blob: 0d457e7db8d6df5e3bd185ac2b2333830bccf7b4 [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,
Shaohua Li33806f02013-02-22 16:34:37 -080040 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070041};
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
44
45static struct {
46 unsigned long add_total;
47 unsigned long del_total;
48 unsigned long find_success;
49 unsigned long find_total;
Linus Torvalds1da177e2005-04-16 15:20:36 -070050} swap_cache_info;
51
Shaohua Li33806f02013-02-22 16:34:37 -080052unsigned long total_swapcache_pages(void)
53{
54 int i;
55 unsigned long ret = 0;
56
57 for (i = 0; i < MAX_SWAPFILES; i++)
58 ret += swapper_spaces[i].nrpages;
59 return ret;
60}
61
Shaohua Li579f8292014-02-06 12:04:21 -080062static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
63
Linus Torvalds1da177e2005-04-16 15:20:36 -070064void show_swap_cache_info(void)
65{
Shaohua Li33806f02013-02-22 16:34:37 -080066 printk("%lu pages in swap cache\n", total_swapcache_pages());
Johannes Weiner2c97b7f2008-07-25 19:46:01 -070067 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 swap_cache_info.add_total, swap_cache_info.del_total,
Hugh Dickinsbb63be02008-02-04 22:28:49 -080069 swap_cache_info.find_success, swap_cache_info.find_total);
Shaohua Liec8acf22013-02-22 16:34:38 -080070 printk("Free swap = %ldkB\n",
71 get_nr_swap_pages() << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
73}
74
75/*
Daisuke Nishimura31a56392009-09-21 17:02:50 -070076 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 * but sets SwapCache flag and private instead of mapping and index.
78 */
Seth Jennings2f772e62013-04-29 15:08:34 -070079int __add_to_swap_cache(struct page *page, swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070080{
81 int error;
Shaohua Li33806f02013-02-22 16:34:37 -080082 struct address_space *address_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
Sasha Levin309381fea2014-01-23 15:52:54 -080084 VM_BUG_ON_PAGE(!PageLocked(page), page);
85 VM_BUG_ON_PAGE(PageSwapCache(page), page);
86 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins51726b12009-01-06 14:39:25 -080087
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030088 get_page(page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -070089 SetPageSwapCache(page);
90 set_page_private(page, entry.val);
91
Shaohua Li33806f02013-02-22 16:34:37 -080092 address_space = swap_address_space(entry);
93 spin_lock_irq(&address_space->tree_lock);
94 error = radix_tree_insert(&address_space->page_tree,
95 entry.val, page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -070096 if (likely(!error)) {
Shaohua Li33806f02013-02-22 16:34:37 -080097 address_space->nrpages++;
Daisuke Nishimura31a56392009-09-21 17:02:50 -070098 __inc_zone_page_state(page, NR_FILE_PAGES);
99 INC_CACHE_INFO(add_total);
100 }
Shaohua Li33806f02013-02-22 16:34:37 -0800101 spin_unlock_irq(&address_space->tree_lock);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700102
103 if (unlikely(error)) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700104 /*
105 * Only the context which have set SWAP_HAS_CACHE flag
106 * would call add_to_swap_cache().
107 * So add_to_swap_cache() doesn't returns -EEXIST.
108 */
109 VM_BUG_ON(error == -EEXIST);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700110 set_page_private(page, 0UL);
111 ClearPageSwapCache(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300112 put_page(page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700113 }
114
115 return error;
116}
117
118
119int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
120{
121 int error;
122
Jan Kara5e4c0d972013-09-11 14:26:05 -0700123 error = radix_tree_maybe_preload(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 if (!error) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700125 error = __add_to_swap_cache(page, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 radix_tree_preload_end();
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800127 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 return error;
129}
130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131/*
132 * This must be called only on pages that have
133 * been verified to be in the swap cache.
