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 | ||
241994ed JW |
94 | /* Can cgroups be reclaimed below their normal consumption range? */ |
95 | unsigned int may_thrash:1; | |
96 | ||
ee814fe2 JW |
97 | unsigned int hibernation_mode:1; |
98 | ||
99 | /* One of the zones is ready for compaction */ | |
100 | unsigned int compaction_ready:1; | |
101 | ||
102 | /* Incremented by the number of inactive pages that were scanned */ | |
103 | unsigned long nr_scanned; | |
104 | ||
105 | /* Number of pages freed so far during a call to shrink_zones() */ | |
106 | unsigned long nr_reclaimed; | |
1da177e4 LT |
107 | }; |
108 | ||
1da177e4 LT |
109 | #ifdef ARCH_HAS_PREFETCH |
110 | #define prefetch_prev_lru_page(_page, _base, _field) \ | |
111 | do { \ | |
112 | if ((_page)->lru.prev != _base) { \ | |
113 | struct page *prev; \ | |
114 | \ | |
115 | prev = lru_to_page(&(_page->lru)); \ | |
116 | prefetch(&prev->_field); \ | |
117 | } \ | |
118 | } while (0) | |
119 | #else | |
120 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) | |
121 | #endif | |
122 | ||
123 | #ifdef ARCH_HAS_PREFETCHW | |
124 | #define prefetchw_prev_lru_page(_page, _base, _field) \ | |
125 | do { \ | |
126 | if ((_page)->lru.prev != _base) { \ | |
127 | struct page *prev; \ | |
128 | \ | |
129 | prev = lru_to_page(&(_page->lru)); \ | |
130 | prefetchw(&prev->_field); \ | |
131 | } \ | |
132 | } while (0) | |
133 | #else | |
134 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) | |
135 | #endif | |
136 | ||
137 | /* | |
138 | * From 0 .. 100. Higher means more swappy. | |
139 | */ | |
140 | int vm_swappiness = 60; | |
d0480be4 WSH |
141 | /* |
142 | * The total number of pages which are beyond the high watermark within all | |
143 | * zones. | |
144 | */ | |
145 | unsigned long vm_total_pages; | |
1da177e4 LT |
146 | |
147 | static LIST_HEAD(shrinker_list); | |
148 | static DECLARE_RWSEM(shrinker_rwsem); | |
149 | ||
c255a458 | 150 | #ifdef CONFIG_MEMCG |
89b5fae5 JW |
151 | static bool global_reclaim(struct scan_control *sc) |
152 | { | |
f16015fb | 153 | return !sc->target_mem_cgroup; |
89b5fae5 | 154 | } |
97c9341f TH |
155 | |
156 | /** | |
157 | * sane_reclaim - is the usual dirty throttling mechanism operational? | |
158 | * @sc: scan_control in question | |
159 | * | |
160 | * The normal page dirty throttling mechanism in balance_dirty_pages() is | |
161 | * completely broken with the legacy memcg and direct stalling in | |
162 | * shrink_page_list() is used for throttling instead, which lacks all the | |
163 | * niceties such as fairness, adaptive pausing, bandwidth proportional | |
164 | * allocation and configurability. | |
165 | * | |
166 | * This function tests whether the vmscan currently in progress can assume | |
167 | * that the normal dirty throttling mechanism is operational. | |
168 | */ | |
169 | static bool sane_reclaim(struct scan_control *sc) | |
170 | { | |
171 | struct mem_cgroup *memcg = sc->target_mem_cgroup; | |
172 | ||
173 | if (!memcg) | |
174 | return true; | |
175 | #ifdef CONFIG_CGROUP_WRITEBACK | |
69234ace | 176 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
97c9341f TH |
177 | return true; |
178 | #endif | |
179 | return false; | |
180 | } | |
91a45470 | 181 | #else |
89b5fae5 JW |
182 | static bool global_reclaim(struct scan_control *sc) |
183 | { | |
184 | return true; | |
185 | } | |
97c9341f TH |
186 | |
187 | static bool sane_reclaim(struct scan_control *sc) | |
188 | { | |
189 | return true; | |
190 | } | |
91a45470 KH |
191 | #endif |
192 | ||
a1c3bfb2 | 193 | static unsigned long zone_reclaimable_pages(struct zone *zone) |
6e543d57 | 194 | { |
d031a157 | 195 | unsigned long nr; |
6e543d57 LD |
196 | |
197 | nr = zone_page_state(zone, NR_ACTIVE_FILE) + | |
9f6c399d MH |
198 | zone_page_state(zone, NR_INACTIVE_FILE) + |
199 | zone_page_state(zone, NR_ISOLATED_FILE); | |
6e543d57 LD |
200 | |
201 | if (get_nr_swap_pages() > 0) | |
202 | nr += zone_page_state(zone, NR_ACTIVE_ANON) + | |
9f6c399d MH |
203 | zone_page_state(zone, NR_INACTIVE_ANON) + |
204 | zone_page_state(zone, NR_ISOLATED_ANON); | |
6e543d57 LD |
205 | |
206 | return nr; | |
207 | } | |
208 | ||
209 | bool zone_reclaimable(struct zone *zone) | |
210 | { | |
0d5d823a MG |
211 | return zone_page_state(zone, NR_PAGES_SCANNED) < |
212 | zone_reclaimable_pages(zone) * 6; | |
6e543d57 LD |
213 | } |
214 | ||
4d7dcca2 | 215 | static unsigned long get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
c9f299d9 | 216 | { |
c3c787e8 | 217 | if (!mem_cgroup_disabled()) |
4d7dcca2 | 218 | return mem_cgroup_get_lru_size(lruvec, lru); |
a3d8e054 | 219 | |
074291fe | 220 | return zone_page_state(lruvec_zone(lruvec), NR_LRU_BASE + lru); |
c9f299d9 KM |
221 | } |
222 | ||
1da177e4 | 223 | /* |
1d3d4437 | 224 | * Add a shrinker callback to be called from the vm. |
1da177e4 | 225 | */ |
1d3d4437 | 226 | int register_shrinker(struct shrinker *shrinker) |
1da177e4 | 227 | { |
1d3d4437 GC |
228 | size_t size = sizeof(*shrinker->nr_deferred); |
229 | ||
230 | /* | |
231 | * If we only have one possible node in the system anyway, save | |
232 | * ourselves the trouble and disable NUMA aware behavior. This way we | |
233 | * will save memory and some small loop time later. | |
234 | */ | |
235 | if (nr_node_ids == 1) | |
236 | shrinker->flags &= ~SHRINKER_NUMA_AWARE; | |
237 | ||
238 | if (shrinker->flags & SHRINKER_NUMA_AWARE) | |
239 | size *= nr_node_ids; | |
240 | ||
241 | shrinker->nr_deferred = kzalloc(size, GFP_KERNEL); | |
242 | if (!shrinker->nr_deferred) | |
243 | return -ENOMEM; | |
244 | ||
8e1f936b RR |
245 | down_write(&shrinker_rwsem); |
246 | list_add_tail(&shrinker->list, &shrinker_list); | |
247 | up_write(&shrinker_rwsem); | |
1d3d4437 | 248 | return 0; |
1da177e4 | 249 | } |
8e1f936b | 250 | EXPORT_SYMBOL(register_shrinker); |
1da177e4 LT |
251 | |
252 | /* | |
253 | * Remove one | |
254 | */ | |
8e1f936b | 255 | void unregister_shrinker(struct shrinker *shrinker) |
1da177e4 LT |
256 | { |
257 | down_write(&shrinker_rwsem); | |
258 | list_del(&shrinker->list); | |
259 | up_write(&shrinker_rwsem); | |
ae393321 | 260 | kfree(shrinker->nr_deferred); |
1da177e4 | 261 | } |
8e1f936b | 262 | EXPORT_SYMBOL(unregister_shrinker); |
1da177e4 LT |
263 | |
264 | #define SHRINK_BATCH 128 | |
1d3d4437 | 265 | |
cb731d6c VD |
266 | static unsigned long do_shrink_slab(struct shrink_control *shrinkctl, |
267 | struct shrinker *shrinker, | |
268 | unsigned long nr_scanned, | |
269 | unsigned long nr_eligible) | |
1d3d4437 GC |
270 | { |
271 | unsigned long freed = 0; | |
272 | unsigned long long delta; | |
273 | long total_scan; | |
d5bc5fd3 | 274 | long freeable; |
1d3d4437 GC |
275 | long nr; |
276 | long new_nr; | |
277 | int nid = shrinkctl->nid; | |
278 | long batch_size = shrinker->batch ? shrinker->batch | |
279 | : SHRINK_BATCH; | |
280 | ||
d5bc5fd3 VD |
281 | freeable = shrinker->count_objects(shrinker, shrinkctl); |
282 | if (freeable == 0) | |
1d3d4437 GC |
283 | return 0; |
284 | ||
285 | /* | |
286 | * copy the current shrinker scan count into a local variable | |
287 | * and zero it so that other concurrent shrinker invocations | |
288 | * don't also do this scanning work. | |
289 | */ | |
290 | nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0); | |
291 | ||
292 | total_scan = nr; | |
6b4f7799 | 293 | delta = (4 * nr_scanned) / shrinker->seeks; |
d5bc5fd3 | 294 | delta *= freeable; |
6b4f7799 | 295 | do_div(delta, nr_eligible + 1); |
1d3d4437 GC |
296 | total_scan += delta; |
297 | if (total_scan < 0) { | |
8612c663 | 298 | pr_err("shrink_slab: %pF negative objects to delete nr=%ld\n", |
a0b02131 | 299 | shrinker->scan_objects, total_scan); |
d5bc5fd3 | 300 | total_scan = freeable; |
1d3d4437 GC |
301 | } |
302 | ||
303 | /* | |
304 | * We need to avoid excessive windup on filesystem shrinkers | |
305 | * due to large numbers of GFP_NOFS allocations causing the | |
306 | * shrinkers to return -1 all the time. This results in a large | |
307 | * nr being built up so when a shrink that can do some work | |
308 | * comes along it empties the entire cache due to nr >>> | |
d5bc5fd3 | 309 | * freeable. This is bad for sustaining a working set in |
1d3d4437 GC |
310 | * memory. |
311 | * | |
312 | * Hence only allow the shrinker to scan the entire cache when | |
313 | * a large delta change is calculated directly. | |
314 | */ | |
d5bc5fd3 VD |
315 | if (delta < freeable / 4) |
316 | total_scan = min(total_scan, freeable / 2); | |
1d3d4437 GC |
317 | |
318 | /* | |
319 | * Avoid risking looping forever due to too large nr value: | |
320 | * never try to free more than twice the estimate number of | |
321 | * freeable entries. | |
322 | */ | |
d5bc5fd3 VD |
323 | if (total_scan > freeable * 2) |
324 | total_scan = freeable * 2; | |
1d3d4437 GC |
325 | |
326 | trace_mm_shrink_slab_start(shrinker, shrinkctl, nr, | |
6b4f7799 JW |
327 | nr_scanned, nr_eligible, |
328 | freeable, delta, total_scan); | |
1d3d4437 | 329 | |
0b1fb40a VD |
330 | /* |
331 | * Normally, we should not scan less than batch_size objects in one | |
332 | * pass to avoid too frequent shrinker calls, but if the slab has less | |
333 | * than batch_size objects in total and we are really tight on memory, | |
334 | * we will try to reclaim all available objects, otherwise we can end | |
335 | * up failing allocations although there are plenty of reclaimable | |
336 | * objects spread over several slabs with usage less than the | |
337 | * batch_size. | |
338 | * | |
339 | * We detect the "tight on memory" situations by looking at the total | |
340 | * number of objects we want to scan (total_scan). If it is greater | |
d5bc5fd3 | 341 | * than the total number of objects on slab (freeable), we must be |
0b1fb40a VD |
342 | * scanning at high prio and therefore should try to reclaim as much as |
343 | * possible. | |
344 | */ | |
345 | while (total_scan >= batch_size || | |
d5bc5fd3 | 346 | total_scan >= freeable) { |
a0b02131 | 347 | unsigned long ret; |
0b1fb40a | 348 | unsigned long nr_to_scan = min(batch_size, total_scan); |
1d3d4437 | 349 | |
0b1fb40a | 350 | shrinkctl->nr_to_scan = nr_to_scan; |
a0b02131 DC |
351 | ret = shrinker->scan_objects(shrinker, shrinkctl); |
352 | if (ret == SHRINK_STOP) | |
353 | break; | |
354 | freed += ret; | |
1d3d4437 | 355 | |
0b1fb40a VD |
356 | count_vm_events(SLABS_SCANNED, nr_to_scan); |
357 | total_scan -= nr_to_scan; | |
1d3d4437 GC |
358 | |
359 | cond_resched(); | |
360 | } | |
361 | ||
362 | /* | |
363 | * move the unused scan count back into the shrinker in a | |
364 | * manner that handles concurrent updates. If we exhausted the | |
365 | * scan, there is no need to do an update. | |
366 | */ | |
367 | if (total_scan > 0) | |
368 | new_nr = atomic_long_add_return(total_scan, | |
369 | &shrinker->nr_deferred[nid]); | |
370 | else | |
371 | new_nr = atomic_long_read(&shrinker->nr_deferred[nid]); | |
372 | ||
df9024a8 | 373 | trace_mm_shrink_slab_end(shrinker, nid, freed, nr, new_nr, total_scan); |
1d3d4437 | 374 | return freed; |
1495f230 YH |
375 | } |
376 | ||
6b4f7799 | 377 | /** |
cb731d6c | 378 | * shrink_slab - shrink slab caches |
6b4f7799 JW |
379 | * @gfp_mask: allocation context |
380 | * @nid: node whose slab caches to target | |
cb731d6c | 381 | * @memcg: memory cgroup whose slab caches to target |
6b4f7799 JW |
382 | * @nr_scanned: pressure numerator |
383 | * @nr_eligible: pressure denominator | |
1da177e4 | 384 | * |
6b4f7799 | 385 | * Call the shrink functions to age shrinkable caches. |
1da177e4 | 386 | * |
6b4f7799 JW |
387 | * @nid is passed along to shrinkers with SHRINKER_NUMA_AWARE set, |
388 | * unaware shrinkers will receive a node id of 0 instead. | |
1da177e4 | 389 | * |
cb731d6c VD |
390 | * @memcg specifies the memory cgroup to target. If it is not NULL, |
391 | * only shrinkers with SHRINKER_MEMCG_AWARE set will be called to scan | |
392 | * objects from the memory cgroup specified. Otherwise all shrinkers | |
393 | * are called, and memcg aware shrinkers are supposed to scan the | |
394 | * global list then. | |
395 | * | |
6b4f7799 JW |
396 | * @nr_scanned and @nr_eligible form a ratio that indicate how much of |
397 | * the available objects should be scanned. Page reclaim for example | |
398 | * passes the number of pages scanned and the number of pages on the | |
399 | * LRU lists that it considered on @nid, plus a bias in @nr_scanned | |
400 | * when it encountered mapped pages. The ratio is further biased by | |
401 | * the ->seeks setting of the shrink function, which indicates the | |
402 | * cost to recreate an object relative to that of an LRU page. | |
b15e0905 | 403 | * |
6b4f7799 | 404 | * Returns the number of reclaimed slab objects. |
1da177e4 | 405 | */ |
cb731d6c VD |
406 | static unsigned long shrink_slab(gfp_t gfp_mask, int nid, |
407 | struct mem_cgroup *memcg, | |
408 | unsigned long nr_scanned, | |
409 | unsigned long nr_eligible) | |
1da177e4 LT |
410 | { |
411 | struct shrinker *shrinker; | |
24f7c6b9 | 412 | unsigned long freed = 0; |
1da177e4 | 413 | |
567e9ab2 | 414 | if (memcg && !memcg_kmem_online(memcg)) |
cb731d6c VD |
415 | return 0; |
416 | ||
6b4f7799 JW |
417 | if (nr_scanned == 0) |
418 | nr_scanned = SWAP_CLUSTER_MAX; | |
1da177e4 | 419 | |
f06590bd | 420 | if (!down_read_trylock(&shrinker_rwsem)) { |
24f7c6b9 DC |
421 | /* |
422 | * If we would return 0, our callers would understand that we | |
423 | * have nothing else to shrink and give up trying. By returning | |
424 | * 1 we keep it going and assume we'll be able to shrink next | |
425 | * time. | |
426 | */ | |
427 | freed = 1; | |
f06590bd MK |
428 | goto out; |
429 | } | |
1da177e4 LT |
430 | |
431 | list_for_each_entry(shrinker, &shrinker_list, list) { | |
6b4f7799 JW |
432 | struct shrink_control sc = { |
433 | .gfp_mask = gfp_mask, | |
434 | .nid = nid, | |
cb731d6c | 435 | .memcg = memcg, |
6b4f7799 | 436 | }; |
ec97097b | 437 | |
cb731d6c VD |
438 | if (memcg && !(shrinker->flags & SHRINKER_MEMCG_AWARE)) |
439 | continue; | |
440 | ||
6b4f7799 JW |
441 | if (!(shrinker->flags & SHRINKER_NUMA_AWARE)) |
442 | sc.nid = 0; | |
1da177e4 | 443 | |
cb731d6c | 444 | freed += do_shrink_slab(&sc, shrinker, nr_scanned, nr_eligible); |
1da177e4 | 445 | } |
6b4f7799 | 446 | |
1da177e4 | 447 | up_read(&shrinker_rwsem); |
f06590bd MK |
448 | out: |
449 | cond_resched(); | |
24f7c6b9 | 450 | return freed; |
1da177e4 LT |
451 | } |
452 | ||
cb731d6c VD |
453 | void drop_slab_node(int nid) |
454 | { | |
455 | unsigned long freed; | |
456 | ||
457 | do { | |
458 | struct mem_cgroup *memcg = NULL; | |
459 | ||
460 | freed = 0; | |
461 | do { | |
462 | freed += shrink_slab(GFP_KERNEL, nid, memcg, | |
463 | 1000, 1000); | |
464 | } while ((memcg = mem_cgroup_iter(NULL, memcg, NULL)) != NULL); | |
465 | } while (freed > 10); | |
466 | } | |
467 | ||
468 | void drop_slab(void) | |
469 | { | |
470 | int nid; | |
471 | ||
472 | for_each_online_node(nid) | |
473 | drop_slab_node(nid); | |
474 | } | |
475 | ||
1da177e4 LT |
476 | static inline int is_page_cache_freeable(struct page *page) |
477 | { | |
ceddc3a5 JW |
478 | /* |
479 | * A freeable page cache page is referenced only by the caller | |
480 | * that isolated the page, the page cache radix tree and | |
481 | * optional buffer heads at page->private. | |
482 | */ | |
edcf4748 | 483 | return page_count(page) - page_has_private(page) == 2; |
1da177e4 LT |
484 | } |
485 | ||
703c2708 | 486 | static int may_write_to_inode(struct inode *inode, struct scan_control *sc) |
1da177e4 | 487 | { |
930d9152 | 488 | if (current->flags & PF_SWAPWRITE) |
1da177e4 | 489 | return 1; |
703c2708 | 490 | if (!inode_write_congested(inode)) |
1da177e4 | 491 | return 1; |
703c2708 | 492 | if (inode_to_bdi(inode) == current->backing_dev_info) |
1da177e4 LT |
493 | return 1; |
494 | return 0; | |
495 | } | |
496 | ||
497 | /* | |
498 | * We detected a synchronous write error writing a page out. Probably | |
499 | * -ENOSPC. We need to propagate that into the address_space for a subsequent | |
500 | * fsync(), msync() or close(). | |
501 | * | |
502 | * The tricky part is that after writepage we cannot touch the mapping: nothing | |
503 | * prevents it from being freed up. But we have a ref on the page and once | |
504 | * that page is locked, the mapping is pinned. | |
505 | * | |
506 | * We're allowed to run sleeping lock_page() here because we know the caller has | |
507 | * __GFP_FS. | |
508 | */ | |
509 | static void handle_write_error(struct address_space *mapping, | |
510 | struct page *page, int error) | |
511 | { | |
7eaceacc | 512 | lock_page(page); |
3e9f45bd GC |
513 | if (page_mapping(page) == mapping) |
514 | mapping_set_error(mapping, error); | |
1da177e4 LT |
515 | unlock_page(page); |
516 | } | |
517 | ||
04e62a29 CL |
518 | /* possible outcome of pageout() */ |
519 | typedef enum { | |
520 | /* failed to write page out, page is locked */ | |
521 | PAGE_KEEP, | |
522 | /* move page to the active list, page is locked */ | |
523 | PAGE_ACTIVATE, | |
524 | /* page has been sent to the disk successfully, page is unlocked */ | |
525 | PAGE_SUCCESS, | |
526 | /* page is clean and locked */ | |
527 | PAGE_CLEAN, | |
528 | } pageout_t; | |
529 | ||
1da177e4 | 530 | /* |
1742f19f AM |
531 | * pageout is called by shrink_page_list() for each dirty page. |
532 | * Calls ->writepage(). | |
1da177e4 | 533 | */ |
c661b078 | 534 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
7d3579e8 | 535 | struct scan_control *sc) |
1da177e4 LT |
536 | { |
537 | /* | |
538 | * If the page is dirty, only perform writeback if that write | |
539 | * will be non-blocking. To prevent this allocation from being | |
540 | * stalled by pagecache activity. But note that there may be | |
541 | * stalls if we need to run get_block(). We could test | |
542 | * PagePrivate for that. | |
543 | * | |
8174202b | 544 | * If this process is currently in __generic_file_write_iter() against |
1da177e4 LT |
545 | * this page's queue, we can perform writeback even if that |
546 | * will block. | |
547 | * | |
548 | * If the page is swapcache, write it back even if that would | |
549 | * block, for some throttling. This happens by accident, because | |
550 | * swap_backing_dev_info is bust: it doesn't reflect the | |
551 | * congestion state of the swapdevs. Easy to fix, if needed. | |
1da177e4 LT |
552 | */ |
553 | if (!is_page_cache_freeable(page)) | |
554 | return PAGE_KEEP; | |
555 | if (!mapping) { | |
556 | /* | |
557 | * Some data journaling orphaned pages can have | |
558 | * page->mapping == NULL while being dirty with clean buffers. | |
559 | */ | |
266cf658 | 560 | if (page_has_private(page)) { |
1da177e4 LT |
561 | if (try_to_free_buffers(page)) { |
562 | ClearPageDirty(page); | |
b1de0d13 | 563 | pr_info("%s: orphaned page\n", __func__); |
1da177e4 LT |
564 | return PAGE_CLEAN; |
565 | } | |
566 | } | |
567 | return PAGE_KEEP; | |
568 | } | |
569 | if (mapping->a_ops->writepage == NULL) | |
570 | return PAGE_ACTIVATE; | |
703c2708 | 571 | if (!may_write_to_inode(mapping->host, sc)) |
1da177e4 LT |
572 | return PAGE_KEEP; |
573 | ||
574 | if (clear_page_dirty_for_io(page)) { | |
575 | int res; | |
576 | struct writeback_control wbc = { | |
577 | .sync_mode = WB_SYNC_NONE, | |
578 | .nr_to_write = SWAP_CLUSTER_MAX, | |
111ebb6e OH |
579 | .range_start = 0, |
580 | .range_end = LLONG_MAX, | |
1da177e4 LT |
581 | .for_reclaim = 1, |
582 | }; | |
583 | ||
584 | SetPageReclaim(page); | |
585 | res = mapping->a_ops->writepage(page, &wbc); | |
586 | if (res < 0) | |
587 | handle_write_error(mapping, page, res); | |
994fc28c | 588 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
1da177e4 LT |
589 | ClearPageReclaim(page); |
590 | return PAGE_ACTIVATE; | |
591 | } | |
c661b078 | 592 | |
1da177e4 LT |
593 | if (!PageWriteback(page)) { |
594 | /* synchronous write or broken a_ops? */ | |
595 | ClearPageReclaim(page); | |
596 | } | |
3aa23851 | 597 | trace_mm_vmscan_writepage(page); |
e129b5c2 | 598 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
1da177e4 LT |
