Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/vmscan.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | |
5 | * | |
6 | * Swap reorganised 29.12.95, Stephen Tweedie. | |
7 | * kswapd added: 7.1.96 sct | |
8 | * Removed kswapd_ctl limits, and swap out as many pages as needed | |
9 | * to bring the system back to freepages.high: 2.4.97, Rik van Riel. | |
10 | * Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com). | |
11 | * Multiqueue VM started 5.8.00, Rik van Riel. | |
12 | */ | |
13 | ||
14 | #include <linux/mm.h> | |
15 | #include <linux/module.h> | |
5a0e3ad6 | 16 | #include <linux/gfp.h> |
1da177e4 LT |
17 | #include <linux/kernel_stat.h> |
18 | #include <linux/swap.h> | |
19 | #include <linux/pagemap.h> | |
20 | #include <linux/init.h> | |
21 | #include <linux/highmem.h> | |
70ddf637 | 22 | #include <linux/vmpressure.h> |
e129b5c2 | 23 | #include <linux/vmstat.h> |
1da177e4 LT |
24 | #include <linux/file.h> |
25 | #include <linux/writeback.h> | |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/buffer_head.h> /* for try_to_release_page(), | |
28 | buffer_heads_over_limit */ | |
29 | #include <linux/mm_inline.h> | |
1da177e4 LT |
30 | #include <linux/backing-dev.h> |
31 | #include <linux/rmap.h> | |
32 | #include <linux/topology.h> | |
33 | #include <linux/cpu.h> | |
34 | #include <linux/cpuset.h> | |
3e7d3449 | 35 | #include <linux/compaction.h> |
1da177e4 LT |
36 | #include <linux/notifier.h> |
37 | #include <linux/rwsem.h> | |
248a0301 | 38 | #include <linux/delay.h> |
3218ae14 | 39 | #include <linux/kthread.h> |
7dfb7103 | 40 | #include <linux/freezer.h> |
66e1707b | 41 | #include <linux/memcontrol.h> |
873b4771 | 42 | #include <linux/delayacct.h> |
af936a16 | 43 | #include <linux/sysctl.h> |
929bea7c | 44 | #include <linux/oom.h> |
268bb0ce | 45 | #include <linux/prefetch.h> |
1da177e4 LT |
46 | |
47 | #include <asm/tlbflush.h> | |
48 | #include <asm/div64.h> | |
49 | ||
50 | #include <linux/swapops.h> | |
51 | ||
0f8053a5 NP |
52 | #include "internal.h" |
53 | ||
33906bc5 MG |
54 | #define CREATE_TRACE_POINTS |
55 | #include <trace/events/vmscan.h> | |
56 | ||
1da177e4 | 57 | struct scan_control { |
1da177e4 LT |
58 | /* Incremented by the number of inactive pages that were scanned */ |
59 | unsigned long nr_scanned; | |
60 | ||
a79311c1 RR |
61 | /* Number of pages freed so far during a call to shrink_zones() */ |
62 | unsigned long nr_reclaimed; | |
63 | ||
22fba335 KM |
64 | /* How many pages shrink_list() should reclaim */ |
65 | unsigned long nr_to_reclaim; | |
66 | ||
7b51755c KM |
67 | unsigned long hibernation_mode; |
68 | ||
1da177e4 | 69 | /* This context's GFP mask */ |
6daa0e28 | 70 | gfp_t gfp_mask; |
1da177e4 LT |
71 | |
72 | int may_writepage; | |
73 | ||
a6dc60f8 JW |
74 | /* Can mapped pages be reclaimed? */ |
75 | int may_unmap; | |
f1fd1067 | 76 | |
2e2e4259 KM |
77 | /* Can pages be swapped as part of reclaim? */ |
78 | int may_swap; | |
79 | ||
5ad333eb | 80 | int order; |
66e1707b | 81 | |
9e3b2f8c KK |
82 | /* Scan (total_size >> priority) pages at once */ |
83 | int priority; | |
84 | ||
f16015fb JW |
85 | /* |
86 | * The memory cgroup that hit its limit and as a result is the | |
87 | * primary target of this reclaim invocation. | |
88 | */ | |
89 | struct mem_cgroup *target_mem_cgroup; | |
66e1707b | 90 | |
327c0e96 KH |
91 | /* |
92 | * Nodemask of nodes allowed by the caller. If NULL, all nodes | |
93 | * are scanned. | |
94 | */ | |
95 | nodemask_t *nodemask; | |
1da177e4 LT |
96 | }; |
97 | ||
1da177e4 LT |
98 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
99 | ||
100 | #ifdef ARCH_HAS_PREFETCH | |
101 | #define prefetch_prev_lru_page(_page, _base, _field) \ | |
102 | do { \ | |
103 | if ((_page)->lru.prev != _base) { \ | |
104 | struct page *prev; \ | |
105 | \ | |
106 | prev = lru_to_page(&(_page->lru)); \ | |
107 | prefetch(&prev->_field); \ | |
108 | } \ | |
109 | } while (0) | |
110 | #else | |
111 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) | |
112 | #endif | |
113 | ||
114 | #ifdef ARCH_HAS_PREFETCHW | |
115 | #define prefetchw_prev_lru_page(_page, _base, _field) \ | |
116 | do { \ | |
117 | if ((_page)->lru.prev != _base) { \ | |
118 | struct page *prev; \ | |
119 | \ | |
120 | prev = lru_to_page(&(_page->lru)); \ | |
121 | prefetchw(&prev->_field); \ | |
122 | } \ | |
123 | } while (0) | |
124 | #else | |
125 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) | |
126 | #endif | |
127 | ||
128 | /* | |
129 | * From 0 .. 100. Higher means more swappy. | |
130 | */ | |
131 | int vm_swappiness = 60; | |
b21e0b90 | 132 | unsigned long vm_total_pages; /* The total number of pages which the VM controls */ |
1da177e4 LT |
133 | |
134 | static LIST_HEAD(shrinker_list); | |
135 | static DECLARE_RWSEM(shrinker_rwsem); | |
136 | ||
c255a458 | 137 | #ifdef CONFIG_MEMCG |
89b5fae5 JW |
138 | static bool global_reclaim(struct scan_control *sc) |
139 | { | |
f16015fb | 140 | return !sc->target_mem_cgroup; |
89b5fae5 | 141 | } |
91a45470 | 142 | #else |
89b5fae5 JW |
143 | static bool global_reclaim(struct scan_control *sc) |
144 | { | |
145 | return true; | |
146 | } | |
91a45470 KH |
147 | #endif |
148 | ||
4d7dcca2 | 149 | static unsigned long get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
c9f299d9 | 150 | { |
c3c787e8 | 151 | if (!mem_cgroup_disabled()) |
4d7dcca2 | 152 | return mem_cgroup_get_lru_size(lruvec, lru); |
a3d8e054 | 153 | |
074291fe | 154 | return zone_page_state(lruvec_zone(lruvec), NR_LRU_BASE + lru); |
c9f299d9 KM |
155 | } |
156 | ||
1da177e4 | 157 | /* |
1d3d4437 | 158 | * Add a shrinker callback to be called from the vm. |
1da177e4 | 159 | */ |
1d3d4437 | 160 | int register_shrinker(struct shrinker *shrinker) |
1da177e4 | 161 | { |
1d3d4437 GC |
162 | size_t size = sizeof(*shrinker->nr_deferred); |
163 | ||
164 | /* | |
165 | * If we only have one possible node in the system anyway, save | |
166 | * ourselves the trouble and disable NUMA aware behavior. This way we | |
167 | * will save memory and some small loop time later. | |
168 | */ | |
169 | if (nr_node_ids == 1) | |
170 | shrinker->flags &= ~SHRINKER_NUMA_AWARE; | |
171 | ||
172 | if (shrinker->flags & SHRINKER_NUMA_AWARE) | |
173 | size *= nr_node_ids; | |
174 | ||
175 | shrinker->nr_deferred = kzalloc(size, GFP_KERNEL); | |
176 | if (!shrinker->nr_deferred) | |
177 | return -ENOMEM; | |
178 | ||
8e1f936b RR |
179 | down_write(&shrinker_rwsem); |
180 | list_add_tail(&shrinker->list, &shrinker_list); | |
181 | up_write(&shrinker_rwsem); | |
1d3d4437 | 182 | return 0; |
1da177e4 | 183 | } |
8e1f936b | 184 | EXPORT_SYMBOL(register_shrinker); |
1da177e4 LT |
185 | |
186 | /* | |
187 | * Remove one | |
188 | */ | |
8e1f936b | 189 | void unregister_shrinker(struct shrinker *shrinker) |
1da177e4 LT |
190 | { |
191 | down_write(&shrinker_rwsem); | |
192 | list_del(&shrinker->list); | |
193 | up_write(&shrinker_rwsem); | |
1da177e4 | 194 | } |
8e1f936b | 195 | EXPORT_SYMBOL(unregister_shrinker); |
1da177e4 | 196 | |
1495f230 YH |
197 | static inline int do_shrinker_shrink(struct shrinker *shrinker, |
198 | struct shrink_control *sc, | |
199 | unsigned long nr_to_scan) | |
200 | { | |
201 | sc->nr_to_scan = nr_to_scan; | |
202 | return (*shrinker->shrink)(shrinker, sc); | |
203 | } | |
204 | ||
1da177e4 | 205 | #define SHRINK_BATCH 128 |
1d3d4437 GC |
206 | |
207 | static unsigned long | |
208 | shrink_slab_node(struct shrink_control *shrinkctl, struct shrinker *shrinker, | |
209 | unsigned long nr_pages_scanned, unsigned long lru_pages) | |
210 | { | |
211 | unsigned long freed = 0; | |
212 | unsigned long long delta; | |
213 | long total_scan; | |
214 | long max_pass; | |
215 | long nr; | |
216 | long new_nr; | |
217 | int nid = shrinkctl->nid; | |
218 | long batch_size = shrinker->batch ? shrinker->batch | |
219 | : SHRINK_BATCH; | |
220 | ||
221 | if (shrinker->count_objects) | |
222 | max_pass = shrinker->count_objects(shrinker, shrinkctl); | |
223 | else | |
224 | max_pass = do_shrinker_shrink(shrinker, shrinkctl, 0); | |
225 | if (max_pass == 0) | |
226 | return 0; | |
227 | ||
228 | /* | |
229 | * copy the current shrinker scan count into a local variable | |
230 | * and zero it so that other concurrent shrinker invocations | |
231 | * don't also do this scanning work. | |
232 | */ | |
233 | nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0); | |
234 | ||
235 | total_scan = nr; | |
236 | delta = (4 * nr_pages_scanned) / shrinker->seeks; | |
237 | delta *= max_pass; | |
238 | do_div(delta, lru_pages + 1); | |
239 | total_scan += delta; | |
240 | if (total_scan < 0) { | |
241 | printk(KERN_ERR | |
242 | "shrink_slab: %pF negative objects to delete nr=%ld\n", | |
243 | shrinker->shrink, total_scan); | |
244 | total_scan = max_pass; | |
245 | } | |
246 | ||
247 | /* | |
248 | * We need to avoid excessive windup on filesystem shrinkers | |
249 | * due to large numbers of GFP_NOFS allocations causing the | |
250 | * shrinkers to return -1 all the time. This results in a large | |
251 | * nr being built up so when a shrink that can do some work | |
252 | * comes along it empties the entire cache due to nr >>> | |
253 | * max_pass. This is bad for sustaining a working set in | |
254 | * memory. | |
255 | * | |
256 | * Hence only allow the shrinker to scan the entire cache when | |
257 | * a large delta change is calculated directly. | |
258 | */ | |
259 | if (delta < max_pass / 4) | |
260 | total_scan = min(total_scan, max_pass / 2); | |
261 | ||
262 | /* | |
263 | * Avoid risking looping forever due to too large nr value: | |
264 | * never try to free more than twice the estimate number of | |
265 | * freeable entries. | |
266 | */ | |
267 | if (total_scan > max_pass * 2) | |
268 | total_scan = max_pass * 2; | |
269 | ||
270 | trace_mm_shrink_slab_start(shrinker, shrinkctl, nr, | |
271 | nr_pages_scanned, lru_pages, | |
272 | max_pass, delta, total_scan); | |
273 | ||
274 | while (total_scan >= batch_size) { | |
275 | ||
276 | if (shrinker->scan_objects) { | |
277 | unsigned long ret; | |
278 | shrinkctl->nr_to_scan = batch_size; | |
279 | ret = shrinker->scan_objects(shrinker, shrinkctl); | |
280 | ||
281 | if (ret == SHRINK_STOP) | |
282 | break; | |
283 | freed += ret; | |
284 | } else { | |
285 | int nr_before; | |
286 | long ret; | |
287 | ||
288 | nr_before = do_shrinker_shrink(shrinker, shrinkctl, 0); | |
289 | ret = do_shrinker_shrink(shrinker, shrinkctl, | |
290 | batch_size); | |
291 | if (ret == -1) | |
292 | break; | |
293 | if (ret < nr_before) | |
294 | freed += nr_before - ret; | |
295 | } | |
296 | ||
297 | count_vm_events(SLABS_SCANNED, batch_size); | |
298 | total_scan -= batch_size; | |
299 | ||
300 | cond_resched(); | |
301 | } | |
302 | ||
303 | /* | |
304 | * move the unused scan count back into the shrinker in a | |
305 | * manner that handles concurrent updates. If we exhausted the | |
306 | * scan, there is no need to do an update. | |
307 | */ | |
308 | if (total_scan > 0) | |
309 | new_nr = atomic_long_add_return(total_scan, | |
310 | &shrinker->nr_deferred[nid]); | |
311 | else | |
312 | new_nr = atomic_long_read(&shrinker->nr_deferred[nid]); | |
313 | ||
314 | trace_mm_shrink_slab_end(shrinker, freed, nr, new_nr); | |
315 | return freed; | |
316 | } | |
317 | ||
1da177e4 LT |
318 | /* |
319 | * Call the shrink functions to age shrinkable caches | |
320 | * | |
321 | * Here we assume it costs one seek to replace a lru page and that it also | |
322 | * takes a seek to recreate a cache object. With this in mind we age equal | |
323 | * percentages of the lru and ageable caches. This should balance the seeks | |
324 | * generated by these structures. | |
325 | * | |
183ff22b | 326 | * If the vm encountered mapped pages on the LRU it increase the pressure on |
1da177e4 LT |
327 | * slab to avoid swapping. |
328 | * | |
329 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. | |
330 | * | |
331 | * `lru_pages' represents the number of on-LRU pages in all the zones which | |
332 | * are eligible for the caller's allocation attempt. It is used for balancing | |
333 | * slab reclaim versus page reclaim. | |
b15e0905 | 334 | * |
335 | * Returns the number of slab objects which we shrunk. | |
1da177e4 | 336 | */ |
24f7c6b9 | 337 | unsigned long shrink_slab(struct shrink_control *shrinkctl, |
1495f230 | 338 | unsigned long nr_pages_scanned, |
a09ed5e0 | 339 | unsigned long lru_pages) |
1da177e4 LT |
340 | { |
341 | struct shrinker *shrinker; | |
24f7c6b9 | 342 | unsigned long freed = 0; |
1da177e4 | 343 | |
1495f230 YH |
344 | if (nr_pages_scanned == 0) |
345 | nr_pages_scanned = SWAP_CLUSTER_MAX; | |
1da177e4 | 346 | |
f06590bd | 347 | if (!down_read_trylock(&shrinker_rwsem)) { |
24f7c6b9 DC |
348 | /* |
349 | * If we would return 0, our callers would understand that we | |
350 | * have nothing else to shrink and give up trying. By returning | |
351 | * 1 we keep it going and assume we'll be able to shrink next | |
352 | * time. | |
353 | */ | |
354 | freed = 1; | |
f06590bd MK |
355 | goto out; |
356 | } | |
1da177e4 LT |
357 | |
358 | list_for_each_entry(shrinker, &shrinker_list, list) { | |
1d3d4437 GC |
359 | for_each_node_mask(shrinkctl->nid, shrinkctl->nodes_to_scan) { |
360 | if (!node_online(shrinkctl->nid)) | |
361 | continue; | |
635697c6 | 362 | |
1d3d4437 GC |
363 | if (!(shrinker->flags & SHRINKER_NUMA_AWARE) && |
364 | (shrinkctl->nid != 0)) | |
365 | break; | |
ea164d73 | 366 | |
1d3d4437 GC |
367 | freed += shrink_slab_node(shrinkctl, shrinker, |
368 | nr_pages_scanned, lru_pages); | |
3567b59a | 369 | |
1da177e4 | 370 | } |
1da177e4 LT |
371 | } |
372 | up_read(&shrinker_rwsem); | |
f06590bd MK |
373 | out: |
374 | cond_resched(); | |
24f7c6b9 | 375 | return freed; |
1da177e4 LT |
376 | } |
377 | ||
1da177e4 LT |
378 | static inline int is_page_cache_freeable(struct page *page) |
379 | { | |
ceddc3a5 JW |
380 | /* |
381 | * A freeable page cache page is referenced only by the caller | |
382 | * that isolated the page, the page cache radix tree and | |
383 | * optional buffer heads at page->private. | |
384 | */ | |
edcf4748 | 385 | return page_count(page) - page_has_private(page) == 2; |
1da177e4 LT |
386 | } |
387 | ||
7d3579e8 KM |
388 | static int may_write_to_queue(struct backing_dev_info *bdi, |
389 | struct scan_control *sc) | |
1da177e4 | 390 | { |
930d9152 | 391 | if (current->flags & PF_SWAPWRITE) |
1da177e4 LT |
392 | return 1; |
393 | if (!bdi_write_congested(bdi)) | |
394 | return 1; | |
395 | if (bdi == current->backing_dev_info) | |
396 | return 1; | |
397 | return 0; | |
398 | } | |
399 | ||
400 | /* | |
401 | * We detected a synchronous write error writing a page out. Probably | |
402 | * -ENOSPC. We need to propagate that into the address_space for a subsequent | |
403 | * fsync(), msync() or close(). | |
404 | * | |
405 | * The tricky part is that after writepage we cannot touch the mapping: nothing | |
406 | * prevents it from being freed up. But we have a ref on the page and once | |
407 | * that page is locked, the mapping is pinned. | |
408 | * | |
409 | * We're allowed to run sleeping lock_page() here because we know the caller has | |
410 | * __GFP_FS. | |
411 | */ | |
412 | static void handle_write_error(struct address_space *mapping, | |
413 | struct page *page, int error) | |
414 | { | |
7eaceacc | 415 | lock_page(page); |
3e9f45bd GC |
416 | if (page_mapping(page) == mapping) |
417 | mapping_set_error(mapping, error); | |
1da177e4 LT |
418 | unlock_page(page); |
419 | } | |
420 | ||
04e62a29 CL |
421 | /* possible outcome of pageout() */ |
422 | typedef enum { | |
423 | /* failed to write page out, page is locked */ | |
424 | PAGE_KEEP, | |
425 | /* move page to the active list, page is locked */ | |
426 | PAGE_ACTIVATE, | |
427 | /* page has been sent to the disk successfully, page is unlocked */ | |
428 | PAGE_SUCCESS, | |
429 | /* page is clean and locked */ | |
430 | PAGE_CLEAN, | |
431 | } pageout_t; | |
432 | ||
1da177e4 | 433 | /* |
1742f19f AM |
434 | * pageout is called by shrink_page_list() for each dirty page. |
435 | * Calls ->writepage(). | |
1da177e4 | 436 | */ |
c661b078 | 437 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
7d3579e8 | 438 | struct scan_control *sc) |
1da177e4 LT |
439 | { |
440 | /* | |
441 | * If the page is dirty, only perform writeback if that write | |
442 | * will be non-blocking. To prevent this allocation from being | |
443 | * stalled by pagecache activity. But note that there may be | |
444 | * stalls if we need to run get_block(). We could test | |
445 | * PagePrivate for that. | |
446 | * | |
6aceb53b | 447 | * If this process is currently in __generic_file_aio_write() against |
1da177e4 LT |
448 | * this page's queue, we can perform writeback even if that |
449 | * will block. | |
450 | * | |
451 | * If the page is swapcache, write it back even if that would | |
452 | * block, for some throttling. This happens by accident, because | |
453 | * swap_backing_dev_info is bust: it doesn't reflect the | |
454 | * congestion state of the swapdevs. Easy to fix, if needed. | |
1da177e4 LT |
455 | */ |
456 | if (!is_page_cache_freeable(page)) | |
457 | return PAGE_KEEP; | |
458 | if (!mapping) { | |
459 | /* | |
460 | * Some data journaling orphaned pages can have | |
461 | * page->mapping == NULL while being dirty with clean buffers. | |
462 | */ | |
266cf658 | 463 | if (page_has_private(page)) { |
1da177e4 LT |
464 | if (try_to_free_buffers(page)) { |
465 | ClearPageDirty(page); | |
d40cee24 | 466 | printk("%s: orphaned page\n", __func__); |
1da177e4 LT |
467 | return PAGE_CLEAN; |
468 | } | |
469 | } | |
470 | return PAGE_KEEP; | |
471 | } | |
472 | if (mapping->a_ops->writepage == NULL) | |
473 | return PAGE_ACTIVATE; | |
0e093d99 | 474 | if (!may_write_to_queue(mapping->backing_dev_info, sc)) |
1da177e4 LT |
475 | return PAGE_KEEP; |
476 | ||
477 | if (clear_page_dirty_for_io(page)) { | |
478 | int res; | |
479 | struct writeback_control wbc = { | |
480 | .sync_mode = WB_SYNC_NONE, | |
481 | .nr_to_write = SWAP_CLUSTER_MAX, | |
111ebb6e OH |
482 | .range_start = 0, |
483 | .range_end = LLONG_MAX, | |
1da177e4 LT |
484 | .for_reclaim = 1, |
485 | }; | |
486 | ||
487 | SetPageReclaim(page); | |
488 | res = mapping->a_ops->writepage(page, &wbc); | |
489 | if (res < 0) | |
490 | handle_write_error(mapping, page, res); | |
994fc28c | 491 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
1da177e4 LT |
492 | ClearPageReclaim(page); |
493 | return PAGE_ACTIVATE; | |
494 | } | |
c661b078 | 495 | |
1da177e4 LT |
496 | if (!PageWriteback(page)) { |
497 | /* synchronous write or broken a_ops? */ | |
498 | ClearPageReclaim(page); | |
499 | } | |
23b9da55 | 500 | trace_mm_vmscan_writepage(page, trace_reclaim_flags(page)); |
e129b5c2 | 501 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
1da177e4 LT |
502 | return PAGE_SUCCESS; |
503 | } | |
504 | ||
505 | return PAGE_CLEAN; | |
506 | } | |
507 | ||
a649fd92 | 508 | /* |
e286781d NP |
509 | * Same as remove_mapping, but if the page is removed from the mapping, it |
510 | * gets returned with a refcount of 0. | |
a649fd92 | 511 | */ |
e286781d | 512 | static int __remove_mapping(struct address_space *mapping, struct page *page) |
49d2e9cc | 513 | { |
28e4d965 NP |
514 | BUG_ON(!PageLocked(page)); |
515 | BUG_ON(mapping != page_mapping(page)); | |
49d2e9cc | 516 | |
19fd6231 | 517 | spin_lock_irq(&mapping->tree_lock); |
49d2e9cc | 518 | /* |
0fd0e6b0 NP |
519 | * The non racy check for a busy page. |
520 | * | |
521 | * Must be careful with the order of the tests. When someone has | |
522 | * a ref to the page, it may be possible that they dirty it then | |
