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
10 #include <linux/gfp.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/backing-dev.h>
17 #include <linux/pagevec.h>
18 #include <linux/migrate.h>
19 #include <linux/page_cgroup.h>
21 #include <asm/pgtable.h>
24 * swapper_space is a fiction, retained to simplify the path through
25 * vmscan's shrink_page_list.
27 static const struct address_space_operations swap_aops = {
28 .writepage = swap_writepage,
29 .set_page_dirty = __set_page_dirty_nobuffers,
30 .migratepage = migrate_page,
33 static struct backing_dev_info swap_backing_dev_info = {
35 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
38 struct address_space swapper_space = {
39 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
40 .tree_lock = __SPIN_LOCK_UNLOCKED(swapper_space.tree_lock),
42 .i_mmap_nonlinear = LIST_HEAD_INIT(swapper_space.i_mmap_nonlinear),
43 .backing_dev_info = &swap_backing_dev_info,
46 #define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
49 unsigned long add_total;
50 unsigned long del_total;
51 unsigned long find_success;
52 unsigned long find_total;
55 void show_swap_cache_info(void)
57 printk("%lu pages in swap cache\n", total_swapcache_pages);
58 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
59 swap_cache_info.add_total, swap_cache_info.del_total,
60 swap_cache_info.find_success, swap_cache_info.find_total);
61 printk("Free swap = %ldkB\n", nr_swap_pages << (PAGE_SHIFT - 10));
62 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
66 * __add_to_swap_cache resembles add_to_page_cache_locked on swapper_space,
67 * but sets SwapCache flag and private instead of mapping and index.
69 static int __add_to_swap_cache(struct page *page, swp_entry_t entry)
73 VM_BUG_ON(!PageLocked(page));
74 VM_BUG_ON(PageSwapCache(page));
75 VM_BUG_ON(!PageSwapBacked(page));
78 SetPageSwapCache(page);
79 set_page_private(page, entry.val);
81 spin_lock_irq(&swapper_space.tree_lock);
82 error = radix_tree_insert(&swapper_space.page_tree, entry.val, page);
84 total_swapcache_pages++;
85 __inc_zone_page_state(page, NR_FILE_PAGES);
86 INC_CACHE_INFO(add_total);
88 spin_unlock_irq(&swapper_space.tree_lock);
90 if (unlikely(error)) {
92 * Only the context which have set SWAP_HAS_CACHE flag
93 * would call add_to_swap_cache().
94 * So add_to_swap_cache() doesn't returns -EEXIST.
96 VM_BUG_ON(error == -EEXIST);
97 set_page_private(page, 0UL);
98 ClearPageSwapCache(page);
99 page_cache_release(page);
106 int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
110 error = radix_tree_preload(gfp_mask);
112 error = __add_to_swap_cache(page, entry);
113 radix_tree_preload_end();
119 * This must be called only on pages that have
120 * been verified to be in the swap cache.
122 void __delete_from_swap_cache(struct page *page)
124 VM_BUG_ON(!PageLocked(page));
125 VM_BUG_ON(!PageSwapCache(page));
126 VM_BUG_ON(PageWriteback(page));
128 radix_tree_delete(&swapper_space.page_tree, page_private(page));
129 set_page_private(page, 0);
130 ClearPageSwapCache(page);
131 total_swapcache_pages--;
132 __dec_zone_page_state(page, NR_FILE_PAGES);
133 INC_CACHE_INFO(del_total);
137 * add_to_swap - allocate swap space for a page
138 * @page: page we want to move to swap
140 * Allocate swap space for the page and add the page to the
141 * swap cache. Caller needs to hold the page lock.
143 int add_to_swap(struct page *page)
148 VM_BUG_ON(!PageLocked(page));
149 VM_BUG_ON(!PageUptodate(page));
151 entry = get_swap_page();
155 if (unlikely(PageTransHuge(page)))
156 if (unlikely(split_huge_page(page))) {
157 swapcache_free(entry, NULL);
162 * Radix-tree node allocations from PF_MEMALLOC contexts could
163 * completely exhaust the page allocator. __GFP_NOMEMALLOC
164 * stops emergency reserves from being allocated.
166 * TODO: this could cause a theoretical memory reclaim
167 * deadlock in the swap out path.
170 * Add it to the swap cache and mark it dirty
172 err = add_to_swap_cache(page, entry,
173 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
175 if (!err) { /* Success */
178 } else { /* -ENOMEM radix-tree allocation failure */
180 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
181 * clear SWAP_HAS_CACHE flag.
183 swapcache_free(entry, NULL);
189 * This must be called only on pages that have
190 * been verified to be in the swap cache and locked.
191 * It will never put the page into the free list,
192 * the caller has a reference on the page.