599 | return PAGE_SUCCESS; |
600 | } | |
601 | ||
602 | return PAGE_CLEAN; | |
603 | } | |
604 | ||
a649fd92 | 605 | /* |
e286781d NP |
606 | * Same as remove_mapping, but if the page is removed from the mapping, it |
607 | * gets returned with a refcount of 0. | |
a649fd92 | 608 | */ |
a528910e JW |
609 | static int __remove_mapping(struct address_space *mapping, struct page *page, |
610 | bool reclaimed) | |
49d2e9cc | 611 | { |
c4843a75 GT |
612 | unsigned long flags; |
613 | struct mem_cgroup *memcg; | |
614 | ||
28e4d965 NP |
615 | BUG_ON(!PageLocked(page)); |
616 | BUG_ON(mapping != page_mapping(page)); | |
49d2e9cc | 617 | |
c4843a75 GT |
618 | memcg = mem_cgroup_begin_page_stat(page); |
619 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
49d2e9cc | 620 | /* |
0fd0e6b0 NP |
621 | * The non racy check for a busy page. |
622 | * | |
623 | * Must be careful with the order of the tests. When someone has | |
624 | * a ref to the page, it may be possible that they dirty it then | |
625 | * drop the reference. So if PageDirty is tested before page_count | |
626 | * here, then the following race may occur: | |
627 | * | |
628 | * get_user_pages(&page); | |
629 | * [user mapping goes away] | |
630 | * write_to(page); | |
631 | * !PageDirty(page) [good] | |
632 | * SetPageDirty(page); | |
633 | * put_page(page); | |
634 | * !page_count(page) [good, discard it] | |
635 | * | |
636 | * [oops, our write_to data is lost] | |
637 | * | |
638 | * Reversing the order of the tests ensures such a situation cannot | |
639 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags | |
640 | * load is not satisfied before that of page->_count. | |
641 | * | |
642 | * Note that if SetPageDirty is always performed via set_page_dirty, | |
643 | * and thus under tree_lock, then this ordering is not required. | |
49d2e9cc | 644 | */ |
e286781d | 645 | if (!page_freeze_refs(page, 2)) |
49d2e9cc | 646 | goto cannot_free; |
e286781d NP |
647 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
648 | if (unlikely(PageDirty(page))) { | |
649 | page_unfreeze_refs(page, 2); | |
49d2e9cc | 650 | goto cannot_free; |
e286781d | 651 | } |
49d2e9cc CL |
652 | |
653 | if (PageSwapCache(page)) { | |
654 | swp_entry_t swap = { .val = page_private(page) }; | |
0a31bc97 | 655 | mem_cgroup_swapout(page, swap); |
49d2e9cc | 656 | __delete_from_swap_cache(page); |
c4843a75 GT |
657 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
658 | mem_cgroup_end_page_stat(memcg); | |
0a31bc97 | 659 | swapcache_free(swap); |
e286781d | 660 | } else { |
6072d13c | 661 | void (*freepage)(struct page *); |
a528910e | 662 | void *shadow = NULL; |
6072d13c LT |
663 | |
664 | freepage = mapping->a_ops->freepage; | |
a528910e JW |
665 | /* |
666 | * Remember a shadow entry for reclaimed file cache in | |
667 | * order to detect refaults, thus thrashing, later on. | |
668 | * | |
669 | * But don't store shadows in an address space that is | |
670 | * already exiting. This is not just an optizimation, | |
671 | * inode reclaim needs to empty out the radix tree or | |
672 | * the nodes are lost. Don't plant shadows behind its | |
673 | * back. | |
674 | */ | |
675 | if (reclaimed && page_is_file_cache(page) && | |
676 | !mapping_exiting(mapping)) | |
677 | shadow = workingset_eviction(mapping, page); | |
c4843a75 GT |
678 | __delete_from_page_cache(page, shadow, memcg); |
679 | spin_unlock_irqrestore(&mapping->tree_lock, flags); | |
680 | mem_cgroup_end_page_stat(memcg); | |
6072d13c LT |
681 | |
682 | if (freepage != NULL) | |
683 | freepage(page); | |
49d2e9cc CL |
684 | } |
685 | ||
49d2e9cc CL |
686 | return 1; |
687 | ||
688 | cannot_free: | |
c4843a75 GT |
689 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
690 | mem_cgroup_end_page_stat(memcg); | |
49d2e9cc CL |
691 | return 0; |
692 | } | |
693 | ||
e286781d NP |
694 | /* |
695 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if | |
696 | * someone else has a ref on the page, abort and return 0. If it was | |
697 | * successfully detached, return 1. Assumes the caller has a single ref on | |
698 | * this page. | |
699 | */ | |
700 | int remove_mapping(struct address_space *mapping, struct page *page) | |
701 | { | |
a528910e | 702 | if (__remove_mapping(mapping, page, false)) { |
e286781d NP |
703 | /* |
704 | * Unfreezing the refcount with 1 rather than 2 effectively | |
705 | * drops the pagecache ref for us without requiring another | |
706 | * atomic operation. | |
707 | */ | |
708 | page_unfreeze_refs(page, 1); | |
709 | return 1; | |
710 | } | |
711 | return 0; | |
712 | } | |
713 | ||
894bc310 LS |
714 | /** |
715 | * putback_lru_page - put previously isolated page onto appropriate LRU list | |
716 | * @page: page to be put back to appropriate lru list | |
717 | * | |
718 | * Add previously isolated @page to appropriate LRU list. | |
719 | * Page may still be unevictable for other reasons. | |
720 | * | |
721 | * lru_lock must not be held, interrupts must be enabled. | |
722 | */ | |
894bc310 LS |
723 | void putback_lru_page(struct page *page) |
724 | { | |
0ec3b74c | 725 | bool is_unevictable; |
bbfd28ee | 726 | int was_unevictable = PageUnevictable(page); |
894bc310 | 727 | |
309381fe | 728 | VM_BUG_ON_PAGE(PageLRU(page), page); |
894bc310 LS |
729 | |
730 | redo: | |
731 | ClearPageUnevictable(page); | |
732 | ||
39b5f29a | 733 | if (page_evictable(page)) { |
894bc310 LS |
734 | /* |
735 | * For evictable pages, we can use the cache. | |
736 | * In event of a race, worst case is we end up with an | |
737 | * unevictable page on [in]active list. | |
738 | * We know how to handle that. | |
739 | */ | |
0ec3b74c | 740 | is_unevictable = false; |
c53954a0 | 741 | lru_cache_add(page); |
894bc310 LS |
742 | } else { |
743 | /* | |
744 | * Put unevictable pages directly on zone's unevictable | |
745 | * list. | |
746 | */ | |
0ec3b74c | 747 | is_unevictable = true; |
894bc310 | 748 | add_page_to_unevictable_list(page); |
6a7b9548 | 749 | /* |
21ee9f39 MK |
750 | * When racing with an mlock or AS_UNEVICTABLE clearing |
751 | * (page is unlocked) make sure that if the other thread | |
752 | * does not observe our setting of PG_lru and fails | |
24513264 | 753 | * isolation/check_move_unevictable_pages, |
21ee9f39 | 754 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
6a7b9548 JW |
755 | * the page back to the evictable list. |
756 | * | |
21ee9f39 | 757 | * The other side is TestClearPageMlocked() or shmem_lock(). |
6a7b9548 JW |
758 | */ |
759 | smp_mb(); | |
894bc310 | 760 | } |
894bc310 LS |
761 | |
762 | /* | |
763 | * page's status can change while we move it among lru. If an evictable | |
764 | * page is on unevictable list, it never be freed. To avoid that, | |
765 | * check after we added it to the list, again. | |
766 | */ | |
0ec3b74c | 767 | if (is_unevictable && page_evictable(page)) { |
894bc310 LS |
768 | if (!isolate_lru_page(page)) { |
769 | put_page(page); | |
770 | goto redo; | |
771 | } | |
772 | /* This means someone else dropped this page from LRU | |
773 | * So, it will be freed or putback to LRU again. There is | |
774 | * nothing to do here. | |
775 | */ | |
776 | } | |
777 | ||
0ec3b74c | 778 | if (was_unevictable && !is_unevictable) |
bbfd28ee | 779 | count_vm_event(UNEVICTABLE_PGRESCUED); |
0ec3b74c | 780 | else if (!was_unevictable && is_unevictable) |
bbfd28ee LS |
781 | count_vm_event(UNEVICTABLE_PGCULLED); |
782 | ||
894bc310 LS |
783 | put_page(page); /* drop ref from isolate */ |
784 | } | |
785 | ||
dfc8d636 JW |
786 | enum page_references { |
787 | PAGEREF_RECLAIM, | |
788 | PAGEREF_RECLAIM_CLEAN, | |
64574746 | 789 | PAGEREF_KEEP, |
dfc8d636 JW |
790 | PAGEREF_ACTIVATE, |
791 | }; | |
792 | ||
793 | static enum page_references page_check_references(struct page *page, | |
794 | struct scan_control *sc) | |
795 | { | |
64574746 | 796 | int referenced_ptes, referenced_page; |
dfc8d636 | 797 | unsigned long vm_flags; |
dfc8d636 | 798 | |
c3ac9a8a JW |
799 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, |
800 | &vm_flags); | |
64574746 | 801 | referenced_page = TestClearPageReferenced(page); |
dfc8d636 | 802 | |
dfc8d636 JW |
803 | /* |
804 | * Mlock lost the isolation race with us. Let try_to_unmap() | |
805 | * move the page to the unevictable list. | |
806 | */ | |
807 | if (vm_flags & VM_LOCKED) | |
808 | return PAGEREF_RECLAIM; | |
809 | ||
64574746 | 810 | if (referenced_ptes) { |
e4898273 | 811 | if (PageSwapBacked(page)) |
64574746 JW |
812 | return PAGEREF_ACTIVATE; |
813 | /* | |
814 | * All mapped pages start out with page table | |
815 | * references from the instantiating fault, so we need | |
816 | * to look twice if a mapped file page is used more | |
817 | * than once. | |
818 | * | |
819 | * Mark it and spare it for another trip around the | |
820 | * inactive list. Another page table reference will | |
821 | * lead to its activation. | |
822 | * | |
823 | * Note: the mark is set for activated pages as well | |
824 | * so that recently deactivated but used pages are | |
825 | * quickly recovered. | |
826 | */ | |
827 | SetPageReferenced(page); | |
828 | ||
34dbc67a | 829 | if (referenced_page || referenced_ptes > 1) |
64574746 JW |
830 | return PAGEREF_ACTIVATE; |
831 | ||
c909e993 KK |
832 | /* |
833 | * Activate file-backed executable pages after first usage. | |
834 | */ | |
835 | if (vm_flags & VM_EXEC) | |
836 | return PAGEREF_ACTIVATE; | |
837 | ||
64574746 JW |
838 | return PAGEREF_KEEP; |
839 | } | |
dfc8d636 JW |
840 | |
841 | /* Reclaim if clean, defer dirty pages to writeback */ | |
2e30244a | 842 | if (referenced_page && !PageSwapBacked(page)) |
64574746 JW |
843 | return PAGEREF_RECLAIM_CLEAN; |
844 | ||
845 | return PAGEREF_RECLAIM; | |
dfc8d636 JW |
846 | } |
847 | ||
e2be15f6 MG |
848 | /* Check if a page is dirty or under writeback */ |
849 | static void page_check_dirty_writeback(struct page *page, | |
850 | bool *dirty, bool *writeback) | |
851 | { | |
b4597226 MG |
852 | struct address_space *mapping; |
853 | ||
e2be15f6 MG |
854 | /* |
855 | * Anonymous pages are not handled by flushers and must be written | |
856 | * from reclaim context. Do not stall reclaim based on them | |
857 | */ | |
858 | if (!page_is_file_cache(page)) { | |
859 | *dirty = false; | |
860 | *writeback = false; | |
861 | return; | |
862 | } | |
863 | ||
864 | /* By default assume that the page flags are accurate */ | |
865 | *dirty = PageDirty(page); | |
866 | *writeback = PageWriteback(page); | |
b4597226 MG |
867 | |
868 | /* Verify dirty/writeback state if the filesystem supports it */ | |
869 | if (!page_has_private(page)) | |
870 | return; | |
871 | ||
872 | mapping = page_mapping(page); | |
873 | if (mapping && mapping->a_ops->is_dirty_writeback) | |
874 | mapping->a_ops->is_dirty_writeback(page, dirty, writeback); | |
e2be15f6 MG |
875 | } |
876 | ||
1da177e4 | 877 | /* |
1742f19f | 878 | * shrink_page_list() returns the number of reclaimed pages |
1da177e4 | 879 | */ |
1742f19f | 880 | static unsigned long shrink_page_list(struct list_head *page_list, |
6a18adb3 | 881 | struct zone *zone, |
f84f6e2b | 882 | struct scan_control *sc, |
02c6de8d | 883 | enum ttu_flags ttu_flags, |
8e950282 | 884 | unsigned long *ret_nr_dirty, |
d43006d5 | 885 | unsigned long *ret_nr_unqueued_dirty, |
8e950282 | 886 | unsigned long *ret_nr_congested, |
02c6de8d | 887 | unsigned long *ret_nr_writeback, |
b1a6f21e | 888 | unsigned long *ret_nr_immediate, |
02c6de8d | 889 | bool force_reclaim) |
1da177e4 LT |
890 | { |
891 | LIST_HEAD(ret_pages); | |
abe4c3b5 | 892 | LIST_HEAD(free_pages); |
1da177e4 | 893 | int pgactivate = 0; |
d43006d5 | 894 | unsigned long nr_unqueued_dirty = 0; |
0e093d99 MG |
895 | unsigned long nr_dirty = 0; |
896 | unsigned long nr_congested = 0; | |
05ff5137 | 897 | unsigned long nr_reclaimed = 0; |
92df3a72 | 898 | unsigned long nr_writeback = 0; |
b1a6f21e | 899 | unsigned long nr_immediate = 0; |
1da177e4 LT |
900 | |
901 | cond_resched(); | |
902 | ||
1da177e4 LT |
903 | while (!list_empty(page_list)) { |
904 | struct address_space *mapping; | |
905 | struct page *page; | |
906 | int may_enter_fs; | |
02c6de8d | 907 | enum page_references references = PAGEREF_RECLAIM_CLEAN; |
e2be15f6 | 908 | bool dirty, writeback; |
854e9ed0 MK |
909 | bool lazyfree = false; |
910 | int ret = SWAP_SUCCESS; | |
1da177e4 LT |
911 | |
912 | cond_resched(); | |
913 | ||
914 | page = lru_to_page(page_list); | |
915 | list_del(&page->lru); | |
916 | ||
529ae9aa | 917 | if (!trylock_page(page)) |
1da177e4 LT |
918 | goto keep; |
919 | ||
309381fe SL |
920 | VM_BUG_ON_PAGE(PageActive(page), page); |
921 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
1da177e4 LT |
922 | |
923 | sc->nr_scanned++; | |
80e43426 | 924 | |
39b5f29a | 925 | if (unlikely(!page_evictable(page))) |
b291f000 | 926 | goto cull_mlocked; |
894bc310 | 927 | |
a6dc60f8 | 928 | if (!sc->may_unmap && page_mapped(page)) |
80e43426 CL |
929 | goto keep_locked; |
930 | ||
1da177e4 LT |
931 | /* Double the slab pressure for mapped and swapcache pages */ |
932 | if (page_mapped(page) || PageSwapCache(page)) | |
933 | sc->nr_scanned++; | |
934 | ||
c661b078 AW |
935 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
936 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); | |
937 | ||
e2be15f6 MG |
938 | /* |
939 | * The number of dirty pages determines if a zone is marked | |
940 | * reclaim_congested which affects wait_iff_congested. kswapd | |
941 | * will stall and start writing pages if the tail of the LRU | |
942 | * is all dirty unqueued pages. | |
943 | */ | |
944 | page_check_dirty_writeback(page, &dirty, &writeback); | |
945 | if (dirty || writeback) | |
946 | nr_dirty++; | |
947 | ||
948 | if (dirty && !writeback) | |
949 | nr_unqueued_dirty++; | |
950 | ||
d04e8acd MG |
951 | /* |
952 | * Treat this page as congested if the underlying BDI is or if | |
953 | * pages are cycling through the LRU so quickly that the | |
954 | * pages marked for immediate reclaim are making it to the | |
955 | * end of the LRU a second time. | |
956 | */ | |
e2be15f6 | 957 | mapping = page_mapping(page); |
1da58ee2 | 958 | if (((dirty || writeback) && mapping && |
703c2708 | 959 | inode_write_congested(mapping->host)) || |
d04e8acd | 960 | (writeback && PageReclaim(page))) |
e2be15f6 MG |
961 | nr_congested++; |
962 | ||
283aba9f MG |
963 | /* |
964 | * If a page at the tail of the LRU is under writeback, there | |
965 | * are three cases to consider. | |
966 | * | |
967 | * 1) If reclaim is encountering an excessive number of pages | |
968 | * under writeback and this page is both under writeback and | |
969 | * PageReclaim then it indicates that pages are being queued | |
970 | * for IO but are being recycled through the LRU before the | |
971 | * IO can complete. Waiting on the page itself risks an | |
972 | * indefinite stall if it is impossible to writeback the | |
973 | * page due to IO error or disconnected storage so instead | |
b1a6f21e MG |
974 | * note that the LRU is being scanned too quickly and the |
975 | * caller can stall after page list has been processed. | |
283aba9f | 976 | * |
97c9341f | 977 | * 2) Global or new memcg reclaim encounters a page that is |
ecf5fc6e MH |
978 | * not marked for immediate reclaim, or the caller does not |
979 | * have __GFP_FS (or __GFP_IO if it's simply going to swap, | |
980 | * not to fs). In this case mark the page for immediate | |
97c9341f | 981 | * reclaim and continue scanning. |
283aba9f | 982 | * |
ecf5fc6e MH |
983 | * Require may_enter_fs because we would wait on fs, which |
984 | * may not have submitted IO yet. And the loop driver might | |
283aba9f MG |
985 | * enter reclaim, and deadlock if it waits on a page for |
986 | * which it is needed to do the write (loop masks off | |
987 | * __GFP_IO|__GFP_FS for this reason); but more thought | |
988 | * would probably show more reasons. | |
989 | * | |
7fadc820 | 990 | * 3) Legacy memcg encounters a page that is already marked |
283aba9f MG |
991 | * PageReclaim. memcg does not have any dirty pages |
992 | * throttling so we could easily OOM just because too many | |
993 | * pages are in writeback and there is nothing else to | |
994 | * reclaim. Wait for the writeback to complete. | |
995 | */ | |
c661b078 | 996 | if (PageWriteback(page)) { |
283aba9f MG |
997 | /* Case 1 above */ |
998 | if (current_is_kswapd() && | |
999 | PageReclaim(page) && | |
57054651 | 1000 | test_bit(ZONE_WRITEBACK, &zone->flags)) { |
b1a6f21e MG |
1001 | nr_immediate++; |
1002 | goto keep_locked; | |
283aba9f MG |
1003 | |
1004 | /* Case 2 above */ | |
97c9341f | 1005 | } else if (sane_reclaim(sc) || |
ecf5fc6e | 1006 | !PageReclaim(page) || !may_enter_fs) { |
c3b94f44 HD |
1007 | /* |
1008 | * This is slightly racy - end_page_writeback() | |
1009 | * might have just cleared PageReclaim, then | |
1010 | * setting PageReclaim here end up interpreted | |
1011 | * as PageReadahead - but that does not matter | |
1012 | * enough to care. What we do want is for this | |
1013 | * page to have PageReclaim set next time memcg | |
1014 | * reclaim reaches the tests above, so it will | |
1015 | * then wait_on_page_writeback() to avoid OOM; | |
1016 | * and it's also appropriate in global reclaim. | |
1017 | */ | |
1018 | SetPageReclaim(page); | |
e62e384e | 1019 | nr_writeback++; |
c3b94f44 | 1020 | goto keep_locked; |
283aba9f MG |
1021 | |
1022 | /* Case 3 above */ | |
1023 | } else { | |
7fadc820 | 1024 | unlock_page(page); |
283aba9f | 1025 | wait_on_page_writeback(page); |
7fadc820 HD |
1026 | /* then go back and try same page again */ |
1027 | list_add_tail(&page->lru, page_list); | |
1028 | continue; | |
e62e384e | 1029 | } |
c661b078 | 1030 | } |
1da177e4 | 1031 | |
02c6de8d MK |
1032 | if (!force_reclaim) |
1033 | references = page_check_references(page, sc); | |
1034 | ||
dfc8d636 JW |
1035 | switch (references) { |
1036 | case PAGEREF_ACTIVATE: | |
1da177e4 | 1037 | goto activate_locked; |
64574746 JW |
1038 | case PAGEREF_KEEP: |
1039 | goto keep_locked; | |
dfc8d636 JW |
1040 | case PAGEREF_RECLAIM: |
1041 | case PAGEREF_RECLAIM_CLEAN: | |
1042 | ; /* try to reclaim the page below */ | |
1043 | } | |
1da177e4 | 1044 | |
1da177e4 LT |
1045 | /* |
1046 | * Anonymous process memory has backing store? | |
1047 | * Try to allocate it some swap space here. | |
1048 | */ | |
b291f000 | 1049 | if (PageAnon(page) && !PageSwapCache(page)) { |
63eb6b93 HD |
1050 | if (!(sc->gfp_mask & __GFP_IO)) |
1051 | goto keep_locked; | |
5bc7b8ac | 1052 | if (!add_to_swap(page, page_list)) |
1da177e4 | 1053 | goto activate_locked; |
854e9ed0 | 1054 | lazyfree = true; |
63eb6b93 | 1055 | may_enter_fs = 1; |
1da177e4 | 1056 | |
e2be15f6 MG |
1057 | /* Adding to swap updated mapping */ |
1058 | mapping = page_mapping(page); | |
1059 | } | |
1da177e4 LT |
1060 | |
1061 | /* | |
1062 | * The page is mapped into the page tables of one or more | |
1063 | * processes. Try to unmap it here. | |
1064 | */ | |
1065 | if (page_mapped(page) && mapping) { | |
854e9ed0 MK |
1066 | switch (ret = try_to_unmap(page, lazyfree ? |
1067 | (ttu_flags | TTU_BATCH_FLUSH | TTU_LZFREE) : | |
1068 | (ttu_flags | TTU_BATCH_FLUSH))) { | |
1da177e4 LT |
1069 | case SWAP_FAIL: |
1070 | goto activate_locked; | |
1071 | case SWAP_AGAIN: | |
1072 | goto keep_locked; | |
b291f000 NP |
1073 | case SWAP_MLOCK: |
1074 | goto cull_mlocked; | |
854e9ed0 MK |
1075 | case SWAP_LZFREE: |
1076 | goto lazyfree; | |
1da177e4 LT |
1077 | case SWAP_SUCCESS: |
1078 | ; /* try to free the page below */ | |
1079 | } | |
1080 | } | |
1081 | ||
1082 | if (PageDirty(page)) { | |
ee72886d MG |
1083 | /* |
1084 | * Only kswapd can writeback filesystem pages to | |
d43006d5 MG |
1085 | * avoid risk of stack overflow but only writeback |
1086 | * if many dirty pages have been encountered. | |
ee72886d | 1087 | */ |
f84f6e2b | 1088 | if (page_is_file_cache(page) && |
9e3b2f8c | 1089 | (!current_is_kswapd() || |
57054651 | 1090 | !test_bit(ZONE_DIRTY, &zone->flags))) { |
49ea7eb6 MG |
1091 | /* |
1092 | * Immediately reclaim when written back. | |