523 | * drop the reference. So if PageDirty is tested before page_count | |
524 | * here, then the following race may occur: | |
525 | * | |
526 | * get_user_pages(&page); | |
527 | * [user mapping goes away] | |
528 | * write_to(page); | |
529 | * !PageDirty(page) [good] | |
530 | * SetPageDirty(page); | |
531 | * put_page(page); | |
532 | * !page_count(page) [good, discard it] | |
533 | * | |
534 | * [oops, our write_to data is lost] | |
535 | * | |
536 | * Reversing the order of the tests ensures such a situation cannot | |
537 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags | |
538 | * load is not satisfied before that of page->_count. | |
539 | * | |
540 | * Note that if SetPageDirty is always performed via set_page_dirty, | |
541 | * and thus under tree_lock, then this ordering is not required. | |
49d2e9cc | 542 | */ |
e286781d | 543 | if (!page_freeze_refs(page, 2)) |
49d2e9cc | 544 | goto cannot_free; |
e286781d NP |
545 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
546 | if (unlikely(PageDirty(page))) { | |
547 | page_unfreeze_refs(page, 2); | |
49d2e9cc | 548 | goto cannot_free; |
e286781d | 549 | } |
49d2e9cc CL |
550 | |
551 | if (PageSwapCache(page)) { | |
552 | swp_entry_t swap = { .val = page_private(page) }; | |
553 | __delete_from_swap_cache(page); | |
19fd6231 | 554 | spin_unlock_irq(&mapping->tree_lock); |
cb4b86ba | 555 | swapcache_free(swap, page); |
e286781d | 556 | } else { |
6072d13c LT |
557 | void (*freepage)(struct page *); |
558 | ||
559 | freepage = mapping->a_ops->freepage; | |
560 | ||
e64a782f | 561 | __delete_from_page_cache(page); |
19fd6231 | 562 | spin_unlock_irq(&mapping->tree_lock); |
e767e056 | 563 | mem_cgroup_uncharge_cache_page(page); |
6072d13c LT |
564 | |
565 | if (freepage != NULL) | |
566 | freepage(page); | |
49d2e9cc CL |
567 | } |
568 | ||
49d2e9cc CL |
569 | return 1; |
570 | ||
571 | cannot_free: | |
19fd6231 | 572 | spin_unlock_irq(&mapping->tree_lock); |
49d2e9cc CL |
573 | return 0; |
574 | } | |
575 | ||
e286781d NP |
576 | /* |
577 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if | |
578 | * someone else has a ref on the page, abort and return 0. If it was | |
579 | * successfully detached, return 1. Assumes the caller has a single ref on | |
580 | * this page. | |
581 | */ | |
582 | int remove_mapping(struct address_space *mapping, struct page *page) | |
583 | { | |
584 | if (__remove_mapping(mapping, page)) { | |
585 | /* | |
586 | * Unfreezing the refcount with 1 rather than 2 effectively | |
587 | * drops the pagecache ref for us without requiring another | |
588 | * atomic operation. | |
589 | */ | |
590 | page_unfreeze_refs(page, 1); | |
591 | return 1; | |
592 | } | |
593 | return 0; | |
594 | } | |
595 | ||
894bc310 LS |
596 | /** |
597 | * putback_lru_page - put previously isolated page onto appropriate LRU list | |
598 | * @page: page to be put back to appropriate lru list | |
599 | * | |
600 | * Add previously isolated @page to appropriate LRU list. | |
601 | * Page may still be unevictable for other reasons. | |
602 | * | |
603 | * lru_lock must not be held, interrupts must be enabled. | |
604 | */ | |
894bc310 LS |
605 | void putback_lru_page(struct page *page) |
606 | { | |
607 | int lru; | |
bbfd28ee | 608 | int was_unevictable = PageUnevictable(page); |
894bc310 LS |
609 | |
610 | VM_BUG_ON(PageLRU(page)); | |
611 | ||
612 | redo: | |
613 | ClearPageUnevictable(page); | |
614 | ||
39b5f29a | 615 | if (page_evictable(page)) { |
894bc310 LS |
616 | /* |
617 | * For evictable pages, we can use the cache. | |
618 | * In event of a race, worst case is we end up with an | |
619 | * unevictable page on [in]active list. | |
620 | * We know how to handle that. | |
621 | */ | |
c53954a0 MG |
622 | lru = page_lru_base_type(page); |
623 | lru_cache_add(page); | |
894bc310 LS |
624 | } else { |
625 | /* | |
626 | * Put unevictable pages directly on zone's unevictable | |
627 | * list. | |
628 | */ | |
629 | lru = LRU_UNEVICTABLE; | |
630 | add_page_to_unevictable_list(page); | |
6a7b9548 | 631 | /* |
21ee9f39 MK |
632 | * When racing with an mlock or AS_UNEVICTABLE clearing |
633 | * (page is unlocked) make sure that if the other thread | |
634 | * does not observe our setting of PG_lru and fails | |
24513264 | 635 | * isolation/check_move_unevictable_pages, |
21ee9f39 | 636 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
6a7b9548 JW |
637 | * the page back to the evictable list. |
638 | * | |
21ee9f39 | 639 | * The other side is TestClearPageMlocked() or shmem_lock(). |
6a7b9548 JW |
640 | */ |
641 | smp_mb(); | |
894bc310 | 642 | } |
894bc310 LS |
643 | |
644 | /* | |
645 | * page's status can change while we move it among lru. If an evictable | |
646 | * page is on unevictable list, it never be freed. To avoid that, | |
647 | * check after we added it to the list, again. | |
648 | */ | |
39b5f29a | 649 | if (lru == LRU_UNEVICTABLE && page_evictable(page)) { |
894bc310 LS |
650 | if (!isolate_lru_page(page)) { |
651 | put_page(page); | |
652 | goto redo; | |
653 | } | |
654 | /* This means someone else dropped this page from LRU | |
655 | * So, it will be freed or putback to LRU again. There is | |
656 | * nothing to do here. | |
657 | */ | |
658 | } | |
659 | ||
bbfd28ee LS |
660 | if (was_unevictable && lru != LRU_UNEVICTABLE) |
661 | count_vm_event(UNEVICTABLE_PGRESCUED); | |
662 | else if (!was_unevictable && lru == LRU_UNEVICTABLE) | |
663 | count_vm_event(UNEVICTABLE_PGCULLED); | |
664 | ||
894bc310 LS |
665 | put_page(page); /* drop ref from isolate */ |
666 | } | |
667 | ||
dfc8d636 JW |
668 | enum page_references { |
669 | PAGEREF_RECLAIM, | |
670 | PAGEREF_RECLAIM_CLEAN, | |
64574746 | 671 | PAGEREF_KEEP, |
dfc8d636 JW |
672 | PAGEREF_ACTIVATE, |
673 | }; | |
674 | ||
675 | static enum page_references page_check_references(struct page *page, | |
676 | struct scan_control *sc) | |
677 | { | |
64574746 | 678 | int referenced_ptes, referenced_page; |
dfc8d636 | 679 | unsigned long vm_flags; |
dfc8d636 | 680 | |
c3ac9a8a JW |
681 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, |
682 | &vm_flags); | |
64574746 | 683 | referenced_page = TestClearPageReferenced(page); |
dfc8d636 | 684 | |
dfc8d636 JW |
685 | /* |
686 | * Mlock lost the isolation race with us. Let try_to_unmap() | |
687 | * move the page to the unevictable list. | |
688 | */ | |
689 | if (vm_flags & VM_LOCKED) | |
690 | return PAGEREF_RECLAIM; | |
691 | ||
64574746 | 692 | if (referenced_ptes) { |
e4898273 | 693 | if (PageSwapBacked(page)) |
64574746 JW |
694 | return PAGEREF_ACTIVATE; |
695 | /* | |
696 | * All mapped pages start out with page table | |
697 | * references from the instantiating fault, so we need | |
698 | * to look twice if a mapped file page is used more | |
699 | * than once. | |
700 | * | |
701 | * Mark it and spare it for another trip around the | |
702 | * inactive list. Another page table reference will | |
703 | * lead to its activation. | |
704 | * | |
705 | * Note: the mark is set for activated pages as well | |
706 | * so that recently deactivated but used pages are | |
707 | * quickly recovered. | |
708 | */ | |
709 | SetPageReferenced(page); | |
710 | ||
34dbc67a | 711 | if (referenced_page || referenced_ptes > 1) |
64574746 JW |
712 | return PAGEREF_ACTIVATE; |
713 | ||
c909e993 KK |
714 | /* |
715 | * Activate file-backed executable pages after first usage. | |
716 | */ | |
717 | if (vm_flags & VM_EXEC) | |
718 | return PAGEREF_ACTIVATE; | |
719 | ||
64574746 JW |
720 | return PAGEREF_KEEP; |
721 | } | |
dfc8d636 JW |
722 | |
723 | /* Reclaim if clean, defer dirty pages to writeback */ | |
2e30244a | 724 | if (referenced_page && !PageSwapBacked(page)) |
64574746 JW |
725 | return PAGEREF_RECLAIM_CLEAN; |
726 | ||
727 | return PAGEREF_RECLAIM; | |
dfc8d636 JW |
728 | } |
729 | ||
e2be15f6 MG |
730 | /* Check if a page is dirty or under writeback */ |
731 | static void page_check_dirty_writeback(struct page *page, | |
732 | bool *dirty, bool *writeback) | |
733 | { | |
b4597226 MG |
734 | struct address_space *mapping; |
735 | ||
e2be15f6 MG |
736 | /* |
737 | * Anonymous pages are not handled by flushers and must be written | |
738 | * from reclaim context. Do not stall reclaim based on them | |
739 | */ | |
740 | if (!page_is_file_cache(page)) { | |
741 | *dirty = false; | |
742 | *writeback = false; | |
743 | return; | |
744 | } | |
745 | ||
746 | /* By default assume that the page flags are accurate */ | |
747 | *dirty = PageDirty(page); | |
748 | *writeback = PageWriteback(page); | |
b4597226 MG |
749 | |
750 | /* Verify dirty/writeback state if the filesystem supports it */ | |
751 | if (!page_has_private(page)) | |
752 | return; | |
753 | ||
754 | mapping = page_mapping(page); | |
755 | if (mapping && mapping->a_ops->is_dirty_writeback) | |
756 | mapping->a_ops->is_dirty_writeback(page, dirty, writeback); | |
e2be15f6 MG |
757 | } |
758 | ||
1da177e4 | 759 | /* |
1742f19f | 760 | * shrink_page_list() returns the number of reclaimed pages |
1da177e4 | 761 | */ |
1742f19f | 762 | static unsigned long shrink_page_list(struct list_head *page_list, |
6a18adb3 | 763 | struct zone *zone, |
f84f6e2b | 764 | struct scan_control *sc, |
02c6de8d | 765 | enum ttu_flags ttu_flags, |
8e950282 | 766 | unsigned long *ret_nr_dirty, |
d43006d5 | 767 | unsigned long *ret_nr_unqueued_dirty, |
8e950282 | 768 | unsigned long *ret_nr_congested, |
02c6de8d | 769 | unsigned long *ret_nr_writeback, |
b1a6f21e | 770 | unsigned long *ret_nr_immediate, |
02c6de8d | 771 | bool force_reclaim) |
1da177e4 LT |
772 | { |
773 | LIST_HEAD(ret_pages); | |
abe4c3b5 | 774 | LIST_HEAD(free_pages); |
1da177e4 | 775 | int pgactivate = 0; |
d43006d5 | 776 | unsigned long nr_unqueued_dirty = 0; |
0e093d99 MG |
777 | unsigned long nr_dirty = 0; |
778 | unsigned long nr_congested = 0; | |
05ff5137 | 779 | unsigned long nr_reclaimed = 0; |
92df3a72 | 780 | unsigned long nr_writeback = 0; |
b1a6f21e | 781 | unsigned long nr_immediate = 0; |
1da177e4 LT |
782 | |
783 | cond_resched(); | |
784 | ||
69980e31 | 785 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
786 | while (!list_empty(page_list)) { |
787 | struct address_space *mapping; | |
788 | struct page *page; | |
789 | int may_enter_fs; | |
02c6de8d | 790 | enum page_references references = PAGEREF_RECLAIM_CLEAN; |
e2be15f6 | 791 | bool dirty, writeback; |
1da177e4 LT |
792 | |
793 | cond_resched(); | |
794 | ||
795 | page = lru_to_page(page_list); | |
796 | list_del(&page->lru); | |
797 | ||
529ae9aa | 798 | if (!trylock_page(page)) |
1da177e4 LT |
799 | goto keep; |
800 | ||
725d704e | 801 | VM_BUG_ON(PageActive(page)); |
6a18adb3 | 802 | VM_BUG_ON(page_zone(page) != zone); |
1da177e4 LT |
803 | |
804 | sc->nr_scanned++; | |
80e43426 | 805 | |
39b5f29a | 806 | if (unlikely(!page_evictable(page))) |
b291f000 | 807 | goto cull_mlocked; |
894bc310 | 808 | |
a6dc60f8 | 809 | if (!sc->may_unmap && page_mapped(page)) |
80e43426 CL |
810 | goto keep_locked; |
811 | ||
1da177e4 LT |
812 | /* Double the slab pressure for mapped and swapcache pages */ |
813 | if (page_mapped(page) || PageSwapCache(page)) | |
814 | sc->nr_scanned++; | |
815 | ||
c661b078 AW |
816 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
817 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); | |
818 | ||
e2be15f6 MG |
819 | /* |
820 | * The number of dirty pages determines if a zone is marked | |
821 | * reclaim_congested which affects wait_iff_congested. kswapd | |
822 | * will stall and start writing pages if the tail of the LRU | |
823 | * is all dirty unqueued pages. | |
824 | */ | |
825 | page_check_dirty_writeback(page, &dirty, &writeback); | |
826 | if (dirty || writeback) | |
827 | nr_dirty++; | |
828 | ||
829 | if (dirty && !writeback) | |
830 | nr_unqueued_dirty++; | |
831 | ||
d04e8acd MG |
832 | /* |
833 | * Treat this page as congested if the underlying BDI is or if | |
834 | * pages are cycling through the LRU so quickly that the | |
835 | * pages marked for immediate reclaim are making it to the | |
836 | * end of the LRU a second time. | |
837 | */ | |
e2be15f6 | 838 | mapping = page_mapping(page); |
d04e8acd MG |
839 | if ((mapping && bdi_write_congested(mapping->backing_dev_info)) || |
840 | (writeback && PageReclaim(page))) | |
e2be15f6 MG |
841 | nr_congested++; |
842 | ||
283aba9f MG |
843 | /* |
844 | * If a page at the tail of the LRU is under writeback, there | |
845 | * are three cases to consider. | |
846 | * | |
847 | * 1) If reclaim is encountering an excessive number of pages | |
848 | * under writeback and this page is both under writeback and | |
849 | * PageReclaim then it indicates that pages are being queued | |
850 | * for IO but are being recycled through the LRU before the | |
851 | * IO can complete. Waiting on the page itself risks an | |
852 | * indefinite stall if it is impossible to writeback the | |
853 | * page due to IO error or disconnected storage so instead | |
b1a6f21e MG |
854 | * note that the LRU is being scanned too quickly and the |
855 | * caller can stall after page list has been processed. | |
283aba9f MG |
856 | * |
857 | * 2) Global reclaim encounters a page, memcg encounters a | |
858 | * page that is not marked for immediate reclaim or | |
859 | * the caller does not have __GFP_IO. In this case mark | |
860 | * the page for immediate reclaim and continue scanning. | |
861 | * | |
862 | * __GFP_IO is checked because a loop driver thread might | |
863 | * enter reclaim, and deadlock if it waits on a page for | |
864 | * which it is needed to do the write (loop masks off | |
865 | * __GFP_IO|__GFP_FS for this reason); but more thought | |
866 | * would probably show more reasons. | |
867 | * | |
868 | * Don't require __GFP_FS, since we're not going into the | |
869 | * FS, just waiting on its writeback completion. Worryingly, | |
870 | * ext4 gfs2 and xfs allocate pages with | |
871 | * grab_cache_page_write_begin(,,AOP_FLAG_NOFS), so testing | |
872 | * may_enter_fs here is liable to OOM on them. | |
873 | * | |
874 | * 3) memcg encounters a page that is not already marked | |
875 | * PageReclaim. memcg does not have any dirty pages | |
876 | * throttling so we could easily OOM just because too many | |
877 | * pages are in writeback and there is nothing else to | |
878 | * reclaim. Wait for the writeback to complete. | |
879 | */ | |
c661b078 | 880 | if (PageWriteback(page)) { |
283aba9f MG |
881 | /* Case 1 above */ |
882 | if (current_is_kswapd() && | |
883 | PageReclaim(page) && | |
884 | zone_is_reclaim_writeback(zone)) { | |
b1a6f21e MG |
885 | nr_immediate++; |
886 | goto keep_locked; | |
283aba9f MG |
887 | |
888 | /* Case 2 above */ | |
889 | } else if (global_reclaim(sc) || | |
c3b94f44 HD |
890 | !PageReclaim(page) || !(sc->gfp_mask & __GFP_IO)) { |
891 | /* | |
892 | * This is slightly racy - end_page_writeback() | |
893 | * might have just cleared PageReclaim, then | |
894 | * setting PageReclaim here end up interpreted | |
895 | * as PageReadahead - but that does not matter | |
896 | * enough to care. What we do want is for this | |
897 | * page to have PageReclaim set next time memcg | |
898 | * reclaim reaches the tests above, so it will | |
899 | * then wait_on_page_writeback() to avoid OOM; | |
900 | * and it's also appropriate in global reclaim. | |
901 | */ | |
902 | SetPageReclaim(page); | |
e62e384e | 903 | nr_writeback++; |
283aba9f | 904 | |
c3b94f44 | 905 | goto keep_locked; |
283aba9f MG |
906 | |
907 | /* Case 3 above */ | |
908 | } else { | |
909 | wait_on_page_writeback(page); | |
e62e384e | 910 | } |
c661b078 | 911 | } |
1da177e4 | 912 | |
02c6de8d MK |
913 | if (!force_reclaim) |
914 | references = page_check_references(page, sc); | |
915 | ||
dfc8d636 JW |
916 | switch (references) { |
917 | case PAGEREF_ACTIVATE: | |
1da177e4 | 918 | goto activate_locked; |
64574746 JW |
919 | case PAGEREF_KEEP: |
920 | goto keep_locked; | |
dfc8d636 JW |
921 | case PAGEREF_RECLAIM: |
922 | case PAGEREF_RECLAIM_CLEAN: | |
923 | ; /* try to reclaim the page below */ | |
924 | } | |
1da177e4 | 925 | |
1da177e4 LT |
926 | /* |
927 | * Anonymous process memory has backing store? | |
928 | * Try to allocate it some swap space here. | |
929 | */ | |
b291f000 | 930 | if (PageAnon(page) && !PageSwapCache(page)) { |
63eb6b93 HD |
931 | if (!(sc->gfp_mask & __GFP_IO)) |
932 | goto keep_locked; | |
5bc7b8ac | 933 | if (!add_to_swap(page, page_list)) |
1da177e4 | 934 | goto activate_locked; |
63eb6b93 | 935 | may_enter_fs = 1; |
1da177e4 | 936 | |
e2be15f6 MG |
937 | /* Adding to swap updated mapping */ |
938 | mapping = page_mapping(page); | |
939 | } | |
1da177e4 LT |
940 | |
941 | /* | |
942 | * The page is mapped into the page tables of one or more | |
943 | * processes. Try to unmap it here. | |
944 | */ | |
945 | if (page_mapped(page) && mapping) { | |
02c6de8d | 946 | switch (try_to_unmap(page, ttu_flags)) { |
1da177e4 LT |
947 | case SWAP_FAIL: |
948 | goto activate_locked; | |
949 | case SWAP_AGAIN: | |
950 | goto keep_locked; | |
b291f000 NP |
951 | case SWAP_MLOCK: |
952 | goto cull_mlocked; | |
1da177e4 LT |
953 | case SWAP_SUCCESS: |
954 | ; /* try to free the page below */ | |
955 | } | |
956 | } | |
957 | ||
958 | if (PageDirty(page)) { | |
ee72886d MG |
959 | /* |
960 | * Only kswapd can writeback filesystem pages to | |
d43006d5 MG |
961 | * avoid risk of stack overflow but only writeback |
962 | * if many dirty pages have been encountered. | |
ee72886d | 963 | */ |
f84f6e2b | 964 | if (page_is_file_cache(page) && |
9e3b2f8c | 965 | (!current_is_kswapd() || |
d43006d5 | 966 | !zone_is_reclaim_dirty(zone))) { |
49ea7eb6 MG |
967 | /* |
968 | * Immediately reclaim when written back. | |
969 | * Similar in principal to deactivate_page() | |
970 | * except we already have the page isolated | |
971 | * and know it's dirty | |
972 | */ | |
973 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); | |
974 | SetPageReclaim(page); | |
975 | ||
ee72886d MG |
976 | goto keep_locked; |
977 | } | |
978 | ||
dfc8d636 | 979 | if (references == PAGEREF_RECLAIM_CLEAN) |
1da177e4 | 980 | goto keep_locked; |
4dd4b920 | 981 | if (!may_enter_fs) |
1da177e4 | 982 | goto keep_locked; |
52a8363e | 983 | if (!sc->may_writepage) |
1da177e4 LT |
984 | goto keep_locked; |
985 | ||
986 | /* Page is dirty, try to write it out here */ | |
7d3579e8 | 987 | switch (pageout(page, mapping, sc)) { |
1da177e4 LT |
988 | case PAGE_KEEP: |
989 | goto keep_locked; | |
990 | case PAGE_ACTIVATE: | |
991 | goto activate_locked; | |
992 | case PAGE_SUCCESS: | |
7d3579e8 | 993 | if (PageWriteback(page)) |
41ac1999 | 994 | goto keep; |
7d3579e8 | 995 | if (PageDirty(page)) |
1da177e4 | 996 | goto keep; |
7d3579e8 | 997 | |
1da177e4 LT |
998 | /* |
999 | * A synchronous write - probably a ramdisk. Go | |
1000 | * ahead and try to reclaim the page. | |
1001 | */ | |
529ae9aa | 1002 | if (!trylock_page(page)) |
1da177e4 LT |
1003 | goto keep; |
1004 | if (PageDirty(page) || PageWriteback(page)) | |
1005 | goto keep_locked; | |
1006 | mapping = page_mapping(page); | |
1007 | case PAGE_CLEAN: | |
1008 | ; /* try to free the page below */ | |
1009 | } | |
1010 | } | |
1011 | ||
1012 | /* | |
1013 | * If the page has buffers, try to free the buffer mappings | |
1014 | * associated with this page. If we succeed we try to free | |
1015 | * the page as well. | |
1016 | * | |
1017 | * We do this even if the page is PageDirty(). | |
1018 | * try_to_release_page() does not perform I/O, but it is | |
1019 | * possible for a page to have PageDirty set, but it is actually | |
1020 | * clean (all its buffers are clean). This happens if the | |
