blob: 8bc8e66138da1baec2b1420564deb5745f19c990 [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
Daisuke Nishimura31a56392009-09-21 17:02:50 -070088 page_cache_get(page);
89 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);
112 page_cache_release(page);
113 }
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
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800173 if (unlikely(PageTransHuge(page)))
Shaohua Li5bc7b8a2013-04-29 15:08:36 -0700174 if (unlikely(split_huge_page_to_list(page, list))) {
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700175 swapcache_free(entry);
Andrea Arcangeli3f04f622011-01-13 15:46:47 -0800176 return 0;
177 }
178
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700179 /*
180 * Radix-tree node allocations from PF_MEMALLOC contexts could
181 * completely exhaust the page allocator. __GFP_NOMEMALLOC
182 * stops emergency reserves from being allocated.
183 *
184 * TODO: this could cause a theoretical memory reclaim
185 * deadlock in the swap out path.
186 */
187 /*
188 * Add it to the swap cache and mark it dirty
189 */
190 err = add_to_swap_cache(page, entry,
191 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700193 if (!err) { /* Success */
194 SetPageDirty(page);
195 return 1;
196 } else { /* -ENOMEM radix-tree allocation failure */
197 /*
198 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
199 * clear SWAP_HAS_CACHE flag.
200 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700201 swapcache_free(entry);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700202 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 }
204}
205
206/*
207 * This must be called only on pages that have
208 * been verified to be in the swap cache and locked.
209 * It will never put the page into the free list,
210 * the caller has a reference on the page.
211 */
212void delete_from_swap_cache(struct page *page)
213{
214 swp_entry_t entry;
Shaohua Li33806f02013-02-22 16:34:37 -0800215 struct address_space *address_space;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700217 entry.val = page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Shaohua Li33806f02013-02-22 16:34:37 -0800219 address_space = swap_address_space(entry);
220 spin_lock_irq(&address_space->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 __delete_from_swap_cache(page);
Shaohua Li33806f02013-02-22 16:34:37 -0800222 spin_unlock_irq(&address_space->tree_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700224 swapcache_free(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 page_cache_release(page);
226}
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228/*
229 * If we are the only user, then try to free up the swap cache.
230 *
231 * Its ok to check for PageSwapCache without the page lock
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800232 * here because we are going to recheck again inside
233 * try_to_free_swap() _with_ the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 * - Marcelo
235 */
236static inline void free_swap_cache(struct page *page)
237{
Hugh Dickinsa2c43ee2009-01-06 14:39:36 -0800238 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
239 try_to_free_swap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 unlock_page(page);
241 }
242}
243
244/*
245 * Perform a free_page(), also freeing any swap cache associated with
Hugh Dickinsb8072f02005-10-29 18:16:41 -0700246 * this page if it is the last user of the page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 */
248void free_page_and_swap_cache(struct page *page)
249{
250 free_swap_cache(page);
251 page_cache_release(page);
252}
253
254/*
255 * Passed an array of pages, drop them all from swapcache and then release
256 * them. They are removed from the LRU and freed if this is their last use.
257 */
258void free_pages_and_swap_cache(struct page **pages, int nr)
259{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 struct page **pagep = pages;
Michal Hockoaabfb572014-10-09 15:28:52 -0700261 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
263 lru_add_drain();
Michal Hockoaabfb572014-10-09 15:28:52 -0700264 for (i = 0; i < nr; i++)
265 free_swap_cache(pagep[i]);
266 release_pages(pagep, nr, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
268
269/*
270 * Lookup a swap entry in the swap cache. A found page will be returned
271 * unlocked and with its refcount incremented - we rely on the kernel
272 * lock getting page table operations atomic even if we drop the page
273 * lock before returning.
274 */
275struct page * lookup_swap_cache(swp_entry_t entry)
276{
277 struct page *page;
278
Shaohua Li33806f02013-02-22 16:34:37 -0800279 page = find_get_page(swap_address_space(entry), entry.val);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280
Shaohua Li579f8292014-02-06 12:04:21 -0800281 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 INC_CACHE_INFO(find_success);
Shaohua Li579f8292014-02-06 12:04:21 -0800283 if (TestClearPageReadahead(page))
284 atomic_inc(&swapin_readahead_hits);
285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287 INC_CACHE_INFO(find_total);
288 return page;
289}
290
291/*
292 * Locate a page of swap in physical memory, reserving swap cache space
293 * and reading the disk if it is not already cached.
294 * A failure return means that either the page allocation failed or that
295 * the swap entry is no longer in use.
296 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800297struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 struct vm_area_struct *vma, unsigned long addr)
299{
300 struct page *found_page, *new_page = NULL;
301 int err;
302
303 do {
304 /*
305 * First check the swap cache. Since this is normally
306 * called after lookup_swap_cache() failed, re-calling
307 * that would confuse statistics.
308 */
Shaohua Li33806f02013-02-22 16:34:37 -0800309 found_page = find_get_page(swap_address_space(entry),
310 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. */
Nick Pigginf45840b2008-10-18 20:26:57 -0700360 __set_page_locked(new_page);
Rik van Rielb2e18532008-10-18 20:26:30 -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);
Minchan Kimaca8bf32009-06-16 15:33:02 -0700369 swap_readpage(new_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 return new_page;
371 }
Daisuke Nishimura31a56392009-09-21 17:02:50 -0700372 radix_tree_preload_end();
Rik van Rielb2e18532008-10-18 20:26:30 -0700373 ClearPageSwapBacked(new_page);
Nick Pigginf45840b2008-10-18 20:26:57 -0700374 __clear_page_locked(new_page);
Daisuke Nishimura2ca45322009-09-21 17:02:52 -0700375 /*
376 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
377 * clear SWAP_HAS_CACHE flag.
