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 | ||
b1de0d13 MH |
14 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
15 | ||
1da177e4 LT |
16 | #include <linux/mm.h> |
17 | #include <linux/module.h> | |
5a0e3ad6 | 18 | #include <linux/gfp.h> |
1da177e4 LT |
19 | #include <linux/kernel_stat.h> |
20 | #include <linux/swap.h> | |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/highmem.h> | |
70ddf637 | 24 | #include <linux/vmpressure.h> |
e129b5c2 | 25 | #include <linux/vmstat.h> |
1da177e4 LT |
26 | #include <linux/file.h> |
27 | #include <linux/writeback.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/buffer_head.h> /* for try_to_release_page(), | |
30 | buffer_heads_over_limit */ | |
31 | #include <linux/mm_inline.h> | |
1da177e4 LT |
32 | #include <linux/backing-dev.h> |
33 | #include <linux/rmap.h> | |
34 | #include <linux/topology.h> | |
35 | #include <linux/cpu.h> | |
36 | #include <linux/cpuset.h> | |
3e7d3449 | 37 | #include <linux/compaction.h> |
1da177e4 LT |
38 | #include <linux/notifier.h> |
39 | #include <linux/rwsem.h> | |
248a0301 | 40 | #include <linux/delay.h> |
3218ae14 | 41 | #include <linux/kthread.h> |
7dfb7103 | 42 | #include <linux/freezer.h> |
66e1707b | 43 | #include <linux/memcontrol.h> |
873b4771 | 44 | #include <linux/delayacct.h> |
af936a16 | 45 | #include <linux/sysctl.h> |
929bea7c | 46 | #include <linux/oom.h> |
268bb0ce | 47 | #include <linux/prefetch.h> |
b1de0d13 | 48 | #include <linux/printk.h> |
1da177e4 LT |
49 | |
50 | #include <asm/tlbflush.h> | |
51 | #include <asm/div64.h> | |
52 | ||
53 | #include <linux/swapops.h> | |
117aad1e | 54 | #include <linux/balloon_compaction.h> |
1da177e4 | 55 | |
0f8053a5 NP |
56 | #include "internal.h" |
57 | ||
33906bc5 MG |
58 | #define CREATE_TRACE_POINTS |
59 | #include <trace/events/vmscan.h> | |
60 | ||
1da177e4 | 61 | struct scan_control { |
22fba335 KM |
62 | /* How many pages shrink_list() should reclaim */ |
63 | unsigned long nr_to_reclaim; | |
64 | ||
1da177e4 | 65 | /* This context's GFP mask */ |
6daa0e28 | 66 | gfp_t gfp_mask; |
1da177e4 | 67 | |
ee814fe2 | 68 | /* Allocation order */ |
5ad333eb | 69 | int order; |
66e1707b | 70 | |
ee814fe2 JW |
71 | /* |
72 | * Nodemask of nodes allowed by the caller. If NULL, all nodes | |
73 | * are scanned. | |
74 | */ | |
75 | nodemask_t *nodemask; | |
9e3b2f8c | 76 | |
f16015fb JW |
77 | /* |
78 | * The memory cgroup that hit its limit and as a result is the | |
79 | * primary target of this reclaim invocation. | |
80 | */ | |
81 | struct mem_cgroup *target_mem_cgroup; | |
66e1707b | 82 | |
ee814fe2 JW |
83 | /* Scan (total_size >> priority) pages at once */ |
84 | int priority; | |
85 | ||
86 | unsigned int may_writepage:1; | |
87 | ||
88 | /* Can mapped pages be reclaimed? */ | |
89 | unsigned int may_unmap:1; | |
90 | ||
91 | /* Can pages be swapped as part of reclaim? */ | |
92 | unsigned int may_swap:1; | |
93 | ||
94 | unsigned int hibernation_mode:1; | |
95 | ||
96 | /* One of the zones is ready for compaction */ | |
97 | unsigned int compaction_ready:1; | |
98 | ||
99 | /* Incremented by the number of inactive pages that were scanned */ | |
100 | unsigned long nr_scanned; | |
101 | ||
102 | /* Number of pages freed so far during a call to shrink_zones() */ | |
103 | unsigned long nr_reclaimed; | |
1da177e4 LT |
104 | }; |
105 | ||
1da177e4 LT |
106 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
107 | ||
108 | #ifdef ARCH_HAS_PREFETCH | |
109 | #define prefetch_prev_lru_page(_page, _base, _field) \ | |
110 | do { \ | |
111 | if ((_page)->lru.prev != _base) { \ | |
112 | struct page *prev; \ | |
113 | \ | |
114 | prev = lru_to_page(&(_page->lru)); \ | |
115 | prefetch(&prev->_field); \ | |
116 | } \ | |
117 | } while (0) | |
118 | #else | |
119 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) | |
120 | #endif | |
121 | ||
122 | #ifdef ARCH_HAS_PREFETCHW | |
123 | #define prefetchw_prev_lru_page(_page, _base, _field) \ | |
124 | do { \ | |
125 | if ((_page)->lru.prev != _base) { \ | |
126 | struct page *prev; \ | |
127 | \ | |
128 | prev = lru_to_page(&(_page->lru)); \ | |
129 | prefetchw(&prev->_field); \ | |
130 | } \ | |
131 | } while (0) | |
132 | #else | |
133 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) | |
134 | #endif | |
135 | ||
136 | /* | |
137 | * From 0 .. 100. Higher means more swappy. | |
138 | */ | |
139 | int vm_swappiness = 60; | |
d0480be4 WSH |
140 | /* |
141 | * The total number of pages which are beyond the high watermark within all | |
142 | * zones. | |
143 | */ | |
144 | unsigned long vm_total_pages; | |
1da177e4 LT |
145 | |
146 | static LIST_HEAD(shrinker_list); | |
147 | static DECLARE_RWSEM(shrinker_rwsem); | |
148 | ||
c255a458 | 149 | #ifdef CONFIG_MEMCG |
89b5fae5 JW |
150 | static bool global_reclaim(struct scan_control *sc) |
151 | { | |
f16015fb | 152 | return !sc->target_mem_cgroup; |
89b5fae5 | 153 | } |
91a45470 | 154 | #else |
89b5fae5 JW |
155 | static bool global_reclaim(struct scan_control *sc) |
156 | { | |
157 | return true; | |
158 | } | |
91a45470 KH |
159 | #endif |
160 | ||
a1c3bfb2 | 161 | static unsigned long zone_reclaimable_pages(struct zone *zone) |
6e543d57 LD |
162 | { |
163 | int nr; | |
164 | ||
165 | nr = zone_page_state(zone, NR_ACTIVE_FILE) + | |
166 | zone_page_state(zone, NR_INACTIVE_FILE); | |
167 | ||
168 | if (get_nr_swap_pages() > 0) | |
169 | nr += zone_page_state(zone, NR_ACTIVE_ANON) + | |
170 | zone_page_state(zone, NR_INACTIVE_ANON); | |
171 | ||
172 | return nr; | |
173 | } | |
174 | ||
175 | bool zone_reclaimable(struct zone *zone) | |
176 | { | |
0d5d823a MG |
177 | return zone_page_state(zone, NR_PAGES_SCANNED) < |
178 | zone_reclaimable_pages(zone) * 6; | |
6e543d57 LD |
179 | } |
180 | ||
4d7dcca2 | 181 | static unsigned long get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
c9f299d9 | 182 | { |
c3c787e8 | 183 | if (!mem_cgroup_disabled()) |
4d7dcca2 | 184 | return mem_cgroup_get_lru_size(lruvec, lru); |
a3d8e054 | 185 | |
074291fe | 186 | return zone_page_state(lruvec_zone(lruvec), NR_LRU_BASE + lru); |
c9f299d9 KM |
187 | } |
188 | ||
1da177e4 | 189 | /* |
1d3d4437 | 190 | * Add a shrinker callback to be called from the vm. |
1da177e4 | 191 | */ |
1d3d4437 | 192 | int register_shrinker(struct shrinker *shrinker) |
1da177e4 | 193 | { |
1d3d4437 GC |
194 | size_t size = sizeof(*shrinker->nr_deferred); |
195 | ||
196 | /* | |
197 | * If we only have one possible node in the system anyway, save | |
198 | * ourselves the trouble and disable NUMA aware behavior. This way we | |
199 | * will save memory and some small loop time later. | |
200 | */ | |
201 | if (nr_node_ids == 1) | |
202 | shrinker->flags &= ~SHRINKER_NUMA_AWARE; | |
203 | ||
204 | if (shrinker->flags & SHRINKER_NUMA_AWARE) | |
205 | size *= nr_node_ids; | |
206 | ||
207 | shrinker->nr_deferred = kzalloc(size, GFP_KERNEL); | |
208 | if (!shrinker->nr_deferred) | |
209 | return -ENOMEM; | |
210 | ||
8e1f936b RR |
211 | down_write(&shrinker_rwsem); |
212 | list_add_tail(&shrinker->list, &shrinker_list); | |
213 | up_write(&shrinker_rwsem); | |
1d3d4437 | 214 | return 0; |
1da177e4 | 215 | } |
8e1f936b | 216 | EXPORT_SYMBOL(register_shrinker); |
1da177e4 LT |
217 | |
218 | /* | |
219 | * Remove one | |
220 | */ | |
8e1f936b | 221 | void unregister_shrinker(struct shrinker *shrinker) |
1da177e4 LT |
222 | { |
223 | down_write(&shrinker_rwsem); | |
224 | list_del(&shrinker->list); | |
225 | up_write(&shrinker_rwsem); | |
ae393321 | 226 | kfree(shrinker->nr_deferred); |
1da177e4 | 227 | } |
8e1f936b | 228 | EXPORT_SYMBOL(unregister_shrinker); |
1da177e4 LT |
229 | |
230 | #define SHRINK_BATCH 128 | |
1d3d4437 GC |
231 | |
232 | static unsigned long | |
233 | shrink_slab_node(struct shrink_control *shrinkctl, struct shrinker *shrinker, | |
234 | unsigned long nr_pages_scanned, unsigned long lru_pages) | |
235 | { | |
236 | unsigned long freed = 0; | |
237 | unsigned long long delta; | |
238 | long total_scan; | |
d5bc5fd3 | 239 | long freeable; |
1d3d4437 GC |
240 | long nr; |
241 | long new_nr; | |
242 | int nid = shrinkctl->nid; | |
243 | long batch_size = shrinker->batch ? shrinker->batch | |
244 | : SHRINK_BATCH; | |
245 | ||
d5bc5fd3 VD |
246 | freeable = shrinker->count_objects(shrinker, shrinkctl); |
247 | if (freeable == 0) | |
1d3d4437 GC |
248 | return 0; |
249 | ||
250 | /* | |
251 | * copy the current shrinker scan count into a local variable | |
252 | * and zero it so that other concurrent shrinker invocations | |
253 | * don't also do this scanning work. | |
254 | */ | |
255 | nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0); | |
256 | ||
257 | total_scan = nr; | |
258 | delta = (4 * nr_pages_scanned) / shrinker->seeks; | |
d5bc5fd3 | 259 | delta *= freeable; |
1d3d4437 GC |
260 | do_div(delta, lru_pages + 1); |
261 | total_scan += delta; | |
262 | if (total_scan < 0) { | |
263 | printk(KERN_ERR | |
264 | "shrink_slab: %pF negative objects to delete nr=%ld\n", | |
a0b02131 | 265 | shrinker->scan_objects, total_scan); |
d5bc5fd3 | 266 | total_scan = freeable; |
1d3d4437 GC |
267 | } |
268 | ||
269 | /* | |
270 | * We need to avoid excessive windup on filesystem shrinkers | |
271 | * due to large numbers of GFP_NOFS allocations causing the | |
272 | * shrinkers to return -1 all the time. This results in a large | |
273 | * nr being built up so when a shrink that can do some work | |
274 | * comes along it empties the entire cache due to nr >>> | |
d5bc5fd3 | 275 | * freeable. This is bad for sustaining a working set in |
1d3d4437 GC |
276 | * memory. |
277 | * | |
278 | * Hence only allow the shrinker to scan the entire cache when | |
279 | * a large delta change is calculated directly. | |
280 | */ | |
d5bc5fd3 VD |
281 | if (delta < freeable / 4) |
282 | total_scan = min(total_scan, freeable / 2); | |
1d3d4437 GC |
283 | |
284 | /* | |
285 | * Avoid risking looping forever due to too large nr value: | |
286 | * never try to free more than twice the estimate number of | |
287 | * freeable entries. | |
288 | */ | |
d5bc5fd3 VD |
289 | if (total_scan > freeable * 2) |
290 | total_scan = freeable * 2; | |
1d3d4437 GC |
291 | |
292 | trace_mm_shrink_slab_start(shrinker, shrinkctl, nr, | |
293 | nr_pages_scanned, lru_pages, | |
d5bc5fd3 | 294 | freeable, delta, total_scan); |
1d3d4437 | 295 | |
0b1fb40a VD |
296 | /* |
297 | * Normally, we should not scan less than batch_size objects in one | |
298 | * pass to avoid too frequent shrinker calls, but if the slab has less | |
299 | * than batch_size objects in total and we are really tight on memory, | |
300 | * we will try to reclaim all available objects, otherwise we can end | |
301 | * up failing allocations although there are plenty of reclaimable | |
302 | * objects spread over several slabs with usage less than the | |
303 | * batch_size. | |
304 | * | |
305 | * We detect the "tight on memory" situations by looking at the total | |
306 | * number of objects we want to scan (total_scan). If it is greater | |
d5bc5fd3 | 307 | * than the total number of objects on slab (freeable), we must be |
0b1fb40a VD |
308 | * scanning at high prio and therefore should try to reclaim as much as |
309 | * possible. | |
310 | */ | |
311 | while (total_scan >= batch_size || | |
d5bc5fd3 | 312 | total_scan >= freeable) { |
a0b02131 | 313 | unsigned long ret; |
0b1fb40a | 314 | unsigned long nr_to_scan = min(batch_size, total_scan); |
1d3d4437 | 315 | |
0b1fb40a | 316 | shrinkctl->nr_to_scan = nr_to_scan; |
a0b02131 DC |
317 | ret = shrinker->scan_objects(shrinker, shrinkctl); |
318 | if (ret == SHRINK_STOP) | |
319 | break; | |
320 | freed += ret; | |
1d3d4437 | 321 | |
0b1fb40a VD |
322 | count_vm_events(SLABS_SCANNED, nr_to_scan); |
323 | total_scan -= nr_to_scan; | |
1d3d4437 GC |
324 | |
325 | cond_resched(); | |
326 | } | |
327 | ||
328 | /* | |
329 | * move the unused scan count back into the shrinker in a | |
330 | * manner that handles concurrent updates. If we exhausted the | |
331 | * scan, there is no need to do an update. | |
332 | */ | |
333 | if (total_scan > 0) | |
334 | new_nr = atomic_long_add_return(total_scan, | |
335 | &shrinker->nr_deferred[nid]); | |
336 | else | |
337 | new_nr = atomic_long_read(&shrinker->nr_deferred[nid]); | |
338 | ||
df9024a8 | 339 | trace_mm_shrink_slab_end(shrinker, nid, freed, nr, new_nr, total_scan); |
1d3d4437 | 340 | return freed; |
1495f230 YH |
341 | } |
342 | ||
1da177e4 LT |
343 | /* |
344 | * Call the shrink functions to age shrinkable caches | |
345 | * | |
346 | * Here we assume it costs one seek to replace a lru page and that it also | |
347 | * takes a seek to recreate a cache object. With this in mind we age equal | |
348 | * percentages of the lru and ageable caches. This should balance the seeks | |
349 | * generated by these structures. | |
350 | * | |
183ff22b | 351 | * If the vm encountered mapped pages on the LRU it increase the pressure on |
1da177e4 LT |
352 | * slab to avoid swapping. |
353 | * | |
354 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. | |
355 | * | |
356 | * `lru_pages' represents the number of on-LRU pages in all the zones which | |
357 | * are eligible for the caller's allocation attempt. It is used for balancing | |
358 | * slab reclaim versus page reclaim. | |
b15e0905 | 359 | * |
360 | * Returns the number of slab objects which we shrunk. | |
1da177e4 | 361 | */ |
24f7c6b9 | 362 | unsigned long shrink_slab(struct shrink_control *shrinkctl, |
1495f230 | 363 | unsigned long nr_pages_scanned, |
a09ed5e0 | 364 | unsigned long lru_pages) |
1da177e4 LT |
365 | { |
366 | struct shrinker *shrinker; | |
24f7c6b9 | 367 | unsigned long freed = 0; |
1da177e4 | 368 | |
1495f230 YH |
369 | if (nr_pages_scanned == 0) |
370 | nr_pages_scanned = SWAP_CLUSTER_MAX; | |
1da177e4 | 371 | |
f06590bd | 372 | if (!down_read_trylock(&shrinker_rwsem)) { |
24f7c6b9 DC |
373 | /* |
374 | * If we would return 0, our callers would understand that we | |
375 | * have nothing else to shrink and give up trying. By returning | |
376 | * 1 we keep it going and assume we'll be able to shrink next | |
377 | * time. | |
378 | */ | |
379 | freed = 1; | |
f06590bd MK |
380 | goto out; |
381 | } | |
1da177e4 LT |
382 | |
383 | list_for_each_entry(shrinker, &shrinker_list, list) { | |
ec97097b VD |
384 | if (!(shrinker->flags & SHRINKER_NUMA_AWARE)) { |
385 | shrinkctl->nid = 0; | |
1d3d4437 | 386 | freed += shrink_slab_node(shrinkctl, shrinker, |
ec97097b VD |
387 | nr_pages_scanned, lru_pages); |
388 | continue; | |
389 | } | |
390 | ||
391 | for_each_node_mask(shrinkctl->nid, shrinkctl->nodes_to_scan) { | |
392 | if (node_online(shrinkctl->nid)) | |
393 | freed += shrink_slab_node(shrinkctl, shrinker, | |
394 | nr_pages_scanned, lru_pages); | |
1da177e4 | 395 | |
1da177e4 | 396 | } |
1da177e4 LT |
397 | } |
398 | up_read(&shrinker_rwsem); | |
f06590bd MK |
399 | out: |
400 | cond_resched(); | |
24f7c6b9 | 401 | return freed; |
1da177e4 LT |
402 | } |
403 | ||
1da177e4 LT |
404 | static inline int is_page_cache_freeable(struct page *page) |
405 | { | |
ceddc3a5 JW |
406 | /* |
407 | * A freeable page cache page is referenced only by the caller | |
408 | * that isolated the page, the page cache radix tree and | |
409 | * optional buffer heads at page->private. | |
410 | */ | |
edcf4748 | 411 | return page_count(page) - page_has_private(page) == 2; |
1da177e4 LT |
412 | } |
413 | ||
7d3579e8 KM |
414 | static int may_write_to_queue(struct backing_dev_info *bdi, |
415 | struct scan_control *sc) | |
1da177e4 | 416 | { |
930d9152 | 417 | if (current->flags & PF_SWAPWRITE) |
1da177e4 LT |
418 | return 1; |
419 | if (!bdi_write_congested(bdi)) | |
420 | return 1; | |
421 | if (bdi == current->backing_dev_info) | |
422 | return 1; | |
423 | return 0; | |
424 | } | |
425 | ||
426 | /* | |
427 | * We detected a synchronous write error writing a page out. Probably | |
428 | * -ENOSPC. We need to propagate that into the address_space for a subsequent | |
429 | * fsync(), msync() or close(). | |
430 | * | |
431 | * The tricky part is that after writepage we cannot touch the mapping: nothing | |
432 | * prevents it from being freed up. But we have a ref on the page and once | |
433 | * that page is locked, the mapping is pinned. | |
434 | * | |
435 | * We're allowed to run sleeping lock_page() here because we know the caller has | |
436 | * __GFP_FS. | |
437 | */ | |
438 | static void handle_write_error(struct address_space *mapping, | |
439 | struct page *page, int error) | |
440 | { | |
7eaceacc | 441 | lock_page(page); |
3e9f45bd GC |
442 | if (page_mapping(page) == mapping) |
443 | mapping_set_error(mapping, error); | |
1da177e4 LT |
444 | unlock_page(page); |
445 | } | |
446 | ||
04e62a29 CL |
447 | /* possible outcome of pageout() */ |
448 | typedef enum { | |
449 | /* failed to write page out, page is locked */ | |
450 | PAGE_KEEP, | |
451 | /* move page to the active list, page is locked */ | |
452 | PAGE_ACTIVATE, | |
453 | /* page has been sent to the disk successfully, page is unlocked */ | |
454 | PAGE_SUCCESS, | |
455 | /* page is clean and locked */ | |
456 | PAGE_CLEAN, | |
457 | } pageout_t; | |
458 | ||
1da177e4 | 459 | /* |
1742f19f AM |
460 | * pageout is called by shrink_page_list() for each dirty page. |
461 | * Calls ->writepage(). | |
1da177e4 | 462 | */ |
c661b078 | 463 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
7d3579e8 | 464 | struct scan_control *sc) |
1da177e4 LT |
465 | { |
466 | /* | |
467 | * If the page is dirty, only perform writeback if that write | |
468 | * will be non-blocking. To prevent this allocation from being | |
469 | * stalled by pagecache activity. But note that there may be | |
470 | * stalls if we need to run get_block(). We could test | |
471 | * PagePrivate for that. | |
472 | * | |
8174202b | 473 | * If this process is currently in __generic_file_write_iter() against |
1da177e4 LT |
474 | * this page's queue, we can perform writeback even if that |
475 | * will block. | |
476 | * | |
477 | * If the page is swapcache, write it back even if that would | |
478 | * block, for some throttling. This happens by accident, because | |
479 | * swap_backing_dev_info is bust: it doesn't reflect the | |
480 | * congestion state of the swapdevs. Easy to fix, if needed. | |
1da177e4 LT |
481 | */ |
482 | if (!is_page_cache_freeable(page)) | |
483 | return PAGE_KEEP; | |
484 | if (!mapping) { | |
485 | /* | |
486 | * Some data journaling orphaned pages can have | |
487 | * page->mapping == NULL while being dirty with clean buffers. | |
488 | */ | |
266cf658 | 489 | if (page_has_private(page)) { |
1da177e4 LT |
490 | if (try_to_free_buffers(page)) { |
491 | ClearPageDirty(page); | |
b1de0d13 | 492 | pr_info("%s: orphaned page\n", __func__); |
1da177e4 LT |
493 | return PAGE_CLEAN; |
494 | } | |
495 | } | |
496 | return PAGE_KEEP; | |
497 | } | |
498 | if (mapping->a_ops->writepage == NULL) | |
499 | return PAGE_ACTIVATE; | |
0e093d99 | 500 | if (!may_write_to_queue(mapping->backing_dev_info, sc)) |
1da177e4 LT |
501 | return PAGE_KEEP; |
502 | ||
503 | if (clear_page_dirty_for_io(page)) { | |
504 | int res; | |
505 | struct writeback_control wbc = { | |
506 | .sync_mode = WB_SYNC_NONE, | |
507 | .nr_to_write = SWAP_CLUSTER_MAX, | |
111ebb6e OH |
508 | .range_start = 0, |
509 | .range_end = LLONG_MAX, | |
1da177e4 LT |
510 | .for_reclaim = 1, |
511 | }; | |
512 | ||
513 | SetPageReclaim(page); | |
514 | res = mapping->a_ops->writepage(page, &wbc); | |
515 | if (res < 0) | |
516 | handle_write_error(mapping, page, res); | |
994fc28c | 517 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
1da177e4 LT |
518 | ClearPageReclaim(page); |
519 | return PAGE_ACTIVATE; | |
520 | } | |
c661b078 | 521 | |
1da177e4 LT |
522 | if (!PageWriteback(page)) { |
523 | /* synchronous write or broken a_ops? */ | |
524 | ClearPageReclaim(page); | |
525 | } | |
23b9da55 | 526 | trace_mm_vmscan_writepage(page, trace_reclaim_flags(page)); |
e129b5c2 | 527 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
1da177e4 LT |
528 | return PAGE_SUCCESS; |
