]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/swap.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
183ff22b | 8 | * This file contains the default values for the operation of the |
1da177e4 LT |
9 | * Linux VM subsystem. Fine-tuning documentation can be found in |
10 | * Documentation/sysctl/vm.txt. | |
11 | * Started 18.12.91 | |
12 | * Swap aging added 23.2.95, Stephen Tweedie. | |
13 | * Buffermem limits added 12.3.98, Rik van Riel. | |
14 | */ | |
15 | ||
16 | #include <linux/mm.h> | |
17 | #include <linux/sched.h> | |
18 | #include <linux/kernel_stat.h> | |
19 | #include <linux/swap.h> | |
20 | #include <linux/mman.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/pagevec.h> | |
23 | #include <linux/init.h> | |
b95f1b31 | 24 | #include <linux/export.h> |
1da177e4 | 25 | #include <linux/mm_inline.h> |
1da177e4 | 26 | #include <linux/percpu_counter.h> |
3565fce3 | 27 | #include <linux/memremap.h> |
1da177e4 LT |
28 | #include <linux/percpu.h> |
29 | #include <linux/cpu.h> | |
30 | #include <linux/notifier.h> | |
e0bf68dd | 31 | #include <linux/backing-dev.h> |
66e1707b | 32 | #include <linux/memcontrol.h> |
5a0e3ad6 | 33 | #include <linux/gfp.h> |
a27bb332 | 34 | #include <linux/uio.h> |
822fc613 | 35 | #include <linux/hugetlb.h> |
33c3fc71 | 36 | #include <linux/page_idle.h> |
1da177e4 | 37 | |
64d6519d LS |
38 | #include "internal.h" |
39 | ||
c6286c98 MG |
40 | #define CREATE_TRACE_POINTS |
41 | #include <trace/events/pagemap.h> | |
42 | ||
1da177e4 LT |
43 | /* How many pages do we try to swap or page in/out together? */ |
44 | int page_cluster; | |
45 | ||
13f7f789 | 46 | static DEFINE_PER_CPU(struct pagevec, lru_add_pvec); |
f84f9504 | 47 | static DEFINE_PER_CPU(struct pagevec, lru_rotate_pvecs); |
cc5993bd | 48 | static DEFINE_PER_CPU(struct pagevec, lru_deactivate_file_pvecs); |
f7ad2a6c | 49 | static DEFINE_PER_CPU(struct pagevec, lru_lazyfree_pvecs); |
a4a921aa ML |
50 | #ifdef CONFIG_SMP |
51 | static DEFINE_PER_CPU(struct pagevec, activate_page_pvecs); | |
52 | #endif | |
902aaed0 | 53 | |
b221385b AB |
54 | /* |
55 | * This path almost never happens for VM activity - pages are normally | |
56 | * freed via pagevecs. But it gets used by networking. | |
57 | */ | |
920c7a5d | 58 | static void __page_cache_release(struct page *page) |
b221385b AB |
59 | { |
60 | if (PageLRU(page)) { | |
b221385b | 61 | struct zone *zone = page_zone(page); |
fa9add64 HD |
62 | struct lruvec *lruvec; |
63 | unsigned long flags; | |
b221385b | 64 | |
a52633d8 | 65 | spin_lock_irqsave(zone_lru_lock(zone), flags); |
599d0c95 | 66 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
309381fe | 67 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
b221385b | 68 | __ClearPageLRU(page); |
fa9add64 | 69 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
a52633d8 | 70 | spin_unlock_irqrestore(zone_lru_lock(zone), flags); |
b221385b | 71 | } |
62906027 | 72 | __ClearPageWaiters(page); |
0a31bc97 | 73 | mem_cgroup_uncharge(page); |
91807063 AA |
74 | } |
75 | ||
76 | static void __put_single_page(struct page *page) | |
77 | { | |
78 | __page_cache_release(page); | |
2d4894b5 | 79 | free_unref_page(page); |
b221385b AB |
80 | } |
81 | ||
91807063 | 82 | static void __put_compound_page(struct page *page) |
1da177e4 | 83 | { |
91807063 | 84 | compound_page_dtor *dtor; |
1da177e4 | 85 | |
822fc613 NH |
86 | /* |
87 | * __page_cache_release() is supposed to be called for thp, not for | |
88 | * hugetlb. This is because hugetlb page does never have PageLRU set | |
89 | * (it's never listed to any LRU lists) and no memcg routines should | |
90 | * be called for hugetlb (it has a separate hugetlb_cgroup.) | |
91 | */ | |
92 | if (!PageHuge(page)) | |
93 | __page_cache_release(page); | |
91807063 AA |
94 | dtor = get_compound_page_dtor(page); |
95 | (*dtor)(page); | |
96 | } | |
97 | ||
ddc58f27 | 98 | void __put_page(struct page *page) |
8519fb30 | 99 | { |
71389703 DW |
100 | if (is_zone_device_page(page)) { |
101 | put_dev_pagemap(page->pgmap); | |
102 | ||
103 | /* | |
104 | * The page belongs to the device that created pgmap. Do | |
105 | * not return it to page allocator. | |
106 | */ | |
107 | return; | |
108 | } | |
109 | ||
8519fb30 | 110 | if (unlikely(PageCompound(page))) |
ddc58f27 KS |
111 | __put_compound_page(page); |
112 | else | |
91807063 | 113 | __put_single_page(page); |
1da177e4 | 114 | } |
ddc58f27 | 115 | EXPORT_SYMBOL(__put_page); |
70b50f94 | 116 | |
1d7ea732 | 117 | /** |
7682486b RD |
118 | * put_pages_list() - release a list of pages |
119 | * @pages: list of pages threaded on page->lru | |
1d7ea732 AZ |
120 | * |
121 | * Release a list of pages which are strung together on page.lru. Currently | |
122 | * used by read_cache_pages() and related error recovery code. | |
1d7ea732 AZ |
123 | */ |
124 | void put_pages_list(struct list_head *pages) | |
125 | { | |
126 | while (!list_empty(pages)) { | |
127 | struct page *victim; | |
128 | ||
129 | victim = list_entry(pages->prev, struct page, lru); | |
130 | list_del(&victim->lru); | |
09cbfeaf | 131 | put_page(victim); |
1d7ea732 AZ |
132 | } |
133 | } | |
134 | EXPORT_SYMBOL(put_pages_list); | |
135 | ||
18022c5d MG |
136 | /* |
137 | * get_kernel_pages() - pin kernel pages in memory | |
138 | * @kiov: An array of struct kvec structures | |