378 */
Johannes Weiner0a31bc92014-08-08 14:19:22 -0700379 swapcache_free(entry);
Hugh Dickinsf0009442008-02-04 22:28:49 -0800380 } while (err != -ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
382 if (new_page)
383 page_cache_release(new_page);
384 return found_page;
385}
Hugh Dickins46017e92008-02-04 22:28:41 -0800386
Shaohua Li579f8292014-02-06 12:04:21 -0800387static unsigned long swapin_nr_pages(unsigned long offset)
388{
389 static unsigned long prev_offset;
390 unsigned int pages, max_pages, last_ra;
391 static atomic_t last_readahead_pages;
392
Jason Low4db0c3c2015-04-15 16:14:08 -0700393 max_pages = 1 << READ_ONCE(page_cluster);
Shaohua Li579f8292014-02-06 12:04:21 -0800394 if (max_pages <= 1)
395 return 1;
396
397 /*
398 * This heuristic has been found to work well on both sequential and
399 * random loads, swapping to hard disk or to SSD: please don't ask
400 * what the "+ 2" means, it just happens to work well, that's all.
401 */
402 pages = atomic_xchg(&swapin_readahead_hits, 0) + 2;
403 if (pages == 2) {
404 /*
405 * We can have no readahead hits to judge by: but must not get
406 * stuck here forever, so check for an adjacent offset instead
407 * (and don't even bother to check whether swap type is same).
408 */
409 if (offset != prev_offset + 1 && offset != prev_offset - 1)
410 pages = 1;
411 prev_offset = offset;
412 } else {
413 unsigned int roundup = 4;
414 while (roundup < pages)
415 roundup <<= 1;
416 pages = roundup;
417 }
418
419 if (pages > max_pages)
420 pages = max_pages;
421
422 /* Don't shrink readahead too fast */
423 last_ra = atomic_read(&last_readahead_pages) / 2;
424 if (pages < last_ra)
425 pages = last_ra;
426 atomic_set(&last_readahead_pages, pages);
427
428 return pages;
429}
430
Hugh Dickins46017e92008-02-04 22:28:41 -0800431/**
432 * swapin_readahead - swap in pages in hope we need them soon
433 * @entry: swap entry of this memory
Randy Dunlap76824862008-03-19 17:00:40 -0700434 * @gfp_mask: memory allocation flags
Hugh Dickins46017e92008-02-04 22:28:41 -0800435 * @vma: user vma this address belongs to
436 * @addr: target address for mempolicy
437 *
438 * Returns the struct page for entry and addr, after queueing swapin.
439 *
440 * Primitive swap readahead code. We simply read an aligned block of
441 * (1 << page_cluster) entries in the swap area. This method is chosen
442 * because it doesn't cost us any seek time. We also make sure to queue
443 * the 'original' request together with the readahead ones...
444 *
445 * This has been extended to use the NUMA policies from the mm triggering
446 * the readahead.
447 *
448 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
449 */
Hugh Dickins02098fe2008-02-04 22:28:42 -0800450struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
Hugh Dickins46017e92008-02-04 22:28:41 -0800451 struct vm_area_struct *vma, unsigned long addr)
452{
Hugh Dickins46017e92008-02-04 22:28:41 -0800453 struct page *page;
Shaohua Li579f8292014-02-06 12:04:21 -0800454 unsigned long entry_offset = swp_offset(entry);
455 unsigned long offset = entry_offset;
Rik van Riel67f96aa2012-03-21 16:33:50 -0700456 unsigned long start_offset, end_offset;
Shaohua Li579f8292014-02-06 12:04:21 -0800457 unsigned long mask;
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700458 struct blk_plug plug;
Hugh Dickins46017e92008-02-04 22:28:41 -0800459
Shaohua Li579f8292014-02-06 12:04:21 -0800460 mask = swapin_nr_pages(offset) - 1;
461 if (!mask)
462 goto skip;
463
Rik van Riel67f96aa2012-03-21 16:33:50 -0700464 /* Read a page_cluster sized and aligned cluster around offset. */
465 start_offset = offset & ~mask;
466 end_offset = offset | mask;
467 if (!start_offset) /* First page is swap header. */
468 start_offset++;
469
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700470 blk_start_plug(&plug);
Rik van Riel67f96aa2012-03-21 16:33:50 -0700471 for (offset = start_offset; offset <= end_offset ; offset++) {
Hugh Dickins46017e92008-02-04 22:28:41 -0800472 /* Ok, do the async read-ahead now */
473 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
Hugh Dickins02098fe2008-02-04 22:28:42 -0800474 gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800475 if (!page)
Rik van Riel67f96aa2012-03-21 16:33:50 -0700476 continue;
Shaohua Li579f8292014-02-06 12:04:21 -0800477 if (offset != entry_offset)
478 SetPageReadahead(page);
Hugh Dickins46017e92008-02-04 22:28:41 -0800479 page_cache_release(page);
480 }
Christian Ehrhardt3fb5c292012-07-31 16:41:44 -0700481 blk_finish_plug(&plug);
482
Hugh Dickins46017e92008-02-04 22:28:41 -0800483 lru_add_drain(); /* Push any new pages onto the LRU now */
Shaohua Li579f8292014-02-06 12:04:21 -0800484skip:
Hugh Dickins02098fe2008-02-04 22:28:42 -0800485 return read_swap_cache_async(entry, gfp_mask, vma, addr);
Hugh Dickins46017e92008-02-04 22:28:41 -0800486}