1093 | * Similar in principal to deactivate_page() | |
1094 | * except we already have the page isolated | |
1095 | * and know it's dirty | |
1096 | */ | |
1097 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); | |
1098 | SetPageReclaim(page); | |
1099 | ||
ee72886d MG |
1100 | goto keep_locked; |
1101 | } | |
1102 | ||
dfc8d636 | 1103 | if (references == PAGEREF_RECLAIM_CLEAN) |
1da177e4 | 1104 | goto keep_locked; |
4dd4b920 | 1105 | if (!may_enter_fs) |
1da177e4 | 1106 | goto keep_locked; |
52a8363e | 1107 | if (!sc->may_writepage) |
1da177e4 LT |
1108 | goto keep_locked; |
1109 | ||
d950c947 MG |
1110 | /* |
1111 | * Page is dirty. Flush the TLB if a writable entry | |
1112 | * potentially exists to avoid CPU writes after IO | |
1113 | * starts and then write it out here. | |
1114 | */ | |
1115 | try_to_unmap_flush_dirty(); | |
7d3579e8 | 1116 | switch (pageout(page, mapping, sc)) { |
1da177e4 LT |
1117 | case PAGE_KEEP: |
1118 | goto keep_locked; | |
1119 | case PAGE_ACTIVATE: | |
1120 | goto activate_locked; | |
1121 | case PAGE_SUCCESS: | |
7d3579e8 | 1122 | if (PageWriteback(page)) |
41ac1999 | 1123 | goto keep; |
7d3579e8 | 1124 | if (PageDirty(page)) |
1da177e4 | 1125 | goto keep; |
7d3579e8 | 1126 | |
1da177e4 LT |
1127 | /* |
1128 | * A synchronous write - probably a ramdisk. Go | |
1129 | * ahead and try to reclaim the page. | |
1130 | */ | |
529ae9aa | 1131 | if (!trylock_page(page)) |
1da177e4 LT |
1132 | goto keep; |
1133 | if (PageDirty(page) || PageWriteback(page)) | |
1134 | goto keep_locked; | |
1135 | mapping = page_mapping(page); | |
1136 | case PAGE_CLEAN: | |
1137 | ; /* try to free the page below */ | |
1138 | } | |
1139 | } | |
1140 | ||
1141 | /* | |
1142 | * If the page has buffers, try to free the buffer mappings | |
1143 | * associated with this page. If we succeed we try to free | |
1144 | * the page as well. | |
1145 | * | |
1146 | * We do this even if the page is PageDirty(). | |
1147 | * try_to_release_page() does not perform I/O, but it is | |
1148 | * possible for a page to have PageDirty set, but it is actually | |
1149 | * clean (all its buffers are clean). This happens if the | |
1150 | * buffers were written out directly, with submit_bh(). ext3 | |
894bc310 | 1151 | * will do this, as well as the blockdev mapping. |
1da177e4 LT |
1152 | * try_to_release_page() will discover that cleanness and will |
1153 | * drop the buffers and mark the page clean - it can be freed. | |
1154 | * | |
1155 | * Rarely, pages can have buffers and no ->mapping. These are | |
1156 | * the pages which were not successfully invalidated in | |
1157 | * truncate_complete_page(). We try to drop those buffers here | |
1158 | * and if that worked, and the page is no longer mapped into | |
1159 | * process address space (page_count == 1) it can be freed. | |
1160 | * Otherwise, leave the page on the LRU so it is swappable. | |
1161 | */ | |
266cf658 | 1162 | if (page_has_private(page)) { |
1da177e4 LT |
1163 | if (!try_to_release_page(page, sc->gfp_mask)) |
1164 | goto activate_locked; | |
e286781d NP |
1165 | if (!mapping && page_count(page) == 1) { |
1166 | unlock_page(page); | |
1167 | if (put_page_testzero(page)) | |
1168 | goto free_it; | |
1169 | else { | |
1170 | /* | |
1171 | * rare race with speculative reference. | |
1172 | * the speculative reference will free | |
1173 | * this page shortly, so we may | |
1174 | * increment nr_reclaimed here (and | |
1175 | * leave it off the LRU). | |
1176 | */ | |
1177 | nr_reclaimed++; | |
1178 | continue; | |
1179 | } | |
1180 | } | |
1da177e4 LT |
1181 | } |
1182 | ||
854e9ed0 | 1183 | lazyfree: |
a528910e | 1184 | if (!mapping || !__remove_mapping(mapping, page, true)) |
49d2e9cc | 1185 | goto keep_locked; |
1da177e4 | 1186 | |
a978d6f5 NP |
1187 | /* |
1188 | * At this point, we have no other references and there is | |
1189 | * no way to pick any more up (removed from LRU, removed | |
1190 | * from pagecache). Can use non-atomic bitops now (and | |
1191 | * we obviously don't have to worry about waking up a process | |
1192 | * waiting on the page lock, because there are no references. | |
1193 | */ | |
48c935ad | 1194 | __ClearPageLocked(page); |
e286781d | 1195 | free_it: |
854e9ed0 MK |
1196 | if (ret == SWAP_LZFREE) |
1197 | count_vm_event(PGLAZYFREED); | |
1198 | ||
05ff5137 | 1199 | nr_reclaimed++; |
abe4c3b5 MG |
1200 | |
1201 | /* | |
1202 | * Is there need to periodically free_page_list? It would | |
1203 | * appear not as the counts should be low | |
1204 | */ | |
1205 | list_add(&page->lru, &free_pages); | |
1da177e4 LT |
1206 | continue; |
1207 | ||
b291f000 | 1208 | cull_mlocked: |
63d6c5ad HD |
1209 | if (PageSwapCache(page)) |
1210 | try_to_free_swap(page); | |
b291f000 | 1211 | unlock_page(page); |
c54839a7 | 1212 | list_add(&page->lru, &ret_pages); |
b291f000 NP |
1213 | continue; |
1214 | ||
1da177e4 | 1215 | activate_locked: |
68a22394 RR |
1216 | /* Not a candidate for swapping, so reclaim swap space. */ |
1217 | if (PageSwapCache(page) && vm_swap_full()) | |
a2c43eed | 1218 | try_to_free_swap(page); |
309381fe | 1219 | VM_BUG_ON_PAGE(PageActive(page), page); |
1da177e4 LT |
1220 | SetPageActive(page); |
1221 | pgactivate++; | |
1222 | keep_locked: | |
1223 | unlock_page(page); | |
1224 | keep: | |
1225 | list_add(&page->lru, &ret_pages); | |
309381fe | 1226 | VM_BUG_ON_PAGE(PageLRU(page) || PageUnevictable(page), page); |
1da177e4 | 1227 | } |
abe4c3b5 | 1228 | |
747db954 | 1229 | mem_cgroup_uncharge_list(&free_pages); |
72b252ae | 1230 | try_to_unmap_flush(); |
b745bc85 | 1231 | free_hot_cold_page_list(&free_pages, true); |
abe4c3b5 | 1232 | |
1da177e4 | 1233 | list_splice(&ret_pages, page_list); |
f8891e5e | 1234 | count_vm_events(PGACTIVATE, pgactivate); |
0a31bc97 | 1235 | |
8e950282 MG |
1236 | *ret_nr_dirty += nr_dirty; |
1237 | *ret_nr_congested += nr_congested; | |
d43006d5 | 1238 | *ret_nr_unqueued_dirty += nr_unqueued_dirty; |
92df3a72 | 1239 | *ret_nr_writeback += nr_writeback; |
b1a6f21e | 1240 | *ret_nr_immediate += nr_immediate; |
05ff5137 | 1241 | return nr_reclaimed; |
1da177e4 LT |
1242 | } |
1243 | ||
02c6de8d MK |
1244 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, |
1245 | struct list_head *page_list) | |
1246 | { | |
1247 | struct scan_control sc = { | |
1248 | .gfp_mask = GFP_KERNEL, | |
1249 | .priority = DEF_PRIORITY, | |
1250 | .may_unmap = 1, | |
1251 | }; | |
8e950282 | 1252 | unsigned long ret, dummy1, dummy2, dummy3, dummy4, dummy5; |
02c6de8d MK |
1253 | struct page *page, *next; |
1254 | LIST_HEAD(clean_pages); | |
1255 | ||
1256 | list_for_each_entry_safe(page, next, page_list, lru) { | |
117aad1e RA |
1257 | if (page_is_file_cache(page) && !PageDirty(page) && |
1258 | !isolated_balloon_page(page)) { | |
02c6de8d MK |
1259 | ClearPageActive(page); |
1260 | list_move(&page->lru, &clean_pages); | |
1261 | } | |
1262 | } | |
1263 | ||
1264 | ret = shrink_page_list(&clean_pages, zone, &sc, | |
8e950282 MG |
1265 | TTU_UNMAP|TTU_IGNORE_ACCESS, |
1266 | &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, true); | |
02c6de8d | 1267 | list_splice(&clean_pages, page_list); |
83da7510 | 1268 | mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret); |
02c6de8d MK |
1269 | return ret; |
1270 | } | |
1271 | ||
5ad333eb AW |
1272 | /* |
1273 | * Attempt to remove the specified page from its LRU. Only take this page | |
1274 | * if it is of the appropriate PageActive status. Pages which are being | |
1275 | * freed elsewhere are also ignored. | |
1276 | * | |
1277 | * page: page to consider | |
1278 | * mode: one of the LRU isolation modes defined above | |
1279 | * | |
1280 | * returns 0 on success, -ve errno on failure. | |
1281 | */ | |
f3fd4a61 | 1282 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) |
5ad333eb AW |
1283 | { |
1284 | int ret = -EINVAL; | |
1285 | ||
1286 | /* Only take pages on the LRU. */ | |
1287 | if (!PageLRU(page)) | |
1288 | return ret; | |
1289 | ||
e46a2879 MK |
1290 | /* Compaction should not handle unevictable pages but CMA can do so */ |
1291 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) | |
894bc310 LS |
1292 | return ret; |
1293 | ||
5ad333eb | 1294 | ret = -EBUSY; |
08e552c6 | 1295 | |
c8244935 MG |
1296 | /* |
1297 | * To minimise LRU disruption, the caller can indicate that it only | |
1298 | * wants to isolate pages it will be able to operate on without | |
1299 | * blocking - clean pages for the most part. | |
1300 | * | |
1301 | * ISOLATE_CLEAN means that only clean pages should be isolated. This | |
1302 | * is used by reclaim when it is cannot write to backing storage | |
1303 | * | |
1304 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages | |
1305 | * that it is possible to migrate without blocking | |
1306 | */ | |
1307 | if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) { | |
1308 | /* All the caller can do on PageWriteback is block */ | |
1309 | if (PageWriteback(page)) | |
1310 | return ret; | |
1311 | ||
1312 | if (PageDirty(page)) { | |
1313 | struct address_space *mapping; | |
1314 | ||
1315 | /* ISOLATE_CLEAN means only clean pages */ | |
1316 | if (mode & ISOLATE_CLEAN) | |
1317 | return ret; | |
1318 | ||
1319 | /* | |
1320 | * Only pages without mappings or that have a | |
1321 | * ->migratepage callback are possible to migrate | |
1322 | * without blocking | |
1323 | */ | |
1324 | mapping = page_mapping(page); | |
1325 | if (mapping && !mapping->a_ops->migratepage) | |
1326 | return ret; | |
1327 | } | |
1328 | } | |
39deaf85 | 1329 | |
f80c0673 MK |
1330 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
1331 | return ret; | |
1332 | ||
5ad333eb AW |
1333 | if (likely(get_page_unless_zero(page))) { |
1334 | /* | |
1335 | * Be careful not to clear PageLRU until after we're | |
1336 | * sure the page is not being freed elsewhere -- the | |
1337 | * page release code relies on it. | |
1338 | */ | |
1339 | ClearPageLRU(page); | |
1340 | ret = 0; | |
1341 | } | |
1342 | ||
1343 | return ret; | |
1344 | } | |
1345 | ||
1da177e4 LT |
1346 | /* |
1347 | * zone->lru_lock is heavily contended. Some of the functions that | |
1348 | * shrink the lists perform better by taking out a batch of pages | |
1349 | * and working on them outside the LRU lock. | |
1350 | * | |
1351 | * For pagecache intensive workloads, this function is the hottest | |
1352 | * spot in the kernel (apart from copy_*_user functions). | |
1353 | * | |
1354 | * Appropriate locks must be held before calling this function. | |
1355 | * | |
1356 | * @nr_to_scan: The number of pages to look through on the list. | |
5dc35979 | 1357 | * @lruvec: The LRU vector to pull pages from. |
1da177e4 | 1358 | * @dst: The temp list to put pages on to. |
f626012d | 1359 | * @nr_scanned: The number of pages that were scanned. |
fe2c2a10 | 1360 | * @sc: The scan_control struct for this reclaim session |
5ad333eb | 1361 | * @mode: One of the LRU isolation modes |
3cb99451 | 1362 | * @lru: LRU list id for isolating |
1da177e4 LT |
1363 | * |
1364 | * returns how many pages were moved onto *@dst. | |
1365 | */ | |
69e05944 | 1366 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
5dc35979 | 1367 | struct lruvec *lruvec, struct list_head *dst, |
fe2c2a10 | 1368 | unsigned long *nr_scanned, struct scan_control *sc, |
3cb99451 | 1369 | isolate_mode_t mode, enum lru_list lru) |
1da177e4 | 1370 | { |
75b00af7 | 1371 | struct list_head *src = &lruvec->lists[lru]; |
69e05944 | 1372 | unsigned long nr_taken = 0; |
c9b02d97 | 1373 | unsigned long scan; |
1da177e4 | 1374 | |
0b802f10 VD |
1375 | for (scan = 0; scan < nr_to_scan && nr_taken < nr_to_scan && |
1376 | !list_empty(src); scan++) { | |
5ad333eb | 1377 | struct page *page; |
fa9add64 | 1378 | int nr_pages; |
5ad333eb | 1379 | |
1da177e4 LT |
1380 | page = lru_to_page(src); |
1381 | prefetchw_prev_lru_page(page, src, flags); | |
1382 | ||
309381fe | 1383 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
8d438f96 | 1384 | |
f3fd4a61 | 1385 | switch (__isolate_lru_page(page, mode)) { |
5ad333eb | 1386 | case 0: |
fa9add64 HD |
1387 | nr_pages = hpage_nr_pages(page); |
1388 | mem_cgroup_update_lru_size(lruvec, lru, -nr_pages); | |
5ad333eb | 1389 | list_move(&page->lru, dst); |
fa9add64 | 1390 | nr_taken += nr_pages; |
5ad333eb AW |
1391 | break; |
1392 | ||
1393 | case -EBUSY: | |
1394 | /* else it is being freed elsewhere */ | |
1395 | list_move(&page->lru, src); | |
1396 | continue; | |
46453a6e | 1397 | |
5ad333eb AW |
1398 | default: |
1399 | BUG(); | |
1400 | } | |
1da177e4 LT |
1401 | } |
1402 | ||
f626012d | 1403 | *nr_scanned = scan; |
75b00af7 HD |
1404 | trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan, |
1405 | nr_taken, mode, is_file_lru(lru)); | |
1da177e4 LT |
1406 | return nr_taken; |
1407 | } | |
1408 | ||
62695a84 NP |
1409 | /** |
1410 | * isolate_lru_page - tries to isolate a page from its LRU list | |
1411 | * @page: page to isolate from its LRU list | |
1412 | * | |
1413 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the | |
1414 | * vmstat statistic corresponding to whatever LRU list the page was on. | |
1415 | * | |
1416 | * Returns 0 if the page was removed from an LRU list. | |
1417 | * Returns -EBUSY if the page was not on an LRU list. | |
1418 | * | |
1419 | * The returned page will have PageLRU() cleared. If it was found on | |
894bc310 LS |
1420 | * the active list, it will have PageActive set. If it was found on |
1421 | * the unevictable list, it will have the PageUnevictable bit set. That flag | |
1422 | * may need to be cleared by the caller before letting the page go. | |
62695a84 NP |
1423 | * |
1424 | * The vmstat statistic corresponding to the list on which the page was | |
1425 | * found will be decremented. | |
1426 | * | |
1427 | * Restrictions: | |
1428 | * (1) Must be called with an elevated refcount on the page. This is a | |
1429 | * fundamentnal difference from isolate_lru_pages (which is called | |
1430 | * without a stable reference). | |
1431 | * (2) the lru_lock must not be held. | |
1432 | * (3) interrupts must be enabled. | |
1433 | */ | |
1434 | int isolate_lru_page(struct page *page) | |
1435 | { | |
1436 | int ret = -EBUSY; | |
1437 | ||
309381fe | 1438 | VM_BUG_ON_PAGE(!page_count(page), page); |
bb5b8589 | 1439 | VM_BUG_ON_PAGE(PageTail(page), page); |
0c917313 | 1440 | |
62695a84 NP |
1441 | if (PageLRU(page)) { |
1442 | struct zone *zone = page_zone(page); | |
fa9add64 | 1443 | struct lruvec *lruvec; |
62695a84 NP |
1444 | |
1445 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 1446 | lruvec = mem_cgroup_page_lruvec(page, zone); |
0c917313 | 1447 | if (PageLRU(page)) { |
894bc310 | 1448 | int lru = page_lru(page); |
0c917313 | 1449 | get_page(page); |
62695a84 | 1450 | ClearPageLRU(page); |
fa9add64 HD |
1451 | del_page_from_lru_list(page, lruvec, lru); |
1452 | ret = 0; | |
62695a84 NP |
1453 | } |
1454 | spin_unlock_irq(&zone->lru_lock); | |
1455 | } | |
1456 | return ret; | |
1457 | } | |
1458 | ||
35cd7815 | 1459 | /* |
d37dd5dc FW |
1460 | * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and |
1461 | * then get resheduled. When there are massive number of tasks doing page | |
1462 | * allocation, such sleeping direct reclaimers may keep piling up on each CPU, | |
1463 | * the LRU list will go small and be scanned faster than necessary, leading to | |
1464 | * unnecessary swapping, thrashing and OOM. | |
35cd7815 RR |
1465 | */ |
1466 | static int too_many_isolated(struct zone *zone, int file, | |
1467 | struct scan_control *sc) | |
1468 | { | |
1469 | unsigned long inactive, isolated; | |
1470 | ||
1471 | if (current_is_kswapd()) | |
1472 | return 0; | |
1473 | ||
97c9341f | 1474 | if (!sane_reclaim(sc)) |
35cd7815 RR |
1475 | return 0; |
1476 | ||
1477 | if (file) { | |
1478 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); | |
1479 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); | |
1480 | } else { | |
1481 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1482 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); | |
1483 | } | |
1484 | ||
3cf23841 FW |
1485 | /* |
1486 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they | |
1487 | * won't get blocked by normal direct-reclaimers, forming a circular | |
1488 | * deadlock. | |
1489 | */ | |
d0164adc | 1490 | if ((sc->gfp_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
3cf23841 FW |
1491 | inactive >>= 3; |
1492 | ||
35cd7815 RR |
1493 | return isolated > inactive; |
1494 | } | |
1495 | ||
66635629 | 1496 | static noinline_for_stack void |
75b00af7 | 1497 | putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list) |
66635629 | 1498 | { |
27ac81d8 KK |
1499 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1500 | struct zone *zone = lruvec_zone(lruvec); | |
3f79768f | 1501 | LIST_HEAD(pages_to_free); |
66635629 | 1502 | |
66635629 MG |
1503 | /* |
1504 | * Put back any unfreeable pages. | |
1505 | */ | |
66635629 | 1506 | while (!list_empty(page_list)) { |
3f79768f | 1507 | struct page *page = lru_to_page(page_list); |
66635629 | 1508 | int lru; |
3f79768f | 1509 | |
309381fe | 1510 | VM_BUG_ON_PAGE(PageLRU(page), page); |
66635629 | 1511 | list_del(&page->lru); |
39b5f29a | 1512 | if (unlikely(!page_evictable(page))) { |
66635629 MG |
1513 | spin_unlock_irq(&zone->lru_lock); |
1514 | putback_lru_page(page); | |
1515 | spin_lock_irq(&zone->lru_lock); | |
1516 | continue; | |
1517 | } | |
fa9add64 HD |
1518 | |
1519 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
1520 | ||
7a608572 | 1521 | SetPageLRU(page); |
66635629 | 1522 | lru = page_lru(page); |
fa9add64 HD |
1523 | add_page_to_lru_list(page, lruvec, lru); |
1524 | ||
66635629 MG |
1525 | if (is_active_lru(lru)) { |
1526 | int file = is_file_lru(lru); | |
9992af10 RR |
1527 | int numpages = hpage_nr_pages(page); |
1528 | reclaim_stat->recent_rotated[file] += numpages; | |
66635629 | 1529 | } |
2bcf8879 HD |
1530 | if (put_page_testzero(page)) { |
1531 | __ClearPageLRU(page); | |
1532 | __ClearPageActive(page); | |
fa9add64 | 1533 | del_page_from_lru_list(page, lruvec, lru); |
2bcf8879 HD |
1534 | |
1535 | if (unlikely(PageCompound(page))) { | |
1536 | spin_unlock_irq(&zone->lru_lock); | |
747db954 | 1537 | mem_cgroup_uncharge(page); |
2bcf8879 HD |
1538 | (*get_compound_page_dtor(page))(page); |
1539 | spin_lock_irq(&zone->lru_lock); | |
1540 | } else | |
1541 | list_add(&page->lru, &pages_to_free); | |
66635629 MG |
1542 | } |
1543 | } | |
66635629 | 1544 | |
3f79768f HD |
1545 | /* |
1546 | * To save our caller's stack, now use input list for pages to free. | |
1547 | */ | |
1548 | list_splice(&pages_to_free, page_list); | |
66635629 MG |
1549 | } |
1550 | ||
399ba0b9 N |
1551 | /* |
1552 | * If a kernel thread (such as nfsd for loop-back mounts) services | |
1553 | * a backing device by writing to the page cache it sets PF_LESS_THROTTLE. | |
1554 | * In that case we should only throttle if the backing device it is | |
1555 | * writing to is congested. In other cases it is safe to throttle. | |
1556 | */ | |
1557 | static int current_may_throttle(void) | |
1558 | { | |
1559 | return !(current->flags & PF_LESS_THROTTLE) || | |
1560 | current->backing_dev_info == NULL || | |
1561 | bdi_write_congested(current->backing_dev_info); | |
1562 | } | |
1563 | ||
1da177e4 | 1564 | /* |
1742f19f AM |
1565 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
1566 | * of reclaimed pages | |
1da177e4 | 1567 | */ |
66635629 | 1568 | static noinline_for_stack unsigned long |
1a93be0e | 1569 | shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec, |
9e3b2f8c | 1570 | struct scan_control *sc, enum lru_list lru) |
1da177e4 LT |
1571 | { |
1572 | LIST_HEAD(page_list); | |
e247dbce | 1573 | unsigned long nr_scanned; |
05ff5137 | 1574 | unsigned long nr_reclaimed = 0; |
e247dbce | 1575 | unsigned long nr_taken; |
8e950282 MG |
1576 | unsigned long nr_dirty = 0; |
1577 | unsigned long nr_congested = 0; | |
e2be15f6 | 1578 | unsigned long nr_unqueued_dirty = 0; |
92df3a72 | 1579 | unsigned long nr_writeback = 0; |
b1a6f21e | 1580 | unsigned long nr_immediate = 0; |
f3fd4a61 | 1581 | isolate_mode_t isolate_mode = 0; |