134 */
135void __delete_from_swap_cache(struct page *page)
136{
Shaohua Li33806f02013-02-22 16:34:37 -0800137 swp_entry_t entry;
138 struct address_space *address_space;
139
Sasha Levin309381fea2014-01-23 15:52:54 -0800140 VM_BUG_ON_PAGE(!PageLocked(page), page);
141 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
142 VM_BUG_ON_PAGE(PageWriteback(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Shaohua Li33806f02013-02-22 16:34:37 -0800144 entry.val = page_private(page);
145 address_space = swap_address_space(entry);
146 radix_tree_delete(&address_space->page_tree, page_private(page));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700147 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 ClearPageSwapCache(page);
Shaohua Li33806f02013-02-22 16:34:37 -0800149 address_space->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700150 __dec_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 INC_CACHE_INFO(del_total);
152}
153
154/**
155 * add_to_swap - allocate swap space for a page
156 * @page: page we want to move to swap
157 *
158 * Allocate swap space for the page and add the page to the
159 * swap cache. Caller needs to hold the page lock.
160 */
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700161int add_to_swap(struct page *page, struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162{
163 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 int err;
165
Sasha Levin309381fea2014-01-23 15:52:54 -0800166 VM_BUG_ON_PAGE(!PageLocked(page), page);
167 VM_BUG_ON_PAGE(!PageUptodate(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700169 entry = get_swap_page();
170 if (!entry.val)
171 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Vladimir Davydov37e84352016-01-20 15:02:56 -0800173 if (mem_cgroup_try_charge_swap(page, entry)) {
174 swapcache_free(entry);
175 return 0;
176 }
177
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800178 if (unlikely(PageTransHuge(page)))
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700179 if (unlikely(split_huge_page_to_list(page, list))) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700180 swapcache_free(entry);
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800181 return 0;
182 }
183
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700184 /*
185 * Radix-tree node allocations from PF_MEMALLOC contexts could
186 * completely exhaust the page allocator. __GFP_NOMEMALLOC
187 * stops emergency reserves from being allocated.
188 *
189 * TODO: this could cause a theoretical memory reclaim
190 * deadlock in the swap out path.
191 */
192 /*
Minchan Kim854e9ed2016-01-15 16:54:53 -0800193 * Add it to the swap cache.
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700194 */
195 err = add_to_swap_cache(page, entry,
196 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
Minchan Kim854e9ed2016-01-15 16:54:53 -0800198 if (!err) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700199 return 1;
200 } else { /* -ENOMEM radix-tree allocation failure */
201 /*
202 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
203 * clear SWAP_HAS_CACHE flag.
204 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700205 swapcache_free(entry);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700206 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 }
208}
209
210/*
211 * This must be called only on pages that have
212 * been verified to be in the swap cache and locked.
213 * It will never put the page into the free list,
214 * the caller has a reference on the page.
215 */
216void delete_from_swap_cache(struct page *page)
217{
218 swp_entry_t entry;
Shaohua Li33806f02013-02-22 16:34:37 -0800219 struct address_space *address_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700221 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Shaohua Li33806f02013-02-22 16:34:37 -0800223 address_space = swap_address_space(entry);
224 spin_lock_irq(&address_space->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 __delete_from_swap_cache(page);
Shaohua Li33806f02013-02-22 16:34:37 -0800226 spin_unlock_irq(&address_space->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700228 swapcache_free(entry);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300229 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230}
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232/*
233 * If we are the only user, then try to free up the swap cache.
234 *
235 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800236 * here because we are going to recheck again inside
237 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 * - Marcelo
239 */
240static inline void free_swap_cache(struct page *page)
241{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800242 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
243 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 unlock_page(page);
245 }
246}
247
248/*
249 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700250 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 */
252void free_page_and_swap_cache(struct page *page)
253{
254 free_swap_cache(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300255 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
258/*
259 * Passed an array of pages, drop them all from swapcache and then release
260 * them. They are removed from the LRU and freed if this is their last use.
261 */
262void free_pages_and_swap_cache(struct page **pages, int nr)
263{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 struct page **pagep = pages;
Michal Hockoaabfb572014-10-09 15:28:52 -0700265 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
267 lru_add_drain();
Michal Hockoaabfb572014-10-09 15:28:52 -0700268 for (i = 0; i < nr; i++)
269 free_swap_cache(pagep[i]);
270 release_pages(pagep, nr, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
272
273/*
274 * Lookup a swap entry in the swap cache. A found page will be returned
275 * unlocked and with its refcount incremented - we rely on the kernel
276 * lock getting page table operations atomic even if we drop the page
277 * lock before returning.
