2 * linux/mm/swap_state.c
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 * Swap reorganised 29.12.95, Stephen Tweedie
7 * Rewritten to use page cache, (C) 1998 Stephen Tweedie
9 #include <linux/module.h>
11 #include <linux/kernel_stat.h>
12 #include <linux/swap.h>
13 #include <linux/swapops.h>
14 #include <linux/init.h>
15 #include <linux/pagemap.h>
16 #include <linux/buffer_head.h>
17 #include <linux/backing-dev.h>
18 #include <linux/pagevec.h>
19 #include <linux/migrate.h>
20 #include <linux/memcontrol.h>
22 #include <asm/pgtable.h>
25 * swapper_space is a fiction, retained to simplify the path through
26 * vmscan's shrink_page_list, to make sync_page look nicer, and to allow
27 * future use of radix_tree tags in the swap cache.
29 static const struct address_space_operations swap_aops
= {
30 .writepage
= swap_writepage
,
31 .sync_page
= block_sync_page
,
32 .set_page_dirty
= __set_page_dirty_nobuffers
,
33 .migratepage
= migrate_page
,
36 static struct backing_dev_info swap_backing_dev_info
= {
37 .capabilities
= BDI_CAP_NO_ACCT_DIRTY
| BDI_CAP_NO_WRITEBACK
,
38 .unplug_io_fn
= swap_unplug_io_fn
,
41 struct address_space swapper_space
= {
42 .page_tree
= RADIX_TREE_INIT(GFP_ATOMIC
|__GFP_NOWARN
),
43 .tree_lock
= __RW_LOCK_UNLOCKED(swapper_space
.tree_lock
),
45 .i_mmap_nonlinear
= LIST_HEAD_INIT(swapper_space
.i_mmap_nonlinear
),
46 .backing_dev_info
= &swap_backing_dev_info
,
49 #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
52 unsigned long add_total
;
53 unsigned long del_total
;
54 unsigned long find_success
;
55 unsigned long find_total
;
58 void show_swap_cache_info(void)
60 printk("Swap cache: add %lu, delete %lu, find %lu/%lu\n",
61 swap_cache_info
.add_total
, swap_cache_info
.del_total
,
62 swap_cache_info
.find_success
, swap_cache_info
.find_total
);
63 printk("Free swap = %lukB\n", nr_swap_pages
<< (PAGE_SHIFT
- 10));
64 printk("Total swap = %lukB\n", total_swap_pages
<< (PAGE_SHIFT
- 10));
68 * add_to_swap_cache resembles add_to_page_cache on swapper_space,
69 * but sets SwapCache flag and private instead of mapping and index.
71 int add_to_swap_cache(struct page
*page
, swp_entry_t entry
, gfp_t gfp_mask
)
75 BUG_ON(!PageLocked(page
));
76 BUG_ON(PageSwapCache(page
));
77 BUG_ON(PagePrivate(page
));
78 error
= radix_tree_preload(gfp_mask
);
81 error
= mem_cgroup_charge(page
, current
->mm
);
85 write_lock_irq(&swapper_space
.tree_lock
);
86 error
= radix_tree_insert(&swapper_space
.page_tree
,
90 SetPageSwapCache(page
);
91 set_page_private(page
, entry
.val
);
92 total_swapcache_pages
++;
93 __inc_zone_page_state(page
, NR_FILE_PAGES
);
94 INC_CACHE_INFO(add_total
);
96 mem_cgroup_uncharge_page(page
);
98 write_unlock_irq(&swapper_space
.tree_lock
);
99 radix_tree_preload_end();
106 * This must be called only on pages that have
107 * been verified to be in the swap cache.
109 void __delete_from_swap_cache(struct page
*page
)
111 BUG_ON(!PageLocked(page
));
112 BUG_ON(!PageSwapCache(page
));
113 BUG_ON(PageWriteback(page
));
114 BUG_ON(PagePrivate(page
));
116 mem_cgroup_uncharge_page(page
);
117 radix_tree_delete(&swapper_space
.page_tree
, page_private(page
));
118 set_page_private(page
, 0);
119 ClearPageSwapCache(page
);
120 total_swapcache_pages
--;
121 __dec_zone_page_state(page
, NR_FILE_PAGES
);
122 INC_CACHE_INFO(del_total
);
126 * add_to_swap - allocate swap space for a page
127 * @page: page we want to move to swap
129 * Allocate swap space for the page and add the page to the
130 * swap cache. Caller needs to hold the page lock.
132 int add_to_swap(struct page
* page
, gfp_t gfp_mask
)
137 BUG_ON(!PageLocked(page
));
138 BUG_ON(!PageUptodate(page
));
141 entry
= get_swap_page();
146 * Radix-tree node allocations from PF_MEMALLOC contexts could
147 * completely exhaust the page allocator. __GFP_NOMEMALLOC
148 * stops emergency reserves from being allocated.
150 * TODO: this could cause a theoretical memory reclaim
151 * deadlock in the swap out path.
154 * Add it to the swap cache and mark it dirty
156 err
= add_to_swap_cache(page
, entry
,
157 gfp_mask
|__GFP_NOMEMALLOC
|__GFP_NOWARN
);
160 case 0: /* Success */
164 /* Raced with "speculative" read_swap_cache_async */
168 /* -ENOMEM radix-tree allocation failure */
176 * This must be called only on pages that have
177 * been verified to be in the swap cache and locked.
