blob: 2972eee184a44c1dc1df3eeed766d7d3140e353b [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>
KAMEZAWA Hiroyuki8c7c6e342009-01-07 18:08:00 -080020#include <linux/page_cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021
22#include <asm/pgtable.h>
23
24/*
25 * swapper_space is a fiction, retained to simplify the path through
Jens Axboe7eaceac2011-03-10 08:52:07 +010026 * vmscan's shrink_page_list.
Linus Torvalds1da177e2005-04-16 15:20:36 -070027 */
Christoph Hellwigf5e54d62006-06-28 04:26:44 -070028static const struct address_space_operations swap_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070029 .writepage = swap_writepage,
Mel Gorman62c230b2012-07-31 16:44:55 -070030 .set_page_dirty = swap_set_page_dirty,
Christoph Lametere965f962006-02-01 03:05:41 -080031 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -070032};
33
34static struct backing_dev_info swap_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +020035 .name = "swap",
Rik van Riel4f98a2f2008-10-18 20:26:32 -070036 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -070037};
38
Shaohua Li33806f02013-02-22 16:34:37 -080039struct address_space swapper_spaces[MAX_SWAPFILES] = {
40 [0 ... MAX_SWAPFILES - 1] = {
41 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
42 .a_ops = &swap_aops,
43 .backing_dev_info = &swap_backing_dev_info,
44 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070045};
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
48
49static struct {
50 unsigned long add_total;
51 unsigned long del_total;
52 unsigned long find_success;
53 unsigned long find_total;
Linus Torvalds1da177e2005-04-16 15:20:36 -070054} swap_cache_info;
55
Shaohua Li33806f02013-02-22 16:34:37 -080056unsigned long total_swapcache_pages(void)
57{
58 int i;
59 unsigned long ret = 0;
60
61 for (i = 0; i < MAX_SWAPFILES; i++)
62 ret += swapper_spaces[i].nrpages;
63 return ret;
64}
65
Shaohua Li579f8292014-02-06 12:04:21 -080066static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
67
Linus Torvalds1da177e2005-04-16 15:20:36 -070068void show_swap_cache_info(void)
69{
Shaohua Li33806f02013-02-22 16:34:37 -080070 printk("%lu pages in swap cache\n", total_swapcache_pages());
Johannes Weiner2c97b7f2008-07-25 19:46:01 -070071 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 swap_cache_info.add_total, swap_cache_info.del_total,
Hugh Dickinsbb63be02008-02-04 22:28:49 -080073 swap_cache_info.find_success, swap_cache_info.find_total);
Shaohua Liec8acf22013-02-22 16:34:38 -080074 printk("Free swap = %ldkB\n",
75 get_nr_swap_pages() << (PAGE_SHIFT - 10));
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
77}
78
79/*
Daisuke Nishimura31a56392009-09-21 17:02:50 -070080 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 * but sets SwapCache flag and private instead of mapping and index.
82 */
Seth Jennings2f772e62013-04-29 15:08:34 -070083int __add_to_swap_cache(struct page *page, swp_entry_t entry)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084{
85 int error;
Shaohua Li33806f02013-02-22 16:34:37 -080086 struct address_space *address_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Sasha Levin309381fea2014-01-23 15:52:54 -080088 VM_BUG_ON_PAGE(!PageLocked(page), page);
89 VM_BUG_ON_PAGE(PageSwapCache(page), page);
90 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Hugh Dickins51726b12009-01-06 14:39:25 -080091
Daisuke Nishimura31a56392009-09-21 17:02:50 -070092 page_cache_get(page);
93 SetPageSwapCache(page);
94 set_page_private(page, entry.val);
95
Shaohua Li33806f02013-02-22 16:34:37 -080096 address_space = swap_address_space(entry);
97 spin_lock_irq(&address_space->tree_lock);
98 error = radix_tree_insert(&address_space->page_tree,
99 entry.val, page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700100 if (likely(!error)) {
Shaohua Li33806f02013-02-22 16:34:37 -0800101 address_space->nrpages++;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700102 __inc_zone_page_state(page, NR_FILE_PAGES);
103 INC_CACHE_INFO(add_total);
104 }
Shaohua Li33806f02013-02-22 16:34:37 -0800105 spin_unlock_irq(&address_space->tree_lock);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700106
107 if (unlikely(error)) {
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700108 /*
109 * Only the context which have set SWAP_HAS_CACHE flag
110 * would call add_to_swap_cache().