278 */
279struct page * lookup_swap_cache(swp_entry_t entry)
280{
281 struct page *page;
282
Shaohua Li33806f02013-02-22 16:34:37 -0800283 page = find_get_page(swap_address_space(entry), entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
Shaohua Li579f8292014-02-06 12:04:21 -0800285 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 INC_CACHE_INFO(find_success);
Shaohua Li579f8292014-02-06 12:04:21 -0800287 if (TestClearPageReadahead(page))
288 atomic_inc(&swapin_readahead_hits);
289 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
291 INC_CACHE_INFO(find_total);
292 return page;
293}
294
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700295struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
296 struct vm_area_struct *vma, unsigned long addr,
297 bool *new_page_allocated)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
299 struct page *found_page, *new_page = NULL;
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700300 struct address_space *swapper_space = swap_address_space(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 int err;
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700302 *new_page_allocated = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
304 do {
305 /*
306 * First check the swap cache. Since this is normally
307 * called after lookup_swap_cache() failed, re-calling
308 * that would confuse statistics.
309 */
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700310 found_page = find_get_page(swapper_space, entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 if (found_page)
312 break;
313
314 /*
315 * Get a new page to read into from swap.
316 */
317 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800318 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 if (!new_page)
320 break; /* Out of memory */
321 }
322
323 /*
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700324 * call radix_tree_preload() while we can wait.
325 */
Jan Kara5e4c0d972013-09-11 14:26:05 -0700326 err = radix_tree_maybe_preload(gfp_mask & GFP_KERNEL);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700327 if (err)
328 break;
329
330 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800331 * Swap entry may have been freed since our caller observed it.
332 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700333 err = swapcache_prepare(entry);
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700334 if (err == -EEXIST) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700335 radix_tree_preload_end();
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700336 /*
337 * We might race against get_swap_page() and stumble
338 * across a SWAP_HAS_CACHE swap_map entry whose page
339 * has not been brought into the swapcache yet, while
340 * the other end is scheduled away waiting on discard
341 * I/O completion at scan_swap_map().
342 *
343 * In order to avoid turning this transitory state
344 * into a permanent loop around this -EEXIST case
345 * if !CONFIG_PREEMPT and the I/O completion happens
346 * to be waiting on the CPU waitqueue where we are now
347 * busy looping, we just conditionally invoke the
348 * scheduler here, if there are some more important
349 * tasks to run.
350 */
351 cond_resched();
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700352 continue;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700353 }
354 if (err) { /* swp entry is obsolete ? */
355 radix_tree_preload_end();
Hugh Dickinsf0009442008-02-04 22:28:49 -0800356 break;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700357 }
Hugh Dickinsf0009442008-02-04 22:28:49 -0800358
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700359 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800360 __SetPageLocked(new_page);
Hugh Dickinsfa9949d2016-05-19 17:12:41 -0700361 __SetPageSwapBacked(new_page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700362 err = __add_to_swap_cache(new_page, entry);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200363 if (likely(!err)) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700364 radix_tree_preload_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 /*
366 * Initiate read into locked page and return.
367 */
Rik van Rielc5fdae42008-10-18 20:26:36 -0700368 lru_cache_add_anon(new_page);
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700369 *new_page_allocated = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 return new_page;
371 }
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700372 radix_tree_preload_end();
Kirill A. Shutemov48c935a2016-01-15 16:51:24 -0800373 __ClearPageLocked(new_page);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700374 /*
375 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
376 * clear SWAP_HAS_CACHE flag.
377 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700378 swapcache_free(entry);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800379 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
381 if (new_page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300382 put_page(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 return found_page;
384}
Hugh Dickins46017e92008-02-04 22:28:41 -0800385
Dmitry Safonov5b999aa2015-09-08 15:05:00 -0700386/*
387 * Locate a page of swap in physical memory, reserving swap cache space
388 * and reading the disk if it is not already cached.
389 * A failure return means that either the page allocation failed or that
390 * the swap entry is no longer in use.
