_ON_PAGE(!PageLocked(page), page);
87 VM_BUG_ON_PAGE(PageSwapCache(page), page);
88 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins51726b12009-01-06 14:39:25 -080089
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030090 get_page(page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -070091 SetPageSwapCache(page);
92 set_page_private(page, entry.val);
93
Shaohua Li33806f02013-02-22 16:34:37 -080094 address_space = swap_address_space(entry);
95 spin_lock_irq(&address_space->tree_lock);
96 error = radix_tree_insert(&address_space->page_tree,
Huang Yingf6ab1f72016-10-07 17:00:21 -070097 swp_offset(entry), page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -070098 if (likely(!error)) {
Shaohua Li33806f02013-02-22 16:34:37 -080099 address_space->nrpages++;
Mel Gorman11fb9982016-07-28 15:46:20 -0700100 __inc_node_page_state(page, NR_FILE_PAGES);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700101 INC_CACHE_INFO(add_total);
102 }
Shaohua Li33806f02013-02-22 16:34:37 -0800103 spin_unlock_irq(&address_space->tree_lock);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700104
105 if (unlikely(error)) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700106 /*
107 * Only the context which have set SWAP_HAS_CACHE flag
108 * would call add_to_swap_cache().
109 * So add_to_swap_cache() doesn't returns -EEXIST.
110 */
111 VM_BUG_ON(error == -EEXIST);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700112 set_page_private(page, 0UL);
113 ClearPageSwapCache(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300114 put_page(page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700115 }
116
117 return error;
118}
119
120
121int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
122{
123 int error;
124
Jan Kara5e4c0d972013-09-11 14:26:05 -0700125 error = radix_tree_maybe_preload(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 if (!error) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700127 error = __add_to_swap_cache(page, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 radix_tree_preload_end();
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800129 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 return error;
131}
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133/*
134 * This must be called only on pages that have
135 * been verified to be in the swap cache.
136 */
137void __delete_from_swap_cache(struct page *page)
138{
Shaohua Li33806f02013-02-22 16:34:37 -0800139 swp_entry_t entry;
140 struct address_space *address_space;
141
Sasha Levin309381fea2014-01-23 15:52:54 -0800142 VM_BUG_ON_PAGE(!PageLocked(page), page);
143 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
144 VM_BUG_ON_PAGE(PageWriteback(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Shaohua Li33806f02013-02-22 16:34:37 -0800146 entry.val = page_private(page);
147 address_space = swap_address_space(entry);
Huang Yingf6ab1f72016-10-07 17:00:21 -0700148 radix_tree_delete(&address_space->page_tree, swp_offset(entry));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700149 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 ClearPageSwapCache(page);
Shaohua Li33806f02013-02-22 16:34:37 -0800151 address_space->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700152 __dec_node_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 INC_CACHE_INFO(del_total);
154}
155
156/**
157 * add_to_swap - allocate swap space for a page
158 * @page: page we want to move to swap
159 *
160 * Allocate swap space for the page and add the page to the
161 * swap cache. Caller needs to hold the page lock.
162 */
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700163int add_to_swap(struct page *page, struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
165 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 int err;
167
Sasha Levin309381fea2014-01-23 15:52:54 -0800168 VM_BUG_ON_PAGE(!PageLocked(page), page);
169 VM_BUG_ON_PAGE(!PageUptodate(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700171 entry = get_swap_page();
172 if (!entry.val)
173 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Vladimir Davydov37e84352016-01-20 15:02:56 -0800175 if (mem_cgroup_try_charge_swap(page, entry)) {
176 swapcache_free(entry);
177 return 0;
178 }
179
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800180 if (unlikely(PageTransHuge(page)))
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700181 if (unlikely(split_huge_page_to_list(page, list))) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700182 swapcache_free(entry);
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800183 return 0;
184 }
185
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700186 /*
187 * Radix-tree node allocations from PF_MEMALLOC contexts could
188 * completely exhaust the page allocator. __GFP_NOMEMALLOC
189 * stops emergency reserves from being allocated.
