1
2
3
4
5
6
7
8
9#include <linux/mm.h>
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/blkdev.h>
18#include <linux/pagevec.h>
19#include <linux/migrate.h>
20
21#include <asm/pgtable.h>
22
23
24
25
26
27static const struct address_space_operations swap_aops = {
28 .writepage = swap_writepage,
29 .set_page_dirty = swap_set_page_dirty,
30#ifdef CONFIG_MIGRATION
31 .migratepage = migrate_page,
32#endif
33};
34
35struct address_space swapper_spaces[MAX_SWAPFILES] = {
36 [0 ... MAX_SWAPFILES - 1] = {
37 .page_tree = RADIX_TREE_INIT(GFP_ATOMIC|__GFP_NOWARN),
38 .i_mmap_writable = ATOMIC_INIT(0),
39 .a_ops = &swap_aops,
40
41 .flags = 1 << AS_NO_WRITEBACK_TAGS,
42 }
43};
44
45#define INC_CACHE_INFO(x) do { swap_cache_info.x++; } while (0)
46
47static struct {
48 unsigned long add_total;
49 unsigned long del_total;
50 unsigned long find_success;
51 unsigned long find_total;
52} swap_cache_info;
53
54unsigned long total_swapcache_pages(void)
55{
56 int i;
57 unsigned long ret = 0;
58
59 for (i = 0; i < MAX_SWAPFILES; i++)
60 ret += swapper_spaces[i].nrpages;
61 return ret;
62}
63
64static atomic_t swapin_readahead_hits = ATOMIC_INIT(4);
65
66void show_swap_cache_info(void)
67{
68 printk("%lu pages in swap cache\n", total_swapcache_pages());
69 printk("Swap cache stats: add %lu, delete %lu, find %lu/%lu\n",
70 swap_cache_info.add_total, swap_cache_info.del_total,
71 swap_cache_info.find_success, swap_cache_info.find_total);
72 printk("Free swap = %ldkB\n",
73 get_nr_swap_pages() << (PAGE_SHIFT - 10));
74 printk("Total swap = %lukB\n", total_swap_pages << (PAGE_SHIFT - 10));
75}
76
77
78
79
80
81int __add_to_swap_cache(struct page *page, swp_entry_t entry)
82{
83 int error;
84 struct address_space *address_space;
85
86 VM_BUG_ON_PAGE(!PageLocked(page), page);
87 VM_BUG_ON_PAGE(PageSwapCache(page), page);
88 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
89
90 get_page(page);
91 SetPageSwapCache(page);
92 set_page_private(page, entry.val);
93
94 address_space = swap_address_space(entry);
95 spin_lock_irq(&address_space->tree_lock);
96 error = radix_tree_insert(&address_space->page_tree,
97 swp_offset(entry), page);
98 if (likely(!error)) {
99 address_space->nrpages++;
100 __inc_node_page_state(page, NR_FILE_PAGES);
101 INC_CACHE_INFO(add_total);
102 }
103 spin_unlock_irq(&address_space->tree_lock);
104
105 if (unlikely(error)) {
106
107
108
109
110
111 VM_BUG_ON(error == -EEXIST);
112 set_page_private(page, 0UL);
113 ClearPageSwapCache(page);
114 put_page(page);
115 }
116
117 return error;
118}
119
120
121int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
122{
123 int error;
124
125 error = radix_tree_maybe_preload(gfp_mask);
126 if (!error) {
127 error = __add_to_swap_cache(page, entry);
128 radix_tree_preload_end();
129 }
130 return error;
131}
132
133
134
135
136
137void __delete_from_swap_cache(struct page *page)
138{
139 swp_entry_t entry;
140 struct address_space *address_space;
141
142 VM_BUG_ON_PAGE(!PageLocked(page), page);
143 VM_BUG_ON_PAGE(!PageSwapCache(page), page);
144 VM_BUG_ON_PAGE(PageWriteback(page), page);
145
146 entry.val = page_private(page);
147 address_space = swap_address_space(entry);
148 radix_tree_delete(&address_space->page_tree, swp_offset(entry));
149 set_page_private(page, 0);
150 ClearPageSwapCache(page);
151 address_space->nrpages--;
152 __dec_node_page_state(page, NR_FILE_PAGES);
153 INC_CACHE_INFO(del_total);
154}
155
156
157
158
159
160
161
162
163int add_to_swap(struct page *page, struct list_head *list)
164{
165 swp_entry_t entry;
166 int err;
167
168 VM_BUG_ON_PAGE(!PageLocked(page), page);
169 VM_BUG_ON_PAGE(!PageUptodate(page), page);
170
171 entry = get_swap_page();
172 if (!entry.val)
173 return 0;
174
175 if (mem_cgroup_try_charge_swap(page, entry)) {
176 swapcache_free(entry);
177 return 0;
178 }
179
180 if (unlikely(PageTransHuge(page)))
181 if (unlikely(split_huge_page_to_list(page, list))) {
182 swapcache_free(entry);
183 return 0;
184 }
185
186
187
188
189
190
191
192
193
194
195
196
197 err = add_to_swap_cache(page, entry,
198 __GFP_HIGH|__GFP_NOMEMALLOC|__GFP_NOWARN);
199
200 if (!err) {
201 return 1;
202 } else {
203
204
205
206
207 swapcache_free(entry);
208 return 0;
209 }
210}
211
212
213
214
215
216
217
218void delete_from_swap_cache(struct page *page)
219{
220 swp_entry_t entry;
221 struct address_space *address_space;
222
223 entry.val = page_private(page);
224
225 address_space = swap_address_space(entry);
226 spin_lock_irq(&address_space->tree_lock);
227 __delete_from_swap_cache(page);
228 spin_unlock_irq(&address_space->tree_lock);
229
230 swapcache_free(entry);
231 put_page(page);
232}
233
234
235
236
237
238
239
240
241