529 | } | |
530 | ||
531 | return PAGE_CLEAN; | |
532 | } | |
533 | ||
a649fd92 | 534 | /* |
e286781d NP |
535 | * Same as remove_mapping, but if the page is removed from the mapping, it |
536 | * gets returned with a refcount of 0. | |
a649fd92 | 537 | */ |
a528910e JW |
538 | static int __remove_mapping(struct address_space *mapping, struct page *page, |
539 | bool reclaimed) | |
49d2e9cc | 540 | { |
28e4d965 NP |
541 | BUG_ON(!PageLocked(page)); |
542 | BUG_ON(mapping != page_mapping(page)); | |
49d2e9cc | 543 | |
19fd6231 | 544 | spin_lock_irq(&mapping->tree_lock); |
49d2e9cc | 545 | /* |
0fd0e6b0 NP |
546 | * The non racy check for a busy page. |
547 | * | |
548 | * Must be careful with the order of the tests. When someone has | |
549 | * a ref to the page, it may be possible that they dirty it then | |
550 | * drop the reference. So if PageDirty is tested before page_count | |
551 | * here, then the following race may occur: | |
552 | * | |
553 | * get_user_pages(&page); | |
554 | * [user mapping goes away] | |
555 | * write_to(page); | |
556 | * !PageDirty(page) [good] | |
557 | * SetPageDirty(page); | |
558 | * put_page(page); | |
559 | * !page_count(page) [good, discard it] | |
560 | * | |
561 | * [oops, our write_to data is lost] | |
562 | * | |
563 | * Reversing the order of the tests ensures such a situation cannot | |
564 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags | |
565 | * load is not satisfied before that of page->_count. | |
566 | * | |
567 | * Note that if SetPageDirty is always performed via set_page_dirty, | |
568 | * and thus under tree_lock, then this ordering is not required. | |
49d2e9cc | 569 | */ |
e286781d | 570 | if (!page_freeze_refs(page, 2)) |
49d2e9cc | 571 | goto cannot_free; |
e286781d NP |
572 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
573 | if (unlikely(PageDirty(page))) { | |
574 | page_unfreeze_refs(page, 2); | |
49d2e9cc | 575 | goto cannot_free; |
e286781d | 576 | } |
49d2e9cc CL |
577 | |
578 | if (PageSwapCache(page)) { | |
579 | swp_entry_t swap = { .val = page_private(page) }; | |
0a31bc97 | 580 | mem_cgroup_swapout(page, swap); |
49d2e9cc | 581 | __delete_from_swap_cache(page); |
19fd6231 | 582 | spin_unlock_irq(&mapping->tree_lock); |
0a31bc97 | 583 | swapcache_free(swap); |
e286781d | 584 | } else { |
6072d13c | 585 | void (*freepage)(struct page *); |
a528910e | 586 | void *shadow = NULL; |
6072d13c LT |
587 | |
588 | freepage = mapping->a_ops->freepage; | |
a528910e JW |
589 | /* |
590 | * Remember a shadow entry for reclaimed file cache in | |
591 | * order to detect refaults, thus thrashing, later on. | |
592 | * | |
593 | * But don't store shadows in an address space that is | |
594 | * already exiting. This is not just an optizimation, | |
595 | * inode reclaim needs to empty out the radix tree or | |
596 | * the nodes are lost. Don't plant shadows behind its | |
597 | * back. | |
598 | */ | |
599 | if (reclaimed && page_is_file_cache(page) && | |
600 | !mapping_exiting(mapping)) | |
601 | shadow = workingset_eviction(mapping, page); | |
602 | __delete_from_page_cache(page, shadow); | |
19fd6231 | 603 | spin_unlock_irq(&mapping->tree_lock); |
6072d13c LT |
604 | |
605 | if (freepage != NULL) | |
606 | freepage(page); | |
49d2e9cc CL |
607 | } |
608 | ||
49d2e9cc CL |
609 | return 1; |
610 | ||
611 | cannot_free: | |
19fd6231 | 612 | spin_unlock_irq(&mapping->tree_lock); |
49d2e9cc CL |
613 | return 0; |
614 | } | |
615 | ||
e286781d NP |
616 | /* |
617 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if | |
618 | * someone else has a ref on the page, abort and return 0. If it was | |
619 | * successfully detached, return 1. Assumes the caller has a single ref on | |
620 | * this page. | |
621 | */ | |
622 | int remove_mapping(struct address_space *mapping, struct page *page) | |
623 | { | |
a528910e | 624 | if (__remove_mapping(mapping, page, false)) { |
e286781d NP |
625 | /* |
626 | * Unfreezing the refcount with 1 rather than 2 effectively | |
627 | * drops the pagecache ref for us without requiring another | |
628 | * atomic operation. | |
629 | */ | |
630 | page_unfreeze_refs(page, 1); | |
631 | return 1; | |
632 | } | |
633 | return 0; | |
634 | } | |
635 | ||
894bc310 LS |
636 | /** |
637 | * putback_lru_page - put previously isolated page onto appropriate LRU list | |
638 | * @page: page to be put back to appropriate lru list | |
639 | * | |
640 | * Add previously isolated @page to appropriate LRU list. | |
641 | * Page may still be unevictable for other reasons. | |
642 | * | |
643 | * lru_lock must not be held, interrupts must be enabled. | |
644 | */ | |
894bc310 LS |
645 | void putback_lru_page(struct page *page) |
646 | { | |
0ec3b74c | 647 | bool is_unevictable; |
bbfd28ee | 648 | int was_unevictable = PageUnevictable(page); |
894bc310 | 649 | |
309381fe | 650 | VM_BUG_ON_PAGE(PageLRU(page), page); |
894bc310 LS |
651 | |
652 | redo: | |
653 | ClearPageUnevictable(page); | |
654 | ||
39b5f29a | 655 | if (page_evictable(page)) { |
894bc310 LS |
656 | /* |
657 | * For evictable pages, we can use the cache. | |
658 | * In event of a race, worst case is we end up with an | |
659 | * unevictable page on [in]active list. | |
660 | * We know how to handle that. | |
661 | */ | |
0ec3b74c | 662 | is_unevictable = false; |
c53954a0 | 663 | lru_cache_add(page); |
894bc310 LS |
664 | } else { |
665 | /* | |
666 | * Put unevictable pages directly on zone's unevictable | |
667 | * list. | |
668 | */ | |
0ec3b74c | 669 | is_unevictable = true; |
894bc310 | 670 | add_page_to_unevictable_list(page); |
6a7b9548 | 671 | /* |
21ee9f39 MK |
672 | * When racing with an mlock or AS_UNEVICTABLE clearing |
673 | * (page is unlocked) make sure that if the other thread | |
674 | * does not observe our setting of PG_lru and fails | |
24513264 | 675 | * isolation/check_move_unevictable_pages, |
21ee9f39 | 676 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
6a7b9548 JW |
677 | * the page back to the evictable list. |
678 | * | |
21ee9f39 | 679 | * The other side is TestClearPageMlocked() or shmem_lock(). |
6a7b9548 JW |
680 | */ |
681 | smp_mb(); | |
894bc310 | 682 | } |
894bc310 LS |
683 | |
684 | /* | |
685 | * page's status can change while we move it among lru. If an evictable | |
686 | * page is on unevictable list, it never be freed. To avoid that, | |
687 | * check after we added it to the list, again. | |
688 | */ | |
0ec3b74c | 689 | if (is_unevictable && page_evictable(page)) { |
894bc310 LS |
690 | if (!isolate_lru_page(page)) { |
691 | put_page(page); | |
692 | goto redo; | |
693 | } | |
694 | /* This means someone else dropped this page from LRU | |
695 | * So, it will be freed or putback to LRU again. There is | |
696 | * nothing to do here. | |
697 | */ | |
698 | } | |
699 | ||
0ec3b74c | 700 | if (was_unevictable && !is_unevictable) |
bbfd28ee | 701 | count_vm_event(UNEVICTABLE_PGRESCUED); |
0ec3b74c | 702 | else if (!was_unevictable && is_unevictable) |
bbfd28ee LS |
703 | count_vm_event(UNEVICTABLE_PGCULLED); |
704 | ||
894bc310 LS |
705 | put_page(page); /* drop ref from isolate */ |
706 | } | |
707 | ||
dfc8d636 JW |
708 | enum page_references { |
709 | PAGEREF_RECLAIM, | |
710 | PAGEREF_RECLAIM_CLEAN, | |
64574746 | 711 | PAGEREF_KEEP, |
dfc8d636 JW |
712 | PAGEREF_ACTIVATE, |
713 | }; | |
714 | ||
715 | static enum page_references page_check_references(struct page *page, | |
716 | struct scan_control *sc) | |
717 | { | |
64574746 | 718 | int referenced_ptes, referenced_page; |
dfc8d636 | 719 | unsigned long vm_flags; |
dfc8d636 | 720 | |
c3ac9a8a JW |
721 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, |
722 | &vm_flags); | |
64574746 | 723 | referenced_page = TestClearPageReferenced(page); |
dfc8d636 | 724 | |
dfc8d636 JW |
725 | /* |
726 | * Mlock lost the isolation race with us. Let try_to_unmap() | |
727 | * move the page to the unevictable list. | |
728 | */ | |
729 | if (vm_flags & VM_LOCKED) | |
730 | return PAGEREF_RECLAIM; | |
731 | ||
64574746 | 732 | if (referenced_ptes) { |
e4898273 | 733 | if (PageSwapBacked(page)) |
64574746 JW |
734 | return PAGEREF_ACTIVATE; |
735 | /* | |
736 | * All mapped pages start out with page table | |
737 | * references from the instantiating fault, so we need | |
738 | * to look twice if a mapped file page is used more | |
739 | * than once. | |
740 | * | |
741 | * Mark it and spare it for another trip around the | |
742 | * inactive list. Another page table reference will | |
743 | * lead to its activation. | |
744 | * | |
745 | * Note: the mark is set for activated pages as well | |
746 | * so that recently deactivated but used pages are | |
747 | * quickly recovered. | |
748 | */ | |
749 | SetPageReferenced(page); | |
750 | ||
34dbc67a | 751 | if (referenced_page || referenced_ptes > 1) |
64574746 JW |
752 | return PAGEREF_ACTIVATE; |
753 | ||
c909e993 KK |
754 | /* |
755 | * Activate file-backed executable pages after first usage. | |
756 | */ | |
757 | if (vm_flags & VM_EXEC) | |
758 | return PAGEREF_ACTIVATE; | |
759 | ||
64574746 JW |
760 | return PAGEREF_KEEP; |
761 | } | |
dfc8d636 JW |
762 | |
763 | /* Reclaim if clean, defer dirty pages to writeback */ | |
2e30244a | 764 | if (referenced_page && !PageSwapBacked(page)) |
64574746 JW |
765 | return PAGEREF_RECLAIM_CLEAN; |
766 | ||
767 | return PAGEREF_RECLAIM; | |
dfc8d636 JW |
768 | } |
769 | ||
e2be15f6 MG |
770 | /* Check if a page is dirty or under writeback */ |
771 | static void page_check_dirty_writeback(struct page *page, | |
772 | bool *dirty, bool *writeback) | |
773 | { | |
b4597226 MG |
774 | struct address_space *mapping; |
775 | ||
e2be15f6 MG |
776 | /* |
777 | * Anonymous pages are not handled by flushers and must be written | |
778 | * from reclaim context. Do not stall reclaim based on them | |
779 | */ | |
780 | if (!page_is_file_cache(page)) { | |
781 | *dirty = false; | |
782 | *writeback = false; | |
783 | return; | |
784 | } | |
785 | ||
786 | /* By default assume that the page flags are accurate */ | |
787 | *dirty = PageDirty(page); | |
788 | *writeback = PageWriteback(page); | |
b4597226 MG |
789 | |
790 | /* Verify dirty/writeback state if the filesystem supports it */ | |
791 | if (!page_has_private(page)) | |
792 | return; | |
793 | ||
794 | mapping = page_mapping(page); | |
795 | if (mapping && mapping->a_ops->is_dirty_writeback) | |
796 | mapping->a_ops->is_dirty_writeback(page, dirty, writeback); | |
e2be15f6 MG |
797 | } |
798 | ||
1da177e4 | 799 | /* |
1742f19f | 800 | * shrink_page_list() returns the number of reclaimed pages |
1da177e4 | 801 | */ |
1742f19f | 802 | static unsigned long shrink_page_list(struct list_head *page_list, |
6a18adb3 | 803 | struct zone *zone, |
f84f6e2b | 804 | struct scan_control *sc, |
02c6de8d | 805 | enum ttu_flags ttu_flags, |
8e950282 | 806 | unsigned long *ret_nr_dirty, |
d43006d5 | 807 | unsigned long *ret_nr_unqueued_dirty, |
8e950282 | 808 | unsigned long *ret_nr_congested, |
02c6de8d | 809 | unsigned long *ret_nr_writeback, |
b1a6f21e | 810 | unsigned long *ret_nr_immediate, |
02c6de8d | 811 | bool force_reclaim) |
1da177e4 LT |
812 | { |
813 | LIST_HEAD(ret_pages); | |
abe4c3b5 | 814 | LIST_HEAD(free_pages); |
1da177e4 | 815 | int pgactivate = 0; |
d43006d5 | 816 | unsigned long nr_unqueued_dirty = 0; |
0e093d99 MG |
817 | unsigned long nr_dirty = 0; |
818 | unsigned long nr_congested = 0; | |
05ff5137 | 819 | unsigned long nr_reclaimed = 0; |
92df3a72 | 820 | unsigned long nr_writeback = 0; |
b1a6f21e | 821 | unsigned long nr_immediate = 0; |
1da177e4 LT |
822 | |
823 | cond_resched(); | |
824 | ||
69980e31 | 825 | mem_cgroup_uncharge_start(); |
1da177e4 LT |
826 | while (!list_empty(page_list)) { |
827 | struct address_space *mapping; | |
828 | struct page *page; | |
829 | int may_enter_fs; | |
02c6de8d | 830 | enum page_references references = PAGEREF_RECLAIM_CLEAN; |
e2be15f6 | 831 | bool dirty, writeback; |
1da177e4 LT |
832 | |
833 | cond_resched(); | |
834 | ||
835 | page = lru_to_page(page_list); | |
836 | list_del(&page->lru); | |
837 | ||
529ae9aa | 838 | if (!trylock_page(page)) |
1da177e4 LT |
839 | goto keep; |
840 | ||
309381fe SL |
841 | VM_BUG_ON_PAGE(PageActive(page), page); |
842 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
1da177e4 LT |
843 | |
844 | sc->nr_scanned++; | |
80e43426 | 845 | |
39b5f29a | 846 | if (unlikely(!page_evictable(page))) |
b291f000 | 847 | goto cull_mlocked; |
894bc310 | 848 | |
a6dc60f8 | 849 | if (!sc->may_unmap && page_mapped(page)) |
80e43426 CL |
850 | goto keep_locked; |
851 | ||
1da177e4 LT |
852 | /* Double the slab pressure for mapped and swapcache pages */ |
853 | if (page_mapped(page) || PageSwapCache(page)) | |
854 | sc->nr_scanned++; | |
855 | ||
c661b078 AW |
856 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
857 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); | |
858 | ||
e2be15f6 MG |
859 | /* |
860 | * The number of dirty pages determines if a zone is marked | |
861 | * reclaim_congested which affects wait_iff_congested. kswapd | |
862 | * will stall and start writing pages if the tail of the LRU | |
863 | * is all dirty unqueued pages. | |
864 | */ | |
865 | page_check_dirty_writeback(page, &dirty, &writeback); | |
866 | if (dirty || writeback) | |
867 | nr_dirty++; | |
868 | ||
869 | if (dirty && !writeback) | |
870 | nr_unqueued_dirty++; | |
871 | ||
d04e8acd MG |
872 | /* |
873 | * Treat this page as congested if the underlying BDI is or if | |
874 | * pages are cycling through the LRU so quickly that the | |
875 | * pages marked for immediate reclaim are making it to the | |
876 | * end of the LRU a second time. | |
877 | */ | |
e2be15f6 | 878 | mapping = page_mapping(page); |
d04e8acd MG |
879 | if ((mapping && bdi_write_congested(mapping->backing_dev_info)) || |
880 | (writeback && PageReclaim(page))) | |
e2be15f6 MG |
881 | nr_congested++; |
882 | ||
283aba9f MG |
883 | /* |
884 | * If a page at the tail of the LRU is under writeback, there | |
885 | * are three cases to consider. | |
886 | * | |
887 | * 1) If reclaim is encountering an excessive number of pages | |
888 | * under writeback and this page is both under writeback and | |
889 | * PageReclaim then it indicates that pages are being queued | |
890 | * for IO but are being recycled through the LRU before the | |
891 | * IO can complete. Waiting on the page itself risks an | |
892 | * indefinite stall if it is impossible to writeback the | |
893 | * page due to IO error or disconnected storage so instead | |
b1a6f21e MG |
894 | * note that the LRU is being scanned too quickly and the |
895 | * caller can stall after page list has been processed. | |
283aba9f MG |
896 | * |
897 | * 2) Global reclaim encounters a page, memcg encounters a | |
898 | * page that is not marked for immediate reclaim or | |
899 | * the caller does not have __GFP_IO. In this case mark | |
900 | * the page for immediate reclaim and continue scanning. | |
901 | * | |
902 | * __GFP_IO is checked because a loop driver thread might | |
903 | * enter reclaim, and deadlock if it waits on a page for | |
904 | * which it is needed to do the write (loop masks off | |
905 | * __GFP_IO|__GFP_FS for this reason); but more thought | |
906 | * would probably show more reasons. | |
907 | * | |
908 | * Don't require __GFP_FS, since we're not going into the | |
909 | * FS, just waiting on its writeback completion. Worryingly, | |
910 | * ext4 gfs2 and xfs allocate pages with | |
911 | * grab_cache_page_write_begin(,,AOP_FLAG_NOFS), so testing | |
912 | * may_enter_fs here is liable to OOM on them. | |
913 | * | |
914 | * 3) memcg encounters a page that is not already marked | |
915 | * PageReclaim. memcg does not have any dirty pages | |
916 | * throttling so we could easily OOM just because too many | |
917 | * pages are in writeback and there is nothing else to | |
918 | * reclaim. Wait for the writeback to complete. | |
919 | */ | |
c661b078 | 920 | if (PageWriteback(page)) { |
283aba9f MG |
921 | /* Case 1 above */ |
922 | if (current_is_kswapd() && | |
923 | PageReclaim(page) && | |
924 | zone_is_reclaim_writeback(zone)) { | |
b1a6f21e MG |
925 | nr_immediate++; |
926 | goto keep_locked; | |
283aba9f MG |
927 | |
928 | /* Case 2 above */ | |
929 | } else if (global_reclaim(sc) || | |
c3b94f44 HD |
930 | !PageReclaim(page) || !(sc->gfp_mask & __GFP_IO)) { |
931 | /* | |
932 | * This is slightly racy - end_page_writeback() | |
933 | * might have just cleared PageReclaim, then | |
934 | * setting PageReclaim here end up interpreted | |
935 | * as PageReadahead - but that does not matter | |
936 | * enough to care. What we do want is for this | |
937 | * page to have PageReclaim set next time memcg | |
938 | * reclaim reaches the tests above, so it will | |
939 | * then wait_on_page_writeback() to avoid OOM; | |
940 | * and it's also appropriate in global reclaim. | |
941 | */ | |
942 | SetPageReclaim(page); | |
e62e384e | 943 | nr_writeback++; |
283aba9f | 944 | |
c3b94f44 | 945 | goto keep_locked; |
283aba9f MG |
946 | |
947 | /* Case 3 above */ | |
948 | } else { | |
949 | wait_on_page_writeback(page); | |
e62e384e | 950 | } |
c661b078 | 951 | } |
1da177e4 | 952 | |
02c6de8d MK |
953 | if (!force_reclaim) |
954 | references = page_check_references(page, sc); | |
955 | ||
dfc8d636 JW |
956 | switch (references) { |
957 | case PAGEREF_ACTIVATE: | |
1da177e4 | 958 | goto activate_locked; |
64574746 JW |
959 | case PAGEREF_KEEP: |
960 | goto keep_locked; | |
dfc8d636 JW |
961 | case PAGEREF_RECLAIM: |
962 | case PAGEREF_RECLAIM_CLEAN: | |
963 | ; /* try to reclaim the page below */ | |
964 | } | |
1da177e4 | 965 | |
1da177e4 LT |
966 | /* |
967 | * Anonymous process memory has backing store? | |
968 | * Try to allocate it some swap space here. | |
969 | */ | |
b291f000 | 970 | if (PageAnon(page) && !PageSwapCache(page)) { |
63eb6b93 HD |
971 | if (!(sc->gfp_mask & __GFP_IO)) |
972 | goto keep_locked; | |
5bc7b8ac | 973 | if (!add_to_swap(page, page_list)) |
1da177e4 | 974 | goto activate_locked; |
63eb6b93 | 975 | may_enter_fs = 1; |
1da177e4 | 976 | |
e2be15f6 MG |
977 | /* Adding to swap updated mapping */ |
978 | mapping = page_mapping(page); | |
979 | } | |
1da177e4 LT |
980 | |
981 | /* | |
982 | * The page is mapped into the page tables of one or more | |
983 | * processes. Try to unmap it here. | |
984 | */ | |
985 | if (page_mapped(page) && mapping) { | |
02c6de8d | 986 | switch (try_to_unmap(page, ttu_flags)) { |
1da177e4 LT |
987 | case SWAP_FAIL: |
988 | goto activate_locked; | |
989 | case SWAP_AGAIN: | |
990 | goto keep_locked; | |
b291f000 NP |
991 | case SWAP_MLOCK: |
992 | goto cull_mlocked; | |
1da177e4 LT |
993 | case SWAP_SUCCESS: |
994 | ; /* try to free the page below */ | |
995 | } | |
996 | } | |
997 | ||
998 | if (PageDirty(page)) { | |
ee72886d MG |
999 | /* |
1000 | * Only kswapd can writeback filesystem pages to | |
d43006d5 MG |
1001 | * avoid risk of stack overflow but only writeback |
1002 | * if many dirty pages have been encountered. | |
ee72886d | 1003 | */ |
f84f6e2b | 1004 | if (page_is_file_cache(page) && |
9e3b2f8c | 1005 | (!current_is_kswapd() || |
d43006d5 | 1006 | !zone_is_reclaim_dirty(zone))) { |
49ea7eb6 MG |
1007 | /* |
1008 | * Immediately reclaim when written back. | |
1009 | * Similar in principal to deactivate_page() | |
1010 | * except we already have the page isolated | |
1011 | * and know it's dirty | |
1012 | */ | |
1013 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); | |
1014 | SetPageReclaim(page); | |
1015 | ||
ee72886d MG |
1016 | goto keep_locked; |
1017 | } | |
1018 | ||
dfc8d636 | 1019 | if (references == PAGEREF_RECLAIM_CLEAN) |
1da177e4 | 1020 | goto keep_locked; |
4dd4b920 | 1021 | if (!may_enter_fs) |
1da177e4 | 1022 | goto keep_locked; |
52a8363e | 1023 | if (!sc->may_writepage) |
1da177e4 LT |
1024 | goto keep_locked; |
1025 | ||
1026 | /* Page is dirty, try to write it out here */ | |
7d3579e8 | 1027 | switch (pageout(page, mapping, sc)) { |
1da177e4 LT |
1028 | case PAGE_KEEP: |
1029 | goto keep_locked; | |