1021 | * buffers were written out directly, with submit_bh(). ext3 | |
894bc310 | 1022 | * will do this, as well as the blockdev mapping. |
1da177e4 LT |
1023 | * try_to_release_page() will discover that cleanness and will |
1024 | * drop the buffers and mark the page clean - it can be freed. | |
1025 | * | |
1026 | * Rarely, pages can have buffers and no ->mapping. These are | |
1027 | * the pages which were not successfully invalidated in | |
1028 | * truncate_complete_page(). We try to drop those buffers here | |
1029 | * and if that worked, and the page is no longer mapped into | |
1030 | * process address space (page_count == 1) it can be freed. | |
1031 | * Otherwise, leave the page on the LRU so it is swappable. | |
1032 | */ | |
266cf658 | 1033 | if (page_has_private(page)) { |
1da177e4 LT |
1034 | if (!try_to_release_page(page, sc->gfp_mask)) |
1035 | goto activate_locked; | |
e286781d NP |
1036 | if (!mapping && page_count(page) == 1) { |
1037 | unlock_page(page); | |
1038 | if (put_page_testzero(page)) | |
1039 | goto free_it; | |
1040 | else { | |
1041 | /* | |
1042 | * rare race with speculative reference. | |
1043 | * the speculative reference will free | |
1044 | * this page shortly, so we may | |
1045 | * increment nr_reclaimed here (and | |
1046 | * leave it off the LRU). | |
1047 | */ | |
1048 | nr_reclaimed++; | |
1049 | continue; | |
1050 | } | |
1051 | } | |
1da177e4 LT |
1052 | } |
1053 | ||
e286781d | 1054 | if (!mapping || !__remove_mapping(mapping, page)) |
49d2e9cc | 1055 | goto keep_locked; |
1da177e4 | 1056 | |
a978d6f5 NP |
1057 | /* |
1058 | * At this point, we have no other references and there is | |
1059 | * no way to pick any more up (removed from LRU, removed | |
1060 | * from pagecache). Can use non-atomic bitops now (and | |
1061 | * we obviously don't have to worry about waking up a process | |
1062 | * waiting on the page lock, because there are no references. | |
1063 | */ | |
1064 | __clear_page_locked(page); | |
e286781d | 1065 | free_it: |
05ff5137 | 1066 | nr_reclaimed++; |
abe4c3b5 MG |
1067 | |
1068 | /* | |
1069 | * Is there need to periodically free_page_list? It would | |
1070 | * appear not as the counts should be low | |
1071 | */ | |
1072 | list_add(&page->lru, &free_pages); | |
1da177e4 LT |
1073 | continue; |
1074 | ||
b291f000 | 1075 | cull_mlocked: |
63d6c5ad HD |
1076 | if (PageSwapCache(page)) |
1077 | try_to_free_swap(page); | |
b291f000 NP |
1078 | unlock_page(page); |
1079 | putback_lru_page(page); | |
1080 | continue; | |
1081 | ||
1da177e4 | 1082 | activate_locked: |
68a22394 RR |
1083 | /* Not a candidate for swapping, so reclaim swap space. */ |
1084 | if (PageSwapCache(page) && vm_swap_full()) | |
a2c43eed | 1085 | try_to_free_swap(page); |
894bc310 | 1086 | VM_BUG_ON(PageActive(page)); |
1da177e4 LT |
1087 | SetPageActive(page); |
1088 | pgactivate++; | |
1089 | keep_locked: | |
1090 | unlock_page(page); | |
1091 | keep: | |
1092 | list_add(&page->lru, &ret_pages); | |
b291f000 | 1093 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); |
1da177e4 | 1094 | } |
abe4c3b5 | 1095 | |
cc59850e | 1096 | free_hot_cold_page_list(&free_pages, 1); |
abe4c3b5 | 1097 | |
1da177e4 | 1098 | list_splice(&ret_pages, page_list); |
f8891e5e | 1099 | count_vm_events(PGACTIVATE, pgactivate); |
69980e31 | 1100 | mem_cgroup_uncharge_end(); |
8e950282 MG |
1101 | *ret_nr_dirty += nr_dirty; |
1102 | *ret_nr_congested += nr_congested; | |
d43006d5 | 1103 | *ret_nr_unqueued_dirty += nr_unqueued_dirty; |
92df3a72 | 1104 | *ret_nr_writeback += nr_writeback; |
b1a6f21e | 1105 | *ret_nr_immediate += nr_immediate; |
05ff5137 | 1106 | return nr_reclaimed; |
1da177e4 LT |
1107 | } |
1108 | ||
02c6de8d MK |
1109 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, |
1110 | struct list_head *page_list) | |
1111 | { | |
1112 | struct scan_control sc = { | |
1113 | .gfp_mask = GFP_KERNEL, | |
1114 | .priority = DEF_PRIORITY, | |
1115 | .may_unmap = 1, | |
1116 | }; | |
8e950282 | 1117 | unsigned long ret, dummy1, dummy2, dummy3, dummy4, dummy5; |
02c6de8d MK |
1118 | struct page *page, *next; |
1119 | LIST_HEAD(clean_pages); | |
1120 | ||
1121 | list_for_each_entry_safe(page, next, page_list, lru) { | |
1122 | if (page_is_file_cache(page) && !PageDirty(page)) { | |
1123 | ClearPageActive(page); | |
1124 | list_move(&page->lru, &clean_pages); | |
1125 | } | |
1126 | } | |
1127 | ||
1128 | ret = shrink_page_list(&clean_pages, zone, &sc, | |
8e950282 MG |
1129 | TTU_UNMAP|TTU_IGNORE_ACCESS, |
1130 | &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, true); | |
02c6de8d MK |
1131 | list_splice(&clean_pages, page_list); |
1132 | __mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret); | |
1133 | return ret; | |
1134 | } | |
1135 | ||
5ad333eb AW |
1136 | /* |
1137 | * Attempt to remove the specified page from its LRU. Only take this page | |
1138 | * if it is of the appropriate PageActive status. Pages which are being | |
1139 | * freed elsewhere are also ignored. | |
1140 | * | |
1141 | * page: page to consider | |
1142 | * mode: one of the LRU isolation modes defined above | |
1143 | * | |
1144 | * returns 0 on success, -ve errno on failure. | |
1145 | */ | |
f3fd4a61 | 1146 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) |
5ad333eb AW |
1147 | { |
1148 | int ret = -EINVAL; | |
1149 | ||
1150 | /* Only take pages on the LRU. */ | |
1151 | if (!PageLRU(page)) | |
1152 | return ret; | |
1153 | ||
e46a2879 MK |
1154 | /* Compaction should not handle unevictable pages but CMA can do so */ |
1155 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) | |
894bc310 LS |
1156 | return ret; |
1157 | ||
5ad333eb | 1158 | ret = -EBUSY; |
08e552c6 | 1159 | |
c8244935 MG |
1160 | /* |
1161 | * To minimise LRU disruption, the caller can indicate that it only | |
1162 | * wants to isolate pages it will be able to operate on without | |
1163 | * blocking - clean pages for the most part. | |
1164 | * | |
1165 | * ISOLATE_CLEAN means that only clean pages should be isolated. This | |
1166 | * is used by reclaim when it is cannot write to backing storage | |
1167 | * | |
1168 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages | |
1169 | * that it is possible to migrate without blocking | |
1170 | */ | |
1171 | if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) { | |
1172 | /* All the caller can do on PageWriteback is block */ | |
1173 | if (PageWriteback(page)) | |
1174 | return ret; | |
1175 | ||
1176 | if (PageDirty(page)) { | |
1177 | struct address_space *mapping; | |
1178 | ||
1179 | /* ISOLATE_CLEAN means only clean pages */ | |
1180 | if (mode & ISOLATE_CLEAN) | |
1181 | return ret; | |
1182 | ||
1183 | /* | |
1184 | * Only pages without mappings or that have a | |
1185 | * ->migratepage callback are possible to migrate | |
1186 | * without blocking | |
1187 | */ | |
1188 | mapping = page_mapping(page); | |
1189 | if (mapping && !mapping->a_ops->migratepage) | |
1190 | return ret; | |
1191 | } | |
1192 | } | |
39deaf85 | 1193 | |
f80c0673 MK |
1194 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
1195 | return ret; | |
1196 | ||
5ad333eb AW |
1197 | if (likely(get_page_unless_zero(page))) { |
1198 | /* | |
1199 | * Be careful not to clear PageLRU until after we're | |
1200 | * sure the page is not being freed elsewhere -- the | |
1201 | * page release code relies on it. | |
1202 | */ | |
1203 | ClearPageLRU(page); | |
1204 | ret = 0; | |
1205 | } | |
1206 | ||
1207 | return ret; | |
1208 | } | |
1209 | ||
1da177e4 LT |
1210 | /* |
1211 | * zone->lru_lock is heavily contended. Some of the functions that | |
1212 | * shrink the lists perform better by taking out a batch of pages | |
1213 | * and working on them outside the LRU lock. | |
1214 | * | |
1215 | * For pagecache intensive workloads, this function is the hottest | |
1216 | * spot in the kernel (apart from copy_*_user functions). | |
1217 | * | |
1218 | * Appropriate locks must be held before calling this function. | |
1219 | * | |
1220 | * @nr_to_scan: The number of pages to look through on the list. | |
5dc35979 | 1221 | * @lruvec: The LRU vector to pull pages from. |
1da177e4 | 1222 | * @dst: The temp list to put pages on to. |
f626012d | 1223 | * @nr_scanned: The number of pages that were scanned. |
fe2c2a10 | 1224 | * @sc: The scan_control struct for this reclaim session |
5ad333eb | 1225 | * @mode: One of the LRU isolation modes |
3cb99451 | 1226 | * @lru: LRU list id for isolating |
1da177e4 LT |
1227 | * |
1228 | * returns how many pages were moved onto *@dst. | |
1229 | */ | |
69e05944 | 1230 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
5dc35979 | 1231 | struct lruvec *lruvec, struct list_head *dst, |
fe2c2a10 | 1232 | unsigned long *nr_scanned, struct scan_control *sc, |
3cb99451 | 1233 | isolate_mode_t mode, enum lru_list lru) |
1da177e4 | 1234 | { |
75b00af7 | 1235 | struct list_head *src = &lruvec->lists[lru]; |
69e05944 | 1236 | unsigned long nr_taken = 0; |
c9b02d97 | 1237 | unsigned long scan; |
1da177e4 | 1238 | |
c9b02d97 | 1239 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
5ad333eb | 1240 | struct page *page; |
fa9add64 | 1241 | int nr_pages; |
5ad333eb | 1242 | |
1da177e4 LT |
1243 | page = lru_to_page(src); |
1244 | prefetchw_prev_lru_page(page, src, flags); | |
1245 | ||
725d704e | 1246 | VM_BUG_ON(!PageLRU(page)); |
8d438f96 | 1247 | |
f3fd4a61 | 1248 | switch (__isolate_lru_page(page, mode)) { |
5ad333eb | 1249 | case 0: |
fa9add64 HD |
1250 | nr_pages = hpage_nr_pages(page); |
1251 | mem_cgroup_update_lru_size(lruvec, lru, -nr_pages); | |
5ad333eb | 1252 | list_move(&page->lru, dst); |
fa9add64 | 1253 | nr_taken += nr_pages; |
5ad333eb AW |
1254 | break; |
1255 | ||
1256 | case -EBUSY: | |
1257 | /* else it is being freed elsewhere */ | |
1258 | list_move(&page->lru, src); | |
1259 | continue; | |
46453a6e | 1260 | |
5ad333eb AW |
1261 | default: |
1262 | BUG(); | |
1263 | } | |
1da177e4 LT |
1264 | } |
1265 | ||
f626012d | 1266 | *nr_scanned = scan; |
75b00af7 HD |
1267 | trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan, |
1268 | nr_taken, mode, is_file_lru(lru)); | |
1da177e4 LT |
1269 | return nr_taken; |
1270 | } | |
1271 | ||
62695a84 NP |
1272 | /** |
1273 | * isolate_lru_page - tries to isolate a page from its LRU list | |
1274 | * @page: page to isolate from its LRU list | |
1275 | * | |
1276 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the | |
1277 | * vmstat statistic corresponding to whatever LRU list the page was on. | |
1278 | * | |
1279 | * Returns 0 if the page was removed from an LRU list. | |
1280 | * Returns -EBUSY if the page was not on an LRU list. | |
1281 | * | |
1282 | * The returned page will have PageLRU() cleared. If it was found on | |
894bc310 LS |
1283 | * the active list, it will have PageActive set. If it was found on |
1284 | * the unevictable list, it will have the PageUnevictable bit set. That flag | |
1285 | * may need to be cleared by the caller before letting the page go. | |
62695a84 NP |
1286 | * |
1287 | * The vmstat statistic corresponding to the list on which the page was | |
1288 | * found will be decremented. | |
1289 | * | |
1290 | * Restrictions: | |
1291 | * (1) Must be called with an elevated refcount on the page. This is a | |
1292 | * fundamentnal difference from isolate_lru_pages (which is called | |
1293 | * without a stable reference). | |
1294 | * (2) the lru_lock must not be held. | |
1295 | * (3) interrupts must be enabled. | |
1296 | */ | |
1297 | int isolate_lru_page(struct page *page) | |
1298 | { | |
1299 | int ret = -EBUSY; | |
1300 | ||
0c917313 KK |
1301 | VM_BUG_ON(!page_count(page)); |
1302 | ||
62695a84 NP |
1303 | if (PageLRU(page)) { |
1304 | struct zone *zone = page_zone(page); | |
fa9add64 | 1305 | struct lruvec *lruvec; |
62695a84 NP |
1306 | |
1307 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 1308 | lruvec = mem_cgroup_page_lruvec(page, zone); |
0c917313 | 1309 | if (PageLRU(page)) { |
894bc310 | 1310 | int lru = page_lru(page); |
0c917313 | 1311 | get_page(page); |
62695a84 | 1312 | ClearPageLRU(page); |
fa9add64 HD |
1313 | del_page_from_lru_list(page, lruvec, lru); |
1314 | ret = 0; | |
62695a84 NP |
1315 | } |
1316 | spin_unlock_irq(&zone->lru_lock); | |
1317 | } | |
1318 | return ret; | |
1319 | } | |
1320 | ||
35cd7815 | 1321 | /* |
d37dd5dc FW |
1322 | * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and |
1323 | * then get resheduled. When there are massive number of tasks doing page | |
1324 | * allocation, such sleeping direct reclaimers may keep piling up on each CPU, | |
1325 | * the LRU list will go small and be scanned faster than necessary, leading to | |
1326 | * unnecessary swapping, thrashing and OOM. | |
35cd7815 RR |
1327 | */ |
1328 | static int too_many_isolated(struct zone *zone, int file, | |
1329 | struct scan_control *sc) | |
1330 | { | |
1331 | unsigned long inactive, isolated; | |
1332 | ||
1333 | if (current_is_kswapd()) | |
1334 | return 0; | |
1335 | ||
89b5fae5 | 1336 | if (!global_reclaim(sc)) |
35cd7815 RR |
1337 | return 0; |
1338 | ||
1339 | if (file) { | |
1340 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); | |
1341 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); | |
1342 | } else { | |
1343 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1344 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); | |
1345 | } | |
1346 | ||
3cf23841 FW |
1347 | /* |
1348 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they | |
1349 | * won't get blocked by normal direct-reclaimers, forming a circular | |
1350 | * deadlock. | |
1351 | */ | |
1352 | if ((sc->gfp_mask & GFP_IOFS) == GFP_IOFS) | |
1353 | inactive >>= 3; | |
1354 | ||
35cd7815 RR |
1355 | return isolated > inactive; |
1356 | } | |
1357 | ||
66635629 | 1358 | static noinline_for_stack void |
75b00af7 | 1359 | putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list) |
66635629 | 1360 | { |
27ac81d8 KK |
1361 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1362 | struct zone *zone = lruvec_zone(lruvec); | |
3f79768f | 1363 | LIST_HEAD(pages_to_free); |
66635629 | 1364 | |
66635629 MG |
1365 | /* |
1366 | * Put back any unfreeable pages. | |
1367 | */ | |
66635629 | 1368 | while (!list_empty(page_list)) { |
3f79768f | 1369 | struct page *page = lru_to_page(page_list); |
66635629 | 1370 | int lru; |
3f79768f | 1371 | |
66635629 MG |
1372 | VM_BUG_ON(PageLRU(page)); |
1373 | list_del(&page->lru); | |
39b5f29a | 1374 | if (unlikely(!page_evictable(page))) { |
66635629 MG |
1375 | spin_unlock_irq(&zone->lru_lock); |
1376 | putback_lru_page(page); | |
1377 | spin_lock_irq(&zone->lru_lock); | |
1378 | continue; | |
1379 | } | |
fa9add64 HD |
1380 | |
1381 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
1382 | ||
7a608572 | 1383 | SetPageLRU(page); |
66635629 | 1384 | lru = page_lru(page); |
fa9add64 HD |
1385 | add_page_to_lru_list(page, lruvec, lru); |
1386 | ||
66635629 MG |
1387 | if (is_active_lru(lru)) { |
1388 | int file = is_file_lru(lru); | |
9992af10 RR |
1389 | int numpages = hpage_nr_pages(page); |
1390 | reclaim_stat->recent_rotated[file] += numpages; | |
66635629 | 1391 | } |
2bcf8879 HD |
1392 | if (put_page_testzero(page)) { |
1393 | __ClearPageLRU(page); | |
1394 | __ClearPageActive(page); | |
fa9add64 | 1395 | del_page_from_lru_list(page, lruvec, lru); |
2bcf8879 HD |
1396 | |
1397 | if (unlikely(PageCompound(page))) { | |
1398 | spin_unlock_irq(&zone->lru_lock); | |
1399 | (*get_compound_page_dtor(page))(page); | |
1400 | spin_lock_irq(&zone->lru_lock); | |
1401 | } else | |
1402 | list_add(&page->lru, &pages_to_free); | |
66635629 MG |
1403 | } |
1404 | } | |
66635629 | 1405 | |
3f79768f HD |
1406 | /* |
1407 | * To save our caller's stack, now use input list for pages to free. | |
1408 | */ | |
1409 | list_splice(&pages_to_free, page_list); | |
66635629 MG |
1410 | } |
1411 | ||
1da177e4 | 1412 | /* |
1742f19f AM |
1413 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
1414 | * of reclaimed pages | |
1da177e4 | 1415 | */ |
66635629 | 1416 | static noinline_for_stack unsigned long |
1a93be0e | 1417 | shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec, |
9e3b2f8c | 1418 | struct scan_control *sc, enum lru_list lru) |
1da177e4 LT |
1419 | { |
1420 | LIST_HEAD(page_list); | |
e247dbce | 1421 | unsigned long nr_scanned; |
05ff5137 | 1422 | unsigned long nr_reclaimed = 0; |
e247dbce | 1423 | unsigned long nr_taken; |
8e950282 MG |
1424 | unsigned long nr_dirty = 0; |
1425 | unsigned long nr_congested = 0; | |
e2be15f6 | 1426 | unsigned long nr_unqueued_dirty = 0; |
92df3a72 | 1427 | unsigned long nr_writeback = 0; |
b1a6f21e | 1428 | unsigned long nr_immediate = 0; |
f3fd4a61 | 1429 | isolate_mode_t isolate_mode = 0; |
3cb99451 | 1430 | int file = is_file_lru(lru); |
1a93be0e KK |
1431 | struct zone *zone = lruvec_zone(lruvec); |
1432 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; | |
78dc583d | 1433 | |
35cd7815 | 1434 | while (unlikely(too_many_isolated(zone, file, sc))) { |
58355c78 | 1435 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
35cd7815 RR |
1436 | |
1437 | /* We are about to die and free our memory. Return now. */ | |
1438 | if (fatal_signal_pending(current)) | |
1439 | return SWAP_CLUSTER_MAX; | |
1440 | } | |
1441 | ||
1da177e4 | 1442 | lru_add_drain(); |
f80c0673 MK |
1443 | |
1444 | if (!sc->may_unmap) | |
61317289 | 1445 | isolate_mode |= ISOLATE_UNMAPPED; |
f80c0673 | 1446 | if (!sc->may_writepage) |
61317289 | 1447 | isolate_mode |= ISOLATE_CLEAN; |
f80c0673 | 1448 | |
1da177e4 | 1449 | spin_lock_irq(&zone->lru_lock); |
b35ea17b | 1450 | |
5dc35979 KK |
1451 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, |
1452 | &nr_scanned, sc, isolate_mode, lru); | |
95d918fc KK |
1453 | |
1454 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); | |
1455 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); | |
1456 | ||
89b5fae5 | 1457 | if (global_reclaim(sc)) { |
e247dbce KM |
1458 | zone->pages_scanned += nr_scanned; |
1459 | if (current_is_kswapd()) | |
75b00af7 | 1460 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned); |
e247dbce | 1461 | else |
75b00af7 | 1462 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned); |
e247dbce | 1463 | } |
d563c050 | 1464 | spin_unlock_irq(&zone->lru_lock); |
b35ea17b | 1465 | |
d563c050 | 1466 | if (nr_taken == 0) |
66635629 | 1467 | return 0; |
5ad333eb | 1468 | |
02c6de8d | 1469 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP, |
8e950282 MG |
1470 | &nr_dirty, &nr_unqueued_dirty, &nr_congested, |
1471 | &nr_writeback, &nr_immediate, | |
1472 | false); | |
c661b078 | 1473 | |
3f79768f HD |
1474 | spin_lock_irq(&zone->lru_lock); |
1475 | ||
95d918fc | 1476 | reclaim_stat->recent_scanned[file] += nr_taken; |
d563c050 | 1477 | |
904249aa YH |
1478 | if (global_reclaim(sc)) { |
1479 | if (current_is_kswapd()) | |
1480 | __count_zone_vm_events(PGSTEAL_KSWAPD, zone, | |
1481 | nr_reclaimed); | |
1482 | else | |
1483 | __count_zone_vm_events(PGSTEAL_DIRECT, zone, | |
1484 | nr_reclaimed); | |
1485 | } | |
a74609fa | 1486 | |
27ac81d8 | 1487 | putback_inactive_pages(lruvec, &page_list); |
3f79768f | 1488 | |
95d918fc | 1489 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
3f79768f HD |
1490 | |
1491 | spin_unlock_irq(&zone->lru_lock); | |
1492 | ||
1493 | free_hot_cold_page_list(&page_list, 1); | |
e11da5b4 | 1494 | |
92df3a72 MG |
1495 | /* |
1496 | * If reclaim is isolating dirty pages under writeback, it implies | |
1497 | * that the long-lived page allocation rate is exceeding the page | |
1498 | * laundering rate. Either the global limits are not being effective | |
