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> |
f9fe48be | 49 | #include <linux/dax.h> |
1da177e4 LT |
50 | |
51 | #include <asm/tlbflush.h> | |
52 | #include <asm/div64.h> | |
53 | ||
54 | #include <linux/swapops.h> | |
117aad1e | 55 | #include <linux/balloon_compaction.h> |
1da177e4 | 56 | |
0f8053a5 NP |
57 | #include "internal.h" |
58 | ||
33906bc5 MG |
59 | #define CREATE_TRACE_POINTS |
60 | #include <trace/events/vmscan.h> | |
61 | ||
1da177e4 | 62 | struct scan_control { |
22fba335 KM |
63 | /* How many pages shrink_list() should reclaim */ |
64 | unsigned long nr_to_reclaim; | |
65 | ||
1da177e4 | 66 | /* This context's GFP mask */ |
6daa0e28 | 67 | gfp_t gfp_mask; |
1da177e4 | 68 | |
ee814fe2 | 69 | /* Allocation order */ |
5ad333eb | 70 | int order; |
66e1707b | 71 | |
ee814fe2 JW |
72 | /* |
73 | * Nodemask of nodes allowed by the caller. If NULL, all nodes | |
74 | * are scanned. | |
75 | */ | |
76 | nodemask_t *nodemask; | |
9e3b2f8c | 77 | |
f16015fb JW |
78 | /* |
79 | * The memory cgroup that hit its limit and as a result is the | |
80 | * primary target of this reclaim invocation. | |
81 | */ | |
82 | struct mem_cgroup *target_mem_cgroup; | |
66e1707b | 83 | |
ee814fe2 JW |
84 | /* Scan (total_size >> priority) pages at once */ |
85 | int priority; | |
86 | ||
87 | unsigned int may_writepage:1; | |
88 | ||
89 | /* Can mapped pages be reclaimed? */ | |
90 | unsigned int may_unmap:1; | |
91 | ||
92 | /* Can pages be swapped as part of reclaim? */ | |
93 | unsigned int may_swap:1; | |
94 | ||
241994ed JW |
95 | /* Can cgroups be reclaimed below their normal consumption range? */ |
96 | unsigned int may_thrash:1; | |
97 | ||
ee814fe2 JW |
98 | unsigned int hibernation_mode:1; |
99 | ||
100 | /* One of the zones is ready for compaction */ | |
101 | unsigned int compaction_ready:1; | |
102 | ||
103 | /* Incremented by the number of inactive pages that were scanned */ | |
104 | unsigned long nr_scanned; | |
105 | ||
106 | /* Number of pages freed so far during a call to shrink_zones() */ | |
107 | unsigned long nr_reclaimed; | |
1da177e4 LT |
108 | }; |
109 | ||
1da177e4 LT |
110 | #ifdef ARCH_HAS_PREFETCH |
111 | #define prefetch_prev_lru_page(_page, _base, _field) \ | |
112 | do { \ | |
113 | if ((_page)->lru.prev != _base) { \ | |
114 | struct page *prev; \ | |
115 | \ | |
116 | prev = lru_to_page(&(_page->lru)); \ | |
117 | prefetch(&prev->_field); \ | |
118 | } \ | |
119 | } while (0) | |
120 | #else | |
121 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) | |
122 | #endif | |
123 | ||
124 | #ifdef ARCH_HAS_PREFETCHW | |
125 | #define prefetchw_prev_lru_page(_page, _base, _field) \ | |
126 | do { \ | |
127 | if ((_page)->lru.prev != _base) { \ | |
128 | struct page *prev; \ | |
129 | \ | |
130 | prev = lru_to_page(&(_page->lru)); \ | |
131 | prefetchw(&prev->_field); \ | |
132 | } \ | |
133 | } while (0) | |
134 | #else | |
135 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) | |
136 | #endif | |
137 | ||
138 | /* | |
139 | * From 0 .. 100. Higher means more swappy. | |
140 | */ | |
141 | int vm_swappiness = 60; | |
d0480be4 WSH |
142 | /* |
143 | * The total number of pages which are beyond the high watermark within all | |
144 | * zones. | |
145 | */ | |
146 | unsigned long vm_total_pages; | |
1da177e4 LT |
147 | |
148 | static LIST_HEAD(shrinker_list); | |
149 | static DECLARE_RWSEM(shrinker_rwsem); | |
150 | ||
c255a458 | 151 | #ifdef CONFIG_MEMCG |
89b5fae5 JW |
152 | static bool global_reclaim(struct scan_control *sc) |
153 | { | |
f16015fb | 154 | return !sc->target_mem_cgroup; |
89b5fae5 | 155 | } |
97c9341f TH |
156 | |
157 | /** | |
158 | * sane_reclaim - is the usual dirty throttling mechanism operational? | |
159 | * @sc: scan_control in question | |
160 | * | |
161 | * The normal page dirty throttling mechanism in balance_dirty_pages() is | |
162 | * completely broken with the legacy memcg and direct stalling in | |
163 | * shrink_page_list() is used for throttling instead, which lacks all the | |
164 | * niceties such as fairness, adaptive pausing, bandwidth proportional | |
165 | * allocation and configurability. | |
166 | * | |
167 | * This function tests whether the vmscan currently in progress can assume | |
168 | * that the normal dirty throttling mechanism is operational. | |
169 | */ | |
170 | static bool sane_reclaim(struct scan_control *sc) | |
171 | { | |
172 | struct mem_cgroup *memcg = sc->target_mem_cgroup; | |
173 | ||
174 | if (!memcg) | |
175 | return true; | |
176 | #ifdef CONFIG_CGROUP_WRITEBACK | |
69234ace | 177 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
97c9341f TH |
178 | return true; |
179 | #endif | |
180 | return false; | |
181 | } | |
91a45470 | 182 | #else |
89b5fae5 JW |
183 | static bool global_reclaim(struct scan_control *sc) |
184 | { | |
185 | return true; | |
186 | } | |
97c9341f TH |
187 | |
188 | static bool sane_reclaim(struct scan_control *sc) | |
189 | { | |
190 | return true; | |
191 | } | |
91a45470 KH |
192 | #endif |
193 | ||
a1c3bfb2 | 194 | static unsigned long zone_reclaimable_pages(struct zone *zone) |
6e543d57 | 195 | { |
d031a157 | 196 | unsigned long nr; |
6e543d57 | 197 | |
0db2cb8d MH |
198 | nr = zone_page_state_snapshot(zone, NR_ACTIVE_FILE) + |
199 | zone_page_state_snapshot(zone, NR_INACTIVE_FILE) + | |
200 | zone_page_state_snapshot(zone, NR_ISOLATED_FILE); | |
6e543d57 LD |
201 | |
202 | if (get_nr_swap_pages() > 0) | |
0db2cb8d MH |
203 | nr += zone_page_state_snapshot(zone, NR_ACTIVE_ANON) + |
204 | zone_page_state_snapshot(zone, NR_INACTIVE_ANON) + | |
205 | zone_page_state_snapshot(zone, NR_ISOLATED_ANON); | |
6e543d57 LD |
206 | |
207 | return nr; | |
208 | } | |
209 | ||
210 | bool zone_reclaimable(struct zone *zone) | |
211 | { | |
0db2cb8d | 212 | return zone_page_state_snapshot(zone, NR_PAGES_SCANNED) < |
0d5d823a | 213 | zone_reclaimable_pages(zone) * 6; |
6e543d57 LD |
214 | } |
215 | ||
23047a96 | 216 | unsigned long lruvec_lru_size(struct lruvec *lruvec, enum lru_list lru) |
c9f299d9 | 217 | { |
c3c787e8 | 218 | if (!mem_cgroup_disabled()) |
4d7dcca2 | 219 | return mem_cgroup_get_lru_size(lruvec, lru); |
a3d8e054 | 220 | |
074291fe | 221 | return zone_page_state(lruvec_zone(lruvec), NR_LRU_BASE + lru); |
c9f299d9 KM |
222 | } |
223 | ||
1da177e4 | 224 | /* |
1d3d4437 | 225 | * Add a shrinker callback to be called from the vm. |
1da177e4 | 226 | */ |
1d3d4437 | 227 | int register_shrinker(struct shrinker *shrinker) |
1da177e4 | 228 | { |
1d3d4437 GC |
229 | size_t size = sizeof(*shrinker->nr_deferred); |
230 | ||
1d3d4437 GC |
231 | if (shrinker->flags & SHRINKER_NUMA_AWARE) |
232 | size *= nr_node_ids; | |
233 | ||
234 | shrinker->nr_deferred = kzalloc(size, GFP_KERNEL); | |
235 | if (!shrinker->nr_deferred) | |
236 | return -ENOMEM; | |
237 | ||
8e1f936b RR |
238 | down_write(&shrinker_rwsem); |
239 | list_add_tail(&shrinker->list, &shrinker_list); | |
240 | up_write(&shrinker_rwsem); | |
1d3d4437 | 241 | return 0; |
1da177e4 | 242 | } |
8e1f936b | 243 | EXPORT_SYMBOL(register_shrinker); |
1da177e4 LT |
244 | |
245 | /* | |
246 | * Remove one | |
247 | */ | |
8e1f936b | 248 | void unregister_shrinker(struct shrinker *shrinker) |
1da177e4 LT |
249 | { |
250 | down_write(&shrinker_rwsem); | |
251 | list_del(&shrinker->list); | |
252 | up_write(&shrinker_rwsem); | |
ae393321 | 253 | kfree(shrinker->nr_deferred); |
1da177e4 | 254 | } |
8e1f936b | 255 | EXPORT_SYMBOL(unregister_shrinker); |
1da177e4 LT |
256 | |
257 | #define SHRINK_BATCH 128 | |
1d3d4437 | 258 | |
cb731d6c VD |
259 | static unsigned long do_shrink_slab(struct shrink_control *shrinkctl, |
260 | struct shrinker *shrinker, | |
261 | unsigned long nr_scanned, | |
262 | unsigned long nr_eligible) | |
1d3d4437 GC |
263 | { |
264 | unsigned long freed = 0; | |
265 | unsigned long long delta; | |
266 | long total_scan; | |
d5bc5fd3 | 267 | long freeable; |
1d3d4437 GC |
268 | long nr; |
269 | long new_nr; | |
270 | int nid = shrinkctl->nid; | |
271 | long batch_size = shrinker->batch ? shrinker->batch | |
272 | : SHRINK_BATCH; | |
273 | ||
d5bc5fd3 VD |
274 | freeable = shrinker->count_objects(shrinker, shrinkctl); |
275 | if (freeable == 0) | |
1d3d4437 GC |
276 | return 0; |
277 | ||
278 | /* | |
279 | * copy the current shrinker scan count into a local variable | |
280 | * and zero it so that other concurrent shrinker invocations | |
281 | * don't also do this scanning work. | |
282 | */ | |
283 | nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0); | |
284 | ||
285 | total_scan = nr; | |
6b4f7799 | 286 | delta = (4 * nr_scanned) / shrinker->seeks; |
d5bc5fd3 | 287 | delta *= freeable; |
6b4f7799 | 288 | do_div(delta, nr_eligible + 1); |
1d3d4437 GC |
289 | total_scan += delta; |
290 | if (total_scan < 0) { | |
8612c663 | 291 | pr_err("shrink_slab: %pF negative objects to delete nr=%ld\n", |
a0b02131 | 292 | shrinker->scan_objects, total_scan); |
d5bc5fd3 | 293 | total_scan = freeable; |
1d3d4437 GC |
294 | } |
295 | ||
296 | /* | |
297 | * We need to avoid excessive windup on filesystem shrinkers | |
298 | * due to large numbers of GFP_NOFS allocations causing the | |
299 | * shrinkers to return -1 all the time. This results in a large | |
300 | * nr being built up so when a shrink that can do some work | |
301 | * comes along it empties the entire cache due to nr >>> | |
d5bc5fd3 | 302 | * freeable. This is bad for sustaining a working set in |
1d3d4437 GC |
303 | * memory. |
304 | * | |
305 | * Hence only allow the shrinker to scan the entire cache when | |
306 | * a large delta change is calculated directly. | |
307 | */ | |
d5bc5fd3 VD |
308 | if (delta < freeable / 4) |
309 | total_scan = min(total_scan, freeable / 2); | |
1d3d4437 GC |
310 | |
311 | /* | |
312 | * Avoid risking looping forever due to too large nr value: | |
313 | * never try to free more than twice the estimate number of | |
314 | * freeable entries. | |
315 | */ | |
d5bc5fd3 VD |
316 | if (total_scan > freeable * 2) |
317 | total_scan = freeable * 2; | |
1d3d4437 GC |
318 | |
319 | trace_mm_shrink_slab_start(shrinker, shrinkctl, nr, | |
6b4f7799 JW |
320 | nr_scanned, nr_eligible, |
321 | freeable, delta, total_scan); | |
1d3d4437 | 322 | |
0b1fb40a VD |
323 | /* |
324 | * Normally, we should not scan less than batch_size objects in one | |
325 | * pass to avoid too frequent shrinker calls, but if the slab has less | |
326 | * than batch_size objects in total and we are really tight on memory, | |
327 | * we will try to reclaim all available objects, otherwise we can end | |
328 | * up failing allocations although there are plenty of reclaimable | |
329 | * objects spread over several slabs with usage less than the | |
330 | * batch_size. | |
331 | * | |
332 | * We detect the "tight on memory" situations by looking at the total | |
333 | * number of objects we want to scan (total_scan). If it is greater | |
d5bc5fd3 | 334 | * than the total number of objects on slab (freeable), we must be |
0b1fb40a VD |
335 | * scanning at high prio and therefore should try to reclaim as much as |
336 | * possible. | |
337 | */ | |
338 | while (total_scan >= batch_size || | |
d5bc5fd3 | 339 | total_scan >= freeable) { |
a0b02131 | 340 | unsigned long ret; |
0b1fb40a | 341 | unsigned long nr_to_scan = min(batch_size, total_scan); |
1d3d4437 | 342 | |
0b1fb40a | 343 | shrinkctl->nr_to_scan = nr_to_scan; |
a0b02131 DC |
344 | ret = shrinker->scan_objects(shrinker, shrinkctl); |
345 | if (ret == SHRINK_STOP) | |
346 | break; | |
347 | freed += ret; | |
1d3d4437 | 348 | |
0b1fb40a VD |
349 | count_vm_events(SLABS_SCANNED, nr_to_scan); |
350 | total_scan -= nr_to_scan; | |
1d3d4437 GC |
351 | |
352 | cond_resched(); | |
353 | } | |
354 | ||
355 | /* | |
356 | * move the unused scan count back into the shrinker in a | |
357 | * manner that handles concurrent updates. If we exhausted the | |
358 | * scan, there is no need to do an update. | |
359 | */ | |
360 | if (total_scan > 0) | |
361 | new_nr = atomic_long_add_return(total_scan, | |
362 | &shrinker->nr_deferred[nid]); | |
363 | else | |
364 | new_nr = atomic_long_read(&shrinker->nr_deferred[nid]); | |
365 | ||
df9024a8 | 366 | trace_mm_shrink_slab_end(shrinker, nid, freed, nr, new_nr, total_scan); |
1d3d4437 | 367 | return freed; |
1495f230 YH |
368 | } |
369 | ||
6b4f7799 | 370 | /** |
cb731d6c | 371 | * shrink_slab - shrink slab caches |
6b4f7799 JW |
372 | * @gfp_mask: allocation context |
373 | * @nid: node whose slab caches to target | |
cb731d6c | 374 | * @memcg: memory cgroup whose slab caches to target |
6b4f7799 JW |