139 | * @nr_segs: number of segments to pin | |
140 | * @write: pinning for read/write, currently ignored | |
141 | * @pages: array that receives pointers to the pages pinned. | |
142 | * Should be at least nr_segs long. | |
143 | * | |
144 | * Returns number of pages pinned. This may be fewer than the number | |
145 | * requested. If nr_pages is 0 or negative, returns 0. If no pages | |
146 | * were pinned, returns -errno. Each page returned must be released | |
147 | * with a put_page() call when it is finished with. | |
148 | */ | |
149 | int get_kernel_pages(const struct kvec *kiov, int nr_segs, int write, | |
150 | struct page **pages) | |
151 | { | |
152 | int seg; | |
153 | ||
154 | for (seg = 0; seg < nr_segs; seg++) { | |
155 | if (WARN_ON(kiov[seg].iov_len != PAGE_SIZE)) | |
156 | return seg; | |
157 | ||
5a178119 | 158 | pages[seg] = kmap_to_page(kiov[seg].iov_base); |
09cbfeaf | 159 | get_page(pages[seg]); |
18022c5d MG |
160 | } |
161 | ||
162 | return seg; | |
163 | } | |
164 | EXPORT_SYMBOL_GPL(get_kernel_pages); | |
165 | ||
166 | /* | |
167 | * get_kernel_page() - pin a kernel page in memory | |
168 | * @start: starting kernel address | |
169 | * @write: pinning for read/write, currently ignored | |
170 | * @pages: array that receives pointer to the page pinned. | |
171 | * Must be at least nr_segs long. | |
172 | * | |
173 | * Returns 1 if page is pinned. If the page was not pinned, returns | |
174 | * -errno. The page returned must be released with a put_page() call | |
175 | * when it is finished with. | |
176 | */ | |
177 | int get_kernel_page(unsigned long start, int write, struct page **pages) | |
178 | { | |
179 | const struct kvec kiov = { | |
180 | .iov_base = (void *)start, | |
181 | .iov_len = PAGE_SIZE | |
182 | }; | |
183 | ||
184 | return get_kernel_pages(&kiov, 1, write, pages); | |
185 | } | |
186 | EXPORT_SYMBOL_GPL(get_kernel_page); | |
187 | ||
3dd7ae8e | 188 | static void pagevec_lru_move_fn(struct pagevec *pvec, |
fa9add64 HD |
189 | void (*move_fn)(struct page *page, struct lruvec *lruvec, void *arg), |
190 | void *arg) | |
902aaed0 HH |
191 | { |
192 | int i; | |
68eb0731 | 193 | struct pglist_data *pgdat = NULL; |
fa9add64 | 194 | struct lruvec *lruvec; |
3dd7ae8e | 195 | unsigned long flags = 0; |
902aaed0 HH |
196 | |
197 | for (i = 0; i < pagevec_count(pvec); i++) { | |
198 | struct page *page = pvec->pages[i]; | |
68eb0731 | 199 | struct pglist_data *pagepgdat = page_pgdat(page); |
902aaed0 | 200 | |
68eb0731 MG |
201 | if (pagepgdat != pgdat) { |
202 | if (pgdat) | |
203 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); | |
204 | pgdat = pagepgdat; | |
205 | spin_lock_irqsave(&pgdat->lru_lock, flags); | |
902aaed0 | 206 | } |
3dd7ae8e | 207 | |
68eb0731 | 208 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
fa9add64 | 209 | (*move_fn)(page, lruvec, arg); |
902aaed0 | 210 | } |
68eb0731 MG |
211 | if (pgdat) |
212 | spin_unlock_irqrestore(&pgdat->lru_lock, flags); | |
c6f92f9f | 213 | release_pages(pvec->pages, pvec->nr); |
83896fb5 | 214 | pagevec_reinit(pvec); |
d8505dee SL |
215 | } |
216 | ||
fa9add64 HD |
217 | static void pagevec_move_tail_fn(struct page *page, struct lruvec *lruvec, |
218 | void *arg) | |
3dd7ae8e SL |
219 | { |
220 | int *pgmoved = arg; | |
3dd7ae8e | 221 | |
c55e8d03 JW |
222 | if (PageLRU(page) && !PageUnevictable(page)) { |
223 | del_page_from_lru_list(page, lruvec, page_lru(page)); | |
224 | ClearPageActive(page); | |
225 | add_page_to_lru_list_tail(page, lruvec, page_lru(page)); | |
3dd7ae8e SL |
226 | (*pgmoved)++; |
227 | } | |
228 | } | |
229 | ||
230 | /* | |
231 | * pagevec_move_tail() must be called with IRQ disabled. | |
232 | * Otherwise this may cause nasty races. | |
233 | */ | |
234 | static void pagevec_move_tail(struct pagevec *pvec) | |
235 | { | |
236 | int pgmoved = 0; | |
237 | ||
238 | pagevec_lru_move_fn(pvec, pagevec_move_tail_fn, &pgmoved); | |
239 | __count_vm_events(PGROTATED, pgmoved); | |
240 | } | |
241 | ||
1da177e4 LT |
242 | /* |
243 | * Writeback is about to end against a page which has been marked for immediate | |
244 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
902aaed0 | 245 | * inactive list. |
1da177e4 | 246 | */ |
3dd7ae8e | 247 | void rotate_reclaimable_page(struct page *page) |
1da177e4 | 248 | { |
c55e8d03 | 249 | if (!PageLocked(page) && !PageDirty(page) && |
894bc310 | 250 | !PageUnevictable(page) && PageLRU(page)) { |
ac6aadb2 MS |
251 | struct pagevec *pvec; |
252 | unsigned long flags; | |
253 | ||
09cbfeaf | 254 | get_page(page); |
ac6aadb2 | 255 | local_irq_save(flags); |
7c8e0181 | 256 | pvec = this_cpu_ptr(&lru_rotate_pvecs); |
8f182270 | 257 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
ac6aadb2 MS |
258 | pagevec_move_tail(pvec); |
259 | local_irq_restore(flags); | |
260 | } | |
1da177e4 LT |
261 | } |
262 | ||
fa9add64 | 263 | static void update_page_reclaim_stat(struct lruvec *lruvec, |
3e2f41f1 KM |
264 | int file, int rotated) |
265 | { | |
fa9add64 | 266 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
3e2f41f1 KM |
267 | |
268 | reclaim_stat->recent_scanned[file]++; | |
269 | if (rotated) | |
270 | reclaim_stat->recent_rotated[file]++; | |
3e2f41f1 KM |
271 | } |
272 | ||
fa9add64 HD |
273 | static void __activate_page(struct page *page, struct lruvec *lruvec, |