3cb99451 | 1582 | int file = is_file_lru(lru); |
1a93be0e KK |
1583 | struct zone *zone = lruvec_zone(lruvec); |
1584 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; | |
78dc583d | 1585 | |
35cd7815 | 1586 | while (unlikely(too_many_isolated(zone, file, sc))) { |
58355c78 | 1587 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
35cd7815 RR |
1588 | |
1589 | /* We are about to die and free our memory. Return now. */ | |
1590 | if (fatal_signal_pending(current)) | |
1591 | return SWAP_CLUSTER_MAX; | |
1592 | } | |
1593 | ||
1da177e4 | 1594 | lru_add_drain(); |
f80c0673 MK |
1595 | |
1596 | if (!sc->may_unmap) | |
61317289 | 1597 | isolate_mode |= ISOLATE_UNMAPPED; |
f80c0673 | 1598 | if (!sc->may_writepage) |
61317289 | 1599 | isolate_mode |= ISOLATE_CLEAN; |
f80c0673 | 1600 | |
1da177e4 | 1601 | spin_lock_irq(&zone->lru_lock); |
b35ea17b | 1602 | |
5dc35979 KK |
1603 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, |
1604 | &nr_scanned, sc, isolate_mode, lru); | |
95d918fc KK |
1605 | |
1606 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); | |
1607 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); | |
1608 | ||
89b5fae5 | 1609 | if (global_reclaim(sc)) { |
0d5d823a | 1610 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, nr_scanned); |
e247dbce | 1611 | if (current_is_kswapd()) |
75b00af7 | 1612 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned); |
e247dbce | 1613 | else |
75b00af7 | 1614 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned); |
e247dbce | 1615 | } |
d563c050 | 1616 | spin_unlock_irq(&zone->lru_lock); |
b35ea17b | 1617 | |
d563c050 | 1618 | if (nr_taken == 0) |
66635629 | 1619 | return 0; |
5ad333eb | 1620 | |
02c6de8d | 1621 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP, |
8e950282 MG |
1622 | &nr_dirty, &nr_unqueued_dirty, &nr_congested, |
1623 | &nr_writeback, &nr_immediate, | |
1624 | false); | |
c661b078 | 1625 | |
3f79768f HD |
1626 | spin_lock_irq(&zone->lru_lock); |
1627 | ||
95d918fc | 1628 | reclaim_stat->recent_scanned[file] += nr_taken; |
d563c050 | 1629 | |
904249aa YH |
1630 | if (global_reclaim(sc)) { |
1631 | if (current_is_kswapd()) | |
1632 | __count_zone_vm_events(PGSTEAL_KSWAPD, zone, | |
1633 | nr_reclaimed); | |
1634 | else | |
1635 | __count_zone_vm_events(PGSTEAL_DIRECT, zone, | |
1636 | nr_reclaimed); | |
1637 | } | |
a74609fa | 1638 | |
27ac81d8 | 1639 | putback_inactive_pages(lruvec, &page_list); |
3f79768f | 1640 | |
95d918fc | 1641 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
3f79768f HD |
1642 | |
1643 | spin_unlock_irq(&zone->lru_lock); | |
1644 | ||
747db954 | 1645 | mem_cgroup_uncharge_list(&page_list); |
b745bc85 | 1646 | free_hot_cold_page_list(&page_list, true); |
e11da5b4 | 1647 | |
92df3a72 MG |
1648 | /* |
1649 | * If reclaim is isolating dirty pages under writeback, it implies | |
1650 | * that the long-lived page allocation rate is exceeding the page | |
1651 | * laundering rate. Either the global limits are not being effective | |
1652 | * at throttling processes due to the page distribution throughout | |
1653 | * zones or there is heavy usage of a slow backing device. The | |
1654 | * only option is to throttle from reclaim context which is not ideal | |
1655 | * as there is no guarantee the dirtying process is throttled in the | |
1656 | * same way balance_dirty_pages() manages. | |
1657 | * | |
8e950282 MG |
1658 | * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number |
1659 | * of pages under pages flagged for immediate reclaim and stall if any | |
1660 | * are encountered in the nr_immediate check below. | |
92df3a72 | 1661 | */ |
918fc718 | 1662 | if (nr_writeback && nr_writeback == nr_taken) |
57054651 | 1663 | set_bit(ZONE_WRITEBACK, &zone->flags); |
92df3a72 | 1664 | |
d43006d5 | 1665 | /* |
97c9341f TH |
1666 | * Legacy memcg will stall in page writeback so avoid forcibly |
1667 | * stalling here. | |
d43006d5 | 1668 | */ |
97c9341f | 1669 | if (sane_reclaim(sc)) { |
8e950282 MG |
1670 | /* |
1671 | * Tag a zone as congested if all the dirty pages scanned were | |
1672 | * backed by a congested BDI and wait_iff_congested will stall. | |
1673 | */ | |
1674 | if (nr_dirty && nr_dirty == nr_congested) | |
57054651 | 1675 | set_bit(ZONE_CONGESTED, &zone->flags); |
8e950282 | 1676 | |
b1a6f21e MG |
1677 | /* |
1678 | * If dirty pages are scanned that are not queued for IO, it | |
1679 | * implies that flushers are not keeping up. In this case, flag | |
57054651 JW |
1680 | * the zone ZONE_DIRTY and kswapd will start writing pages from |
1681 | * reclaim context. | |
b1a6f21e MG |
1682 | */ |
1683 | if (nr_unqueued_dirty == nr_taken) | |
57054651 | 1684 | set_bit(ZONE_DIRTY, &zone->flags); |
b1a6f21e MG |
1685 | |
1686 | /* | |
b738d764 LT |
1687 | * If kswapd scans pages marked marked for immediate |
1688 | * reclaim and under writeback (nr_immediate), it implies | |
1689 | * that pages are cycling through the LRU faster than | |
b1a6f21e MG |
1690 | * they are written so also forcibly stall. |
1691 | */ | |
b738d764 | 1692 | if (nr_immediate && current_may_throttle()) |
b1a6f21e | 1693 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
e2be15f6 | 1694 | } |
d43006d5 | 1695 | |
8e950282 MG |
1696 | /* |
1697 | * Stall direct reclaim for IO completions if underlying BDIs or zone | |
1698 | * is congested. Allow kswapd to continue until it starts encountering | |
1699 | * unqueued dirty pages or cycling through the LRU too quickly. | |
1700 | */ | |
399ba0b9 N |
1701 | if (!sc->hibernation_mode && !current_is_kswapd() && |
1702 | current_may_throttle()) | |
8e950282 MG |
1703 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); |
1704 | ||
ba5e9579 | 1705 | trace_mm_vmscan_lru_shrink_inactive(zone, nr_scanned, nr_reclaimed, |
1706 | sc->priority, file); | |
05ff5137 | 1707 | return nr_reclaimed; |
1da177e4 LT |
1708 | } |
1709 | ||
1710 | /* | |
1711 | * This moves pages from the active list to the inactive list. | |
1712 | * | |
1713 | * We move them the other way if the page is referenced by one or more | |
1714 | * processes, from rmap. | |
1715 | * | |
1716 | * If the pages are mostly unmapped, the processing is fast and it is | |
1717 | * appropriate to hold zone->lru_lock across the whole operation. But if | |
1718 | * the pages are mapped, the processing is slow (page_referenced()) so we | |
1719 | * should drop zone->lru_lock around each page. It's impossible to balance | |
1720 | * this, so instead we remove the pages from the LRU while processing them. | |
1721 | * It is safe to rely on PG_active against the non-LRU pages in here because | |
1722 | * nobody will play with that bit on a non-LRU page. | |
1723 | * | |
1724 | * The downside is that we have to touch page->_count against each page. | |
1725 | * But we had to alter page->flags anyway. | |
1726 | */ | |
1cfb419b | 1727 | |
fa9add64 | 1728 | static void move_active_pages_to_lru(struct lruvec *lruvec, |
3eb4140f | 1729 | struct list_head *list, |
2bcf8879 | 1730 | struct list_head *pages_to_free, |
3eb4140f WF |
1731 | enum lru_list lru) |
1732 | { | |
fa9add64 | 1733 | struct zone *zone = lruvec_zone(lruvec); |
3eb4140f | 1734 | unsigned long pgmoved = 0; |
3eb4140f | 1735 | struct page *page; |
fa9add64 | 1736 | int nr_pages; |
3eb4140f | 1737 | |
3eb4140f WF |
1738 | while (!list_empty(list)) { |
1739 | page = lru_to_page(list); | |
fa9add64 | 1740 | lruvec = mem_cgroup_page_lruvec(page, zone); |
3eb4140f | 1741 | |
309381fe | 1742 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3eb4140f WF |
1743 | SetPageLRU(page); |
1744 | ||
fa9add64 HD |
1745 | nr_pages = hpage_nr_pages(page); |
1746 | mem_cgroup_update_lru_size(lruvec, lru, nr_pages); | |
925b7673 | 1747 | list_move(&page->lru, &lruvec->lists[lru]); |
fa9add64 | 1748 | pgmoved += nr_pages; |
3eb4140f | 1749 | |
2bcf8879 HD |
1750 | if (put_page_testzero(page)) { |
1751 | __ClearPageLRU(page); | |
1752 | __ClearPageActive(page); | |
fa9add64 | 1753 | del_page_from_lru_list(page, lruvec, lru); |
2bcf8879 HD |
1754 | |
1755 | if (unlikely(PageCompound(page))) { | |
1756 | spin_unlock_irq(&zone->lru_lock); | |
747db954 | 1757 | mem_cgroup_uncharge(page); |
2bcf8879 HD |
1758 | (*get_compound_page_dtor(page))(page); |
1759 | spin_lock_irq(&zone->lru_lock); | |
1760 | } else | |
1761 | list_add(&page->lru, pages_to_free); | |
3eb4140f WF |
1762 | } |
1763 | } | |
1764 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); | |
1765 | if (!is_active_lru(lru)) | |
1766 | __count_vm_events(PGDEACTIVATE, pgmoved); | |
1767 | } | |
1cfb419b | 1768 | |
f626012d | 1769 | static void shrink_active_list(unsigned long nr_to_scan, |
1a93be0e | 1770 | struct lruvec *lruvec, |
f16015fb | 1771 | struct scan_control *sc, |
9e3b2f8c | 1772 | enum lru_list lru) |
1da177e4 | 1773 | { |
44c241f1 | 1774 | unsigned long nr_taken; |
f626012d | 1775 | unsigned long nr_scanned; |
6fe6b7e3 | 1776 | unsigned long vm_flags; |
1da177e4 | 1777 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
8cab4754 | 1778 | LIST_HEAD(l_active); |
b69408e8 | 1779 | LIST_HEAD(l_inactive); |
1da177e4 | 1780 | struct page *page; |
1a93be0e | 1781 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
44c241f1 | 1782 | unsigned long nr_rotated = 0; |
f3fd4a61 | 1783 | isolate_mode_t isolate_mode = 0; |
3cb99451 | 1784 | int file = is_file_lru(lru); |
1a93be0e | 1785 | struct zone *zone = lruvec_zone(lruvec); |
1da177e4 LT |
1786 | |
1787 | lru_add_drain(); | |
f80c0673 MK |
1788 | |
1789 | if (!sc->may_unmap) | |
61317289 | 1790 | isolate_mode |= ISOLATE_UNMAPPED; |
f80c0673 | 1791 | if (!sc->may_writepage) |
61317289 | 1792 | isolate_mode |= ISOLATE_CLEAN; |
f80c0673 | 1793 | |
1da177e4 | 1794 | spin_lock_irq(&zone->lru_lock); |
925b7673 | 1795 | |
5dc35979 KK |
1796 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, |
1797 | &nr_scanned, sc, isolate_mode, lru); | |
89b5fae5 | 1798 | if (global_reclaim(sc)) |
0d5d823a | 1799 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, nr_scanned); |
89b5fae5 | 1800 | |
b7c46d15 | 1801 | reclaim_stat->recent_scanned[file] += nr_taken; |
1cfb419b | 1802 | |
f626012d | 1803 | __count_zone_vm_events(PGREFILL, zone, nr_scanned); |
3cb99451 | 1804 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
a731286d | 1805 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
1da177e4 LT |
1806 | spin_unlock_irq(&zone->lru_lock); |
1807 | ||
1da177e4 LT |
1808 | while (!list_empty(&l_hold)) { |
1809 | cond_resched(); | |
1810 | page = lru_to_page(&l_hold); | |
1811 | list_del(&page->lru); | |
7e9cd484 | 1812 | |
39b5f29a | 1813 | if (unlikely(!page_evictable(page))) { |
894bc310 LS |
1814 | putback_lru_page(page); |
1815 | continue; | |
1816 | } | |
1817 | ||
cc715d99 MG |
1818 | if (unlikely(buffer_heads_over_limit)) { |
1819 | if (page_has_private(page) && trylock_page(page)) { | |
1820 | if (page_has_private(page)) | |
1821 | try_to_release_page(page, 0); | |
1822 | unlock_page(page); | |
1823 | } | |
1824 | } | |
1825 | ||
c3ac9a8a JW |
1826 | if (page_referenced(page, 0, sc->target_mem_cgroup, |
1827 | &vm_flags)) { | |
9992af10 | 1828 | nr_rotated += hpage_nr_pages(page); |
8cab4754 WF |
1829 | /* |
1830 | * Identify referenced, file-backed active pages and | |
1831 | * give them one more trip around the active list. So | |
1832 | * that executable code get better chances to stay in | |
1833 | * memory under moderate memory pressure. Anon pages | |
1834 | * are not likely to be evicted by use-once streaming | |
1835 | * IO, plus JVM can create lots of anon VM_EXEC pages, | |
1836 | * so we ignore them here. | |
1837 | */ | |
41e20983 | 1838 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
8cab4754 WF |
1839 | list_add(&page->lru, &l_active); |
1840 | continue; | |
1841 | } | |
1842 | } | |
7e9cd484 | 1843 | |
5205e56e | 1844 | ClearPageActive(page); /* we are de-activating */ |
1da177e4 LT |
1845 | list_add(&page->lru, &l_inactive); |
1846 | } | |
1847 | ||
b555749a | 1848 | /* |
8cab4754 | 1849 | * Move pages back to the lru list. |
b555749a | 1850 | */ |
2a1dc509 | 1851 | spin_lock_irq(&zone->lru_lock); |
556adecb | 1852 | /* |
8cab4754 WF |
1853 | * Count referenced pages from currently used mappings as rotated, |
1854 | * even though only some of them are actually re-activated. This | |
1855 | * helps balance scan pressure between file and anonymous pages in | |
7c0db9e9 | 1856 | * get_scan_count. |
7e9cd484 | 1857 | */ |
b7c46d15 | 1858 | reclaim_stat->recent_rotated[file] += nr_rotated; |
556adecb | 1859 | |
fa9add64 HD |
1860 | move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru); |
1861 | move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE); | |
a731286d | 1862 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
f8891e5e | 1863 | spin_unlock_irq(&zone->lru_lock); |
2bcf8879 | 1864 | |
747db954 | 1865 | mem_cgroup_uncharge_list(&l_hold); |
b745bc85 | 1866 | free_hot_cold_page_list(&l_hold, true); |
1da177e4 LT |
1867 | } |
1868 | ||
74e3f3c3 | 1869 | #ifdef CONFIG_SWAP |
42e2e457 | 1870 | static bool inactive_anon_is_low_global(struct zone *zone) |
f89eb90e KM |
1871 | { |
1872 | unsigned long active, inactive; | |
1873 | ||
1874 | active = zone_page_state(zone, NR_ACTIVE_ANON); | |
1875 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1876 | ||
29d06bbb | 1877 | return inactive * zone->inactive_ratio < active; |
f89eb90e KM |
1878 | } |
1879 | ||
14797e23 KM |
1880 | /** |
1881 | * inactive_anon_is_low - check if anonymous pages need to be deactivated | |
c56d5c7d | 1882 | * @lruvec: LRU vector to check |
14797e23 KM |
1883 | * |
1884 | * Returns true if the zone does not have enough inactive anon pages, | |
1885 | * meaning some active anon pages need to be deactivated. | |
1886 | */ | |
42e2e457 | 1887 | static bool inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1888 | { |
74e3f3c3 MK |
1889 | /* |
1890 | * If we don't have swap space, anonymous page deactivation | |
1891 | * is pointless. | |
1892 | */ | |
1893 | if (!total_swap_pages) | |
42e2e457 | 1894 | return false; |
74e3f3c3 | 1895 | |
c3c787e8 | 1896 | if (!mem_cgroup_disabled()) |
c56d5c7d | 1897 | return mem_cgroup_inactive_anon_is_low(lruvec); |
f16015fb | 1898 | |
c56d5c7d | 1899 | return inactive_anon_is_low_global(lruvec_zone(lruvec)); |
14797e23 | 1900 | } |
74e3f3c3 | 1901 | #else |
42e2e457 | 1902 | static inline bool inactive_anon_is_low(struct lruvec *lruvec) |
74e3f3c3 | 1903 | { |
42e2e457 | 1904 | return false; |
74e3f3c3 MK |
1905 | } |
1906 | #endif | |
14797e23 | 1907 | |
56e49d21 RR |
1908 | /** |
1909 | * inactive_file_is_low - check if file pages need to be deactivated | |
c56d5c7d | 1910 | * @lruvec: LRU vector to check |
56e49d21 RR |
1911 | * |
1912 | * When the system is doing streaming IO, memory pressure here | |
1913 | * ensures that active file pages get deactivated, until more | |
1914 | * than half of the file pages are on the inactive list. | |
1915 | * | |
1916 | * Once we get to that situation, protect the system's working | |
1917 | * set from being evicted by disabling active file page aging. | |
1918 | * | |
1919 | * This uses a different ratio than the anonymous pages, because | |
1920 | * the page cache uses a use-once replacement algorithm. | |
1921 | */ | |
42e2e457 | 1922 | static bool inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 | 1923 | { |
e3790144 JW |
1924 | unsigned long inactive; |
1925 | unsigned long active; | |
1926 | ||
1927 | inactive = get_lru_size(lruvec, LRU_INACTIVE_FILE); | |
1928 | active = get_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 | 1929 | |
e3790144 | 1930 | return active > inactive; |
56e49d21 RR |
1931 | } |
1932 | ||
42e2e457 | 1933 | static bool inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru) |
b39415b2 | 1934 | { |
75b00af7 | 1935 | if (is_file_lru(lru)) |
c56d5c7d | 1936 | return inactive_file_is_low(lruvec); |
b39415b2 | 1937 | else |
c56d5c7d | 1938 | return inactive_anon_is_low(lruvec); |
b39415b2 RR |
1939 | } |
1940 | ||
4f98a2fe | 1941 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
1a93be0e | 1942 | struct lruvec *lruvec, struct scan_control *sc) |
b69408e8 | 1943 | { |
b39415b2 | 1944 | if (is_active_lru(lru)) { |
75b00af7 | 1945 | if (inactive_list_is_low(lruvec, lru)) |
1a93be0e | 1946 | shrink_active_list(nr_to_scan, lruvec, sc, lru); |
556adecb RR |
1947 | return 0; |
1948 | } | |
1949 | ||
1a93be0e | 1950 | return shrink_inactive_list(nr_to_scan, lruvec, sc, lru); |
4f98a2fe RR |
1951 | } |
1952 | ||
9a265114 JW |
1953 | enum scan_balance { |
1954 | SCAN_EQUAL, | |
1955 | SCAN_FRACT, | |
1956 | SCAN_ANON, | |
1957 | SCAN_FILE, | |
1958 | }; | |
1959 | ||
4f98a2fe RR |
1960 | /* |
1961 | * Determine how aggressively the anon and file LRU lists should be | |
1962 | * scanned. The relative value of each set of LRU lists is determined | |
1963 | * by looking at the fraction of the pages scanned we did rotate back | |
1964 | * onto the active list instead of evict. | |
1965 | * | |
be7bd59d WL |
1966 | * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan |
1967 | * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan | |
4f98a2fe | 1968 | */ |
33377678 | 1969 | static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg, |
6b4f7799 JW |
1970 | struct scan_control *sc, unsigned long *nr, |
1971 | unsigned long *lru_pages) | |
4f98a2fe | 1972 | { |
33377678 | 1973 | int swappiness = mem_cgroup_swappiness(memcg); |
9a265114 JW |
1974 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1975 | u64 fraction[2]; | |
1976 | u64 denominator = 0; /* gcc */ | |
1977 | struct zone *zone = lruvec_zone(lruvec); | |
4f98a2fe | 1978 | unsigned long anon_prio, file_prio; |
9a265114 | 1979 | enum scan_balance scan_balance; |
0bf1457f | 1980 | unsigned long anon, file; |
9a265114 | 1981 | bool force_scan = false; |
4f98a2fe | 1982 | unsigned long ap, fp; |
4111304d | 1983 | enum lru_list lru; |
6f04f48d SS |
1984 | bool some_scanned; |
1985 | int pass; | |
246e87a9 | 1986 | |
f11c0ca5 JW |
1987 | /* |
1988 | * If the zone or memcg is small, nr[l] can be 0. This | |
1989 | * results in no scanning on this priority and a potential | |
1990 | * priority drop. Global direct reclaim can go to the next | |
1991 | * zone and tends to have no problems. Global kswapd is for | |
1992 | * zone balancing and it needs to scan a minimum amount. When | |
1993 | * reclaiming for a memcg, a priority drop can cause high | |
1994 | * latencies, so it's better to scan a minimum amount there as | |
1995 | * well. | |
1996 | */ | |
90cbc250 VD |
1997 | if (current_is_kswapd()) { |
1998 | if (!zone_reclaimable(zone)) | |
1999 | force_scan = true; | |
eb01aaab | 2000 | if (!mem_cgroup_online(memcg)) |
90cbc250 VD |
2001 | force_scan = true; |
2002 | } | |
89b5fae5 | 2003 | if (!global_reclaim(sc)) |
a4d3e9e7 | 2004 | force_scan = true; |
76a33fc3 SL |
2005 | |
2006 | /* If we have no swap space, do not bother scanning anon pages. */ | |
d8b38438 | 2007 | if (!sc->may_swap || mem_cgroup_get_nr_swap_pages(memcg) <= 0) { |
9a265114 | 2008 | scan_balance = SCAN_FILE; |
76a33fc3 SL |
2009 | goto out; |
2010 | } | |
4f98a2fe | 2011 | |
10316b31 JW |
2012 | /* |
2013 | * Global reclaim will swap to prevent OOM even with no | |
2014 | * swappiness, but memcg users want to use this knob to | |
2015 | * disable swapping for individual groups completely when | |
2016 | * using the memory controller's swap limit feature would be | |
2017 | * too expensive. | |
2018 | */ | |
02695175 | 2019 | if (!global_reclaim(sc) && !swappiness) { |
9a265114 | 2020 | scan_balance = SCAN_FILE; |
10316b31 JW |
2021 | goto out; |
2022 | } | |
2023 | ||
2024 | /* | |
2025 | * Do not apply any pressure balancing cleverness when the | |
2026 | * system is close to OOM, scan both anon and file equally | |
2027 | * (unless the swappiness setting disagrees with swapping). | |
2028 | */ | |
02695175 | 2029 | if (!sc->priority && swappiness) { |
9a265114 | 2030 | scan_balance = SCAN_EQUAL; |
10316b31 JW |
2031 | goto out; |
2032 | } | |
2033 | ||
62376251 JW |