375 | * @nr_scanned: pressure numerator |
376 | * @nr_eligible: pressure denominator | |
1da177e4 | 377 | * |
6b4f7799 | 378 | * Call the shrink functions to age shrinkable caches. |
1da177e4 | 379 | * |
6b4f7799 JW |
380 | * @nid is passed along to shrinkers with SHRINKER_NUMA_AWARE set, |
381 | * unaware shrinkers will receive a node id of 0 instead. | |
1da177e4 | 382 | * |
cb731d6c VD |
383 | * @memcg specifies the memory cgroup to target. If it is not NULL, |
384 | * only shrinkers with SHRINKER_MEMCG_AWARE set will be called to scan | |
385 | * objects from the memory cgroup specified. Otherwise all shrinkers | |
386 | * are called, and memcg aware shrinkers are supposed to scan the | |
387 | * global list then. | |
388 | * | |
6b4f7799 JW |
389 | * @nr_scanned and @nr_eligible form a ratio that indicate how much of |
390 | * the available objects should be scanned. Page reclaim for example | |
391 | * passes the number of pages scanned and the number of pages on the | |
392 | * LRU lists that it considered on @nid, plus a bias in @nr_scanned | |
393 | * when it encountered mapped pages. The ratio is further biased by | |
394 | * the ->seeks setting of the shrink function, which indicates the | |
395 | * cost to recreate an object relative to that of an LRU page. | |
b15e0905 | 396 | * |
6b4f7799 | 397 | * Returns the number of reclaimed slab objects. |
1da177e4 | 398 | */ |
cb731d6c VD |
399 | static unsigned long shrink_slab(gfp_t gfp_mask, int nid, |
400 | struct mem_cgroup *memcg, | |
401 | unsigned long nr_scanned, | |
402 | unsigned long nr_eligible) | |
1da177e4 LT |
403 | { |
404 | struct shrinker *shrinker; | |
24f7c6b9 | 405 | unsigned long freed = 0; |
1da177e4 | 406 | |
567e9ab2 | 407 | if (memcg && !memcg_kmem_online(memcg)) |
cb731d6c VD |
408 | return 0; |
409 | ||
6b4f7799 JW |
410 | if (nr_scanned == 0) |
411 | nr_scanned = SWAP_CLUSTER_MAX; | |
1da177e4 | 412 | |
f06590bd | 413 | if (!down_read_trylock(&shrinker_rwsem)) { |
24f7c6b9 DC |
414 | /* |
415 | * If we would return 0, our callers would understand that we | |
416 | * have nothing else to shrink and give up trying. By returning | |
417 | * 1 we keep it going and assume we'll be able to shrink next | |
418 | * time. | |
419 | */ | |
420 | freed = 1; | |
f06590bd MK |
421 | goto out; |
422 | } | |
1da177e4 LT |
423 | |
424 | list_for_each_entry(shrinker, &shrinker_list, list) { | |
6b4f7799 JW |
425 | struct shrink_control sc = { |
426 | .gfp_mask = gfp_mask, | |
427 | .nid = nid, | |
cb731d6c | 428 | .memcg = memcg, |
6b4f7799 | 429 | }; |
ec97097b | 430 | |
cb731d6c VD |
431 | if (memcg && !(shrinker->flags & SHRINKER_MEMCG_AWARE)) |
432 | continue; | |
433 | ||
6b4f7799 JW |
434 | if (!(shrinker->flags & SHRINKER_NUMA_AWARE)) |
435 | sc.nid = 0; | |
1da177e4 | 436 | |
cb731d6c | 437 | freed += do_shrink_slab(&sc, shrinker, nr_scanned, nr_eligible); |
1da177e4 | 438 | } |
6b4f7799 | 439 | |
1da177e4 | 440 | up_read(&shrinker_rwsem); |
f06590bd MK |
441 | out: |
442 | cond_resched(); | |
24f7c6b9 | 443 | return freed; |
1da177e4 LT |
444 | } |
445 | ||
cb731d6c VD |
446 | void drop_slab_node(int nid) |
447 | { | |
448 | unsigned long freed; | |
449 | ||
450 | do { | |
451 | struct mem_cgroup *memcg = NULL; | |
452 | ||
453 | freed = 0; | |
454 | do { | |
455 | freed += shrink_slab(GFP_KERNEL, nid, memcg, | |
456 | 1000, 1000); | |
457 | } while ((memcg = mem_cgroup_iter(NULL, memcg, NULL)) != NULL); | |
458 | } while (freed > 10); | |
459 | } | |
460 | ||
461 | void drop_slab(void) | |
462 | { | |
463 | int nid; | |
464 | ||
465 | for_each_online_node(nid) | |
466 | drop_slab_node(nid); | |
467 | } | |
468 | ||
1da177e4 LT |
469 | static inline int is_page_cache_freeable(struct page *page) |
470 | { | |
ceddc3a5 JW |
471 | /* |
472 | * A freeable page cache page is referenced only by the caller | |
473 | * that isolated the page, the page cache radix tree and | |
474 | * optional buffer heads at page->private. | |
475 | */ | |
edcf4748 | 476 | return page_count(page) - page_has_private(page) == 2; |
1da177e4 LT |
477 | } |
478 | ||
703c2708 | 479 | static int may_write_to_inode(struct inode *inode, struct scan_control *sc) |
1da177e4 | 480 | { |
930d9152 | 481 | if (current->flags & PF_SWAPWRITE) |
1da177e4 | 482 | return 1; |
703c2708 | 483 | if (!inode_write_congested(inode)) |
1da177e4 | 484 | return 1; |
703c2708 | 485 | if (inode_to_bdi(inode) == current->backing_dev_info) |
1da177e4 LT |
486 | return 1; |
487 | return 0; | |
488 | } | |
489 | ||
490 | /* | |
491 | * We detected a synchronous write error writing a page out. Probably | |
492 | * -ENOSPC. We need to propagate that into the address_space for a subsequent | |
493 | * fsync(), msync() or close(). | |
494 | * | |
495 | * The tricky part is that after writepage we cannot touch the mapping: nothing | |
496 | * prevents it from being freed up. But we have a ref on the page and once | |
497 | * that page is locked, the mapping is pinned. | |
498 | * | |
499 | * We're allowed to run sleeping lock_page() here because we know the caller has | |
500 | * __GFP_FS. | |
501 | */ | |
502 | static void handle_write_error(struct address_space *mapping, | |
503 | struct page *page, int error) | |
504 | { | |
7eaceacc | 505 | lock_page(page); |
3e9f45bd GC |
506 | if (page_mapping(page) == mapping) |
507 | mapping_set_error(mapping, error); | |
1da177e4 LT |
508 | unlock_page(page); |
509 | } | |
510 | ||
04e62a29 CL |
511 | /* possible outcome of pageout() */ |
512 | typedef enum { | |
513 | /* failed to write page out, page is locked */ | |
514 | PAGE_KEEP, | |
515 | /* move page to the active list, page is locked */ | |
516 | PAGE_ACTIVATE, | |
517 | /* page has been sent to the disk successfully, page is unlocked */ | |
518 | PAGE_SUCCESS, | |
519 | /* page is clean and locked */ | |
520 | PAGE_CLEAN, | |
521 | } pageout_t; | |
522 | ||
1da177e4 | 523 | /* |
1742f19f AM |
524 | * pageout is called by shrink_page_list() for each dirty page. |
525 | * Calls ->writepage(). | |
1da177e4 | 526 | */ |
c661b078 | 527 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
7d3579e8 | 528 | struct scan_control *sc) |
1da177e4 LT |
529 | { |
530 | /* | |
531 | * If the page is dirty, only perform writeback if that write | |
532 | * will be non-blocking. To prevent this allocation from being | |
533 | * stalled by pagecache activity. But note that there may be | |
534 | * stalls if we need to run get_block(). We could test | |
535 | * PagePrivate for that. | |
536 | * | |
8174202b | 537 | * If this process is currently in __generic_file_write_iter() against |
1da177e4 LT |
538 | * this page's queue, we can perform writeback even if that |
539 | * will block. | |
540 | * | |
541 | * If the page is swapcache, write it back even if that would | |
542 | * block, for some throttling. This happens by accident, because | |
543 | * swap_backing_dev_info is bust: it doesn't reflect the | |
544 | * congestion state of the swapdevs. Easy to fix, if needed. | |
1da177e4 LT |
545 | */ |
546 | if (!is_page_cache_freeable(page)) | |
547 | return PAGE_KEEP; | |
548 | if (!mapping) { | |
549 | /* | |
550 | * Some data journaling orphaned pages can have | |
551 | * page->mapping == NULL while being dirty with clean buffers. | |
552 | */ | |
266cf658 | 553 | if (page_has_private(page)) { |
1da177e4 LT |
554 | if (try_to_free_buffers(page)) { |
555 | ClearPageDirty(page); | |
b1de0d13 | 556 | pr_info("%s: orphaned page\n", __func__); |
1da177e4 LT |
557 | return PAGE_CLEAN; |
558 | } | |
559 | } | |
560 | return PAGE_KEEP; | |
561 | } | |
562 | if (mapping->a_ops->writepage == NULL) | |
563 | return PAGE_ACTIVATE; | |
703c2708 | 564 | if (!may_write_to_inode(mapping->host, sc)) |
1da177e4 LT |
565 | return PAGE_KEEP; |
566 | ||
567 | if (clear_page_dirty_for_io(page)) { | |
568 | int res; | |
569 | struct writeback_control wbc = { | |
570 | .sync_mode = WB_SYNC_NONE, | |
571 | .nr_to_write = SWAP_CLUSTER_MAX, | |
111ebb6e OH |
572 | .range_start = 0, |
573 | .range_end = LLONG_MAX, | |
1da177e4 LT |
574 | .for_reclaim = 1, |
575 | }; | |
576 | ||
577 | SetPageReclaim(page); | |
578 | res = mapping->a_ops->writepage(page, &wbc); | |
579 | if (res < 0) | |
580 | handle_write_error(mapping, page, res); | |
994fc28c | 581 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
1da177e4 LT |
582 | ClearPageReclaim(page); |
583 | return PAGE_ACTIVATE; | |
584 | } | |
c661b078 | 585 | |
1da177e4 LT |
586 | if (!PageWriteback(page)) { |
587 | /* synchronous write or broken a_ops? */ | |
588 | ClearPageReclaim(page); | |
589 | } | |
3aa23851 | 590 | trace_mm_vmscan_writepage(page); |
e129b5c2 | 591 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
1da177e4 LT |
592 | return PAGE_SUCCESS; |
593 | } | |
594 | ||
595 | return PAGE_CLEAN; | |
596 | } | |
597 | ||
a649fd92 | 598 | /* |
e286781d NP |
599 | * Same as remove_mapping, but if the page is removed from the mapping, it |
600 | * gets returned with a refcount of 0. | |
a649fd92 | 601 | */ |
a528910e JW |
602 | static int __remove_mapping(struct address_space *mapping, struct page *page, |
603 | bool reclaimed) | |
49d2e9cc | 604 | { |
c4843a75 | 605 | unsigned long flags; |
c4843a75 | 606 | |
28e4d965 NP |
607 | BUG_ON(!PageLocked(page)); |
608 | BUG_ON(mapping != page_mapping(page)); | |
49d2e9cc | 609 | |
c4843a75 | 610 | spin_lock_irqsave(&mapping->tree_lock, flags); |
49d2e9cc | 611 | /* |
0fd0e6b0 NP |
612 | * The non racy check for a busy page. |
613 | * | |
614 | * Must be careful with the order of the tests. When someone has | |
615 | * a ref to the page, it may be possible that they dirty it then | |
616 | * drop the reference. So if PageDirty is tested before page_count | |
617 | * here, then the following race may occur: | |
618 | * | |
619 | * get_user_pages(&page); | |
620 | * [user mapping goes away] | |
621 | * write_to(page); | |
622 | * !PageDirty(page) [good] | |
623 | * SetPageDirty(page); | |
624 | * put_page(page); | |
625 | * !page_count(page) [good, discard it] | |
626 | * | |
627 | * [oops, our write_to data is lost] | |
628 | * | |
629 | * Reversing the order of the tests ensures such a situation cannot | |
630 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags | |
631 | * load is not satisfied before that of page->_count. | |
632 | * | |
633 | * Note that if SetPageDirty is always performed via set_page_dirty, | |
634 | * and thus under tree_lock, then this ordering is not required. | |
49d2e9cc | 635 | */ |
e286781d | 636 | if (!page_freeze_refs(page, 2)) |
49d2e9cc | 637 | goto cannot_free; |
e286781d NP |
638 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
639 | if (unlikely(PageDirty(page))) { | |
640 | page_unfreeze_refs(page, 2); | |
49d2e9cc | 641 | goto cannot_free; |
e286781d | 642 | } |
49d2e9cc CL |
643 | |
644 | if (PageSwapCache(page)) { | |
645 | swp_entry_t swap = { .val = page_private(page) }; | |
0a31bc97 | 646 | mem_cgroup_swapout(page, swap); |
49d2e9cc | 647 | __delete_from_swap_cache(page); |
c4843a75 | 648 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
0a31bc97 | 649 | swapcache_free(swap); |
e286781d | 650 | } else { |
6072d13c | 651 | void (*freepage)(struct page *); |
a528910e | 652 | void *shadow = NULL; |
6072d13c LT |
653 | |
654 | freepage = mapping->a_ops->freepage; | |
a528910e JW |
655 | /* |
656 | * Remember a shadow entry for reclaimed file cache in | |
657 | * order to detect refaults, thus thrashing, later on. | |
658 | * | |
659 | * But don't store shadows in an address space that is | |
660 | * already exiting. This is not just an optizimation, | |
661 | * inode reclaim needs to empty out the radix tree or | |
662 | * the nodes are lost. Don't plant shadows behind its | |
663 | * back. | |
f9fe48be RZ |
664 | * |
665 | * We also don't store shadows for DAX mappings because the | |
666 | * only page cache pages found in these are zero pages | |
667 | * covering holes, and because we don't want to mix DAX | |
668 | * exceptional entries and shadow exceptional entries in the | |
669 | * same page_tree. | |
a528910e JW |
670 | */ |
671 | if (reclaimed && page_is_file_cache(page) && | |
f9fe48be | 672 | !mapping_exiting(mapping) && !dax_mapping(mapping)) |
a528910e | 673 | shadow = workingset_eviction(mapping, page); |
62cccb8c | 674 | __delete_from_page_cache(page, shadow); |
c4843a75 | 675 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
6072d13c LT |
676 | |
677 | if (freepage != NULL) | |
678 | freepage(page); | |
49d2e9cc CL |
679 | } |
680 | ||
49d2e9cc CL |
681 | return 1; |