1030 | case PAGE_ACTIVATE: | |
1031 | goto activate_locked; | |
1032 | case PAGE_SUCCESS: | |
7d3579e8 | 1033 | if (PageWriteback(page)) |
41ac1999 | 1034 | goto keep; |
7d3579e8 | 1035 | if (PageDirty(page)) |
1da177e4 | 1036 | goto keep; |
7d3579e8 | 1037 | |
1da177e4 LT |
1038 | /* |
1039 | * A synchronous write - probably a ramdisk. Go | |
1040 | * ahead and try to reclaim the page. | |
1041 | */ | |
529ae9aa | 1042 | if (!trylock_page(page)) |
1da177e4 LT |
1043 | goto keep; |
1044 | if (PageDirty(page) || PageWriteback(page)) | |
1045 | goto keep_locked; | |
1046 | mapping = page_mapping(page); | |
1047 | case PAGE_CLEAN: | |
1048 | ; /* try to free the page below */ | |
1049 | } | |
1050 | } | |
1051 | ||
1052 | /* | |
1053 | * If the page has buffers, try to free the buffer mappings | |
1054 | * associated with this page. If we succeed we try to free | |
1055 | * the page as well. | |
1056 | * | |
1057 | * We do this even if the page is PageDirty(). | |
1058 | * try_to_release_page() does not perform I/O, but it is | |
1059 | * possible for a page to have PageDirty set, but it is actually | |
1060 | * clean (all its buffers are clean). This happens if the | |
1061 | * buffers were written out directly, with submit_bh(). ext3 | |
894bc310 | 1062 | * will do this, as well as the blockdev mapping. |
1da177e4 LT |
1063 | * try_to_release_page() will discover that cleanness and will |
1064 | * drop the buffers and mark the page clean - it can be freed. | |
1065 | * | |
1066 | * Rarely, pages can have buffers and no ->mapping. These are | |
1067 | * the pages which were not successfully invalidated in | |
1068 | * truncate_complete_page(). We try to drop those buffers here | |
1069 | * and if that worked, and the page is no longer mapped into | |
1070 | * process address space (page_count == 1) it can be freed. | |
1071 | * Otherwise, leave the page on the LRU so it is swappable. | |
1072 | */ | |
266cf658 | 1073 | if (page_has_private(page)) { |
1da177e4 LT |
1074 | if (!try_to_release_page(page, sc->gfp_mask)) |
1075 | goto activate_locked; | |
e286781d NP |
1076 | if (!mapping && page_count(page) == 1) { |
1077 | unlock_page(page); | |
1078 | if (put_page_testzero(page)) | |
1079 | goto free_it; | |
1080 | else { | |
1081 | /* | |
1082 | * rare race with speculative reference. | |
1083 | * the speculative reference will free | |
1084 | * this page shortly, so we may | |
1085 | * increment nr_reclaimed here (and | |
1086 | * leave it off the LRU). | |
1087 | */ | |
1088 | nr_reclaimed++; | |
1089 | continue; | |
1090 | } | |
1091 | } | |
1da177e4 LT |
1092 | } |
1093 | ||
a528910e | 1094 | if (!mapping || !__remove_mapping(mapping, page, true)) |
49d2e9cc | 1095 | goto keep_locked; |
1da177e4 | 1096 | |
a978d6f5 NP |
1097 | /* |
1098 | * At this point, we have no other references and there is | |
1099 | * no way to pick any more up (removed from LRU, removed | |
1100 | * from pagecache). Can use non-atomic bitops now (and | |
1101 | * we obviously don't have to worry about waking up a process | |
1102 | * waiting on the page lock, because there are no references. | |
1103 | */ | |
1104 | __clear_page_locked(page); | |
e286781d | 1105 | free_it: |
0a31bc97 | 1106 | mem_cgroup_uncharge(page); |
05ff5137 | 1107 | nr_reclaimed++; |
abe4c3b5 MG |
1108 | |
1109 | /* | |
1110 | * Is there need to periodically free_page_list? It would | |
1111 | * appear not as the counts should be low | |
1112 | */ | |
1113 | list_add(&page->lru, &free_pages); | |
1da177e4 LT |
1114 | continue; |
1115 | ||
b291f000 | 1116 | cull_mlocked: |
63d6c5ad HD |
1117 | if (PageSwapCache(page)) |
1118 | try_to_free_swap(page); | |
b291f000 NP |
1119 | unlock_page(page); |
1120 | putback_lru_page(page); | |
1121 | continue; | |
1122 | ||
1da177e4 | 1123 | activate_locked: |
68a22394 RR |
1124 | /* Not a candidate for swapping, so reclaim swap space. */ |
1125 | if (PageSwapCache(page) && vm_swap_full()) | |
a2c43eed | 1126 | try_to_free_swap(page); |
309381fe | 1127 | VM_BUG_ON_PAGE(PageActive(page), page); |
1da177e4 LT |
1128 | SetPageActive(page); |
1129 | pgactivate++; | |
1130 | keep_locked: | |
1131 | unlock_page(page); | |
1132 | keep: | |
1133 | list_add(&page->lru, &ret_pages); | |
309381fe | 1134 | VM_BUG_ON_PAGE(PageLRU(page) || PageUnevictable(page), page); |
1da177e4 | 1135 | } |
0a31bc97 | 1136 | mem_cgroup_uncharge_end(); |
abe4c3b5 | 1137 | |
b745bc85 | 1138 | free_hot_cold_page_list(&free_pages, true); |
abe4c3b5 | 1139 | |
1da177e4 | 1140 | list_splice(&ret_pages, page_list); |
f8891e5e | 1141 | count_vm_events(PGACTIVATE, pgactivate); |
0a31bc97 | 1142 | |
8e950282 MG |
1143 | *ret_nr_dirty += nr_dirty; |
1144 | *ret_nr_congested += nr_congested; | |
d43006d5 | 1145 | *ret_nr_unqueued_dirty += nr_unqueued_dirty; |
92df3a72 | 1146 | *ret_nr_writeback += nr_writeback; |
b1a6f21e | 1147 | *ret_nr_immediate += nr_immediate; |
05ff5137 | 1148 | return nr_reclaimed; |
1da177e4 LT |
1149 | } |
1150 | ||
02c6de8d MK |
1151 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, |
1152 | struct list_head *page_list) | |
1153 | { | |
1154 | struct scan_control sc = { | |
1155 | .gfp_mask = GFP_KERNEL, | |
1156 | .priority = DEF_PRIORITY, | |
1157 | .may_unmap = 1, | |
1158 | }; | |
8e950282 | 1159 | unsigned long ret, dummy1, dummy2, dummy3, dummy4, dummy5; |
02c6de8d MK |
1160 | struct page *page, *next; |
1161 | LIST_HEAD(clean_pages); | |
1162 | ||
1163 | list_for_each_entry_safe(page, next, page_list, lru) { | |
117aad1e RA |
1164 | if (page_is_file_cache(page) && !PageDirty(page) && |
1165 | !isolated_balloon_page(page)) { | |
02c6de8d MK |
1166 | ClearPageActive(page); |
1167 | list_move(&page->lru, &clean_pages); | |
1168 | } | |
1169 | } | |
1170 | ||
1171 | ret = shrink_page_list(&clean_pages, zone, &sc, | |
8e950282 MG |
1172 | TTU_UNMAP|TTU_IGNORE_ACCESS, |
1173 | &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, true); | |
02c6de8d | 1174 | list_splice(&clean_pages, page_list); |
83da7510 | 1175 | mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret); |
02c6de8d MK |
1176 | return ret; |
1177 | } | |
1178 | ||
5ad333eb AW |
1179 | /* |
1180 | * Attempt to remove the specified page from its LRU. Only take this page | |
1181 | * if it is of the appropriate PageActive status. Pages which are being | |
1182 | * freed elsewhere are also ignored. | |
1183 | * | |
1184 | * page: page to consider | |
1185 | * mode: one of the LRU isolation modes defined above | |
1186 | * | |
1187 | * returns 0 on success, -ve errno on failure. | |
1188 | */ | |
f3fd4a61 | 1189 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) |
5ad333eb AW |
1190 | { |
1191 | int ret = -EINVAL; | |
1192 | ||
1193 | /* Only take pages on the LRU. */ | |
1194 | if (!PageLRU(page)) | |
1195 | return ret; | |
1196 | ||
e46a2879 MK |
1197 | /* Compaction should not handle unevictable pages but CMA can do so */ |
1198 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) | |
894bc310 LS |
1199 | return ret; |
1200 | ||
5ad333eb | 1201 | ret = -EBUSY; |
08e552c6 | 1202 | |
c8244935 MG |
1203 | /* |
1204 | * To minimise LRU disruption, the caller can indicate that it only | |
1205 | * wants to isolate pages it will be able to operate on without | |
1206 | * blocking - clean pages for the most part. | |
1207 | * | |
1208 | * ISOLATE_CLEAN means that only clean pages should be isolated. This | |
1209 | * is used by reclaim when it is cannot write to backing storage | |
1210 | * | |
1211 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages | |
1212 | * that it is possible to migrate without blocking | |
1213 | */ | |
1214 | if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) { | |
1215 | /* All the caller can do on PageWriteback is block */ | |
1216 | if (PageWriteback(page)) | |
1217 | return ret; | |
1218 | ||
1219 | if (PageDirty(page)) { | |
1220 | struct address_space *mapping; | |
1221 | ||
1222 | /* ISOLATE_CLEAN means only clean pages */ | |
1223 | if (mode & ISOLATE_CLEAN) | |
1224 | return ret; | |
1225 | ||
1226 | /* | |
1227 | * Only pages without mappings or that have a | |
1228 | * ->migratepage callback are possible to migrate | |
1229 | * without blocking | |
1230 | */ | |
1231 | mapping = page_mapping(page); | |
1232 | if (mapping && !mapping->a_ops->migratepage) | |
1233 | return ret; | |
1234 | } | |
1235 | } | |
39deaf85 | 1236 | |
f80c0673 MK |
1237 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
1238 | return ret; | |
1239 | ||
5ad333eb AW |
1240 | if (likely(get_page_unless_zero(page))) { |
1241 | /* | |
1242 | * Be careful not to clear PageLRU until after we're | |
1243 | * sure the page is not being freed elsewhere -- the | |
1244 | * page release code relies on it. | |
1245 | */ | |
1246 | ClearPageLRU(page); | |
1247 | ret = 0; | |
1248 | } | |
1249 | ||
1250 | return ret; | |
1251 | } | |
1252 | ||
1da177e4 LT |
1253 | /* |
1254 | * zone->lru_lock is heavily contended. Some of the functions that | |
1255 | * shrink the lists perform better by taking out a batch of pages | |
1256 | * and working on them outside the LRU lock. | |
1257 | * | |
1258 | * For pagecache intensive workloads, this function is the hottest | |
1259 | * spot in the kernel (apart from copy_*_user functions). | |
1260 | * | |
1261 | * Appropriate locks must be held before calling this function. | |
1262 | * | |
1263 | * @nr_to_scan: The number of pages to look through on the list. | |
5dc35979 | 1264 | * @lruvec: The LRU vector to pull pages from. |
1da177e4 | 1265 | * @dst: The temp list to put pages on to. |
f626012d | 1266 | * @nr_scanned: The number of pages that were scanned. |
fe2c2a10 | 1267 | * @sc: The scan_control struct for this reclaim session |
5ad333eb | 1268 | * @mode: One of the LRU isolation modes |
3cb99451 | 1269 | * @lru: LRU list id for isolating |
1da177e4 LT |
1270 | * |
1271 | * returns how many pages were moved onto *@dst. | |
1272 | */ | |
69e05944 | 1273 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
5dc35979 | 1274 | struct lruvec *lruvec, struct list_head *dst, |
fe2c2a10 | 1275 | unsigned long *nr_scanned, struct scan_control *sc, |
3cb99451 | 1276 | isolate_mode_t mode, enum lru_list lru) |
1da177e4 | 1277 | { |
75b00af7 | 1278 | struct list_head *src = &lruvec->lists[lru]; |
69e05944 | 1279 | unsigned long nr_taken = 0; |
c9b02d97 | 1280 | unsigned long scan; |
1da177e4 | 1281 | |
c9b02d97 | 1282 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
5ad333eb | 1283 | struct page *page; |
fa9add64 | 1284 | int nr_pages; |
5ad333eb | 1285 | |
1da177e4 LT |
1286 | page = lru_to_page(src); |
1287 | prefetchw_prev_lru_page(page, src, flags); | |
1288 | ||
309381fe | 1289 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
8d438f96 | 1290 | |
f3fd4a61 | 1291 | switch (__isolate_lru_page(page, mode)) { |
5ad333eb | 1292 | case 0: |
fa9add64 HD |
1293 | nr_pages = hpage_nr_pages(page); |
1294 | mem_cgroup_update_lru_size(lruvec, lru, -nr_pages); | |
5ad333eb | 1295 | list_move(&page->lru, dst); |
fa9add64 | 1296 | nr_taken += nr_pages; |
5ad333eb AW |
1297 | break; |
1298 | ||
1299 | case -EBUSY: | |
1300 | /* else it is being freed elsewhere */ | |
1301 | list_move(&page->lru, src); | |
1302 | continue; | |
46453a6e | 1303 | |
5ad333eb AW |
1304 | default: |
1305 | BUG(); | |
1306 | } | |
1da177e4 LT |
1307 | } |
1308 | ||
f626012d | 1309 | *nr_scanned = scan; |
75b00af7 HD |
1310 | trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan, |
1311 | nr_taken, mode, is_file_lru(lru)); | |
1da177e4 LT |
1312 | return nr_taken; |
1313 | } | |
1314 | ||
62695a84 NP |
1315 | /** |
1316 | * isolate_lru_page - tries to isolate a page from its LRU list | |
1317 | * @page: page to isolate from its LRU list | |
1318 | * | |
1319 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the | |
1320 | * vmstat statistic corresponding to whatever LRU list the page was on. | |
1321 | * | |
1322 | * Returns 0 if the page was removed from an LRU list. | |
1323 | * Returns -EBUSY if the page was not on an LRU list. | |
1324 | * | |
1325 | * The returned page will have PageLRU() cleared. If it was found on | |
894bc310 LS |
1326 | * the active list, it will have PageActive set. If it was found on |
1327 | * the unevictable list, it will have the PageUnevictable bit set. That flag | |
1328 | * may need to be cleared by the caller before letting the page go. | |
62695a84 NP |
1329 | * |
1330 | * The vmstat statistic corresponding to the list on which the page was | |
1331 | * found will be decremented. | |
1332 | * | |
1333 | * Restrictions: | |
1334 | * (1) Must be called with an elevated refcount on the page. This is a | |
1335 | * fundamentnal difference from isolate_lru_pages (which is called | |
1336 | * without a stable reference). | |
1337 | * (2) the lru_lock must not be held. | |
1338 | * (3) interrupts must be enabled. | |
1339 | */ | |
1340 | int isolate_lru_page(struct page *page) | |
1341 | { | |
1342 | int ret = -EBUSY; | |
1343 | ||
309381fe | 1344 | VM_BUG_ON_PAGE(!page_count(page), page); |
0c917313 | 1345 | |
62695a84 NP |
1346 | if (PageLRU(page)) { |
1347 | struct zone *zone = page_zone(page); | |
fa9add64 | 1348 | struct lruvec *lruvec; |
62695a84 NP |
1349 | |
1350 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 1351 | lruvec = mem_cgroup_page_lruvec(page, zone); |
0c917313 | 1352 | if (PageLRU(page)) { |
894bc310 | 1353 | int lru = page_lru(page); |
0c917313 | 1354 | get_page(page); |
62695a84 | 1355 | ClearPageLRU(page); |
fa9add64 HD |
1356 | del_page_from_lru_list(page, lruvec, lru); |
1357 | ret = 0; | |
62695a84 NP |
1358 | } |
1359 | spin_unlock_irq(&zone->lru_lock); | |
1360 | } | |
1361 | return ret; | |
1362 | } | |
1363 | ||
35cd7815 | 1364 | /* |
d37dd5dc FW |
1365 | * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and |
1366 | * then get resheduled. When there are massive number of tasks doing page | |
1367 | * allocation, such sleeping direct reclaimers may keep piling up on each CPU, | |
1368 | * the LRU list will go small and be scanned faster than necessary, leading to | |
1369 | * unnecessary swapping, thrashing and OOM. | |
35cd7815 RR |
1370 | */ |
1371 | static int too_many_isolated(struct zone *zone, int file, | |
1372 | struct scan_control *sc) | |
1373 | { | |
1374 | unsigned long inactive, isolated; | |
1375 | ||
1376 | if (current_is_kswapd()) | |
1377 | return 0; | |
1378 | ||
89b5fae5 | 1379 | if (!global_reclaim(sc)) |
35cd7815 RR |
1380 | return 0; |
1381 | ||
1382 | if (file) { | |
1383 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); | |
1384 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); | |
1385 | } else { | |
1386 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1387 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); | |
1388 | } | |
1389 | ||
3cf23841 FW |
1390 | /* |
1391 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they | |
1392 | * won't get blocked by normal direct-reclaimers, forming a circular | |
1393 | * deadlock. | |
1394 | */ | |
1395 | if ((sc->gfp_mask & GFP_IOFS) == GFP_IOFS) | |
1396 | inactive >>= 3; | |
1397 | ||
35cd7815 RR |
1398 | return isolated > inactive; |
1399 | } | |
1400 | ||
66635629 | 1401 | static noinline_for_stack void |
75b00af7 | 1402 | putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list) |
66635629 | 1403 | { |
27ac81d8 KK |
1404 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1405 | struct zone *zone = lruvec_zone(lruvec); | |
3f79768f | 1406 | LIST_HEAD(pages_to_free); |
66635629 | 1407 | |
66635629 MG |
1408 | /* |
1409 | * Put back any unfreeable pages. | |
1410 | */ | |
66635629 | 1411 | while (!list_empty(page_list)) { |
3f79768f | 1412 | struct page *page = lru_to_page(page_list); |
66635629 | 1413 | int lru; |
3f79768f | 1414 | |
309381fe | 1415 | VM_BUG_ON_PAGE(PageLRU(page), page); |
66635629 | 1416 | list_del(&page->lru); |
39b5f29a | 1417 | if (unlikely(!page_evictable(page))) { |
66635629 MG |
1418 | spin_unlock_irq(&zone->lru_lock); |
1419 | putback_lru_page(page); | |
1420 | spin_lock_irq(&zone->lru_lock); | |
1421 | continue; | |
1422 | } | |
fa9add64 HD |
1423 | |
1424 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
1425 | ||
7a608572 | 1426 | SetPageLRU(page); |
66635629 | 1427 | lru = page_lru(page); |
fa9add64 HD |
1428 | add_page_to_lru_list(page, lruvec, lru); |
1429 | ||
66635629 MG |
1430 | if (is_active_lru(lru)) { |
1431 | int file = is_file_lru(lru); | |
9992af10 RR |
1432 | int numpages = hpage_nr_pages(page); |
1433 | reclaim_stat->recent_rotated[file] += numpages; | |
66635629 | 1434 | } |
2bcf8879 HD |
1435 | if (put_page_testzero(page)) { |
1436 | __ClearPageLRU(page); | |
1437 | __ClearPageActive(page); | |
fa9add64 | 1438 | del_page_from_lru_list(page, lruvec, lru); |
2bcf8879 | 1439 | |
0a31bc97 JW |
1440 | mem_cgroup_uncharge(page); |
1441 | ||
2bcf8879 HD |
1442 | if (unlikely(PageCompound(page))) { |
1443 | spin_unlock_irq(&zone->lru_lock); | |
1444 | (*get_compound_page_dtor(page))(page); | |
1445 | spin_lock_irq(&zone->lru_lock); | |
1446 | } else | |
1447 | list_add(&page->lru, &pages_to_free); | |
66635629 MG |
1448 | } |
1449 | } | |
66635629 | 1450 | |
3f79768f HD |
1451 | /* |
1452 | * To save our caller's stack, now use input list for pages to free. | |
1453 | */ | |
1454 | list_splice(&pages_to_free, page_list); | |
66635629 MG |
1455 | } |
1456 | ||
399ba0b9 N |
1457 | /* |
1458 | * If a kernel thread (such as nfsd for loop-back mounts) services | |
1459 | * a backing device by writing to the page cache it sets PF_LESS_THROTTLE. | |
1460 | * In that case we should only throttle if the backing device it is | |
1461 | * writing to is congested. In other cases it is safe to throttle. | |
1462 | */ | |
1463 | static int current_may_throttle(void) | |
1464 | { | |
1465 | return !(current->flags & PF_LESS_THROTTLE) || | |
1466 | current->backing_dev_info == NULL || | |
1467 | bdi_write_congested(current->backing_dev_info); | |
1468 | } | |
1469 | ||
1da177e4 | 1470 | /* |
1742f19f AM |
1471 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
1472 | * of reclaimed pages | |
1da177e4 | 1473 | */ |
66635629 | 1474 | static noinline_for_stack unsigned long |
1a93be0e | 1475 | shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec, |
9e3b2f8c | 1476 | struct scan_control *sc, enum lru_list lru) |
1da177e4 LT |
1477 | { |
1478 | LIST_HEAD(page_list); | |
e247dbce | 1479 | unsigned long nr_scanned; |
05ff5137 | 1480 | unsigned long nr_reclaimed = 0; |
e247dbce | 1481 | unsigned long nr_taken; |
8e950282 MG |
1482 | unsigned long nr_dirty = 0; |
1483 | unsigned long nr_congested = 0; | |
e2be15f6 | 1484 | unsigned long nr_unqueued_dirty = 0; |
92df3a72 | 1485 | unsigned long nr_writeback = 0; |
b1a6f21e | 1486 | unsigned long nr_immediate = 0; |
f3fd4a61 | 1487 | isolate_mode_t isolate_mode = 0; |
3cb99451 | 1488 | int file = is_file_lru(lru); |
1a93be0e KK |
1489 | struct zone *zone = lruvec_zone(lruvec); |
1490 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; | |
78dc583d | 1491 | |
35cd7815 | 1492 | while (unlikely(too_many_isolated(zone, file, sc))) { |
58355c78 | 1493 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
35cd7815 RR |
1494 | |
1495 | /* We are about to die and free our memory. Return now. */ | |
1496 | if (fatal_signal_pending(current)) | |
1497 | return SWAP_CLUSTER_MAX; | |
1498 | } | |
1499 | ||
1da177e4 | 1500 | lru_add_drain(); |
f80c0673 MK |
1501 | |
1502 | if (!sc->may_unmap) | |
61317289 | 1503 | isolate_mode |= ISOLATE_UNMAPPED; |
f80c0673 | 1504 | if (!sc->may_writepage) |
61317289 | 1505 | isolate_mode |= ISOLATE_CLEAN; |
f80c0673 | 1506 | |
1da177e4 | 1507 | spin_lock_irq(&zone->lru_lock); |
b35ea17b | 1508 | |
5dc35979 KK |
1509 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, |
1510 | &nr_scanned, sc, isolate_mode, lru); | |
95d918fc KK |
1511 | |
1512 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); | |
1513 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); | |
1514 | ||
89b5fae5 | 1515 | if (global_reclaim(sc)) { |