2034 | /* |
2035 | * Prevent the reclaimer from falling into the cache trap: as | |
2036 | * cache pages start out inactive, every cache fault will tip | |
2037 | * the scan balance towards the file LRU. And as the file LRU | |
2038 | * shrinks, so does the window for rotation from references. | |
2039 | * This means we have a runaway feedback loop where a tiny | |
2040 | * thrashing file LRU becomes infinitely more attractive than | |
2041 | * anon pages. Try to detect this based on file LRU size. | |
2042 | */ | |
2043 | if (global_reclaim(sc)) { | |
2ab051e1 JM |
2044 | unsigned long zonefile; |
2045 | unsigned long zonefree; | |
2046 | ||
2047 | zonefree = zone_page_state(zone, NR_FREE_PAGES); | |
2048 | zonefile = zone_page_state(zone, NR_ACTIVE_FILE) + | |
2049 | zone_page_state(zone, NR_INACTIVE_FILE); | |
62376251 | 2050 | |
2ab051e1 | 2051 | if (unlikely(zonefile + zonefree <= high_wmark_pages(zone))) { |
62376251 JW |
2052 | scan_balance = SCAN_ANON; |
2053 | goto out; | |
2054 | } | |
2055 | } | |
2056 | ||
7c5bd705 | 2057 | /* |
316bda0e VD |
2058 | * If there is enough inactive page cache, i.e. if the size of the |
2059 | * inactive list is greater than that of the active list *and* the | |
2060 | * inactive list actually has some pages to scan on this priority, we | |
2061 | * do not reclaim anything from the anonymous working set right now. | |
2062 | * Without the second condition we could end up never scanning an | |
2063 | * lruvec even if it has plenty of old anonymous pages unless the | |
2064 | * system is under heavy pressure. | |
7c5bd705 | 2065 | */ |
316bda0e VD |
2066 | if (!inactive_file_is_low(lruvec) && |
2067 | get_lru_size(lruvec, LRU_INACTIVE_FILE) >> sc->priority) { | |
9a265114 | 2068 | scan_balance = SCAN_FILE; |
7c5bd705 JW |
2069 | goto out; |
2070 | } | |
2071 | ||
9a265114 JW |
2072 | scan_balance = SCAN_FRACT; |
2073 | ||
58c37f6e KM |
2074 | /* |
2075 | * With swappiness at 100, anonymous and file have the same priority. | |
2076 | * This scanning priority is essentially the inverse of IO cost. | |
2077 | */ | |
02695175 | 2078 | anon_prio = swappiness; |
75b00af7 | 2079 | file_prio = 200 - anon_prio; |
58c37f6e | 2080 | |
4f98a2fe RR |
2081 | /* |
2082 | * OK, so we have swap space and a fair amount of page cache | |
2083 | * pages. We use the recently rotated / recently scanned | |
2084 | * ratios to determine how valuable each cache is. | |
2085 | * | |
2086 | * Because workloads change over time (and to avoid overflow) | |
2087 | * we keep these statistics as a floating average, which ends | |
2088 | * up weighing recent references more than old ones. | |
2089 | * | |
2090 | * anon in [0], file in [1] | |
2091 | */ | |
2ab051e1 JM |
2092 | |
2093 | anon = get_lru_size(lruvec, LRU_ACTIVE_ANON) + | |
2094 | get_lru_size(lruvec, LRU_INACTIVE_ANON); | |
2095 | file = get_lru_size(lruvec, LRU_ACTIVE_FILE) + | |
2096 | get_lru_size(lruvec, LRU_INACTIVE_FILE); | |
2097 | ||
90126375 | 2098 | spin_lock_irq(&zone->lru_lock); |
6e901571 | 2099 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
6e901571 KM |
2100 | reclaim_stat->recent_scanned[0] /= 2; |
2101 | reclaim_stat->recent_rotated[0] /= 2; | |
4f98a2fe RR |
2102 | } |
2103 | ||
6e901571 | 2104 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
6e901571 KM |
2105 | reclaim_stat->recent_scanned[1] /= 2; |
2106 | reclaim_stat->recent_rotated[1] /= 2; | |
4f98a2fe RR |
2107 | } |
2108 | ||
4f98a2fe | 2109 | /* |
00d8089c RR |
2110 | * The amount of pressure on anon vs file pages is inversely |
2111 | * proportional to the fraction of recently scanned pages on | |
2112 | * each list that were recently referenced and in active use. | |
4f98a2fe | 2113 | */ |
fe35004f | 2114 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); |
6e901571 | 2115 | ap /= reclaim_stat->recent_rotated[0] + 1; |
4f98a2fe | 2116 | |
fe35004f | 2117 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); |
6e901571 | 2118 | fp /= reclaim_stat->recent_rotated[1] + 1; |
90126375 | 2119 | spin_unlock_irq(&zone->lru_lock); |
4f98a2fe | 2120 | |
76a33fc3 SL |
2121 | fraction[0] = ap; |
2122 | fraction[1] = fp; | |
2123 | denominator = ap + fp + 1; | |
2124 | out: | |
6f04f48d SS |
2125 | some_scanned = false; |
2126 | /* Only use force_scan on second pass. */ | |
2127 | for (pass = 0; !some_scanned && pass < 2; pass++) { | |
6b4f7799 | 2128 | *lru_pages = 0; |
6f04f48d SS |
2129 | for_each_evictable_lru(lru) { |
2130 | int file = is_file_lru(lru); | |
2131 | unsigned long size; | |
2132 | unsigned long scan; | |
6e08a369 | 2133 | |
6f04f48d SS |
2134 | size = get_lru_size(lruvec, lru); |
2135 | scan = size >> sc->priority; | |
9a265114 | 2136 | |
6f04f48d SS |
2137 | if (!scan && pass && force_scan) |
2138 | scan = min(size, SWAP_CLUSTER_MAX); | |
9a265114 | 2139 | |
6f04f48d SS |
2140 | switch (scan_balance) { |
2141 | case SCAN_EQUAL: | |
2142 | /* Scan lists relative to size */ | |
2143 | break; | |
2144 | case SCAN_FRACT: | |
2145 | /* | |
2146 | * Scan types proportional to swappiness and | |
2147 | * their relative recent reclaim efficiency. | |
2148 | */ | |
2149 | scan = div64_u64(scan * fraction[file], | |
2150 | denominator); | |
2151 | break; | |
2152 | case SCAN_FILE: | |
2153 | case SCAN_ANON: | |
2154 | /* Scan one type exclusively */ | |
6b4f7799 JW |
2155 | if ((scan_balance == SCAN_FILE) != file) { |
2156 | size = 0; | |
6f04f48d | 2157 | scan = 0; |
6b4f7799 | 2158 | } |
6f04f48d SS |
2159 | break; |
2160 | default: | |
2161 | /* Look ma, no brain */ | |
2162 | BUG(); | |
2163 | } | |
6b4f7799 JW |
2164 | |
2165 | *lru_pages += size; | |
6f04f48d | 2166 | nr[lru] = scan; |
6b4f7799 | 2167 | |
9a265114 | 2168 | /* |
6f04f48d SS |
2169 | * Skip the second pass and don't force_scan, |
2170 | * if we found something to scan. | |
9a265114 | 2171 | */ |
6f04f48d | 2172 | some_scanned |= !!scan; |
9a265114 | 2173 | } |
76a33fc3 | 2174 | } |
6e08a369 | 2175 | } |
4f98a2fe | 2176 | |
72b252ae MG |
2177 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
2178 | static void init_tlb_ubc(void) | |
2179 | { | |
2180 | /* | |
2181 | * This deliberately does not clear the cpumask as it's expensive | |
2182 | * and unnecessary. If there happens to be data in there then the | |
2183 | * first SWAP_CLUSTER_MAX pages will send an unnecessary IPI and | |
2184 | * then will be cleared. | |
2185 | */ | |
2186 | current->tlb_ubc.flush_required = false; | |
2187 | } | |
2188 | #else | |
2189 | static inline void init_tlb_ubc(void) | |
2190 | { | |
2191 | } | |
2192 | #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ | |
2193 | ||
9b4f98cd JW |
2194 | /* |
2195 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. | |
2196 | */ | |
33377678 VD |
2197 | static void shrink_zone_memcg(struct zone *zone, struct mem_cgroup *memcg, |
2198 | struct scan_control *sc, unsigned long *lru_pages) | |
9b4f98cd | 2199 | { |
33377678 | 2200 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
9b4f98cd | 2201 | unsigned long nr[NR_LRU_LISTS]; |
e82e0561 | 2202 | unsigned long targets[NR_LRU_LISTS]; |
9b4f98cd JW |
2203 | unsigned long nr_to_scan; |
2204 | enum lru_list lru; | |
2205 | unsigned long nr_reclaimed = 0; | |
2206 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; | |
2207 | struct blk_plug plug; | |
1a501907 | 2208 | bool scan_adjusted; |
9b4f98cd | 2209 | |
33377678 | 2210 | get_scan_count(lruvec, memcg, sc, nr, lru_pages); |
9b4f98cd | 2211 | |
e82e0561 MG |
2212 | /* Record the original scan target for proportional adjustments later */ |
2213 | memcpy(targets, nr, sizeof(nr)); | |
2214 | ||
1a501907 MG |
2215 | /* |
2216 | * Global reclaiming within direct reclaim at DEF_PRIORITY is a normal | |
2217 | * event that can occur when there is little memory pressure e.g. | |
2218 | * multiple streaming readers/writers. Hence, we do not abort scanning | |
2219 | * when the requested number of pages are reclaimed when scanning at | |
2220 | * DEF_PRIORITY on the assumption that the fact we are direct | |
2221 | * reclaiming implies that kswapd is not keeping up and it is best to | |
2222 | * do a batch of work at once. For memcg reclaim one check is made to | |
2223 | * abort proportional reclaim if either the file or anon lru has already | |
2224 | * dropped to zero at the first pass. | |
2225 | */ | |
2226 | scan_adjusted = (global_reclaim(sc) && !current_is_kswapd() && | |
2227 | sc->priority == DEF_PRIORITY); | |
2228 | ||
72b252ae MG |
2229 | init_tlb_ubc(); |
2230 | ||
9b4f98cd JW |
2231 | blk_start_plug(&plug); |
2232 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || | |
2233 | nr[LRU_INACTIVE_FILE]) { | |
e82e0561 MG |
2234 | unsigned long nr_anon, nr_file, percentage; |
2235 | unsigned long nr_scanned; | |
2236 | ||
9b4f98cd JW |
2237 | for_each_evictable_lru(lru) { |
2238 | if (nr[lru]) { | |
2239 | nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX); | |
2240 | nr[lru] -= nr_to_scan; | |
2241 | ||
2242 | nr_reclaimed += shrink_list(lru, nr_to_scan, | |
2243 | lruvec, sc); | |
2244 | } | |
2245 | } | |
e82e0561 MG |
2246 | |
2247 | if (nr_reclaimed < nr_to_reclaim || scan_adjusted) | |
2248 | continue; | |
2249 | ||
e82e0561 MG |
2250 | /* |
2251 | * For kswapd and memcg, reclaim at least the number of pages | |
1a501907 | 2252 | * requested. Ensure that the anon and file LRUs are scanned |
e82e0561 MG |
2253 | * proportionally what was requested by get_scan_count(). We |
2254 | * stop reclaiming one LRU and reduce the amount scanning | |
2255 | * proportional to the original scan target. | |
2256 | */ | |
2257 | nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE]; | |
2258 | nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON]; | |
2259 | ||
1a501907 MG |
2260 | /* |
2261 | * It's just vindictive to attack the larger once the smaller | |
2262 | * has gone to zero. And given the way we stop scanning the | |
2263 | * smaller below, this makes sure that we only make one nudge | |
2264 | * towards proportionality once we've got nr_to_reclaim. | |
2265 | */ | |
2266 | if (!nr_file || !nr_anon) | |
2267 | break; | |
2268 | ||
e82e0561 MG |
2269 | if (nr_file > nr_anon) { |
2270 | unsigned long scan_target = targets[LRU_INACTIVE_ANON] + | |
2271 | targets[LRU_ACTIVE_ANON] + 1; | |
2272 | lru = LRU_BASE; | |
2273 | percentage = nr_anon * 100 / scan_target; | |
2274 | } else { | |
2275 | unsigned long scan_target = targets[LRU_INACTIVE_FILE] + | |
2276 | targets[LRU_ACTIVE_FILE] + 1; | |
2277 | lru = LRU_FILE; | |
2278 | percentage = nr_file * 100 / scan_target; | |
2279 | } | |
2280 | ||
2281 | /* Stop scanning the smaller of the LRU */ | |
2282 | nr[lru] = 0; | |
2283 | nr[lru + LRU_ACTIVE] = 0; | |
2284 | ||
2285 | /* | |
2286 | * Recalculate the other LRU scan count based on its original | |
2287 | * scan target and the percentage scanning already complete | |
2288 | */ | |
2289 | lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE; | |
2290 | nr_scanned = targets[lru] - nr[lru]; | |
2291 | nr[lru] = targets[lru] * (100 - percentage) / 100; | |
2292 | nr[lru] -= min(nr[lru], nr_scanned); | |
2293 | ||
2294 | lru += LRU_ACTIVE; | |
2295 | nr_scanned = targets[lru] - nr[lru]; | |
2296 | nr[lru] = targets[lru] * (100 - percentage) / 100; | |
2297 | nr[lru] -= min(nr[lru], nr_scanned); | |
2298 | ||
2299 | scan_adjusted = true; | |
9b4f98cd JW |
2300 | } |
2301 | blk_finish_plug(&plug); | |
2302 | sc->nr_reclaimed += nr_reclaimed; | |
2303 | ||
2304 | /* | |
2305 | * Even if we did not try to evict anon pages at all, we want to | |
2306 | * rebalance the anon lru active/inactive ratio. | |
2307 | */ | |
2308 | if (inactive_anon_is_low(lruvec)) | |
2309 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, | |
2310 | sc, LRU_ACTIVE_ANON); | |
2311 | ||
2312 | throttle_vm_writeout(sc->gfp_mask); | |
2313 | } | |
2314 | ||
23b9da55 | 2315 | /* Use reclaim/compaction for costly allocs or under memory pressure */ |
9e3b2f8c | 2316 | static bool in_reclaim_compaction(struct scan_control *sc) |
23b9da55 | 2317 | { |
d84da3f9 | 2318 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && |
23b9da55 | 2319 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || |
9e3b2f8c | 2320 | sc->priority < DEF_PRIORITY - 2)) |
23b9da55 MG |
2321 | return true; |
2322 | ||
2323 | return false; | |
2324 | } | |
2325 | ||
3e7d3449 | 2326 | /* |
23b9da55 MG |
2327 | * Reclaim/compaction is used for high-order allocation requests. It reclaims |
2328 | * order-0 pages before compacting the zone. should_continue_reclaim() returns | |
2329 | * true if more pages should be reclaimed such that when the page allocator | |
2330 | * calls try_to_compact_zone() that it will have enough free pages to succeed. | |
2331 | * It will give up earlier than that if there is difficulty reclaiming pages. | |
3e7d3449 | 2332 | */ |
9b4f98cd | 2333 | static inline bool should_continue_reclaim(struct zone *zone, |
3e7d3449 MG |
2334 | unsigned long nr_reclaimed, |
2335 | unsigned long nr_scanned, | |
2336 | struct scan_control *sc) | |
2337 | { | |
2338 | unsigned long pages_for_compaction; | |
2339 | unsigned long inactive_lru_pages; | |
2340 | ||
2341 | /* If not in reclaim/compaction mode, stop */ | |
9e3b2f8c | 2342 | if (!in_reclaim_compaction(sc)) |
3e7d3449 MG |
2343 | return false; |
2344 | ||
2876592f MG |
2345 | /* Consider stopping depending on scan and reclaim activity */ |
2346 | if (sc->gfp_mask & __GFP_REPEAT) { | |
2347 | /* | |
2348 | * For __GFP_REPEAT allocations, stop reclaiming if the | |
2349 | * full LRU list has been scanned and we are still failing | |
2350 | * to reclaim pages. This full LRU scan is potentially | |
2351 | * expensive but a __GFP_REPEAT caller really wants to succeed | |
2352 | */ | |
2353 | if (!nr_reclaimed && !nr_scanned) | |
2354 | return false; | |
2355 | } else { | |
2356 | /* | |
2357 | * For non-__GFP_REPEAT allocations which can presumably | |
2358 | * fail without consequence, stop if we failed to reclaim | |
2359 | * any pages from the last SWAP_CLUSTER_MAX number of | |
2360 | * pages that were scanned. This will return to the | |
2361 | * caller faster at the risk reclaim/compaction and | |
2362 | * the resulting allocation attempt fails | |
2363 | */ | |
2364 | if (!nr_reclaimed) | |
2365 | return false; | |
2366 | } | |
3e7d3449 MG |
2367 | |
2368 | /* | |
2369 | * If we have not reclaimed enough pages for compaction and the | |
2370 | * inactive lists are large enough, continue reclaiming | |
2371 | */ | |
2372 | pages_for_compaction = (2UL << sc->order); | |
9b4f98cd | 2373 | inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE); |
ec8acf20 | 2374 | if (get_nr_swap_pages() > 0) |
9b4f98cd | 2375 | inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON); |
3e7d3449 MG |
2376 | if (sc->nr_reclaimed < pages_for_compaction && |
2377 | inactive_lru_pages > pages_for_compaction) | |
2378 | return true; | |
2379 | ||
2380 | /* If compaction would go ahead or the allocation would succeed, stop */ | |
ebff3980 | 2381 | switch (compaction_suitable(zone, sc->order, 0, 0)) { |
3e7d3449 MG |
2382 | case COMPACT_PARTIAL: |
2383 | case COMPACT_CONTINUE: | |
2384 | return false; | |
2385 | default: | |
2386 | return true; | |
2387 | } | |
2388 | } | |
2389 | ||
6b4f7799 JW |
2390 | static bool shrink_zone(struct zone *zone, struct scan_control *sc, |
2391 | bool is_classzone) | |
1da177e4 | 2392 | { |
cb731d6c | 2393 | struct reclaim_state *reclaim_state = current->reclaim_state; |
f0fdc5e8 | 2394 | unsigned long nr_reclaimed, nr_scanned; |
2344d7e4 | 2395 | bool reclaimable = false; |
1da177e4 | 2396 | |
9b4f98cd JW |
2397 | do { |
2398 | struct mem_cgroup *root = sc->target_mem_cgroup; | |
2399 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2400 | .zone = zone, | |
2401 | .priority = sc->priority, | |
2402 | }; | |
6b4f7799 | 2403 | unsigned long zone_lru_pages = 0; |
694fbc0f | 2404 | struct mem_cgroup *memcg; |
3e7d3449 | 2405 | |
9b4f98cd JW |
2406 | nr_reclaimed = sc->nr_reclaimed; |
2407 | nr_scanned = sc->nr_scanned; | |
1da177e4 | 2408 | |
694fbc0f AM |
2409 | memcg = mem_cgroup_iter(root, NULL, &reclaim); |
2410 | do { | |
6b4f7799 | 2411 | unsigned long lru_pages; |
8e8ae645 | 2412 | unsigned long reclaimed; |
cb731d6c | 2413 | unsigned long scanned; |
5660048c | 2414 | |
241994ed JW |
2415 | if (mem_cgroup_low(root, memcg)) { |
2416 | if (!sc->may_thrash) | |
2417 | continue; | |
2418 | mem_cgroup_events(memcg, MEMCG_LOW, 1); | |
2419 | } | |
2420 | ||
8e8ae645 | 2421 | reclaimed = sc->nr_reclaimed; |
cb731d6c | 2422 | scanned = sc->nr_scanned; |
f9be23d6 | 2423 | |
33377678 | 2424 | shrink_zone_memcg(zone, memcg, sc, &lru_pages); |
6b4f7799 | 2425 | zone_lru_pages += lru_pages; |
f16015fb | 2426 | |
cb731d6c VD |
2427 | if (memcg && is_classzone) |
2428 | shrink_slab(sc->gfp_mask, zone_to_nid(zone), | |
2429 | memcg, sc->nr_scanned - scanned, | |
2430 | lru_pages); | |
2431 | ||
8e8ae645 JW |
2432 | /* Record the group's reclaim efficiency */ |
2433 | vmpressure(sc->gfp_mask, memcg, false, | |
2434 | sc->nr_scanned - scanned, | |
2435 | sc->nr_reclaimed - reclaimed); | |
2436 | ||
9b4f98cd | 2437 | /* |
a394cb8e MH |
2438 | * Direct reclaim and kswapd have to scan all memory |
2439 | * cgroups to fulfill the overall scan target for the | |
9b4f98cd | 2440 | * zone. |
a394cb8e MH |
2441 | * |
2442 | * Limit reclaim, on the other hand, only cares about | |
2443 | * nr_to_reclaim pages to be reclaimed and it will | |
2444 | * retry with decreasing priority if one round over the | |
2445 | * whole hierarchy is not sufficient. | |
9b4f98cd | 2446 | */ |
a394cb8e MH |
2447 | if (!global_reclaim(sc) && |
2448 | sc->nr_reclaimed >= sc->nr_to_reclaim) { | |
9b4f98cd JW |
2449 | mem_cgroup_iter_break(root, memcg); |
2450 | break; | |
2451 | } | |
241994ed | 2452 | } while ((memcg = mem_cgroup_iter(root, memcg, &reclaim))); |
70ddf637 | 2453 | |
6b4f7799 JW |
2454 | /* |
2455 | * Shrink the slab caches in the same proportion that | |
2456 | * the eligible LRU pages were scanned. | |
2457 | */ | |
cb731d6c VD |
2458 | if (global_reclaim(sc) && is_classzone) |
2459 | shrink_slab(sc->gfp_mask, zone_to_nid(zone), NULL, | |
2460 | sc->nr_scanned - nr_scanned, | |
2461 | zone_lru_pages); | |
2462 | ||
2463 | if (reclaim_state) { | |
2464 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; | |
2465 | reclaim_state->reclaimed_slab = 0; | |
6b4f7799 JW |
2466 | } |
2467 | ||
8e8ae645 JW |
2468 | /* Record the subtree's reclaim efficiency */ |
2469 | vmpressure(sc->gfp_mask, sc->target_mem_cgroup, true, | |
70ddf637 AV |
2470 | sc->nr_scanned - nr_scanned, |
2471 | sc->nr_reclaimed - nr_reclaimed); | |
2472 | ||
2344d7e4 JW |
2473 | if (sc->nr_reclaimed - nr_reclaimed) |
2474 | reclaimable = true; | |
2475 | ||
9b4f98cd JW |
2476 | } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed, |
2477 | sc->nr_scanned - nr_scanned, sc)); | |
2344d7e4 JW |
2478 | |
2479 | return reclaimable; | |
f16015fb JW |
2480 | } |
2481 | ||
53853e2d VB |
2482 | /* |
2483 | * Returns true if compaction should go ahead for a high-order request, or | |
2484 | * the high-order allocation would succeed without compaction. | |
2485 | */ | |
0b06496a | 2486 | static inline bool compaction_ready(struct zone *zone, int order) |
fe4b1b24 MG |
2487 | { |
2488 | unsigned long balance_gap, watermark; | |
2489 | bool watermark_ok; | |
2490 | ||
fe4b1b24 MG |
2491 | /* |
2492 | * Compaction takes time to run and there are potentially other | |
2493 | * callers using the pages just freed. Continue reclaiming until | |
2494 | * there is a buffer of free pages available to give compaction | |
2495 | * a reasonable chance of completing and allocating the page | |
2496 | */ | |
4be89a34 JZ |
2497 | balance_gap = min(low_wmark_pages(zone), DIV_ROUND_UP( |
2498 | zone->managed_pages, KSWAPD_ZONE_BALANCE_GAP_RATIO)); | |
0b06496a | 2499 | watermark = high_wmark_pages(zone) + balance_gap + (2UL << order); |
e2b19197 | 2500 | watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0); |
fe4b1b24 MG |
2501 | |
2502 | /* | |