1499 | * at throttling processes due to the page distribution throughout | |
1500 | * zones or there is heavy usage of a slow backing device. The | |
1501 | * only option is to throttle from reclaim context which is not ideal | |
1502 | * as there is no guarantee the dirtying process is throttled in the | |
1503 | * same way balance_dirty_pages() manages. | |
1504 | * | |
8e950282 MG |
1505 | * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number |
1506 | * of pages under pages flagged for immediate reclaim and stall if any | |
1507 | * are encountered in the nr_immediate check below. | |
92df3a72 | 1508 | */ |
918fc718 | 1509 | if (nr_writeback && nr_writeback == nr_taken) |
283aba9f | 1510 | zone_set_flag(zone, ZONE_WRITEBACK); |
92df3a72 | 1511 | |
d43006d5 | 1512 | /* |
b1a6f21e MG |
1513 | * memcg will stall in page writeback so only consider forcibly |
1514 | * stalling for global reclaim | |
d43006d5 | 1515 | */ |
b1a6f21e | 1516 | if (global_reclaim(sc)) { |
8e950282 MG |
1517 | /* |
1518 | * Tag a zone as congested if all the dirty pages scanned were | |
1519 | * backed by a congested BDI and wait_iff_congested will stall. | |
1520 | */ | |
1521 | if (nr_dirty && nr_dirty == nr_congested) | |
1522 | zone_set_flag(zone, ZONE_CONGESTED); | |
1523 | ||
b1a6f21e MG |
1524 | /* |
1525 | * If dirty pages are scanned that are not queued for IO, it | |
1526 | * implies that flushers are not keeping up. In this case, flag | |
1527 | * the zone ZONE_TAIL_LRU_DIRTY and kswapd will start writing | |
1528 | * pages from reclaim context. It will forcibly stall in the | |
1529 | * next check. | |
1530 | */ | |
1531 | if (nr_unqueued_dirty == nr_taken) | |
1532 | zone_set_flag(zone, ZONE_TAIL_LRU_DIRTY); | |
1533 | ||
1534 | /* | |
1535 | * In addition, if kswapd scans pages marked marked for | |
1536 | * immediate reclaim and under writeback (nr_immediate), it | |
1537 | * implies that pages are cycling through the LRU faster than | |
1538 | * they are written so also forcibly stall. | |
1539 | */ | |
1540 | if (nr_unqueued_dirty == nr_taken || nr_immediate) | |
1541 | congestion_wait(BLK_RW_ASYNC, HZ/10); | |
e2be15f6 | 1542 | } |
d43006d5 | 1543 | |
8e950282 MG |
1544 | /* |
1545 | * Stall direct reclaim for IO completions if underlying BDIs or zone | |
1546 | * is congested. Allow kswapd to continue until it starts encountering | |
1547 | * unqueued dirty pages or cycling through the LRU too quickly. | |
1548 | */ | |
1549 | if (!sc->hibernation_mode && !current_is_kswapd()) | |
1550 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); | |
1551 | ||
e11da5b4 MG |
1552 | trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id, |
1553 | zone_idx(zone), | |
1554 | nr_scanned, nr_reclaimed, | |
9e3b2f8c | 1555 | sc->priority, |
23b9da55 | 1556 | trace_shrink_flags(file)); |
05ff5137 | 1557 | return nr_reclaimed; |
1da177e4 LT |
1558 | } |
1559 | ||
1560 | /* | |
1561 | * This moves pages from the active list to the inactive list. | |
1562 | * | |
1563 | * We move them the other way if the page is referenced by one or more | |
1564 | * processes, from rmap. | |
1565 | * | |
1566 | * If the pages are mostly unmapped, the processing is fast and it is | |
1567 | * appropriate to hold zone->lru_lock across the whole operation. But if | |
1568 | * the pages are mapped, the processing is slow (page_referenced()) so we | |
1569 | * should drop zone->lru_lock around each page. It's impossible to balance | |
1570 | * this, so instead we remove the pages from the LRU while processing them. | |
1571 | * It is safe to rely on PG_active against the non-LRU pages in here because | |
1572 | * nobody will play with that bit on a non-LRU page. | |
1573 | * | |
1574 | * The downside is that we have to touch page->_count against each page. | |
1575 | * But we had to alter page->flags anyway. | |
1576 | */ | |
1cfb419b | 1577 | |
fa9add64 | 1578 | static void move_active_pages_to_lru(struct lruvec *lruvec, |
3eb4140f | 1579 | struct list_head *list, |
2bcf8879 | 1580 | struct list_head *pages_to_free, |
3eb4140f WF |
1581 | enum lru_list lru) |
1582 | { | |
fa9add64 | 1583 | struct zone *zone = lruvec_zone(lruvec); |
3eb4140f | 1584 | unsigned long pgmoved = 0; |
3eb4140f | 1585 | struct page *page; |
fa9add64 | 1586 | int nr_pages; |
3eb4140f | 1587 | |
3eb4140f WF |
1588 | while (!list_empty(list)) { |
1589 | page = lru_to_page(list); | |
fa9add64 | 1590 | lruvec = mem_cgroup_page_lruvec(page, zone); |
3eb4140f WF |
1591 | |
1592 | VM_BUG_ON(PageLRU(page)); | |
1593 | SetPageLRU(page); | |
1594 | ||
fa9add64 HD |
1595 | nr_pages = hpage_nr_pages(page); |
1596 | mem_cgroup_update_lru_size(lruvec, lru, nr_pages); | |
925b7673 | 1597 | list_move(&page->lru, &lruvec->lists[lru]); |
fa9add64 | 1598 | pgmoved += nr_pages; |
3eb4140f | 1599 | |
2bcf8879 HD |
1600 | if (put_page_testzero(page)) { |
1601 | __ClearPageLRU(page); | |
1602 | __ClearPageActive(page); | |
fa9add64 | 1603 | del_page_from_lru_list(page, lruvec, lru); |
2bcf8879 HD |
1604 | |
1605 | if (unlikely(PageCompound(page))) { | |
1606 | spin_unlock_irq(&zone->lru_lock); | |
1607 | (*get_compound_page_dtor(page))(page); | |
1608 | spin_lock_irq(&zone->lru_lock); | |
1609 | } else | |
1610 | list_add(&page->lru, pages_to_free); | |
3eb4140f WF |
1611 | } |
1612 | } | |
1613 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); | |
1614 | if (!is_active_lru(lru)) | |
1615 | __count_vm_events(PGDEACTIVATE, pgmoved); | |
1616 | } | |
1cfb419b | 1617 | |
f626012d | 1618 | static void shrink_active_list(unsigned long nr_to_scan, |
1a93be0e | 1619 | struct lruvec *lruvec, |
f16015fb | 1620 | struct scan_control *sc, |
9e3b2f8c | 1621 | enum lru_list lru) |
1da177e4 | 1622 | { |
44c241f1 | 1623 | unsigned long nr_taken; |
f626012d | 1624 | unsigned long nr_scanned; |
6fe6b7e3 | 1625 | unsigned long vm_flags; |
1da177e4 | 1626 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
8cab4754 | 1627 | LIST_HEAD(l_active); |
b69408e8 | 1628 | LIST_HEAD(l_inactive); |
1da177e4 | 1629 | struct page *page; |
1a93be0e | 1630 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
44c241f1 | 1631 | unsigned long nr_rotated = 0; |
f3fd4a61 | 1632 | isolate_mode_t isolate_mode = 0; |
3cb99451 | 1633 | int file = is_file_lru(lru); |
1a93be0e | 1634 | struct zone *zone = lruvec_zone(lruvec); |
1da177e4 LT |
1635 | |
1636 | lru_add_drain(); | |
f80c0673 MK |
1637 | |
1638 | if (!sc->may_unmap) | |
61317289 | 1639 | isolate_mode |= ISOLATE_UNMAPPED; |
f80c0673 | 1640 | if (!sc->may_writepage) |
61317289 | 1641 | isolate_mode |= ISOLATE_CLEAN; |
f80c0673 | 1642 | |
1da177e4 | 1643 | spin_lock_irq(&zone->lru_lock); |
925b7673 | 1644 | |
5dc35979 KK |
1645 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, |
1646 | &nr_scanned, sc, isolate_mode, lru); | |
89b5fae5 | 1647 | if (global_reclaim(sc)) |
f626012d | 1648 | zone->pages_scanned += nr_scanned; |
89b5fae5 | 1649 | |
b7c46d15 | 1650 | reclaim_stat->recent_scanned[file] += nr_taken; |
1cfb419b | 1651 | |
f626012d | 1652 | __count_zone_vm_events(PGREFILL, zone, nr_scanned); |
3cb99451 | 1653 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
a731286d | 1654 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
1da177e4 LT |
1655 | spin_unlock_irq(&zone->lru_lock); |
1656 | ||
1da177e4 LT |
1657 | while (!list_empty(&l_hold)) { |
1658 | cond_resched(); | |
1659 | page = lru_to_page(&l_hold); | |
1660 | list_del(&page->lru); | |
7e9cd484 | 1661 | |
39b5f29a | 1662 | if (unlikely(!page_evictable(page))) { |
894bc310 LS |
1663 | putback_lru_page(page); |
1664 | continue; | |
1665 | } | |
1666 | ||
cc715d99 MG |
1667 | if (unlikely(buffer_heads_over_limit)) { |
1668 | if (page_has_private(page) && trylock_page(page)) { | |
1669 | if (page_has_private(page)) | |
1670 | try_to_release_page(page, 0); | |
1671 | unlock_page(page); | |
1672 | } | |
1673 | } | |
1674 | ||
c3ac9a8a JW |
1675 | if (page_referenced(page, 0, sc->target_mem_cgroup, |
1676 | &vm_flags)) { | |
9992af10 | 1677 | nr_rotated += hpage_nr_pages(page); |
8cab4754 WF |
1678 | /* |
1679 | * Identify referenced, file-backed active pages and | |
1680 | * give them one more trip around the active list. So | |
1681 | * that executable code get better chances to stay in | |
1682 | * memory under moderate memory pressure. Anon pages | |
1683 | * are not likely to be evicted by use-once streaming | |
1684 | * IO, plus JVM can create lots of anon VM_EXEC pages, | |
1685 | * so we ignore them here. | |
1686 | */ | |
41e20983 | 1687 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
8cab4754 WF |
1688 | list_add(&page->lru, &l_active); |
1689 | continue; | |
1690 | } | |
1691 | } | |
7e9cd484 | 1692 | |
5205e56e | 1693 | ClearPageActive(page); /* we are de-activating */ |
1da177e4 LT |
1694 | list_add(&page->lru, &l_inactive); |
1695 | } | |
1696 | ||
b555749a | 1697 | /* |
8cab4754 | 1698 | * Move pages back to the lru list. |
b555749a | 1699 | */ |
2a1dc509 | 1700 | spin_lock_irq(&zone->lru_lock); |
556adecb | 1701 | /* |
8cab4754 WF |
1702 | * Count referenced pages from currently used mappings as rotated, |
1703 | * even though only some of them are actually re-activated. This | |
1704 | * helps balance scan pressure between file and anonymous pages in | |
1705 | * get_scan_ratio. | |
7e9cd484 | 1706 | */ |
b7c46d15 | 1707 | reclaim_stat->recent_rotated[file] += nr_rotated; |
556adecb | 1708 | |
fa9add64 HD |
1709 | move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru); |
1710 | move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE); | |
a731286d | 1711 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
f8891e5e | 1712 | spin_unlock_irq(&zone->lru_lock); |
2bcf8879 HD |
1713 | |
1714 | free_hot_cold_page_list(&l_hold, 1); | |
1da177e4 LT |
1715 | } |
1716 | ||
74e3f3c3 | 1717 | #ifdef CONFIG_SWAP |
14797e23 | 1718 | static int inactive_anon_is_low_global(struct zone *zone) |
f89eb90e KM |
1719 | { |
1720 | unsigned long active, inactive; | |
1721 | ||
1722 | active = zone_page_state(zone, NR_ACTIVE_ANON); | |
1723 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1724 | ||
1725 | if (inactive * zone->inactive_ratio < active) | |
1726 | return 1; | |
1727 | ||
1728 | return 0; | |
1729 | } | |
1730 | ||
14797e23 KM |
1731 | /** |
1732 | * inactive_anon_is_low - check if anonymous pages need to be deactivated | |
c56d5c7d | 1733 | * @lruvec: LRU vector to check |
14797e23 KM |
1734 | * |
1735 | * Returns true if the zone does not have enough inactive anon pages, | |
1736 | * meaning some active anon pages need to be deactivated. | |
1737 | */ | |
c56d5c7d | 1738 | static int inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1739 | { |
74e3f3c3 MK |
1740 | /* |
1741 | * If we don't have swap space, anonymous page deactivation | |
1742 | * is pointless. | |
1743 | */ | |
1744 | if (!total_swap_pages) | |
1745 | return 0; | |
1746 | ||
c3c787e8 | 1747 | if (!mem_cgroup_disabled()) |
c56d5c7d | 1748 | return mem_cgroup_inactive_anon_is_low(lruvec); |
f16015fb | 1749 | |
c56d5c7d | 1750 | return inactive_anon_is_low_global(lruvec_zone(lruvec)); |
14797e23 | 1751 | } |
74e3f3c3 | 1752 | #else |
c56d5c7d | 1753 | static inline int inactive_anon_is_low(struct lruvec *lruvec) |
74e3f3c3 MK |
1754 | { |
1755 | return 0; | |
1756 | } | |
1757 | #endif | |
14797e23 | 1758 | |
56e49d21 RR |
1759 | /** |
1760 | * inactive_file_is_low - check if file pages need to be deactivated | |
c56d5c7d | 1761 | * @lruvec: LRU vector to check |
56e49d21 RR |
1762 | * |
1763 | * When the system is doing streaming IO, memory pressure here | |
1764 | * ensures that active file pages get deactivated, until more | |
1765 | * than half of the file pages are on the inactive list. | |
1766 | * | |
1767 | * Once we get to that situation, protect the system's working | |
1768 | * set from being evicted by disabling active file page aging. | |
1769 | * | |
1770 | * This uses a different ratio than the anonymous pages, because | |
1771 | * the page cache uses a use-once replacement algorithm. | |
1772 | */ | |
c56d5c7d | 1773 | static int inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 | 1774 | { |
e3790144 JW |
1775 | unsigned long inactive; |
1776 | unsigned long active; | |
1777 | ||
1778 | inactive = get_lru_size(lruvec, LRU_INACTIVE_FILE); | |
1779 | active = get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 | 1780 | |
e3790144 | 1781 | return active > inactive; |
56e49d21 RR |
1782 | } |
1783 | ||
75b00af7 | 1784 | static int inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru) |
b39415b2 | 1785 | { |
75b00af7 | 1786 | if (is_file_lru(lru)) |
c56d5c7d | 1787 | return inactive_file_is_low(lruvec); |
b39415b2 | 1788 | else |
c56d5c7d | 1789 | return inactive_anon_is_low(lruvec); |
b39415b2 RR |
1790 | } |
1791 | ||
4f98a2fe | 1792 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
1a93be0e | 1793 | struct lruvec *lruvec, struct scan_control *sc) |
b69408e8 | 1794 | { |
b39415b2 | 1795 | if (is_active_lru(lru)) { |
75b00af7 | 1796 | if (inactive_list_is_low(lruvec, lru)) |
1a93be0e | 1797 | shrink_active_list(nr_to_scan, lruvec, sc, lru); |
556adecb RR |
1798 | return 0; |
1799 | } | |
1800 | ||
1a93be0e | 1801 | return shrink_inactive_list(nr_to_scan, lruvec, sc, lru); |
4f98a2fe RR |
1802 | } |
1803 | ||
3d58ab5c | 1804 | static int vmscan_swappiness(struct scan_control *sc) |
1f4c025b | 1805 | { |
89b5fae5 | 1806 | if (global_reclaim(sc)) |
1f4c025b | 1807 | return vm_swappiness; |
3d58ab5c | 1808 | return mem_cgroup_swappiness(sc->target_mem_cgroup); |
1f4c025b KH |
1809 | } |
1810 | ||
9a265114 JW |
1811 | enum scan_balance { |
1812 | SCAN_EQUAL, | |
1813 | SCAN_FRACT, | |
1814 | SCAN_ANON, | |
1815 | SCAN_FILE, | |
1816 | }; | |
1817 | ||
4f98a2fe RR |
1818 | /* |
1819 | * Determine how aggressively the anon and file LRU lists should be | |
1820 | * scanned. The relative value of each set of LRU lists is determined | |
1821 | * by looking at the fraction of the pages scanned we did rotate back | |
1822 | * onto the active list instead of evict. | |
1823 | * | |
be7bd59d WL |
1824 | * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan |
1825 | * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan | |
4f98a2fe | 1826 | */ |
90126375 | 1827 | static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc, |
9e3b2f8c | 1828 | unsigned long *nr) |
4f98a2fe | 1829 | { |
9a265114 JW |
1830 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1831 | u64 fraction[2]; | |
1832 | u64 denominator = 0; /* gcc */ | |
1833 | struct zone *zone = lruvec_zone(lruvec); | |
4f98a2fe | 1834 | unsigned long anon_prio, file_prio; |
9a265114 JW |
1835 | enum scan_balance scan_balance; |
1836 | unsigned long anon, file, free; | |
1837 | bool force_scan = false; | |
4f98a2fe | 1838 | unsigned long ap, fp; |
4111304d | 1839 | enum lru_list lru; |
246e87a9 | 1840 | |
f11c0ca5 JW |
1841 | /* |
1842 | * If the zone or memcg is small, nr[l] can be 0. This | |
1843 | * results in no scanning on this priority and a potential | |
1844 | * priority drop. Global direct reclaim can go to the next | |
1845 | * zone and tends to have no problems. Global kswapd is for | |
1846 | * zone balancing and it needs to scan a minimum amount. When | |
1847 | * reclaiming for a memcg, a priority drop can cause high | |
1848 | * latencies, so it's better to scan a minimum amount there as | |
1849 | * well. | |
1850 | */ | |
90126375 | 1851 | if (current_is_kswapd() && zone->all_unreclaimable) |
a4d3e9e7 | 1852 | force_scan = true; |
89b5fae5 | 1853 | if (!global_reclaim(sc)) |
a4d3e9e7 | 1854 | force_scan = true; |
76a33fc3 SL |
1855 | |
1856 | /* If we have no swap space, do not bother scanning anon pages. */ | |
ec8acf20 | 1857 | if (!sc->may_swap || (get_nr_swap_pages() <= 0)) { |
9a265114 | 1858 | scan_balance = SCAN_FILE; |
76a33fc3 SL |
1859 | goto out; |
1860 | } | |
4f98a2fe | 1861 | |
10316b31 JW |
1862 | /* |
1863 | * Global reclaim will swap to prevent OOM even with no | |
1864 | * swappiness, but memcg users want to use this knob to | |
1865 | * disable swapping for individual groups completely when | |
1866 | * using the memory controller's swap limit feature would be | |
1867 | * too expensive. | |
1868 | */ | |
1869 | if (!global_reclaim(sc) && !vmscan_swappiness(sc)) { | |
9a265114 | 1870 | scan_balance = SCAN_FILE; |
10316b31 JW |
1871 | goto out; |
1872 | } | |
1873 | ||
1874 | /* | |
1875 | * Do not apply any pressure balancing cleverness when the | |
1876 | * system is close to OOM, scan both anon and file equally | |
1877 | * (unless the swappiness setting disagrees with swapping). | |
1878 | */ | |
1879 | if (!sc->priority && vmscan_swappiness(sc)) { | |
9a265114 | 1880 | scan_balance = SCAN_EQUAL; |
10316b31 JW |
1881 | goto out; |
1882 | } | |
1883 | ||
4d7dcca2 HD |
1884 | anon = get_lru_size(lruvec, LRU_ACTIVE_ANON) + |
1885 | get_lru_size(lruvec, LRU_INACTIVE_ANON); | |
1886 | file = get_lru_size(lruvec, LRU_ACTIVE_FILE) + | |
1887 | get_lru_size(lruvec, LRU_INACTIVE_FILE); | |
a4d3e9e7 | 1888 | |
11d16c25 JW |
1889 | /* |
1890 | * If it's foreseeable that reclaiming the file cache won't be | |
1891 | * enough to get the zone back into a desirable shape, we have | |
1892 | * to swap. Better start now and leave the - probably heavily | |
1893 | * thrashing - remaining file pages alone. | |
1894 | */ | |
89b5fae5 | 1895 | if (global_reclaim(sc)) { |
11d16c25 | 1896 | free = zone_page_state(zone, NR_FREE_PAGES); |
90126375 | 1897 | if (unlikely(file + free <= high_wmark_pages(zone))) { |
9a265114 | 1898 | scan_balance = SCAN_ANON; |
76a33fc3 | 1899 | goto out; |
eeee9a8c | 1900 | } |
4f98a2fe RR |
1901 | } |
1902 | ||
7c5bd705 JW |
1903 | /* |
1904 | * There is enough inactive page cache, do not reclaim | |
1905 | * anything from the anonymous working set right now. | |
1906 | */ | |
1907 | if (!inactive_file_is_low(lruvec)) { | |
9a265114 | 1908 | scan_balance = SCAN_FILE; |
7c5bd705 JW |
1909 | goto out; |
1910 | } | |
1911 | ||
9a265114 JW |
1912 | scan_balance = SCAN_FRACT; |
1913 | ||
58c37f6e KM |
1914 | /* |
1915 | * With swappiness at 100, anonymous and file have the same priority. | |
1916 | * This scanning priority is essentially the inverse of IO cost. | |
1917 | */ | |
3d58ab5c | 1918 | anon_prio = vmscan_swappiness(sc); |
75b00af7 | 1919 | file_prio = 200 - anon_prio; |
58c37f6e | 1920 | |
4f98a2fe RR |
1921 | /* |
1922 | * OK, so we have swap space and a fair amount of page cache | |
1923 | * pages. We use the recently rotated / recently scanned | |
1924 | * ratios to determine how valuable each cache is. | |
1925 | * | |
1926 | * Because workloads change over time (and to avoid overflow) | |
1927 | * we keep these statistics as a floating average, which ends | |
1928 | * up weighing recent references more than old ones. | |
1929 | * | |
1930 | * anon in [0], file in [1] | |
1931 | */ | |
90126375 | 1932 | spin_lock_irq(&zone->lru_lock); |
6e901571 | 1933 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
6e901571 KM |
1934 | reclaim_stat->recent_scanned[0] /= 2; |
1935 | reclaim_stat->recent_rotated[0] /= 2; | |
4f98a2fe RR |
1936 | } |
1937 | ||
6e901571 | 1938 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
6e901571 KM |
1939 | reclaim_stat->recent_scanned[1] /= 2; |
1940 | reclaim_stat->recent_rotated[1] /= 2; | |
4f98a2fe RR |
1941 | } |
1942 | ||
4f98a2fe | 1943 | /* |
00d8089c RR |
1944 | * The amount of pressure on anon vs file pages is inversely |
1945 | * proportional to the fraction of recently scanned pages on | |
1946 | * each list that were recently referenced and in active use. | |