0d5d823a | 1516 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, nr_scanned); |
e247dbce | 1517 | if (current_is_kswapd()) |
75b00af7 | 1518 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned); |
e247dbce | 1519 | else |
75b00af7 | 1520 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned); |
e247dbce | 1521 | } |
d563c050 | 1522 | spin_unlock_irq(&zone->lru_lock); |
b35ea17b | 1523 | |
d563c050 | 1524 | if (nr_taken == 0) |
66635629 | 1525 | return 0; |
5ad333eb | 1526 | |
02c6de8d | 1527 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP, |
8e950282 MG |
1528 | &nr_dirty, &nr_unqueued_dirty, &nr_congested, |
1529 | &nr_writeback, &nr_immediate, | |
1530 | false); | |
c661b078 | 1531 | |
3f79768f HD |
1532 | spin_lock_irq(&zone->lru_lock); |
1533 | ||
95d918fc | 1534 | reclaim_stat->recent_scanned[file] += nr_taken; |
d563c050 | 1535 | |
904249aa YH |
1536 | if (global_reclaim(sc)) { |
1537 | if (current_is_kswapd()) | |
1538 | __count_zone_vm_events(PGSTEAL_KSWAPD, zone, | |
1539 | nr_reclaimed); | |
1540 | else | |
1541 | __count_zone_vm_events(PGSTEAL_DIRECT, zone, | |
1542 | nr_reclaimed); | |
1543 | } | |
a74609fa | 1544 | |
27ac81d8 | 1545 | putback_inactive_pages(lruvec, &page_list); |
3f79768f | 1546 | |
95d918fc | 1547 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
3f79768f HD |
1548 | |
1549 | spin_unlock_irq(&zone->lru_lock); | |
1550 | ||
b745bc85 | 1551 | free_hot_cold_page_list(&page_list, true); |
e11da5b4 | 1552 | |
92df3a72 MG |
1553 | /* |
1554 | * If reclaim is isolating dirty pages under writeback, it implies | |
1555 | * that the long-lived page allocation rate is exceeding the page | |
1556 | * laundering rate. Either the global limits are not being effective | |
1557 | * at throttling processes due to the page distribution throughout | |
1558 | * zones or there is heavy usage of a slow backing device. The | |
1559 | * only option is to throttle from reclaim context which is not ideal | |
1560 | * as there is no guarantee the dirtying process is throttled in the | |
1561 | * same way balance_dirty_pages() manages. | |
1562 | * | |
8e950282 MG |
1563 | * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number |
1564 | * of pages under pages flagged for immediate reclaim and stall if any | |
1565 | * are encountered in the nr_immediate check below. | |
92df3a72 | 1566 | */ |
918fc718 | 1567 | if (nr_writeback && nr_writeback == nr_taken) |
283aba9f | 1568 | zone_set_flag(zone, ZONE_WRITEBACK); |
92df3a72 | 1569 | |
d43006d5 | 1570 | /* |
b1a6f21e MG |
1571 | * memcg will stall in page writeback so only consider forcibly |
1572 | * stalling for global reclaim | |
d43006d5 | 1573 | */ |
b1a6f21e | 1574 | if (global_reclaim(sc)) { |
8e950282 MG |
1575 | /* |
1576 | * Tag a zone as congested if all the dirty pages scanned were | |
1577 | * backed by a congested BDI and wait_iff_congested will stall. | |
1578 | */ | |
1579 | if (nr_dirty && nr_dirty == nr_congested) | |
1580 | zone_set_flag(zone, ZONE_CONGESTED); | |
1581 | ||
b1a6f21e MG |
1582 | /* |
1583 | * If dirty pages are scanned that are not queued for IO, it | |
1584 | * implies that flushers are not keeping up. In this case, flag | |
1585 | * the zone ZONE_TAIL_LRU_DIRTY and kswapd will start writing | |
b738d764 | 1586 | * pages from reclaim context. |
b1a6f21e MG |
1587 | */ |
1588 | if (nr_unqueued_dirty == nr_taken) | |
1589 | zone_set_flag(zone, ZONE_TAIL_LRU_DIRTY); | |
1590 | ||
1591 | /* | |
b738d764 LT |
1592 | * If kswapd scans pages marked marked for immediate |
1593 | * reclaim and under writeback (nr_immediate), it implies | |
1594 | * that pages are cycling through the LRU faster than | |
b1a6f21e MG |
1595 | * they are written so also forcibly stall. |
1596 | */ | |
b738d764 | 1597 | if (nr_immediate && current_may_throttle()) |
b1a6f21e | 1598 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
e2be15f6 | 1599 | } |
d43006d5 | 1600 | |
8e950282 MG |
1601 | /* |
1602 | * Stall direct reclaim for IO completions if underlying BDIs or zone | |
1603 | * is congested. Allow kswapd to continue until it starts encountering | |
1604 | * unqueued dirty pages or cycling through the LRU too quickly. | |
1605 | */ | |
399ba0b9 N |
1606 | if (!sc->hibernation_mode && !current_is_kswapd() && |
1607 | current_may_throttle()) | |
8e950282 MG |
1608 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); |
1609 | ||
e11da5b4 MG |
1610 | trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id, |
1611 | zone_idx(zone), | |
1612 | nr_scanned, nr_reclaimed, | |
9e3b2f8c | 1613 | sc->priority, |
23b9da55 | 1614 | trace_shrink_flags(file)); |
05ff5137 | 1615 | return nr_reclaimed; |
1da177e4 LT |
1616 | } |
1617 | ||
1618 | /* | |
1619 | * This moves pages from the active list to the inactive list. | |
1620 | * | |
1621 | * We move them the other way if the page is referenced by one or more | |
1622 | * processes, from rmap. | |
1623 | * | |
1624 | * If the pages are mostly unmapped, the processing is fast and it is | |
1625 | * appropriate to hold zone->lru_lock across the whole operation. But if | |
1626 | * the pages are mapped, the processing is slow (page_referenced()) so we | |
1627 | * should drop zone->lru_lock around each page. It's impossible to balance | |
1628 | * this, so instead we remove the pages from the LRU while processing them. | |
1629 | * It is safe to rely on PG_active against the non-LRU pages in here because | |
1630 | * nobody will play with that bit on a non-LRU page. | |
1631 | * | |
1632 | * The downside is that we have to touch page->_count against each page. | |
1633 | * But we had to alter page->flags anyway. | |
1634 | */ | |
1cfb419b | 1635 | |
fa9add64 | 1636 | static void move_active_pages_to_lru(struct lruvec *lruvec, |
3eb4140f | 1637 | struct list_head *list, |
2bcf8879 | 1638 | struct list_head *pages_to_free, |
3eb4140f WF |
1639 | enum lru_list lru) |
1640 | { | |
fa9add64 | 1641 | struct zone *zone = lruvec_zone(lruvec); |
3eb4140f | 1642 | unsigned long pgmoved = 0; |
3eb4140f | 1643 | struct page *page; |
fa9add64 | 1644 | int nr_pages; |
3eb4140f | 1645 | |
3eb4140f WF |
1646 | while (!list_empty(list)) { |
1647 | page = lru_to_page(list); | |
fa9add64 | 1648 | lruvec = mem_cgroup_page_lruvec(page, zone); |
3eb4140f | 1649 | |
309381fe | 1650 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3eb4140f WF |
1651 | SetPageLRU(page); |
1652 | ||
fa9add64 HD |
1653 | nr_pages = hpage_nr_pages(page); |
1654 | mem_cgroup_update_lru_size(lruvec, lru, nr_pages); | |
925b7673 | 1655 | list_move(&page->lru, &lruvec->lists[lru]); |
fa9add64 | 1656 | pgmoved += nr_pages; |
3eb4140f | 1657 | |
2bcf8879 HD |
1658 | if (put_page_testzero(page)) { |
1659 | __ClearPageLRU(page); | |
1660 | __ClearPageActive(page); | |
fa9add64 | 1661 | del_page_from_lru_list(page, lruvec, lru); |
2bcf8879 | 1662 | |
0a31bc97 JW |
1663 | mem_cgroup_uncharge(page); |
1664 | ||
2bcf8879 HD |
1665 | if (unlikely(PageCompound(page))) { |
1666 | spin_unlock_irq(&zone->lru_lock); | |
1667 | (*get_compound_page_dtor(page))(page); | |
1668 | spin_lock_irq(&zone->lru_lock); | |
1669 | } else | |
1670 | list_add(&page->lru, pages_to_free); | |
3eb4140f WF |
1671 | } |
1672 | } | |
1673 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); | |
1674 | if (!is_active_lru(lru)) | |
1675 | __count_vm_events(PGDEACTIVATE, pgmoved); | |
1676 | } | |
1cfb419b | 1677 | |
f626012d | 1678 | static void shrink_active_list(unsigned long nr_to_scan, |
1a93be0e | 1679 | struct lruvec *lruvec, |
f16015fb | 1680 | struct scan_control *sc, |
9e3b2f8c | 1681 | enum lru_list lru) |
1da177e4 | 1682 | { |
44c241f1 | 1683 | unsigned long nr_taken; |
f626012d | 1684 | unsigned long nr_scanned; |
6fe6b7e3 | 1685 | unsigned long vm_flags; |
1da177e4 | 1686 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
8cab4754 | 1687 | LIST_HEAD(l_active); |
b69408e8 | 1688 | LIST_HEAD(l_inactive); |
1da177e4 | 1689 | struct page *page; |
1a93be0e | 1690 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
44c241f1 | 1691 | unsigned long nr_rotated = 0; |
f3fd4a61 | 1692 | isolate_mode_t isolate_mode = 0; |
3cb99451 | 1693 | int file = is_file_lru(lru); |
1a93be0e | 1694 | struct zone *zone = lruvec_zone(lruvec); |
1da177e4 LT |
1695 | |
1696 | lru_add_drain(); | |
f80c0673 MK |
1697 | |
1698 | if (!sc->may_unmap) | |
61317289 | 1699 | isolate_mode |= ISOLATE_UNMAPPED; |
f80c0673 | 1700 | if (!sc->may_writepage) |
61317289 | 1701 | isolate_mode |= ISOLATE_CLEAN; |
f80c0673 | 1702 | |
1da177e4 | 1703 | spin_lock_irq(&zone->lru_lock); |
925b7673 | 1704 | |
5dc35979 KK |
1705 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, |
1706 | &nr_scanned, sc, isolate_mode, lru); | |
89b5fae5 | 1707 | if (global_reclaim(sc)) |
0d5d823a | 1708 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, nr_scanned); |
89b5fae5 | 1709 | |
b7c46d15 | 1710 | reclaim_stat->recent_scanned[file] += nr_taken; |
1cfb419b | 1711 | |
f626012d | 1712 | __count_zone_vm_events(PGREFILL, zone, nr_scanned); |
3cb99451 | 1713 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
a731286d | 1714 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
1da177e4 LT |
1715 | spin_unlock_irq(&zone->lru_lock); |
1716 | ||
1da177e4 LT |
1717 | while (!list_empty(&l_hold)) { |
1718 | cond_resched(); | |
1719 | page = lru_to_page(&l_hold); | |
1720 | list_del(&page->lru); | |
7e9cd484 | 1721 | |
39b5f29a | 1722 | if (unlikely(!page_evictable(page))) { |
894bc310 LS |
1723 | putback_lru_page(page); |
1724 | continue; | |
1725 | } | |
1726 | ||
cc715d99 MG |
1727 | if (unlikely(buffer_heads_over_limit)) { |
1728 | if (page_has_private(page) && trylock_page(page)) { | |
1729 | if (page_has_private(page)) | |
1730 | try_to_release_page(page, 0); | |
1731 | unlock_page(page); | |
1732 | } | |
1733 | } | |
1734 | ||
c3ac9a8a JW |
1735 | if (page_referenced(page, 0, sc->target_mem_cgroup, |
1736 | &vm_flags)) { | |
9992af10 | 1737 | nr_rotated += hpage_nr_pages(page); |
8cab4754 WF |
1738 | /* |
1739 | * Identify referenced, file-backed active pages and | |
1740 | * give them one more trip around the active list. So | |
1741 | * that executable code get better chances to stay in | |
1742 | * memory under moderate memory pressure. Anon pages | |
1743 | * are not likely to be evicted by use-once streaming | |
1744 | * IO, plus JVM can create lots of anon VM_EXEC pages, | |
1745 | * so we ignore them here. | |
1746 | */ | |
41e20983 | 1747 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
8cab4754 WF |
1748 | list_add(&page->lru, &l_active); |
1749 | continue; | |
1750 | } | |
1751 | } | |
7e9cd484 | 1752 | |
5205e56e | 1753 | ClearPageActive(page); /* we are de-activating */ |
1da177e4 LT |
1754 | list_add(&page->lru, &l_inactive); |
1755 | } | |
1756 | ||
b555749a | 1757 | /* |
8cab4754 | 1758 | * Move pages back to the lru list. |
b555749a | 1759 | */ |
2a1dc509 | 1760 | spin_lock_irq(&zone->lru_lock); |
556adecb | 1761 | /* |
8cab4754 WF |
1762 | * Count referenced pages from currently used mappings as rotated, |
1763 | * even though only some of them are actually re-activated. This | |
1764 | * helps balance scan pressure between file and anonymous pages in | |
7c0db9e9 | 1765 | * get_scan_count. |
7e9cd484 | 1766 | */ |
b7c46d15 | 1767 | reclaim_stat->recent_rotated[file] += nr_rotated; |
556adecb | 1768 | |
fa9add64 HD |
1769 | move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru); |
1770 | move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE); | |
a731286d | 1771 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
f8891e5e | 1772 | spin_unlock_irq(&zone->lru_lock); |
2bcf8879 | 1773 | |
b745bc85 | 1774 | free_hot_cold_page_list(&l_hold, true); |
1da177e4 LT |
1775 | } |
1776 | ||
74e3f3c3 | 1777 | #ifdef CONFIG_SWAP |
14797e23 | 1778 | static int inactive_anon_is_low_global(struct zone *zone) |
f89eb90e KM |
1779 | { |
1780 | unsigned long active, inactive; | |
1781 | ||
1782 | active = zone_page_state(zone, NR_ACTIVE_ANON); | |
1783 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1784 | ||
1785 | if (inactive * zone->inactive_ratio < active) | |
1786 | return 1; | |
1787 | ||
1788 | return 0; | |
1789 | } | |
1790 | ||
14797e23 KM |
1791 | /** |
1792 | * inactive_anon_is_low - check if anonymous pages need to be deactivated | |
c56d5c7d | 1793 | * @lruvec: LRU vector to check |
14797e23 KM |
1794 | * |
1795 | * Returns true if the zone does not have enough inactive anon pages, | |
1796 | * meaning some active anon pages need to be deactivated. | |
1797 | */ | |
c56d5c7d | 1798 | static int inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1799 | { |
74e3f3c3 MK |
1800 | /* |
1801 | * If we don't have swap space, anonymous page deactivation | |
1802 | * is pointless. | |
1803 | */ | |
1804 | if (!total_swap_pages) | |
1805 | return 0; | |
1806 | ||
c3c787e8 | 1807 | if (!mem_cgroup_disabled()) |
c56d5c7d | 1808 | return mem_cgroup_inactive_anon_is_low(lruvec); |
f16015fb | 1809 | |
c56d5c7d | 1810 | return inactive_anon_is_low_global(lruvec_zone(lruvec)); |
14797e23 | 1811 | } |
74e3f3c3 | 1812 | #else |
c56d5c7d | 1813 | static inline int inactive_anon_is_low(struct lruvec *lruvec) |
74e3f3c3 MK |
1814 | { |
1815 | return 0; | |
1816 | } | |
1817 | #endif | |
14797e23 | 1818 | |
56e49d21 RR |
1819 | /** |
1820 | * inactive_file_is_low - check if file pages need to be deactivated | |
c56d5c7d | 1821 | * @lruvec: LRU vector to check |
56e49d21 RR |
1822 | * |
1823 | * When the system is doing streaming IO, memory pressure here | |
1824 | * ensures that active file pages get deactivated, until more | |
1825 | * than half of the file pages are on the inactive list. | |
1826 | * | |
1827 | * Once we get to that situation, protect the system's working | |
1828 | * set from being evicted by disabling active file page aging. | |
1829 | * | |
1830 | * This uses a different ratio than the anonymous pages, because | |
1831 | * the page cache uses a use-once replacement algorithm. | |
1832 | */ | |
c56d5c7d | 1833 | static int inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 | 1834 | { |
e3790144 JW |
1835 | unsigned long inactive; |
1836 | unsigned long active; | |
1837 | ||
1838 | inactive = get_lru_size(lruvec, LRU_INACTIVE_FILE); | |
1839 | active = get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 | 1840 | |
e3790144 | 1841 | return active > inactive; |
56e49d21 RR |
1842 | } |
1843 | ||
75b00af7 | 1844 | static int inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru) |
b39415b2 | 1845 | { |
75b00af7 | 1846 | if (is_file_lru(lru)) |
c56d5c7d | 1847 | return inactive_file_is_low(lruvec); |
b39415b2 | 1848 | else |
c56d5c7d | 1849 | return inactive_anon_is_low(lruvec); |
b39415b2 RR |
1850 | } |
1851 | ||
4f98a2fe | 1852 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
1a93be0e | 1853 | struct lruvec *lruvec, struct scan_control *sc) |
b69408e8 | 1854 | { |
b39415b2 | 1855 | if (is_active_lru(lru)) { |
75b00af7 | 1856 | if (inactive_list_is_low(lruvec, lru)) |
1a93be0e | 1857 | shrink_active_list(nr_to_scan, lruvec, sc, lru); |
556adecb RR |
1858 | return 0; |
1859 | } | |
1860 | ||
1a93be0e | 1861 | return shrink_inactive_list(nr_to_scan, lruvec, sc, lru); |
4f98a2fe RR |
1862 | } |
1863 | ||
9a265114 JW |
1864 | enum scan_balance { |
1865 | SCAN_EQUAL, | |
1866 | SCAN_FRACT, | |
1867 | SCAN_ANON, | |
1868 | SCAN_FILE, | |
1869 | }; | |
1870 | ||
4f98a2fe RR |
1871 | /* |
1872 | * Determine how aggressively the anon and file LRU lists should be | |
1873 | * scanned. The relative value of each set of LRU lists is determined | |
1874 | * by looking at the fraction of the pages scanned we did rotate back | |
1875 | * onto the active list instead of evict. | |
1876 | * | |
be7bd59d WL |
1877 | * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan |
1878 | * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan | |
4f98a2fe | 1879 | */ |
02695175 JW |
1880 | static void get_scan_count(struct lruvec *lruvec, int swappiness, |
1881 | struct scan_control *sc, unsigned long *nr) | |
4f98a2fe | 1882 | { |
9a265114 JW |
1883 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1884 | u64 fraction[2]; | |
1885 | u64 denominator = 0; /* gcc */ | |
1886 | struct zone *zone = lruvec_zone(lruvec); | |
4f98a2fe | 1887 | unsigned long anon_prio, file_prio; |
9a265114 | 1888 | enum scan_balance scan_balance; |
0bf1457f | 1889 | unsigned long anon, file; |
9a265114 | 1890 | bool force_scan = false; |
4f98a2fe | 1891 | unsigned long ap, fp; |
4111304d | 1892 | enum lru_list lru; |
6f04f48d SS |
1893 | bool some_scanned; |
1894 | int pass; | |
246e87a9 | 1895 | |
f11c0ca5 JW |
1896 | /* |
1897 | * If the zone or memcg is small, nr[l] can be 0. This | |
1898 | * results in no scanning on this priority and a potential | |
1899 | * priority drop. Global direct reclaim can go to the next | |
1900 | * zone and tends to have no problems. Global kswapd is for | |
1901 | * zone balancing and it needs to scan a minimum amount. When | |
1902 | * reclaiming for a memcg, a priority drop can cause high | |
1903 | * latencies, so it's better to scan a minimum amount there as | |
1904 | * well. | |
1905 | */ | |
6e543d57 | 1906 | if (current_is_kswapd() && !zone_reclaimable(zone)) |
a4d3e9e7 | 1907 | force_scan = true; |
89b5fae5 | 1908 | if (!global_reclaim(sc)) |
a4d3e9e7 | 1909 | force_scan = true; |
76a33fc3 SL |
1910 | |
1911 | /* If we have no swap space, do not bother scanning anon pages. */ | |
ec8acf20 | 1912 | if (!sc->may_swap || (get_nr_swap_pages() <= 0)) { |
9a265114 | 1913 | scan_balance = SCAN_FILE; |
76a33fc3 SL |
1914 | goto out; |
1915 | } | |
4f98a2fe | 1916 | |
10316b31 JW |
1917 | /* |
1918 | * Global reclaim will swap to prevent OOM even with no | |
1919 | * swappiness, but memcg users want to use this knob to | |
1920 | * disable swapping for individual groups completely when | |
1921 | * using the memory controller's swap limit feature would be | |
1922 | * too expensive. | |
1923 | */ | |
02695175 | 1924 | if (!global_reclaim(sc) && !swappiness) { |
9a265114 | 1925 | scan_balance = SCAN_FILE; |
10316b31 JW |
1926 | goto out; |
1927 | } | |
1928 | ||
1929 | /* | |
1930 | * Do not apply any pressure balancing cleverness when the | |
1931 | * system is close to OOM, scan both anon and file equally | |
1932 | * (unless the swappiness setting disagrees with swapping). | |
1933 | */ | |
02695175 | 1934 | if (!sc->priority && swappiness) { |
9a265114 | 1935 | scan_balance = SCAN_EQUAL; |
10316b31 JW |
1936 | goto out; |
1937 | } | |
1938 | ||
62376251 JW |
1939 | /* |
1940 | * Prevent the reclaimer from falling into the cache trap: as | |
1941 | * cache pages start out inactive, every cache fault will tip | |
1942 | * the scan balance towards the file LRU. And as the file LRU | |
1943 | * shrinks, so does the window for rotation from references. | |
1944 | * This means we have a runaway feedback loop where a tiny | |
1945 | * thrashing file LRU becomes infinitely more attractive than | |
1946 | * anon pages. Try to detect this based on file LRU size. | |
1947 | */ | |
1948 | if (global_reclaim(sc)) { | |
2ab051e1 JM |
1949 | unsigned long zonefile; |
1950 | unsigned long zonefree; | |
1951 | ||
1952 | zonefree = zone_page_state(zone, NR_FREE_PAGES); | |
1953 | zonefile = zone_page_state(zone, NR_ACTIVE_FILE) + | |
1954 | zone_page_state(zone, NR_INACTIVE_FILE); | |
62376251 | 1955 | |
2ab051e1 | 1956 | if (unlikely(zonefile + zonefree <= high_wmark_pages(zone))) { |
62376251 JW |
1957 | scan_balance = SCAN_ANON; |
1958 | goto out; | |
1959 | } | |
1960 | } | |
1961 | ||
7c5bd705 JW |
1962 | /* |
1963 | * There is enough inactive page cache, do not reclaim | |
1964 | * anything from the anonymous working set right now. | |
1965 | */ | |
1966 | if (!inactive_file_is_low(lruvec)) { | |
9a265114 | 1967 | scan_balance = SCAN_FILE; |
7c5bd705 JW |
1968 | goto out; |
1969 | } | |
1970 | ||
9a265114 JW |
1971 | scan_balance = SCAN_FRACT; |
1972 | ||
58c37f6e KM |
1973 | /* |
1974 | * With swappiness at 100, anonymous and file have the same priority. | |