682 | ||
683 | cannot_free: | |
c4843a75 | 684 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
49d2e9cc CL |
685 | return 0; |
686 | } | |
687 | ||
e286781d NP |
688 | /* |
689 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if | |
690 | * someone else has a ref on the page, abort and return 0. If it was | |
691 | * successfully detached, return 1. Assumes the caller has a single ref on | |
692 | * this page. | |
693 | */ | |
694 | int remove_mapping(struct address_space *mapping, struct page *page) | |
695 | { | |
a528910e | 696 | if (__remove_mapping(mapping, page, false)) { |
e286781d NP |
697 | /* |
698 | * Unfreezing the refcount with 1 rather than 2 effectively | |
699 | * drops the pagecache ref for us without requiring another | |
700 | * atomic operation. | |
701 | */ | |
702 | page_unfreeze_refs(page, 1); | |
703 | return 1; | |
704 | } | |
705 | return 0; | |
706 | } | |
707 | ||
894bc310 LS |
708 | /** |
709 | * putback_lru_page - put previously isolated page onto appropriate LRU list | |
710 | * @page: page to be put back to appropriate lru list | |
711 | * | |
712 | * Add previously isolated @page to appropriate LRU list. | |
713 | * Page may still be unevictable for other reasons. | |
714 | * | |
715 | * lru_lock must not be held, interrupts must be enabled. | |
716 | */ | |
894bc310 LS |
717 | void putback_lru_page(struct page *page) |
718 | { | |
0ec3b74c | 719 | bool is_unevictable; |
bbfd28ee | 720 | int was_unevictable = PageUnevictable(page); |
894bc310 | 721 | |
309381fe | 722 | VM_BUG_ON_PAGE(PageLRU(page), page); |
894bc310 LS |
723 | |
724 | redo: | |
725 | ClearPageUnevictable(page); | |
726 | ||
39b5f29a | 727 | if (page_evictable(page)) { |
894bc310 LS |
728 | /* |
729 | * For evictable pages, we can use the cache. | |
730 | * In event of a race, worst case is we end up with an | |
731 | * unevictable page on [in]active list. | |
732 | * We know how to handle that. | |
733 | */ | |
0ec3b74c | 734 | is_unevictable = false; |
c53954a0 | 735 | lru_cache_add(page); |
894bc310 LS |
736 | } else { |
737 | /* | |
738 | * Put unevictable pages directly on zone's unevictable | |
739 | * list. | |
740 | */ | |
0ec3b74c | 741 | is_unevictable = true; |
894bc310 | 742 | add_page_to_unevictable_list(page); |
6a7b9548 | 743 | /* |
21ee9f39 MK |
744 | * When racing with an mlock or AS_UNEVICTABLE clearing |
745 | * (page is unlocked) make sure that if the other thread | |
746 | * does not observe our setting of PG_lru and fails | |
24513264 | 747 | * isolation/check_move_unevictable_pages, |
21ee9f39 | 748 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
6a7b9548 JW |
749 | * the page back to the evictable list. |
750 | * | |
21ee9f39 | 751 | * The other side is TestClearPageMlocked() or shmem_lock(). |
6a7b9548 JW |
752 | */ |
753 | smp_mb(); | |
894bc310 | 754 | } |
894bc310 LS |
755 | |
756 | /* | |
757 | * page's status can change while we move it among lru. If an evictable | |
758 | * page is on unevictable list, it never be freed. To avoid that, | |
759 | * check after we added it to the list, again. | |
760 | */ | |
0ec3b74c | 761 | if (is_unevictable && page_evictable(page)) { |
894bc310 LS |
762 | if (!isolate_lru_page(page)) { |
763 | put_page(page); | |
764 | goto redo; | |
765 | } | |
766 | /* This means someone else dropped this page from LRU | |
767 | * So, it will be freed or putback to LRU again. There is | |
768 | * nothing to do here. | |
769 | */ | |
770 | } | |
771 | ||
0ec3b74c | 772 | if (was_unevictable && !is_unevictable) |
bbfd28ee | 773 | count_vm_event(UNEVICTABLE_PGRESCUED); |
0ec3b74c | 774 | else if (!was_unevictable && is_unevictable) |
bbfd28ee LS |
775 | count_vm_event(UNEVICTABLE_PGCULLED); |
776 | ||
894bc310 LS |
777 | put_page(page); /* drop ref from isolate */ |
778 | } | |
779 | ||
dfc8d636 JW |
780 | enum page_references { |
781 | PAGEREF_RECLAIM, | |
782 | PAGEREF_RECLAIM_CLEAN, | |
64574746 | 783 | PAGEREF_KEEP, |
dfc8d636 JW |
784 | PAGEREF_ACTIVATE, |
785 | }; | |
786 | ||
787 | static enum page_references page_check_references(struct page *page, | |
788 | struct scan_control *sc) | |
789 | { | |
64574746 | 790 | int referenced_ptes, referenced_page; |
dfc8d636 | 791 | unsigned long vm_flags; |
dfc8d636 | 792 | |
c3ac9a8a JW |
793 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, |
794 | &vm_flags); | |
64574746 | 795 | referenced_page = TestClearPageReferenced(page); |
dfc8d636 | 796 | |
dfc8d636 JW |
797 | /* |
798 | * Mlock lost the isolation race with us. Let try_to_unmap() | |
799 | * move the page to the unevictable list. | |
800 | */ | |
801 | if (vm_flags & VM_LOCKED) | |
802 | return PAGEREF_RECLAIM; | |
803 | ||
64574746 | 804 | if (referenced_ptes) { |
e4898273 | 805 | if (PageSwapBacked(page)) |
64574746 JW |
806 | return PAGEREF_ACTIVATE; |
807 | /* | |
808 | * All mapped pages start out with page table | |
809 | * references from the instantiating fault, so we need | |
810 | * to look twice if a mapped file page is used more | |
811 | * than once. | |
812 | * | |
813 | * Mark it and spare it for another trip around the | |
814 | * inactive list. Another page table reference will | |
815 | * lead to its activation. | |
816 | * | |
817 | * Note: the mark is set for activated pages as well | |
818 | * so that recently deactivated but used pages are | |
819 | * quickly recovered. | |
820 | */ | |
821 | SetPageReferenced(page); | |
822 | ||
34dbc67a | 823 | if (referenced_page || referenced_ptes > 1) |
64574746 JW |
824 | return PAGEREF_ACTIVATE; |
825 | ||
c909e993 KK |
826 | /* |
827 | * Activate file-backed executable pages after first usage. | |
828 | */ | |
829 | if (vm_flags & VM_EXEC) | |
830 | return PAGEREF_ACTIVATE; | |
831 | ||
64574746 JW |
832 | return PAGEREF_KEEP; |
833 | } | |
dfc8d636 JW |
834 | |
835 | /* Reclaim if clean, defer dirty pages to writeback */ | |
2e30244a | 836 | if (referenced_page && !PageSwapBacked(page)) |
64574746 JW |
837 | return PAGEREF_RECLAIM_CLEAN; |
838 | ||
839 | return PAGEREF_RECLAIM; | |
dfc8d636 JW |
840 | } |
841 | ||
e2be15f6 MG |
842 | /* Check if a page is dirty or under writeback */ |
843 | static void page_check_dirty_writeback(struct page *page, | |
844 | bool *dirty, bool *writeback) | |
845 | { | |
b4597226 MG |
846 | struct address_space *mapping; |
847 | ||
e2be15f6 MG |
848 | /* |
849 | * Anonymous pages are not handled by flushers and must be written | |
850 | * from reclaim context. Do not stall reclaim based on them | |
851 | */ | |
852 | if (!page_is_file_cache(page)) { | |
853 | *dirty = false; | |
854 | *writeback = false; | |
855 | return; | |
856 | } | |
857 | ||
858 | /* By default assume that the page flags are accurate */ | |
859 | *dirty = PageDirty(page); | |
860 | *writeback = PageWriteback(page); | |
b4597226 MG |
861 | |
862 | /* Verify dirty/writeback state if the filesystem supports it */ | |
863 | if (!page_has_private(page)) | |
864 | return; | |
865 | ||
866 | mapping = page_mapping(page); | |
867 | if (mapping && mapping->a_ops->is_dirty_writeback) | |
868 | mapping->a_ops->is_dirty_writeback(page, dirty, writeback); | |
e2be15f6 MG |
869 | } |
870 | ||
1da177e4 | 871 | /* |
1742f19f | 872 | * shrink_page_list() returns the number of reclaimed pages |
1da177e4 | 873 | */ |
1742f19f | 874 | static unsigned long shrink_page_list(struct list_head *page_list, |
6a18adb3 | 875 | struct zone *zone, |
f84f6e2b | 876 | struct scan_control *sc, |
02c6de8d | 877 | enum ttu_flags ttu_flags, |
8e950282 | 878 | unsigned long *ret_nr_dirty, |
d43006d5 | 879 | unsigned long *ret_nr_unqueued_dirty, |
8e950282 | 880 | unsigned long *ret_nr_congested, |
02c6de8d | 881 | unsigned long *ret_nr_writeback, |
b1a6f21e | 882 | unsigned long *ret_nr_immediate, |
02c6de8d | 883 | bool force_reclaim) |
1da177e4 LT |
884 | { |
885 | LIST_HEAD(ret_pages); | |
abe4c3b5 | 886 | LIST_HEAD(free_pages); |
1da177e4 | 887 | int pgactivate = 0; |
d43006d5 | 888 | unsigned long nr_unqueued_dirty = 0; |
0e093d99 MG |
889 | unsigned long nr_dirty = 0; |
890 | unsigned long nr_congested = 0; | |
05ff5137 | 891 | unsigned long nr_reclaimed = 0; |
92df3a72 | 892 | unsigned long nr_writeback = 0; |
b1a6f21e | 893 | unsigned long nr_immediate = 0; |
1da177e4 LT |
894 | |
895 | cond_resched(); | |
896 | ||
1da177e4 LT |
897 | while (!list_empty(page_list)) { |
898 | struct address_space *mapping; | |
899 | struct page *page; | |
900 | int may_enter_fs; | |
02c6de8d | 901 | enum page_references references = PAGEREF_RECLAIM_CLEAN; |
e2be15f6 | 902 | bool dirty, writeback; |
854e9ed0 MK |
903 | bool lazyfree = false; |
904 | int ret = SWAP_SUCCESS; | |
1da177e4 LT |
905 | |
906 | cond_resched(); | |
907 | ||
908 | page = lru_to_page(page_list); | |
909 | list_del(&page->lru); | |
910 | ||
529ae9aa | 911 | if (!trylock_page(page)) |
1da177e4 LT |
912 | goto keep; |
913 | ||
309381fe SL |
914 | VM_BUG_ON_PAGE(PageActive(page), page); |
915 | VM_BUG_ON_PAGE(page_zone(page) != zone, page); | |
1da177e4 LT |
916 | |
917 | sc->nr_scanned++; | |
80e43426 | 918 | |
39b5f29a | 919 | if (unlikely(!page_evictable(page))) |
b291f000 | 920 | goto cull_mlocked; |
894bc310 | 921 | |
a6dc60f8 | 922 | if (!sc->may_unmap && page_mapped(page)) |
80e43426 CL |
923 | goto keep_locked; |
924 | ||
1da177e4 LT |
925 | /* Double the slab pressure for mapped and swapcache pages */ |
926 | if (page_mapped(page) || PageSwapCache(page)) | |
927 | sc->nr_scanned++; | |
928 | ||
c661b078 AW |
929 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
930 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); | |
931 | ||
e2be15f6 MG |
932 | /* |
933 | * The number of dirty pages determines if a zone is marked | |
934 | * reclaim_congested which affects wait_iff_congested. kswapd | |
935 | * will stall and start writing pages if the tail of the LRU | |
936 | * is all dirty unqueued pages. | |
937 | */ | |
938 | page_check_dirty_writeback(page, &dirty, &writeback); | |
939 | if (dirty || writeback) | |
940 | nr_dirty++; | |
941 | ||
942 | if (dirty && !writeback) | |
943 | nr_unqueued_dirty++; | |
944 | ||
d04e8acd MG |
945 | /* |
946 | * Treat this page as congested if the underlying BDI is or if | |
947 | * pages are cycling through the LRU so quickly that the | |
948 | * pages marked for immediate reclaim are making it to the | |
949 | * end of the LRU a second time. | |
950 | */ | |
e2be15f6 | 951 | mapping = page_mapping(page); |
1da58ee2 | 952 | if (((dirty || writeback) && mapping && |
703c2708 | 953 | inode_write_congested(mapping->host)) || |
d04e8acd | 954 | (writeback && PageReclaim(page))) |
e2be15f6 MG |
955 | nr_congested++; |
956 | ||
283aba9f MG |
957 | /* |
958 | * If a page at the tail of the LRU is under writeback, there | |
959 | * are three cases to consider. | |
960 | * | |
961 | * 1) If reclaim is encountering an excessive number of pages | |
962 | * under writeback and this page is both under writeback and | |
963 | * PageReclaim then it indicates that pages are being queued | |
964 | * for IO but are being recycled through the LRU before the | |
965 | * IO can complete. Waiting on the page itself risks an | |
966 | * indefinite stall if it is impossible to writeback the | |
967 | * page due to IO error or disconnected storage so instead | |
b1a6f21e MG |
968 | * note that the LRU is being scanned too quickly and the |
969 | * caller can stall after page list has been processed. | |
283aba9f | 970 | * |
97c9341f | 971 | * 2) Global or new memcg reclaim encounters a page that is |
ecf5fc6e MH |
972 | * not marked for immediate reclaim, or the caller does not |
973 | * have __GFP_FS (or __GFP_IO if it's simply going to swap, | |
974 | * not to fs). In this case mark the page for immediate | |
97c9341f | 975 | * reclaim and continue scanning. |
283aba9f | 976 | * |
ecf5fc6e MH |
977 | * Require may_enter_fs because we would wait on fs, which |
978 | * may not have submitted IO yet. And the loop driver might | |
283aba9f MG |
979 | * enter reclaim, and deadlock if it waits on a page for |
980 | * which it is needed to do the write (loop masks off | |
981 | * __GFP_IO|__GFP_FS for this reason); but more thought | |
982 | * would probably show more reasons. | |
983 | * | |
7fadc820 | 984 | * 3) Legacy memcg encounters a page that is already marked |
283aba9f MG |
985 | * PageReclaim. memcg does not have any dirty pages |
986 | * throttling so we could easily OOM just because too many | |
987 | * pages are in writeback and there is nothing else to | |
988 | * reclaim. Wait for the writeback to complete. | |
989 | */ | |