194 void delete_from_swap_cache(struct page *page)
198 entry.val = page_private(page);
200 spin_lock_irq(&swapper_space.tree_lock);
201 __delete_from_swap_cache(page);
202 spin_unlock_irq(&swapper_space.tree_lock);
204 swapcache_free(entry, page);
205 page_cache_release(page);
209 * If we are the only user, then try to free up the swap cache.
211 * Its ok to check for PageSwapCache without the page lock
212 * here because we are going to recheck again inside
213 * try_to_free_swap() _with_ the lock.
216 static inline void free_swap_cache(struct page *page)
218 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
219 try_to_free_swap(page);
225 * Perform a free_page(), also freeing any swap cache associated with
226 * this page if it is the last user of the page.
228 void free_page_and_swap_cache(struct page *page)
230 free_swap_cache(page);
231 page_cache_release(page);
235 * Passed an array of pages, drop them all from swapcache and then release
236 * them. They are removed from the LRU and freed if this is their last use.
238 void free_pages_and_swap_cache(struct page **pages, int nr)
240 struct page **pagep = pages;
244 int todo = min(nr, PAGEVEC_SIZE);
247 for (i = 0; i < todo; i++)
248 free_swap_cache(pagep[i]);
249 release_pages(pagep, todo, 0);
256 * Lookup a swap entry in the swap cache. A found page will be returned
257 * unlocked and with its refcount incremented - we rely on the kernel
258 * lock getting page table operations atomic even if we drop the page
259 * lock before returning.
261 struct page * lookup_swap_cache(swp_entry_t entry)
265 page = find_get_page(&swapper_space, entry.val);
268 INC_CACHE_INFO(find_success);
270 INC_CACHE_INFO(find_total);
275 * Locate a page of swap in physical memory, reserving swap cache space
276 * and reading the disk if it is not already cached.
277 * A failure return means that either the page allocation failed or that
278 * the swap entry is no longer in use.
280 struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
281 struct vm_area_struct *vma, unsigned long addr)
283 struct page *found_page, *new_page = NULL;
288 * First check the swap cache. Since this is normally
289 * called after lookup_swap_cache() failed, re-calling
290 * that would confuse statistics.
292 found_page = find_get_page(&swapper_space, entry.val);
297 * Get a new page to read into from swap.
300 new_page = alloc_page_vma(gfp_mask, vma, addr);
302 break; /* Out of memory */
306 * call radix_tree_preload() while we can wait.
308 err = radix_tree_preload(gfp_mask & GFP_KERNEL);
313 * Swap entry may have been freed since our caller observed it.
315 err = swapcache_prepare(entry);
316 if (err == -EEXIST) { /* seems racy */
317 radix_tree_preload_end();
320 if (err) { /* swp entry is obsolete ? */
321 radix_tree_preload_end();
325 /* May fail (-ENOMEM) if radix-tree node allocation failed. */
326 __set_page_locked(new_page);
327 SetPageSwapBacked(new_page);
328 err = __add_to_swap_cache(new_page, entry);
330 radix_tree_preload_end();
332 * Initiate read into locked page and return.
334 lru_cache_add_anon(new_page);
335 swap_readpage(new_page);
338 radix_tree_preload_end();
339 ClearPageSwapBacked(new_page);
340 __clear_page_locked(new_page);
342 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
343 * clear SWAP_HAS_CACHE flag.
345 swapcache_free(entry, NULL);
346 } while (err != -ENOMEM);
349 page_cache_release(new_page);
354 * swapin_readahead - swap in pages in hope we need them soon
355 * @entry: swap entry of this memory
356 * @gfp_mask: memory allocation flags
357 * @vma: user vma this address belongs to
358 * @addr: target address for mempolicy
360 * Returns the struct page for entry and addr, after queueing swapin.
362 * Primitive swap readahead code. We simply read an aligned block of
363 * (1 << page_cluster) entries in the swap area. This method is chosen
364 * because it doesn't cost us any seek time. We also make sure to queue
365 * the 'original' request together with the readahead ones...
367 * This has been extended to use the NUMA policies from the mm triggering
370 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
372 struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
373 struct vm_area_struct *vma, unsigned long addr)
377 unsigned long offset;
378 unsigned long end_offset;
381 * Get starting offset for readaround, and number of pages to read.
382 * Adjust starting address by readbehind (for NUMA interleave case)?
383 * No, it's very unlikely that swap layout would follow vma layout,
384 * more likely that neighbouring swap pages came from the same node:
385 * so use the same "addr" to choose the same node for each swap read.
387 nr_pages = valid_swaphandles(entry, &offset);
388 for (end_offset = offset + nr_pages; offset < end_offset; offset++) {
389 /* Ok, do the async read-ahead now */
390 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
391 gfp_mask, vma, addr);
394 page_cache_release(page);
396 lru_add_drain(); /* Push any new pages onto the LRU now */
397 return read_swap_cache_async(entry, gfp_mask, vma, addr);