2503 | * If compaction is deferred, reclaim up to a point where | |
2504 | * compaction will have a chance of success when re-enabled | |
2505 | */ | |
0b06496a | 2506 | if (compaction_deferred(zone, order)) |
fe4b1b24 MG |
2507 | return watermark_ok; |
2508 | ||
53853e2d VB |
2509 | /* |
2510 | * If compaction is not ready to start and allocation is not likely | |
2511 | * to succeed without it, then keep reclaiming. | |
2512 | */ | |
ebff3980 | 2513 | if (compaction_suitable(zone, order, 0, 0) == COMPACT_SKIPPED) |
fe4b1b24 MG |
2514 | return false; |
2515 | ||
2516 | return watermark_ok; | |
2517 | } | |
2518 | ||
1da177e4 LT |
2519 | /* |
2520 | * This is the direct reclaim path, for page-allocating processes. We only | |
2521 | * try to reclaim pages from zones which will satisfy the caller's allocation | |
2522 | * request. | |
2523 | * | |
41858966 MG |
2524 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). |
2525 | * Because: | |
1da177e4 LT |
2526 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
2527 | * allocation or | |
41858966 MG |
2528 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones |
2529 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' | |
2530 | * zone defense algorithm. | |
1da177e4 | 2531 | * |
1da177e4 LT |
2532 | * If a zone is deemed to be full of pinned pages then just give it a light |
2533 | * scan then give up on it. | |
2344d7e4 JW |
2534 | * |
2535 | * Returns true if a zone was reclaimable. | |
1da177e4 | 2536 | */ |
2344d7e4 | 2537 | static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc) |
1da177e4 | 2538 | { |
dd1a239f | 2539 | struct zoneref *z; |
54a6eb5c | 2540 | struct zone *zone; |
0608f43d AM |
2541 | unsigned long nr_soft_reclaimed; |
2542 | unsigned long nr_soft_scanned; | |
619d0d76 | 2543 | gfp_t orig_mask; |
9bbc04ee | 2544 | enum zone_type requested_highidx = gfp_zone(sc->gfp_mask); |
2344d7e4 | 2545 | bool reclaimable = false; |
1cfb419b | 2546 | |
cc715d99 MG |
2547 | /* |
2548 | * If the number of buffer_heads in the machine exceeds the maximum | |
2549 | * allowed level, force direct reclaim to scan the highmem zone as | |
2550 | * highmem pages could be pinning lowmem pages storing buffer_heads | |
2551 | */ | |
619d0d76 | 2552 | orig_mask = sc->gfp_mask; |
cc715d99 MG |
2553 | if (buffer_heads_over_limit) |
2554 | sc->gfp_mask |= __GFP_HIGHMEM; | |
2555 | ||
d4debc66 | 2556 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
6b4f7799 JW |
2557 | requested_highidx, sc->nodemask) { |
2558 | enum zone_type classzone_idx; | |
2559 | ||
f3fe6512 | 2560 | if (!populated_zone(zone)) |
1da177e4 | 2561 | continue; |
6b4f7799 JW |
2562 | |
2563 | classzone_idx = requested_highidx; | |
2564 | while (!populated_zone(zone->zone_pgdat->node_zones + | |
2565 | classzone_idx)) | |
2566 | classzone_idx--; | |
2567 | ||
1cfb419b KH |
2568 | /* |
2569 | * Take care memory controller reclaiming has small influence | |
2570 | * to global LRU. | |
2571 | */ | |
89b5fae5 | 2572 | if (global_reclaim(sc)) { |
344736f2 VD |
2573 | if (!cpuset_zone_allowed(zone, |
2574 | GFP_KERNEL | __GFP_HARDWALL)) | |
1cfb419b | 2575 | continue; |
65ec02cb | 2576 | |
6e543d57 LD |
2577 | if (sc->priority != DEF_PRIORITY && |
2578 | !zone_reclaimable(zone)) | |
1cfb419b | 2579 | continue; /* Let kswapd poll it */ |
0b06496a JW |
2580 | |
2581 | /* | |
2582 | * If we already have plenty of memory free for | |
2583 | * compaction in this zone, don't free any more. | |
2584 | * Even though compaction is invoked for any | |
2585 | * non-zero order, only frequent costly order | |
2586 | * reclamation is disruptive enough to become a | |
2587 | * noticeable problem, like transparent huge | |
2588 | * page allocations. | |
2589 | */ | |
2590 | if (IS_ENABLED(CONFIG_COMPACTION) && | |
2591 | sc->order > PAGE_ALLOC_COSTLY_ORDER && | |
2592 | zonelist_zone_idx(z) <= requested_highidx && | |
2593 | compaction_ready(zone, sc->order)) { | |
2594 | sc->compaction_ready = true; | |
2595 | continue; | |
e0887c19 | 2596 | } |
0b06496a | 2597 | |
0608f43d AM |
2598 | /* |
2599 | * This steals pages from memory cgroups over softlimit | |
2600 | * and returns the number of reclaimed pages and | |
2601 | * scanned pages. This works for global memory pressure | |
2602 | * and balancing, not for a memcg's limit. | |
2603 | */ | |
2604 | nr_soft_scanned = 0; | |
2605 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | |
2606 | sc->order, sc->gfp_mask, | |
2607 | &nr_soft_scanned); | |
2608 | sc->nr_reclaimed += nr_soft_reclaimed; | |
2609 | sc->nr_scanned += nr_soft_scanned; | |
2344d7e4 JW |
2610 | if (nr_soft_reclaimed) |
2611 | reclaimable = true; | |
ac34a1a3 | 2612 | /* need some check for avoid more shrink_zone() */ |
1cfb419b | 2613 | } |
408d8544 | 2614 | |
6b4f7799 | 2615 | if (shrink_zone(zone, sc, zone_idx(zone) == classzone_idx)) |
2344d7e4 JW |
2616 | reclaimable = true; |
2617 | ||
2618 | if (global_reclaim(sc) && | |
2619 | !reclaimable && zone_reclaimable(zone)) | |
2620 | reclaimable = true; | |
1da177e4 | 2621 | } |
e0c23279 | 2622 | |
619d0d76 WY |
2623 | /* |
2624 | * Restore to original mask to avoid the impact on the caller if we | |
2625 | * promoted it to __GFP_HIGHMEM. | |
2626 | */ | |
2627 | sc->gfp_mask = orig_mask; | |
d1908362 | 2628 | |
2344d7e4 | 2629 | return reclaimable; |
1da177e4 | 2630 | } |
4f98a2fe | 2631 | |
1da177e4 LT |
2632 | /* |
2633 | * This is the main entry point to direct page reclaim. | |
2634 | * | |
2635 | * If a full scan of the inactive list fails to free enough memory then we | |
2636 | * are "out of memory" and something needs to be killed. | |
2637 | * | |
2638 | * If the caller is !__GFP_FS then the probability of a failure is reasonably | |
2639 | * high - the zone may be full of dirty or under-writeback pages, which this | |
5b0830cb JA |
2640 | * caller can't do much about. We kick the writeback threads and take explicit |
2641 | * naps in the hope that some of these pages can be written. But if the | |
2642 | * allocating task holds filesystem locks which prevent writeout this might not | |
2643 | * work, and the allocation attempt will fail. | |
a41f24ea NA |
2644 | * |
2645 | * returns: 0, if no pages reclaimed | |
2646 | * else, the number of pages reclaimed | |
1da177e4 | 2647 | */ |
dac1d27b | 2648 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
3115cd91 | 2649 | struct scan_control *sc) |
1da177e4 | 2650 | { |
241994ed | 2651 | int initial_priority = sc->priority; |
69e05944 | 2652 | unsigned long total_scanned = 0; |
22fba335 | 2653 | unsigned long writeback_threshold; |
2344d7e4 | 2654 | bool zones_reclaimable; |
241994ed | 2655 | retry: |
873b4771 KK |
2656 | delayacct_freepages_start(); |
2657 | ||
89b5fae5 | 2658 | if (global_reclaim(sc)) |
1cfb419b | 2659 | count_vm_event(ALLOCSTALL); |
1da177e4 | 2660 | |
9e3b2f8c | 2661 | do { |
70ddf637 AV |
2662 | vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup, |
2663 | sc->priority); | |
66e1707b | 2664 | sc->nr_scanned = 0; |
2344d7e4 | 2665 | zones_reclaimable = shrink_zones(zonelist, sc); |
c6a8a8c5 | 2666 | |
66e1707b | 2667 | total_scanned += sc->nr_scanned; |
bb21c7ce | 2668 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
0b06496a JW |
2669 | break; |
2670 | ||
2671 | if (sc->compaction_ready) | |
2672 | break; | |
1da177e4 | 2673 | |
0e50ce3b MK |
2674 | /* |
2675 | * If we're getting trouble reclaiming, start doing | |
2676 | * writepage even in laptop mode. | |
2677 | */ | |
2678 | if (sc->priority < DEF_PRIORITY - 2) | |
2679 | sc->may_writepage = 1; | |
2680 | ||
1da177e4 LT |
2681 | /* |
2682 | * Try to write back as many pages as we just scanned. This | |
2683 | * tends to cause slow streaming writers to write data to the | |
2684 | * disk smoothly, at the dirtying rate, which is nice. But | |
2685 | * that's undesirable in laptop mode, where we *want* lumpy | |
2686 | * writeout. So in laptop mode, write out the whole world. | |
2687 | */ | |
22fba335 KM |
2688 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; |
2689 | if (total_scanned > writeback_threshold) { | |
0e175a18 CW |
2690 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, |
2691 | WB_REASON_TRY_TO_FREE_PAGES); | |
66e1707b | 2692 | sc->may_writepage = 1; |
1da177e4 | 2693 | } |
0b06496a | 2694 | } while (--sc->priority >= 0); |
bb21c7ce | 2695 | |
873b4771 KK |
2696 | delayacct_freepages_end(); |
2697 | ||
bb21c7ce KM |
2698 | if (sc->nr_reclaimed) |
2699 | return sc->nr_reclaimed; | |
2700 | ||
0cee34fd | 2701 | /* Aborted reclaim to try compaction? don't OOM, then */ |
0b06496a | 2702 | if (sc->compaction_ready) |
7335084d MG |
2703 | return 1; |
2704 | ||
241994ed JW |
2705 | /* Untapped cgroup reserves? Don't OOM, retry. */ |
2706 | if (!sc->may_thrash) { | |
2707 | sc->priority = initial_priority; | |
2708 | sc->may_thrash = 1; | |
2709 | goto retry; | |
2710 | } | |
2711 | ||
2344d7e4 JW |
2712 | /* Any of the zones still reclaimable? Don't OOM. */ |
2713 | if (zones_reclaimable) | |
bb21c7ce KM |
2714 | return 1; |
2715 | ||
2716 | return 0; | |
1da177e4 LT |
2717 | } |
2718 | ||
5515061d MG |
2719 | static bool pfmemalloc_watermark_ok(pg_data_t *pgdat) |
2720 | { | |
2721 | struct zone *zone; | |
2722 | unsigned long pfmemalloc_reserve = 0; | |
2723 | unsigned long free_pages = 0; | |
2724 | int i; | |
2725 | bool wmark_ok; | |
2726 | ||
2727 | for (i = 0; i <= ZONE_NORMAL; i++) { | |
2728 | zone = &pgdat->node_zones[i]; | |
f012a84a NA |
2729 | if (!populated_zone(zone) || |
2730 | zone_reclaimable_pages(zone) == 0) | |
675becce MG |
2731 | continue; |
2732 | ||
5515061d MG |
2733 | pfmemalloc_reserve += min_wmark_pages(zone); |
2734 | free_pages += zone_page_state(zone, NR_FREE_PAGES); | |
2735 | } | |
2736 | ||
675becce MG |
2737 | /* If there are no reserves (unexpected config) then do not throttle */ |
2738 | if (!pfmemalloc_reserve) | |
2739 | return true; | |
2740 | ||
5515061d MG |
2741 | wmark_ok = free_pages > pfmemalloc_reserve / 2; |
2742 | ||
2743 | /* kswapd must be awake if processes are being throttled */ | |
2744 | if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) { | |
2745 | pgdat->classzone_idx = min(pgdat->classzone_idx, | |
2746 | (enum zone_type)ZONE_NORMAL); | |
2747 | wake_up_interruptible(&pgdat->kswapd_wait); | |
2748 | } | |
2749 | ||
2750 | return wmark_ok; | |
2751 | } | |
2752 | ||
2753 | /* | |
2754 | * Throttle direct reclaimers if backing storage is backed by the network | |
2755 | * and the PFMEMALLOC reserve for the preferred node is getting dangerously | |
2756 | * depleted. kswapd will continue to make progress and wake the processes | |
50694c28 MG |
2757 | * when the low watermark is reached. |
2758 | * | |
2759 | * Returns true if a fatal signal was delivered during throttling. If this | |
2760 | * happens, the page allocator should not consider triggering the OOM killer. | |
5515061d | 2761 | */ |
50694c28 | 2762 | static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist, |
5515061d MG |
2763 | nodemask_t *nodemask) |
2764 | { | |
675becce | 2765 | struct zoneref *z; |
5515061d | 2766 | struct zone *zone; |
675becce | 2767 | pg_data_t *pgdat = NULL; |
5515061d MG |
2768 | |
2769 | /* | |
2770 | * Kernel threads should not be throttled as they may be indirectly | |
2771 | * responsible for cleaning pages necessary for reclaim to make forward | |
2772 | * progress. kjournald for example may enter direct reclaim while | |
2773 | * committing a transaction where throttling it could forcing other | |
2774 | * processes to block on log_wait_commit(). | |
2775 | */ | |
2776 | if (current->flags & PF_KTHREAD) | |
50694c28 MG |
2777 | goto out; |
2778 | ||
2779 | /* | |
2780 | * If a fatal signal is pending, this process should not throttle. | |
2781 | * It should return quickly so it can exit and free its memory | |
2782 | */ | |
2783 | if (fatal_signal_pending(current)) | |
2784 | goto out; | |
5515061d | 2785 | |
675becce MG |
2786 | /* |
2787 | * Check if the pfmemalloc reserves are ok by finding the first node | |
2788 | * with a usable ZONE_NORMAL or lower zone. The expectation is that | |
2789 | * GFP_KERNEL will be required for allocating network buffers when | |
2790 | * swapping over the network so ZONE_HIGHMEM is unusable. | |
2791 | * | |
2792 | * Throttling is based on the first usable node and throttled processes | |
2793 | * wait on a queue until kswapd makes progress and wakes them. There | |
2794 | * is an affinity then between processes waking up and where reclaim | |
2795 | * progress has been made assuming the process wakes on the same node. | |
2796 | * More importantly, processes running on remote nodes will not compete | |
2797 | * for remote pfmemalloc reserves and processes on different nodes | |
2798 | * should make reasonable progress. | |
2799 | */ | |
2800 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
17636faa | 2801 | gfp_zone(gfp_mask), nodemask) { |
675becce MG |
2802 | if (zone_idx(zone) > ZONE_NORMAL) |
2803 | continue; | |
2804 | ||
2805 | /* Throttle based on the first usable node */ | |
2806 | pgdat = zone->zone_pgdat; | |
2807 | if (pfmemalloc_watermark_ok(pgdat)) | |
2808 | goto out; | |
2809 | break; | |
2810 | } | |
2811 | ||
2812 | /* If no zone was usable by the allocation flags then do not throttle */ | |
2813 | if (!pgdat) | |
50694c28 | 2814 | goto out; |
5515061d | 2815 | |
68243e76 MG |
2816 | /* Account for the throttling */ |
2817 | count_vm_event(PGSCAN_DIRECT_THROTTLE); | |
2818 | ||
5515061d MG |
2819 | /* |
2820 | * If the caller cannot enter the filesystem, it's possible that it | |
2821 | * is due to the caller holding an FS lock or performing a journal | |
2822 | * transaction in the case of a filesystem like ext[3|4]. In this case, | |
2823 | * it is not safe to block on pfmemalloc_wait as kswapd could be | |
2824 | * blocked waiting on the same lock. Instead, throttle for up to a | |
2825 | * second before continuing. | |
2826 | */ | |
2827 | if (!(gfp_mask & __GFP_FS)) { | |
2828 | wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, | |
2829 | pfmemalloc_watermark_ok(pgdat), HZ); | |
50694c28 MG |
2830 | |
2831 | goto check_pending; | |
5515061d MG |
2832 | } |
2833 | ||
2834 | /* Throttle until kswapd wakes the process */ | |
2835 | wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, | |
2836 | pfmemalloc_watermark_ok(pgdat)); | |
50694c28 MG |
2837 | |
2838 | check_pending: | |
2839 | if (fatal_signal_pending(current)) | |
2840 | return true; | |
2841 | ||
2842 | out: | |
2843 | return false; | |
5515061d MG |
2844 | } |
2845 | ||
dac1d27b | 2846 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
327c0e96 | 2847 | gfp_t gfp_mask, nodemask_t *nodemask) |
66e1707b | 2848 | { |
33906bc5 | 2849 | unsigned long nr_reclaimed; |
66e1707b | 2850 | struct scan_control sc = { |
ee814fe2 | 2851 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
21caf2fc | 2852 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
ee814fe2 JW |
2853 | .order = order, |
2854 | .nodemask = nodemask, | |
2855 | .priority = DEF_PRIORITY, | |
66e1707b | 2856 | .may_writepage = !laptop_mode, |
a6dc60f8 | 2857 | .may_unmap = 1, |
2e2e4259 | 2858 | .may_swap = 1, |
66e1707b BS |
2859 | }; |
2860 | ||
5515061d | 2861 | /* |
50694c28 MG |
2862 | * Do not enter reclaim if fatal signal was delivered while throttled. |
2863 | * 1 is returned so that the page allocator does not OOM kill at this | |
2864 | * point. | |
5515061d | 2865 | */ |
50694c28 | 2866 | if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask)) |
5515061d MG |
2867 | return 1; |
2868 | ||
33906bc5 MG |
2869 | trace_mm_vmscan_direct_reclaim_begin(order, |
2870 | sc.may_writepage, | |
2871 | gfp_mask); | |
2872 | ||
3115cd91 | 2873 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
33906bc5 MG |
2874 | |
2875 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); | |
2876 | ||
2877 | return nr_reclaimed; | |
66e1707b BS |
2878 | } |
2879 | ||
c255a458 | 2880 | #ifdef CONFIG_MEMCG |
66e1707b | 2881 | |
72835c86 | 2882 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg, |
4e416953 | 2883 | gfp_t gfp_mask, bool noswap, |
0ae5e89c YH |
2884 | struct zone *zone, |
2885 | unsigned long *nr_scanned) | |
4e416953 BS |
2886 | { |
2887 | struct scan_control sc = { | |
b8f5c566 | 2888 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
ee814fe2 | 2889 | .target_mem_cgroup = memcg, |
4e416953 BS |
2890 | .may_writepage = !laptop_mode, |
2891 | .may_unmap = 1, | |
2892 | .may_swap = !noswap, | |
4e416953 | 2893 | }; |
6b4f7799 | 2894 | unsigned long lru_pages; |
0ae5e89c | 2895 | |
4e416953 BS |
2896 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2897 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); | |
bdce6d9e | 2898 | |
9e3b2f8c | 2899 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, |
bdce6d9e KM |
2900 | sc.may_writepage, |
2901 | sc.gfp_mask); | |
2902 | ||
4e416953 BS |
2903 | /* |
2904 | * NOTE: Although we can get the priority field, using it | |
2905 | * here is not a good idea, since it limits the pages we can scan. | |
2906 | * if we don't reclaim here, the shrink_zone from balance_pgdat | |
2907 | * will pick up pages from other mem cgroup's as well. We hack | |
2908 | * the priority and make it zero. | |
2909 | */ | |
33377678 | 2910 | shrink_zone_memcg(zone, memcg, &sc, &lru_pages); |
bdce6d9e KM |
2911 | |
2912 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); | |
2913 | ||
0ae5e89c | 2914 | *nr_scanned = sc.nr_scanned; |
4e416953 BS |
2915 | return sc.nr_reclaimed; |
2916 | } | |
2917 | ||
72835c86 | 2918 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, |
b70a2a21 | 2919 | unsigned long nr_pages, |
a7885eb8 | 2920 | gfp_t gfp_mask, |
b70a2a21 | 2921 | bool may_swap) |
66e1707b | 2922 | { |
4e416953 | 2923 | struct zonelist *zonelist; |
bdce6d9e | 2924 | unsigned long nr_reclaimed; |
889976db | 2925 | int nid; |
66e1707b | 2926 | struct scan_control sc = { |
b70a2a21 | 2927 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
a09ed5e0 YH |
2928 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2929 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), | |
ee814fe2 JW |
2930 | .target_mem_cgroup = memcg, |
2931 | .priority = DEF_PRIORITY, | |
2932 | .may_writepage = !laptop_mode, | |
2933 | .may_unmap = 1, | |
b70a2a21 | 2934 | .may_swap = may_swap, |
a09ed5e0 | 2935 | }; |
66e1707b | 2936 | |
889976db YH |
2937 | /* |
2938 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't | |
2939 | * take care of from where we get pages. So the node where we start the | |
2940 | * scan does not need to be the current node. | |
2941 | */ | |
72835c86 | 2942 | nid = mem_cgroup_select_victim_node(memcg); |
889976db YH |
2943 | |
2944 | zonelist = NODE_DATA(nid)->node_zonelists; | |
bdce6d9e KM |
2945 | |
2946 | trace_mm_vmscan_memcg_reclaim_begin(0, | |
2947 | sc.may_writepage, | |
2948 | sc.gfp_mask); | |
2949 | ||
3115cd91 | 2950 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
bdce6d9e KM |
2951 | |
2952 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); | |
2953 | ||
2954 | return nr_reclaimed; | |
66e1707b BS |
2955 | } |
2956 | #endif | |
2957 | ||
9e3b2f8c | 2958 | static void age_active_anon(struct zone *zone, struct scan_control *sc) |
f16015fb | 2959 | { |
b95a2f2d | 2960 | struct mem_cgroup *memcg; |
f16015fb | 2961 | |
b95a2f2d JW |
2962 | if (!total_swap_pages) |
2963 | return; | |
2964 | ||
2965 | memcg = mem_cgroup_iter(NULL, NULL, NULL); | |
2966 | do { | |
c56d5c7d | 2967 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
b95a2f2d | 2968 | |
c56d5c7d | 2969 | if (inactive_anon_is_low(lruvec)) |
1a93be0e | 2970 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
9e3b2f8c | 2971 | sc, LRU_ACTIVE_ANON); |
b95a2f2d JW |
2972 | |
2973 | memcg = mem_cgroup_iter(NULL, memcg, NULL); | |
2974 | } while (memcg); | |
f16015fb JW |
2975 | } |
2976 | ||
60cefed4 JW |
2977 | static bool zone_balanced(struct zone *zone, int order, |
2978 | unsigned long balance_gap, int classzone_idx) | |
2979 | { | |
2980 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) + | |