242static inline void free_swap_cache(struct page *page)
243{
244 if (PageSwapCache(page) && !page_mapped(page) && trylock_page(page)) {
245 try_to_free_swap(page);
246 unlock_page(page);
247 }
248}
249
250
251
252
253
254void free_page_and_swap_cache(struct page *page)
255{
256 free_swap_cache(page);
257 if (!is_huge_zero_page(page))
258 put_page(page);
259}
260
261
262
263
264
265void free_pages_and_swap_cache(struct page **pages, int nr)
266{
267 struct page **pagep = pages;
268 int i;
269
270 lru_add_drain();
271 for (i = 0; i < nr; i++)
272 free_swap_cache(pagep[i]);
273 release_pages(pagep, nr, false);
274}
275
276
277
278
279
280
281
282struct page * lookup_swap_cache(swp_entry_t entry)
283{
284 struct page *page;
285
286 page = find_get_page(swap_address_space(entry), swp_offset(entry));
287
288 if (page) {
289 INC_CACHE_INFO(find_success);
290 if (TestClearPageReadahead(page))
291 atomic_inc(&swapin_readahead_hits);
292 }
293
294 INC_CACHE_INFO(find_total);
295 return page;
296}
297
298struct page *__read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
299 struct vm_area_struct *vma, unsigned long addr,
300 bool *new_page_allocated)
301{
302 struct page *found_page, *new_page = NULL;
303 struct address_space *swapper_space = swap_address_space(entry);
304 int err;
305 *new_page_allocated = false;
306
307 do {
308
309
310
311
312
313 found_page = find_get_page(swapper_space, swp_offset(entry));
314 if (found_page)
315 break;
316
317
318
319
320 if (!new_page) {
321 new_page = alloc_page_vma(gfp_mask, vma, addr);
322 if (!new_page)
323 break;
324 }
325
326
327
328
329 err = radix_tree_maybe_preload(gfp_mask & GFP_KERNEL);
330 if (err)
331 break;
332
333
334
335
336 err = swapcache_prepare(entry);
337 if (err == -EEXIST) {
338 radix_tree_preload_end();
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354 cond_resched();
355 continue;
356 }
357 if (err) {
358 radix_tree_preload_end();
359 break;
360 }
361
362
363 __SetPageLocked(new_page);
364 __SetPageSwapBacked(new_page);
365 err = __add_to_swap_cache(new_page, entry);
366 if (likely(!err)) {
367 radix_tree_preload_end();
368
369
370
371 lru_cache_add_anon(new_page);
372 *new_page_allocated = true;
373 return new_page;
374 }
375 radix_tree_preload_end();
376 __ClearPageLocked(new_page);
377
378
379
380
381 swapcache_free(entry);
382 } while (err != -ENOMEM);
383
384 if (new_page)
385 put_page(new_page);
386 return found_page;
387}
388
389
390
391
392
393
394
395struct page *read_swap_cache_async(swp_entry_t entry, gfp_t gfp_mask,
396 struct vm_area_struct *vma, unsigned long addr)
397{
398 bool page_was_allocated;
399 struct page *retpage = __read_swap_cache_async(entry, gfp_mask,
400 vma, addr, &page_was_allocated);
401
402 if (page_was_allocated)
403 swap_readpage(retpage);
404
405 return retpage;
406}
407
408static unsigned long swapin_nr_pages(unsigned long offset)
409{
410 static unsigned long prev_offset;
411 unsigned int pages, max_pages, last_ra;
412 static atomic_t last_readahead_pages;
413
414 max_pages = 1 << READ_ONCE(page_cluster);
415 if (max_pages <= 1)
416 return 1;
417
418
419
420
421
422
423 pages = atomic_xchg(&swapin_readahead_hits, 0) + 2;
424 if (pages == 2) {
425
426
427
428
429
430 if (offset != prev_offset + 1 && offset != prev_offset - 1)
431 pages = 1;
432 prev_offset = offset;
433 } else {
434 unsigned int roundup = 4;
435 while (roundup < pages)
436 roundup <<= 1;
437 pages = roundup;
438 }
439
440 if (pages > max_pages)
441 pages = max_pages;
442
443
444 last_ra = atomic_read(&last_readahead_pages) / 2;
445 if (pages < last_ra)
446 pages = last_ra;
447 atomic_set(&last_readahead_pages, pages);
448
449 return pages;
450}
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471struct page *swapin_readahead(swp_entry_t entry, gfp_t gfp_mask,
472 struct vm_area_struct *vma, unsigned long addr)
473{
474 struct page *page;
475 unsigned long entry_offset = swp_offset(entry);
476 unsigned long offset = entry_offset;
477 unsigned long start_offset, end_offset;
478 unsigned long mask;
479 struct blk_plug plug;
480
481 mask = swapin_nr_pages(offset) - 1;
482 if (!mask)
483 goto skip;
484
485
486 start_offset = offset & ~mask;
487 end_offset = offset | mask;
488 if (!start_offset)
489 start_offset++;
490
491 blk_start_plug(&plug);
492 for (offset = start_offset; offset <= end_offset ; offset++) {
493
494 page = read_swap_cache_async(swp_entry(swp_type(entry), offset),
495 gfp_mask, vma, addr);
496 if (!page)
497 continue;
498 if (offset != entry_offset)
499 SetPageReadahead(page);
500 put_page(page);
501 }
502 blk_finish_plug(&plug);
503
504 lru_add_drain();
505skip:
506 return read_swap_cache_async(entry, gfp_mask, vma, addr);
507}
508