4f98a2fe | 1947 | */ |
fe35004f | 1948 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); |
6e901571 | 1949 | ap /= reclaim_stat->recent_rotated[0] + 1; |
4f98a2fe | 1950 | |
fe35004f | 1951 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); |
6e901571 | 1952 | fp /= reclaim_stat->recent_rotated[1] + 1; |
90126375 | 1953 | spin_unlock_irq(&zone->lru_lock); |
4f98a2fe | 1954 | |
76a33fc3 SL |
1955 | fraction[0] = ap; |
1956 | fraction[1] = fp; | |
1957 | denominator = ap + fp + 1; | |
1958 | out: | |
4111304d HD |
1959 | for_each_evictable_lru(lru) { |
1960 | int file = is_file_lru(lru); | |
d778df51 | 1961 | unsigned long size; |
76a33fc3 | 1962 | unsigned long scan; |
6e08a369 | 1963 | |
d778df51 | 1964 | size = get_lru_size(lruvec, lru); |
10316b31 | 1965 | scan = size >> sc->priority; |
9a265114 | 1966 | |
10316b31 JW |
1967 | if (!scan && force_scan) |
1968 | scan = min(size, SWAP_CLUSTER_MAX); | |
9a265114 JW |
1969 | |
1970 | switch (scan_balance) { | |
1971 | case SCAN_EQUAL: | |
1972 | /* Scan lists relative to size */ | |
1973 | break; | |
1974 | case SCAN_FRACT: | |
1975 | /* | |
1976 | * Scan types proportional to swappiness and | |
1977 | * their relative recent reclaim efficiency. | |
1978 | */ | |
1979 | scan = div64_u64(scan * fraction[file], denominator); | |
1980 | break; | |
1981 | case SCAN_FILE: | |
1982 | case SCAN_ANON: | |
1983 | /* Scan one type exclusively */ | |
1984 | if ((scan_balance == SCAN_FILE) != file) | |
1985 | scan = 0; | |
1986 | break; | |
1987 | default: | |
1988 | /* Look ma, no brain */ | |
1989 | BUG(); | |
1990 | } | |
4111304d | 1991 | nr[lru] = scan; |
76a33fc3 | 1992 | } |
6e08a369 | 1993 | } |
4f98a2fe | 1994 | |
9b4f98cd JW |
1995 | /* |
1996 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. | |
1997 | */ | |
1998 | static void shrink_lruvec(struct lruvec *lruvec, struct scan_control *sc) | |
1999 | { | |
2000 | unsigned long nr[NR_LRU_LISTS]; | |
e82e0561 | 2001 | unsigned long targets[NR_LRU_LISTS]; |
9b4f98cd JW |
2002 | unsigned long nr_to_scan; |
2003 | enum lru_list lru; | |
2004 | unsigned long nr_reclaimed = 0; | |
2005 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; | |
2006 | struct blk_plug plug; | |
e82e0561 | 2007 | bool scan_adjusted = false; |
9b4f98cd JW |
2008 | |
2009 | get_scan_count(lruvec, sc, nr); | |
2010 | ||
e82e0561 MG |
2011 | /* Record the original scan target for proportional adjustments later */ |
2012 | memcpy(targets, nr, sizeof(nr)); | |
2013 | ||
9b4f98cd JW |
2014 | blk_start_plug(&plug); |
2015 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || | |
2016 | nr[LRU_INACTIVE_FILE]) { | |
e82e0561 MG |
2017 | unsigned long nr_anon, nr_file, percentage; |
2018 | unsigned long nr_scanned; | |
2019 | ||
9b4f98cd JW |
2020 | for_each_evictable_lru(lru) { |
2021 | if (nr[lru]) { | |
2022 | nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX); | |
2023 | nr[lru] -= nr_to_scan; | |
2024 | ||
2025 | nr_reclaimed += shrink_list(lru, nr_to_scan, | |
2026 | lruvec, sc); | |
2027 | } | |
2028 | } | |
e82e0561 MG |
2029 | |
2030 | if (nr_reclaimed < nr_to_reclaim || scan_adjusted) | |
2031 | continue; | |
2032 | ||
9b4f98cd | 2033 | /* |
e82e0561 MG |
2034 | * For global direct reclaim, reclaim only the number of pages |
2035 | * requested. Less care is taken to scan proportionally as it | |
2036 | * is more important to minimise direct reclaim stall latency | |
2037 | * than it is to properly age the LRU lists. | |
9b4f98cd | 2038 | */ |
e82e0561 | 2039 | if (global_reclaim(sc) && !current_is_kswapd()) |
9b4f98cd | 2040 | break; |
e82e0561 MG |
2041 | |
2042 | /* | |
2043 | * For kswapd and memcg, reclaim at least the number of pages | |
2044 | * requested. Ensure that the anon and file LRUs shrink | |
2045 | * proportionally what was requested by get_scan_count(). We | |
2046 | * stop reclaiming one LRU and reduce the amount scanning | |
2047 | * proportional to the original scan target. | |
2048 | */ | |
2049 | nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE]; | |
2050 | nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON]; | |
2051 | ||
2052 | if (nr_file > nr_anon) { | |
2053 | unsigned long scan_target = targets[LRU_INACTIVE_ANON] + | |
2054 | targets[LRU_ACTIVE_ANON] + 1; | |
2055 | lru = LRU_BASE; | |
2056 | percentage = nr_anon * 100 / scan_target; | |
2057 | } else { | |
2058 | unsigned long scan_target = targets[LRU_INACTIVE_FILE] + | |
2059 | targets[LRU_ACTIVE_FILE] + 1; | |
2060 | lru = LRU_FILE; | |
2061 | percentage = nr_file * 100 / scan_target; | |
2062 | } | |
2063 | ||
2064 | /* Stop scanning the smaller of the LRU */ | |
2065 | nr[lru] = 0; | |
2066 | nr[lru + LRU_ACTIVE] = 0; | |
2067 | ||
2068 | /* | |
2069 | * Recalculate the other LRU scan count based on its original | |
2070 | * scan target and the percentage scanning already complete | |
2071 | */ | |
2072 | lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE; | |
2073 | nr_scanned = targets[lru] - nr[lru]; | |
2074 | nr[lru] = targets[lru] * (100 - percentage) / 100; | |
2075 | nr[lru] -= min(nr[lru], nr_scanned); | |
2076 | ||
2077 | lru += LRU_ACTIVE; | |
2078 | nr_scanned = targets[lru] - nr[lru]; | |
2079 | nr[lru] = targets[lru] * (100 - percentage) / 100; | |
2080 | nr[lru] -= min(nr[lru], nr_scanned); | |
2081 | ||
2082 | scan_adjusted = true; | |
9b4f98cd JW |
2083 | } |
2084 | blk_finish_plug(&plug); | |
2085 | sc->nr_reclaimed += nr_reclaimed; | |
2086 | ||
2087 | /* | |
2088 | * Even if we did not try to evict anon pages at all, we want to | |
2089 | * rebalance the anon lru active/inactive ratio. | |
2090 | */ | |
2091 | if (inactive_anon_is_low(lruvec)) | |
2092 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, | |
2093 | sc, LRU_ACTIVE_ANON); | |
2094 | ||
2095 | throttle_vm_writeout(sc->gfp_mask); | |
2096 | } | |
2097 | ||
23b9da55 | 2098 | /* Use reclaim/compaction for costly allocs or under memory pressure */ |
9e3b2f8c | 2099 | static bool in_reclaim_compaction(struct scan_control *sc) |
23b9da55 | 2100 | { |
d84da3f9 | 2101 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && |
23b9da55 | 2102 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || |
9e3b2f8c | 2103 | sc->priority < DEF_PRIORITY - 2)) |
23b9da55 MG |
2104 | return true; |
2105 | ||
2106 | return false; | |
2107 | } | |
2108 | ||
3e7d3449 | 2109 | /* |
23b9da55 MG |
2110 | * Reclaim/compaction is used for high-order allocation requests. It reclaims |
2111 | * order-0 pages before compacting the zone. should_continue_reclaim() returns | |
2112 | * true if more pages should be reclaimed such that when the page allocator | |
2113 | * calls try_to_compact_zone() that it will have enough free pages to succeed. | |
2114 | * It will give up earlier than that if there is difficulty reclaiming pages. | |
3e7d3449 | 2115 | */ |
9b4f98cd | 2116 | static inline bool should_continue_reclaim(struct zone *zone, |
3e7d3449 MG |
2117 | unsigned long nr_reclaimed, |
2118 | unsigned long nr_scanned, | |
2119 | struct scan_control *sc) | |
2120 | { | |
2121 | unsigned long pages_for_compaction; | |
2122 | unsigned long inactive_lru_pages; | |
2123 | ||
2124 | /* If not in reclaim/compaction mode, stop */ | |
9e3b2f8c | 2125 | if (!in_reclaim_compaction(sc)) |
3e7d3449 MG |
2126 | return false; |
2127 | ||
2876592f MG |
2128 | /* Consider stopping depending on scan and reclaim activity */ |
2129 | if (sc->gfp_mask & __GFP_REPEAT) { | |
2130 | /* | |
2131 | * For __GFP_REPEAT allocations, stop reclaiming if the | |
2132 | * full LRU list has been scanned and we are still failing | |
2133 | * to reclaim pages. This full LRU scan is potentially | |
2134 | * expensive but a __GFP_REPEAT caller really wants to succeed | |
2135 | */ | |
2136 | if (!nr_reclaimed && !nr_scanned) | |
2137 | return false; | |
2138 | } else { | |
2139 | /* | |
2140 | * For non-__GFP_REPEAT allocations which can presumably | |
2141 | * fail without consequence, stop if we failed to reclaim | |
2142 | * any pages from the last SWAP_CLUSTER_MAX number of | |
2143 | * pages that were scanned. This will return to the | |
2144 | * caller faster at the risk reclaim/compaction and | |
2145 | * the resulting allocation attempt fails | |
2146 | */ | |
2147 | if (!nr_reclaimed) | |
2148 | return false; | |
2149 | } | |
3e7d3449 MG |
2150 | |
2151 | /* | |
2152 | * If we have not reclaimed enough pages for compaction and the | |
2153 | * inactive lists are large enough, continue reclaiming | |
2154 | */ | |
2155 | pages_for_compaction = (2UL << sc->order); | |
9b4f98cd | 2156 | inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE); |
ec8acf20 | 2157 | if (get_nr_swap_pages() > 0) |
9b4f98cd | 2158 | inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON); |
3e7d3449 MG |
2159 | if (sc->nr_reclaimed < pages_for_compaction && |
2160 | inactive_lru_pages > pages_for_compaction) | |
2161 | return true; | |
2162 | ||
2163 | /* If compaction would go ahead or the allocation would succeed, stop */ | |
9b4f98cd | 2164 | switch (compaction_suitable(zone, sc->order)) { |
3e7d3449 MG |
2165 | case COMPACT_PARTIAL: |
2166 | case COMPACT_CONTINUE: | |
2167 | return false; | |
2168 | default: | |
2169 | return true; | |
2170 | } | |
2171 | } | |
2172 | ||
9b4f98cd | 2173 | static void shrink_zone(struct zone *zone, struct scan_control *sc) |
1da177e4 | 2174 | { |
f0fdc5e8 | 2175 | unsigned long nr_reclaimed, nr_scanned; |
1da177e4 | 2176 | |
9b4f98cd JW |
2177 | do { |
2178 | struct mem_cgroup *root = sc->target_mem_cgroup; | |
2179 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2180 | .zone = zone, | |
2181 | .priority = sc->priority, | |
2182 | }; | |
2183 | struct mem_cgroup *memcg; | |
3e7d3449 | 2184 | |
9b4f98cd JW |
2185 | nr_reclaimed = sc->nr_reclaimed; |
2186 | nr_scanned = sc->nr_scanned; | |
1da177e4 | 2187 | |
9b4f98cd JW |
2188 | memcg = mem_cgroup_iter(root, NULL, &reclaim); |
2189 | do { | |
2190 | struct lruvec *lruvec; | |
5660048c | 2191 | |
9b4f98cd | 2192 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
f9be23d6 | 2193 | |
9b4f98cd | 2194 | shrink_lruvec(lruvec, sc); |
f16015fb | 2195 | |
9b4f98cd | 2196 | /* |
a394cb8e MH |
2197 | * Direct reclaim and kswapd have to scan all memory |
2198 | * cgroups to fulfill the overall scan target for the | |
9b4f98cd | 2199 | * zone. |
a394cb8e MH |
2200 | * |
2201 | * Limit reclaim, on the other hand, only cares about | |
2202 | * nr_to_reclaim pages to be reclaimed and it will | |
2203 | * retry with decreasing priority if one round over the | |
2204 | * whole hierarchy is not sufficient. | |
9b4f98cd | 2205 | */ |
a394cb8e MH |
2206 | if (!global_reclaim(sc) && |
2207 | sc->nr_reclaimed >= sc->nr_to_reclaim) { | |
9b4f98cd JW |
2208 | mem_cgroup_iter_break(root, memcg); |
2209 | break; | |
2210 | } | |
2211 | memcg = mem_cgroup_iter(root, memcg, &reclaim); | |
2212 | } while (memcg); | |
70ddf637 AV |
2213 | |
2214 | vmpressure(sc->gfp_mask, sc->target_mem_cgroup, | |
2215 | sc->nr_scanned - nr_scanned, | |
2216 | sc->nr_reclaimed - nr_reclaimed); | |
2217 | ||
9b4f98cd JW |
2218 | } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed, |
2219 | sc->nr_scanned - nr_scanned, sc)); | |
f16015fb JW |
2220 | } |
2221 | ||
fe4b1b24 MG |
2222 | /* Returns true if compaction should go ahead for a high-order request */ |
2223 | static inline bool compaction_ready(struct zone *zone, struct scan_control *sc) | |
2224 | { | |
2225 | unsigned long balance_gap, watermark; | |
2226 | bool watermark_ok; | |
2227 | ||
2228 | /* Do not consider compaction for orders reclaim is meant to satisfy */ | |
2229 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER) | |
2230 | return false; | |
2231 | ||
2232 | /* | |
2233 | * Compaction takes time to run and there are potentially other | |
2234 | * callers using the pages just freed. Continue reclaiming until | |
2235 | * there is a buffer of free pages available to give compaction | |
2236 | * a reasonable chance of completing and allocating the page | |
2237 | */ | |
2238 | balance_gap = min(low_wmark_pages(zone), | |
b40da049 | 2239 | (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / |
fe4b1b24 MG |
2240 | KSWAPD_ZONE_BALANCE_GAP_RATIO); |
2241 | watermark = high_wmark_pages(zone) + balance_gap + (2UL << sc->order); | |
2242 | watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0); | |
2243 | ||
2244 | /* | |
2245 | * If compaction is deferred, reclaim up to a point where | |
2246 | * compaction will have a chance of success when re-enabled | |
2247 | */ | |
aff62249 | 2248 | if (compaction_deferred(zone, sc->order)) |
fe4b1b24 MG |
2249 | return watermark_ok; |
2250 | ||
2251 | /* If compaction is not ready to start, keep reclaiming */ | |
2252 | if (!compaction_suitable(zone, sc->order)) | |
2253 | return false; | |
2254 | ||
2255 | return watermark_ok; | |
2256 | } | |
2257 | ||
1da177e4 LT |
2258 | /* |
2259 | * This is the direct reclaim path, for page-allocating processes. We only | |
2260 | * try to reclaim pages from zones which will satisfy the caller's allocation | |
2261 | * request. | |
2262 | * | |
41858966 MG |
2263 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). |
2264 | * Because: | |
1da177e4 LT |
2265 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
2266 | * allocation or | |
41858966 MG |
2267 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones |
2268 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' | |
2269 | * zone defense algorithm. | |
1da177e4 | 2270 | * |
1da177e4 LT |
2271 | * If a zone is deemed to be full of pinned pages then just give it a light |
2272 | * scan then give up on it. | |
e0c23279 MG |
2273 | * |
2274 | * This function returns true if a zone is being reclaimed for a costly | |
fe4b1b24 | 2275 | * high-order allocation and compaction is ready to begin. This indicates to |
0cee34fd MG |
2276 | * the caller that it should consider retrying the allocation instead of |
2277 | * further reclaim. | |
1da177e4 | 2278 | */ |
9e3b2f8c | 2279 | static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc) |
1da177e4 | 2280 | { |
dd1a239f | 2281 | struct zoneref *z; |
54a6eb5c | 2282 | struct zone *zone; |
d149e3b2 YH |
2283 | unsigned long nr_soft_reclaimed; |
2284 | unsigned long nr_soft_scanned; | |
0cee34fd | 2285 | bool aborted_reclaim = false; |
1cfb419b | 2286 | |
cc715d99 MG |
2287 | /* |
2288 | * If the number of buffer_heads in the machine exceeds the maximum | |
2289 | * allowed level, force direct reclaim to scan the highmem zone as | |
2290 | * highmem pages could be pinning lowmem pages storing buffer_heads | |
2291 | */ | |
2292 | if (buffer_heads_over_limit) | |
2293 | sc->gfp_mask |= __GFP_HIGHMEM; | |
2294 | ||
d4debc66 MG |
2295 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
2296 | gfp_zone(sc->gfp_mask), sc->nodemask) { | |
f3fe6512 | 2297 | if (!populated_zone(zone)) |
1da177e4 | 2298 | continue; |
1cfb419b KH |
2299 | /* |
2300 | * Take care memory controller reclaiming has small influence | |
2301 | * to global LRU. | |
2302 | */ | |
89b5fae5 | 2303 | if (global_reclaim(sc)) { |
1cfb419b KH |
2304 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
2305 | continue; | |
9e3b2f8c KK |
2306 | if (zone->all_unreclaimable && |
2307 | sc->priority != DEF_PRIORITY) | |
1cfb419b | 2308 | continue; /* Let kswapd poll it */ |
d84da3f9 | 2309 | if (IS_ENABLED(CONFIG_COMPACTION)) { |
e0887c19 | 2310 | /* |
e0c23279 MG |
2311 | * If we already have plenty of memory free for |
2312 | * compaction in this zone, don't free any more. | |
2313 | * Even though compaction is invoked for any | |
2314 | * non-zero order, only frequent costly order | |
2315 | * reclamation is disruptive enough to become a | |
c7cfa37b CA |
2316 | * noticeable problem, like transparent huge |
2317 | * page allocations. | |
e0887c19 | 2318 | */ |
fe4b1b24 | 2319 | if (compaction_ready(zone, sc)) { |
0cee34fd | 2320 | aborted_reclaim = true; |
e0887c19 | 2321 | continue; |
e0c23279 | 2322 | } |
e0887c19 | 2323 | } |
ac34a1a3 KH |
2324 | /* |
2325 | * This steals pages from memory cgroups over softlimit | |
2326 | * and returns the number of reclaimed pages and | |
2327 | * scanned pages. This works for global memory pressure | |
2328 | * and balancing, not for a memcg's limit. | |
2329 | */ | |
2330 | nr_soft_scanned = 0; | |
2331 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | |
2332 | sc->order, sc->gfp_mask, | |
2333 | &nr_soft_scanned); | |
2334 | sc->nr_reclaimed += nr_soft_reclaimed; | |
2335 | sc->nr_scanned += nr_soft_scanned; | |
2336 | /* need some check for avoid more shrink_zone() */ | |
1cfb419b | 2337 | } |
408d8544 | 2338 | |
9e3b2f8c | 2339 | shrink_zone(zone, sc); |
1da177e4 | 2340 | } |
e0c23279 | 2341 | |
0cee34fd | 2342 | return aborted_reclaim; |
d1908362 MK |
2343 | } |
2344 | ||
2345 | static bool zone_reclaimable(struct zone *zone) | |
2346 | { | |
2347 | return zone->pages_scanned < zone_reclaimable_pages(zone) * 6; | |
2348 | } | |
2349 | ||
929bea7c | 2350 | /* All zones in zonelist are unreclaimable? */ |
d1908362 MK |
2351 | static bool all_unreclaimable(struct zonelist *zonelist, |
2352 | struct scan_control *sc) | |
2353 | { | |
2354 | struct zoneref *z; | |
2355 | struct zone *zone; | |
d1908362 MK |
2356 | |
2357 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
2358 | gfp_zone(sc->gfp_mask), sc->nodemask) { | |
2359 | if (!populated_zone(zone)) | |
2360 | continue; | |
2361 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) | |
2362 | continue; | |
929bea7c KM |
2363 | if (!zone->all_unreclaimable) |
2364 | return false; | |
d1908362 MK |
2365 | } |
2366 | ||
929bea7c | 2367 | return true; |
1da177e4 | 2368 | } |
4f98a2fe | 2369 | |
1da177e4 LT |
2370 | /* |
2371 | * This is the main entry point to direct page reclaim. | |
2372 | * | |
2373 | * If a full scan of the inactive list fails to free enough memory then we | |
2374 | * are "out of memory" and something needs to be killed. | |
2375 | * | |
2376 | * If the caller is !__GFP_FS then the probability of a failure is reasonably | |
2377 | * high - the zone may be full of dirty or under-writeback pages, which this | |
5b0830cb JA |
2378 | * caller can't do much about. We kick the writeback threads and take explicit |
2379 | * naps in the hope that some of these pages can be written. But if the | |
2380 | * allocating task holds filesystem locks which prevent writeout this might not | |
2381 | * work, and the allocation attempt will fail. | |
a41f24ea NA |
2382 | * |
2383 | * returns: 0, if no pages reclaimed | |
2384 | * else, the number of pages reclaimed | |
1da177e4 | 2385 | */ |
dac1d27b | 2386 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
a09ed5e0 YH |
2387 | struct scan_control *sc, |
2388 | struct shrink_control *shrink) | |
1da177e4 | 2389 | { |
69e05944 | 2390 | unsigned long total_scanned = 0; |
1da177e4 | 2391 | struct reclaim_state *reclaim_state = current->reclaim_state; |
dd1a239f | 2392 | struct zoneref *z; |
54a6eb5c | 2393 | struct zone *zone; |
22fba335 | 2394 | unsigned long writeback_threshold; |
0cee34fd | 2395 | bool aborted_reclaim; |
1da177e4 | 2396 | |
873b4771 KK |
2397 | delayacct_freepages_start(); |
2398 | ||
89b5fae5 | 2399 | if (global_reclaim(sc)) |
1cfb419b | 2400 | count_vm_event(ALLOCSTALL); |
1da177e4 | 2401 | |
9e3b2f8c | 2402 | do { |
70ddf637 AV |
2403 | vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup, |
2404 | sc->priority); | |
66e1707b | 2405 | sc->nr_scanned = 0; |
9e3b2f8c | 2406 | aborted_reclaim = shrink_zones(zonelist, sc); |
e0c23279 | 2407 | |
66e1707b | 2408 | /* |
5a1c9cbc MG |