c661b078 | 990 | if (PageWriteback(page)) { |
283aba9f MG |
991 | /* Case 1 above */ |
992 | if (current_is_kswapd() && | |
993 | PageReclaim(page) && | |
57054651 | 994 | test_bit(ZONE_WRITEBACK, &zone->flags)) { |
b1a6f21e MG |
995 | nr_immediate++; |
996 | goto keep_locked; | |
283aba9f MG |
997 | |
998 | /* Case 2 above */ | |
97c9341f | 999 | } else if (sane_reclaim(sc) || |
ecf5fc6e | 1000 | !PageReclaim(page) || !may_enter_fs) { |
c3b94f44 HD |
1001 | /* |
1002 | * This is slightly racy - end_page_writeback() | |
1003 | * might have just cleared PageReclaim, then | |
1004 | * setting PageReclaim here end up interpreted | |
1005 | * as PageReadahead - but that does not matter | |
1006 | * enough to care. What we do want is for this | |
1007 | * page to have PageReclaim set next time memcg | |
1008 | * reclaim reaches the tests above, so it will | |
1009 | * then wait_on_page_writeback() to avoid OOM; | |
1010 | * and it's also appropriate in global reclaim. | |
1011 | */ | |
1012 | SetPageReclaim(page); | |
e62e384e | 1013 | nr_writeback++; |
c3b94f44 | 1014 | goto keep_locked; |
283aba9f MG |
1015 | |
1016 | /* Case 3 above */ | |
1017 | } else { | |
7fadc820 | 1018 | unlock_page(page); |
283aba9f | 1019 | wait_on_page_writeback(page); |
7fadc820 HD |
1020 | /* then go back and try same page again */ |
1021 | list_add_tail(&page->lru, page_list); | |
1022 | continue; | |
e62e384e | 1023 | } |
c661b078 | 1024 | } |
1da177e4 | 1025 | |
02c6de8d MK |
1026 | if (!force_reclaim) |
1027 | references = page_check_references(page, sc); | |
1028 | ||
dfc8d636 JW |
1029 | switch (references) { |
1030 | case PAGEREF_ACTIVATE: | |
1da177e4 | 1031 | goto activate_locked; |
64574746 JW |
1032 | case PAGEREF_KEEP: |
1033 | goto keep_locked; | |
dfc8d636 JW |
1034 | case PAGEREF_RECLAIM: |
1035 | case PAGEREF_RECLAIM_CLEAN: | |
1036 | ; /* try to reclaim the page below */ | |
1037 | } | |
1da177e4 | 1038 | |
1da177e4 LT |
1039 | /* |
1040 | * Anonymous process memory has backing store? | |
1041 | * Try to allocate it some swap space here. | |
1042 | */ | |
b291f000 | 1043 | if (PageAnon(page) && !PageSwapCache(page)) { |
63eb6b93 HD |
1044 | if (!(sc->gfp_mask & __GFP_IO)) |
1045 | goto keep_locked; | |
5bc7b8ac | 1046 | if (!add_to_swap(page, page_list)) |
1da177e4 | 1047 | goto activate_locked; |
854e9ed0 | 1048 | lazyfree = true; |
63eb6b93 | 1049 | may_enter_fs = 1; |
1da177e4 | 1050 | |
e2be15f6 MG |
1051 | /* Adding to swap updated mapping */ |
1052 | mapping = page_mapping(page); | |
1053 | } | |
1da177e4 LT |
1054 | |
1055 | /* | |
1056 | * The page is mapped into the page tables of one or more | |
1057 | * processes. Try to unmap it here. | |
1058 | */ | |
1059 | if (page_mapped(page) && mapping) { | |
854e9ed0 MK |
1060 | switch (ret = try_to_unmap(page, lazyfree ? |
1061 | (ttu_flags | TTU_BATCH_FLUSH | TTU_LZFREE) : | |
1062 | (ttu_flags | TTU_BATCH_FLUSH))) { | |
1da177e4 LT |
1063 | case SWAP_FAIL: |
1064 | goto activate_locked; | |
1065 | case SWAP_AGAIN: | |
1066 | goto keep_locked; | |
b291f000 NP |
1067 | case SWAP_MLOCK: |
1068 | goto cull_mlocked; | |
854e9ed0 MK |
1069 | case SWAP_LZFREE: |
1070 | goto lazyfree; | |
1da177e4 LT |
1071 | case SWAP_SUCCESS: |
1072 | ; /* try to free the page below */ | |
1073 | } | |
1074 | } | |
1075 | ||
1076 | if (PageDirty(page)) { | |
ee72886d MG |
1077 | /* |
1078 | * Only kswapd can writeback filesystem pages to | |
d43006d5 MG |
1079 | * avoid risk of stack overflow but only writeback |
1080 | * if many dirty pages have been encountered. | |
ee72886d | 1081 | */ |
f84f6e2b | 1082 | if (page_is_file_cache(page) && |
9e3b2f8c | 1083 | (!current_is_kswapd() || |
57054651 | 1084 | !test_bit(ZONE_DIRTY, &zone->flags))) { |
49ea7eb6 MG |
1085 | /* |
1086 | * Immediately reclaim when written back. | |
1087 | * Similar in principal to deactivate_page() | |
1088 | * except we already have the page isolated | |
1089 | * and know it's dirty | |
1090 | */ | |
1091 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); | |
1092 | SetPageReclaim(page); | |
1093 | ||
ee72886d MG |
1094 | goto keep_locked; |
1095 | } | |
1096 | ||
dfc8d636 | 1097 | if (references == PAGEREF_RECLAIM_CLEAN) |
1da177e4 | 1098 | goto keep_locked; |
4dd4b920 | 1099 | if (!may_enter_fs) |
1da177e4 | 1100 | goto keep_locked; |
52a8363e | 1101 | if (!sc->may_writepage) |
1da177e4 LT |
1102 | goto keep_locked; |
1103 | ||
d950c947 MG |
1104 | /* |
1105 | * Page is dirty. Flush the TLB if a writable entry | |
1106 | * potentially exists to avoid CPU writes after IO | |
1107 | * starts and then write it out here. | |
1108 | */ | |
1109 | try_to_unmap_flush_dirty(); | |
7d3579e8 | 1110 | switch (pageout(page, mapping, sc)) { |
1da177e4 LT |
1111 | case PAGE_KEEP: |
1112 | goto keep_locked; | |
1113 | case PAGE_ACTIVATE: | |
1114 | goto activate_locked; | |
1115 | case PAGE_SUCCESS: | |
7d3579e8 | 1116 | if (PageWriteback(page)) |
41ac1999 | 1117 | goto keep; |
7d3579e8 | 1118 | if (PageDirty(page)) |
1da177e4 | 1119 | goto keep; |
7d3579e8 | 1120 | |
1da177e4 LT |
1121 | /* |
1122 | * A synchronous write - probably a ramdisk. Go | |
1123 | * ahead and try to reclaim the page. | |
1124 | */ | |
529ae9aa | 1125 | if (!trylock_page(page)) |
1da177e4 LT |
1126 | goto keep; |
1127 | if (PageDirty(page) || PageWriteback(page)) | |
1128 | goto keep_locked; | |
1129 | mapping = page_mapping(page); | |
1130 | case PAGE_CLEAN: | |
1131 | ; /* try to free the page below */ | |
1132 | } | |
1133 | } | |
1134 | ||
1135 | /* | |
1136 | * If the page has buffers, try to free the buffer mappings | |
1137 | * associated with this page. If we succeed we try to free | |
1138 | * the page as well. | |
1139 | * | |
1140 | * We do this even if the page is PageDirty(). | |
1141 | * try_to_release_page() does not perform I/O, but it is | |
1142 | * possible for a page to have PageDirty set, but it is actually | |
1143 | * clean (all its buffers are clean). This happens if the | |
1144 | * buffers were written out directly, with submit_bh(). ext3 | |
894bc310 | 1145 | * will do this, as well as the blockdev mapping. |
1da177e4 LT |
1146 | * try_to_release_page() will discover that cleanness and will |
1147 | * drop the buffers and mark the page clean - it can be freed. | |
1148 | * | |
1149 | * Rarely, pages can have buffers and no ->mapping. These are | |
1150 | * the pages which were not successfully invalidated in | |
1151 | * truncate_complete_page(). We try to drop those buffers here | |
1152 | * and if that worked, and the page is no longer mapped into | |
1153 | * process address space (page_count == 1) it can be freed. | |
1154 | * Otherwise, leave the page on the LRU so it is swappable. | |
1155 | */ | |
266cf658 | 1156 | if (page_has_private(page)) { |
1da177e4 LT |
1157 | if (!try_to_release_page(page, sc->gfp_mask)) |
1158 | goto activate_locked; | |
e286781d NP |
1159 | if (!mapping && page_count(page) == 1) { |
1160 | unlock_page(page); | |
1161 | if (put_page_testzero(page)) | |
1162 | goto free_it; | |
1163 | else { | |
1164 | /* | |
1165 | * rare race with speculative reference. | |
1166 | * the speculative reference will free | |
1167 | * this page shortly, so we may | |
1168 | * increment nr_reclaimed here (and | |
1169 | * leave it off the LRU). | |
1170 | */ | |
1171 | nr_reclaimed++; | |
1172 | continue; | |
1173 | } | |
1174 | } | |
1da177e4 LT |
1175 | } |
1176 | ||
854e9ed0 | 1177 | lazyfree: |
a528910e | 1178 | if (!mapping || !__remove_mapping(mapping, page, true)) |
49d2e9cc | 1179 | goto keep_locked; |
1da177e4 | 1180 | |
a978d6f5 NP |
1181 | /* |
1182 | * At this point, we have no other references and there is | |
1183 | * no way to pick any more up (removed from LRU, removed | |
1184 | * from pagecache). Can use non-atomic bitops now (and | |
1185 | * we obviously don't have to worry about waking up a process | |
1186 | * waiting on the page lock, because there are no references. | |
1187 | */ | |
48c935ad | 1188 | __ClearPageLocked(page); |
e286781d | 1189 | free_it: |
854e9ed0 MK |
1190 | if (ret == SWAP_LZFREE) |
1191 | count_vm_event(PGLAZYFREED); | |
1192 | ||
05ff5137 | 1193 | nr_reclaimed++; |
abe4c3b5 MG |
1194 | |
1195 | /* | |
1196 | * Is there need to periodically free_page_list? It would | |
1197 | * appear not as the counts should be low | |
1198 | */ | |
1199 | list_add(&page->lru, &free_pages); | |
1da177e4 LT |
1200 | continue; |
1201 | ||
b291f000 | 1202 | cull_mlocked: |
63d6c5ad HD |
1203 | if (PageSwapCache(page)) |
1204 | try_to_free_swap(page); | |
b291f000 | 1205 | unlock_page(page); |
c54839a7 | 1206 | list_add(&page->lru, &ret_pages); |
b291f000 NP |
1207 | continue; |
1208 | ||
1da177e4 | 1209 | activate_locked: |
68a22394 | 1210 | /* Not a candidate for swapping, so reclaim swap space. */ |
5ccc5aba | 1211 | if (PageSwapCache(page) && mem_cgroup_swap_full(page)) |
a2c43eed | 1212 | try_to_free_swap(page); |
309381fe | 1213 | VM_BUG_ON_PAGE(PageActive(page), page); |
1da177e4 LT |
1214 | SetPageActive(page); |
1215 | pgactivate++; | |
1216 | keep_locked: | |
1217 | unlock_page(page); | |
1218 | keep: | |
1219 | list_add(&page->lru, &ret_pages); | |
309381fe | 1220 | VM_BUG_ON_PAGE(PageLRU(page) || PageUnevictable(page), page); |
1da177e4 | 1221 | } |
abe4c3b5 | 1222 | |
747db954 | 1223 | mem_cgroup_uncharge_list(&free_pages); |
72b252ae | 1224 | try_to_unmap_flush(); |
b745bc85 | 1225 | free_hot_cold_page_list(&free_pages, true); |
abe4c3b5 | 1226 | |
1da177e4 | 1227 | list_splice(&ret_pages, page_list); |
f8891e5e | 1228 | count_vm_events(PGACTIVATE, pgactivate); |
0a31bc97 | 1229 | |
8e950282 MG |
1230 | *ret_nr_dirty += nr_dirty; |
1231 | *ret_nr_congested += nr_congested; | |
d43006d5 | 1232 | *ret_nr_unqueued_dirty += nr_unqueued_dirty; |
92df3a72 | 1233 | *ret_nr_writeback += nr_writeback; |
b1a6f21e | 1234 | *ret_nr_immediate += nr_immediate; |
05ff5137 | 1235 | return nr_reclaimed; |
1da177e4 LT |
1236 | } |
1237 | ||
02c6de8d MK |
1238 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, |
1239 | struct list_head *page_list) | |
1240 | { | |
1241 | struct scan_control sc = { | |
1242 | .gfp_mask = GFP_KERNEL, | |
1243 | .priority = DEF_PRIORITY, | |
1244 | .may_unmap = 1, | |
1245 | }; | |
8e950282 | 1246 | unsigned long ret, dummy1, dummy2, dummy3, dummy4, dummy5; |
02c6de8d MK |
1247 | struct page *page, *next; |
1248 | LIST_HEAD(clean_pages); | |
1249 | ||
1250 | list_for_each_entry_safe(page, next, page_list, lru) { | |
117aad1e RA |
1251 | if (page_is_file_cache(page) && !PageDirty(page) && |
1252 | !isolated_balloon_page(page)) { | |
02c6de8d MK |
1253 | ClearPageActive(page); |
1254 | list_move(&page->lru, &clean_pages); | |
1255 | } | |
1256 | } | |
1257 | ||
1258 | ret = shrink_page_list(&clean_pages, zone, &sc, | |
8e950282 MG |
1259 | TTU_UNMAP|TTU_IGNORE_ACCESS, |
1260 | &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, true); | |
02c6de8d | 1261 | list_splice(&clean_pages, page_list); |
83da7510 | 1262 | mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret); |
02c6de8d MK |
1263 | return ret; |
1264 | } | |
1265 | ||
5ad333eb AW |
1266 | /* |
1267 | * Attempt to remove the specified page from its LRU. Only take this page | |
1268 | * if it is of the appropriate PageActive status. Pages which are being | |
1269 | * freed elsewhere are also ignored. | |
1270 | * | |
1271 | * page: page to consider | |
1272 | * mode: one of the LRU isolation modes defined above | |
1273 | * | |
1274 | * returns 0 on success, -ve errno on failure. | |
1275 | */ | |
f3fd4a61 | 1276 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) |
5ad333eb AW |
1277 | { |
1278 | int ret = -EINVAL; | |
1279 | ||
1280 | /* Only take pages on the LRU. */ | |
1281 | if (!PageLRU(page)) | |
1282 | return ret; | |
1283 | ||
e46a2879 MK |
1284 | /* Compaction should not handle unevictable pages but CMA can do so */ |
1285 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) | |
894bc310 LS |
1286 | return ret; |
1287 | ||
5ad333eb | 1288 | ret = -EBUSY; |
08e552c6 | 1289 | |
c8244935 MG |
1290 | /* |
1291 | * To minimise LRU disruption, the caller can indicate that it only | |
1292 | * wants to isolate pages it will be able to operate on without | |
1293 | * blocking - clean pages for the most part. | |
1294 | * | |
1295 | * ISOLATE_CLEAN means that only clean pages should be isolated. This | |
1296 | * is used by reclaim when it is cannot write to backing storage | |
1297 | * | |
1298 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages | |
1299 | * that it is possible to migrate without blocking | |
1300 | */ | |
1301 | if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) { | |
1302 | /* All the caller can do on PageWriteback is block */ | |
1303 | if (PageWriteback(page)) | |
1304 | return ret; | |
1305 | ||
1306 | if (PageDirty(page)) { | |
1307 | struct address_space *mapping; | |
1308 | ||
1309 | /* ISOLATE_CLEAN means only clean pages */ | |
1310 | if (mode & ISOLATE_CLEAN) | |
1311 | return ret; | |
1312 | ||
1313 | /* | |
1314 | * Only pages without mappings or that have a | |
1315 | * ->migratepage callback are possible to migrate | |
1316 | * without blocking | |
1317 | */ | |
1318 | mapping = page_mapping(page); | |
1319 | if (mapping && !mapping->a_ops->migratepage) | |
1320 | return ret; | |
1321 | } | |
1322 | } | |
39deaf85 | 1323 | |
f80c0673 MK |
1324 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
1325 | return ret; | |
1326 | ||
5ad333eb AW |
1327 | if (likely(get_page_unless_zero(page))) { |
1328 | /* | |
1329 | * Be careful not to clear PageLRU until after we're | |
1330 | * sure the page is not being freed elsewhere -- the | |
1331 | * page release code relies on it. | |
1332 | */ | |
1333 | ClearPageLRU(page); | |
1334 | ret = 0; | |
1335 | } | |
1336 | ||
1337 | return ret; | |
1338 | } | |
1339 | ||
1da177e4 LT |
1340 | /* |
1341 | * zone->lru_lock is heavily contended. Some of the functions that | |
1342 | * shrink the lists perform better by taking out a batch of pages | |
1343 | * and working on them outside the LRU lock. | |
1344 | * | |
1345 | * For pagecache intensive workloads, this function is the hottest | |
1346 | * spot in the kernel (apart from copy_*_user functions). | |
1347 | * | |
1348 | * Appropriate locks must be held before calling this function. | |
1349 | * | |
1350 | * @nr_to_scan: The number of pages to look through on the list. | |
5dc35979 | 1351 | * @lruvec: The LRU vector to pull pages from. |
1da177e4 | 1352 | * @dst: The temp list to put pages on to. |
f626012d | 1353 | * @nr_scanned: The number of pages that were scanned. |
fe2c2a10 | 1354 | * @sc: The scan_control struct for this reclaim session |
5ad333eb | 1355 | * @mode: One of the LRU isolation modes |
3cb99451 | 1356 | * @lru: LRU list id for isolating |
1da177e4 LT |
1357 | * |
1358 | * returns how many pages were moved onto *@dst. | |
1359 | */ | |
69e05944 | 1360 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
5dc35979 | 1361 | struct lruvec *lruvec, struct list_head *dst, |
fe2c2a10 | 1362 | unsigned long *nr_scanned, struct scan_control *sc, |
3cb99451 | 1363 | isolate_mode_t mode, enum lru_list lru) |
1da177e4 | 1364 | { |
75b00af7 | 1365 | struct list_head *src = &lruvec->lists[lru]; |
69e05944 | 1366 | unsigned long nr_taken = 0; |
c9b02d97 | 1367 | unsigned long scan; |
1da177e4 | 1368 | |
0b802f10 VD |
1369 | for (scan = 0; scan < nr_to_scan && nr_taken < nr_to_scan && |
1370 | !list_empty(src); scan++) { | |
5ad333eb | 1371 | struct page *page; |
fa9add64 | 1372 | int nr_pages; |
5ad333eb | 1373 | |
1da177e4 LT |
1374 | page = lru_to_page(src); |
1375 | prefetchw_prev_lru_page(page, src, flags); | |
1376 | ||
309381fe | 1377 | VM_BUG_ON_PAGE(!PageLRU(page), page); |
8d438f96 | 1378 | |
f3fd4a61 | 1379 | switch (__isolate_lru_page(page, mode)) { |
5ad333eb | 1380 | case 0: |
fa9add64 HD |
1381 | nr_pages = hpage_nr_pages(page); |
1382 | mem_cgroup_update_lru_size(lruvec, lru, -nr_pages); | |
5ad333eb | 1383 | list_move(&page->lru, dst); |
fa9add64 | 1384 | nr_taken += nr_pages; |
5ad333eb AW |
1385 | break; |
1386 | ||
1387 | case -EBUSY: | |
1388 | /* else it is being freed elsewhere */ | |
1389 | list_move(&page->lru, src); | |
1390 | continue; | |
46453a6e | 1391 | |
5ad333eb AW |
1392 | default: |
1393 | BUG(); | |
1394 | } | |
1da177e4 LT |
1395 | } |
1396 | ||
f626012d | 1397 | *nr_scanned = scan; |
75b00af7 HD |
1398 | trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan, |
1399 | nr_taken, mode, is_file_lru(lru)); | |
1da177e4 LT |
1400 | return nr_taken; |
1401 | } | |
1402 | ||
62695a84 NP |
1403 | /** |
1404 | * isolate_lru_page - tries to isolate a page from its LRU list | |
1405 | * @page: page to isolate from its LRU list | |
1406 | * | |
1407 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the | |
1408 | * vmstat statistic corresponding to whatever LRU list the page was on. | |
1409 | * | |
1410 | * Returns 0 if the page was removed from an LRU list. | |
1411 | * Returns -EBUSY if the page was not on an LRU list. | |
1412 | * | |
1413 | * The returned page will have PageLRU() cleared. If it was found on | |
894bc310 LS |
1414 | * the active list, it will have PageActive set. If it was found on |
1415 | * the unevictable list, it will have the PageUnevictable bit set. That flag | |
1416 | * may need to be cleared by the caller before letting the page go. | |
62695a84 NP |
1417 | * |
1418 | * The vmstat statistic corresponding to the list on which the page was | |
1419 | * found will be decremented. | |
1420 | * | |
1421 | * Restrictions: | |
1422 | * (1) Must be called with an elevated refcount on the page. This is a | |
1423 | * fundamentnal difference from isolate_lru_pages (which is called | |
1424 | * without a stable reference). | |
1425 | * (2) the lru_lock must not be held. | |
1426 | * (3) interrupts must be enabled. | |
1427 | */ | |
1428 | int isolate_lru_page(struct page *page) | |
1429 | { | |
1430 | int ret = -EBUSY; | |
1431 | ||
309381fe | 1432 | VM_BUG_ON_PAGE(!page_count(page), page); |
cf2a82ee | 1433 | WARN_RATELIMIT(PageTail(page), "trying to isolate tail page"); |
0c917313 | 1434 | |
62695a84 NP |
1435 | if (PageLRU(page)) { |
1436 | struct zone *zone = page_zone(page); | |
fa9add64 | 1437 | struct lruvec *lruvec; |
62695a84 NP |
1438 | |
1439 | spin_lock_irq(&zone->lru_lock); | |
fa9add64 | 1440 | lruvec = mem_cgroup_page_lruvec(page, zone); |
0c917313 | 1441 | if (PageLRU(page)) { |
894bc310 | 1442 | int lru = page_lru(page); |
0c917313 | 1443 | get_page(page); |
62695a84 | 1444 | ClearPageLRU(page); |
fa9add64 HD |
1445 | del_page_from_lru_list(page, lruvec, lru); |
1446 | ret = 0; | |
62695a84 NP |
1447 | } |
1448 | spin_unlock_irq(&zone->lru_lock); | |
1449 | } | |
1450 | return ret; | |
1451 | } | |
1452 | ||
35cd7815 | 1453 | /* |
d37dd5dc FW |
1454 | * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and |
1455 | * then get resheduled. When there are massive number of tasks doing page | |
1456 | * allocation, such sleeping direct reclaimers may keep piling up on each CPU, | |
1457 | * the LRU list will go small and be scanned faster than necessary, leading to | |
1458 | * unnecessary swapping, thrashing and OOM. | |
35cd7815 RR |
1459 | */ |
1460 | static int too_many_isolated(struct zone *zone, int file, | |
1461 | struct scan_control *sc) | |
1462 | { | |
1463 | unsigned long inactive, isolated; | |
1464 | ||
1465 | if (current_is_kswapd()) | |
1466 | return 0; | |
1467 | ||
97c9341f | 1468 | if (!sane_reclaim(sc)) |
35cd7815 RR |
1469 | return 0; |
1470 | ||
1471 | if (file) { | |
1472 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); | |
1473 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); | |
1474 | } else { | |
1475 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1476 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); | |
1477 | } | |
1478 | ||
3cf23841 FW |
1479 | /* |
1480 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they | |
1481 | * won't get blocked by normal direct-reclaimers, forming a circular | |
1482 | * deadlock. | |
1483 | */ | |
d0164adc | 1484 | if ((sc->gfp_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS)) |
3cf23841 FW |
1485 | inactive >>= 3; |
1486 | ||
35cd7815 RR |
1487 | return isolated > inactive; |
1488 | } | |
1489 | ||
66635629 | 1490 | static noinline_for_stack void |
75b00af7 | 1491 | putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list) |
66635629 | 1492 | { |
27ac81d8 KK |
1493 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1494 | struct zone *zone = lruvec_zone(lruvec); | |
3f79768f | 1495 | LIST_HEAD(pages_to_free); |
66635629 | 1496 | |
66635629 MG |
1497 | /* |
1498 | * Put back any unfreeable pages. | |
1499 | */ | |
66635629 | 1500 | while (!list_empty(page_list)) { |
3f79768f | 1501 | struct page *page = lru_to_page(page_list); |
66635629 | 1502 | int lru; |
3f79768f | 1503 | |
309381fe | 1504 | VM_BUG_ON_PAGE(PageLRU(page), page); |
66635629 | 1505 | list_del(&page->lru); |
39b5f29a | 1506 | if (unlikely(!page_evictable(page))) { |
66635629 MG |
1507 | spin_unlock_irq(&zone->lru_lock); |
1508 | putback_lru_page(page); | |
1509 | spin_lock_irq(&zone->lru_lock); | |
1510 | continue; | |
1511 | } | |
fa9add64 HD |
1512 | |
1513 | lruvec = mem_cgroup_page_lruvec(page, zone); | |
1514 | ||
7a608572 | 1515 | SetPageLRU(page); |
66635629 | 1516 | lru = page_lru(page); |
fa9add64 HD |
1517 | add_page_to_lru_list(page, lruvec, lru); |
1518 | ||
66635629 MG |
1519 | if (is_active_lru(lru)) { |
1520 | int file = is_file_lru(lru); | |
9992af10 RR |
1521 | int numpages = hpage_nr_pages(page); |
1522 | reclaim_stat->recent_rotated[file] += numpages; | |
66635629 | 1523 | } |
2bcf8879 HD |
1524 | if (put_page_testzero(page)) { |
1525 | __ClearPageLRU(page); | |
1526 | __ClearPageActive(page); | |
fa9add64 | 1527 | del_page_from_lru_list(page, lruvec, lru); |
2bcf8879 HD |
1528 | |
1529 | if (unlikely(PageCompound(page))) { | |
1530 | spin_unlock_irq(&zone->lru_lock); | |
747db954 | 1531 | mem_cgroup_uncharge(page); |
2bcf8879 HD |
1532 | (*get_compound_page_dtor(page))(page); |
1533 | spin_lock_irq(&zone->lru_lock); | |
1534 | } else | |
1535 | list_add(&page->lru, &pages_to_free); | |
66635629 MG |
1536 | } |
1537 | } | |
66635629 | 1538 | |
3f79768f HD |
1539 | /* |
1540 | * To save our caller's stack, now use input list for pages to free. | |
1541 | */ | |
1542 | list_splice(&pages_to_free, page_list); | |
66635629 MG |
1543 | } |
1544 | ||
399ba0b9 N |
1545 | /* |
1546 | * If a kernel thread (such as nfsd for loop-back mounts) services | |
1547 | * a backing device by writing to the page cache it sets PF_LESS_THROTTLE. | |
1548 | * In that case we should only throttle if the backing device it is | |
1549 | * writing to is congested. In other cases it is safe to throttle. | |
1550 | */ | |
1551 | static int current_may_throttle(void) | |
1552 | { | |
1553 | return !(current->flags & PF_LESS_THROTTLE) || | |
1554 | current->backing_dev_info == NULL || | |
1555 | bdi_write_congested(current->backing_dev_info); | |
1556 | } | |
1557 | ||
1da177e4 | 1558 | /* |
1742f19f AM |
1559 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
1560 | * of reclaimed pages | |
1da177e4 | 1561 | */ |
66635629 | 1562 | static noinline_for_stack unsigned long |
1a93be0e | 1563 | shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec, |
9e3b2f8c | 1564 | struct scan_control *sc, enum lru_list lru) |
1da177e4 LT |
1565 | { |
1566 | LIST_HEAD(page_list); | |
e247dbce | 1567 | unsigned long nr_scanned; |
05ff5137 | 1568 | unsigned long nr_reclaimed = 0; |
e247dbce | 1569 | unsigned long nr_taken; |
8e950282 MG |
1570 | unsigned long nr_dirty = 0; |
1571 | unsigned long nr_congested = 0; | |
e2be15f6 | 1572 | unsigned long nr_unqueued_dirty = 0; |
92df3a72 | 1573 | unsigned long nr_writeback = 0; |
b1a6f21e | 1574 | unsigned long nr_immediate = 0; |
f3fd4a61 | 1575 | isolate_mode_t isolate_mode = 0; |
3cb99451 | 1576 | int file = is_file_lru(lru); |
1a93be0e KK |
1577 | struct zone *zone = lruvec_zone(lruvec); |
1578 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; | |
78dc583d | 1579 | |
35cd7815 | 1580 | while (unlikely(too_many_isolated(zone, file, sc))) { |
58355c78 | 1581 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
35cd7815 RR |
1582 | |
1583 | /* We are about to die and free our memory. Return now. */ | |
1584 | if (fatal_signal_pending(current)) | |
1585 | return SWAP_CLUSTER_MAX; | |
1586 | } | |
1587 | ||
1da177e4 | 1588 | lru_add_drain(); |
f80c0673 MK |
1589 | |
1590 | if (!sc->may_unmap) | |
61317289 | 1591 | isolate_mode |= ISOLATE_UNMAPPED; |
f80c0673 | 1592 | if (!sc->may_writepage) |
61317289 | 1593 | isolate_mode |= ISOLATE_CLEAN; |
f80c0673 | 1594 | |
1da177e4 | 1595 | spin_lock_irq(&zone->lru_lock); |
b35ea17b | 1596 | |
5dc35979 KK |
1597 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, |
1598 | &nr_scanned, sc, isolate_mode, lru); | |
95d918fc KK |
1599 | |
1600 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); | |
1601 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); | |
1602 | ||