190 *
191 * TODO: this could cause a theoretical memory reclaim
192 * deadlock in the swap out path.
193 */
194 /*
Minchan Kim854e9ed2016-01-15 16:54:53 -0800195 * Add it to the swap cache.
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700196 */
197 err = add_to_swap_cache(page, entry,
198 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
Minchan Kim854e9ed2016-01-15 16:54:53 -0800200 if (!err) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700201 return 1;
202 } else { /* -ENOMEM radix-tree allocation failure */
203 /*
204 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
205 * clear SWAP_HAS_CACHE flag.
206 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700207 swapcache_free(entry);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700208 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 }
210}
211
212/*
213 * This must be called only on pages that have
214 * been verified to be in the swap cache and locked.
215 * It will never put the page into the free list,
216 * the caller has a reference on the page.
217 */
218void delete_from_swap_cache(struct page *page)
219{
220 swp_entry_t entry;
Shaohua Li33806f02013-02-22 16:34:37 -0800221 struct address_space *address_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700223 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
Shaohua Li33806f02013-02-22 16:34:37 -0800225 address_space = swap_address_space(entry);
226 spin_lock_irq(&address_space->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 __delete_from_swap_cache(page);
Shaohua Li33806f02013-02-22 16:34:37 -0800228 spin_unlock_irq(&address_space->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700230 swapcache_free(entry);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300231 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234/*
235 * If we are the only user, then try to free up the swap cache.
236 *
237 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800238 * here because we are going to recheck again inside
239 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 * - Marcelo
241 */
242static inline void free_swap_cache(struct page *page)
243{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800244 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
245 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 unlock_page(page);
247 }
248}
249
250/*
251 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700252 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 */
254void free_page_and_swap_cache(struct page *page)
255{
256 free_swap_cache(page);
Aaron Lu6fcb52a2016-10-07 17:00:08 -0700257 if (!is_huge_zero_page(page))
Gerald Schaefer770a5372016-06-08 15:33:50 -0700258 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259}
260
261/*
262 * Passed an array of pages, drop them all from swapcache and then release
263 * them. They are removed from the LRU and freed if this is their last use.
264 */
265void free_pages_and_swap_cache(struct page **pages, int nr)
266{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 struct page **pagep = pages;
Michal Hockoaabfb572014-10-09 15:28:52 -0700268 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270 lru_add_drain();
Michal Hockoaabfb572014-10-09 15:28:52 -0700271 for (i = 0; i < nr; i++)
272 free_swap_cache(pagep[i]);
273 release_pages(pagep, nr, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274}
275
276/*
277 * Lookup a swap entry in the swap cache. A found page will be returned
278 * unlocked and with its refcount incremented - we rely on the kernel
279 * lock getting page table operations atomic even if we drop the page
280 * lock before returning.
281 */
282struct page * lookup_swap_cache(swp_entry_t entry)
283{
284 struct page *page;
285
Huang Yingf6ab1f72016-10-07 17:00:21 -0700286 page = find_get_page(swap_address_space(entry), swp_offset(entry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Shaohua Li579f8292014-02-06 12:04:21 -0800288 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 INC_CACHE_INFO(find_success);
Shaohua Li579f8292014-02-06 12:04:21 -0800290 if (TestClearPageReadahead(page))
291 atomic_inc(&swapin_readahead_hits);
292 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
294 INC_CACHE_INFO(find_total);
295 return page;
296}
297
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700298struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
299 struct vm_area_struct *vma, unsigned long addr,
300 bool *new_page_allocated)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301{
302 struct page *found_page, *new_page = NULL;
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700303 struct address_space *swapper_space = swap_address_space(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 int err;
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700305 *new_page_allocated = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
307 do {
308 /*
309 * First check the swap cache. Since this is normally
310 * called after lookup_swap_cache() failed, re-calling
311 * that would confuse statistics.