2409 | * Don't shrink slabs when reclaiming memory from over limit |
2410 | * cgroups but do shrink slab at least once when aborting | |
2411 | * reclaim for compaction to avoid unevenly scanning file/anon | |
2412 | * LRU pages over slab pages. | |
66e1707b | 2413 | */ |
89b5fae5 | 2414 | if (global_reclaim(sc)) { |
c6a8a8c5 | 2415 | unsigned long lru_pages = 0; |
0ce3d744 DC |
2416 | |
2417 | nodes_clear(shrink->nodes_to_scan); | |
d4debc66 MG |
2418 | for_each_zone_zonelist(zone, z, zonelist, |
2419 | gfp_zone(sc->gfp_mask)) { | |
c6a8a8c5 KM |
2420 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
2421 | continue; | |
2422 | ||
2423 | lru_pages += zone_reclaimable_pages(zone); | |
0ce3d744 DC |
2424 | node_set(zone_to_nid(zone), |
2425 | shrink->nodes_to_scan); | |
c6a8a8c5 KM |
2426 | } |
2427 | ||
1495f230 | 2428 | shrink_slab(shrink, sc->nr_scanned, lru_pages); |
91a45470 | 2429 | if (reclaim_state) { |
a79311c1 | 2430 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
91a45470 KH |
2431 | reclaim_state->reclaimed_slab = 0; |
2432 | } | |
1da177e4 | 2433 | } |
66e1707b | 2434 | total_scanned += sc->nr_scanned; |
bb21c7ce | 2435 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
1da177e4 | 2436 | goto out; |
1da177e4 | 2437 | |
0e50ce3b MK |
2438 | /* |
2439 | * If we're getting trouble reclaiming, start doing | |
2440 | * writepage even in laptop mode. | |
2441 | */ | |
2442 | if (sc->priority < DEF_PRIORITY - 2) | |
2443 | sc->may_writepage = 1; | |
2444 | ||
1da177e4 LT |
2445 | /* |
2446 | * Try to write back as many pages as we just scanned. This | |
2447 | * tends to cause slow streaming writers to write data to the | |
2448 | * disk smoothly, at the dirtying rate, which is nice. But | |
2449 | * that's undesirable in laptop mode, where we *want* lumpy | |
2450 | * writeout. So in laptop mode, write out the whole world. | |
2451 | */ | |
22fba335 KM |
2452 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; |
2453 | if (total_scanned > writeback_threshold) { | |
0e175a18 CW |
2454 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, |
2455 | WB_REASON_TRY_TO_FREE_PAGES); | |
66e1707b | 2456 | sc->may_writepage = 1; |
1da177e4 | 2457 | } |
5a1c9cbc | 2458 | } while (--sc->priority >= 0 && !aborted_reclaim); |
bb21c7ce | 2459 | |
1da177e4 | 2460 | out: |
873b4771 KK |
2461 | delayacct_freepages_end(); |
2462 | ||
bb21c7ce KM |
2463 | if (sc->nr_reclaimed) |
2464 | return sc->nr_reclaimed; | |
2465 | ||
929bea7c KM |
2466 | /* |
2467 | * As hibernation is going on, kswapd is freezed so that it can't mark | |
2468 | * the zone into all_unreclaimable. Thus bypassing all_unreclaimable | |
2469 | * check. | |
2470 | */ | |
2471 | if (oom_killer_disabled) | |
2472 | return 0; | |
2473 | ||
0cee34fd MG |
2474 | /* Aborted reclaim to try compaction? don't OOM, then */ |
2475 | if (aborted_reclaim) | |
7335084d MG |
2476 | return 1; |
2477 | ||
bb21c7ce | 2478 | /* top priority shrink_zones still had more to do? don't OOM, then */ |
89b5fae5 | 2479 | if (global_reclaim(sc) && !all_unreclaimable(zonelist, sc)) |
bb21c7ce KM |
2480 | return 1; |
2481 | ||
2482 | return 0; | |
1da177e4 LT |
2483 | } |
2484 | ||
5515061d MG |
2485 | static bool pfmemalloc_watermark_ok(pg_data_t *pgdat) |
2486 | { | |
2487 | struct zone *zone; | |
2488 | unsigned long pfmemalloc_reserve = 0; | |
2489 | unsigned long free_pages = 0; | |
2490 | int i; | |
2491 | bool wmark_ok; | |
2492 | ||
2493 | for (i = 0; i <= ZONE_NORMAL; i++) { | |
2494 | zone = &pgdat->node_zones[i]; | |
2495 | pfmemalloc_reserve += min_wmark_pages(zone); | |
2496 | free_pages += zone_page_state(zone, NR_FREE_PAGES); | |
2497 | } | |
2498 | ||
2499 | wmark_ok = free_pages > pfmemalloc_reserve / 2; | |
2500 | ||
2501 | /* kswapd must be awake if processes are being throttled */ | |
2502 | if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) { | |
2503 | pgdat->classzone_idx = min(pgdat->classzone_idx, | |
2504 | (enum zone_type)ZONE_NORMAL); | |
2505 | wake_up_interruptible(&pgdat->kswapd_wait); | |
2506 | } | |
2507 | ||
2508 | return wmark_ok; | |
2509 | } | |
2510 | ||
2511 | /* | |
2512 | * Throttle direct reclaimers if backing storage is backed by the network | |
2513 | * and the PFMEMALLOC reserve for the preferred node is getting dangerously | |
2514 | * depleted. kswapd will continue to make progress and wake the processes | |
50694c28 MG |
2515 | * when the low watermark is reached. |
2516 | * | |
2517 | * Returns true if a fatal signal was delivered during throttling. If this | |
2518 | * happens, the page allocator should not consider triggering the OOM killer. | |
5515061d | 2519 | */ |
50694c28 | 2520 | static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist, |
5515061d MG |
2521 | nodemask_t *nodemask) |
2522 | { | |
2523 | struct zone *zone; | |
2524 | int high_zoneidx = gfp_zone(gfp_mask); | |
2525 | pg_data_t *pgdat; | |
2526 | ||
2527 | /* | |
2528 | * Kernel threads should not be throttled as they may be indirectly | |
2529 | * responsible for cleaning pages necessary for reclaim to make forward | |
2530 | * progress. kjournald for example may enter direct reclaim while | |
2531 | * committing a transaction where throttling it could forcing other | |
2532 | * processes to block on log_wait_commit(). | |
2533 | */ | |
2534 | if (current->flags & PF_KTHREAD) | |
50694c28 MG |
2535 | goto out; |
2536 | ||
2537 | /* | |
2538 | * If a fatal signal is pending, this process should not throttle. | |
2539 | * It should return quickly so it can exit and free its memory | |
2540 | */ | |
2541 | if (fatal_signal_pending(current)) | |
2542 | goto out; | |
5515061d MG |
2543 | |
2544 | /* Check if the pfmemalloc reserves are ok */ | |
2545 | first_zones_zonelist(zonelist, high_zoneidx, NULL, &zone); | |
2546 | pgdat = zone->zone_pgdat; | |
2547 | if (pfmemalloc_watermark_ok(pgdat)) | |
50694c28 | 2548 | goto out; |
5515061d | 2549 | |
68243e76 MG |
2550 | /* Account for the throttling */ |
2551 | count_vm_event(PGSCAN_DIRECT_THROTTLE); | |
2552 | ||
5515061d MG |
2553 | /* |
2554 | * If the caller cannot enter the filesystem, it's possible that it | |
2555 | * is due to the caller holding an FS lock or performing a journal | |
2556 | * transaction in the case of a filesystem like ext[3|4]. In this case, | |
2557 | * it is not safe to block on pfmemalloc_wait as kswapd could be | |
2558 | * blocked waiting on the same lock. Instead, throttle for up to a | |
2559 | * second before continuing. | |
2560 | */ | |
2561 | if (!(gfp_mask & __GFP_FS)) { | |
2562 | wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, | |
2563 | pfmemalloc_watermark_ok(pgdat), HZ); | |
50694c28 MG |
2564 | |
2565 | goto check_pending; | |
5515061d MG |
2566 | } |
2567 | ||
2568 | /* Throttle until kswapd wakes the process */ | |
2569 | wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, | |
2570 | pfmemalloc_watermark_ok(pgdat)); | |
50694c28 MG |
2571 | |
2572 | check_pending: | |
2573 | if (fatal_signal_pending(current)) | |
2574 | return true; | |
2575 | ||
2576 | out: | |
2577 | return false; | |
5515061d MG |
2578 | } |
2579 | ||
dac1d27b | 2580 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
327c0e96 | 2581 | gfp_t gfp_mask, nodemask_t *nodemask) |
66e1707b | 2582 | { |
33906bc5 | 2583 | unsigned long nr_reclaimed; |
66e1707b | 2584 | struct scan_control sc = { |
21caf2fc | 2585 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
66e1707b | 2586 | .may_writepage = !laptop_mode, |
22fba335 | 2587 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
a6dc60f8 | 2588 | .may_unmap = 1, |
2e2e4259 | 2589 | .may_swap = 1, |
66e1707b | 2590 | .order = order, |
9e3b2f8c | 2591 | .priority = DEF_PRIORITY, |
f16015fb | 2592 | .target_mem_cgroup = NULL, |
327c0e96 | 2593 | .nodemask = nodemask, |
66e1707b | 2594 | }; |
a09ed5e0 YH |
2595 | struct shrink_control shrink = { |
2596 | .gfp_mask = sc.gfp_mask, | |
2597 | }; | |
66e1707b | 2598 | |
5515061d | 2599 | /* |
50694c28 MG |
2600 | * Do not enter reclaim if fatal signal was delivered while throttled. |
2601 | * 1 is returned so that the page allocator does not OOM kill at this | |
2602 | * point. | |
5515061d | 2603 | */ |
50694c28 | 2604 | if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask)) |
5515061d MG |
2605 | return 1; |
2606 | ||
33906bc5 MG |
2607 | trace_mm_vmscan_direct_reclaim_begin(order, |
2608 | sc.may_writepage, | |
2609 | gfp_mask); | |
2610 | ||
a09ed5e0 | 2611 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
33906bc5 MG |
2612 | |
2613 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); | |
2614 | ||
2615 | return nr_reclaimed; | |
66e1707b BS |
2616 | } |
2617 | ||
c255a458 | 2618 | #ifdef CONFIG_MEMCG |
66e1707b | 2619 | |
72835c86 | 2620 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg, |
4e416953 | 2621 | gfp_t gfp_mask, bool noswap, |
0ae5e89c YH |
2622 | struct zone *zone, |
2623 | unsigned long *nr_scanned) | |
4e416953 BS |
2624 | { |
2625 | struct scan_control sc = { | |
0ae5e89c | 2626 | .nr_scanned = 0, |
b8f5c566 | 2627 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
4e416953 BS |
2628 | .may_writepage = !laptop_mode, |
2629 | .may_unmap = 1, | |
2630 | .may_swap = !noswap, | |
4e416953 | 2631 | .order = 0, |
9e3b2f8c | 2632 | .priority = 0, |
72835c86 | 2633 | .target_mem_cgroup = memcg, |
4e416953 | 2634 | }; |
f9be23d6 | 2635 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
0ae5e89c | 2636 | |
4e416953 BS |
2637 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2638 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); | |
bdce6d9e | 2639 | |
9e3b2f8c | 2640 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, |
bdce6d9e KM |
2641 | sc.may_writepage, |
2642 | sc.gfp_mask); | |
2643 | ||
4e416953 BS |
2644 | /* |
2645 | * NOTE: Although we can get the priority field, using it | |
2646 | * here is not a good idea, since it limits the pages we can scan. | |
2647 | * if we don't reclaim here, the shrink_zone from balance_pgdat | |
2648 | * will pick up pages from other mem cgroup's as well. We hack | |
2649 | * the priority and make it zero. | |
2650 | */ | |
f9be23d6 | 2651 | shrink_lruvec(lruvec, &sc); |
bdce6d9e KM |
2652 | |
2653 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); | |
2654 | ||
0ae5e89c | 2655 | *nr_scanned = sc.nr_scanned; |
4e416953 BS |
2656 | return sc.nr_reclaimed; |
2657 | } | |
2658 | ||
72835c86 | 2659 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, |
a7885eb8 | 2660 | gfp_t gfp_mask, |
185efc0f | 2661 | bool noswap) |
66e1707b | 2662 | { |
4e416953 | 2663 | struct zonelist *zonelist; |
bdce6d9e | 2664 | unsigned long nr_reclaimed; |
889976db | 2665 | int nid; |
66e1707b | 2666 | struct scan_control sc = { |
66e1707b | 2667 | .may_writepage = !laptop_mode, |
a6dc60f8 | 2668 | .may_unmap = 1, |
2e2e4259 | 2669 | .may_swap = !noswap, |
22fba335 | 2670 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
66e1707b | 2671 | .order = 0, |
9e3b2f8c | 2672 | .priority = DEF_PRIORITY, |
72835c86 | 2673 | .target_mem_cgroup = memcg, |
327c0e96 | 2674 | .nodemask = NULL, /* we don't care the placement */ |
a09ed5e0 YH |
2675 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2676 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), | |
2677 | }; | |
2678 | struct shrink_control shrink = { | |
2679 | .gfp_mask = sc.gfp_mask, | |
66e1707b | 2680 | }; |
66e1707b | 2681 | |
889976db YH |
2682 | /* |
2683 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't | |
2684 | * take care of from where we get pages. So the node where we start the | |
2685 | * scan does not need to be the current node. | |
2686 | */ | |
72835c86 | 2687 | nid = mem_cgroup_select_victim_node(memcg); |
889976db YH |
2688 | |
2689 | zonelist = NODE_DATA(nid)->node_zonelists; | |
bdce6d9e KM |
2690 | |
2691 | trace_mm_vmscan_memcg_reclaim_begin(0, | |
2692 | sc.may_writepage, | |
2693 | sc.gfp_mask); | |
2694 | ||
a09ed5e0 | 2695 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
bdce6d9e KM |
2696 | |
2697 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); | |
2698 | ||
2699 | return nr_reclaimed; | |
66e1707b BS |
2700 | } |
2701 | #endif | |
2702 | ||
9e3b2f8c | 2703 | static void age_active_anon(struct zone *zone, struct scan_control *sc) |
f16015fb | 2704 | { |
b95a2f2d | 2705 | struct mem_cgroup *memcg; |
f16015fb | 2706 | |
b95a2f2d JW |
2707 | if (!total_swap_pages) |
2708 | return; | |
2709 | ||
2710 | memcg = mem_cgroup_iter(NULL, NULL, NULL); | |
2711 | do { | |
c56d5c7d | 2712 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
b95a2f2d | 2713 | |
c56d5c7d | 2714 | if (inactive_anon_is_low(lruvec)) |
1a93be0e | 2715 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
9e3b2f8c | 2716 | sc, LRU_ACTIVE_ANON); |
b95a2f2d JW |
2717 | |
2718 | memcg = mem_cgroup_iter(NULL, memcg, NULL); | |
2719 | } while (memcg); | |
f16015fb JW |
2720 | } |
2721 | ||
60cefed4 JW |
2722 | static bool zone_balanced(struct zone *zone, int order, |
2723 | unsigned long balance_gap, int classzone_idx) | |
2724 | { | |
2725 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) + | |
2726 | balance_gap, classzone_idx, 0)) | |
2727 | return false; | |
2728 | ||
d84da3f9 KS |
2729 | if (IS_ENABLED(CONFIG_COMPACTION) && order && |
2730 | !compaction_suitable(zone, order)) | |
60cefed4 JW |
2731 | return false; |
2732 | ||
2733 | return true; | |
2734 | } | |
2735 | ||
1741c877 | 2736 | /* |
4ae0a48b ZC |
2737 | * pgdat_balanced() is used when checking if a node is balanced. |
2738 | * | |
2739 | * For order-0, all zones must be balanced! | |
2740 | * | |
2741 | * For high-order allocations only zones that meet watermarks and are in a | |
2742 | * zone allowed by the callers classzone_idx are added to balanced_pages. The | |
2743 | * total of balanced pages must be at least 25% of the zones allowed by | |
2744 | * classzone_idx for the node to be considered balanced. Forcing all zones to | |
2745 | * be balanced for high orders can cause excessive reclaim when there are | |
2746 | * imbalanced zones. | |
1741c877 MG |
2747 | * The choice of 25% is due to |
2748 | * o a 16M DMA zone that is balanced will not balance a zone on any | |
2749 | * reasonable sized machine | |
2750 | * o On all other machines, the top zone must be at least a reasonable | |
25985edc | 2751 | * percentage of the middle zones. For example, on 32-bit x86, highmem |
1741c877 MG |
2752 | * would need to be at least 256M for it to be balance a whole node. |
2753 | * Similarly, on x86-64 the Normal zone would need to be at least 1G | |
2754 | * to balance a node on its own. These seemed like reasonable ratios. | |
2755 | */ | |
4ae0a48b | 2756 | static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx) |
1741c877 | 2757 | { |
b40da049 | 2758 | unsigned long managed_pages = 0; |
4ae0a48b | 2759 | unsigned long balanced_pages = 0; |
1741c877 MG |
2760 | int i; |
2761 | ||
4ae0a48b ZC |
2762 | /* Check the watermark levels */ |
2763 | for (i = 0; i <= classzone_idx; i++) { | |
2764 | struct zone *zone = pgdat->node_zones + i; | |
1741c877 | 2765 | |
4ae0a48b ZC |
2766 | if (!populated_zone(zone)) |
2767 | continue; | |
2768 | ||
b40da049 | 2769 | managed_pages += zone->managed_pages; |
4ae0a48b ZC |
2770 | |
2771 | /* | |
2772 | * A special case here: | |
2773 | * | |
2774 | * balance_pgdat() skips over all_unreclaimable after | |
2775 | * DEF_PRIORITY. Effectively, it considers them balanced so | |
2776 | * they must be considered balanced here as well! | |
2777 | */ | |
2778 | if (zone->all_unreclaimable) { | |
b40da049 | 2779 | balanced_pages += zone->managed_pages; |
4ae0a48b ZC |
2780 | continue; |
2781 | } | |
2782 | ||
2783 | if (zone_balanced(zone, order, 0, i)) | |
b40da049 | 2784 | balanced_pages += zone->managed_pages; |
4ae0a48b ZC |
2785 | else if (!order) |
2786 | return false; | |
2787 | } | |
2788 | ||
2789 | if (order) | |
b40da049 | 2790 | return balanced_pages >= (managed_pages >> 2); |
4ae0a48b ZC |
2791 | else |
2792 | return true; | |
1741c877 MG |
2793 | } |
2794 | ||
5515061d MG |
2795 | /* |
2796 | * Prepare kswapd for sleeping. This verifies that there are no processes | |
2797 | * waiting in throttle_direct_reclaim() and that watermarks have been met. | |
2798 | * | |
2799 | * Returns true if kswapd is ready to sleep | |
2800 | */ | |
2801 | static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining, | |
dc83edd9 | 2802 | int classzone_idx) |
f50de2d3 | 2803 | { |
f50de2d3 MG |
2804 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ |
2805 | if (remaining) | |
5515061d MG |
2806 | return false; |
2807 | ||
2808 | /* | |
2809 | * There is a potential race between when kswapd checks its watermarks | |
2810 | * and a process gets throttled. There is also a potential race if | |
2811 | * processes get throttled, kswapd wakes, a large process exits therby | |
2812 | * balancing the zones that causes kswapd to miss a wakeup. If kswapd | |
2813 | * is going to sleep, no process should be sleeping on pfmemalloc_wait | |
2814 | * so wake them now if necessary. If necessary, processes will wake | |
2815 | * kswapd and get throttled again | |
2816 | */ | |
2817 | if (waitqueue_active(&pgdat->pfmemalloc_wait)) { | |
2818 | wake_up(&pgdat->pfmemalloc_wait); | |
2819 | return false; | |
2820 | } | |
f50de2d3 | 2821 | |
4ae0a48b | 2822 | return pgdat_balanced(pgdat, order, classzone_idx); |
f50de2d3 MG |
2823 | } |
2824 | ||
75485363 MG |
2825 | /* |
2826 | * kswapd shrinks the zone by the number of pages required to reach | |
2827 | * the high watermark. | |
b8e83b94 MG |
2828 | * |
2829 | * Returns true if kswapd scanned at least the requested number of pages to | |
283aba9f MG |
2830 | * reclaim or if the lack of progress was due to pages under writeback. |
2831 | * This is used to determine if the scanning priority needs to be raised. | |
75485363 | 2832 | */ |
b8e83b94 | 2833 | static bool kswapd_shrink_zone(struct zone *zone, |
7c954f6d | 2834 | int classzone_idx, |
75485363 | 2835 | struct scan_control *sc, |
2ab44f43 MG |
2836 | unsigned long lru_pages, |
2837 | unsigned long *nr_attempted) | |
75485363 MG |
2838 | { |
2839 | unsigned long nr_slab; | |
7c954f6d MG |
2840 | int testorder = sc->order; |
2841 | unsigned long balance_gap; | |
75485363 MG |
2842 | struct reclaim_state *reclaim_state = current->reclaim_state; |
2843 | struct shrink_control shrink = { | |
2844 | .gfp_mask = sc->gfp_mask, | |
2845 | }; | |
7c954f6d | 2846 | bool lowmem_pressure; |
75485363 MG |
2847 | |
2848 | /* Reclaim above the high watermark. */ | |
2849 | sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone)); | |
7c954f6d MG |
2850 | |
2851 | /* | |
2852 | * Kswapd reclaims only single pages with compaction enabled. Trying | |
2853 | * too hard to reclaim until contiguous free pages have become | |
2854 | * available can hurt performance by evicting too much useful data | |
2855 | * from memory. Do not reclaim more than needed for compaction. | |
2856 | */ | |
2857 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && | |
2858 | compaction_suitable(zone, sc->order) != | |
2859 | COMPACT_SKIPPED) | |
2860 | testorder = 0; | |
2861 | ||
2862 | /* | |
2863 | * We put equal pressure on every zone, unless one zone has way too | |
2864 | * many pages free already. The "too many pages" is defined as the | |
2865 | * high wmark plus a "gap" where the gap is either the low | |
2866 | * watermark or 1% of the zone, whichever is smaller. | |
2867 | */ | |
2868 | balance_gap = min(low_wmark_pages(zone), | |
2869 | (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / | |
2870 | KSWAPD_ZONE_BALANCE_GAP_RATIO); | |