89b5fae5 | 1603 | if (global_reclaim(sc)) { |
0d5d823a | 1604 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, nr_scanned); |
e247dbce | 1605 | if (current_is_kswapd()) |
75b00af7 | 1606 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned); |
e247dbce | 1607 | else |
75b00af7 | 1608 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned); |
e247dbce | 1609 | } |
d563c050 | 1610 | spin_unlock_irq(&zone->lru_lock); |
b35ea17b | 1611 | |
d563c050 | 1612 | if (nr_taken == 0) |
66635629 | 1613 | return 0; |
5ad333eb | 1614 | |
02c6de8d | 1615 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP, |
8e950282 MG |
1616 | &nr_dirty, &nr_unqueued_dirty, &nr_congested, |
1617 | &nr_writeback, &nr_immediate, | |
1618 | false); | |
c661b078 | 1619 | |
3f79768f HD |
1620 | spin_lock_irq(&zone->lru_lock); |
1621 | ||
95d918fc | 1622 | reclaim_stat->recent_scanned[file] += nr_taken; |
d563c050 | 1623 | |
904249aa YH |
1624 | if (global_reclaim(sc)) { |
1625 | if (current_is_kswapd()) | |
1626 | __count_zone_vm_events(PGSTEAL_KSWAPD, zone, | |
1627 | nr_reclaimed); | |
1628 | else | |
1629 | __count_zone_vm_events(PGSTEAL_DIRECT, zone, | |
1630 | nr_reclaimed); | |
1631 | } | |
a74609fa | 1632 | |
27ac81d8 | 1633 | putback_inactive_pages(lruvec, &page_list); |
3f79768f | 1634 | |
95d918fc | 1635 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
3f79768f HD |
1636 | |
1637 | spin_unlock_irq(&zone->lru_lock); | |
1638 | ||
747db954 | 1639 | mem_cgroup_uncharge_list(&page_list); |
b745bc85 | 1640 | free_hot_cold_page_list(&page_list, true); |
e11da5b4 | 1641 | |
92df3a72 MG |
1642 | /* |
1643 | * If reclaim is isolating dirty pages under writeback, it implies | |
1644 | * that the long-lived page allocation rate is exceeding the page | |
1645 | * laundering rate. Either the global limits are not being effective | |
1646 | * at throttling processes due to the page distribution throughout | |
1647 | * zones or there is heavy usage of a slow backing device. The | |
1648 | * only option is to throttle from reclaim context which is not ideal | |
1649 | * as there is no guarantee the dirtying process is throttled in the | |
1650 | * same way balance_dirty_pages() manages. | |
1651 | * | |
8e950282 MG |
1652 | * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number |
1653 | * of pages under pages flagged for immediate reclaim and stall if any | |
1654 | * are encountered in the nr_immediate check below. | |
92df3a72 | 1655 | */ |
918fc718 | 1656 | if (nr_writeback && nr_writeback == nr_taken) |
57054651 | 1657 | set_bit(ZONE_WRITEBACK, &zone->flags); |
92df3a72 | 1658 | |
d43006d5 | 1659 | /* |
97c9341f TH |
1660 | * Legacy memcg will stall in page writeback so avoid forcibly |
1661 | * stalling here. | |
d43006d5 | 1662 | */ |
97c9341f | 1663 | if (sane_reclaim(sc)) { |
8e950282 MG |
1664 | /* |
1665 | * Tag a zone as congested if all the dirty pages scanned were | |
1666 | * backed by a congested BDI and wait_iff_congested will stall. | |
1667 | */ | |
1668 | if (nr_dirty && nr_dirty == nr_congested) | |
57054651 | 1669 | set_bit(ZONE_CONGESTED, &zone->flags); |
8e950282 | 1670 | |
b1a6f21e MG |
1671 | /* |
1672 | * If dirty pages are scanned that are not queued for IO, it | |
1673 | * implies that flushers are not keeping up. In this case, flag | |
57054651 JW |
1674 | * the zone ZONE_DIRTY and kswapd will start writing pages from |
1675 | * reclaim context. | |
b1a6f21e MG |
1676 | */ |
1677 | if (nr_unqueued_dirty == nr_taken) | |
57054651 | 1678 | set_bit(ZONE_DIRTY, &zone->flags); |
b1a6f21e MG |
1679 | |
1680 | /* | |
b738d764 LT |
1681 | * If kswapd scans pages marked marked for immediate |
1682 | * reclaim and under writeback (nr_immediate), it implies | |
1683 | * that pages are cycling through the LRU faster than | |
b1a6f21e MG |
1684 | * they are written so also forcibly stall. |
1685 | */ | |
b738d764 | 1686 | if (nr_immediate && current_may_throttle()) |
b1a6f21e | 1687 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
e2be15f6 | 1688 | } |
d43006d5 | 1689 | |
8e950282 MG |
1690 | /* |
1691 | * Stall direct reclaim for IO completions if underlying BDIs or zone | |
1692 | * is congested. Allow kswapd to continue until it starts encountering | |
1693 | * unqueued dirty pages or cycling through the LRU too quickly. | |
1694 | */ | |
399ba0b9 N |
1695 | if (!sc->hibernation_mode && !current_is_kswapd() && |
1696 | current_may_throttle()) | |
8e950282 MG |
1697 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); |
1698 | ||
ba5e9579 | 1699 | trace_mm_vmscan_lru_shrink_inactive(zone, nr_scanned, nr_reclaimed, |
1700 | sc->priority, file); | |
05ff5137 | 1701 | return nr_reclaimed; |
1da177e4 LT |
1702 | } |
1703 | ||
1704 | /* | |
1705 | * This moves pages from the active list to the inactive list. | |
1706 | * | |
1707 | * We move them the other way if the page is referenced by one or more | |
1708 | * processes, from rmap. | |
1709 | * | |
1710 | * If the pages are mostly unmapped, the processing is fast and it is | |
1711 | * appropriate to hold zone->lru_lock across the whole operation. But if | |
1712 | * the pages are mapped, the processing is slow (page_referenced()) so we | |
1713 | * should drop zone->lru_lock around each page. It's impossible to balance | |
1714 | * this, so instead we remove the pages from the LRU while processing them. | |
1715 | * It is safe to rely on PG_active against the non-LRU pages in here because | |
1716 | * nobody will play with that bit on a non-LRU page. | |
1717 | * | |
1718 | * The downside is that we have to touch page->_count against each page. | |
1719 | * But we had to alter page->flags anyway. | |
1720 | */ | |
1cfb419b | 1721 | |
fa9add64 | 1722 | static void move_active_pages_to_lru(struct lruvec *lruvec, |
3eb4140f | 1723 | struct list_head *list, |
2bcf8879 | 1724 | struct list_head *pages_to_free, |
3eb4140f WF |
1725 | enum lru_list lru) |
1726 | { | |
fa9add64 | 1727 | struct zone *zone = lruvec_zone(lruvec); |
3eb4140f | 1728 | unsigned long pgmoved = 0; |
3eb4140f | 1729 | struct page *page; |
fa9add64 | 1730 | int nr_pages; |
3eb4140f | 1731 | |
3eb4140f WF |
1732 | while (!list_empty(list)) { |
1733 | page = lru_to_page(list); | |
fa9add64 | 1734 | lruvec = mem_cgroup_page_lruvec(page, zone); |
3eb4140f | 1735 | |
309381fe | 1736 | VM_BUG_ON_PAGE(PageLRU(page), page); |
3eb4140f WF |
1737 | SetPageLRU(page); |
1738 | ||
fa9add64 HD |
1739 | nr_pages = hpage_nr_pages(page); |
1740 | mem_cgroup_update_lru_size(lruvec, lru, nr_pages); | |
925b7673 | 1741 | list_move(&page->lru, &lruvec->lists[lru]); |
fa9add64 | 1742 | pgmoved += nr_pages; |
3eb4140f | 1743 | |
2bcf8879 HD |
1744 | if (put_page_testzero(page)) { |
1745 | __ClearPageLRU(page); | |
1746 | __ClearPageActive(page); | |
fa9add64 | 1747 | del_page_from_lru_list(page, lruvec, lru); |
2bcf8879 HD |
1748 | |
1749 | if (unlikely(PageCompound(page))) { | |
1750 | spin_unlock_irq(&zone->lru_lock); | |
747db954 | 1751 | mem_cgroup_uncharge(page); |
2bcf8879 HD |
1752 | (*get_compound_page_dtor(page))(page); |
1753 | spin_lock_irq(&zone->lru_lock); | |
1754 | } else | |
1755 | list_add(&page->lru, pages_to_free); | |
3eb4140f WF |
1756 | } |
1757 | } | |
1758 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); | |
1759 | if (!is_active_lru(lru)) | |
1760 | __count_vm_events(PGDEACTIVATE, pgmoved); | |
1761 | } | |
1cfb419b | 1762 | |
f626012d | 1763 | static void shrink_active_list(unsigned long nr_to_scan, |
1a93be0e | 1764 | struct lruvec *lruvec, |
f16015fb | 1765 | struct scan_control *sc, |
9e3b2f8c | 1766 | enum lru_list lru) |
1da177e4 | 1767 | { |
44c241f1 | 1768 | unsigned long nr_taken; |
f626012d | 1769 | unsigned long nr_scanned; |
6fe6b7e3 | 1770 | unsigned long vm_flags; |
1da177e4 | 1771 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
8cab4754 | 1772 | LIST_HEAD(l_active); |
b69408e8 | 1773 | LIST_HEAD(l_inactive); |
1da177e4 | 1774 | struct page *page; |
1a93be0e | 1775 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
44c241f1 | 1776 | unsigned long nr_rotated = 0; |
f3fd4a61 | 1777 | isolate_mode_t isolate_mode = 0; |
3cb99451 | 1778 | int file = is_file_lru(lru); |
1a93be0e | 1779 | struct zone *zone = lruvec_zone(lruvec); |
1da177e4 LT |
1780 | |
1781 | lru_add_drain(); | |
f80c0673 MK |
1782 | |
1783 | if (!sc->may_unmap) | |
61317289 | 1784 | isolate_mode |= ISOLATE_UNMAPPED; |
f80c0673 | 1785 | if (!sc->may_writepage) |
61317289 | 1786 | isolate_mode |= ISOLATE_CLEAN; |
f80c0673 | 1787 | |
1da177e4 | 1788 | spin_lock_irq(&zone->lru_lock); |
925b7673 | 1789 | |
5dc35979 KK |
1790 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, |
1791 | &nr_scanned, sc, isolate_mode, lru); | |
89b5fae5 | 1792 | if (global_reclaim(sc)) |
0d5d823a | 1793 | __mod_zone_page_state(zone, NR_PAGES_SCANNED, nr_scanned); |
89b5fae5 | 1794 | |
b7c46d15 | 1795 | reclaim_stat->recent_scanned[file] += nr_taken; |
1cfb419b | 1796 | |
f626012d | 1797 | __count_zone_vm_events(PGREFILL, zone, nr_scanned); |
3cb99451 | 1798 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
a731286d | 1799 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
1da177e4 LT |
1800 | spin_unlock_irq(&zone->lru_lock); |
1801 | ||
1da177e4 LT |
1802 | while (!list_empty(&l_hold)) { |
1803 | cond_resched(); | |
1804 | page = lru_to_page(&l_hold); | |
1805 | list_del(&page->lru); | |
7e9cd484 | 1806 | |
39b5f29a | 1807 | if (unlikely(!page_evictable(page))) { |
894bc310 LS |
1808 | putback_lru_page(page); |
1809 | continue; | |
1810 | } | |
1811 | ||
cc715d99 MG |
1812 | if (unlikely(buffer_heads_over_limit)) { |
1813 | if (page_has_private(page) && trylock_page(page)) { | |
1814 | if (page_has_private(page)) | |
1815 | try_to_release_page(page, 0); | |
1816 | unlock_page(page); | |
1817 | } | |
1818 | } | |
1819 | ||
c3ac9a8a JW |
1820 | if (page_referenced(page, 0, sc->target_mem_cgroup, |
1821 | &vm_flags)) { | |
9992af10 | 1822 | nr_rotated += hpage_nr_pages(page); |
8cab4754 WF |
1823 | /* |
1824 | * Identify referenced, file-backed active pages and | |
1825 | * give them one more trip around the active list. So | |
1826 | * that executable code get better chances to stay in | |
1827 | * memory under moderate memory pressure. Anon pages | |
1828 | * are not likely to be evicted by use-once streaming | |
1829 | * IO, plus JVM can create lots of anon VM_EXEC pages, | |
1830 | * so we ignore them here. | |
1831 | */ | |
41e20983 | 1832 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
8cab4754 WF |
1833 | list_add(&page->lru, &l_active); |
1834 | continue; | |
1835 | } | |
1836 | } | |
7e9cd484 | 1837 | |
5205e56e | 1838 | ClearPageActive(page); /* we are de-activating */ |
1da177e4 LT |
1839 | list_add(&page->lru, &l_inactive); |
1840 | } | |
1841 | ||
b555749a | 1842 | /* |
8cab4754 | 1843 | * Move pages back to the lru list. |
b555749a | 1844 | */ |
2a1dc509 | 1845 | spin_lock_irq(&zone->lru_lock); |
556adecb | 1846 | /* |
8cab4754 WF |
1847 | * Count referenced pages from currently used mappings as rotated, |
1848 | * even though only some of them are actually re-activated. This | |
1849 | * helps balance scan pressure between file and anonymous pages in | |
7c0db9e9 | 1850 | * get_scan_count. |
7e9cd484 | 1851 | */ |
b7c46d15 | 1852 | reclaim_stat->recent_rotated[file] += nr_rotated; |
556adecb | 1853 | |
fa9add64 HD |
1854 | move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru); |
1855 | move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE); | |
a731286d | 1856 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
f8891e5e | 1857 | spin_unlock_irq(&zone->lru_lock); |
2bcf8879 | 1858 | |
747db954 | 1859 | mem_cgroup_uncharge_list(&l_hold); |
b745bc85 | 1860 | free_hot_cold_page_list(&l_hold, true); |
1da177e4 LT |
1861 | } |
1862 | ||
74e3f3c3 | 1863 | #ifdef CONFIG_SWAP |
42e2e457 | 1864 | static bool inactive_anon_is_low_global(struct zone *zone) |
f89eb90e KM |
1865 | { |
1866 | unsigned long active, inactive; | |
1867 | ||
1868 | active = zone_page_state(zone, NR_ACTIVE_ANON); | |
1869 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | |
1870 | ||
29d06bbb | 1871 | return inactive * zone->inactive_ratio < active; |
f89eb90e KM |
1872 | } |
1873 | ||
14797e23 KM |
1874 | /** |
1875 | * inactive_anon_is_low - check if anonymous pages need to be deactivated | |
c56d5c7d | 1876 | * @lruvec: LRU vector to check |
14797e23 KM |
1877 | * |
1878 | * Returns true if the zone does not have enough inactive anon pages, | |
1879 | * meaning some active anon pages need to be deactivated. | |