111 * So add_to_swap_cache() doesn't returns -EEXIST.
112 */
113 VM_BUG_ON(error == -EEXIST);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700114 set_page_private(page, 0UL);
115 ClearPageSwapCache(page);
116 page_cache_release(page);
117 }
118
119 return error;
120}
121
122
123int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
124{
125 int error;
126
Jan Kara5e4c0d972013-09-11 14:26:05 -0700127 error = radix_tree_maybe_preload(gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 if (!error) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700129 error = __add_to_swap_cache(page, entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 radix_tree_preload_end();
Hugh Dickinsfa1de902008-02-07 00:14:13 -0800131 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 return error;
133}
134
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135/*
136 * This must be called only on pages that have
137 * been verified to be in the swap cache.
138 */
139void __delete_from_swap_cache(struct page *page)
140{
Shaohua Li33806f02013-02-22 16:34:37 -0800141 swp_entry_t entry;
142 struct address_space *address_space;
143
Sasha Levin309381fea2014-01-23 15:52:54 -0800144 VM_BUG_ON_PAGE(!PageLocked(page), page);
145 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
146 VM_BUG_ON_PAGE(PageWriteback(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
Shaohua Li33806f02013-02-22 16:34:37 -0800148 entry.val = page_private(page);
149 address_space = swap_address_space(entry);
150 radix_tree_delete(&address_space->page_tree, page_private(page));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700151 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 ClearPageSwapCache(page);
Shaohua Li33806f02013-02-22 16:34:37 -0800153 address_space->nrpages--;
Christoph Lameter347ce432006-06-30 01:55:35 -0700154 __dec_zone_page_state(page, NR_FILE_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 INC_CACHE_INFO(del_total);
156}
157
158/**
159 * add_to_swap - allocate swap space for a page
160 * @page: page we want to move to swap
161 *
162 * Allocate swap space for the page and add the page to the
163 * swap cache. Caller needs to hold the page lock.
164 */
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700165int add_to_swap(struct page *page, struct list_head *list)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
167 swp_entry_t entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 int err;
169
Sasha Levin309381fea2014-01-23 15:52:54 -0800170 VM_BUG_ON_PAGE(!PageLocked(page), page);
171 VM_BUG_ON_PAGE(!PageUptodate(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700173 entry = get_swap_page();
174 if (!entry.val)
175 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800177 if (unlikely(PageTransHuge(page)))
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700178 if (unlikely(split_huge_page_to_list(page, list))) {
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800179 swapcache_free(entry, NULL);
180 return 0;
181 }
182
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700183 /*
184 * Radix-tree node allocations from PF_MEMALLOC contexts could
185 * completely exhaust the page allocator. __GFP_NOMEMALLOC
186 * stops emergency reserves from being allocated.
187 *
188 * TODO: this could cause a theoretical memory reclaim
189 * deadlock in the swap out path.
190 */
191 /*
192 * Add it to the swap cache and mark it dirty
193 */
194 err = add_to_swap_cache(page, entry,
195 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700197 if (!err) { /* Success */
198 SetPageDirty(page);
199 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 */
205 swapcache_free(entry, NULL);
206 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
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700228 swapcache_free(entry, page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 page_cache_release(page);
230}
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);
255 page_cache_release(page);
256}
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;
265
266 lru_add_drain();
267 while (nr) {
Hugh Dickinsc484d412006-01-06 00:10:55 -0800268 int todo = min(nr, PAGEVEC_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 int i;
270
271 for (i = 0; i < todo; i++)
272 free_swap_cache(pagep[i]);
Mel Gormanb745bc82014-06-04 16:10:22 -0700273 release_pages(pagep, todo, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 pagep += todo;
275 nr -= todo;
276 }
277}
278
279/*
280 * Lookup a swap entry in the swap cache. A found page will be returned
281 * unlocked and with its refcount incremented - we rely on the kernel
282 * lock getting page table operations atomic even if we drop the page
283 * lock before returning.