1975 | * This scanning priority is essentially the inverse of IO cost. | |
1976 | */ | |
02695175 | 1977 | anon_prio = swappiness; |
75b00af7 | 1978 | file_prio = 200 - anon_prio; |
58c37f6e | 1979 | |
4f98a2fe RR |
1980 | /* |
1981 | * OK, so we have swap space and a fair amount of page cache | |
1982 | * pages. We use the recently rotated / recently scanned | |
1983 | * ratios to determine how valuable each cache is. | |
1984 | * | |
1985 | * Because workloads change over time (and to avoid overflow) | |
1986 | * we keep these statistics as a floating average, which ends | |
1987 | * up weighing recent references more than old ones. | |
1988 | * | |
1989 | * anon in [0], file in [1] | |
1990 | */ | |
2ab051e1 JM |
1991 | |
1992 | anon = get_lru_size(lruvec, LRU_ACTIVE_ANON) + | |
1993 | get_lru_size(lruvec, LRU_INACTIVE_ANON); | |
1994 | file = get_lru_size(lruvec, LRU_ACTIVE_FILE) + | |
1995 | get_lru_size(lruvec, LRU_INACTIVE_FILE); | |
1996 | ||
90126375 | 1997 | spin_lock_irq(&zone->lru_lock); |
6e901571 | 1998 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
6e901571 KM |
1999 | reclaim_stat->recent_scanned[0] /= 2; |
2000 | reclaim_stat->recent_rotated[0] /= 2; | |
4f98a2fe RR |
2001 | } |
2002 | ||
6e901571 | 2003 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
6e901571 KM |
2004 | reclaim_stat->recent_scanned[1] /= 2; |
2005 | reclaim_stat->recent_rotated[1] /= 2; | |
4f98a2fe RR |
2006 | } |
2007 | ||
4f98a2fe | 2008 | /* |
00d8089c RR |
2009 | * The amount of pressure on anon vs file pages is inversely |
2010 | * proportional to the fraction of recently scanned pages on | |
2011 | * each list that were recently referenced and in active use. | |
4f98a2fe | 2012 | */ |
fe35004f | 2013 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); |
6e901571 | 2014 | ap /= reclaim_stat->recent_rotated[0] + 1; |
4f98a2fe | 2015 | |
fe35004f | 2016 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); |
6e901571 | 2017 | fp /= reclaim_stat->recent_rotated[1] + 1; |
90126375 | 2018 | spin_unlock_irq(&zone->lru_lock); |
4f98a2fe | 2019 | |
76a33fc3 SL |
2020 | fraction[0] = ap; |
2021 | fraction[1] = fp; | |
2022 | denominator = ap + fp + 1; | |
2023 | out: | |
6f04f48d SS |
2024 | some_scanned = false; |
2025 | /* Only use force_scan on second pass. */ | |
2026 | for (pass = 0; !some_scanned && pass < 2; pass++) { | |
2027 | for_each_evictable_lru(lru) { | |
2028 | int file = is_file_lru(lru); | |
2029 | unsigned long size; | |
2030 | unsigned long scan; | |
6e08a369 | 2031 | |
6f04f48d SS |
2032 | size = get_lru_size(lruvec, lru); |
2033 | scan = size >> sc->priority; | |
9a265114 | 2034 | |
6f04f48d SS |
2035 | if (!scan && pass && force_scan) |
2036 | scan = min(size, SWAP_CLUSTER_MAX); | |
9a265114 | 2037 | |
6f04f48d SS |
2038 | switch (scan_balance) { |
2039 | case SCAN_EQUAL: | |
2040 | /* Scan lists relative to size */ | |
2041 | break; | |
2042 | case SCAN_FRACT: | |
2043 | /* | |
2044 | * Scan types proportional to swappiness and | |
2045 | * their relative recent reclaim efficiency. | |
2046 | */ | |
2047 | scan = div64_u64(scan * fraction[file], | |
2048 | denominator); | |
2049 | break; | |
2050 | case SCAN_FILE: | |
2051 | case SCAN_ANON: | |
2052 | /* Scan one type exclusively */ | |
2053 | if ((scan_balance == SCAN_FILE) != file) | |
2054 | scan = 0; | |
2055 | break; | |
2056 | default: | |
2057 | /* Look ma, no brain */ | |
2058 | BUG(); | |
2059 | } | |
2060 | nr[lru] = scan; | |
9a265114 | 2061 | /* |
6f04f48d SS |
2062 | * Skip the second pass and don't force_scan, |
2063 | * if we found something to scan. | |
9a265114 | 2064 | */ |
6f04f48d | 2065 | some_scanned |= !!scan; |
9a265114 | 2066 | } |
76a33fc3 | 2067 | } |
6e08a369 | 2068 | } |
4f98a2fe | 2069 | |
9b4f98cd JW |
2070 | /* |
2071 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. | |
2072 | */ | |
02695175 JW |
2073 | static void shrink_lruvec(struct lruvec *lruvec, int swappiness, |
2074 | struct scan_control *sc) | |
9b4f98cd JW |
2075 | { |
2076 | unsigned long nr[NR_LRU_LISTS]; | |
e82e0561 | 2077 | unsigned long targets[NR_LRU_LISTS]; |
9b4f98cd JW |
2078 | unsigned long nr_to_scan; |
2079 | enum lru_list lru; | |
2080 | unsigned long nr_reclaimed = 0; | |
2081 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; | |
2082 | struct blk_plug plug; | |
1a501907 | 2083 | bool scan_adjusted; |
9b4f98cd | 2084 | |
02695175 | 2085 | get_scan_count(lruvec, swappiness, sc, nr); |
9b4f98cd | 2086 | |
e82e0561 MG |
2087 | /* Record the original scan target for proportional adjustments later */ |
2088 | memcpy(targets, nr, sizeof(nr)); | |
2089 | ||
1a501907 MG |
2090 | /* |
2091 | * Global reclaiming within direct reclaim at DEF_PRIORITY is a normal | |
2092 | * event that can occur when there is little memory pressure e.g. | |
2093 | * multiple streaming readers/writers. Hence, we do not abort scanning | |
2094 | * when the requested number of pages are reclaimed when scanning at | |
2095 | * DEF_PRIORITY on the assumption that the fact we are direct | |
2096 | * reclaiming implies that kswapd is not keeping up and it is best to | |
2097 | * do a batch of work at once. For memcg reclaim one check is made to | |
2098 | * abort proportional reclaim if either the file or anon lru has already | |
2099 | * dropped to zero at the first pass. | |
2100 | */ | |
2101 | scan_adjusted = (global_reclaim(sc) && !current_is_kswapd() && | |
2102 | sc->priority == DEF_PRIORITY); | |
2103 | ||
9b4f98cd JW |
2104 | blk_start_plug(&plug); |
2105 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || | |
2106 | nr[LRU_INACTIVE_FILE]) { | |
e82e0561 MG |
2107 | unsigned long nr_anon, nr_file, percentage; |
2108 | unsigned long nr_scanned; | |
2109 | ||
9b4f98cd JW |
2110 | for_each_evictable_lru(lru) { |
2111 | if (nr[lru]) { | |
2112 | nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX); | |
2113 | nr[lru] -= nr_to_scan; | |
2114 | ||
2115 | nr_reclaimed += shrink_list(lru, nr_to_scan, | |
2116 | lruvec, sc); | |
2117 | } | |
2118 | } | |
e82e0561 MG |
2119 | |
2120 | if (nr_reclaimed < nr_to_reclaim || scan_adjusted) | |
2121 | continue; | |
2122 | ||
e82e0561 MG |
2123 | /* |
2124 | * For kswapd and memcg, reclaim at least the number of pages | |
1a501907 | 2125 | * requested. Ensure that the anon and file LRUs are scanned |
e82e0561 MG |
2126 | * proportionally what was requested by get_scan_count(). We |
2127 | * stop reclaiming one LRU and reduce the amount scanning | |
2128 | * proportional to the original scan target. | |
2129 | */ | |
2130 | nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE]; | |
2131 | nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON]; | |
2132 | ||
1a501907 MG |
2133 | /* |
2134 | * It's just vindictive to attack the larger once the smaller | |
2135 | * has gone to zero. And given the way we stop scanning the | |
2136 | * smaller below, this makes sure that we only make one nudge | |
2137 | * towards proportionality once we've got nr_to_reclaim. | |
2138 | */ | |
2139 | if (!nr_file || !nr_anon) | |
2140 | break; | |
2141 | ||
e82e0561 MG |
2142 | if (nr_file > nr_anon) { |
2143 | unsigned long scan_target = targets[LRU_INACTIVE_ANON] + | |
2144 | targets[LRU_ACTIVE_ANON] + 1; | |
2145 | lru = LRU_BASE; | |
2146 | percentage = nr_anon * 100 / scan_target; | |
2147 | } else { | |
2148 | unsigned long scan_target = targets[LRU_INACTIVE_FILE] + | |
2149 | targets[LRU_ACTIVE_FILE] + 1; | |
2150 | lru = LRU_FILE; | |
2151 | percentage = nr_file * 100 / scan_target; | |
2152 | } | |
2153 | ||
2154 | /* Stop scanning the smaller of the LRU */ | |
2155 | nr[lru] = 0; | |
2156 | nr[lru + LRU_ACTIVE] = 0; | |
2157 | ||
2158 | /* | |
2159 | * Recalculate the other LRU scan count based on its original | |
2160 | * scan target and the percentage scanning already complete | |
2161 | */ | |
2162 | lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE; | |
2163 | nr_scanned = targets[lru] - nr[lru]; | |
2164 | nr[lru] = targets[lru] * (100 - percentage) / 100; | |
2165 | nr[lru] -= min(nr[lru], nr_scanned); | |
2166 | ||
2167 | lru += LRU_ACTIVE; | |
2168 | nr_scanned = targets[lru] - nr[lru]; | |
2169 | nr[lru] = targets[lru] * (100 - percentage) / 100; | |
2170 | nr[lru] -= min(nr[lru], nr_scanned); | |
2171 | ||
2172 | scan_adjusted = true; | |
9b4f98cd JW |
2173 | } |
2174 | blk_finish_plug(&plug); | |
2175 | sc->nr_reclaimed += nr_reclaimed; | |
2176 | ||
2177 | /* | |
2178 | * Even if we did not try to evict anon pages at all, we want to | |
2179 | * rebalance the anon lru active/inactive ratio. | |
2180 | */ | |
2181 | if (inactive_anon_is_low(lruvec)) | |
2182 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, | |
2183 | sc, LRU_ACTIVE_ANON); | |
2184 | ||
2185 | throttle_vm_writeout(sc->gfp_mask); | |
2186 | } | |
2187 | ||
23b9da55 | 2188 | /* Use reclaim/compaction for costly allocs or under memory pressure */ |
9e3b2f8c | 2189 | static bool in_reclaim_compaction(struct scan_control *sc) |
23b9da55 | 2190 | { |
d84da3f9 | 2191 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && |
23b9da55 | 2192 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || |
9e3b2f8c | 2193 | sc->priority < DEF_PRIORITY - 2)) |
23b9da55 MG |
2194 | return true; |
2195 | ||
2196 | return false; | |
2197 | } | |
2198 | ||
3e7d3449 | 2199 | /* |
23b9da55 MG |
2200 | * Reclaim/compaction is used for high-order allocation requests. It reclaims |
2201 | * order-0 pages before compacting the zone. should_continue_reclaim() returns | |
2202 | * true if more pages should be reclaimed such that when the page allocator | |
2203 | * calls try_to_compact_zone() that it will have enough free pages to succeed. | |
2204 | * It will give up earlier than that if there is difficulty reclaiming pages. | |
3e7d3449 | 2205 | */ |
9b4f98cd | 2206 | static inline bool should_continue_reclaim(struct zone *zone, |
3e7d3449 MG |
2207 | unsigned long nr_reclaimed, |
2208 | unsigned long nr_scanned, | |
2209 | struct scan_control *sc) | |
2210 | { | |
2211 | unsigned long pages_for_compaction; | |
2212 | unsigned long inactive_lru_pages; | |
2213 | ||
2214 | /* If not in reclaim/compaction mode, stop */ | |
9e3b2f8c | 2215 | if (!in_reclaim_compaction(sc)) |
3e7d3449 MG |
2216 | return false; |
2217 | ||
2876592f MG |
2218 | /* Consider stopping depending on scan and reclaim activity */ |
2219 | if (sc->gfp_mask & __GFP_REPEAT) { | |
2220 | /* | |
2221 | * For __GFP_REPEAT allocations, stop reclaiming if the | |
2222 | * full LRU list has been scanned and we are still failing | |
2223 | * to reclaim pages. This full LRU scan is potentially | |
2224 | * expensive but a __GFP_REPEAT caller really wants to succeed | |
2225 | */ | |
2226 | if (!nr_reclaimed && !nr_scanned) | |
2227 | return false; | |
2228 | } else { | |
2229 | /* | |
2230 | * For non-__GFP_REPEAT allocations which can presumably | |
2231 | * fail without consequence, stop if we failed to reclaim | |
2232 | * any pages from the last SWAP_CLUSTER_MAX number of | |
2233 | * pages that were scanned. This will return to the | |
2234 | * caller faster at the risk reclaim/compaction and | |
2235 | * the resulting allocation attempt fails | |
2236 | */ | |
2237 | if (!nr_reclaimed) | |
2238 | return false; | |
2239 | } | |
3e7d3449 MG |
2240 | |
2241 | /* | |
2242 | * If we have not reclaimed enough pages for compaction and the | |
2243 | * inactive lists are large enough, continue reclaiming | |
2244 | */ | |
2245 | pages_for_compaction = (2UL << sc->order); | |
9b4f98cd | 2246 | inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE); |
ec8acf20 | 2247 | if (get_nr_swap_pages() > 0) |
9b4f98cd | 2248 | inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON); |
3e7d3449 MG |
2249 | if (sc->nr_reclaimed < pages_for_compaction && |
2250 | inactive_lru_pages > pages_for_compaction) | |
2251 | return true; | |
2252 | ||
2253 | /* If compaction would go ahead or the allocation would succeed, stop */ | |
9b4f98cd | 2254 | switch (compaction_suitable(zone, sc->order)) { |
3e7d3449 MG |
2255 | case COMPACT_PARTIAL: |
2256 | case COMPACT_CONTINUE: | |
2257 | return false; | |
2258 | default: | |
2259 | return true; | |
2260 | } | |
2261 | } | |
2262 | ||
2344d7e4 | 2263 | static bool shrink_zone(struct zone *zone, struct scan_control *sc) |
1da177e4 | 2264 | { |
f0fdc5e8 | 2265 | unsigned long nr_reclaimed, nr_scanned; |
2344d7e4 | 2266 | bool reclaimable = false; |
1da177e4 | 2267 | |
9b4f98cd JW |
2268 | do { |
2269 | struct mem_cgroup *root = sc->target_mem_cgroup; | |
2270 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2271 | .zone = zone, | |
2272 | .priority = sc->priority, | |
2273 | }; | |
694fbc0f | 2274 | struct mem_cgroup *memcg; |
3e7d3449 | 2275 | |
9b4f98cd JW |
2276 | nr_reclaimed = sc->nr_reclaimed; |
2277 | nr_scanned = sc->nr_scanned; | |
1da177e4 | 2278 | |
694fbc0f AM |
2279 | memcg = mem_cgroup_iter(root, NULL, &reclaim); |
2280 | do { | |
9b4f98cd | 2281 | struct lruvec *lruvec; |
02695175 | 2282 | int swappiness; |
5660048c | 2283 | |
9b4f98cd | 2284 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
02695175 | 2285 | swappiness = mem_cgroup_swappiness(memcg); |
f9be23d6 | 2286 | |
02695175 | 2287 | shrink_lruvec(lruvec, swappiness, sc); |
f16015fb | 2288 | |
9b4f98cd | 2289 | /* |
a394cb8e MH |
2290 | * Direct reclaim and kswapd have to scan all memory |
2291 | * cgroups to fulfill the overall scan target for the | |
9b4f98cd | 2292 | * zone. |
a394cb8e MH |
2293 | * |
2294 | * Limit reclaim, on the other hand, only cares about | |
2295 | * nr_to_reclaim pages to be reclaimed and it will | |
2296 | * retry with decreasing priority if one round over the | |
2297 | * whole hierarchy is not sufficient. | |
9b4f98cd | 2298 | */ |
a394cb8e MH |
2299 | if (!global_reclaim(sc) && |
2300 | sc->nr_reclaimed >= sc->nr_to_reclaim) { | |
9b4f98cd JW |
2301 | mem_cgroup_iter_break(root, memcg); |
2302 | break; | |
2303 | } | |
694fbc0f AM |
2304 | memcg = mem_cgroup_iter(root, memcg, &reclaim); |
2305 | } while (memcg); | |
70ddf637 AV |
2306 | |
2307 | vmpressure(sc->gfp_mask, sc->target_mem_cgroup, | |
2308 | sc->nr_scanned - nr_scanned, | |
2309 | sc->nr_reclaimed - nr_reclaimed); | |
2310 | ||
2344d7e4 JW |
2311 | if (sc->nr_reclaimed - nr_reclaimed) |
2312 | reclaimable = true; | |
2313 | ||
9b4f98cd JW |
2314 | } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed, |
2315 | sc->nr_scanned - nr_scanned, sc)); | |
2344d7e4 JW |
2316 | |
2317 | return reclaimable; | |
f16015fb JW |
2318 | } |
2319 | ||
fe4b1b24 | 2320 | /* Returns true if compaction should go ahead for a high-order request */ |
0b06496a | 2321 | static inline bool compaction_ready(struct zone *zone, int order) |
fe4b1b24 MG |
2322 | { |
2323 | unsigned long balance_gap, watermark; | |
2324 | bool watermark_ok; | |
2325 | ||
fe4b1b24 MG |
2326 | /* |
2327 | * Compaction takes time to run and there are potentially other | |
2328 | * callers using the pages just freed. Continue reclaiming until | |
2329 | * there is a buffer of free pages available to give compaction | |
2330 | * a reasonable chance of completing and allocating the page | |
2331 | */ | |
4be89a34 JZ |
2332 | balance_gap = min(low_wmark_pages(zone), DIV_ROUND_UP( |
2333 | zone->managed_pages, KSWAPD_ZONE_BALANCE_GAP_RATIO)); | |
0b06496a | 2334 | watermark = high_wmark_pages(zone) + balance_gap + (2UL << order); |
fe4b1b24 MG |
2335 | watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0); |
2336 | ||
2337 | /* | |
2338 | * If compaction is deferred, reclaim up to a point where | |
2339 | * compaction will have a chance of success when re-enabled | |
2340 | */ | |
0b06496a | 2341 | if (compaction_deferred(zone, order)) |
fe4b1b24 MG |
2342 | return watermark_ok; |
2343 | ||
2344 | /* If compaction is not ready to start, keep reclaiming */ | |
0b06496a | 2345 | if (!compaction_suitable(zone, order)) |
fe4b1b24 MG |
2346 | return false; |
2347 | ||
2348 | return watermark_ok; | |
2349 | } | |
2350 | ||
1da177e4 LT |
2351 | /* |
2352 | * This is the direct reclaim path, for page-allocating processes. We only | |
2353 | * try to reclaim pages from zones which will satisfy the caller's allocation | |
2354 | * request. | |
2355 | * | |
41858966 MG |
2356 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). |
2357 | * Because: | |
1da177e4 LT |
2358 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
2359 | * allocation or | |
41858966 MG |
2360 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones |
2361 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' | |
2362 | * zone defense algorithm. | |
1da177e4 | 2363 | * |
1da177e4 LT |
2364 | * If a zone is deemed to be full of pinned pages then just give it a light |
2365 | * scan then give up on it. | |
2344d7e4 JW |
2366 | * |
2367 | * Returns true if a zone was reclaimable. | |
1da177e4 | 2368 | */ |
2344d7e4 | 2369 | static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc) |
1da177e4 | 2370 | { |
dd1a239f | 2371 | struct zoneref *z; |
54a6eb5c | 2372 | struct zone *zone; |
0608f43d AM |
2373 | unsigned long nr_soft_reclaimed; |
2374 | unsigned long nr_soft_scanned; | |
65ec02cb | 2375 | unsigned long lru_pages = 0; |
65ec02cb | 2376 | struct reclaim_state *reclaim_state = current->reclaim_state; |
619d0d76 | 2377 | gfp_t orig_mask; |
3115cd91 VD |
2378 | struct shrink_control shrink = { |
2379 | .gfp_mask = sc->gfp_mask, | |
2380 | }; | |
9bbc04ee | 2381 | enum zone_type requested_highidx = gfp_zone(sc->gfp_mask); |
2344d7e4 | 2382 | bool reclaimable = false; |
1cfb419b | 2383 | |
cc715d99 MG |
2384 | /* |
2385 | * If the number of buffer_heads in the machine exceeds the maximum | |
2386 | * allowed level, force direct reclaim to scan the highmem zone as | |
2387 | * highmem pages could be pinning lowmem pages storing buffer_heads | |
2388 | */ | |
619d0d76 | 2389 | orig_mask = sc->gfp_mask; |
cc715d99 MG |
2390 | if (buffer_heads_over_limit) |
2391 | sc->gfp_mask |= __GFP_HIGHMEM; | |
2392 | ||
3115cd91 | 2393 | nodes_clear(shrink.nodes_to_scan); |
65ec02cb | 2394 | |
d4debc66 MG |
2395 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
2396 | gfp_zone(sc->gfp_mask), sc->nodemask) { | |
f3fe6512 | 2397 | if (!populated_zone(zone)) |
1da177e4 | 2398 | continue; |
1cfb419b KH |
2399 | /* |
2400 | * Take care memory controller reclaiming has small influence | |
2401 | * to global LRU. | |
2402 | */ | |
89b5fae5 | 2403 | if (global_reclaim(sc)) { |
1cfb419b KH |
2404 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
2405 | continue; | |
65ec02cb VD |
2406 | |
2407 | lru_pages += zone_reclaimable_pages(zone); | |
3115cd91 | 2408 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); |
65ec02cb | 2409 | |
6e543d57 LD |
2410 | if (sc->priority != DEF_PRIORITY && |
2411 | !zone_reclaimable(zone)) | |
1cfb419b | 2412 | continue; /* Let kswapd poll it */ |
0b06496a JW |
2413 | |
2414 | /* | |
2415 | * If we already have plenty of memory free for | |
2416 | * compaction in this zone, don't free any more. | |
2417 | * Even though compaction is invoked for any | |
2418 | * non-zero order, only frequent costly order | |
2419 | * reclamation is disruptive enough to become a | |
2420 | * noticeable problem, like transparent huge | |
2421 | * page allocations. | |
2422 | */ | |
2423 | if (IS_ENABLED(CONFIG_COMPACTION) && | |
2424 | sc->order > PAGE_ALLOC_COSTLY_ORDER && | |
2425 | zonelist_zone_idx(z) <= requested_highidx && | |
2426 | compaction_ready(zone, sc->order)) { | |
2427 | sc->compaction_ready = true; | |
2428 | continue; | |
e0887c19 | 2429 | } |
0b06496a | 2430 | |
0608f43d AM |
2431 | /* |
2432 | * This steals pages from memory cgroups over softlimit | |
2433 | * and returns the number of reclaimed pages and | |
2434 | * scanned pages. This works for global memory pressure | |
2435 | * and balancing, not for a memcg's limit. | |
2436 | */ | |
2437 | nr_soft_scanned = 0; | |
2438 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | |
2439 | sc->order, sc->gfp_mask, | |