1880 | */ | |
42e2e457 | 1881 | static bool inactive_anon_is_low(struct lruvec *lruvec) |
14797e23 | 1882 | { |
74e3f3c3 MK |
1883 | /* |
1884 | * If we don't have swap space, anonymous page deactivation | |
1885 | * is pointless. | |
1886 | */ | |
1887 | if (!total_swap_pages) | |
42e2e457 | 1888 | return false; |
74e3f3c3 | 1889 | |
c3c787e8 | 1890 | if (!mem_cgroup_disabled()) |
c56d5c7d | 1891 | return mem_cgroup_inactive_anon_is_low(lruvec); |
f16015fb | 1892 | |
c56d5c7d | 1893 | return inactive_anon_is_low_global(lruvec_zone(lruvec)); |
14797e23 | 1894 | } |
74e3f3c3 | 1895 | #else |
42e2e457 | 1896 | static inline bool inactive_anon_is_low(struct lruvec *lruvec) |
74e3f3c3 | 1897 | { |
42e2e457 | 1898 | return false; |
74e3f3c3 MK |
1899 | } |
1900 | #endif | |
14797e23 | 1901 | |
56e49d21 RR |
1902 | /** |
1903 | * inactive_file_is_low - check if file pages need to be deactivated | |
c56d5c7d | 1904 | * @lruvec: LRU vector to check |
56e49d21 RR |
1905 | * |
1906 | * When the system is doing streaming IO, memory pressure here | |
1907 | * ensures that active file pages get deactivated, until more | |
1908 | * than half of the file pages are on the inactive list. | |
1909 | * | |
1910 | * Once we get to that situation, protect the system's working | |
1911 | * set from being evicted by disabling active file page aging. | |
1912 | * | |
1913 | * This uses a different ratio than the anonymous pages, because | |
1914 | * the page cache uses a use-once replacement algorithm. | |
1915 | */ | |
42e2e457 | 1916 | static bool inactive_file_is_low(struct lruvec *lruvec) |
56e49d21 | 1917 | { |
e3790144 JW |
1918 | unsigned long inactive; |
1919 | unsigned long active; | |
1920 | ||
23047a96 JW |
1921 | inactive = lruvec_lru_size(lruvec, LRU_INACTIVE_FILE); |
1922 | active = lruvec_lru_size(lruvec, LRU_ACTIVE_FILE); | |
56e49d21 | 1923 | |
e3790144 | 1924 | return active > inactive; |
56e49d21 RR |
1925 | } |
1926 | ||
42e2e457 | 1927 | static bool inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru) |
b39415b2 | 1928 | { |
75b00af7 | 1929 | if (is_file_lru(lru)) |
c56d5c7d | 1930 | return inactive_file_is_low(lruvec); |
b39415b2 | 1931 | else |
c56d5c7d | 1932 | return inactive_anon_is_low(lruvec); |
b39415b2 RR |
1933 | } |
1934 | ||
4f98a2fe | 1935 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
1a93be0e | 1936 | struct lruvec *lruvec, struct scan_control *sc) |
b69408e8 | 1937 | { |
b39415b2 | 1938 | if (is_active_lru(lru)) { |
75b00af7 | 1939 | if (inactive_list_is_low(lruvec, lru)) |
1a93be0e | 1940 | shrink_active_list(nr_to_scan, lruvec, sc, lru); |
556adecb RR |
1941 | return 0; |
1942 | } | |
1943 | ||
1a93be0e | 1944 | return shrink_inactive_list(nr_to_scan, lruvec, sc, lru); |
4f98a2fe RR |
1945 | } |
1946 | ||
9a265114 JW |
1947 | enum scan_balance { |
1948 | SCAN_EQUAL, | |
1949 | SCAN_FRACT, | |
1950 | SCAN_ANON, | |
1951 | SCAN_FILE, | |
1952 | }; | |
1953 | ||
4f98a2fe RR |
1954 | /* |
1955 | * Determine how aggressively the anon and file LRU lists should be | |
1956 | * scanned. The relative value of each set of LRU lists is determined | |
1957 | * by looking at the fraction of the pages scanned we did rotate back | |
1958 | * onto the active list instead of evict. | |
1959 | * | |
be7bd59d WL |
1960 | * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan |
1961 | * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan | |
4f98a2fe | 1962 | */ |
33377678 | 1963 | static void get_scan_count(struct lruvec *lruvec, struct mem_cgroup *memcg, |
6b4f7799 JW |
1964 | struct scan_control *sc, unsigned long *nr, |
1965 | unsigned long *lru_pages) | |
4f98a2fe | 1966 | { |
33377678 | 1967 | int swappiness = mem_cgroup_swappiness(memcg); |
9a265114 JW |
1968 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1969 | u64 fraction[2]; | |
1970 | u64 denominator = 0; /* gcc */ | |
1971 | struct zone *zone = lruvec_zone(lruvec); | |
4f98a2fe | 1972 | unsigned long anon_prio, file_prio; |
9a265114 | 1973 | enum scan_balance scan_balance; |
0bf1457f | 1974 | unsigned long anon, file; |
9a265114 | 1975 | bool force_scan = false; |
4f98a2fe | 1976 | unsigned long ap, fp; |
4111304d | 1977 | enum lru_list lru; |
6f04f48d SS |
1978 | bool some_scanned; |
1979 | int pass; | |
246e87a9 | 1980 | |
f11c0ca5 JW |
1981 | /* |
1982 | * If the zone or memcg is small, nr[l] can be 0. This | |
1983 | * results in no scanning on this priority and a potential | |
1984 | * priority drop. Global direct reclaim can go to the next | |
1985 | * zone and tends to have no problems. Global kswapd is for | |
1986 | * zone balancing and it needs to scan a minimum amount. When | |
1987 | * reclaiming for a memcg, a priority drop can cause high | |
1988 | * latencies, so it's better to scan a minimum amount there as | |
1989 | * well. | |
1990 | */ | |
90cbc250 VD |
1991 | if (current_is_kswapd()) { |
1992 | if (!zone_reclaimable(zone)) | |
1993 | force_scan = true; | |
eb01aaab | 1994 | if (!mem_cgroup_online(memcg)) |
90cbc250 VD |
1995 | force_scan = true; |
1996 | } | |
89b5fae5 | 1997 | if (!global_reclaim(sc)) |
a4d3e9e7 | 1998 | force_scan = true; |
76a33fc3 SL |
1999 | |
2000 | /* If we have no swap space, do not bother scanning anon pages. */ | |
d8b38438 | 2001 | if (!sc->may_swap || mem_cgroup_get_nr_swap_pages(memcg) <= 0) { |
9a265114 | 2002 | scan_balance = SCAN_FILE; |
76a33fc3 SL |
2003 | goto out; |
2004 | } | |
4f98a2fe | 2005 | |
10316b31 JW |
2006 | /* |
2007 | * Global reclaim will swap to prevent OOM even with no | |
2008 | * swappiness, but memcg users want to use this knob to | |
2009 | * disable swapping for individual groups completely when | |
2010 | * using the memory controller's swap limit feature would be | |
2011 | * too expensive. | |
2012 | */ | |
02695175 | 2013 | if (!global_reclaim(sc) && !swappiness) { |
9a265114 | 2014 | scan_balance = SCAN_FILE; |
10316b31 JW |
2015 | goto out; |
2016 | } | |
2017 | ||
2018 | /* | |
2019 | * Do not apply any pressure balancing cleverness when the | |
2020 | * system is close to OOM, scan both anon and file equally | |
2021 | * (unless the swappiness setting disagrees with swapping). | |
2022 | */ | |
02695175 | 2023 | if (!sc->priority && swappiness) { |
9a265114 | 2024 | scan_balance = SCAN_EQUAL; |
10316b31 JW |
2025 | goto out; |
2026 | } | |
2027 | ||
62376251 JW |
2028 | /* |
2029 | * Prevent the reclaimer from falling into the cache trap: as | |
2030 | * cache pages start out inactive, every cache fault will tip | |
2031 | * the scan balance towards the file LRU. And as the file LRU | |
2032 | * shrinks, so does the window for rotation from references. | |
2033 | * This means we have a runaway feedback loop where a tiny | |
2034 | * thrashing file LRU becomes infinitely more attractive than | |
2035 | * anon pages. Try to detect this based on file LRU size. | |
2036 | */ | |
2037 | if (global_reclaim(sc)) { | |
2ab051e1 JM |
2038 | unsigned long zonefile; |
2039 | unsigned long zonefree; | |
2040 | ||
2041 | zonefree = zone_page_state(zone, NR_FREE_PAGES); | |
2042 | zonefile = zone_page_state(zone, NR_ACTIVE_FILE) + | |
2043 | zone_page_state(zone, NR_INACTIVE_FILE); | |
62376251 | 2044 | |
2ab051e1 | 2045 | if (unlikely(zonefile + zonefree <= high_wmark_pages(zone))) { |
62376251 JW |
2046 | scan_balance = SCAN_ANON; |
2047 | goto out; | |
2048 | } | |
2049 | } | |
2050 | ||
7c5bd705 | 2051 | /* |
316bda0e VD |
2052 | * If there is enough inactive page cache, i.e. if the size of the |
2053 | * inactive list is greater than that of the active list *and* the | |
2054 | * inactive list actually has some pages to scan on this priority, we | |
2055 | * do not reclaim anything from the anonymous working set right now. | |
2056 | * Without the second condition we could end up never scanning an | |
2057 | * lruvec even if it has plenty of old anonymous pages unless the | |
2058 | * system is under heavy pressure. | |
7c5bd705 | 2059 | */ |
316bda0e | 2060 | if (!inactive_file_is_low(lruvec) && |
23047a96 | 2061 | lruvec_lru_size(lruvec, LRU_INACTIVE_FILE) >> sc->priority) { |
9a265114 | 2062 | scan_balance = SCAN_FILE; |
7c5bd705 JW |
2063 | goto out; |
2064 | } | |
2065 | ||
9a265114 JW |
2066 | scan_balance = SCAN_FRACT; |
2067 | ||
58c37f6e KM |
2068 | /* |
2069 | * With swappiness at 100, anonymous and file have the same priority. | |
2070 | * This scanning priority is essentially the inverse of IO cost. | |
2071 | */ | |
02695175 | 2072 | anon_prio = swappiness; |
75b00af7 | 2073 | file_prio = 200 - anon_prio; |
58c37f6e | 2074 | |
4f98a2fe RR |
2075 | /* |
2076 | * OK, so we have swap space and a fair amount of page cache | |
2077 | * pages. We use the recently rotated / recently scanned | |
2078 | * ratios to determine how valuable each cache is. | |
2079 | * | |
2080 | * Because workloads change over time (and to avoid overflow) | |
2081 | * we keep these statistics as a floating average, which ends | |
2082 | * up weighing recent references more than old ones. | |
2083 | * | |
2084 | * anon in [0], file in [1] | |
2085 | */ | |
2ab051e1 | 2086 | |
23047a96 JW |
2087 | anon = lruvec_lru_size(lruvec, LRU_ACTIVE_ANON) + |
2088 | lruvec_lru_size(lruvec, LRU_INACTIVE_ANON); | |
2089 | file = lruvec_lru_size(lruvec, LRU_ACTIVE_FILE) + | |
2090 | lruvec_lru_size(lruvec, LRU_INACTIVE_FILE); | |
2ab051e1 | 2091 | |
90126375 | 2092 | spin_lock_irq(&zone->lru_lock); |
6e901571 | 2093 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
6e901571 KM |
2094 | reclaim_stat->recent_scanned[0] /= 2; |
2095 | reclaim_stat->recent_rotated[0] /= 2; | |
4f98a2fe RR |
2096 | } |
2097 | ||
6e901571 | 2098 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
6e901571 KM |
2099 | reclaim_stat->recent_scanned[1] /= 2; |
2100 | reclaim_stat->recent_rotated[1] /= 2; | |
4f98a2fe RR |
2101 | } |
2102 | ||
4f98a2fe | 2103 | /* |
00d8089c RR |
2104 | * The amount of pressure on anon vs file pages is inversely |
2105 | * proportional to the fraction of recently scanned pages on | |
2106 | * each list that were recently referenced and in active use. | |
4f98a2fe | 2107 | */ |
fe35004f | 2108 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); |
6e901571 | 2109 | ap /= reclaim_stat->recent_rotated[0] + 1; |
4f98a2fe | 2110 | |
fe35004f | 2111 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); |
6e901571 | 2112 | fp /= reclaim_stat->recent_rotated[1] + 1; |
90126375 | 2113 | spin_unlock_irq(&zone->lru_lock); |
4f98a2fe | 2114 | |
76a33fc3 SL |
2115 | fraction[0] = ap; |
2116 | fraction[1] = fp; | |
2117 | denominator = ap + fp + 1; | |
2118 | out: | |
6f04f48d SS |
2119 | some_scanned = false; |
2120 | /* Only use force_scan on second pass. */ | |
2121 | for (pass = 0; !some_scanned && pass < 2; pass++) { | |
6b4f7799 | 2122 | *lru_pages = 0; |
6f04f48d SS |
2123 | for_each_evictable_lru(lru) { |
2124 | int file = is_file_lru(lru); | |
2125 | unsigned long size; | |
2126 | unsigned long scan; | |
6e08a369 | 2127 | |
23047a96 | 2128 | size = lruvec_lru_size(lruvec, lru); |
6f04f48d | 2129 | scan = size >> sc->priority; |
9a265114 | 2130 | |
6f04f48d SS |
2131 | if (!scan && pass && force_scan) |
2132 | scan = min(size, SWAP_CLUSTER_MAX); | |
9a265114 | 2133 | |
6f04f48d SS |
2134 | switch (scan_balance) { |
2135 | case SCAN_EQUAL: | |
2136 | /* Scan lists relative to size */ | |
2137 | break; | |
2138 | case SCAN_FRACT: | |
2139 | /* | |
2140 | * Scan types proportional to swappiness and | |
2141 | * their relative recent reclaim efficiency. | |
2142 | */ | |
2143 | scan = div64_u64(scan * fraction[file], | |
2144 | denominator); | |
2145 | break; | |
2146 | case SCAN_FILE: | |
2147 | case SCAN_ANON: | |
2148 | /* Scan one type exclusively */ | |
6b4f7799 JW |
2149 | if ((scan_balance == SCAN_FILE) != file) { |
2150 | size = 0; | |
6f04f48d | 2151 | scan = 0; |
6b4f7799 | 2152 | } |
6f04f48d SS |
2153 | break; |
2154 | default: | |
2155 | /* Look ma, no brain */ | |
2156 | BUG(); | |
2157 | } | |
6b4f7799 JW |
2158 | |
2159 | *lru_pages += size; | |
6f04f48d | 2160 | nr[lru] = scan; |
6b4f7799 | 2161 | |
9a265114 | 2162 | /* |
6f04f48d SS |
2163 | * Skip the second pass and don't force_scan, |
2164 | * if we found something to scan. | |
9a265114 | 2165 | */ |
6f04f48d | 2166 | some_scanned |= !!scan; |
9a265114 | 2167 | } |
76a33fc3 | 2168 | } |
6e08a369 | 2169 | } |
4f98a2fe | 2170 | |
72b252ae MG |
2171 | #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH |
2172 | static void init_tlb_ubc(void) | |
2173 | { | |
2174 | /* | |