2871 | ||
2872 | /* | |
2873 | * If there is no low memory pressure or the zone is balanced then no | |
2874 | * reclaim is necessary | |
2875 | */ | |
2876 | lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone)); | |
2877 | if (!lowmem_pressure && zone_balanced(zone, testorder, | |
2878 | balance_gap, classzone_idx)) | |
2879 | return true; | |
2880 | ||
75485363 | 2881 | shrink_zone(zone, sc); |
0ce3d744 DC |
2882 | nodes_clear(shrink.nodes_to_scan); |
2883 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); | |
75485363 MG |
2884 | |
2885 | reclaim_state->reclaimed_slab = 0; | |
2886 | nr_slab = shrink_slab(&shrink, sc->nr_scanned, lru_pages); | |
2887 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; | |
2888 | ||
2ab44f43 MG |
2889 | /* Account for the number of pages attempted to reclaim */ |
2890 | *nr_attempted += sc->nr_to_reclaim; | |
2891 | ||
75485363 MG |
2892 | if (nr_slab == 0 && !zone_reclaimable(zone)) |
2893 | zone->all_unreclaimable = 1; | |
b8e83b94 | 2894 | |
283aba9f MG |
2895 | zone_clear_flag(zone, ZONE_WRITEBACK); |
2896 | ||
7c954f6d MG |
2897 | /* |
2898 | * If a zone reaches its high watermark, consider it to be no longer | |
2899 | * congested. It's possible there are dirty pages backed by congested | |
2900 | * BDIs but as pressure is relieved, speculatively avoid congestion | |
2901 | * waits. | |
2902 | */ | |
2903 | if (!zone->all_unreclaimable && | |
2904 | zone_balanced(zone, testorder, 0, classzone_idx)) { | |
2905 | zone_clear_flag(zone, ZONE_CONGESTED); | |
2906 | zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY); | |
2907 | } | |
2908 | ||
b8e83b94 | 2909 | return sc->nr_scanned >= sc->nr_to_reclaim; |
75485363 MG |
2910 | } |
2911 | ||
1da177e4 LT |
2912 | /* |
2913 | * For kswapd, balance_pgdat() will work across all this node's zones until | |
41858966 | 2914 | * they are all at high_wmark_pages(zone). |
1da177e4 | 2915 | * |
0abdee2b | 2916 | * Returns the final order kswapd was reclaiming at |
1da177e4 LT |
2917 | * |
2918 | * There is special handling here for zones which are full of pinned pages. | |
2919 | * This can happen if the pages are all mlocked, or if they are all used by | |
2920 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. | |
2921 | * What we do is to detect the case where all pages in the zone have been | |
2922 | * scanned twice and there has been zero successful reclaim. Mark the zone as | |
2923 | * dead and from now on, only perform a short scan. Basically we're polling | |
2924 | * the zone for when the problem goes away. | |
2925 | * | |
2926 | * kswapd scans the zones in the highmem->normal->dma direction. It skips | |
41858966 MG |
2927 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
2928 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the | |
2929 | * lower zones regardless of the number of free pages in the lower zones. This | |
2930 | * interoperates with the page allocator fallback scheme to ensure that aging | |
2931 | * of pages is balanced across the zones. | |
1da177e4 | 2932 | */ |
99504748 | 2933 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, |
dc83edd9 | 2934 | int *classzone_idx) |
1da177e4 | 2935 | { |
1da177e4 | 2936 | int i; |
99504748 | 2937 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
0ae5e89c YH |
2938 | unsigned long nr_soft_reclaimed; |
2939 | unsigned long nr_soft_scanned; | |
179e9639 AM |
2940 | struct scan_control sc = { |
2941 | .gfp_mask = GFP_KERNEL, | |
b8e83b94 | 2942 | .priority = DEF_PRIORITY, |
a6dc60f8 | 2943 | .may_unmap = 1, |
2e2e4259 | 2944 | .may_swap = 1, |
b8e83b94 | 2945 | .may_writepage = !laptop_mode, |
5ad333eb | 2946 | .order = order, |
f16015fb | 2947 | .target_mem_cgroup = NULL, |
179e9639 | 2948 | }; |
f8891e5e | 2949 | count_vm_event(PAGEOUTRUN); |
1da177e4 | 2950 | |
9e3b2f8c | 2951 | do { |
1da177e4 | 2952 | unsigned long lru_pages = 0; |
2ab44f43 | 2953 | unsigned long nr_attempted = 0; |
b8e83b94 | 2954 | bool raise_priority = true; |
2ab44f43 | 2955 | bool pgdat_needs_compaction = (order > 0); |
b8e83b94 MG |
2956 | |
2957 | sc.nr_reclaimed = 0; | |
1da177e4 | 2958 | |
d6277db4 RW |
2959 | /* |
2960 | * Scan in the highmem->dma direction for the highest | |
2961 | * zone which needs scanning | |
2962 | */ | |
2963 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { | |
2964 | struct zone *zone = pgdat->node_zones + i; | |
1da177e4 | 2965 | |
d6277db4 RW |
2966 | if (!populated_zone(zone)) |
2967 | continue; | |
1da177e4 | 2968 | |
9e3b2f8c KK |
2969 | if (zone->all_unreclaimable && |
2970 | sc.priority != DEF_PRIORITY) | |
d6277db4 | 2971 | continue; |
1da177e4 | 2972 | |
556adecb RR |
2973 | /* |
2974 | * Do some background aging of the anon list, to give | |
2975 | * pages a chance to be referenced before reclaiming. | |
2976 | */ | |
9e3b2f8c | 2977 | age_active_anon(zone, &sc); |
556adecb | 2978 | |
cc715d99 MG |
2979 | /* |
2980 | * If the number of buffer_heads in the machine | |
2981 | * exceeds the maximum allowed level and this node | |
2982 | * has a highmem zone, force kswapd to reclaim from | |
2983 | * it to relieve lowmem pressure. | |
2984 | */ | |
2985 | if (buffer_heads_over_limit && is_highmem_idx(i)) { | |
2986 | end_zone = i; | |
2987 | break; | |
2988 | } | |
2989 | ||
60cefed4 | 2990 | if (!zone_balanced(zone, order, 0, 0)) { |
d6277db4 | 2991 | end_zone = i; |
e1dbeda6 | 2992 | break; |
439423f6 | 2993 | } else { |
d43006d5 MG |
2994 | /* |
2995 | * If balanced, clear the dirty and congested | |
2996 | * flags | |
2997 | */ | |
439423f6 | 2998 | zone_clear_flag(zone, ZONE_CONGESTED); |
d43006d5 | 2999 | zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY); |
1da177e4 | 3000 | } |
1da177e4 | 3001 | } |
dafcb73e | 3002 | |
b8e83b94 | 3003 | if (i < 0) |
e1dbeda6 AM |
3004 | goto out; |
3005 | ||
1da177e4 LT |
3006 | for (i = 0; i <= end_zone; i++) { |
3007 | struct zone *zone = pgdat->node_zones + i; | |
3008 | ||
2ab44f43 MG |
3009 | if (!populated_zone(zone)) |
3010 | continue; | |
3011 | ||
adea02a1 | 3012 | lru_pages += zone_reclaimable_pages(zone); |
2ab44f43 MG |
3013 | |
3014 | /* | |
3015 | * If any zone is currently balanced then kswapd will | |
3016 | * not call compaction as it is expected that the | |
3017 | * necessary pages are already available. | |
3018 | */ | |
3019 | if (pgdat_needs_compaction && | |
3020 | zone_watermark_ok(zone, order, | |
3021 | low_wmark_pages(zone), | |
3022 | *classzone_idx, 0)) | |
3023 | pgdat_needs_compaction = false; | |
1da177e4 LT |
3024 | } |
3025 | ||
b7ea3c41 MG |
3026 | /* |
3027 | * If we're getting trouble reclaiming, start doing writepage | |
3028 | * even in laptop mode. | |
3029 | */ | |
3030 | if (sc.priority < DEF_PRIORITY - 2) | |
3031 | sc.may_writepage = 1; | |
3032 | ||
1da177e4 LT |
3033 | /* |
3034 | * Now scan the zone in the dma->highmem direction, stopping | |
3035 | * at the last zone which needs scanning. | |
3036 | * | |
3037 | * We do this because the page allocator works in the opposite | |
3038 | * direction. This prevents the page allocator from allocating | |
3039 | * pages behind kswapd's direction of progress, which would | |
3040 | * cause too much scanning of the lower zones. | |
3041 | */ | |
3042 | for (i = 0; i <= end_zone; i++) { | |
3043 | struct zone *zone = pgdat->node_zones + i; | |
3044 | ||
f3fe6512 | 3045 | if (!populated_zone(zone)) |
1da177e4 LT |
3046 | continue; |
3047 | ||
9e3b2f8c KK |
3048 | if (zone->all_unreclaimable && |
3049 | sc.priority != DEF_PRIORITY) | |
1da177e4 LT |
3050 | continue; |
3051 | ||
1da177e4 | 3052 | sc.nr_scanned = 0; |
4e416953 | 3053 | |
0ae5e89c | 3054 | nr_soft_scanned = 0; |
4e416953 BS |
3055 | /* |
3056 | * Call soft limit reclaim before calling shrink_zone. | |
4e416953 | 3057 | */ |
0ae5e89c YH |
3058 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
3059 | order, sc.gfp_mask, | |
3060 | &nr_soft_scanned); | |
3061 | sc.nr_reclaimed += nr_soft_reclaimed; | |
00918b6a | 3062 | |
32a4330d | 3063 | /* |
7c954f6d MG |
3064 | * There should be no need to raise the scanning |
3065 | * priority if enough pages are already being scanned | |
3066 | * that that high watermark would be met at 100% | |
3067 | * efficiency. | |
fe2c2a10 | 3068 | */ |
7c954f6d MG |
3069 | if (kswapd_shrink_zone(zone, end_zone, &sc, |
3070 | lru_pages, &nr_attempted)) | |
3071 | raise_priority = false; | |
1da177e4 | 3072 | } |
5515061d MG |
3073 | |
3074 | /* | |
3075 | * If the low watermark is met there is no need for processes | |
3076 | * to be throttled on pfmemalloc_wait as they should not be | |
3077 | * able to safely make forward progress. Wake them | |
3078 | */ | |
3079 | if (waitqueue_active(&pgdat->pfmemalloc_wait) && | |
3080 | pfmemalloc_watermark_ok(pgdat)) | |
3081 | wake_up(&pgdat->pfmemalloc_wait); | |
3082 | ||
1da177e4 | 3083 | /* |
b8e83b94 MG |
3084 | * Fragmentation may mean that the system cannot be rebalanced |
3085 | * for high-order allocations in all zones. If twice the | |
3086 | * allocation size has been reclaimed and the zones are still | |
3087 | * not balanced then recheck the watermarks at order-0 to | |
3088 | * prevent kswapd reclaiming excessively. Assume that a | |
3089 | * process requested a high-order can direct reclaim/compact. | |
1da177e4 | 3090 | */ |
b8e83b94 MG |
3091 | if (order && sc.nr_reclaimed >= 2UL << order) |
3092 | order = sc.order = 0; | |
8357376d | 3093 | |
b8e83b94 MG |
3094 | /* Check if kswapd should be suspending */ |
3095 | if (try_to_freeze() || kthread_should_stop()) | |
3096 | break; | |
8357376d | 3097 | |
2ab44f43 MG |
3098 | /* |
3099 | * Compact if necessary and kswapd is reclaiming at least the | |
3100 | * high watermark number of pages as requsted | |
3101 | */ | |
3102 | if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted) | |
3103 | compact_pgdat(pgdat, order); | |
3104 | ||
73ce02e9 | 3105 | /* |
b8e83b94 MG |
3106 | * Raise priority if scanning rate is too low or there was no |
3107 | * progress in reclaiming pages | |
73ce02e9 | 3108 | */ |
b8e83b94 MG |
3109 | if (raise_priority || !sc.nr_reclaimed) |
3110 | sc.priority--; | |
9aa41348 | 3111 | } while (sc.priority >= 1 && |
b8e83b94 | 3112 | !pgdat_balanced(pgdat, order, *classzone_idx)); |
1da177e4 | 3113 | |
b8e83b94 | 3114 | out: |
0abdee2b | 3115 | /* |
5515061d | 3116 | * Return the order we were reclaiming at so prepare_kswapd_sleep() |
0abdee2b MG |
3117 | * makes a decision on the order we were last reclaiming at. However, |
3118 | * if another caller entered the allocator slow path while kswapd | |
3119 | * was awake, order will remain at the higher level | |
3120 | */ | |
dc83edd9 | 3121 | *classzone_idx = end_zone; |
0abdee2b | 3122 | return order; |
1da177e4 LT |
3123 | } |
3124 | ||
dc83edd9 | 3125 | static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx) |
f0bc0a60 KM |
3126 | { |
3127 | long remaining = 0; | |
3128 | DEFINE_WAIT(wait); | |
3129 | ||
3130 | if (freezing(current) || kthread_should_stop()) | |
3131 | return; | |
3132 | ||
3133 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | |
3134 | ||
3135 | /* Try to sleep for a short interval */ | |
5515061d | 3136 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
f0bc0a60 KM |
3137 | remaining = schedule_timeout(HZ/10); |
3138 | finish_wait(&pgdat->kswapd_wait, &wait); | |
3139 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | |
3140 | } | |
3141 | ||
3142 | /* | |
3143 | * After a short sleep, check if it was a premature sleep. If not, then | |
3144 | * go fully to sleep until explicitly woken up. | |
3145 | */ | |
5515061d | 3146 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
f0bc0a60 KM |
3147 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
3148 | ||
3149 | /* | |
3150 | * vmstat counters are not perfectly accurate and the estimated | |
3151 | * value for counters such as NR_FREE_PAGES can deviate from the | |
3152 | * true value by nr_online_cpus * threshold. To avoid the zone | |
3153 | * watermarks being breached while under pressure, we reduce the | |
3154 | * per-cpu vmstat threshold while kswapd is awake and restore | |
3155 | * them before going back to sleep. | |
3156 | */ | |
3157 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); | |
1c7e7f6c | 3158 | |
62997027 MG |
3159 | /* |
3160 | * Compaction records what page blocks it recently failed to | |
3161 | * isolate pages from and skips them in the future scanning. | |
3162 | * When kswapd is going to sleep, it is reasonable to assume | |
3163 | * that pages and compaction may succeed so reset the cache. | |
3164 | */ | |
3165 | reset_isolation_suitable(pgdat); | |
3166 | ||
1c7e7f6c AK |
3167 | if (!kthread_should_stop()) |
3168 | schedule(); | |
3169 | ||
f0bc0a60 KM |
3170 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
3171 | } else { | |
3172 | if (remaining) | |
3173 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); | |
3174 | else | |
3175 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); | |
3176 | } | |
3177 | finish_wait(&pgdat->kswapd_wait, &wait); | |
3178 | } | |
3179 | ||
1da177e4 LT |
3180 | /* |
3181 | * The background pageout daemon, started as a kernel thread | |
4f98a2fe | 3182 | * from the init process. |
1da177e4 LT |
3183 | * |
3184 | * This basically trickles out pages so that we have _some_ | |
3185 | * free memory available even if there is no other activity | |
3186 | * that frees anything up. This is needed for things like routing | |
3187 | * etc, where we otherwise might have all activity going on in | |
3188 | * asynchronous contexts that cannot page things out. | |
3189 | * | |
3190 | * If there are applications that are active memory-allocators | |
3191 | * (most normal use), this basically shouldn't matter. | |
3192 | */ | |
3193 | static int kswapd(void *p) | |
3194 | { | |
215ddd66 | 3195 | unsigned long order, new_order; |
d2ebd0f6 | 3196 | unsigned balanced_order; |
215ddd66 | 3197 | int classzone_idx, new_classzone_idx; |
d2ebd0f6 | 3198 | int balanced_classzone_idx; |
1da177e4 LT |
3199 | pg_data_t *pgdat = (pg_data_t*)p; |
3200 | struct task_struct *tsk = current; | |
f0bc0a60 | 3201 | |
1da177e4 LT |
3202 | struct reclaim_state reclaim_state = { |
3203 | .reclaimed_slab = 0, | |
3204 | }; | |
a70f7302 | 3205 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
1da177e4 | 3206 | |
cf40bd16 NP |
3207 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
3208 | ||
174596a0 | 3209 | if (!cpumask_empty(cpumask)) |
c5f59f08 | 3210 | set_cpus_allowed_ptr(tsk, cpumask); |
1da177e4 LT |
3211 | current->reclaim_state = &reclaim_state; |
3212 | ||
3213 | /* | |
3214 | * Tell the memory management that we're a "memory allocator", | |
3215 | * and that if we need more memory we should get access to it | |
3216 | * regardless (see "__alloc_pages()"). "kswapd" should | |
3217 | * never get caught in the normal page freeing logic. | |
3218 | * | |
3219 | * (Kswapd normally doesn't need memory anyway, but sometimes | |
3220 | * you need a small amount of memory in order to be able to | |
3221 | * page out something else, and this flag essentially protects | |
3222 | * us from recursively trying to free more memory as we're | |
3223 | * trying to free the first piece of memory in the first place). | |
3224 | */ | |
930d9152 | 3225 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
83144186 | 3226 | set_freezable(); |
1da177e4 | 3227 | |
215ddd66 | 3228 | order = new_order = 0; |
d2ebd0f6 | 3229 | balanced_order = 0; |
215ddd66 | 3230 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; |
d2ebd0f6 | 3231 | balanced_classzone_idx = classzone_idx; |
1da177e4 | 3232 | for ( ; ; ) { |
6f6313d4 | 3233 | bool ret; |
3e1d1d28 | 3234 | |
215ddd66 MG |
3235 | /* |
3236 | * If the last balance_pgdat was unsuccessful it's unlikely a | |
3237 | * new request of a similar or harder type will succeed soon | |
3238 | * so consider going to sleep on the basis we reclaimed at | |
3239 | */ | |
d2ebd0f6 AS |
3240 | if (balanced_classzone_idx >= new_classzone_idx && |
3241 | balanced_order == new_order) { | |
215ddd66 MG |
3242 | new_order = pgdat->kswapd_max_order; |
3243 | new_classzone_idx = pgdat->classzone_idx; | |
3244 | pgdat->kswapd_max_order = 0; | |
3245 | pgdat->classzone_idx = pgdat->nr_zones - 1; | |
3246 | } | |
3247 | ||
99504748 | 3248 | if (order < new_order || classzone_idx > new_classzone_idx) { |
1da177e4 LT |
3249 | /* |
3250 | * Don't sleep if someone wants a larger 'order' | |
99504748 | 3251 | * allocation or has tigher zone constraints |
1da177e4 LT |
3252 | */ |
3253 | order = new_order; | |
99504748 | 3254 | classzone_idx = new_classzone_idx; |
1da177e4 | 3255 | } else { |
d2ebd0f6 AS |
3256 | kswapd_try_to_sleep(pgdat, balanced_order, |
3257 | balanced_classzone_idx); | |
1da177e4 | 3258 | order = pgdat->kswapd_max_order; |
99504748 | 3259 | classzone_idx = pgdat->classzone_idx; |
f0dfcde0 AS |
3260 | new_order = order; |
3261 | new_classzone_idx = classzone_idx; | |
4d40502e | 3262 | pgdat->kswapd_max_order = 0; |
215ddd66 | 3263 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
1da177e4 | 3264 | } |
1da177e4 | 3265 | |
8fe23e05 DR |
3266 | ret = try_to_freeze(); |
3267 | if (kthread_should_stop()) | |
3268 | break; | |
3269 | ||
3270 | /* | |
3271 | * We can speed up thawing tasks if we don't call balance_pgdat | |
3272 | * after returning from the refrigerator | |
3273 | */ | |
33906bc5 MG |
3274 | if (!ret) { |
3275 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); | |
d2ebd0f6 AS |
3276 | balanced_classzone_idx = classzone_idx; |
3277 | balanced_order = balance_pgdat(pgdat, order, | |
3278 | &balanced_classzone_idx); | |
33906bc5 | 3279 | } |
1da177e4 | 3280 | } |
b0a8cc58 TY |
3281 | |
3282 | current->reclaim_state = NULL; | |
1da177e4 LT |
3283 | return 0; |
3284 | } | |
3285 | ||
3286 | /* | |
3287 | * A zone is low on free memory, so wake its kswapd task to service it. | |
3288 | */ | |
99504748 | 3289 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx) |
1da177e4 LT |
3290 | { |
3291 | pg_data_t *pgdat; | |
3292 | ||
f3fe6512 | 3293 | if (!populated_zone(zone)) |
1da177e4 LT |
3294 | return; |
3295 | ||
88f5acf8 | 3296 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
1da177e4 | 3297 | return; |
88f5acf8 | 3298 | pgdat = zone->zone_pgdat; |
99504748 | 3299 | if (pgdat->kswapd_max_order < order) { |
1da177e4 | 3300 | pgdat->kswapd_max_order = order; |
99504748 MG |
3301 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); |
3302 | } | |
8d0986e2 | 3303 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
1da177e4 | 3304 | return; |
88f5acf8 MG |
3305 | if (zone_watermark_ok_safe(zone, order, low_wmark_pages(zone), 0, 0)) |
3306 | return; | |
3307 | ||
3308 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); | |
8d0986e2 | 3309 | wake_up_interruptible(&pgdat->kswapd_wait); |
1da177e4 LT |
3310 | } |
3311 | ||
adea02a1 WF |
3312 | /* |
3313 | * The reclaimable count would be mostly accurate. | |
3314 | * The less reclaimable pages may be | |
3315 | * - mlocked pages, which will be moved to unevictable list when encountered | |
3316 | * - mapped pages, which may require several travels to be reclaimed | |
3317 | * - dirty pages, which is not "instantly" reclaimable | |
3318 | */ | |
3319 | unsigned long global_reclaimable_pages(void) | |
4f98a2fe | 3320 | { |
adea02a1 WF |
3321 | int nr; |
3322 | ||
3323 | nr = global_page_state(NR_ACTIVE_FILE) + | |
3324 | global_page_state(NR_INACTIVE_FILE); | |
3325 | ||
ec8acf20 | 3326 | if (get_nr_swap_pages() > 0) |
adea02a1 WF |
3327 | nr += global_page_state(NR_ACTIVE_ANON) + |
3328 | global_page_state(NR_INACTIVE_ANON); | |