2440 | &nr_soft_scanned); | |
2441 | sc->nr_reclaimed += nr_soft_reclaimed; | |
2442 | sc->nr_scanned += nr_soft_scanned; | |
2344d7e4 JW |
2443 | if (nr_soft_reclaimed) |
2444 | reclaimable = true; | |
ac34a1a3 | 2445 | /* need some check for avoid more shrink_zone() */ |
1cfb419b | 2446 | } |
408d8544 | 2447 | |
2344d7e4 JW |
2448 | if (shrink_zone(zone, sc)) |
2449 | reclaimable = true; | |
2450 | ||
2451 | if (global_reclaim(sc) && | |
2452 | !reclaimable && zone_reclaimable(zone)) | |
2453 | reclaimable = true; | |
1da177e4 | 2454 | } |
e0c23279 | 2455 | |
65ec02cb VD |
2456 | /* |
2457 | * Don't shrink slabs when reclaiming memory from over limit cgroups | |
2458 | * but do shrink slab at least once when aborting reclaim for | |
2459 | * compaction to avoid unevenly scanning file/anon LRU pages over slab | |
2460 | * pages. | |
2461 | */ | |
2462 | if (global_reclaim(sc)) { | |
3115cd91 | 2463 | shrink_slab(&shrink, sc->nr_scanned, lru_pages); |
65ec02cb VD |
2464 | if (reclaim_state) { |
2465 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; | |
2466 | reclaim_state->reclaimed_slab = 0; | |
2467 | } | |
2468 | } | |
2469 | ||
619d0d76 WY |
2470 | /* |
2471 | * Restore to original mask to avoid the impact on the caller if we | |
2472 | * promoted it to __GFP_HIGHMEM. | |
2473 | */ | |
2474 | sc->gfp_mask = orig_mask; | |
d1908362 | 2475 | |
2344d7e4 | 2476 | return reclaimable; |
1da177e4 | 2477 | } |
4f98a2fe | 2478 | |
1da177e4 LT |
2479 | /* |
2480 | * This is the main entry point to direct page reclaim. | |
2481 | * | |
2482 | * If a full scan of the inactive list fails to free enough memory then we | |
2483 | * are "out of memory" and something needs to be killed. | |
2484 | * | |
2485 | * If the caller is !__GFP_FS then the probability of a failure is reasonably | |
2486 | * high - the zone may be full of dirty or under-writeback pages, which this | |
5b0830cb JA |
2487 | * caller can't do much about. We kick the writeback threads and take explicit |
2488 | * naps in the hope that some of these pages can be written. But if the | |
2489 | * allocating task holds filesystem locks which prevent writeout this might not | |
2490 | * work, and the allocation attempt will fail. | |
a41f24ea NA |
2491 | * |
2492 | * returns: 0, if no pages reclaimed | |
2493 | * else, the number of pages reclaimed | |
1da177e4 | 2494 | */ |
dac1d27b | 2495 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
3115cd91 | 2496 | struct scan_control *sc) |
1da177e4 | 2497 | { |
69e05944 | 2498 | unsigned long total_scanned = 0; |
22fba335 | 2499 | unsigned long writeback_threshold; |
2344d7e4 | 2500 | bool zones_reclaimable; |
1da177e4 | 2501 | |
873b4771 KK |
2502 | delayacct_freepages_start(); |
2503 | ||
89b5fae5 | 2504 | if (global_reclaim(sc)) |
1cfb419b | 2505 | count_vm_event(ALLOCSTALL); |
1da177e4 | 2506 | |
9e3b2f8c | 2507 | do { |
70ddf637 AV |
2508 | vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup, |
2509 | sc->priority); | |
66e1707b | 2510 | sc->nr_scanned = 0; |
2344d7e4 | 2511 | zones_reclaimable = shrink_zones(zonelist, sc); |
c6a8a8c5 | 2512 | |
66e1707b | 2513 | total_scanned += sc->nr_scanned; |
bb21c7ce | 2514 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
0b06496a JW |
2515 | break; |
2516 | ||
2517 | if (sc->compaction_ready) | |
2518 | break; | |
1da177e4 | 2519 | |
0e50ce3b MK |
2520 | /* |
2521 | * If we're getting trouble reclaiming, start doing | |
2522 | * writepage even in laptop mode. | |
2523 | */ | |
2524 | if (sc->priority < DEF_PRIORITY - 2) | |
2525 | sc->may_writepage = 1; | |
2526 | ||
1da177e4 LT |
2527 | /* |
2528 | * Try to write back as many pages as we just scanned. This | |
2529 | * tends to cause slow streaming writers to write data to the | |
2530 | * disk smoothly, at the dirtying rate, which is nice. But | |
2531 | * that's undesirable in laptop mode, where we *want* lumpy | |
2532 | * writeout. So in laptop mode, write out the whole world. | |
2533 | */ | |
22fba335 KM |
2534 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; |
2535 | if (total_scanned > writeback_threshold) { | |
0e175a18 CW |
2536 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, |
2537 | WB_REASON_TRY_TO_FREE_PAGES); | |
66e1707b | 2538 | sc->may_writepage = 1; |
1da177e4 | 2539 | } |
0b06496a | 2540 | } while (--sc->priority >= 0); |
bb21c7ce | 2541 | |
873b4771 KK |
2542 | delayacct_freepages_end(); |
2543 | ||
bb21c7ce KM |
2544 | if (sc->nr_reclaimed) |
2545 | return sc->nr_reclaimed; | |
2546 | ||
0cee34fd | 2547 | /* Aborted reclaim to try compaction? don't OOM, then */ |
0b06496a | 2548 | if (sc->compaction_ready) |
7335084d MG |
2549 | return 1; |
2550 | ||
2344d7e4 JW |
2551 | /* Any of the zones still reclaimable? Don't OOM. */ |
2552 | if (zones_reclaimable) | |
bb21c7ce KM |
2553 | return 1; |
2554 | ||
2555 | return 0; | |
1da177e4 LT |
2556 | } |
2557 | ||
5515061d MG |
2558 | static bool pfmemalloc_watermark_ok(pg_data_t *pgdat) |
2559 | { | |
2560 | struct zone *zone; | |
2561 | unsigned long pfmemalloc_reserve = 0; | |
2562 | unsigned long free_pages = 0; | |
2563 | int i; | |
2564 | bool wmark_ok; | |
2565 | ||
2566 | for (i = 0; i <= ZONE_NORMAL; i++) { | |
2567 | zone = &pgdat->node_zones[i]; | |
675becce MG |
2568 | if (!populated_zone(zone)) |
2569 | continue; | |
2570 | ||
5515061d MG |
2571 | pfmemalloc_reserve += min_wmark_pages(zone); |
2572 | free_pages += zone_page_state(zone, NR_FREE_PAGES); | |
2573 | } | |
2574 | ||
675becce MG |
2575 | /* If there are no reserves (unexpected config) then do not throttle */ |
2576 | if (!pfmemalloc_reserve) | |
2577 | return true; | |
2578 | ||
5515061d MG |
2579 | wmark_ok = free_pages > pfmemalloc_reserve / 2; |
2580 | ||
2581 | /* kswapd must be awake if processes are being throttled */ | |
2582 | if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) { | |
2583 | pgdat->classzone_idx = min(pgdat->classzone_idx, | |
2584 | (enum zone_type)ZONE_NORMAL); | |
2585 | wake_up_interruptible(&pgdat->kswapd_wait); | |
2586 | } | |
2587 | ||
2588 | return wmark_ok; | |
2589 | } | |
2590 | ||
2591 | /* | |
2592 | * Throttle direct reclaimers if backing storage is backed by the network | |
2593 | * and the PFMEMALLOC reserve for the preferred node is getting dangerously | |
2594 | * depleted. kswapd will continue to make progress and wake the processes | |
50694c28 MG |
2595 | * when the low watermark is reached. |
2596 | * | |
2597 | * Returns true if a fatal signal was delivered during throttling. If this | |
2598 | * happens, the page allocator should not consider triggering the OOM killer. | |
5515061d | 2599 | */ |
50694c28 | 2600 | static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist, |
5515061d MG |
2601 | nodemask_t *nodemask) |
2602 | { | |
675becce | 2603 | struct zoneref *z; |
5515061d | 2604 | struct zone *zone; |
675becce | 2605 | pg_data_t *pgdat = NULL; |
5515061d MG |
2606 | |
2607 | /* | |
2608 | * Kernel threads should not be throttled as they may be indirectly | |
2609 | * responsible for cleaning pages necessary for reclaim to make forward | |
2610 | * progress. kjournald for example may enter direct reclaim while | |
2611 | * committing a transaction where throttling it could forcing other | |
2612 | * processes to block on log_wait_commit(). | |
2613 | */ | |
2614 | if (current->flags & PF_KTHREAD) | |
50694c28 MG |
2615 | goto out; |
2616 | ||
2617 | /* | |
2618 | * If a fatal signal is pending, this process should not throttle. | |
2619 | * It should return quickly so it can exit and free its memory | |
2620 | */ | |
2621 | if (fatal_signal_pending(current)) | |
2622 | goto out; | |
5515061d | 2623 | |
675becce MG |
2624 | /* |
2625 | * Check if the pfmemalloc reserves are ok by finding the first node | |
2626 | * with a usable ZONE_NORMAL or lower zone. The expectation is that | |
2627 | * GFP_KERNEL will be required for allocating network buffers when | |
2628 | * swapping over the network so ZONE_HIGHMEM is unusable. | |
2629 | * | |
2630 | * Throttling is based on the first usable node and throttled processes | |
2631 | * wait on a queue until kswapd makes progress and wakes them. There | |
2632 | * is an affinity then between processes waking up and where reclaim | |
2633 | * progress has been made assuming the process wakes on the same node. | |
2634 | * More importantly, processes running on remote nodes will not compete | |
2635 | * for remote pfmemalloc reserves and processes on different nodes | |
2636 | * should make reasonable progress. | |
2637 | */ | |
2638 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
2639 | gfp_mask, nodemask) { | |
2640 | if (zone_idx(zone) > ZONE_NORMAL) | |
2641 | continue; | |
2642 | ||
2643 | /* Throttle based on the first usable node */ | |
2644 | pgdat = zone->zone_pgdat; | |
2645 | if (pfmemalloc_watermark_ok(pgdat)) | |
2646 | goto out; | |
2647 | break; | |
2648 | } | |
2649 | ||
2650 | /* If no zone was usable by the allocation flags then do not throttle */ | |
2651 | if (!pgdat) | |
50694c28 | 2652 | goto out; |
5515061d | 2653 | |
68243e76 MG |
2654 | /* Account for the throttling */ |
2655 | count_vm_event(PGSCAN_DIRECT_THROTTLE); | |
2656 | ||
5515061d MG |
2657 | /* |
2658 | * If the caller cannot enter the filesystem, it's possible that it | |
2659 | * is due to the caller holding an FS lock or performing a journal | |
2660 | * transaction in the case of a filesystem like ext[3|4]. In this case, | |
2661 | * it is not safe to block on pfmemalloc_wait as kswapd could be | |
2662 | * blocked waiting on the same lock. Instead, throttle for up to a | |
2663 | * second before continuing. | |
2664 | */ | |
2665 | if (!(gfp_mask & __GFP_FS)) { | |
2666 | wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, | |
2667 | pfmemalloc_watermark_ok(pgdat), HZ); | |
50694c28 MG |
2668 | |
2669 | goto check_pending; | |
5515061d MG |
2670 | } |
2671 | ||
2672 | /* Throttle until kswapd wakes the process */ | |
2673 | wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, | |
2674 | pfmemalloc_watermark_ok(pgdat)); | |
50694c28 MG |
2675 | |
2676 | check_pending: | |
2677 | if (fatal_signal_pending(current)) | |
2678 | return true; | |
2679 | ||
2680 | out: | |
2681 | return false; | |
5515061d MG |
2682 | } |
2683 | ||
dac1d27b | 2684 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
327c0e96 | 2685 | gfp_t gfp_mask, nodemask_t *nodemask) |
66e1707b | 2686 | { |
33906bc5 | 2687 | unsigned long nr_reclaimed; |
66e1707b | 2688 | struct scan_control sc = { |
ee814fe2 | 2689 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
21caf2fc | 2690 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
ee814fe2 JW |
2691 | .order = order, |
2692 | .nodemask = nodemask, | |
2693 | .priority = DEF_PRIORITY, | |
66e1707b | 2694 | .may_writepage = !laptop_mode, |
a6dc60f8 | 2695 | .may_unmap = 1, |
2e2e4259 | 2696 | .may_swap = 1, |
66e1707b BS |
2697 | }; |
2698 | ||
5515061d | 2699 | /* |
50694c28 MG |
2700 | * Do not enter reclaim if fatal signal was delivered while throttled. |
2701 | * 1 is returned so that the page allocator does not OOM kill at this | |
2702 | * point. | |
5515061d | 2703 | */ |
50694c28 | 2704 | if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask)) |
5515061d MG |
2705 | return 1; |
2706 | ||
33906bc5 MG |
2707 | trace_mm_vmscan_direct_reclaim_begin(order, |
2708 | sc.may_writepage, | |
2709 | gfp_mask); | |
2710 | ||
3115cd91 | 2711 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
33906bc5 MG |
2712 | |
2713 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); | |
2714 | ||
2715 | return nr_reclaimed; | |
66e1707b BS |
2716 | } |
2717 | ||
c255a458 | 2718 | #ifdef CONFIG_MEMCG |
66e1707b | 2719 | |
72835c86 | 2720 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg, |
4e416953 | 2721 | gfp_t gfp_mask, bool noswap, |
0ae5e89c YH |
2722 | struct zone *zone, |
2723 | unsigned long *nr_scanned) | |
4e416953 BS |
2724 | { |
2725 | struct scan_control sc = { | |
b8f5c566 | 2726 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
ee814fe2 | 2727 | .target_mem_cgroup = memcg, |
4e416953 BS |
2728 | .may_writepage = !laptop_mode, |
2729 | .may_unmap = 1, | |
2730 | .may_swap = !noswap, | |
4e416953 | 2731 | }; |
f9be23d6 | 2732 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
02695175 | 2733 | int swappiness = mem_cgroup_swappiness(memcg); |
0ae5e89c | 2734 | |
4e416953 BS |
2735 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2736 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); | |
bdce6d9e | 2737 | |
9e3b2f8c | 2738 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, |
bdce6d9e KM |
2739 | sc.may_writepage, |
2740 | sc.gfp_mask); | |
2741 | ||
4e416953 BS |
2742 | /* |
2743 | * NOTE: Although we can get the priority field, using it | |
2744 | * here is not a good idea, since it limits the pages we can scan. | |
2745 | * if we don't reclaim here, the shrink_zone from balance_pgdat | |
2746 | * will pick up pages from other mem cgroup's as well. We hack | |
2747 | * the priority and make it zero. | |
2748 | */ | |
02695175 | 2749 | shrink_lruvec(lruvec, swappiness, &sc); |
bdce6d9e KM |
2750 | |
2751 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); | |
2752 | ||
0ae5e89c | 2753 | *nr_scanned = sc.nr_scanned; |
4e416953 BS |
2754 | return sc.nr_reclaimed; |
2755 | } | |
2756 | ||
72835c86 | 2757 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, |
a7885eb8 | 2758 | gfp_t gfp_mask, |
185efc0f | 2759 | bool noswap) |
66e1707b | 2760 | { |
4e416953 | 2761 | struct zonelist *zonelist; |
bdce6d9e | 2762 | unsigned long nr_reclaimed; |
889976db | 2763 | int nid; |
66e1707b | 2764 | struct scan_control sc = { |
22fba335 | 2765 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
a09ed5e0 YH |
2766 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2767 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), | |
ee814fe2 JW |
2768 | .target_mem_cgroup = memcg, |
2769 | .priority = DEF_PRIORITY, | |
2770 | .may_writepage = !laptop_mode, | |
2771 | .may_unmap = 1, | |
2772 | .may_swap = !noswap, | |
a09ed5e0 | 2773 | }; |
66e1707b | 2774 | |
889976db YH |
2775 | /* |
2776 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't | |
2777 | * take care of from where we get pages. So the node where we start the | |
2778 | * scan does not need to be the current node. | |
2779 | */ | |
72835c86 | 2780 | nid = mem_cgroup_select_victim_node(memcg); |
889976db YH |
2781 | |
2782 | zonelist = NODE_DATA(nid)->node_zonelists; | |
bdce6d9e KM |
2783 | |
2784 | trace_mm_vmscan_memcg_reclaim_begin(0, | |
2785 | sc.may_writepage, | |
2786 | sc.gfp_mask); | |
2787 | ||
3115cd91 | 2788 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
bdce6d9e KM |
2789 | |
2790 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); | |
2791 | ||
2792 | return nr_reclaimed; | |
66e1707b BS |
2793 | } |
2794 | #endif | |
2795 | ||
9e3b2f8c | 2796 | static void age_active_anon(struct zone *zone, struct scan_control *sc) |
f16015fb | 2797 | { |
b95a2f2d | 2798 | struct mem_cgroup *memcg; |
f16015fb | 2799 | |
b95a2f2d JW |
2800 | if (!total_swap_pages) |
2801 | return; | |
2802 | ||
2803 | memcg = mem_cgroup_iter(NULL, NULL, NULL); | |
2804 | do { | |
c56d5c7d | 2805 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
b95a2f2d | 2806 | |
c56d5c7d | 2807 | if (inactive_anon_is_low(lruvec)) |
1a93be0e | 2808 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
9e3b2f8c | 2809 | sc, LRU_ACTIVE_ANON); |
b95a2f2d JW |
2810 | |
2811 | memcg = mem_cgroup_iter(NULL, memcg, NULL); | |
2812 | } while (memcg); | |
f16015fb JW |
2813 | } |
2814 | ||
60cefed4 JW |
2815 | static bool zone_balanced(struct zone *zone, int order, |
2816 | unsigned long balance_gap, int classzone_idx) | |
2817 | { | |
2818 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) + | |
2819 | balance_gap, classzone_idx, 0)) | |
2820 | return false; | |
2821 | ||
d84da3f9 KS |
2822 | if (IS_ENABLED(CONFIG_COMPACTION) && order && |
2823 | !compaction_suitable(zone, order)) | |
60cefed4 JW |
2824 | return false; |
2825 | ||
2826 | return true; | |
2827 | } | |
2828 | ||
1741c877 | 2829 | /* |
4ae0a48b ZC |
2830 | * pgdat_balanced() is used when checking if a node is balanced. |
2831 | * | |
2832 | * For order-0, all zones must be balanced! | |
2833 | * | |
2834 | * For high-order allocations only zones that meet watermarks and are in a | |
2835 | * zone allowed by the callers classzone_idx are added to balanced_pages. The | |
2836 | * total of balanced pages must be at least 25% of the zones allowed by | |
2837 | * classzone_idx for the node to be considered balanced. Forcing all zones to | |
2838 | * be balanced for high orders can cause excessive reclaim when there are | |
2839 | * imbalanced zones. | |
1741c877 MG |
2840 | * The choice of 25% is due to |
2841 | * o a 16M DMA zone that is balanced will not balance a zone on any | |
2842 | * reasonable sized machine | |
2843 | * o On all other machines, the top zone must be at least a reasonable | |
25985edc | 2844 | * percentage of the middle zones. For example, on 32-bit x86, highmem |
1741c877 MG |
2845 | * would need to be at least 256M for it to be balance a whole node. |
2846 | * Similarly, on x86-64 the Normal zone would need to be at least 1G | |
2847 | * to balance a node on its own. These seemed like reasonable ratios. | |
2848 | */ | |
4ae0a48b | 2849 | static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx) |
1741c877 | 2850 | { |
b40da049 | 2851 | unsigned long managed_pages = 0; |
4ae0a48b | 2852 | unsigned long balanced_pages = 0; |
1741c877 MG |
2853 | int i; |
2854 | ||
4ae0a48b ZC |
2855 | /* Check the watermark levels */ |
2856 | for (i = 0; i <= classzone_idx; i++) { | |
2857 | struct zone *zone = pgdat->node_zones + i; | |
1741c877 | 2858 | |
4ae0a48b ZC |
2859 | if (!populated_zone(zone)) |
2860 | continue; | |
2861 | ||
b40da049 | 2862 | managed_pages += zone->managed_pages; |
4ae0a48b ZC |
2863 | |
2864 | /* | |
2865 | * A special case here: | |
2866 | * | |
2867 | * balance_pgdat() skips over all_unreclaimable after | |
2868 | * DEF_PRIORITY. Effectively, it considers them balanced so | |
2869 | * they must be considered balanced here as well! | |
2870 | */ | |
6e543d57 | 2871 | if (!zone_reclaimable(zone)) { |
b40da049 | 2872 | balanced_pages += zone->managed_pages; |
4ae0a48b ZC |
2873 | continue; |
2874 | } | |
2875 | ||
2876 | if (zone_balanced(zone, order, 0, i)) | |
b40da049 | 2877 | balanced_pages += zone->managed_pages; |
4ae0a48b ZC |
2878 | else if (!order) |
2879 | return false; | |
2880 | } | |
2881 | ||
2882 | if (order) | |
b40da049 | 2883 | return balanced_pages >= (managed_pages >> 2); |
4ae0a48b ZC |
2884 | else |
2885 | return true; | |
1741c877 MG |
2886 | } |
2887 | ||
5515061d MG |
2888 | /* |
2889 | * Prepare kswapd for sleeping. This verifies that there are no processes | |
2890 | * waiting in throttle_direct_reclaim() and that watermarks have been met. | |
2891 | * | |
2892 | * Returns true if kswapd is ready to sleep | |
2893 | */ | |
2894 | static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining, | |
dc83edd9 | 2895 | int classzone_idx) |
f50de2d3 | 2896 | { |
f50de2d3 MG |
2897 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ |
2898 | if (remaining) | |
5515061d MG |
2899 | return false; |
2900 | ||
2901 | /* | |
2902 | * There is a potential race between when kswapd checks its watermarks | |
2903 | * and a process gets throttled. There is also a potential race if | |
2904 | * processes get throttled, kswapd wakes, a large process exits therby | |
2905 | * balancing the zones that causes kswapd to miss a wakeup. If kswapd | |
2906 | * is going to sleep, no process should be sleeping on pfmemalloc_wait | |
2907 | * so wake them now if necessary. If necessary, processes will wake | |
2908 | * kswapd and get throttled again | |
2909 | */ | |
2910 | if (waitqueue_active(&pgdat->pfmemalloc_wait)) { | |
2911 | wake_up(&pgdat->pfmemalloc_wait); | |
2912 | return false; | |
2913 | } | |
f50de2d3 | 2914 | |