312 */
Huang Yingf6ab1f72016-10-07 17:00:21 -0700313 found_page = find_get_page(swapper_space, swp_offset(entry));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 if (found_page)
315 break;
316
317 /*
318 * Get a new page to read into from swap.
319 */
320 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800321 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 if (!new_page)
323 break; /* Out of memory */
324 }
325
326 /*
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700327 * call radix_tree_preload() while we can wait.
328 */
Jan Kara5e4c0d972013-09-11 14:26:05 -0700329 err = radix_tree_maybe_preload(gfp_mask & GFP_KERNEL);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700330 if (err)
331 break;
332
333 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800334 * Swap entry may have been freed since our caller observed it.
335 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700336 err = swapcache_prepare(entry);
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700337 if (err == -EEXIST) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700338 radix_tree_preload_end();
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700339 /*
340 * We might race against get_swap_page() and stumble
341 * across a SWAP_HAS_CACHE swap_map entry whose page
342 * has not been brought into the swapcache yet, while
343 * the other end is scheduled away waiting on discard
344 * I/O completion at scan_swap_map().
345 *
346 * In order to avoid turning this transitory state
347 * into a permanent loop around this -EEXIST case
348 * if !CONFIG_PREEMPT and the I/O completion happens
349 * to be waiting on the CPU waitqueue where we are now
350 * busy looping, we just conditionally invoke the
351 * scheduler here, if there are some more important
352 * tasks to run.
353 */
354 cond_resched();
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700355 continue;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700356 }
357 if (err) { /* swp entry is obsolete ? */
358 radix_tree_preload_end();
Hugh Dickinsf0009442008-02-04 22:28:49 -0800359 break;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700360 }
Hugh Dickinsf0009442008-02-04 22:28:49 -0800361
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700362 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800363 __SetPageLocked(new_page);
Hugh Dickinsfa9949d2016-05-19 17:12:41 -0700364 __SetPageSwapBacked(new_page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700365 err = __add_to_swap_cache(new_page, entry);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200366 if (likely(!err)) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700367 radix_tree_preload_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 /*
369 * Initiate read into locked page and return.
370 */
Rik van Rielc5fdae42008-10-18 20:26:36 -0700371 lru_cache_add_anon(new_page);
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700372 *new_page_allocated = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 return new_page;
374 }
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700375 radix_tree_preload_end();
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800376 __ClearPageLocked(new_page);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700377 /*
378 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
379 * clear SWAP_HAS_CACHE flag.
380 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700381 swapcache_free(entry);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800382 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383
384 if (new_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300385 put_page(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 return found_page;
387}
Hugh Dickins46017e92008-02-04 22:28:41 -0800388
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700389/*
390 * Locate a page of swap in physical memory, reserving swap cache space
391 * and reading the disk if it is not already cached.
392 * A failure return means that either the page allocation failed or that
393 * the swap entry is no longer in use.
394 */
395struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
396 struct vm_area_struct *vma, unsigned long addr)
397{
398 bool page_was_allocated;
399 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
400 vma, addr, &page_was_allocated);
401
402 if (page_was_allocated)
403 swap_readpage(retpage);
404
405 return retpage;
406}
407
Shaohua Li579f8292014-02-06 12:04:21 -0800408static unsigned long swapin_nr_pages(unsigned long offset)
409{
410 static unsigned long prev_offset;
411 unsigned int pages, max_pages, last_ra;
412 static atomic_t last_readahead_pages;
413
Jason Low4db0c3c2015-04-15 16:14:08 -0700414 max_pages = 1 << READ_ONCE(page_cluster);
Shaohua Li579f8292014-02-06 12:04:21 -0800415 if (max_pages <= 1)
416 return 1;
417
418 /*
419 * This heuristic has been found to work well on both sequential and
420 * random loads, swapping to hard disk or to SSD: please don't ask
421 * what the "+ 2" means, it just happens to work well, that's all.
