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