4ae0a48b | 2915 | return pgdat_balanced(pgdat, order, classzone_idx); |
f50de2d3 MG |
2916 | } |
2917 | ||
75485363 MG |
2918 | /* |
2919 | * kswapd shrinks the zone by the number of pages required to reach | |
2920 | * the high watermark. | |
b8e83b94 MG |
2921 | * |
2922 | * Returns true if kswapd scanned at least the requested number of pages to | |
283aba9f MG |
2923 | * reclaim or if the lack of progress was due to pages under writeback. |
2924 | * This is used to determine if the scanning priority needs to be raised. | |
75485363 | 2925 | */ |
b8e83b94 | 2926 | static bool kswapd_shrink_zone(struct zone *zone, |
7c954f6d | 2927 | int classzone_idx, |
75485363 | 2928 | struct scan_control *sc, |
2ab44f43 MG |
2929 | unsigned long lru_pages, |
2930 | unsigned long *nr_attempted) | |
75485363 | 2931 | { |
7c954f6d MG |
2932 | int testorder = sc->order; |
2933 | unsigned long balance_gap; | |
75485363 MG |
2934 | struct reclaim_state *reclaim_state = current->reclaim_state; |
2935 | struct shrink_control shrink = { | |
2936 | .gfp_mask = sc->gfp_mask, | |
2937 | }; | |
7c954f6d | 2938 | bool lowmem_pressure; |
75485363 MG |
2939 | |
2940 | /* Reclaim above the high watermark. */ | |
2941 | sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone)); | |
7c954f6d MG |
2942 | |
2943 | /* | |
2944 | * Kswapd reclaims only single pages with compaction enabled. Trying | |
2945 | * too hard to reclaim until contiguous free pages have become | |
2946 | * available can hurt performance by evicting too much useful data | |
2947 | * from memory. Do not reclaim more than needed for compaction. | |
2948 | */ | |
2949 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && | |
2950 | compaction_suitable(zone, sc->order) != | |
2951 | COMPACT_SKIPPED) | |
2952 | testorder = 0; | |
2953 | ||
2954 | /* | |
2955 | * We put equal pressure on every zone, unless one zone has way too | |
2956 | * many pages free already. The "too many pages" is defined as the | |
2957 | * high wmark plus a "gap" where the gap is either the low | |
2958 | * watermark or 1% of the zone, whichever is smaller. | |
2959 | */ | |
4be89a34 JZ |
2960 | balance_gap = min(low_wmark_pages(zone), DIV_ROUND_UP( |
2961 | zone->managed_pages, KSWAPD_ZONE_BALANCE_GAP_RATIO)); | |
7c954f6d MG |
2962 | |
2963 | /* | |
2964 | * If there is no low memory pressure or the zone is balanced then no | |
2965 | * reclaim is necessary | |
2966 | */ | |
2967 | lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone)); | |
2968 | if (!lowmem_pressure && zone_balanced(zone, testorder, | |
2969 | balance_gap, classzone_idx)) | |
2970 | return true; | |
2971 | ||
75485363 | 2972 | shrink_zone(zone, sc); |
0ce3d744 DC |
2973 | nodes_clear(shrink.nodes_to_scan); |
2974 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); | |
75485363 MG |
2975 | |
2976 | reclaim_state->reclaimed_slab = 0; | |
6e543d57 | 2977 | shrink_slab(&shrink, sc->nr_scanned, lru_pages); |
75485363 MG |
2978 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
2979 | ||
2ab44f43 MG |
2980 | /* Account for the number of pages attempted to reclaim */ |
2981 | *nr_attempted += sc->nr_to_reclaim; | |
2982 | ||
283aba9f MG |
2983 | zone_clear_flag(zone, ZONE_WRITEBACK); |
2984 | ||
7c954f6d MG |
2985 | /* |
2986 | * If a zone reaches its high watermark, consider it to be no longer | |
2987 | * congested. It's possible there are dirty pages backed by congested | |
2988 | * BDIs but as pressure is relieved, speculatively avoid congestion | |
2989 | * waits. | |
2990 | */ | |
6e543d57 | 2991 | if (zone_reclaimable(zone) && |
7c954f6d MG |
2992 | zone_balanced(zone, testorder, 0, classzone_idx)) { |
2993 | zone_clear_flag(zone, ZONE_CONGESTED); | |
2994 | zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY); | |
2995 | } | |
2996 | ||
b8e83b94 | 2997 | return sc->nr_scanned >= sc->nr_to_reclaim; |
75485363 MG |
2998 | } |
2999 | ||
1da177e4 LT |
3000 | /* |
3001 | * For kswapd, balance_pgdat() will work across all this node's zones until | |
41858966 | 3002 | * they are all at high_wmark_pages(zone). |
1da177e4 | 3003 | * |
0abdee2b | 3004 | * Returns the final order kswapd was reclaiming at |
1da177e4 LT |
3005 | * |
3006 | * There is special handling here for zones which are full of pinned pages. | |
3007 | * This can happen if the pages are all mlocked, or if they are all used by | |
3008 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. | |
3009 | * What we do is to detect the case where all pages in the zone have been | |
3010 | * scanned twice and there has been zero successful reclaim. Mark the zone as | |
3011 | * dead and from now on, only perform a short scan. Basically we're polling | |
3012 | * the zone for when the problem goes away. | |
3013 | * | |
3014 | * kswapd scans the zones in the highmem->normal->dma direction. It skips | |
41858966 MG |
3015 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
3016 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the | |
3017 | * lower zones regardless of the number of free pages in the lower zones. This | |
3018 | * interoperates with the page allocator fallback scheme to ensure that aging | |
3019 | * of pages is balanced across the zones. | |
1da177e4 | 3020 | */ |
99504748 | 3021 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, |
dc83edd9 | 3022 | int *classzone_idx) |
1da177e4 | 3023 | { |
1da177e4 | 3024 | int i; |
99504748 | 3025 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
0608f43d AM |
3026 | unsigned long nr_soft_reclaimed; |
3027 | unsigned long nr_soft_scanned; | |
179e9639 AM |
3028 | struct scan_control sc = { |
3029 | .gfp_mask = GFP_KERNEL, | |
ee814fe2 | 3030 | .order = order, |
b8e83b94 | 3031 | .priority = DEF_PRIORITY, |
ee814fe2 | 3032 | .may_writepage = !laptop_mode, |
a6dc60f8 | 3033 | .may_unmap = 1, |
2e2e4259 | 3034 | .may_swap = 1, |
179e9639 | 3035 | }; |
f8891e5e | 3036 | count_vm_event(PAGEOUTRUN); |
1da177e4 | 3037 | |
9e3b2f8c | 3038 | do { |
1da177e4 | 3039 | unsigned long lru_pages = 0; |
2ab44f43 | 3040 | unsigned long nr_attempted = 0; |
b8e83b94 | 3041 | bool raise_priority = true; |
2ab44f43 | 3042 | bool pgdat_needs_compaction = (order > 0); |
b8e83b94 MG |
3043 | |
3044 | sc.nr_reclaimed = 0; | |
1da177e4 | 3045 | |
d6277db4 RW |
3046 | /* |
3047 | * Scan in the highmem->dma direction for the highest | |
3048 | * zone which needs scanning | |
3049 | */ | |
3050 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { | |
3051 | struct zone *zone = pgdat->node_zones + i; | |
1da177e4 | 3052 | |
d6277db4 RW |
3053 | if (!populated_zone(zone)) |
3054 | continue; | |
1da177e4 | 3055 | |
6e543d57 LD |
3056 | if (sc.priority != DEF_PRIORITY && |
3057 | !zone_reclaimable(zone)) | |
d6277db4 | 3058 | continue; |
1da177e4 | 3059 | |
556adecb RR |
3060 | /* |
3061 | * Do some background aging of the anon list, to give | |
3062 | * pages a chance to be referenced before reclaiming. | |
3063 | */ | |
9e3b2f8c | 3064 | age_active_anon(zone, &sc); |
556adecb | 3065 | |
cc715d99 MG |
3066 | /* |
3067 | * If the number of buffer_heads in the machine | |
3068 | * exceeds the maximum allowed level and this node | |
3069 | * has a highmem zone, force kswapd to reclaim from | |
3070 | * it to relieve lowmem pressure. | |
3071 | */ | |
3072 | if (buffer_heads_over_limit && is_highmem_idx(i)) { | |
3073 | end_zone = i; | |
3074 | break; | |
3075 | } | |
3076 | ||
60cefed4 | 3077 | if (!zone_balanced(zone, order, 0, 0)) { |
d6277db4 | 3078 | end_zone = i; |
e1dbeda6 | 3079 | break; |
439423f6 | 3080 | } else { |
d43006d5 MG |
3081 | /* |
3082 | * If balanced, clear the dirty and congested | |
3083 | * flags | |
3084 | */ | |
439423f6 | 3085 | zone_clear_flag(zone, ZONE_CONGESTED); |
d43006d5 | 3086 | zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY); |
1da177e4 | 3087 | } |
1da177e4 | 3088 | } |
dafcb73e | 3089 | |
b8e83b94 | 3090 | if (i < 0) |
e1dbeda6 AM |
3091 | goto out; |
3092 | ||
1da177e4 LT |
3093 | for (i = 0; i <= end_zone; i++) { |
3094 | struct zone *zone = pgdat->node_zones + i; | |
3095 | ||
2ab44f43 MG |
3096 | if (!populated_zone(zone)) |
3097 | continue; | |
3098 | ||
adea02a1 | 3099 | lru_pages += zone_reclaimable_pages(zone); |
2ab44f43 MG |
3100 | |
3101 | /* | |
3102 | * If any zone is currently balanced then kswapd will | |
3103 | * not call compaction as it is expected that the | |
3104 | * necessary pages are already available. | |
3105 | */ | |
3106 | if (pgdat_needs_compaction && | |
3107 | zone_watermark_ok(zone, order, | |
3108 | low_wmark_pages(zone), | |
3109 | *classzone_idx, 0)) | |
3110 | pgdat_needs_compaction = false; | |
1da177e4 LT |
3111 | } |
3112 | ||
b7ea3c41 MG |
3113 | /* |
3114 | * If we're getting trouble reclaiming, start doing writepage | |
3115 | * even in laptop mode. | |
3116 | */ | |
3117 | if (sc.priority < DEF_PRIORITY - 2) | |
3118 | sc.may_writepage = 1; | |
3119 | ||
1da177e4 LT |
3120 | /* |
3121 | * Now scan the zone in the dma->highmem direction, stopping | |
3122 | * at the last zone which needs scanning. | |
3123 | * | |
3124 | * We do this because the page allocator works in the opposite | |
3125 | * direction. This prevents the page allocator from allocating | |
3126 | * pages behind kswapd's direction of progress, which would | |
3127 | * cause too much scanning of the lower zones. | |
3128 | */ | |
3129 | for (i = 0; i <= end_zone; i++) { | |
3130 | struct zone *zone = pgdat->node_zones + i; | |
3131 | ||
f3fe6512 | 3132 | if (!populated_zone(zone)) |
1da177e4 LT |
3133 | continue; |
3134 | ||
6e543d57 LD |
3135 | if (sc.priority != DEF_PRIORITY && |
3136 | !zone_reclaimable(zone)) | |
1da177e4 LT |
3137 | continue; |
3138 | ||
1da177e4 | 3139 | sc.nr_scanned = 0; |
4e416953 | 3140 | |
0608f43d AM |
3141 | nr_soft_scanned = 0; |
3142 | /* | |
3143 | * Call soft limit reclaim before calling shrink_zone. | |
3144 | */ | |
3145 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | |
3146 | order, sc.gfp_mask, | |
3147 | &nr_soft_scanned); | |
3148 | sc.nr_reclaimed += nr_soft_reclaimed; | |
3149 | ||
32a4330d | 3150 | /* |
7c954f6d MG |
3151 | * There should be no need to raise the scanning |
3152 | * priority if enough pages are already being scanned | |
3153 | * that that high watermark would be met at 100% | |
3154 | * efficiency. | |
fe2c2a10 | 3155 | */ |
7c954f6d MG |
3156 | if (kswapd_shrink_zone(zone, end_zone, &sc, |
3157 | lru_pages, &nr_attempted)) | |
3158 | raise_priority = false; | |
1da177e4 | 3159 | } |
5515061d MG |
3160 | |
3161 | /* | |
3162 | * If the low watermark is met there is no need for processes | |
3163 | * to be throttled on pfmemalloc_wait as they should not be | |
3164 | * able to safely make forward progress. Wake them | |
3165 | */ | |
3166 | if (waitqueue_active(&pgdat->pfmemalloc_wait) && | |
3167 | pfmemalloc_watermark_ok(pgdat)) | |
3168 | wake_up(&pgdat->pfmemalloc_wait); | |
3169 | ||
1da177e4 | 3170 | /* |
b8e83b94 MG |
3171 | * Fragmentation may mean that the system cannot be rebalanced |
3172 | * for high-order allocations in all zones. If twice the | |
3173 | * allocation size has been reclaimed and the zones are still | |
3174 | * not balanced then recheck the watermarks at order-0 to | |
3175 | * prevent kswapd reclaiming excessively. Assume that a | |
3176 | * process requested a high-order can direct reclaim/compact. | |
1da177e4 | 3177 | */ |
b8e83b94 MG |
3178 | if (order && sc.nr_reclaimed >= 2UL << order) |
3179 | order = sc.order = 0; | |
8357376d | 3180 | |
b8e83b94 MG |
3181 | /* Check if kswapd should be suspending */ |
3182 | if (try_to_freeze() || kthread_should_stop()) | |
3183 | break; | |
8357376d | 3184 | |
2ab44f43 MG |
3185 | /* |
3186 | * Compact if necessary and kswapd is reclaiming at least the | |
3187 | * high watermark number of pages as requsted | |
3188 | */ | |
3189 | if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted) | |
3190 | compact_pgdat(pgdat, order); | |
3191 | ||
73ce02e9 | 3192 | /* |
b8e83b94 MG |
3193 | * Raise priority if scanning rate is too low or there was no |
3194 | * progress in reclaiming pages | |
73ce02e9 | 3195 | */ |
b8e83b94 MG |
3196 | if (raise_priority || !sc.nr_reclaimed) |
3197 | sc.priority--; | |
9aa41348 | 3198 | } while (sc.priority >= 1 && |
b8e83b94 | 3199 | !pgdat_balanced(pgdat, order, *classzone_idx)); |
1da177e4 | 3200 | |
b8e83b94 | 3201 | out: |
0abdee2b | 3202 | /* |
5515061d | 3203 | * Return the order we were reclaiming at so prepare_kswapd_sleep() |
0abdee2b MG |
3204 | * makes a decision on the order we were last reclaiming at. However, |
3205 | * if another caller entered the allocator slow path while kswapd | |
3206 | * was awake, order will remain at the higher level | |
3207 | */ | |
dc83edd9 | 3208 | *classzone_idx = end_zone; |
0abdee2b | 3209 | return order; |
1da177e4 LT |
3210 | } |
3211 | ||
dc83edd9 | 3212 | static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx) |
f0bc0a60 KM |
3213 | { |
3214 | long remaining = 0; | |
3215 | DEFINE_WAIT(wait); | |
3216 | ||
3217 | if (freezing(current) || kthread_should_stop()) | |
3218 | return; | |
3219 | ||
3220 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | |
3221 | ||
3222 | /* Try to sleep for a short interval */ | |
5515061d | 3223 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
f0bc0a60 KM |
3224 | remaining = schedule_timeout(HZ/10); |
3225 | finish_wait(&pgdat->kswapd_wait, &wait); | |
3226 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | |
3227 | } | |
3228 | ||
3229 | /* | |
3230 | * After a short sleep, check if it was a premature sleep. If not, then | |
3231 | * go fully to sleep until explicitly woken up. | |
3232 | */ | |
5515061d | 3233 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
f0bc0a60 KM |
3234 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
3235 | ||
3236 | /* | |
3237 | * vmstat counters are not perfectly accurate and the estimated | |
3238 | * value for counters such as NR_FREE_PAGES can deviate from the | |
3239 | * true value by nr_online_cpus * threshold. To avoid the zone | |
3240 | * watermarks being breached while under pressure, we reduce the | |
3241 | * per-cpu vmstat threshold while kswapd is awake and restore | |
3242 | * them before going back to sleep. | |
3243 | */ | |
3244 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); | |
1c7e7f6c | 3245 | |
62997027 MG |
3246 | /* |
3247 | * Compaction records what page blocks it recently failed to | |
3248 | * isolate pages from and skips them in the future scanning. | |
3249 | * When kswapd is going to sleep, it is reasonable to assume | |
3250 | * that pages and compaction may succeed so reset the cache. | |
3251 | */ | |
3252 | reset_isolation_suitable(pgdat); | |
3253 | ||
1c7e7f6c AK |
3254 | if (!kthread_should_stop()) |
3255 | schedule(); | |
3256 | ||
f0bc0a60 KM |
3257 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
3258 | } else { | |
3259 | if (remaining) | |
3260 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); | |
3261 | else | |
3262 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); | |
3263 | } | |
3264 | finish_wait(&pgdat->kswapd_wait, &wait); | |
3265 | } | |
3266 | ||
1da177e4 LT |
3267 | /* |
3268 | * The background pageout daemon, started as a kernel thread | |
4f98a2fe | 3269 | * from the init process. |
1da177e4 LT |
3270 | * |
3271 | * This basically trickles out pages so that we have _some_ | |
3272 | * free memory available even if there is no other activity | |
3273 | * that frees anything up. This is needed for things like routing | |
3274 | * etc, where we otherwise might have all activity going on in | |
3275 | * asynchronous contexts that cannot page things out. | |
3276 | * | |
3277 | * If there are applications that are active memory-allocators | |
3278 | * (most normal use), this basically shouldn't matter. | |
3279 | */ | |
3280 | static int kswapd(void *p) | |
3281 | { | |
215ddd66 | 3282 | unsigned long order, new_order; |
d2ebd0f6 | 3283 | unsigned balanced_order; |
215ddd66 | 3284 | int classzone_idx, new_classzone_idx; |
d2ebd0f6 | 3285 | int balanced_classzone_idx; |
1da177e4 LT |
3286 | pg_data_t *pgdat = (pg_data_t*)p; |
3287 | struct task_struct *tsk = current; | |
f0bc0a60 | 3288 | |
1da177e4 LT |
3289 | struct reclaim_state reclaim_state = { |
3290 | .reclaimed_slab = 0, | |
3291 | }; | |
a70f7302 | 3292 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
1da177e4 | 3293 | |
cf40bd16 NP |
3294 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
3295 | ||
174596a0 | 3296 | if (!cpumask_empty(cpumask)) |
c5f59f08 | 3297 | set_cpus_allowed_ptr(tsk, cpumask); |
1da177e4 LT |
3298 | current->reclaim_state = &reclaim_state; |
3299 | ||
3300 | /* | |
3301 | * Tell the memory management that we're a "memory allocator", | |
3302 | * and that if we need more memory we should get access to it | |
3303 | * regardless (see "__alloc_pages()"). "kswapd" should | |
3304 | * never get caught in the normal page freeing logic. | |
3305 | * | |
3306 | * (Kswapd normally doesn't need memory anyway, but sometimes | |
3307 | * you need a small amount of memory in order to be able to | |
3308 | * page out something else, and this flag essentially protects | |
3309 | * us from recursively trying to free more memory as we're | |
3310 | * trying to free the first piece of memory in the first place). | |
3311 | */ | |
930d9152 | 3312 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
83144186 | 3313 | set_freezable(); |
1da177e4 | 3314 | |
215ddd66 | 3315 | order = new_order = 0; |
d2ebd0f6 | 3316 | balanced_order = 0; |
215ddd66 | 3317 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; |
d2ebd0f6 | 3318 | balanced_classzone_idx = classzone_idx; |
1da177e4 | 3319 | for ( ; ; ) { |
6f6313d4 | 3320 | bool ret; |
3e1d1d28 | 3321 | |
215ddd66 MG |
3322 | /* |
3323 | * If the last balance_pgdat was unsuccessful it's unlikely a | |
3324 | * new request of a similar or harder type will succeed soon | |
3325 | * so consider going to sleep on the basis we reclaimed at | |
3326 | */ | |
d2ebd0f6 AS |
3327 | if (balanced_classzone_idx >= new_classzone_idx && |
3328 | balanced_order == new_order) { | |
215ddd66 MG |
3329 | new_order = pgdat->kswapd_max_order; |
3330 | new_classzone_idx = pgdat->classzone_idx; | |
3331 | pgdat->kswapd_max_order = 0; | |
3332 | pgdat->classzone_idx = pgdat->nr_zones - 1; | |
3333 | } | |
3334 | ||
99504748 | 3335 | if (order < new_order || classzone_idx > new_classzone_idx) { |
1da177e4 LT |
3336 | /* |
3337 | * Don't sleep if someone wants a larger 'order' | |
99504748 | 3338 | * allocation or has tigher zone constraints |
1da177e4 LT |
3339 | */ |
3340 | order = new_order; | |
99504748 | 3341 | classzone_idx = new_classzone_idx; |
1da177e4 | 3342 | } else { |
d2ebd0f6 AS |
3343 | kswapd_try_to_sleep(pgdat, balanced_order, |
3344 | balanced_classzone_idx); | |
1da177e4 | 3345 | order = pgdat->kswapd_max_order; |
99504748 | 3346 | classzone_idx = pgdat->classzone_idx; |
f0dfcde0 AS |
3347 | new_order = order; |
3348 | new_classzone_idx = classzone_idx; | |
4d40502e | 3349 | pgdat->kswapd_max_order = 0; |
215ddd66 | 3350 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
1da177e4 | 3351 | } |
1da177e4 | 3352 | |
8fe23e05 DR |
3353 | ret = try_to_freeze(); |
3354 | if (kthread_should_stop()) | |
3355 | break; | |
3356 | ||
3357 | /* | |
3358 | * We can speed up thawing tasks if we don't call balance_pgdat | |
3359 | * after returning from the refrigerator | |
3360 | */ | |
33906bc5 MG |
3361 | if (!ret) { |
3362 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); | |
d2ebd0f6 AS |
3363 | balanced_classzone_idx = classzone_idx; |
3364 | balanced_order = balance_pgdat(pgdat, order, | |
3365 | &balanced_classzone_idx); | |
33906bc5 | 3366 | } |
1da177e4 | 3367 | } |
b0a8cc58 | 3368 | |
71abdc15 | 3369 | tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD); |
b0a8cc58 | 3370 | current->reclaim_state = NULL; |
71abdc15 JW |
3371 | lockdep_clear_current_reclaim_state(); |
3372 | ||
1da177e4 LT |
3373 | return 0; |
3374 | } | |
3375 | ||