e2b19197 | 2981 | balance_gap, classzone_idx)) |
60cefed4 JW |
2982 | return false; |
2983 | ||
ebff3980 VB |
2984 | if (IS_ENABLED(CONFIG_COMPACTION) && order && compaction_suitable(zone, |
2985 | order, 0, classzone_idx) == COMPACT_SKIPPED) | |
60cefed4 JW |
2986 | return false; |
2987 | ||
2988 | return true; | |
2989 | } | |
2990 | ||
1741c877 | 2991 | /* |
4ae0a48b ZC |
2992 | * pgdat_balanced() is used when checking if a node is balanced. |
2993 | * | |
2994 | * For order-0, all zones must be balanced! | |
2995 | * | |
2996 | * For high-order allocations only zones that meet watermarks and are in a | |
2997 | * zone allowed by the callers classzone_idx are added to balanced_pages. The | |
2998 | * total of balanced pages must be at least 25% of the zones allowed by | |
2999 | * classzone_idx for the node to be considered balanced. Forcing all zones to | |
3000 | * be balanced for high orders can cause excessive reclaim when there are | |
3001 | * imbalanced zones. | |
1741c877 MG |
3002 | * The choice of 25% is due to |
3003 | * o a 16M DMA zone that is balanced will not balance a zone on any | |
3004 | * reasonable sized machine | |
3005 | * o On all other machines, the top zone must be at least a reasonable | |
25985edc | 3006 | * percentage of the middle zones. For example, on 32-bit x86, highmem |
1741c877 MG |
3007 | * would need to be at least 256M for it to be balance a whole node. |
3008 | * Similarly, on x86-64 the Normal zone would need to be at least 1G | |
3009 | * to balance a node on its own. These seemed like reasonable ratios. | |
3010 | */ | |
4ae0a48b | 3011 | static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx) |
1741c877 | 3012 | { |
b40da049 | 3013 | unsigned long managed_pages = 0; |
4ae0a48b | 3014 | unsigned long balanced_pages = 0; |
1741c877 MG |
3015 | int i; |
3016 | ||
4ae0a48b ZC |
3017 | /* Check the watermark levels */ |
3018 | for (i = 0; i <= classzone_idx; i++) { | |
3019 | struct zone *zone = pgdat->node_zones + i; | |
1741c877 | 3020 | |
4ae0a48b ZC |
3021 | if (!populated_zone(zone)) |
3022 | continue; | |
3023 | ||
b40da049 | 3024 | managed_pages += zone->managed_pages; |
4ae0a48b ZC |
3025 | |
3026 | /* | |
3027 | * A special case here: | |
3028 | * | |
3029 | * balance_pgdat() skips over all_unreclaimable after | |
3030 | * DEF_PRIORITY. Effectively, it considers them balanced so | |
3031 | * they must be considered balanced here as well! | |
3032 | */ | |
6e543d57 | 3033 | if (!zone_reclaimable(zone)) { |
b40da049 | 3034 | balanced_pages += zone->managed_pages; |
4ae0a48b ZC |
3035 | continue; |
3036 | } | |
3037 | ||
3038 | if (zone_balanced(zone, order, 0, i)) | |
b40da049 | 3039 | balanced_pages += zone->managed_pages; |
4ae0a48b ZC |
3040 | else if (!order) |
3041 | return false; | |
3042 | } | |
3043 | ||
3044 | if (order) | |
b40da049 | 3045 | return balanced_pages >= (managed_pages >> 2); |
4ae0a48b ZC |
3046 | else |
3047 | return true; | |
1741c877 MG |
3048 | } |
3049 | ||
5515061d MG |
3050 | /* |
3051 | * Prepare kswapd for sleeping. This verifies that there are no processes | |
3052 | * waiting in throttle_direct_reclaim() and that watermarks have been met. | |
3053 | * | |
3054 | * Returns true if kswapd is ready to sleep | |
3055 | */ | |
3056 | static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining, | |
dc83edd9 | 3057 | int classzone_idx) |
f50de2d3 | 3058 | { |
f50de2d3 MG |
3059 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ |
3060 | if (remaining) | |
5515061d MG |
3061 | return false; |
3062 | ||
3063 | /* | |
9e5e3661 VB |
3064 | * The throttled processes are normally woken up in balance_pgdat() as |
3065 | * soon as pfmemalloc_watermark_ok() is true. But there is a potential | |
3066 | * race between when kswapd checks the watermarks and a process gets | |
3067 | * throttled. There is also a potential race if processes get | |
3068 | * throttled, kswapd wakes, a large process exits thereby balancing the | |
3069 | * zones, which causes kswapd to exit balance_pgdat() before reaching | |
3070 | * the wake up checks. If kswapd is going to sleep, no process should | |
3071 | * be sleeping on pfmemalloc_wait, so wake them now if necessary. If | |
3072 | * the wake up is premature, processes will wake kswapd and get | |
3073 | * throttled again. The difference from wake ups in balance_pgdat() is | |
3074 | * that here we are under prepare_to_wait(). | |
5515061d | 3075 | */ |
9e5e3661 VB |
3076 | if (waitqueue_active(&pgdat->pfmemalloc_wait)) |
3077 | wake_up_all(&pgdat->pfmemalloc_wait); | |
f50de2d3 | 3078 | |
4ae0a48b | 3079 | return pgdat_balanced(pgdat, order, classzone_idx); |
f50de2d3 MG |
3080 | } |
3081 | ||
75485363 MG |
3082 | /* |
3083 | * kswapd shrinks the zone by the number of pages required to reach | |
3084 | * the high watermark. | |
b8e83b94 MG |
3085 | * |
3086 | * Returns true if kswapd scanned at least the requested number of pages to | |
283aba9f MG |
3087 | * reclaim or if the lack of progress was due to pages under writeback. |
3088 | * This is used to determine if the scanning priority needs to be raised. | |
75485363 | 3089 | */ |
b8e83b94 | 3090 | static bool kswapd_shrink_zone(struct zone *zone, |
7c954f6d | 3091 | int classzone_idx, |
75485363 | 3092 | struct scan_control *sc, |
2ab44f43 | 3093 | unsigned long *nr_attempted) |
75485363 | 3094 | { |
7c954f6d MG |
3095 | int testorder = sc->order; |
3096 | unsigned long balance_gap; | |
7c954f6d | 3097 | bool lowmem_pressure; |
75485363 MG |
3098 | |
3099 | /* Reclaim above the high watermark. */ | |
3100 | sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone)); | |
7c954f6d MG |
3101 | |
3102 | /* | |
3103 | * Kswapd reclaims only single pages with compaction enabled. Trying | |
3104 | * too hard to reclaim until contiguous free pages have become | |
3105 | * available can hurt performance by evicting too much useful data | |
3106 | * from memory. Do not reclaim more than needed for compaction. | |
3107 | */ | |
3108 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && | |
ebff3980 VB |
3109 | compaction_suitable(zone, sc->order, 0, classzone_idx) |
3110 | != COMPACT_SKIPPED) | |
7c954f6d MG |
3111 | testorder = 0; |
3112 | ||
3113 | /* | |
3114 | * We put equal pressure on every zone, unless one zone has way too | |
3115 | * many pages free already. The "too many pages" is defined as the | |
3116 | * high wmark plus a "gap" where the gap is either the low | |
3117 | * watermark or 1% of the zone, whichever is smaller. | |
3118 | */ | |
4be89a34 JZ |
3119 | balance_gap = min(low_wmark_pages(zone), DIV_ROUND_UP( |
3120 | zone->managed_pages, KSWAPD_ZONE_BALANCE_GAP_RATIO)); | |
7c954f6d MG |
3121 | |
3122 | /* | |
3123 | * If there is no low memory pressure or the zone is balanced then no | |
3124 | * reclaim is necessary | |
3125 | */ | |
3126 | lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone)); | |
3127 | if (!lowmem_pressure && zone_balanced(zone, testorder, | |
3128 | balance_gap, classzone_idx)) | |
3129 | return true; | |
3130 | ||
6b4f7799 | 3131 | shrink_zone(zone, sc, zone_idx(zone) == classzone_idx); |
75485363 | 3132 | |
2ab44f43 MG |
3133 | /* Account for the number of pages attempted to reclaim */ |
3134 | *nr_attempted += sc->nr_to_reclaim; | |
3135 | ||
57054651 | 3136 | clear_bit(ZONE_WRITEBACK, &zone->flags); |
283aba9f | 3137 | |
7c954f6d MG |
3138 | /* |
3139 | * If a zone reaches its high watermark, consider it to be no longer | |
3140 | * congested. It's possible there are dirty pages backed by congested | |
3141 | * BDIs but as pressure is relieved, speculatively avoid congestion | |
3142 | * waits. | |
3143 | */ | |
6e543d57 | 3144 | if (zone_reclaimable(zone) && |
7c954f6d | 3145 | zone_balanced(zone, testorder, 0, classzone_idx)) { |
57054651 JW |
3146 | clear_bit(ZONE_CONGESTED, &zone->flags); |
3147 | clear_bit(ZONE_DIRTY, &zone->flags); | |
7c954f6d MG |
3148 | } |
3149 | ||
b8e83b94 | 3150 | return sc->nr_scanned >= sc->nr_to_reclaim; |
75485363 MG |
3151 | } |
3152 | ||
1da177e4 LT |
3153 | /* |
3154 | * For kswapd, balance_pgdat() will work across all this node's zones until | |
41858966 | 3155 | * they are all at high_wmark_pages(zone). |
1da177e4 | 3156 | * |
0abdee2b | 3157 | * Returns the final order kswapd was reclaiming at |
1da177e4 LT |
3158 | * |
3159 | * There is special handling here for zones which are full of pinned pages. | |
3160 | * This can happen if the pages are all mlocked, or if they are all used by | |
3161 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. | |
3162 | * What we do is to detect the case where all pages in the zone have been | |
3163 | * scanned twice and there has been zero successful reclaim. Mark the zone as | |
3164 | * dead and from now on, only perform a short scan. Basically we're polling | |
3165 | * the zone for when the problem goes away. | |
3166 | * | |
3167 | * kswapd scans the zones in the highmem->normal->dma direction. It skips | |
41858966 MG |
3168 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
3169 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the | |
3170 | * lower zones regardless of the number of free pages in the lower zones. This | |
3171 | * interoperates with the page allocator fallback scheme to ensure that aging | |
3172 | * of pages is balanced across the zones. | |
1da177e4 | 3173 | */ |
99504748 | 3174 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, |
dc83edd9 | 3175 | int *classzone_idx) |
1da177e4 | 3176 | { |
1da177e4 | 3177 | int i; |
99504748 | 3178 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
0608f43d AM |
3179 | unsigned long nr_soft_reclaimed; |
3180 | unsigned long nr_soft_scanned; | |
179e9639 AM |
3181 | struct scan_control sc = { |
3182 | .gfp_mask = GFP_KERNEL, | |
ee814fe2 | 3183 | .order = order, |
b8e83b94 | 3184 | .priority = DEF_PRIORITY, |
ee814fe2 | 3185 | .may_writepage = !laptop_mode, |
a6dc60f8 | 3186 | .may_unmap = 1, |
2e2e4259 | 3187 | .may_swap = 1, |
179e9639 | 3188 | }; |
f8891e5e | 3189 | count_vm_event(PAGEOUTRUN); |
1da177e4 | 3190 | |
9e3b2f8c | 3191 | do { |
2ab44f43 | 3192 | unsigned long nr_attempted = 0; |
b8e83b94 | 3193 | bool raise_priority = true; |
2ab44f43 | 3194 | bool pgdat_needs_compaction = (order > 0); |
b8e83b94 MG |
3195 | |
3196 | sc.nr_reclaimed = 0; | |
1da177e4 | 3197 | |
d6277db4 RW |
3198 | /* |
3199 | * Scan in the highmem->dma direction for the highest | |
3200 | * zone which needs scanning | |
3201 | */ | |
3202 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { | |
3203 | struct zone *zone = pgdat->node_zones + i; | |
1da177e4 | 3204 | |
d6277db4 RW |
3205 | if (!populated_zone(zone)) |
3206 | continue; | |
1da177e4 | 3207 | |
6e543d57 LD |
3208 | if (sc.priority != DEF_PRIORITY && |
3209 | !zone_reclaimable(zone)) | |
d6277db4 | 3210 | continue; |
1da177e4 | 3211 | |
556adecb RR |
3212 | /* |
3213 | * Do some background aging of the anon list, to give | |
3214 | * pages a chance to be referenced before reclaiming. | |
3215 | */ | |
9e3b2f8c | 3216 | age_active_anon(zone, &sc); |
556adecb | 3217 | |
cc715d99 MG |
3218 | /* |
3219 | * If the number of buffer_heads in the machine | |
3220 | * exceeds the maximum allowed level and this node | |
3221 | * has a highmem zone, force kswapd to reclaim from | |
3222 | * it to relieve lowmem pressure. | |
3223 | */ | |
3224 | if (buffer_heads_over_limit && is_highmem_idx(i)) { | |
3225 | end_zone = i; | |
3226 | break; | |
3227 | } | |
3228 | ||
60cefed4 | 3229 | if (!zone_balanced(zone, order, 0, 0)) { |
d6277db4 | 3230 | end_zone = i; |
e1dbeda6 | 3231 | break; |
439423f6 | 3232 | } else { |
d43006d5 MG |
3233 | /* |
3234 | * If balanced, clear the dirty and congested | |
3235 | * flags | |
3236 | */ | |
57054651 JW |
3237 | clear_bit(ZONE_CONGESTED, &zone->flags); |
3238 | clear_bit(ZONE_DIRTY, &zone->flags); | |
1da177e4 | 3239 | } |
1da177e4 | 3240 | } |
dafcb73e | 3241 | |
b8e83b94 | 3242 | if (i < 0) |
e1dbeda6 AM |
3243 | goto out; |
3244 | ||
1da177e4 LT |
3245 | for (i = 0; i <= end_zone; i++) { |
3246 | struct zone *zone = pgdat->node_zones + i; | |
3247 | ||
2ab44f43 MG |
3248 | if (!populated_zone(zone)) |
3249 | continue; | |
3250 | ||
2ab44f43 MG |
3251 | /* |
3252 | * If any zone is currently balanced then kswapd will | |
3253 | * not call compaction as it is expected that the | |
3254 | * necessary pages are already available. | |
3255 | */ | |
3256 | if (pgdat_needs_compaction && | |
3257 | zone_watermark_ok(zone, order, | |
3258 | low_wmark_pages(zone), | |
3259 | *classzone_idx, 0)) | |
3260 | pgdat_needs_compaction = false; | |
1da177e4 LT |
3261 | } |
3262 | ||
b7ea3c41 MG |
3263 | /* |
3264 | * If we're getting trouble reclaiming, start doing writepage | |
3265 | * even in laptop mode. | |
3266 | */ | |
3267 | if (sc.priority < DEF_PRIORITY - 2) | |
3268 | sc.may_writepage = 1; | |
3269 | ||
1da177e4 LT |
3270 | /* |
3271 | * Now scan the zone in the dma->highmem direction, stopping | |
3272 | * at the last zone which needs scanning. | |
3273 | * | |
3274 | * We do this because the page allocator works in the opposite | |
3275 | * direction. This prevents the page allocator from allocating | |
3276 | * pages behind kswapd's direction of progress, which would | |
3277 | * cause too much scanning of the lower zones. | |
3278 | */ | |
3279 | for (i = 0; i <= end_zone; i++) { | |
3280 | struct zone *zone = pgdat->node_zones + i; | |
3281 | ||
f3fe6512 | 3282 | if (!populated_zone(zone)) |
1da177e4 LT |
3283 | continue; |
3284 | ||
6e543d57 LD |
3285 | if (sc.priority != DEF_PRIORITY && |
3286 | !zone_reclaimable(zone)) | |
1da177e4 LT |
3287 | continue; |
3288 | ||
1da177e4 | 3289 | sc.nr_scanned = 0; |
4e416953 | 3290 | |
0608f43d AM |
3291 | nr_soft_scanned = 0; |
3292 | /* | |
3293 | * Call soft limit reclaim before calling shrink_zone. | |
3294 | */ | |
3295 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | |
3296 | order, sc.gfp_mask, | |
3297 | &nr_soft_scanned); | |
3298 | sc.nr_reclaimed += nr_soft_reclaimed; | |
3299 | ||
32a4330d | 3300 | /* |
7c954f6d MG |
3301 | * There should be no need to raise the scanning |
3302 | * priority if enough pages are already being scanned | |
3303 | * that that high watermark would be met at 100% | |
3304 | * efficiency. | |
fe2c2a10 | 3305 | */ |
6b4f7799 JW |
3306 | if (kswapd_shrink_zone(zone, end_zone, |
3307 | &sc, &nr_attempted)) | |
7c954f6d | 3308 | raise_priority = false; |
1da177e4 | 3309 | } |
5515061d MG |
3310 | |
3311 | /* | |
3312 | * If the low watermark is met there is no need for processes | |
3313 | * to be throttled on pfmemalloc_wait as they should not be | |
3314 | * able to safely make forward progress. Wake them | |
3315 | */ | |
3316 | if (waitqueue_active(&pgdat->pfmemalloc_wait) && | |
3317 | pfmemalloc_watermark_ok(pgdat)) | |
cfc51155 | 3318 | wake_up_all(&pgdat->pfmemalloc_wait); |
5515061d | 3319 | |
1da177e4 | 3320 | /* |
b8e83b94 MG |
3321 | * Fragmentation may mean that the system cannot be rebalanced |
3322 | * for high-order allocations in all zones. If twice the | |
3323 | * allocation size has been reclaimed and the zones are still | |
3324 | * not balanced then recheck the watermarks at order-0 to | |
3325 | * prevent kswapd reclaiming excessively. Assume that a | |
3326 | * process requested a high-order can direct reclaim/compact. | |
1da177e4 | 3327 | */ |
b8e83b94 MG |
3328 | if (order && sc.nr_reclaimed >= 2UL << order) |
3329 | order = sc.order = 0; | |
8357376d | 3330 | |
b8e83b94 MG |
3331 | /* Check if kswapd should be suspending */ |
3332 | if (try_to_freeze() || kthread_should_stop()) | |
3333 | break; | |
8357376d | 3334 | |
2ab44f43 MG |
3335 | /* |
3336 | * Compact if necessary and kswapd is reclaiming at least the | |
3337 | * high watermark number of pages as requsted | |
3338 | */ | |
3339 | if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted) | |
3340 | compact_pgdat(pgdat, order); | |
3341 | ||
73ce02e9 | 3342 | /* |
b8e83b94 MG |
3343 | * Raise priority if scanning rate is too low or there was no |
3344 | * progress in reclaiming pages | |
73ce02e9 | 3345 | */ |
b8e83b94 MG |
3346 | if (raise_priority || !sc.nr_reclaimed) |
3347 | sc.priority--; | |
9aa41348 | 3348 | } while (sc.priority >= 1 && |
b8e83b94 | 3349 | !pgdat_balanced(pgdat, order, *classzone_idx)); |
1da177e4 | 3350 | |
b8e83b94 | 3351 | out: |
0abdee2b | 3352 | /* |
5515061d | 3353 | * Return the order we were reclaiming at so prepare_kswapd_sleep() |
0abdee2b MG |
3354 | * makes a decision on the order we were last reclaiming at. However, |
3355 | * if another caller entered the allocator slow path while kswapd | |
3356 | * was awake, order will remain at the higher level | |
3357 | */ | |
dc83edd9 | 3358 | *classzone_idx = end_zone; |
0abdee2b | 3359 | return order; |
1da177e4 LT |
3360 | } |
3361 | ||
dc83edd9 | 3362 | static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx) |
f0bc0a60 KM |
3363 | { |
3364 | long remaining = 0; | |
3365 | DEFINE_WAIT(wait); | |
3366 | ||
3367 | if (freezing(current) || kthread_should_stop()) | |
3368 | return; | |
3369 | ||
3370 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | |
3371 | ||
3372 | /* Try to sleep for a short interval */ | |
5515061d | 3373 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
f0bc0a60 KM |
3374 | remaining = schedule_timeout(HZ/10); |
3375 | finish_wait(&pgdat->kswapd_wait, &wait); | |
3376 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | |
3377 | } | |
3378 | ||
3379 | /* | |
3380 | * After a short sleep, check if it was a premature sleep. If not, then | |
3381 | * go fully to sleep until explicitly woken up. | |
3382 | */ | |
5515061d | 3383 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
f0bc0a60 KM |
3384 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
3385 | ||
3386 | /* | |
3387 | * vmstat counters are not perfectly accurate and the estimated | |
3388 | * value for counters such as NR_FREE_PAGES can deviate from the | |
3389 | * true value by nr_online_cpus * threshold. To avoid the zone | |
3390 | * watermarks being breached while under pressure, we reduce the | |
3391 | * per-cpu vmstat threshold while kswapd is awake and restore | |
3392 | * them before going back to sleep. | |
3393 | */ | |
3394 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); | |
1c7e7f6c | 3395 | |
62997027 MG |
3396 | /* |
3397 | * Compaction records what page blocks it recently failed to | |
3398 | * isolate pages from and skips them in the future scanning. | |
3399 | * When kswapd is going to sleep, it is reasonable to assume | |
3400 | * that pages and compaction may succeed so reset the cache. | |
3401 | */ | |
3402 | reset_isolation_suitable(pgdat); | |
3403 | ||
1c7e7f6c AK |
3404 | if (!kthread_should_stop()) |
3405 | schedule(); | |
3406 | ||
f0bc0a60 KM |
3407 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
3408 | } else { | |
3409 | if (remaining) | |
3410 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); | |
3411 | else | |
3412 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); | |
3413 | } | |
3414 | finish_wait(&pgdat->kswapd_wait, &wait); | |
3415 | } | |
3416 | ||
1da177e4 LT |
3417 | /* |
3418 | * The background pageout daemon, started as a kernel thread | |
4f98a2fe | 3419 | * from the init process. |
1da177e4 LT |
3420 | * |
3421 | * This basically trickles out pages so that we have _some_ | |
3422 | * free memory available even if there is no other activity | |
3423 | * that frees anything up. This is needed for things like routing | |
3424 | * etc, where we otherwise might have all activity going on in | |
3425 | * asynchronous contexts that cannot page things out. | |
3426 | * | |
3427 | * If there are applications that are active memory-allocators | |
3428 | * (most normal use), this basically shouldn't matter. | |
3429 | */ | |
3430 | static int kswapd(void *p) | |
3431 | { | |
215ddd66 | 3432 | unsigned long order, new_order; |
d2ebd0f6 | 3433 | unsigned balanced_order; |
215ddd66 | 3434 | int classzone_idx, new_classzone_idx; |