274 | void *arg) | |
1da177e4 | 275 | { |
744ed144 | 276 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
7a608572 LT |
277 | int file = page_is_file_cache(page); |
278 | int lru = page_lru_base_type(page); | |
744ed144 | 279 | |
fa9add64 | 280 | del_page_from_lru_list(page, lruvec, lru); |
7a608572 LT |
281 | SetPageActive(page); |
282 | lru += LRU_ACTIVE; | |
fa9add64 | 283 | add_page_to_lru_list(page, lruvec, lru); |
24b7e581 | 284 | trace_mm_lru_activate(page); |
4f98a2fe | 285 | |
fa9add64 HD |
286 | __count_vm_event(PGACTIVATE); |
287 | update_page_reclaim_stat(lruvec, file, 1); | |
1da177e4 | 288 | } |
eb709b0d SL |
289 | } |
290 | ||
291 | #ifdef CONFIG_SMP | |
eb709b0d SL |
292 | static void activate_page_drain(int cpu) |
293 | { | |
294 | struct pagevec *pvec = &per_cpu(activate_page_pvecs, cpu); | |
295 | ||
296 | if (pagevec_count(pvec)) | |
297 | pagevec_lru_move_fn(pvec, __activate_page, NULL); | |
298 | } | |
299 | ||
5fbc4616 CM |
300 | static bool need_activate_page_drain(int cpu) |
301 | { | |
302 | return pagevec_count(&per_cpu(activate_page_pvecs, cpu)) != 0; | |
303 | } | |
304 | ||
eb709b0d SL |
305 | void activate_page(struct page *page) |
306 | { | |
800d8c63 | 307 | page = compound_head(page); |
eb709b0d SL |
308 | if (PageLRU(page) && !PageActive(page) && !PageUnevictable(page)) { |
309 | struct pagevec *pvec = &get_cpu_var(activate_page_pvecs); | |
310 | ||
09cbfeaf | 311 | get_page(page); |
8f182270 | 312 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
eb709b0d SL |
313 | pagevec_lru_move_fn(pvec, __activate_page, NULL); |
314 | put_cpu_var(activate_page_pvecs); | |
315 | } | |
316 | } | |
317 | ||
318 | #else | |
319 | static inline void activate_page_drain(int cpu) | |
320 | { | |
321 | } | |
322 | ||
5fbc4616 CM |
323 | static bool need_activate_page_drain(int cpu) |
324 | { | |
325 | return false; | |
326 | } | |
327 | ||
eb709b0d SL |
328 | void activate_page(struct page *page) |
329 | { | |
330 | struct zone *zone = page_zone(page); | |
331 | ||
800d8c63 | 332 | page = compound_head(page); |
a52633d8 | 333 | spin_lock_irq(zone_lru_lock(zone)); |
599d0c95 | 334 | __activate_page(page, mem_cgroup_page_lruvec(page, zone->zone_pgdat), NULL); |
a52633d8 | 335 | spin_unlock_irq(zone_lru_lock(zone)); |
1da177e4 | 336 | } |
eb709b0d | 337 | #endif |
1da177e4 | 338 | |
059285a2 MG |
339 | static void __lru_cache_activate_page(struct page *page) |
340 | { | |
341 | struct pagevec *pvec = &get_cpu_var(lru_add_pvec); | |
342 | int i; | |
343 | ||
344 | /* | |
345 | * Search backwards on the optimistic assumption that the page being | |
346 | * activated has just been added to this pagevec. Note that only | |
347 | * the local pagevec is examined as a !PageLRU page could be in the | |
348 | * process of being released, reclaimed, migrated or on a remote | |
349 | * pagevec that is currently being drained. Furthermore, marking | |
350 | * a remote pagevec's page PageActive potentially hits a race where | |
351 | * a page is marked PageActive just after it is added to the inactive | |
352 | * list causing accounting errors and BUG_ON checks to trigger. | |
353 | */ | |
354 | for (i = pagevec_count(pvec) - 1; i >= 0; i--) { | |
355 | struct page *pagevec_page = pvec->pages[i]; | |
356 | ||
357 | if (pagevec_page == page) { | |
358 | SetPageActive(page); | |
359 | break; | |
360 | } | |
361 | } | |
362 | ||
363 | put_cpu_var(lru_add_pvec); | |
364 | } | |
365 | ||
1da177e4 LT |
366 | /* |
367 | * Mark a page as having seen activity. | |
368 | * | |
369 | * inactive,unreferenced -> inactive,referenced | |
370 | * inactive,referenced -> active,unreferenced | |
371 | * active,unreferenced -> active,referenced | |
eb39d618 HD |
372 | * |
373 | * When a newly allocated page is not yet visible, so safe for non-atomic ops, | |
374 | * __SetPageReferenced(page) may be substituted for mark_page_accessed(page). | |
1da177e4 | 375 | */ |
920c7a5d | 376 | void mark_page_accessed(struct page *page) |
1da177e4 | 377 | { |
e90309c9 | 378 | page = compound_head(page); |
894bc310 | 379 | if (!PageActive(page) && !PageUnevictable(page) && |
059285a2 MG |
380 | PageReferenced(page)) { |
381 | ||
382 | /* | |
383 | * If the page is on the LRU, queue it for activation via | |
384 | * activate_page_pvecs. Otherwise, assume the page is on a | |
385 | * pagevec, mark it active and it'll be moved to the active | |
386 | * LRU on the next drain. | |
387 | */ | |
388 | if (PageLRU(page)) | |
389 | activate_page(page); | |
390 | else | |
391 | __lru_cache_activate_page(page); | |
1da177e4 | 392 | ClearPageReferenced(page); |
a528910e JW |
393 | if (page_is_file_cache(page)) |
394 | workingset_activation(page); | |
1da177e4 LT |
395 | } else if (!PageReferenced(page)) { |
396 | SetPageReferenced(page); | |
397 | } | |
33c3fc71 VD |
398 | if (page_is_idle(page)) |
399 | clear_page_idle(page); | |
1da177e4 | 400 | } |
1da177e4 LT |
401 | EXPORT_SYMBOL(mark_page_accessed); |
402 | ||
2329d375 | 403 | static void __lru_cache_add(struct page *page) |
1da177e4 | 404 | { |
13f7f789 MG |
405 | struct pagevec *pvec = &get_cpu_var(lru_add_pvec); |
406 | ||
09cbfeaf | 407 | get_page(page); |
8f182270 | 408 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
a0b8cab3 | 409 | __pagevec_lru_add(pvec); |
13f7f789 | 410 | put_cpu_var(lru_add_pvec); |