178 * It will never put the page into the free list,
179 * the caller has a reference on the page.
181 void delete_from_swap_cache(struct page
*page
)
185 entry
.val
= page_private(page
);
187 write_lock_irq(&swapper_space
.tree_lock
);
188 __delete_from_swap_cache(page
);
189 write_unlock_irq(&swapper_space
.tree_lock
);
192 page_cache_release(page
);
196 * If we are the only user, then try to free up the swap cache.
198 * Its ok to check for PageSwapCache without the page lock
199 * here because we are going to recheck again inside
200 * exclusive_swap_page() _with_ the lock.
203 static inline void free_swap_cache(struct page
*page
)
205 if (PageSwapCache(page
) && !TestSetPageLocked(page
)) {
206 remove_exclusive_swap_page(page
);
212 * Perform a free_page(), also freeing any swap cache associated with
213 * this page if it is the last user of the page.
215 void free_page_and_swap_cache(struct page
*page
)
217 free_swap_cache(page
);
218 page_cache_release(page
);
222 * Passed an array of pages, drop them all from swapcache and then release
223 * them. They are removed from the LRU and freed if this is their last use.
225 void free_pages_and_swap_cache(struct page
**pages
, int nr
)
227 struct page
**pagep
= pages
;
231 int todo
= min(nr
, PAGEVEC_SIZE
);
234 for (i
= 0; i
< todo
; i
++)
235 free_swap_cache(pagep
[i
]);
236 release_pages(pagep
, todo
, 0);
243 * Lookup a swap entry in the swap cache. A found page will be returned
244 * unlocked and with its refcount incremented - we rely on the kernel
245 * lock getting page table operations atomic even if we drop the page
246 * lock before returning.
248 struct page
* lookup_swap_cache(swp_entry_t entry
)
252 page
= find_get_page(&swapper_space
, entry
.val
);
255 INC_CACHE_INFO(find_success
);
257 INC_CACHE_INFO(find_total
);
262 * Locate a page of swap in physical memory, reserving swap cache space
263 * and reading the disk if it is not already cached.
264 * A failure return means that either the page allocation failed or that
265 * the swap entry is no longer in use.
267 struct page
*read_swap_cache_async(swp_entry_t entry
, gfp_t gfp_mask
,
268 struct vm_area_struct
*vma
, unsigned long addr
)
270 struct page
*found_page
, *new_page
= NULL
;
275 * First check the swap cache. Since this is normally
276 * called after lookup_swap_cache() failed, re-calling
277 * that would confuse statistics.
279 found_page
= find_get_page(&swapper_space
, entry
.val
);
284 * Get a new page to read into from swap.
287 new_page
= alloc_page_vma(gfp_mask
, vma
, addr
);
289 break; /* Out of memory */
293 * Swap entry may have been freed since our caller observed it.
295 if (!swap_duplicate(entry
))
299 * Associate the page with swap entry in the swap cache.
300 * May fail (-EEXIST) if there is already a page associated
301 * with this entry in the swap cache: added by a racing
302 * read_swap_cache_async, or add_to_swap or shmem_writepage
303 * re-using the just freed swap entry for an existing page.
304 * May fail (-ENOMEM) if radix-tree node allocation failed.
306 SetPageLocked(new_page
);
307 err
= add_to_swap_cache(new_page
, entry
, gfp_mask
& GFP_KERNEL
);
310 * Initiate read into locked page and return.
312 lru_cache_add_active(new_page
);
313 swap_readpage(NULL
, new_page
);
316 ClearPageLocked(new_page
);
318 } while (err
!= -ENOMEM
);
321 page_cache_release(new_page
);
326 * swapin_readahead - swap in pages in hope we need them soon
327 * @entry: swap entry of this memory
328 * @vma: user vma this address belongs to
329 * @addr: target address for mempolicy
331 * Returns the struct page for entry and addr, after queueing swapin.
333 * Primitive swap readahead code. We simply read an aligned block of
334 * (1 << page_cluster) entries in the swap area. This method is chosen
335 * because it doesn't cost us any seek time. We also make sure to queue
336 * the 'original' request together with the readahead ones...
338 * This has been extended to use the NUMA policies from the mm triggering
341 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
343 struct page
*swapin_readahead(swp_entry_t entry
, gfp_t gfp_mask
,
344 struct vm_area_struct
*vma
, unsigned long addr
)
348 unsigned long offset
;
349 unsigned long end_offset
;
352 * Get starting offset for readaround, and number of pages to read.
353 * Adjust starting address by readbehind (for NUMA interleave case)?
354 * No, it's very unlikely that swap layout would follow vma layout,
355 * more likely that neighbouring swap pages came from the same node:
356 * so use the same "addr" to choose the same node for each swap read.
358 nr_pages
= valid_swaphandles(entry
, &offset
);
359 for (end_offset
= offset
+ nr_pages
; offset
< end_offset
; offset
++) {
360 /* Ok, do the async read-ahead now */
361 page
= read_swap_cache_async(swp_entry(swp_type(entry
), offset
),
362 gfp_mask
, vma
, addr
);
365 page_cache_release(page
);
367 lru_add_drain(); /* Push any new pages onto the LRU now */
368 return read_swap_cache_async(entry
, gfp_mask
, vma
, addr
);