422 */
423 pages = atomic_xchg(&swapin_readahead_hits, 0) + 2;
424 if (pages == 2) {
425 /*
426 * We can have no readahead hits to judge by: but must not get
427 * stuck here forever, so check for an adjacent offset instead
428 * (and don't even bother to check whether swap type is same).
429 */
430 if (offset != prev_offset + 1 && offset != prev_offset - 1)
431 pages = 1;
432 prev_offset = offset;
433 } else {
434 unsigned int roundup = 4;
435 while (roundup < pages)
436 roundup <<= 1;
437 pages = roundup;
438 }
439
440 if (pages > max_pages)
441 pages = max_pages;
442
443 /* Don't shrink readahead too fast */
444 last_ra = atomic_read(&last_readahead_pages) / 2;
445 if (pages < last_ra)
446 pages = last_ra;
447 atomic_set(&last_readahead_pages, pages);
448
449 return pages;
450}
451
Hugh Dickins46017e92008-02-04 22:28:41 -0800452/**
453 * swapin_readahead - swap in pages in hope we need them soon
454 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700455 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800456 * @vma: user vma this address belongs to
457 * @addr: target address for mempolicy
458 *
459 * Returns the struct page for entry and addr, after queueing swapin.
460 *
461 * Primitive swap readahead code. We simply read an aligned block of
462 * (1 << page_cluster) entries in the swap area. This method is chosen
463 * because it doesn't cost us any seek time. We also make sure to queue
464 * the 'original' request together with the readahead ones...
465 *
466 * This has been extended to use the NUMA policies from the mm triggering
467 * the readahead.
468 *
469 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
470 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800471struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800472 struct vm_area_struct *vma, unsigned long addr)
473{
Hugh Dickins46017e92008-02-04 22:28:41 -0800474 struct page *page;
Shaohua Li579f8292014-02-06 12:04:21 -0800475 unsigned long entry_offset = swp_offset(entry);
476 unsigned long offset = entry_offset;
Rik van Riel67f96aa2012-03-21 16:33:50 -0700477 unsigned long start_offset, end_offset;
Vinayak Menonee081c232015-11-20 13:37:08 +0530478 unsigned long mask;
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700479 struct blk_plug plug;
Hugh Dickins46017e92008-02-04 22:28:41 -0800480
Vinayak Menonee081c232015-11-20 13:37:08 +0530481 mask = is_swap_fast(entry) ? 0 : swapin_nr_pages(offset) - 1;
Shaohua Li579f8292014-02-06 12:04:21 -0800482 if (!mask)
483 goto skip;
484
Rik van Riel67f96aa2012-03-21 16:33:50 -0700485 /* Read a page_cluster sized and aligned cluster around offset. */
486 start_offset = offset & ~mask;
487 end_offset = offset | mask;
488 if (!start_offset) /* First page is swap header. */
489 start_offset++;
490
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700491 blk_start_plug(&plug);
Rik van Riel67f96aa2012-03-21 16:33:50 -0700492 for (offset = start_offset; offset <= end_offset ; offset++) {
Hugh Dickins46017e92008-02-04 22:28:41 -0800493 /* Ok, do the async read-ahead now */
494 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800495 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800496 if (!page)
Rik van Riel67f96aa2012-03-21 16:33:50 -0700497 continue;
Shaohua Li579f8292014-02-06 12:04:21 -0800498 if (offset != entry_offset)
499 SetPageReadahead(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300500 put_page(page);
Hugh Dickins46017e92008-02-04 22:28:41 -0800501 }
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700502 blk_finish_plug(&plug);
503
Hugh Dickins46017e92008-02-04 22:28:41 -0800504 lru_add_drain(); /* Push any new pages onto the LRU now */
Shaohua Li579f8292014-02-06 12:04:21 -0800505skip:
Hugh Dickins02098fe2008-02-04 22:28:42 -0800506 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800507}