1da177e4 | 411 | } |
2329d375 JZ |
412 | |
413 | /** | |
e02a9f04 | 414 | * lru_cache_add_anon - add a page to the page lists |
2329d375 JZ |
415 | * @page: the page to add |
416 | */ | |
417 | void lru_cache_add_anon(struct page *page) | |
418 | { | |
6fb81a17 MG |
419 | if (PageActive(page)) |
420 | ClearPageActive(page); | |
2329d375 JZ |
421 | __lru_cache_add(page); |
422 | } | |
423 | ||
424 | void lru_cache_add_file(struct page *page) | |
425 | { | |
6fb81a17 MG |
426 | if (PageActive(page)) |
427 | ClearPageActive(page); | |
2329d375 JZ |
428 | __lru_cache_add(page); |
429 | } | |
430 | EXPORT_SYMBOL(lru_cache_add_file); | |
1da177e4 | 431 | |
f04e9ebb | 432 | /** |
c53954a0 | 433 | * lru_cache_add - add a page to a page list |
f04e9ebb | 434 | * @page: the page to be added to the LRU. |
2329d375 JZ |
435 | * |
436 | * Queue the page for addition to the LRU via pagevec. The decision on whether | |
437 | * to add the page to the [in]active [file|anon] list is deferred until the | |
438 | * pagevec is drained. This gives a chance for the caller of lru_cache_add() | |
439 | * have the page added to the active list using mark_page_accessed(). | |
f04e9ebb | 440 | */ |
c53954a0 | 441 | void lru_cache_add(struct page *page) |
1da177e4 | 442 | { |
309381fe SL |
443 | VM_BUG_ON_PAGE(PageActive(page) && PageUnevictable(page), page); |
444 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
c53954a0 | 445 | __lru_cache_add(page); |
1da177e4 LT |
446 | } |
447 | ||
00501b53 JW |
448 | /** |
449 | * lru_cache_add_active_or_unevictable | |
450 | * @page: the page to be added to LRU | |
451 | * @vma: vma in which page is mapped for determining reclaimability | |
452 | * | |
453 | * Place @page on the active or unevictable LRU list, depending on its | |
454 | * evictability. Note that if the page is not evictable, it goes | |
455 | * directly back onto it's zone's unevictable list, it does NOT use a | |
456 | * per cpu pagevec. | |
457 | */ | |
458 | void lru_cache_add_active_or_unevictable(struct page *page, | |
459 | struct vm_area_struct *vma) | |
460 | { | |
461 | VM_BUG_ON_PAGE(PageLRU(page), page); | |
462 | ||
9c4e6b1a | 463 | if (likely((vma->vm_flags & (VM_LOCKED | VM_SPECIAL)) != VM_LOCKED)) |
00501b53 | 464 | SetPageActive(page); |
9c4e6b1a | 465 | else if (!TestSetPageMlocked(page)) { |
00501b53 JW |
466 | /* |
467 | * We use the irq-unsafe __mod_zone_page_stat because this | |
468 | * counter is not modified from interrupt context, and the pte | |
469 | * lock is held(spinlock), which implies preemption disabled. | |
470 | */ | |
471 | __mod_zone_page_state(page_zone(page), NR_MLOCK, | |
472 | hpage_nr_pages(page)); | |
473 | count_vm_event(UNEVICTABLE_PGMLOCKED); | |
474 | } | |
9c4e6b1a | 475 | lru_cache_add(page); |
00501b53 JW |
476 | } |
477 | ||
31560180 MK |
478 | /* |
479 | * If the page can not be invalidated, it is moved to the | |
480 | * inactive list to speed up its reclaim. It is moved to the | |
481 | * head of the list, rather than the tail, to give the flusher | |
482 | * threads some time to write it out, as this is much more | |
483 | * effective than the single-page writeout from reclaim. | |
278df9f4 MK |
484 | * |
485 | * If the page isn't page_mapped and dirty/writeback, the page | |
486 | * could reclaim asap using PG_reclaim. | |
487 | * | |
488 | * 1. active, mapped page -> none | |
489 | * 2. active, dirty/writeback page -> inactive, head, PG_reclaim | |
490 | * 3. inactive, mapped page -> none | |
491 | * 4. inactive, dirty/writeback page -> inactive, head, PG_reclaim | |
492 | * 5. inactive, clean -> inactive, tail | |
493 | * 6. Others -> none | |
494 | * | |
495 | * In 4, why it moves inactive's head, the VM expects the page would | |
496 | * be write it out by flusher threads as this is much more effective | |
497 | * than the single-page writeout from reclaim. | |
31560180 | 498 | */ |
cc5993bd | 499 | static void lru_deactivate_file_fn(struct page *page, struct lruvec *lruvec, |
fa9add64 | 500 | void *arg) |
31560180 MK |
501 | { |
502 | int lru, file; | |
278df9f4 | 503 | bool active; |
31560180 | 504 | |
278df9f4 | 505 | if (!PageLRU(page)) |
31560180 MK |
506 | return; |
507 | ||
bad49d9c MK |
508 | if (PageUnevictable(page)) |
509 | return; | |
510 | ||
31560180 MK |
511 | /* Some processes are using the page */ |
512 | if (page_mapped(page)) | |
513 | return; | |
514 | ||
278df9f4 | 515 | active = PageActive(page); |
31560180 MK |
516 | file = page_is_file_cache(page); |
517 | lru = page_lru_base_type(page); | |
fa9add64 HD |
518 | |
519 | del_page_from_lru_list(page, lruvec, lru + active); | |
31560180 MK |
520 | ClearPageActive(page); |
521 | ClearPageReferenced(page); | |
fa9add64 | 522 | add_page_to_lru_list(page, lruvec, lru); |
31560180 | 523 | |
278df9f4 MK |
524 | if (PageWriteback(page) || PageDirty(page)) { |
525 | /* | |
526 | * PG_reclaim could be raced with end_page_writeback | |
527 | * It can make readahead confusing. But race window | |
528 | * is _really_ small and it's non-critical problem. | |
529 | */ | |
530 | SetPageReclaim(page); | |
531 | } else { | |
532 | /* | |
533 | * The page's writeback ends up during pagevec | |
534 | * We moves tha page into tail of inactive. | |
535 | */ | |
925b7673 | 536 | list_move_tail(&page->lru, &lruvec->lists[lru]); |