284 */
285struct page * lookup_swap_cache(swp_entry_t entry)
286{
287 struct page *page;
288
Shaohua Li33806f02013-02-22 16:34:37 -0800289 page = find_get_page(swap_address_space(entry), entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Shaohua Li579f8292014-02-06 12:04:21 -0800291 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 INC_CACHE_INFO(find_success);
Shaohua Li579f8292014-02-06 12:04:21 -0800293 if (TestClearPageReadahead(page))
294 atomic_inc(&swapin_readahead_hits);
295 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296
297 INC_CACHE_INFO(find_total);
298 return page;
299}
300
301/*
302 * Locate a page of swap in physical memory, reserving swap cache space
303 * and reading the disk if it is not already cached.
304 * A failure return means that either the page allocation failed or that
305 * the swap entry is no longer in use.
306 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800307struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 struct vm_area_struct *vma, unsigned long addr)
309{
310 struct page *found_page, *new_page = NULL;
311 int err;
312
313 do {
314 /*
315 * First check the swap cache. Since this is normally
316 * called after lookup_swap_cache() failed, re-calling
317 * that would confuse statistics.
318 */
Shaohua Li33806f02013-02-22 16:34:37 -0800319 found_page = find_get_page(swap_address_space(entry),
320 entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 if (found_page)
322 break;
323
324 /*
325 * Get a new page to read into from swap.
326 */
327 if (!new_page) {
Hugh Dickins02098fe2008-02-04 22:28:42 -0800328 new_page = alloc_page_vma(gfp_mask, vma, addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 if (!new_page)
330 break; /* Out of memory */
331 }
332
333 /*
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700334 * call radix_tree_preload() while we can wait.
335 */
Jan Kara5e4c0d972013-09-11 14:26:05 -0700336 err = radix_tree_maybe_preload(gfp_mask & GFP_KERNEL);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700337 if (err)
338 break;
339
340 /*
Hugh Dickinsf0009442008-02-04 22:28:49 -0800341 * Swap entry may have been freed since our caller observed it.
342 */
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700343 err = swapcache_prepare(entry);
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700344 if (err == -EEXIST) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700345 radix_tree_preload_end();
Rafael Aquinicbab0e42013-06-12 14:04:49 -0700346 /*
347 * We might race against get_swap_page() and stumble
348 * across a SWAP_HAS_CACHE swap_map entry whose page
349 * has not been brought into the swapcache yet, while
350 * the other end is scheduled away waiting on discard
351 * I/O completion at scan_swap_map().
352 *
353 * In order to avoid turning this transitory state
354 * into a permanent loop around this -EEXIST case
355 * if !CONFIG_PREEMPT and the I/O completion happens
356 * to be waiting on the CPU waitqueue where we are now
357 * busy looping, we just conditionally invoke the
358 * scheduler here, if there are some more important
359 * tasks to run.
360 */
361 cond_resched();
KAMEZAWA Hiroyuki355cfa72009-06-16 15:32:53 -0700362 continue;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700363 }
364 if (err) { /* swp entry is obsolete ? */
365 radix_tree_preload_end();
Hugh Dickinsf0009442008-02-04 22:28:49 -0800366 break;
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700367 }
Hugh Dickinsf0009442008-02-04 22:28:49 -0800368
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700369 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
Nick Pigginf45840b2008-10-18 20:26:57 -0700370 __set_page_locked(new_page);
Rik van Rielb2e18532008-10-18 20:26:30 -0700371 SetPageSwapBacked(new_page);
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700372 err = __add_to_swap_cache(new_page, entry);
Nick Piggin529ae9a2008-08-02 12:01:03 +0200373 if (likely(!err)) {
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700374 radix_tree_preload_end();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 /*
376 * Initiate read into locked page and return.
377 */
Rik van Rielc5fdae42008-10-18 20:26:36 -0700378 lru_cache_add_anon(new_page);
Minchan Kimaca8bf32009-06-16 15:33:02 -0700379 swap_readpage(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 return new_page;
381 }
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700382 radix_tree_preload_end();
Rik van Rielb2e18532008-10-18 20:26:30 -0700383 ClearPageSwapBacked(new_page);
Nick Pigginf45840b2008-10-18 20:26:57 -0700384 __clear_page_locked(new_page);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700385 /*
386 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
387 * clear SWAP_HAS_CACHE flag.