3376 | /* | |
3377 | * A zone is low on free memory, so wake its kswapd task to service it. | |
3378 | */ | |
99504748 | 3379 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx) |
1da177e4 LT |
3380 | { |
3381 | pg_data_t *pgdat; | |
3382 | ||
f3fe6512 | 3383 | if (!populated_zone(zone)) |
1da177e4 LT |
3384 | return; |
3385 | ||
88f5acf8 | 3386 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
1da177e4 | 3387 | return; |
88f5acf8 | 3388 | pgdat = zone->zone_pgdat; |
99504748 | 3389 | if (pgdat->kswapd_max_order < order) { |
1da177e4 | 3390 | pgdat->kswapd_max_order = order; |
99504748 MG |
3391 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); |
3392 | } | |
8d0986e2 | 3393 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
1da177e4 | 3394 | return; |
892f795d | 3395 | if (zone_balanced(zone, order, 0, 0)) |
88f5acf8 MG |
3396 | return; |
3397 | ||
3398 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); | |
8d0986e2 | 3399 | wake_up_interruptible(&pgdat->kswapd_wait); |
1da177e4 LT |
3400 | } |
3401 | ||
c6f37f12 | 3402 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3403 | /* |
7b51755c | 3404 | * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of |
d6277db4 RW |
3405 | * freed pages. |
3406 | * | |
3407 | * Rather than trying to age LRUs the aim is to preserve the overall | |
3408 | * LRU order by reclaiming preferentially | |
3409 | * inactive > active > active referenced > active mapped | |
1da177e4 | 3410 | */ |
7b51755c | 3411 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
1da177e4 | 3412 | { |
d6277db4 | 3413 | struct reclaim_state reclaim_state; |
d6277db4 | 3414 | struct scan_control sc = { |
ee814fe2 | 3415 | .nr_to_reclaim = nr_to_reclaim, |
7b51755c | 3416 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
ee814fe2 | 3417 | .priority = DEF_PRIORITY, |
d6277db4 | 3418 | .may_writepage = 1, |
ee814fe2 JW |
3419 | .may_unmap = 1, |
3420 | .may_swap = 1, | |
7b51755c | 3421 | .hibernation_mode = 1, |
1da177e4 | 3422 | }; |
a09ed5e0 | 3423 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); |
7b51755c KM |
3424 | struct task_struct *p = current; |
3425 | unsigned long nr_reclaimed; | |
1da177e4 | 3426 | |
7b51755c KM |
3427 | p->flags |= PF_MEMALLOC; |
3428 | lockdep_set_current_reclaim_state(sc.gfp_mask); | |
3429 | reclaim_state.reclaimed_slab = 0; | |
3430 | p->reclaim_state = &reclaim_state; | |
d6277db4 | 3431 | |
3115cd91 | 3432 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
d979677c | 3433 | |
7b51755c KM |
3434 | p->reclaim_state = NULL; |
3435 | lockdep_clear_current_reclaim_state(); | |
3436 | p->flags &= ~PF_MEMALLOC; | |
d6277db4 | 3437 | |
7b51755c | 3438 | return nr_reclaimed; |
1da177e4 | 3439 | } |
c6f37f12 | 3440 | #endif /* CONFIG_HIBERNATION */ |
1da177e4 | 3441 | |
1da177e4 LT |
3442 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
3443 | not required for correctness. So if the last cpu in a node goes | |
3444 | away, we get changed to run anywhere: as the first one comes back, | |
3445 | restore their cpu bindings. */ | |
fcb35a9b GKH |
3446 | static int cpu_callback(struct notifier_block *nfb, unsigned long action, |
3447 | void *hcpu) | |
1da177e4 | 3448 | { |
58c0a4a7 | 3449 | int nid; |
1da177e4 | 3450 | |
8bb78442 | 3451 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
48fb2e24 | 3452 | for_each_node_state(nid, N_MEMORY) { |
c5f59f08 | 3453 | pg_data_t *pgdat = NODE_DATA(nid); |
a70f7302 RR |
3454 | const struct cpumask *mask; |
3455 | ||
3456 | mask = cpumask_of_node(pgdat->node_id); | |
c5f59f08 | 3457 | |
3e597945 | 3458 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
1da177e4 | 3459 | /* One of our CPUs online: restore mask */ |
c5f59f08 | 3460 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
1da177e4 LT |
3461 | } |
3462 | } | |
3463 | return NOTIFY_OK; | |
3464 | } | |
1da177e4 | 3465 | |
3218ae14 YG |
3466 | /* |
3467 | * This kswapd start function will be called by init and node-hot-add. | |
3468 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. | |
3469 | */ | |
3470 | int kswapd_run(int nid) | |
3471 | { | |
3472 | pg_data_t *pgdat = NODE_DATA(nid); | |
3473 | int ret = 0; | |
3474 | ||
3475 | if (pgdat->kswapd) | |
3476 | return 0; | |
3477 | ||
3478 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); | |
3479 | if (IS_ERR(pgdat->kswapd)) { | |
3480 | /* failure at boot is fatal */ | |
3481 | BUG_ON(system_state == SYSTEM_BOOTING); | |
d5dc0ad9 GS |
3482 | pr_err("Failed to start kswapd on node %d\n", nid); |
3483 | ret = PTR_ERR(pgdat->kswapd); | |
d72515b8 | 3484 | pgdat->kswapd = NULL; |
3218ae14 YG |
3485 | } |
3486 | return ret; | |
3487 | } | |
3488 | ||
8fe23e05 | 3489 | /* |
d8adde17 | 3490 | * Called by memory hotplug when all memory in a node is offlined. Caller must |
bfc8c901 | 3491 | * hold mem_hotplug_begin/end(). |
8fe23e05 DR |
3492 | */ |
3493 | void kswapd_stop(int nid) | |
3494 | { | |
3495 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; | |
3496 | ||
d8adde17 | 3497 | if (kswapd) { |
8fe23e05 | 3498 | kthread_stop(kswapd); |
d8adde17 JL |
3499 | NODE_DATA(nid)->kswapd = NULL; |
3500 | } | |
8fe23e05 DR |
3501 | } |
3502 | ||
1da177e4 LT |
3503 | static int __init kswapd_init(void) |
3504 | { | |
3218ae14 | 3505 | int nid; |
69e05944 | 3506 | |
1da177e4 | 3507 | swap_setup(); |
48fb2e24 | 3508 | for_each_node_state(nid, N_MEMORY) |
3218ae14 | 3509 | kswapd_run(nid); |
1da177e4 LT |
3510 | hotcpu_notifier(cpu_callback, 0); |
3511 | return 0; | |
3512 | } | |
3513 | ||
3514 | module_init(kswapd_init) | |
9eeff239 CL |
3515 | |
3516 | #ifdef CONFIG_NUMA | |
3517 | /* | |
3518 | * Zone reclaim mode | |
3519 | * | |
3520 | * If non-zero call zone_reclaim when the number of free pages falls below | |
3521 | * the watermarks. | |
9eeff239 CL |
3522 | */ |
3523 | int zone_reclaim_mode __read_mostly; | |
3524 | ||
1b2ffb78 | 3525 | #define RECLAIM_OFF 0 |
7d03431c | 3526 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
1b2ffb78 CL |
3527 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
3528 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ | |
3529 | ||
a92f7126 CL |
3530 | /* |
3531 | * Priority for ZONE_RECLAIM. This determines the fraction of pages | |
3532 | * of a node considered for each zone_reclaim. 4 scans 1/16th of | |
3533 | * a zone. | |
3534 | */ | |
3535 | #define ZONE_RECLAIM_PRIORITY 4 | |
3536 | ||
9614634f CL |
3537 | /* |
3538 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to | |
3539 | * occur. | |
3540 | */ | |
3541 | int sysctl_min_unmapped_ratio = 1; | |
3542 | ||
0ff38490 CL |
3543 | /* |
3544 | * If the number of slab pages in a zone grows beyond this percentage then | |
3545 | * slab reclaim needs to occur. | |
3546 | */ | |
3547 | int sysctl_min_slab_ratio = 5; | |
3548 | ||
90afa5de MG |
3549 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) |
3550 | { | |
3551 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); | |
3552 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + | |
3553 | zone_page_state(zone, NR_ACTIVE_FILE); | |
3554 | ||
3555 | /* | |
3556 | * It's possible for there to be more file mapped pages than | |
3557 | * accounted for by the pages on the file LRU lists because | |
3558 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED | |
3559 | */ | |
3560 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; | |
3561 | } | |
3562 | ||
3563 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ | |
3564 | static long zone_pagecache_reclaimable(struct zone *zone) | |
3565 | { | |
3566 | long nr_pagecache_reclaimable; | |
3567 | long delta = 0; | |
3568 | ||
3569 | /* | |
3570 | * If RECLAIM_SWAP is set, then all file pages are considered | |
3571 | * potentially reclaimable. Otherwise, we have to worry about | |
3572 | * pages like swapcache and zone_unmapped_file_pages() provides | |
3573 | * a better estimate | |
3574 | */ | |
3575 | if (zone_reclaim_mode & RECLAIM_SWAP) | |
3576 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); | |
3577 | else | |
3578 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); | |
3579 | ||
3580 | /* If we can't clean pages, remove dirty pages from consideration */ | |
3581 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) | |
3582 | delta += zone_page_state(zone, NR_FILE_DIRTY); | |
3583 | ||
3584 | /* Watch for any possible underflows due to delta */ | |
3585 | if (unlikely(delta > nr_pagecache_reclaimable)) | |
3586 | delta = nr_pagecache_reclaimable; | |
3587 | ||
3588 | return nr_pagecache_reclaimable - delta; | |
3589 | } | |
3590 | ||
9eeff239 CL |
3591 | /* |
3592 | * Try to free up some pages from this zone through reclaim. | |
3593 | */ | |
179e9639 | 3594 | static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
9eeff239 | 3595 | { |
7fb2d46d | 3596 | /* Minimum pages needed in order to stay on node */ |
69e05944 | 3597 | const unsigned long nr_pages = 1 << order; |
9eeff239 CL |
3598 | struct task_struct *p = current; |
3599 | struct reclaim_state reclaim_state; | |
179e9639 | 3600 | struct scan_control sc = { |
62b726c1 | 3601 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
21caf2fc | 3602 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
bd2f6199 | 3603 | .order = order, |
9e3b2f8c | 3604 | .priority = ZONE_RECLAIM_PRIORITY, |
ee814fe2 JW |
3605 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
3606 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP), | |
3607 | .may_swap = 1, | |
179e9639 | 3608 | }; |
a09ed5e0 YH |
3609 | struct shrink_control shrink = { |
3610 | .gfp_mask = sc.gfp_mask, | |
3611 | }; | |
15748048 | 3612 | unsigned long nr_slab_pages0, nr_slab_pages1; |
9eeff239 | 3613 | |
9eeff239 | 3614 | cond_resched(); |
d4f7796e CL |
3615 | /* |
3616 | * We need to be able to allocate from the reserves for RECLAIM_SWAP | |
3617 | * and we also need to be able to write out pages for RECLAIM_WRITE | |
3618 | * and RECLAIM_SWAP. | |
3619 | */ | |
3620 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; | |
76ca542d | 3621 | lockdep_set_current_reclaim_state(gfp_mask); |
9eeff239 CL |
3622 | reclaim_state.reclaimed_slab = 0; |
3623 | p->reclaim_state = &reclaim_state; | |
c84db23c | 3624 | |
90afa5de | 3625 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { |
0ff38490 CL |
3626 | /* |
3627 | * Free memory by calling shrink zone with increasing | |
3628 | * priorities until we have enough memory freed. | |
3629 | */ | |
0ff38490 | 3630 | do { |
9e3b2f8c KK |
3631 | shrink_zone(zone, &sc); |
3632 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); | |
0ff38490 | 3633 | } |
c84db23c | 3634 | |
15748048 KM |
3635 | nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
3636 | if (nr_slab_pages0 > zone->min_slab_pages) { | |
2a16e3f4 | 3637 | /* |
7fb2d46d | 3638 | * shrink_slab() does not currently allow us to determine how |
0ff38490 CL |
3639 | * many pages were freed in this zone. So we take the current |
3640 | * number of slab pages and shake the slab until it is reduced | |
3641 | * by the same nr_pages that we used for reclaiming unmapped | |
3642 | * pages. | |
2a16e3f4 | 3643 | */ |
0ce3d744 DC |
3644 | nodes_clear(shrink.nodes_to_scan); |
3645 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); | |
4dc4b3d9 KM |
3646 | for (;;) { |
3647 | unsigned long lru_pages = zone_reclaimable_pages(zone); | |
3648 | ||
3649 | /* No reclaimable slab or very low memory pressure */ | |
1495f230 | 3650 | if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages)) |
4dc4b3d9 KM |
3651 | break; |
3652 | ||
3653 | /* Freed enough memory */ | |
3654 | nr_slab_pages1 = zone_page_state(zone, | |
3655 | NR_SLAB_RECLAIMABLE); | |
3656 | if (nr_slab_pages1 + nr_pages <= nr_slab_pages0) | |
3657 | break; | |
3658 | } | |
83e33a47 CL |
3659 | |
3660 | /* | |
3661 | * Update nr_reclaimed by the number of slab pages we | |
3662 | * reclaimed from this zone. | |
3663 | */ | |
15748048 KM |
3664 | nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
3665 | if (nr_slab_pages1 < nr_slab_pages0) | |
3666 | sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1; | |
2a16e3f4 CL |
3667 | } |
3668 | ||
9eeff239 | 3669 | p->reclaim_state = NULL; |
d4f7796e | 3670 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
76ca542d | 3671 | lockdep_clear_current_reclaim_state(); |
a79311c1 | 3672 | return sc.nr_reclaimed >= nr_pages; |
9eeff239 | 3673 | } |
179e9639 AM |
3674 | |
3675 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | |
3676 | { | |
179e9639 | 3677 | int node_id; |
d773ed6b | 3678 | int ret; |
179e9639 AM |
3679 | |
3680 | /* | |
0ff38490 CL |
3681 | * Zone reclaim reclaims unmapped file backed pages and |
3682 | * slab pages if we are over the defined limits. | |
34aa1330 | 3683 | * |
9614634f CL |
3684 | * A small portion of unmapped file backed pages is needed for |
3685 | * file I/O otherwise pages read by file I/O will be immediately | |
3686 | * thrown out if the zone is overallocated. So we do not reclaim | |
3687 | * if less than a specified percentage of the zone is used by | |
3688 | * unmapped file backed pages. | |
179e9639 | 3689 | */ |
90afa5de MG |
3690 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && |
3691 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) | |
fa5e084e | 3692 | return ZONE_RECLAIM_FULL; |
179e9639 | 3693 | |
6e543d57 | 3694 | if (!zone_reclaimable(zone)) |
fa5e084e | 3695 | return ZONE_RECLAIM_FULL; |
d773ed6b | 3696 | |
179e9639 | 3697 | /* |
d773ed6b | 3698 | * Do not scan if the allocation should not be delayed. |
179e9639 | 3699 | */ |
d773ed6b | 3700 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
fa5e084e | 3701 | return ZONE_RECLAIM_NOSCAN; |
179e9639 AM |
3702 | |
3703 | /* | |
3704 | * Only run zone reclaim on the local zone or on zones that do not | |
3705 | * have associated processors. This will favor the local processor | |
3706 | * over remote processors and spread off node memory allocations | |
3707 | * as wide as possible. | |
3708 | */ | |
89fa3024 | 3709 | node_id = zone_to_nid(zone); |
37c0708d | 3710 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
fa5e084e | 3711 | return ZONE_RECLAIM_NOSCAN; |
d773ed6b DR |
3712 | |
3713 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) | |
fa5e084e MG |
3714 | return ZONE_RECLAIM_NOSCAN; |
3715 | ||
d773ed6b DR |
3716 | ret = __zone_reclaim(zone, gfp_mask, order); |
3717 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); | |
3718 | ||
24cf7251 MG |
3719 | if (!ret) |
3720 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); | |
3721 | ||
d773ed6b | 3722 | return ret; |
179e9639 | 3723 | } |
9eeff239 | 3724 | #endif |
894bc310 | 3725 | |
894bc310 LS |
3726 | /* |
3727 | * page_evictable - test whether a page is evictable | |
3728 | * @page: the page to test | |
894bc310 LS |
3729 | * |
3730 | * Test whether page is evictable--i.e., should be placed on active/inactive | |
39b5f29a | 3731 | * lists vs unevictable list. |
894bc310 LS |
3732 | * |
3733 | * Reasons page might not be evictable: | |
ba9ddf49 | 3734 | * (1) page's mapping marked unevictable |
b291f000 | 3735 | * (2) page is part of an mlocked VMA |
ba9ddf49 | 3736 | * |
894bc310 | 3737 | */ |
39b5f29a | 3738 | int page_evictable(struct page *page) |
894bc310 | 3739 | { |
39b5f29a | 3740 | return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); |
894bc310 | 3741 | } |
89e004ea | 3742 | |
85046579 | 3743 | #ifdef CONFIG_SHMEM |
89e004ea | 3744 | /** |
24513264 HD |
3745 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list |
3746 | * @pages: array of pages to check | |
3747 | * @nr_pages: number of pages to check | |
89e004ea | 3748 | * |
24513264 | 3749 | * Checks pages for evictability and moves them to the appropriate lru list. |
85046579 HD |
3750 | * |
3751 | * This function is only used for SysV IPC SHM_UNLOCK. | |
89e004ea | 3752 | */ |
24513264 | 3753 | void check_move_unevictable_pages(struct page **pages, int nr_pages) |
89e004ea | 3754 | { |
925b7673 | 3755 | struct lruvec *lruvec; |
24513264 HD |
3756 | struct zone *zone = NULL; |
3757 | int pgscanned = 0; | |
3758 | int pgrescued = 0; | |
3759 | int i; | |
89e004ea | 3760 | |
24513264 HD |
3761 | for (i = 0; i < nr_pages; i++) { |
3762 | struct page *page = pages[i]; | |
3763 | struct zone *pagezone; | |
89e004ea | 3764 | |
24513264 HD |
3765 | pgscanned++; |
3766 | pagezone = page_zone(page); | |
3767 | if (pagezone != zone) { | |
3768 | if (zone) | |
3769 | spin_unlock_irq(&zone->lru_lock); | |
3770 | zone = pagezone; | |
3771 | spin_lock_irq(&zone->lru_lock); | |
3772 | } | |
fa9add64 | 3773 | lruvec = mem_cgroup_page_lruvec(page, zone); |
89e004ea | 3774 | |
24513264 HD |
3775 | if (!PageLRU(page) || !PageUnevictable(page)) |
3776 | continue; | |
89e004ea | 3777 | |
39b5f29a | 3778 | if (page_evictable(page)) { |
24513264 HD |
3779 | enum lru_list lru = page_lru_base_type(page); |
3780 | ||
309381fe | 3781 | VM_BUG_ON_PAGE(PageActive(page), page); |
24513264 | 3782 | ClearPageUnevictable(page); |
fa9add64 HD |
3783 | del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE); |
3784 | add_page_to_lru_list(page, lruvec, lru); | |
24513264 | 3785 | pgrescued++; |
89e004ea | 3786 | } |
24513264 | 3787 | } |
89e004ea | 3788 | |
24513264 HD |
3789 | if (zone) { |
3790 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
3791 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); | |
3792 | spin_unlock_irq(&zone->lru_lock); | |
89e004ea | 3793 | } |
89e004ea | 3794 | } |
85046579 | 3795 | #endif /* CONFIG_SHMEM */ |
af936a16 | 3796 | |
264e56d8 | 3797 | static void warn_scan_unevictable_pages(void) |
af936a16 | 3798 | { |
264e56d8 | 3799 | printk_once(KERN_WARNING |
25bd91bd | 3800 | "%s: The scan_unevictable_pages sysctl/node-interface has been " |
264e56d8 | 3801 | "disabled for lack of a legitimate use case. If you have " |
25bd91bd KM |
3802 | "one, please send an email to linux-mm@kvack.org.\n", |
3803 | current->comm); | |
af936a16 LS |
3804 | } |
3805 | ||
3806 | /* | |
3807 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of | |
3808 | * all nodes' unevictable lists for evictable pages | |
3809 | */ | |
3810 | unsigned long scan_unevictable_pages; | |
3811 | ||
3812 | int scan_unevictable_handler(struct ctl_table *table, int write, | |
8d65af78 | 3813 | void __user *buffer, |
af936a16 LS |
3814 | size_t *length, loff_t *ppos) |
3815 | { | |
264e56d8 | 3816 | warn_scan_unevictable_pages(); |
8d65af78 | 3817 | proc_doulongvec_minmax(table, write, buffer, length, ppos); |
af936a16 LS |
3818 | scan_unevictable_pages = 0; |
3819 | return 0; | |
3820 | } | |
3821 | ||
e4455abb | 3822 | #ifdef CONFIG_NUMA |
af936a16 LS |
3823 | /* |
3824 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of | |
3825 | * a specified node's per zone unevictable lists for evictable pages. | |
3826 | */ | |
3827 | ||
10fbcf4c KS |
3828 | static ssize_t read_scan_unevictable_node(struct device *dev, |
3829 | struct device_attribute *attr, | |
af936a16 LS |
3830 | char *buf) |
3831 | { | |
264e56d8 | 3832 | warn_scan_unevictable_pages(); |
af936a16 LS |
3833 | return sprintf(buf, "0\n"); /* always zero; should fit... */ |
3834 | } | |
3835 | ||
10fbcf4c KS |
3836 | static ssize_t write_scan_unevictable_node(struct device *dev, |
3837 | struct device_attribute *attr, | |
af936a16 LS |
3838 | const char *buf, size_t count) |
3839 | { | |
264e56d8 | 3840 | warn_scan_unevictable_pages(); |
af936a16 LS |
3841 | return 1; |
3842 | } | |
3843 | ||
3844 | ||
10fbcf4c | 3845 | static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, |
af936a16 LS |
3846 | read_scan_unevictable_node, |
3847 | write_scan_unevictable_node); | |
3848 | ||
3849 | int scan_unevictable_register_node(struct node *node) | |
3850 | { | |
10fbcf4c | 3851 | return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages); |
af936a16 LS |
3852 | } |
3853 | ||
3854 | void scan_unevictable_unregister_node(struct node *node) | |
3855 | { | |
10fbcf4c | 3856 | device_remove_file(&node->dev, &dev_attr_scan_unevictable_pages); |
af936a16 | 3857 | } |
e4455abb | 3858 | #endif |