278df9f4 MK |
537 | __count_vm_event(PGROTATED); |
538 | } | |
539 | ||
540 | if (active) | |
541 | __count_vm_event(PGDEACTIVATE); | |
fa9add64 | 542 | update_page_reclaim_stat(lruvec, file, 0); |
31560180 MK |
543 | } |
544 | ||
10853a03 | 545 | |
f7ad2a6c | 546 | static void lru_lazyfree_fn(struct page *page, struct lruvec *lruvec, |
10853a03 MK |
547 | void *arg) |
548 | { | |
f7ad2a6c | 549 | if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 550 | !PageSwapCache(page) && !PageUnevictable(page)) { |
f7ad2a6c | 551 | bool active = PageActive(page); |
10853a03 | 552 | |
f7ad2a6c SL |
553 | del_page_from_lru_list(page, lruvec, |
554 | LRU_INACTIVE_ANON + active); | |
10853a03 MK |
555 | ClearPageActive(page); |
556 | ClearPageReferenced(page); | |
f7ad2a6c SL |
557 | /* |
558 | * lazyfree pages are clean anonymous pages. They have | |
559 | * SwapBacked flag cleared to distinguish normal anonymous | |
560 | * pages | |
561 | */ | |
562 | ClearPageSwapBacked(page); | |
563 | add_page_to_lru_list(page, lruvec, LRU_INACTIVE_FILE); | |
10853a03 | 564 | |
f7ad2a6c | 565 | __count_vm_events(PGLAZYFREE, hpage_nr_pages(page)); |
2262185c | 566 | count_memcg_page_event(page, PGLAZYFREE); |
f7ad2a6c | 567 | update_page_reclaim_stat(lruvec, 1, 0); |
10853a03 MK |
568 | } |
569 | } | |
570 | ||
902aaed0 HH |
571 | /* |
572 | * Drain pages out of the cpu's pagevecs. | |
573 | * Either "cpu" is the current CPU, and preemption has already been | |
574 | * disabled; or "cpu" is being hot-unplugged, and is already dead. | |
575 | */ | |
f0cb3c76 | 576 | void lru_add_drain_cpu(int cpu) |
1da177e4 | 577 | { |
13f7f789 | 578 | struct pagevec *pvec = &per_cpu(lru_add_pvec, cpu); |
1da177e4 | 579 | |
13f7f789 | 580 | if (pagevec_count(pvec)) |
a0b8cab3 | 581 | __pagevec_lru_add(pvec); |
902aaed0 HH |
582 | |
583 | pvec = &per_cpu(lru_rotate_pvecs, cpu); | |
584 | if (pagevec_count(pvec)) { | |
585 | unsigned long flags; | |
586 | ||
587 | /* No harm done if a racing interrupt already did this */ | |
588 | local_irq_save(flags); | |
589 | pagevec_move_tail(pvec); | |
590 | local_irq_restore(flags); | |
591 | } | |
31560180 | 592 | |
cc5993bd | 593 | pvec = &per_cpu(lru_deactivate_file_pvecs, cpu); |
31560180 | 594 | if (pagevec_count(pvec)) |
cc5993bd | 595 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); |
eb709b0d | 596 | |
f7ad2a6c | 597 | pvec = &per_cpu(lru_lazyfree_pvecs, cpu); |
10853a03 | 598 | if (pagevec_count(pvec)) |
f7ad2a6c | 599 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL); |
10853a03 | 600 | |
eb709b0d | 601 | activate_page_drain(cpu); |
31560180 MK |
602 | } |
603 | ||
604 | /** | |
cc5993bd | 605 | * deactivate_file_page - forcefully deactivate a file page |
31560180 MK |
606 | * @page: page to deactivate |
607 | * | |
608 | * This function hints the VM that @page is a good reclaim candidate, | |
609 | * for example if its invalidation fails due to the page being dirty | |
610 | * or under writeback. | |
611 | */ | |
cc5993bd | 612 | void deactivate_file_page(struct page *page) |
31560180 | 613 | { |
821ed6bb | 614 | /* |
cc5993bd MK |
615 | * In a workload with many unevictable page such as mprotect, |
616 | * unevictable page deactivation for accelerating reclaim is pointless. | |
821ed6bb MK |
617 | */ |
618 | if (PageUnevictable(page)) | |
619 | return; | |
620 | ||
31560180 | 621 | if (likely(get_page_unless_zero(page))) { |
cc5993bd | 622 | struct pagevec *pvec = &get_cpu_var(lru_deactivate_file_pvecs); |
31560180 | 623 | |
8f182270 | 624 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
cc5993bd MK |
625 | pagevec_lru_move_fn(pvec, lru_deactivate_file_fn, NULL); |
626 | put_cpu_var(lru_deactivate_file_pvecs); | |
31560180 | 627 | } |
80bfed90 AM |
628 | } |
629 | ||
10853a03 | 630 | /** |
f7ad2a6c | 631 | * mark_page_lazyfree - make an anon page lazyfree |
10853a03 MK |
632 | * @page: page to deactivate |
633 | * | |
f7ad2a6c SL |
634 | * mark_page_lazyfree() moves @page to the inactive file list. |
635 | * This is done to accelerate the reclaim of @page. | |
10853a03 | 636 | */ |
f7ad2a6c | 637 | void mark_page_lazyfree(struct page *page) |
10853a03 | 638 | { |
f7ad2a6c | 639 | if (PageLRU(page) && PageAnon(page) && PageSwapBacked(page) && |
24c92eb7 | 640 | !PageSwapCache(page) && !PageUnevictable(page)) { |
f7ad2a6c | 641 | struct pagevec *pvec = &get_cpu_var(lru_lazyfree_pvecs); |
10853a03 | 642 | |
09cbfeaf | 643 | get_page(page); |
8f182270 | 644 | if (!pagevec_add(pvec, page) || PageCompound(page)) |
f7ad2a6c SL |
645 | pagevec_lru_move_fn(pvec, lru_lazyfree_fn, NULL); |
646 | put_cpu_var(lru_lazyfree_pvecs); | |
10853a03 MK |
647 | } |
648 | } | |
649 | ||
80bfed90 AM |
650 | void lru_add_drain(void) |
651 | { | |
f0cb3c76 | 652 | lru_add_drain_cpu(get_cpu()); |
80bfed90 | 653 | put_cpu(); |
1da177e4 LT |
654 | } |
655 | ||
c4028958 | 656 | static void lru_add_drain_per_cpu(struct work_struct *dummy) |
053837fc NP |
657 | { |
658 | lru_add_drain(); | |
659 | } | |
660 | ||
5fbc4616 CM |
661 | static DEFINE_PER_CPU(struct work_struct, lru_add_drain_work); |
662 | ||
9852a721 MH |
663 | /* |
664 | * Doesn't need any cpu hotplug locking because we do rely on per-cpu | |
665 | * kworkers being shut down before our page_alloc_cpu_dead callback is | |