d2ebd0f6 | 3435 | int balanced_classzone_idx; |
1da177e4 LT |
3436 | pg_data_t *pgdat = (pg_data_t*)p; |
3437 | struct task_struct *tsk = current; | |
f0bc0a60 | 3438 | |
1da177e4 LT |
3439 | struct reclaim_state reclaim_state = { |
3440 | .reclaimed_slab = 0, | |
3441 | }; | |
a70f7302 | 3442 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
1da177e4 | 3443 | |
cf40bd16 NP |
3444 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
3445 | ||
174596a0 | 3446 | if (!cpumask_empty(cpumask)) |
c5f59f08 | 3447 | set_cpus_allowed_ptr(tsk, cpumask); |
1da177e4 LT |
3448 | current->reclaim_state = &reclaim_state; |
3449 | ||
3450 | /* | |
3451 | * Tell the memory management that we're a "memory allocator", | |
3452 | * and that if we need more memory we should get access to it | |
3453 | * regardless (see "__alloc_pages()"). "kswapd" should | |
3454 | * never get caught in the normal page freeing logic. | |
3455 | * | |
3456 | * (Kswapd normally doesn't need memory anyway, but sometimes | |
3457 | * you need a small amount of memory in order to be able to | |
3458 | * page out something else, and this flag essentially protects | |
3459 | * us from recursively trying to free more memory as we're | |
3460 | * trying to free the first piece of memory in the first place). | |
3461 | */ | |
930d9152 | 3462 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
83144186 | 3463 | set_freezable(); |
1da177e4 | 3464 | |
215ddd66 | 3465 | order = new_order = 0; |
d2ebd0f6 | 3466 | balanced_order = 0; |
215ddd66 | 3467 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; |
d2ebd0f6 | 3468 | balanced_classzone_idx = classzone_idx; |
1da177e4 | 3469 | for ( ; ; ) { |
6f6313d4 | 3470 | bool ret; |
3e1d1d28 | 3471 | |
215ddd66 MG |
3472 | /* |
3473 | * If the last balance_pgdat was unsuccessful it's unlikely a | |
3474 | * new request of a similar or harder type will succeed soon | |
3475 | * so consider going to sleep on the basis we reclaimed at | |
3476 | */ | |
d2ebd0f6 AS |
3477 | if (balanced_classzone_idx >= new_classzone_idx && |
3478 | balanced_order == new_order) { | |
215ddd66 MG |
3479 | new_order = pgdat->kswapd_max_order; |
3480 | new_classzone_idx = pgdat->classzone_idx; | |
3481 | pgdat->kswapd_max_order = 0; | |
3482 | pgdat->classzone_idx = pgdat->nr_zones - 1; | |
3483 | } | |
3484 | ||
99504748 | 3485 | if (order < new_order || classzone_idx > new_classzone_idx) { |
1da177e4 LT |
3486 | /* |
3487 | * Don't sleep if someone wants a larger 'order' | |
99504748 | 3488 | * allocation or has tigher zone constraints |
1da177e4 LT |
3489 | */ |
3490 | order = new_order; | |
99504748 | 3491 | classzone_idx = new_classzone_idx; |
1da177e4 | 3492 | } else { |
d2ebd0f6 AS |
3493 | kswapd_try_to_sleep(pgdat, balanced_order, |
3494 | balanced_classzone_idx); | |
1da177e4 | 3495 | order = pgdat->kswapd_max_order; |
99504748 | 3496 | classzone_idx = pgdat->classzone_idx; |
f0dfcde0 AS |
3497 | new_order = order; |
3498 | new_classzone_idx = classzone_idx; | |
4d40502e | 3499 | pgdat->kswapd_max_order = 0; |
215ddd66 | 3500 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
1da177e4 | 3501 | } |
1da177e4 | 3502 | |
8fe23e05 DR |
3503 | ret = try_to_freeze(); |
3504 | if (kthread_should_stop()) | |
3505 | break; | |
3506 | ||
3507 | /* | |
3508 | * We can speed up thawing tasks if we don't call balance_pgdat | |
3509 | * after returning from the refrigerator | |
3510 | */ | |
33906bc5 MG |
3511 | if (!ret) { |
3512 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); | |
d2ebd0f6 AS |
3513 | balanced_classzone_idx = classzone_idx; |
3514 | balanced_order = balance_pgdat(pgdat, order, | |
3515 | &balanced_classzone_idx); | |
33906bc5 | 3516 | } |
1da177e4 | 3517 | } |
b0a8cc58 | 3518 | |
71abdc15 | 3519 | tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD); |
b0a8cc58 | 3520 | current->reclaim_state = NULL; |
71abdc15 JW |
3521 | lockdep_clear_current_reclaim_state(); |
3522 | ||
1da177e4 LT |
3523 | return 0; |
3524 | } | |
3525 | ||
3526 | /* | |
3527 | * A zone is low on free memory, so wake its kswapd task to service it. | |
3528 | */ | |
99504748 | 3529 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx) |
1da177e4 LT |
3530 | { |
3531 | pg_data_t *pgdat; | |
3532 | ||
f3fe6512 | 3533 | if (!populated_zone(zone)) |
1da177e4 LT |
3534 | return; |
3535 | ||
344736f2 | 3536 | if (!cpuset_zone_allowed(zone, GFP_KERNEL | __GFP_HARDWALL)) |
1da177e4 | 3537 | return; |
88f5acf8 | 3538 | pgdat = zone->zone_pgdat; |
99504748 | 3539 | if (pgdat->kswapd_max_order < order) { |
1da177e4 | 3540 | pgdat->kswapd_max_order = order; |
99504748 MG |
3541 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); |
3542 | } | |
8d0986e2 | 3543 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
1da177e4 | 3544 | return; |
892f795d | 3545 | if (zone_balanced(zone, order, 0, 0)) |
88f5acf8 MG |
3546 | return; |
3547 | ||
3548 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); | |
8d0986e2 | 3549 | wake_up_interruptible(&pgdat->kswapd_wait); |
1da177e4 LT |
3550 | } |
3551 | ||
c6f37f12 | 3552 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3553 | /* |
7b51755c | 3554 | * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of |
d6277db4 RW |
3555 | * freed pages. |
3556 | * | |
3557 | * Rather than trying to age LRUs the aim is to preserve the overall | |
3558 | * LRU order by reclaiming preferentially | |
3559 | * inactive > active > active referenced > active mapped | |
1da177e4 | 3560 | */ |
7b51755c | 3561 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
1da177e4 | 3562 | { |
d6277db4 | 3563 | struct reclaim_state reclaim_state; |
d6277db4 | 3564 | struct scan_control sc = { |
ee814fe2 | 3565 | .nr_to_reclaim = nr_to_reclaim, |
7b51755c | 3566 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
ee814fe2 | 3567 | .priority = DEF_PRIORITY, |
d6277db4 | 3568 | .may_writepage = 1, |
ee814fe2 JW |
3569 | .may_unmap = 1, |
3570 | .may_swap = 1, | |
7b51755c | 3571 | .hibernation_mode = 1, |
1da177e4 | 3572 | }; |
a09ed5e0 | 3573 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); |
7b51755c KM |
3574 | struct task_struct *p = current; |
3575 | unsigned long nr_reclaimed; | |
1da177e4 | 3576 | |
7b51755c KM |
3577 | p->flags |= PF_MEMALLOC; |
3578 | lockdep_set_current_reclaim_state(sc.gfp_mask); | |
3579 | reclaim_state.reclaimed_slab = 0; | |
3580 | p->reclaim_state = &reclaim_state; | |
d6277db4 | 3581 | |
3115cd91 | 3582 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
d979677c | 3583 | |
7b51755c KM |
3584 | p->reclaim_state = NULL; |
3585 | lockdep_clear_current_reclaim_state(); | |
3586 | p->flags &= ~PF_MEMALLOC; | |
d6277db4 | 3587 | |
7b51755c | 3588 | return nr_reclaimed; |
1da177e4 | 3589 | } |
c6f37f12 | 3590 | #endif /* CONFIG_HIBERNATION */ |
1da177e4 | 3591 | |
1da177e4 LT |
3592 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
3593 | not required for correctness. So if the last cpu in a node goes | |
3594 | away, we get changed to run anywhere: as the first one comes back, | |
3595 | restore their cpu bindings. */ | |
fcb35a9b GKH |
3596 | static int cpu_callback(struct notifier_block *nfb, unsigned long action, |
3597 | void *hcpu) | |
1da177e4 | 3598 | { |
58c0a4a7 | 3599 | int nid; |
1da177e4 | 3600 | |
8bb78442 | 3601 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
48fb2e24 | 3602 | for_each_node_state(nid, N_MEMORY) { |
c5f59f08 | 3603 | pg_data_t *pgdat = NODE_DATA(nid); |
a70f7302 RR |
3604 | const struct cpumask *mask; |
3605 | ||
3606 | mask = cpumask_of_node(pgdat->node_id); | |
c5f59f08 | 3607 | |
3e597945 | 3608 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
1da177e4 | 3609 | /* One of our CPUs online: restore mask */ |
c5f59f08 | 3610 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
1da177e4 LT |
3611 | } |
3612 | } | |
3613 | return NOTIFY_OK; | |
3614 | } | |
1da177e4 | 3615 | |
3218ae14 YG |
3616 | /* |
3617 | * This kswapd start function will be called by init and node-hot-add. | |
3618 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. | |
3619 | */ | |
3620 | int kswapd_run(int nid) | |
3621 | { | |
3622 | pg_data_t *pgdat = NODE_DATA(nid); | |
3623 | int ret = 0; | |
3624 | ||
3625 | if (pgdat->kswapd) | |
3626 | return 0; | |
3627 | ||
3628 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); | |
3629 | if (IS_ERR(pgdat->kswapd)) { | |
3630 | /* failure at boot is fatal */ | |
3631 | BUG_ON(system_state == SYSTEM_BOOTING); | |
d5dc0ad9 GS |
3632 | pr_err("Failed to start kswapd on node %d\n", nid); |
3633 | ret = PTR_ERR(pgdat->kswapd); | |
d72515b8 | 3634 | pgdat->kswapd = NULL; |
3218ae14 YG |
3635 | } |
3636 | return ret; | |
3637 | } | |
3638 | ||
8fe23e05 | 3639 | /* |
d8adde17 | 3640 | * Called by memory hotplug when all memory in a node is offlined. Caller must |
bfc8c901 | 3641 | * hold mem_hotplug_begin/end(). |
8fe23e05 DR |
3642 | */ |
3643 | void kswapd_stop(int nid) | |
3644 | { | |
3645 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; | |
3646 | ||
d8adde17 | 3647 | if (kswapd) { |
8fe23e05 | 3648 | kthread_stop(kswapd); |
d8adde17 JL |
3649 | NODE_DATA(nid)->kswapd = NULL; |
3650 | } | |
8fe23e05 DR |
3651 | } |
3652 | ||
1da177e4 LT |
3653 | static int __init kswapd_init(void) |
3654 | { | |
3218ae14 | 3655 | int nid; |
69e05944 | 3656 | |
1da177e4 | 3657 | swap_setup(); |
48fb2e24 | 3658 | for_each_node_state(nid, N_MEMORY) |
3218ae14 | 3659 | kswapd_run(nid); |
1da177e4 LT |
3660 | hotcpu_notifier(cpu_callback, 0); |
3661 | return 0; | |
3662 | } | |
3663 | ||
3664 | module_init(kswapd_init) | |
9eeff239 CL |
3665 | |
3666 | #ifdef CONFIG_NUMA | |
3667 | /* | |
3668 | * Zone reclaim mode | |
3669 | * | |
3670 | * If non-zero call zone_reclaim when the number of free pages falls below | |
3671 | * the watermarks. | |
9eeff239 CL |
3672 | */ |
3673 | int zone_reclaim_mode __read_mostly; | |
3674 | ||
1b2ffb78 | 3675 | #define RECLAIM_OFF 0 |
7d03431c | 3676 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
1b2ffb78 | 3677 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
95bbc0c7 | 3678 | #define RECLAIM_UNMAP (1<<2) /* Unmap pages during reclaim */ |
1b2ffb78 | 3679 | |
a92f7126 CL |
3680 | /* |
3681 | * Priority for ZONE_RECLAIM. This determines the fraction of pages | |
3682 | * of a node considered for each zone_reclaim. 4 scans 1/16th of | |
3683 | * a zone. | |
3684 | */ | |
3685 | #define ZONE_RECLAIM_PRIORITY 4 | |
3686 | ||
9614634f CL |
3687 | /* |
3688 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to | |
3689 | * occur. | |
3690 | */ | |
3691 | int sysctl_min_unmapped_ratio = 1; | |
3692 | ||
0ff38490 CL |
3693 | /* |
3694 | * If the number of slab pages in a zone grows beyond this percentage then | |
3695 | * slab reclaim needs to occur. | |
3696 | */ | |
3697 | int sysctl_min_slab_ratio = 5; | |
3698 | ||
90afa5de MG |
3699 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) |
3700 | { | |
3701 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); | |
3702 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + | |
3703 | zone_page_state(zone, NR_ACTIVE_FILE); | |
3704 | ||
3705 | /* | |
3706 | * It's possible for there to be more file mapped pages than | |
3707 | * accounted for by the pages on the file LRU lists because | |
3708 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED | |
3709 | */ | |
3710 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; | |
3711 | } | |
3712 | ||
3713 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ | |
d031a157 | 3714 | static unsigned long zone_pagecache_reclaimable(struct zone *zone) |
90afa5de | 3715 | { |
d031a157 AM |
3716 | unsigned long nr_pagecache_reclaimable; |
3717 | unsigned long delta = 0; | |
90afa5de MG |
3718 | |
3719 | /* | |
95bbc0c7 | 3720 | * If RECLAIM_UNMAP is set, then all file pages are considered |
90afa5de MG |
3721 | * potentially reclaimable. Otherwise, we have to worry about |
3722 | * pages like swapcache and zone_unmapped_file_pages() provides | |
3723 | * a better estimate | |
3724 | */ | |
95bbc0c7 | 3725 | if (zone_reclaim_mode & RECLAIM_UNMAP) |
90afa5de MG |
3726 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); |
3727 | else | |
3728 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); | |
3729 | ||
3730 | /* If we can't clean pages, remove dirty pages from consideration */ | |
3731 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) | |
3732 | delta += zone_page_state(zone, NR_FILE_DIRTY); | |
3733 | ||
3734 | /* Watch for any possible underflows due to delta */ | |
3735 | if (unlikely(delta > nr_pagecache_reclaimable)) | |
3736 | delta = nr_pagecache_reclaimable; | |
3737 | ||
3738 | return nr_pagecache_reclaimable - delta; | |
3739 | } | |
3740 | ||
9eeff239 CL |
3741 | /* |
3742 | * Try to free up some pages from this zone through reclaim. | |
3743 | */ | |
179e9639 | 3744 | static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
9eeff239 | 3745 | { |
7fb2d46d | 3746 | /* Minimum pages needed in order to stay on node */ |
69e05944 | 3747 | const unsigned long nr_pages = 1 << order; |
9eeff239 CL |
3748 | struct task_struct *p = current; |
3749 | struct reclaim_state reclaim_state; | |
179e9639 | 3750 | struct scan_control sc = { |
62b726c1 | 3751 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
21caf2fc | 3752 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
bd2f6199 | 3753 | .order = order, |
9e3b2f8c | 3754 | .priority = ZONE_RECLAIM_PRIORITY, |
ee814fe2 | 3755 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
95bbc0c7 | 3756 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_UNMAP), |
ee814fe2 | 3757 | .may_swap = 1, |
179e9639 | 3758 | }; |
9eeff239 | 3759 | |
9eeff239 | 3760 | cond_resched(); |
d4f7796e | 3761 | /* |
95bbc0c7 | 3762 | * We need to be able to allocate from the reserves for RECLAIM_UNMAP |
d4f7796e | 3763 | * and we also need to be able to write out pages for RECLAIM_WRITE |
95bbc0c7 | 3764 | * and RECLAIM_UNMAP. |
d4f7796e CL |
3765 | */ |
3766 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; | |
76ca542d | 3767 | lockdep_set_current_reclaim_state(gfp_mask); |
9eeff239 CL |
3768 | reclaim_state.reclaimed_slab = 0; |
3769 | p->reclaim_state = &reclaim_state; | |
c84db23c | 3770 | |
90afa5de | 3771 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { |
0ff38490 CL |
3772 | /* |
3773 | * Free memory by calling shrink zone with increasing | |
3774 | * priorities until we have enough memory freed. | |
3775 | */ | |
0ff38490 | 3776 | do { |
6b4f7799 | 3777 | shrink_zone(zone, &sc, true); |
9e3b2f8c | 3778 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); |
0ff38490 | 3779 | } |
c84db23c | 3780 | |
9eeff239 | 3781 | p->reclaim_state = NULL; |
d4f7796e | 3782 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
76ca542d | 3783 | lockdep_clear_current_reclaim_state(); |
a79311c1 | 3784 | return sc.nr_reclaimed >= nr_pages; |
9eeff239 | 3785 | } |
179e9639 AM |
3786 | |
3787 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | |
3788 | { | |
179e9639 | 3789 | int node_id; |
d773ed6b | 3790 | int ret; |
179e9639 AM |
3791 | |
3792 | /* | |
0ff38490 CL |
3793 | * Zone reclaim reclaims unmapped file backed pages and |
3794 | * slab pages if we are over the defined limits. | |
34aa1330 | 3795 | * |
9614634f CL |
3796 | * A small portion of unmapped file backed pages is needed for |
3797 | * file I/O otherwise pages read by file I/O will be immediately | |
3798 | * thrown out if the zone is overallocated. So we do not reclaim | |
3799 | * if less than a specified percentage of the zone is used by | |
3800 | * unmapped file backed pages. | |
179e9639 | 3801 | */ |
90afa5de MG |
3802 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && |
3803 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) | |
fa5e084e | 3804 | return ZONE_RECLAIM_FULL; |
179e9639 | 3805 | |
6e543d57 | 3806 | if (!zone_reclaimable(zone)) |
fa5e084e | 3807 | return ZONE_RECLAIM_FULL; |
d773ed6b | 3808 | |
179e9639 | 3809 | /* |
d773ed6b | 3810 | * Do not scan if the allocation should not be delayed. |
179e9639 | 3811 | */ |
d0164adc | 3812 | if (!gfpflags_allow_blocking(gfp_mask) || (current->flags & PF_MEMALLOC)) |
fa5e084e | 3813 | return ZONE_RECLAIM_NOSCAN; |
179e9639 AM |
3814 | |
3815 | /* | |
3816 | * Only run zone reclaim on the local zone or on zones that do not | |
3817 | * have associated processors. This will favor the local processor | |
3818 | * over remote processors and spread off node memory allocations | |
3819 | * as wide as possible. | |
3820 | */ | |
89fa3024 | 3821 | node_id = zone_to_nid(zone); |
37c0708d | 3822 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
fa5e084e | 3823 | return ZONE_RECLAIM_NOSCAN; |
d773ed6b | 3824 | |
57054651 | 3825 | if (test_and_set_bit(ZONE_RECLAIM_LOCKED, &zone->flags)) |
fa5e084e MG |
3826 | return ZONE_RECLAIM_NOSCAN; |
3827 | ||
d773ed6b | 3828 | ret = __zone_reclaim(zone, gfp_mask, order); |
57054651 | 3829 | clear_bit(ZONE_RECLAIM_LOCKED, &zone->flags); |
d773ed6b | 3830 | |
24cf7251 MG |
3831 | if (!ret) |
3832 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); | |
3833 | ||
d773ed6b | 3834 | return ret; |
179e9639 | 3835 | } |
9eeff239 | 3836 | #endif |
894bc310 | 3837 | |
894bc310 LS |
3838 | /* |
3839 | * page_evictable - test whether a page is evictable | |
3840 | * @page: the page to test | |
894bc310 LS |
3841 | * |
3842 | * Test whether page is evictable--i.e., should be placed on active/inactive | |
39b5f29a | 3843 | * lists vs unevictable list. |
894bc310 LS |
3844 | * |
3845 | * Reasons page might not be evictable: | |
ba9ddf49 | 3846 | * (1) page's mapping marked unevictable |
b291f000 | 3847 | * (2) page is part of an mlocked VMA |
ba9ddf49 | 3848 | * |
894bc310 | 3849 | */ |
39b5f29a | 3850 | int page_evictable(struct page *page) |
894bc310 | 3851 | { |
39b5f29a | 3852 | return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); |
894bc310 | 3853 | } |
89e004ea | 3854 | |
85046579 | 3855 | #ifdef CONFIG_SHMEM |
89e004ea | 3856 | /** |
24513264 HD |
3857 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list |
3858 | * @pages: array of pages to check | |
3859 | * @nr_pages: number of pages to check | |
89e004ea | 3860 | * |
24513264 | 3861 | * Checks pages for evictability and moves them to the appropriate lru list. |
85046579 HD |
3862 | * |
3863 | * This function is only used for SysV IPC SHM_UNLOCK. | |
89e004ea | 3864 | */ |
24513264 | 3865 | void check_move_unevictable_pages(struct page **pages, int nr_pages) |
89e004ea | 3866 | { |
925b7673 | 3867 | struct lruvec *lruvec; |
24513264 HD |
3868 | struct zone *zone = NULL; |
3869 | int pgscanned = 0; | |
3870 | int pgrescued = 0; | |
3871 | int i; | |
89e004ea | 3872 | |
24513264 HD |
3873 | for (i = 0; i < nr_pages; i++) { |
3874 | struct page *page = pages[i]; | |
3875 | struct zone *pagezone; | |
89e004ea | 3876 | |
24513264 HD |
3877 | pgscanned++; |
3878 | pagezone = page_zone(page); | |
3879 | if (pagezone != zone) { | |
3880 | if (zone) | |
3881 | spin_unlock_irq(&zone->lru_lock); | |
3882 | zone = pagezone; | |
3883 | spin_lock_irq(&zone->lru_lock); | |
3884 | } | |
fa9add64 | 3885 | lruvec = mem_cgroup_page_lruvec(page, zone); |
89e004ea | 3886 | |
24513264 HD |
3887 | if (!PageLRU(page) || !PageUnevictable(page)) |
3888 | continue; | |
89e004ea | 3889 | |
39b5f29a | 3890 | if (page_evictable(page)) { |
24513264 HD |
3891 | enum lru_list lru = page_lru_base_type(page); |
3892 | ||
309381fe | 3893 | VM_BUG_ON_PAGE(PageActive(page), page); |
24513264 | 3894 | ClearPageUnevictable(page); |
fa9add64 HD |
3895 | del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE); |
3896 | add_page_to_lru_list(page, lruvec, lru); | |
24513264 | 3897 | pgrescued++; |
89e004ea | 3898 | } |
24513264 | 3899 | } |
89e004ea | 3900 | |
24513264 HD |
3901 | if (zone) { |
3902 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
3903 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); | |
3904 | spin_unlock_irq(&zone->lru_lock); | |
89e004ea | 3905 | } |
89e004ea | 3906 | } |
85046579 | 3907 | #endif /* CONFIG_SHMEM */ |