391 */
392struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
393 struct vm_area_struct *vma, unsigned long addr)
394{
395 bool page_was_allocated;
396 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
397 vma, addr, &page_was_allocated);
398
399 if (page_was_allocated)
400 swap_readpage(retpage);
401
402 return retpage;
403}
404
Shaohua Li579f8292014-02-06 12:04:21 -0800405static unsigned long swapin_nr_pages(unsigned long offset)
406{
407 static unsigned long prev_offset;
408 unsigned int pages, max_pages, last_ra;
409 static atomic_t last_readahead_pages;
410
Jason Low4db0c3c2015-04-15 16:14:08 -0700411 max_pages = 1 << READ_ONCE(page_cluster);
Shaohua Li579f8292014-02-06 12:04:21 -0800412 if (max_pages <= 1)
413 return 1;
414
415 /*
416 * This heuristic has been found to work well on both sequential and
417 * random loads, swapping to hard disk or to SSD: please don't ask
418 * what the "+ 2" means, it just happens to work well, that's all.
419 */
420 pages = atomic_xchg(&swapin_readahead_hits, 0) + 2;
421 if (pages == 2) {
422 /*
423 * We can have no readahead hits to judge by: but must not get
424 * stuck here forever, so check for an adjacent offset instead
425 * (and don't even bother to check whether swap type is same).
426 */
427 if (offset != prev_offset + 1 && offset != prev_offset - 1)
428 pages = 1;
429 prev_offset = offset;
430 } else {
431 unsigned int roundup = 4;
432 while (roundup < pages)
433 roundup <<= 1;
434 pages = roundup;
435 }
436
437 if (pages > max_pages)
438 pages = max_pages;
439
440 /* Don't shrink readahead too fast */
441 last_ra = atomic_read(&last_readahead_pages) / 2;
442 if (pages < last_ra)
443 pages = last_ra;
444 atomic_set(&last_readahead_pages, pages);
445
446 return pages;
447}
448
Hugh Dickins46017e92008-02-04 22:28:41 -0800449/**
450 * swapin_readahead - swap in pages in hope we need them soon
451 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700452 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800453 * @vma: user vma this address belongs to
454 * @addr: target address for mempolicy
455 *
456 * Returns the struct page for entry and addr, after queueing swapin.
457 *
458 * Primitive swap readahead code. We simply read an aligned block of
459 * (1 << page_cluster) entries in the swap area. This method is chosen
460 * because it doesn't cost us any seek time. We also make sure to queue
461 * the 'original' request together with the readahead ones...
462 *
463 * This has been extended to use the NUMA policies from the mm triggering
464 * the readahead.
465 *
466 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
467 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800468struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800469 struct vm_area_struct *vma, unsigned long addr)
470{
Hugh Dickins46017e92008-02-04 22:28:41 -0800471 struct page *page;
Shaohua Li579f8292014-02-06 12:04:21 -0800472 unsigned long entry_offset = swp_offset(entry);
473 unsigned long offset = entry_offset;
Rik van Riel67f96aa2012-03-21 16:33:50 -0700474 unsigned long start_offset, end_offset;
Shaohua Li579f8292014-02-06 12:04:21 -0800475 unsigned long mask;
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700476 struct blk_plug plug;
Hugh Dickins46017e92008-02-04 22:28:41 -0800477
Shaohua Li579f8292014-02-06 12:04:21 -0800478 mask = swapin_nr_pages(offset) - 1;
479 if (!mask)
480 goto skip;
481
Rik van Riel67f96aa2012-03-21 16:33:50 -0700482 /* Read a page_cluster sized and aligned cluster around offset. */
483 start_offset = offset & ~mask;
484 end_offset = offset | mask;
485 if (!start_offset) /* First page is swap header. */
486 start_offset++;
487
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700488 blk_start_plug(&plug);
Rik van Riel67f96aa2012-03-21 16:33:50 -0700489 for (offset = start_offset; offset <= end_offset ; offset++) {
Hugh Dickins46017e92008-02-04 22:28:41 -0800490 /* Ok, do the async read-ahead now */
491 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800492 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800493 if (!page)
Rik van Riel67f96aa2012-03-21 16:33:50 -0700494 continue;
Shaohua Li579f8292014-02-06 12:04:21 -0800495 if (offset != entry_offset)
496 SetPageReadahead(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300497 put_page(page);
Hugh Dickins46017e92008-02-04 22:28:41 -0800498 }
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700499 blk_finish_plug(&plug);
500
Hugh Dickins46017e92008-02-04 22:28:41 -0800501 lru_add_drain(); /* Push any new pages onto the LRU now */
Shaohua Li579f8292014-02-06 12:04:21 -0800502skip:
Hugh Dickins02098fe2008-02-04 22:28:42 -0800503 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800504}