666 | * executed on the offlined cpu. | |
667 | * Calling this function with cpu hotplug locks held can actually lead | |
668 | * to obscure indirect dependencies via WQ context. | |
669 | */ | |
670 | void lru_add_drain_all(void) | |
053837fc | 671 | { |
5fbc4616 CM |
672 | static DEFINE_MUTEX(lock); |
673 | static struct cpumask has_work; | |
674 | int cpu; | |
675 | ||
ce612879 MH |
676 | /* |
677 | * Make sure nobody triggers this path before mm_percpu_wq is fully | |
678 | * initialized. | |
679 | */ | |
680 | if (WARN_ON(!mm_percpu_wq)) | |
681 | return; | |
682 | ||
5fbc4616 | 683 | mutex_lock(&lock); |
5fbc4616 CM |
684 | cpumask_clear(&has_work); |
685 | ||
686 | for_each_online_cpu(cpu) { | |
687 | struct work_struct *work = &per_cpu(lru_add_drain_work, cpu); | |
688 | ||
689 | if (pagevec_count(&per_cpu(lru_add_pvec, cpu)) || | |
690 | pagevec_count(&per_cpu(lru_rotate_pvecs, cpu)) || | |
cc5993bd | 691 | pagevec_count(&per_cpu(lru_deactivate_file_pvecs, cpu)) || |
f7ad2a6c | 692 | pagevec_count(&per_cpu(lru_lazyfree_pvecs, cpu)) || |
5fbc4616 CM |
693 | need_activate_page_drain(cpu)) { |
694 | INIT_WORK(work, lru_add_drain_per_cpu); | |
ce612879 | 695 | queue_work_on(cpu, mm_percpu_wq, work); |
5fbc4616 CM |
696 | cpumask_set_cpu(cpu, &has_work); |
697 | } | |
698 | } | |
699 | ||
700 | for_each_cpu(cpu, &has_work) | |
701 | flush_work(&per_cpu(lru_add_drain_work, cpu)); | |
702 | ||
5fbc4616 | 703 | mutex_unlock(&lock); |
053837fc NP |
704 | } |
705 | ||
aabfb572 | 706 | /** |
ea1754a0 | 707 | * release_pages - batched put_page() |
aabfb572 MH |
708 | * @pages: array of pages to release |
709 | * @nr: number of pages | |
1da177e4 | 710 | * |
aabfb572 MH |
711 | * Decrement the reference count on all the pages in @pages. If it |
712 | * fell to zero, remove the page from the LRU and free it. | |
1da177e4 | 713 | */ |
c6f92f9f | 714 | void release_pages(struct page **pages, int nr) |
1da177e4 LT |
715 | { |
716 | int i; | |
cc59850e | 717 | LIST_HEAD(pages_to_free); |
599d0c95 | 718 | struct pglist_data *locked_pgdat = NULL; |
fa9add64 | 719 | struct lruvec *lruvec; |
902aaed0 | 720 | unsigned long uninitialized_var(flags); |
aabfb572 | 721 | unsigned int uninitialized_var(lock_batch); |
1da177e4 | 722 | |
1da177e4 LT |
723 | for (i = 0; i < nr; i++) { |
724 | struct page *page = pages[i]; | |
1da177e4 | 725 | |
aabfb572 MH |
726 | /* |
727 | * Make sure the IRQ-safe lock-holding time does not get | |
728 | * excessive with a continuous string of pages from the | |
599d0c95 | 729 | * same pgdat. The lock is held only if pgdat != NULL. |
aabfb572 | 730 | */ |
599d0c95 MG |
731 | if (locked_pgdat && ++lock_batch == SWAP_CLUSTER_MAX) { |
732 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
733 | locked_pgdat = NULL; | |
aabfb572 MH |
734 | } |
735 | ||
6fcb52a5 | 736 | if (is_huge_zero_page(page)) |
aa88b68c | 737 | continue; |
aa88b68c | 738 | |
df6ad698 JG |
739 | /* Device public page can not be huge page */ |
740 | if (is_device_public_page(page)) { | |
741 | if (locked_pgdat) { | |
742 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, | |
743 | flags); | |
744 | locked_pgdat = NULL; | |
745 | } | |
e7638488 | 746 | put_devmap_managed_page(page); |
df6ad698 JG |
747 | continue; |
748 | } | |
749 | ||
ddc58f27 | 750 | page = compound_head(page); |
b5810039 | 751 | if (!put_page_testzero(page)) |
1da177e4 LT |
752 | continue; |
753 | ||
ddc58f27 | 754 | if (PageCompound(page)) { |
599d0c95 MG |
755 | if (locked_pgdat) { |
756 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
757 | locked_pgdat = NULL; | |
ddc58f27 KS |
758 | } |
759 | __put_compound_page(page); | |
760 | continue; | |
761 | } | |
762 | ||
46453a6e | 763 | if (PageLRU(page)) { |
599d0c95 | 764 | struct pglist_data *pgdat = page_pgdat(page); |
894bc310 | 765 | |
599d0c95 MG |
766 | if (pgdat != locked_pgdat) { |
767 | if (locked_pgdat) | |
768 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, | |
902aaed0 | 769 | flags); |
aabfb572 | 770 | lock_batch = 0; |
599d0c95 MG |
771 | locked_pgdat = pgdat; |
772 | spin_lock_irqsave(&locked_pgdat->lru_lock, flags); | |
46453a6e | 773 | } |
fa9add64 | 774 | |
599d0c95 | 775 | lruvec = mem_cgroup_page_lruvec(page, locked_pgdat); |
309381fe | 776 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
67453911 | 777 | __ClearPageLRU(page); |
fa9add64 | 778 | del_page_from_lru_list(page, lruvec, page_off_lru(page)); |
46453a6e NP |
779 | } |
780 | ||
c53954a0 | 781 | /* Clear Active bit in case of parallel mark_page_accessed */ |
e3741b50 | 782 | __ClearPageActive(page); |
62906027 | 783 | __ClearPageWaiters(page); |
c53954a0 | 784 | |
cc59850e | 785 | list_add(&page->lru, &pages_to_free); |
1da177e4 | 786 | } |
599d0c95 MG |
787 | if (locked_pgdat) |
788 | spin_unlock_irqrestore(&locked_pgdat->lru_lock, flags); | |
1da177e4 | 789 | |
747db954 | 790 | mem_cgroup_uncharge_list(&pages_to_free); |
2d4894b5 | 791 | free_unref_page_list(&pages_to_free); |
1da177e4 | 792 | } |
0be8557b | 793 | EXPORT_SYMBOL(release_pages); |
1da177e4 LT |
794 | |
795 | /* | |
796 | * The pages which we're about to release may be in the deferred lru-addition | |
797 | * queues. That would prevent them from really being freed right now. That's | |
798 | * OK from a correctness point of view but is inefficient - those pages may be | |