3329 | ||
3330 | return nr; | |
3331 | } | |
3332 | ||
3333 | unsigned long zone_reclaimable_pages(struct zone *zone) | |
3334 | { | |
3335 | int nr; | |
3336 | ||
3337 | nr = zone_page_state(zone, NR_ACTIVE_FILE) + | |
3338 | zone_page_state(zone, NR_INACTIVE_FILE); | |
3339 | ||
ec8acf20 | 3340 | if (get_nr_swap_pages() > 0) |
adea02a1 WF |
3341 | nr += zone_page_state(zone, NR_ACTIVE_ANON) + |
3342 | zone_page_state(zone, NR_INACTIVE_ANON); | |
3343 | ||
3344 | return nr; | |
4f98a2fe RR |
3345 | } |
3346 | ||
c6f37f12 | 3347 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3348 | /* |
7b51755c | 3349 | * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of |
d6277db4 RW |
3350 | * freed pages. |
3351 | * | |
3352 | * Rather than trying to age LRUs the aim is to preserve the overall | |
3353 | * LRU order by reclaiming preferentially | |
3354 | * inactive > active > active referenced > active mapped | |
1da177e4 | 3355 | */ |
7b51755c | 3356 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
1da177e4 | 3357 | { |
d6277db4 | 3358 | struct reclaim_state reclaim_state; |
d6277db4 | 3359 | struct scan_control sc = { |
7b51755c KM |
3360 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
3361 | .may_swap = 1, | |
3362 | .may_unmap = 1, | |
d6277db4 | 3363 | .may_writepage = 1, |
7b51755c KM |
3364 | .nr_to_reclaim = nr_to_reclaim, |
3365 | .hibernation_mode = 1, | |
7b51755c | 3366 | .order = 0, |
9e3b2f8c | 3367 | .priority = DEF_PRIORITY, |
1da177e4 | 3368 | }; |
a09ed5e0 YH |
3369 | struct shrink_control shrink = { |
3370 | .gfp_mask = sc.gfp_mask, | |
3371 | }; | |
3372 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); | |
7b51755c KM |
3373 | struct task_struct *p = current; |
3374 | unsigned long nr_reclaimed; | |
1da177e4 | 3375 | |
7b51755c KM |
3376 | p->flags |= PF_MEMALLOC; |
3377 | lockdep_set_current_reclaim_state(sc.gfp_mask); | |
3378 | reclaim_state.reclaimed_slab = 0; | |
3379 | p->reclaim_state = &reclaim_state; | |
d6277db4 | 3380 | |
a09ed5e0 | 3381 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
d979677c | 3382 | |
7b51755c KM |
3383 | p->reclaim_state = NULL; |
3384 | lockdep_clear_current_reclaim_state(); | |
3385 | p->flags &= ~PF_MEMALLOC; | |
d6277db4 | 3386 | |
7b51755c | 3387 | return nr_reclaimed; |
1da177e4 | 3388 | } |
c6f37f12 | 3389 | #endif /* CONFIG_HIBERNATION */ |
1da177e4 | 3390 | |
1da177e4 LT |
3391 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
3392 | not required for correctness. So if the last cpu in a node goes | |
3393 | away, we get changed to run anywhere: as the first one comes back, | |
3394 | restore their cpu bindings. */ | |
fcb35a9b GKH |
3395 | static int cpu_callback(struct notifier_block *nfb, unsigned long action, |
3396 | void *hcpu) | |
1da177e4 | 3397 | { |
58c0a4a7 | 3398 | int nid; |
1da177e4 | 3399 | |
8bb78442 | 3400 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
48fb2e24 | 3401 | for_each_node_state(nid, N_MEMORY) { |
c5f59f08 | 3402 | pg_data_t *pgdat = NODE_DATA(nid); |
a70f7302 RR |
3403 | const struct cpumask *mask; |
3404 | ||
3405 | mask = cpumask_of_node(pgdat->node_id); | |
c5f59f08 | 3406 | |
3e597945 | 3407 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
1da177e4 | 3408 | /* One of our CPUs online: restore mask */ |
c5f59f08 | 3409 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
1da177e4 LT |
3410 | } |
3411 | } | |
3412 | return NOTIFY_OK; | |
3413 | } | |
1da177e4 | 3414 | |
3218ae14 YG |
3415 | /* |
3416 | * This kswapd start function will be called by init and node-hot-add. | |
3417 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. | |
3418 | */ | |
3419 | int kswapd_run(int nid) | |
3420 | { | |
3421 | pg_data_t *pgdat = NODE_DATA(nid); | |
3422 | int ret = 0; | |
3423 | ||
3424 | if (pgdat->kswapd) | |
3425 | return 0; | |
3426 | ||
3427 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); | |
3428 | if (IS_ERR(pgdat->kswapd)) { | |
3429 | /* failure at boot is fatal */ | |
3430 | BUG_ON(system_state == SYSTEM_BOOTING); | |
d5dc0ad9 GS |
3431 | pr_err("Failed to start kswapd on node %d\n", nid); |
3432 | ret = PTR_ERR(pgdat->kswapd); | |
d72515b8 | 3433 | pgdat->kswapd = NULL; |
3218ae14 YG |
3434 | } |
3435 | return ret; | |
3436 | } | |
3437 | ||
8fe23e05 | 3438 | /* |
d8adde17 JL |
3439 | * Called by memory hotplug when all memory in a node is offlined. Caller must |
3440 | * hold lock_memory_hotplug(). | |
8fe23e05 DR |
3441 | */ |
3442 | void kswapd_stop(int nid) | |
3443 | { | |
3444 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; | |
3445 | ||
d8adde17 | 3446 | if (kswapd) { |
8fe23e05 | 3447 | kthread_stop(kswapd); |
d8adde17 JL |
3448 | NODE_DATA(nid)->kswapd = NULL; |
3449 | } | |
8fe23e05 DR |
3450 | } |
3451 | ||
1da177e4 LT |
3452 | static int __init kswapd_init(void) |
3453 | { | |
3218ae14 | 3454 | int nid; |
69e05944 | 3455 | |
1da177e4 | 3456 | swap_setup(); |
48fb2e24 | 3457 | for_each_node_state(nid, N_MEMORY) |
3218ae14 | 3458 | kswapd_run(nid); |
1da177e4 LT |
3459 | hotcpu_notifier(cpu_callback, 0); |
3460 | return 0; | |
3461 | } | |
3462 | ||
3463 | module_init(kswapd_init) | |
9eeff239 CL |
3464 | |
3465 | #ifdef CONFIG_NUMA | |
3466 | /* | |
3467 | * Zone reclaim mode | |
3468 | * | |
3469 | * If non-zero call zone_reclaim when the number of free pages falls below | |
3470 | * the watermarks. | |
9eeff239 CL |
3471 | */ |
3472 | int zone_reclaim_mode __read_mostly; | |
3473 | ||
1b2ffb78 | 3474 | #define RECLAIM_OFF 0 |
7d03431c | 3475 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
1b2ffb78 CL |
3476 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
3477 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ | |
3478 | ||
a92f7126 CL |
3479 | /* |
3480 | * Priority for ZONE_RECLAIM. This determines the fraction of pages | |
3481 | * of a node considered for each zone_reclaim. 4 scans 1/16th of | |
3482 | * a zone. | |
3483 | */ | |
3484 | #define ZONE_RECLAIM_PRIORITY 4 | |
3485 | ||
9614634f CL |
3486 | /* |
3487 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to | |
3488 | * occur. | |
3489 | */ | |
3490 | int sysctl_min_unmapped_ratio = 1; | |
3491 | ||
0ff38490 CL |
3492 | /* |
3493 | * If the number of slab pages in a zone grows beyond this percentage then | |
3494 | * slab reclaim needs to occur. | |
3495 | */ | |
3496 | int sysctl_min_slab_ratio = 5; | |
3497 | ||
90afa5de MG |
3498 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) |
3499 | { | |
3500 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); | |
3501 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + | |
3502 | zone_page_state(zone, NR_ACTIVE_FILE); | |
3503 | ||
3504 | /* | |
3505 | * It's possible for there to be more file mapped pages than | |
3506 | * accounted for by the pages on the file LRU lists because | |
3507 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED | |
3508 | */ | |
3509 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; | |
3510 | } | |
3511 | ||
3512 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ | |
3513 | static long zone_pagecache_reclaimable(struct zone *zone) | |
3514 | { | |
3515 | long nr_pagecache_reclaimable; | |
3516 | long delta = 0; | |
3517 | ||
3518 | /* | |
3519 | * If RECLAIM_SWAP is set, then all file pages are considered | |
3520 | * potentially reclaimable. Otherwise, we have to worry about | |
3521 | * pages like swapcache and zone_unmapped_file_pages() provides | |
3522 | * a better estimate | |
3523 | */ | |
3524 | if (zone_reclaim_mode & RECLAIM_SWAP) | |
3525 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); | |
3526 | else | |
3527 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); | |
3528 | ||
3529 | /* If we can't clean pages, remove dirty pages from consideration */ | |
3530 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) | |
3531 | delta += zone_page_state(zone, NR_FILE_DIRTY); | |
3532 | ||
3533 | /* Watch for any possible underflows due to delta */ | |
3534 | if (unlikely(delta > nr_pagecache_reclaimable)) | |
3535 | delta = nr_pagecache_reclaimable; | |
3536 | ||
3537 | return nr_pagecache_reclaimable - delta; | |
3538 | } | |
3539 | ||
9eeff239 CL |
3540 | /* |
3541 | * Try to free up some pages from this zone through reclaim. | |
3542 | */ | |
179e9639 | 3543 | static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
9eeff239 | 3544 | { |
7fb2d46d | 3545 | /* Minimum pages needed in order to stay on node */ |
69e05944 | 3546 | const unsigned long nr_pages = 1 << order; |
9eeff239 CL |
3547 | struct task_struct *p = current; |
3548 | struct reclaim_state reclaim_state; | |
179e9639 AM |
3549 | struct scan_control sc = { |
3550 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), | |
a6dc60f8 | 3551 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP), |
2e2e4259 | 3552 | .may_swap = 1, |
62b726c1 | 3553 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
21caf2fc | 3554 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
bd2f6199 | 3555 | .order = order, |
9e3b2f8c | 3556 | .priority = ZONE_RECLAIM_PRIORITY, |
179e9639 | 3557 | }; |
a09ed5e0 YH |
3558 | struct shrink_control shrink = { |
3559 | .gfp_mask = sc.gfp_mask, | |
3560 | }; | |
15748048 | 3561 | unsigned long nr_slab_pages0, nr_slab_pages1; |
9eeff239 | 3562 | |
9eeff239 | 3563 | cond_resched(); |
d4f7796e CL |
3564 | /* |
3565 | * We need to be able to allocate from the reserves for RECLAIM_SWAP | |
3566 | * and we also need to be able to write out pages for RECLAIM_WRITE | |
3567 | * and RECLAIM_SWAP. | |
3568 | */ | |
3569 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; | |
76ca542d | 3570 | lockdep_set_current_reclaim_state(gfp_mask); |
9eeff239 CL |
3571 | reclaim_state.reclaimed_slab = 0; |
3572 | p->reclaim_state = &reclaim_state; | |
c84db23c | 3573 | |
90afa5de | 3574 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { |
0ff38490 CL |
3575 | /* |
3576 | * Free memory by calling shrink zone with increasing | |
3577 | * priorities until we have enough memory freed. | |
3578 | */ | |
0ff38490 | 3579 | do { |
9e3b2f8c KK |
3580 | shrink_zone(zone, &sc); |
3581 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); | |
0ff38490 | 3582 | } |
c84db23c | 3583 | |
15748048 KM |
3584 | nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
3585 | if (nr_slab_pages0 > zone->min_slab_pages) { | |
2a16e3f4 | 3586 | /* |
7fb2d46d | 3587 | * shrink_slab() does not currently allow us to determine how |
0ff38490 CL |
3588 | * many pages were freed in this zone. So we take the current |
3589 | * number of slab pages and shake the slab until it is reduced | |
3590 | * by the same nr_pages that we used for reclaiming unmapped | |
3591 | * pages. | |
2a16e3f4 | 3592 | */ |
0ce3d744 DC |
3593 | nodes_clear(shrink.nodes_to_scan); |
3594 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); | |
4dc4b3d9 KM |
3595 | for (;;) { |
3596 | unsigned long lru_pages = zone_reclaimable_pages(zone); | |
3597 | ||
3598 | /* No reclaimable slab or very low memory pressure */ | |
1495f230 | 3599 | if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages)) |
4dc4b3d9 KM |
3600 | break; |
3601 | ||
3602 | /* Freed enough memory */ | |
3603 | nr_slab_pages1 = zone_page_state(zone, | |
3604 | NR_SLAB_RECLAIMABLE); | |
3605 | if (nr_slab_pages1 + nr_pages <= nr_slab_pages0) | |
3606 | break; | |
3607 | } | |
83e33a47 CL |
3608 | |
3609 | /* | |
3610 | * Update nr_reclaimed by the number of slab pages we | |
3611 | * reclaimed from this zone. | |
3612 | */ | |
15748048 KM |
3613 | nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
3614 | if (nr_slab_pages1 < nr_slab_pages0) | |
3615 | sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1; | |
2a16e3f4 CL |
3616 | } |
3617 | ||
9eeff239 | 3618 | p->reclaim_state = NULL; |
d4f7796e | 3619 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
76ca542d | 3620 | lockdep_clear_current_reclaim_state(); |
a79311c1 | 3621 | return sc.nr_reclaimed >= nr_pages; |
9eeff239 | 3622 | } |
179e9639 AM |
3623 | |
3624 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | |
3625 | { | |
179e9639 | 3626 | int node_id; |
d773ed6b | 3627 | int ret; |
179e9639 AM |
3628 | |
3629 | /* | |
0ff38490 CL |
3630 | * Zone reclaim reclaims unmapped file backed pages and |
3631 | * slab pages if we are over the defined limits. | |
34aa1330 | 3632 | * |
9614634f CL |
3633 | * A small portion of unmapped file backed pages is needed for |
3634 | * file I/O otherwise pages read by file I/O will be immediately | |
3635 | * thrown out if the zone is overallocated. So we do not reclaim | |
3636 | * if less than a specified percentage of the zone is used by | |
3637 | * unmapped file backed pages. | |
179e9639 | 3638 | */ |
90afa5de MG |
3639 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && |
3640 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) | |
fa5e084e | 3641 | return ZONE_RECLAIM_FULL; |
179e9639 | 3642 | |
93e4a89a | 3643 | if (zone->all_unreclaimable) |
fa5e084e | 3644 | return ZONE_RECLAIM_FULL; |
d773ed6b | 3645 | |
179e9639 | 3646 | /* |
d773ed6b | 3647 | * Do not scan if the allocation should not be delayed. |
179e9639 | 3648 | */ |
d773ed6b | 3649 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
fa5e084e | 3650 | return ZONE_RECLAIM_NOSCAN; |
179e9639 AM |
3651 | |
3652 | /* | |
3653 | * Only run zone reclaim on the local zone or on zones that do not | |
3654 | * have associated processors. This will favor the local processor | |
3655 | * over remote processors and spread off node memory allocations | |
3656 | * as wide as possible. | |
3657 | */ | |
89fa3024 | 3658 | node_id = zone_to_nid(zone); |
37c0708d | 3659 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
fa5e084e | 3660 | return ZONE_RECLAIM_NOSCAN; |
d773ed6b DR |
3661 | |
3662 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) | |
fa5e084e MG |
3663 | return ZONE_RECLAIM_NOSCAN; |
3664 | ||
d773ed6b DR |
3665 | ret = __zone_reclaim(zone, gfp_mask, order); |
3666 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); | |
3667 | ||
24cf7251 MG |
3668 | if (!ret) |
3669 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); | |
3670 | ||
d773ed6b | 3671 | return ret; |
179e9639 | 3672 | } |
9eeff239 | 3673 | #endif |
894bc310 | 3674 | |
894bc310 LS |
3675 | /* |
3676 | * page_evictable - test whether a page is evictable | |
3677 | * @page: the page to test | |
894bc310 LS |
3678 | * |
3679 | * Test whether page is evictable--i.e., should be placed on active/inactive | |
39b5f29a | 3680 | * lists vs unevictable list. |
894bc310 LS |
3681 | * |
3682 | * Reasons page might not be evictable: | |
ba9ddf49 | 3683 | * (1) page's mapping marked unevictable |
b291f000 | 3684 | * (2) page is part of an mlocked VMA |
ba9ddf49 | 3685 | * |
894bc310 | 3686 | */ |
39b5f29a | 3687 | int page_evictable(struct page *page) |
894bc310 | 3688 | { |
39b5f29a | 3689 | return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); |
894bc310 | 3690 | } |
89e004ea | 3691 | |
85046579 | 3692 | #ifdef CONFIG_SHMEM |
89e004ea | 3693 | /** |
24513264 HD |
3694 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list |
3695 | * @pages: array of pages to check | |
3696 | * @nr_pages: number of pages to check | |
89e004ea | 3697 | * |
24513264 | 3698 | * Checks pages for evictability and moves them to the appropriate lru list. |
85046579 HD |
3699 | * |
3700 | * This function is only used for SysV IPC SHM_UNLOCK. | |
89e004ea | 3701 | */ |
24513264 | 3702 | void check_move_unevictable_pages(struct page **pages, int nr_pages) |
89e004ea | 3703 | { |
925b7673 | 3704 | struct lruvec *lruvec; |
24513264 HD |
3705 | struct zone *zone = NULL; |
3706 | int pgscanned = 0; | |
3707 | int pgrescued = 0; | |
3708 | int i; | |
89e004ea | 3709 | |
24513264 HD |
3710 | for (i = 0; i < nr_pages; i++) { |
3711 | struct page *page = pages[i]; | |
3712 | struct zone *pagezone; | |
89e004ea | 3713 | |
24513264 HD |
3714 | pgscanned++; |
3715 | pagezone = page_zone(page); | |
3716 | if (pagezone != zone) { | |
3717 | if (zone) | |
3718 | spin_unlock_irq(&zone->lru_lock); | |
3719 | zone = pagezone; | |
3720 | spin_lock_irq(&zone->lru_lock); | |
3721 | } | |
fa9add64 | 3722 | lruvec = mem_cgroup_page_lruvec(page, zone); |
89e004ea | 3723 | |
24513264 HD |
3724 | if (!PageLRU(page) || !PageUnevictable(page)) |
3725 | continue; | |
89e004ea | 3726 | |
39b5f29a | 3727 | if (page_evictable(page)) { |
24513264 HD |
3728 | enum lru_list lru = page_lru_base_type(page); |
3729 | ||
3730 | VM_BUG_ON(PageActive(page)); | |
3731 | ClearPageUnevictable(page); | |
fa9add64 HD |
3732 | del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE); |
3733 | add_page_to_lru_list(page, lruvec, lru); | |
24513264 | 3734 | pgrescued++; |
89e004ea | 3735 | } |
24513264 | 3736 | } |
89e004ea | 3737 | |
24513264 HD |
3738 | if (zone) { |
3739 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
3740 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); | |
3741 | spin_unlock_irq(&zone->lru_lock); | |
89e004ea | 3742 | } |
89e004ea | 3743 | } |
85046579 | 3744 | #endif /* CONFIG_SHMEM */ |
af936a16 | 3745 | |
264e56d8 | 3746 | static void warn_scan_unevictable_pages(void) |
af936a16 | 3747 | { |
264e56d8 | 3748 | printk_once(KERN_WARNING |
25bd91bd | 3749 | "%s: The scan_unevictable_pages sysctl/node-interface has been " |
264e56d8 | 3750 | "disabled for lack of a legitimate use case. If you have " |
25bd91bd KM |
3751 | "one, please send an email to linux-mm@kvack.org.\n", |
3752 | current->comm); | |
af936a16 LS |
3753 | } |
3754 | ||
3755 | /* | |
3756 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of | |
3757 | * all nodes' unevictable lists for evictable pages | |
3758 | */ | |
3759 | unsigned long scan_unevictable_pages; | |
3760 | ||
3761 | int scan_unevictable_handler(struct ctl_table *table, int write, | |
8d65af78 | 3762 | void __user *buffer, |
af936a16 LS |
3763 | size_t *length, loff_t *ppos) |
3764 | { | |
264e56d8 | 3765 | warn_scan_unevictable_pages(); |
8d65af78 | 3766 | proc_doulongvec_minmax(table, write, buffer, length, ppos); |
af936a16 LS |
3767 | scan_unevictable_pages = 0; |
3768 | return 0; | |
3769 | } | |
3770 | ||
e4455abb | 3771 | #ifdef CONFIG_NUMA |
af936a16 LS |
3772 | /* |
3773 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of | |
3774 | * a specified node's per zone unevictable lists for evictable pages. | |
3775 | */ | |
3776 | ||
10fbcf4c KS |
3777 | static ssize_t read_scan_unevictable_node(struct device *dev, |
3778 | struct device_attribute *attr, | |
af936a16 LS |
3779 | char *buf) |
3780 | { | |
264e56d8 | 3781 | warn_scan_unevictable_pages(); |
af936a16 LS |
3782 | return sprintf(buf, "0\n"); /* always zero; should fit... */ |
3783 | } | |
3784 | ||
10fbcf4c KS |
3785 | static ssize_t write_scan_unevictable_node(struct device *dev, |
3786 | struct device_attribute *attr, | |
af936a16 LS |
3787 | const char *buf, size_t count) |
3788 | { | |
264e56d8 | 3789 | warn_scan_unevictable_pages(); |
af936a16 LS |
3790 | return 1; |
3791 | } | |
3792 | ||
3793 | ||
10fbcf4c | 3794 | static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, |
af936a16 LS |
3795 | read_scan_unevictable_node, |
3796 | write_scan_unevictable_node); | |
3797 | ||
3798 | int scan_unevictable_register_node(struct node *node) | |
3799 | { | |
10fbcf4c | 3800 | return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages); |
af936a16 LS |
3801 | } |
3802 | ||
3803 | void scan_unevictable_unregister_node(struct node *node) | |
3804 | { | |
10fbcf4c | 3805 | device_remove_file(&node->dev, &dev_attr_scan_unevictable_pages); |
af936a16 | 3806 | } |
e4455abb | 3807 | #endif |