388 */
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700389 swapcache_free(entry, NULL);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800390 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 if (new_page)
393 page_cache_release(new_page);
394 return found_page;
395}
Hugh Dickins46017e92008-02-04 22:28:41 -0800396
Shaohua Li579f8292014-02-06 12:04:21 -0800397static unsigned long swapin_nr_pages(unsigned long offset)
398{
399 static unsigned long prev_offset;
400 unsigned int pages, max_pages, last_ra;
401 static atomic_t last_readahead_pages;
402
403 max_pages = 1 << ACCESS_ONCE(page_cluster);
404 if (max_pages <= 1)
405 return 1;
406
407 /*
408 * This heuristic has been found to work well on both sequential and
409 * random loads, swapping to hard disk or to SSD: please don't ask
410 * what the "+ 2" means, it just happens to work well, that's all.
411 */
412 pages = atomic_xchg(&swapin_readahead_hits, 0) + 2;
413 if (pages == 2) {
414 /*
415 * We can have no readahead hits to judge by: but must not get
416 * stuck here forever, so check for an adjacent offset instead
417 * (and don't even bother to check whether swap type is same).
418 */
419 if (offset != prev_offset + 1 && offset != prev_offset - 1)
420 pages = 1;
421 prev_offset = offset;
422 } else {
423 unsigned int roundup = 4;
424 while (roundup < pages)
425 roundup <<= 1;
426 pages = roundup;
427 }
428
429 if (pages > max_pages)
430 pages = max_pages;
431
432 /* Don't shrink readahead too fast */
433 last_ra = atomic_read(&last_readahead_pages) / 2;
434 if (pages < last_ra)
435 pages = last_ra;
436 atomic_set(&last_readahead_pages, pages);
437
438 return pages;
439}
440
Hugh Dickins46017e92008-02-04 22:28:41 -0800441/**
442 * swapin_readahead - swap in pages in hope we need them soon
443 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700444 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800445 * @vma: user vma this address belongs to
446 * @addr: target address for mempolicy
447 *
448 * Returns the struct page for entry and addr, after queueing swapin.
449 *
450 * Primitive swap readahead code. We simply read an aligned block of
451 * (1 << page_cluster) entries in the swap area. This method is chosen
452 * because it doesn't cost us any seek time. We also make sure to queue
453 * the 'original' request together with the readahead ones...
454 *
455 * This has been extended to use the NUMA policies from the mm triggering
456 * the readahead.
457 *
458 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
459 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800460struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800461 struct vm_area_struct *vma, unsigned long addr)
462{
Hugh Dickins46017e92008-02-04 22:28:41 -0800463 struct page *page;
Shaohua Li579f8292014-02-06 12:04:21 -0800464 unsigned long entry_offset = swp_offset(entry);
465 unsigned long offset = entry_offset;
Rik van Riel67f96aa2012-03-21 16:33:50 -0700466 unsigned long start_offset, end_offset;
Shaohua Li579f8292014-02-06 12:04:21 -0800467 unsigned long mask;
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700468 struct blk_plug plug;
Hugh Dickins46017e92008-02-04 22:28:41 -0800469
Shaohua Li579f8292014-02-06 12:04:21 -0800470 mask = swapin_nr_pages(offset) - 1;
471 if (!mask)
472 goto skip;
473
Rik van Riel67f96aa2012-03-21 16:33:50 -0700474 /* Read a page_cluster sized and aligned cluster around offset. */
475 start_offset = offset & ~mask;
476 end_offset = offset | mask;
477 if (!start_offset) /* First page is swap header. */
478 start_offset++;
479
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700480 blk_start_plug(&plug);
Rik van Riel67f96aa2012-03-21 16:33:50 -0700481 for (offset = start_offset; offset <= end_offset ; offset++) {
Hugh Dickins46017e92008-02-04 22:28:41 -0800482 /* Ok, do the async read-ahead now */
483 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800484 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800485 if (!page)
Rik van Riel67f96aa2012-03-21 16:33:50 -0700486 continue;
Shaohua Li579f8292014-02-06 12:04:21 -0800487 if (offset != entry_offset)
488 SetPageReadahead(page);
Hugh Dickins46017e92008-02-04 22:28:41 -0800489 page_cache_release(page);
490 }
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700491 blk_finish_plug(&plug);
492
Hugh Dickins46017e92008-02-04 22:28:41 -0800493 lru_add_drain(); /* Push any new pages onto the LRU now */
Shaohua Li579f8292014-02-06 12:04:21 -0800494skip:
Hugh Dickins02098fe2008-02-04 22:28:42 -0800495 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800496}