799 | * cache-warm and we want to give them back to the page allocator ASAP. | |
800 | * | |
801 | * So __pagevec_release() will drain those queues here. __pagevec_lru_add() | |
802 | * and __pagevec_lru_add_active() call release_pages() directly to avoid | |
803 | * mutual recursion. | |
804 | */ | |
805 | void __pagevec_release(struct pagevec *pvec) | |
806 | { | |
7f0b5fb9 | 807 | if (!pvec->percpu_pvec_drained) { |
d9ed0d08 | 808 | lru_add_drain(); |
7f0b5fb9 | 809 | pvec->percpu_pvec_drained = true; |
d9ed0d08 | 810 | } |
c6f92f9f | 811 | release_pages(pvec->pages, pagevec_count(pvec)); |
1da177e4 LT |
812 | pagevec_reinit(pvec); |
813 | } | |
7f285701 SF |
814 | EXPORT_SYMBOL(__pagevec_release); |
815 | ||
12d27107 | 816 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
71e3aac0 | 817 | /* used by __split_huge_page_refcount() */ |
fa9add64 | 818 | void lru_add_page_tail(struct page *page, struct page *page_tail, |
5bc7b8ac | 819 | struct lruvec *lruvec, struct list_head *list) |
71e3aac0 | 820 | { |
71e3aac0 | 821 | const int file = 0; |
71e3aac0 | 822 | |
309381fe SL |
823 | VM_BUG_ON_PAGE(!PageHead(page), page); |
824 | VM_BUG_ON_PAGE(PageCompound(page_tail), page); | |
825 | VM_BUG_ON_PAGE(PageLRU(page_tail), page); | |
35f3aa39 | 826 | lockdep_assert_held(&lruvec_pgdat(lruvec)->lru_lock); |
71e3aac0 | 827 | |
5bc7b8ac SL |
828 | if (!list) |
829 | SetPageLRU(page_tail); | |
71e3aac0 | 830 | |
12d27107 HD |
831 | if (likely(PageLRU(page))) |
832 | list_add_tail(&page_tail->lru, &page->lru); | |
5bc7b8ac SL |
833 | else if (list) { |
834 | /* page reclaim is reclaiming a huge page */ | |
835 | get_page(page_tail); | |
836 | list_add_tail(&page_tail->lru, list); | |
837 | } else { | |
12d27107 HD |
838 | struct list_head *list_head; |
839 | /* | |
840 | * Head page has not yet been counted, as an hpage, | |
841 | * so we must account for each subpage individually. | |
842 | * | |
843 | * Use the standard add function to put page_tail on the list, | |
844 | * but then correct its position so they all end up in order. | |
845 | */ | |
e180cf80 | 846 | add_page_to_lru_list(page_tail, lruvec, page_lru(page_tail)); |
12d27107 HD |
847 | list_head = page_tail->lru.prev; |
848 | list_move_tail(&page_tail->lru, list_head); | |
71e3aac0 | 849 | } |
7512102c HD |
850 | |
851 | if (!PageUnevictable(page)) | |
e180cf80 | 852 | update_page_reclaim_stat(lruvec, file, PageActive(page_tail)); |
71e3aac0 | 853 | } |
12d27107 | 854 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
71e3aac0 | 855 | |
fa9add64 HD |
856 | static void __pagevec_lru_add_fn(struct page *page, struct lruvec *lruvec, |
857 | void *arg) | |
3dd7ae8e | 858 | { |
9c4e6b1a SB |
859 | enum lru_list lru; |
860 | int was_unevictable = TestClearPageUnevictable(page); | |
3dd7ae8e | 861 | |
309381fe | 862 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3dd7ae8e SL |
863 | |
864 | SetPageLRU(page); | |
9c4e6b1a SB |
865 | /* |
866 | * Page becomes evictable in two ways: | |
867 | * 1) Within LRU lock [munlock_vma_pages() and __munlock_pagevec()]. | |
868 | * 2) Before acquiring LRU lock to put the page to correct LRU and then | |
869 | * a) do PageLRU check with lock [check_move_unevictable_pages] | |
870 | * b) do PageLRU check before lock [clear_page_mlock] | |
871 | * | |
872 | * (1) & (2a) are ok as LRU lock will serialize them. For (2b), we need | |
873 | * following strict ordering: | |
874 | * | |
875 | * #0: __pagevec_lru_add_fn #1: clear_page_mlock | |
876 | * | |
877 | * SetPageLRU() TestClearPageMlocked() | |
878 | * smp_mb() // explicit ordering // above provides strict | |
879 | * // ordering | |
880 | * PageMlocked() PageLRU() | |
881 | * | |
882 | * | |
883 | * if '#1' does not observe setting of PG_lru by '#0' and fails | |
884 | * isolation, the explicit barrier will make sure that page_evictable | |
885 | * check will put the page in correct LRU. Without smp_mb(), SetPageLRU | |
886 | * can be reordered after PageMlocked check and can make '#1' to fail | |
887 | * the isolation of the page whose Mlocked bit is cleared (#0 is also | |
888 | * looking at the same page) and the evictable page will be stranded | |
889 | * in an unevictable LRU. | |
890 | */ | |
891 | smp_mb(); | |
892 | ||
893 | if (page_evictable(page)) { | |
894 | lru = page_lru(page); | |
895 | update_page_reclaim_stat(lruvec, page_is_file_cache(page), | |
896 | PageActive(page)); | |
897 | if (was_unevictable) | |
898 | count_vm_event(UNEVICTABLE_PGRESCUED); | |
899 | } else { | |
900 | lru = LRU_UNEVICTABLE; | |
901 | ClearPageActive(page); | |
902 | SetPageUnevictable(page); | |
903 | if (!was_unevictable) | |
904 | count_vm_event(UNEVICTABLE_PGCULLED); | |
905 | } | |
906 | ||
fa9add64 | 907 | add_page_to_lru_list(page, lruvec, lru); |
24b7e581 | 908 | trace_mm_lru_insertion(page, lru); |
3dd7ae8e SL |
909 | } |
910 | ||
1da177e4 LT |
911 | /* |
912 | * Add the passed pages to the LRU, then drop the caller's refcount | |
913 | * on them. Reinitialises the caller's pagevec. | |
914 | */ | |
a0b8cab3 | 915 | void __pagevec_lru_add(struct pagevec *pvec) |
1da177e4 | 916 | { |
a0b8cab3 | 917 | pagevec_lru_move_fn(pvec, __pagevec_lru_add_fn, NULL); |
1da177e4 | 918 | } |
5095ae83 | 919 | EXPORT_SYMBOL(__pagevec_lru_add); |
1da177e4 | 920 | |
0cd6144a JW |
921 | /** |
922 | * pagevec_lookup_entries - gang pagecache lookup | |
923 | * @pvec: Where the resulting entries are placed | |
924 | * @mapping: The address_space to search | |
925 | * @start: The starting entry index | |
cb6f0f34 | 926 | * @nr_entries: The maximum number of pages |
0cd6144a JW |
927 | * @indices: The cache indices corresponding to the entries in @pvec |
928 | * | |
929 | * pagevec_lookup_entries() will search for and return a group of up | |
f144c390 | 930 | * to @nr_pages pages and shadow entries in the mapping. All |
0cd6144a JW |
931 | * entries are placed in @pvec. pagevec_lookup_entries() takes a |
932 | * reference against actual pages in @pvec. | |
933 | * | |
934 | * The search returns a group of mapping-contiguous entries with | |
935 | * ascending indexes. There may be holes in the indices due to | |
936 | * not-present entries. | |
937 | * | |
938 | * pagevec_lookup_entries() returns the number of entries which were | |
939 | * found. | |
940 | */ | |
941 | unsigned pagevec_lookup_entries(struct pagevec *pvec, | |
942 | struct address_space *mapping, | |
e02a9f04 | 943 | pgoff_t start, unsigned nr_entries, |
0cd6144a JW |
944 | pgoff_t *indices) |
945 | { | |
e02a9f04 | 946 | pvec->nr = find_get_entries(mapping, start, nr_entries, |
0cd6144a JW |
947 | pvec->pages, indices); |
948 | return pagevec_count(pvec); | |
949 | } | |
950 | ||
951 | /** | |
952 | * pagevec_remove_exceptionals - pagevec exceptionals pruning | |
953 | * @pvec: The pagevec to prune | |
954 | * | |
955 | * pagevec_lookup_entries() fills both pages and exceptional radix | |
956 | * tree entries into the pagevec. This function prunes all | |
957 | * exceptionals from @pvec without leaving holes, so that it can be | |
958 | * passed on to page-only pagevec operations. | |
959 | */ | |
960 | void pagevec_remove_exceptionals(struct pagevec *pvec) | |
961 | { | |
962 | int i, j; | |
963 | ||
964 | for (i = 0, j = 0; i < pagevec_count(pvec); i++) { | |
965 | struct page *page = pvec->pages[i]; | |
3159f943 | 966 | if (!xa_is_value(page)) |
0cd6144a JW |
967 | pvec->pages[j++] = page; |
968 | } | |
969 | pvec->nr = j; | |
970 | } | |
971 | ||
1da177e4 | 972 | /** |
b947cee4 | 973 | * pagevec_lookup_range - gang pagecache lookup |
1da177e4 LT |
974 | * @pvec: Where the resulting pages are placed |
975 | * @mapping: The address_space to search | |
976 | * @start: The starting page index | |
b947cee4 | 977 | * @end: The final page index |
1da177e4 | 978 | * |
e02a9f04 | 979 | * pagevec_lookup_range() will search for & return a group of up to PAGEVEC_SIZE |
b947cee4 JK |
980 | * pages in the mapping starting from index @start and upto index @end |
981 | * (inclusive). The pages are placed in @pvec. pagevec_lookup() takes a | |
1da177e4 LT |
982 | * reference against the pages in @pvec. |
983 | * | |
984 | * The search returns a group of mapping-contiguous pages with ascending | |
d72dc8a2 JK |
985 | * indexes. There may be holes in the indices due to not-present pages. We |
986 | * also update @start to index the next page for the traversal. | |
1da177e4 | 987 | * |
b947cee4 | 988 | * pagevec_lookup_range() returns the number of pages which were found. If this |
e02a9f04 | 989 | * number is smaller than PAGEVEC_SIZE, the end of specified range has been |
b947cee4 | 990 | * reached. |
1da177e4 | 991 | */ |
b947cee4 | 992 | unsigned pagevec_lookup_range(struct pagevec *pvec, |
397162ff | 993 | struct address_space *mapping, pgoff_t *start, pgoff_t end) |
1da177e4 | 994 | { |
397162ff | 995 | pvec->nr = find_get_pages_range(mapping, start, end, PAGEVEC_SIZE, |
b947cee4 | 996 | pvec->pages); |
1da177e4 LT |
997 | return pagevec_count(pvec); |
998 | } | |
b947cee4 | 999 | EXPORT_SYMBOL(pagevec_lookup_range); |
78539fdf | 1000 | |
72b045ae JK |
1001 | unsigned pagevec_lookup_range_tag(struct pagevec *pvec, |
1002 | struct address_space *mapping, pgoff_t *index, pgoff_t end, | |
10bbd235 | 1003 | xa_mark_t tag) |
1da177e4 | 1004 | { |
72b045ae | 1005 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, |
67fd707f | 1006 | PAGEVEC_SIZE, pvec->pages); |
1da177e4 LT |
1007 | return pagevec_count(pvec); |
1008 | } | |
72b045ae | 1009 | EXPORT_SYMBOL(pagevec_lookup_range_tag); |
1da177e4 | 1010 | |
93d3b714 JK |
1011 | unsigned pagevec_lookup_range_nr_tag(struct pagevec *pvec, |
1012 | struct address_space *mapping, pgoff_t *index, pgoff_t end, | |
10bbd235 | 1013 | xa_mark_t tag, unsigned max_pages) |
93d3b714 JK |
1014 | { |
1015 | pvec->nr = find_get_pages_range_tag(mapping, index, end, tag, | |
1016 | min_t(unsigned int, max_pages, PAGEVEC_SIZE), pvec->pages); | |
1017 | return pagevec_count(pvec); | |
1018 | } | |
1019 | EXPORT_SYMBOL(pagevec_lookup_range_nr_tag); | |
1da177e4 LT |
1020 | /* |
1021 | * Perform any setup for the swap system | |
1022 | */ | |
1023 | void __init swap_setup(void) | |
1024 | { | |
4481374c | 1025 | unsigned long megs = totalram_pages >> (20 - PAGE_SHIFT); |
e0bf68dd | 1026 | |
1da177e4 LT |
1027 | /* Use a smaller cluster for small-memory machines */ |
1028 | if (megs < 16) | |
1029 | page_cluster = 2; | |
1030 | else | |
1031 | page_cluster = 3; | |
1032 | /* | |
1033 | * Right now other parts of the system means that we | |
1034 | * _really_ don't want to cluster much more | |
1035 | */ | |
1da177e4 | 1036 | } |