2175 | * This deliberately does not clear the cpumask as it's expensive | |
2176 | * and unnecessary. If there happens to be data in there then the | |
2177 | * first SWAP_CLUSTER_MAX pages will send an unnecessary IPI and | |
2178 | * then will be cleared. | |
2179 | */ | |
2180 | current->tlb_ubc.flush_required = false; | |
2181 | } | |
2182 | #else | |
2183 | static inline void init_tlb_ubc(void) | |
2184 | { | |
2185 | } | |
2186 | #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */ | |
2187 | ||
9b4f98cd JW |
2188 | /* |
2189 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. | |
2190 | */ | |
33377678 VD |
2191 | static void shrink_zone_memcg(struct zone *zone, struct mem_cgroup *memcg, |
2192 | struct scan_control *sc, unsigned long *lru_pages) | |
9b4f98cd | 2193 | { |
33377678 | 2194 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
9b4f98cd | 2195 | unsigned long nr[NR_LRU_LISTS]; |
e82e0561 | 2196 | unsigned long targets[NR_LRU_LISTS]; |
9b4f98cd JW |
2197 | unsigned long nr_to_scan; |
2198 | enum lru_list lru; | |
2199 | unsigned long nr_reclaimed = 0; | |
2200 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; | |
2201 | struct blk_plug plug; | |
1a501907 | 2202 | bool scan_adjusted; |
9b4f98cd | 2203 | |
33377678 | 2204 | get_scan_count(lruvec, memcg, sc, nr, lru_pages); |
9b4f98cd | 2205 | |
e82e0561 MG |
2206 | /* Record the original scan target for proportional adjustments later */ |
2207 | memcpy(targets, nr, sizeof(nr)); | |
2208 | ||
1a501907 MG |
2209 | /* |
2210 | * Global reclaiming within direct reclaim at DEF_PRIORITY is a normal | |
2211 | * event that can occur when there is little memory pressure e.g. | |
2212 | * multiple streaming readers/writers. Hence, we do not abort scanning | |
2213 | * when the requested number of pages are reclaimed when scanning at | |
2214 | * DEF_PRIORITY on the assumption that the fact we are direct | |
2215 | * reclaiming implies that kswapd is not keeping up and it is best to | |
2216 | * do a batch of work at once. For memcg reclaim one check is made to | |
2217 | * abort proportional reclaim if either the file or anon lru has already | |
2218 | * dropped to zero at the first pass. | |
2219 | */ | |
2220 | scan_adjusted = (global_reclaim(sc) && !current_is_kswapd() && | |
2221 | sc->priority == DEF_PRIORITY); | |
2222 | ||
72b252ae MG |
2223 | init_tlb_ubc(); |
2224 | ||
9b4f98cd JW |
2225 | blk_start_plug(&plug); |
2226 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || | |
2227 | nr[LRU_INACTIVE_FILE]) { | |
e82e0561 MG |
2228 | unsigned long nr_anon, nr_file, percentage; |
2229 | unsigned long nr_scanned; | |
2230 | ||
9b4f98cd JW |
2231 | for_each_evictable_lru(lru) { |
2232 | if (nr[lru]) { | |
2233 | nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX); | |
2234 | nr[lru] -= nr_to_scan; | |
2235 | ||
2236 | nr_reclaimed += shrink_list(lru, nr_to_scan, | |
2237 | lruvec, sc); | |
2238 | } | |
2239 | } | |
e82e0561 MG |
2240 | |
2241 | if (nr_reclaimed < nr_to_reclaim || scan_adjusted) | |
2242 | continue; | |
2243 | ||
e82e0561 MG |
2244 | /* |
2245 | * For kswapd and memcg, reclaim at least the number of pages | |
1a501907 | 2246 | * requested. Ensure that the anon and file LRUs are scanned |
e82e0561 MG |
2247 | * proportionally what was requested by get_scan_count(). We |
2248 | * stop reclaiming one LRU and reduce the amount scanning | |
2249 | * proportional to the original scan target. | |
2250 | */ | |
2251 | nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE]; | |
2252 | nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON]; | |
2253 | ||
1a501907 MG |
2254 | /* |
2255 | * It's just vindictive to attack the larger once the smaller | |
2256 | * has gone to zero. And given the way we stop scanning the | |
2257 | * smaller below, this makes sure that we only make one nudge | |
2258 | * towards proportionality once we've got nr_to_reclaim. | |
2259 | */ | |
2260 | if (!nr_file || !nr_anon) | |
2261 | break; | |
2262 | ||
e82e0561 MG |
2263 | if (nr_file > nr_anon) { |
2264 | unsigned long scan_target = targets[LRU_INACTIVE_ANON] + | |
2265 | targets[LRU_ACTIVE_ANON] + 1; | |
2266 | lru = LRU_BASE; | |
2267 | percentage = nr_anon * 100 / scan_target; | |
2268 | } else { | |
2269 | unsigned long scan_target = targets[LRU_INACTIVE_FILE] + | |
2270 | targets[LRU_ACTIVE_FILE] + 1; | |
2271 | lru = LRU_FILE; | |
2272 | percentage = nr_file * 100 / scan_target; | |
2273 | } | |
2274 | ||
2275 | /* Stop scanning the smaller of the LRU */ | |
2276 | nr[lru] = 0; | |
2277 | nr[lru + LRU_ACTIVE] = 0; | |
2278 | ||
2279 | /* | |
2280 | * Recalculate the other LRU scan count based on its original | |
2281 | * scan target and the percentage scanning already complete | |
2282 | */ | |
2283 | lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE; | |
2284 | nr_scanned = targets[lru] - nr[lru]; | |
2285 | nr[lru] = targets[lru] * (100 - percentage) / 100; | |
2286 | nr[lru] -= min(nr[lru], nr_scanned); | |
2287 | ||
2288 | lru += LRU_ACTIVE; | |
2289 | nr_scanned = targets[lru] - nr[lru]; | |
2290 | nr[lru] = targets[lru] * (100 - percentage) / 100; | |
2291 | nr[lru] -= min(nr[lru], nr_scanned); | |
2292 | ||
2293 | scan_adjusted = true; | |
9b4f98cd JW |
2294 | } |
2295 | blk_finish_plug(&plug); | |
2296 | sc->nr_reclaimed += nr_reclaimed; | |
2297 | ||
2298 | /* | |
2299 | * Even if we did not try to evict anon pages at all, we want to | |
2300 | * rebalance the anon lru active/inactive ratio. | |
2301 | */ | |
2302 | if (inactive_anon_is_low(lruvec)) | |
2303 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, | |
2304 | sc, LRU_ACTIVE_ANON); | |
2305 | ||
2306 | throttle_vm_writeout(sc->gfp_mask); | |
2307 | } | |
2308 | ||
23b9da55 | 2309 | /* Use reclaim/compaction for costly allocs or under memory pressure */ |
9e3b2f8c | 2310 | static bool in_reclaim_compaction(struct scan_control *sc) |
23b9da55 | 2311 | { |
d84da3f9 | 2312 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && |
23b9da55 | 2313 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || |
9e3b2f8c | 2314 | sc->priority < DEF_PRIORITY - 2)) |
23b9da55 MG |
2315 | return true; |
2316 | ||
2317 | return false; | |
2318 | } | |
2319 | ||
3e7d3449 | 2320 | /* |
23b9da55 MG |
2321 | * Reclaim/compaction is used for high-order allocation requests. It reclaims |
2322 | * order-0 pages before compacting the zone. should_continue_reclaim() returns | |
2323 | * true if more pages should be reclaimed such that when the page allocator | |
2324 | * calls try_to_compact_zone() that it will have enough free pages to succeed. | |
2325 | * It will give up earlier than that if there is difficulty reclaiming pages. | |
3e7d3449 | 2326 | */ |
9b4f98cd | 2327 | static inline bool should_continue_reclaim(struct zone *zone, |
3e7d3449 MG |
2328 | unsigned long nr_reclaimed, |
2329 | unsigned long nr_scanned, | |
2330 | struct scan_control *sc) | |
2331 | { | |
2332 | unsigned long pages_for_compaction; | |
2333 | unsigned long inactive_lru_pages; | |
2334 | ||
2335 | /* If not in reclaim/compaction mode, stop */ | |
9e3b2f8c | 2336 | if (!in_reclaim_compaction(sc)) |
3e7d3449 MG |
2337 | return false; |
2338 | ||
2876592f MG |
2339 | /* Consider stopping depending on scan and reclaim activity */ |
2340 | if (sc->gfp_mask & __GFP_REPEAT) { | |
2341 | /* | |
2342 | * For __GFP_REPEAT allocations, stop reclaiming if the | |
2343 | * full LRU list has been scanned and we are still failing | |
2344 | * to reclaim pages. This full LRU scan is potentially | |
2345 | * expensive but a __GFP_REPEAT caller really wants to succeed | |
2346 | */ | |
2347 | if (!nr_reclaimed && !nr_scanned) | |
2348 | return false; | |
2349 | } else { | |
2350 | /* | |
2351 | * For non-__GFP_REPEAT allocations which can presumably | |
2352 | * fail without consequence, stop if we failed to reclaim | |
2353 | * any pages from the last SWAP_CLUSTER_MAX number of | |
2354 | * pages that were scanned. This will return to the | |
2355 | * caller faster at the risk reclaim/compaction and | |
2356 | * the resulting allocation attempt fails | |
2357 | */ | |
2358 | if (!nr_reclaimed) | |
2359 | return false; | |
2360 | } | |
3e7d3449 MG |
2361 | |
2362 | /* | |
2363 | * If we have not reclaimed enough pages for compaction and the | |
2364 | * inactive lists are large enough, continue reclaiming | |
2365 | */ | |
2366 | pages_for_compaction = (2UL << sc->order); | |
9b4f98cd | 2367 | inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE); |
ec8acf20 | 2368 | if (get_nr_swap_pages() > 0) |
9b4f98cd | 2369 | inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON); |
3e7d3449 MG |
2370 | if (sc->nr_reclaimed < pages_for_compaction && |
2371 | inactive_lru_pages > pages_for_compaction) | |
2372 | return true; | |
2373 | ||
2374 | /* If compaction would go ahead or the allocation would succeed, stop */ | |
ebff3980 | 2375 | switch (compaction_suitable(zone, sc->order, 0, 0)) { |
3e7d3449 MG |
2376 | case COMPACT_PARTIAL: |
2377 | case COMPACT_CONTINUE: | |
2378 | return false; | |
2379 | default: | |
2380 | return true; | |
2381 | } | |
2382 | } | |
2383 | ||
6b4f7799 JW |
2384 | static bool shrink_zone(struct zone *zone, struct scan_control *sc, |
2385 | bool is_classzone) | |
1da177e4 | 2386 | { |
cb731d6c | 2387 | struct reclaim_state *reclaim_state = current->reclaim_state; |
f0fdc5e8 | 2388 | unsigned long nr_reclaimed, nr_scanned; |
2344d7e4 | 2389 | bool reclaimable = false; |
1da177e4 | 2390 | |
9b4f98cd JW |
2391 | do { |
2392 | struct mem_cgroup *root = sc->target_mem_cgroup; | |
2393 | struct mem_cgroup_reclaim_cookie reclaim = { | |
2394 | .zone = zone, | |
2395 | .priority = sc->priority, | |
2396 | }; | |
6b4f7799 | 2397 | unsigned long zone_lru_pages = 0; |
694fbc0f | 2398 | struct mem_cgroup *memcg; |
3e7d3449 | 2399 | |
9b4f98cd JW |
2400 | nr_reclaimed = sc->nr_reclaimed; |
2401 | nr_scanned = sc->nr_scanned; | |
1da177e4 | 2402 | |
694fbc0f AM |
2403 | memcg = mem_cgroup_iter(root, NULL, &reclaim); |
2404 | do { | |
6b4f7799 | 2405 | unsigned long lru_pages; |
8e8ae645 | 2406 | unsigned long reclaimed; |
cb731d6c | 2407 | unsigned long scanned; |
5660048c | 2408 | |
241994ed JW |
2409 | if (mem_cgroup_low(root, memcg)) { |
2410 | if (!sc->may_thrash) | |
2411 | continue; | |
2412 | mem_cgroup_events(memcg, MEMCG_LOW, 1); | |
2413 | } | |
2414 | ||
8e8ae645 | 2415 | reclaimed = sc->nr_reclaimed; |
cb731d6c | 2416 | scanned = sc->nr_scanned; |
f9be23d6 | 2417 | |
33377678 | 2418 | shrink_zone_memcg(zone, memcg, sc, &lru_pages); |
6b4f7799 | 2419 | zone_lru_pages += lru_pages; |
f16015fb | 2420 | |
cb731d6c VD |
2421 | if (memcg && is_classzone) |
2422 | shrink_slab(sc->gfp_mask, zone_to_nid(zone), | |
2423 | memcg, sc->nr_scanned - scanned, | |
2424 | lru_pages); | |
2425 | ||
8e8ae645 JW |
2426 | /* Record the group's reclaim efficiency */ |
2427 | vmpressure(sc->gfp_mask, memcg, false, | |
2428 | sc->nr_scanned - scanned, | |
2429 | sc->nr_reclaimed - reclaimed); | |
2430 | ||
9b4f98cd | 2431 | /* |
a394cb8e MH |
2432 | * Direct reclaim and kswapd have to scan all memory |
2433 | * cgroups to fulfill the overall scan target for the | |
9b4f98cd | 2434 | * zone. |
a394cb8e MH |
2435 | * |
2436 | * Limit reclaim, on the other hand, only cares about | |
2437 | * nr_to_reclaim pages to be reclaimed and it will | |
2438 | * retry with decreasing priority if one round over the | |
2439 | * whole hierarchy is not sufficient. | |
9b4f98cd | 2440 | */ |
a394cb8e MH |
2441 | if (!global_reclaim(sc) && |
2442 | sc->nr_reclaimed >= sc->nr_to_reclaim) { | |
9b4f98cd JW |
2443 | mem_cgroup_iter_break(root, memcg); |
2444 | break; | |
2445 | } | |
241994ed | 2446 | } while ((memcg = mem_cgroup_iter(root, memcg, &reclaim))); |
70ddf637 | 2447 | |
6b4f7799 JW |
2448 | /* |
2449 | * Shrink the slab caches in the same proportion that | |
2450 | * the eligible LRU pages were scanned. | |
2451 | */ | |
cb731d6c VD |
2452 | if (global_reclaim(sc) && is_classzone) |
2453 | shrink_slab(sc->gfp_mask, zone_to_nid(zone), NULL, | |
2454 | sc->nr_scanned - nr_scanned, | |
2455 | zone_lru_pages); | |
2456 | ||
2457 | if (reclaim_state) { | |
2458 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; | |
2459 | reclaim_state->reclaimed_slab = 0; | |
6b4f7799 JW |
2460 | } |
2461 | ||
8e8ae645 JW |
2462 | /* Record the subtree's reclaim efficiency */ |
2463 | vmpressure(sc->gfp_mask, sc->target_mem_cgroup, true, | |
70ddf637 AV |
2464 | sc->nr_scanned - nr_scanned, |
2465 | sc->nr_reclaimed - nr_reclaimed); | |
2466 | ||
2344d7e4 JW |
2467 | if (sc->nr_reclaimed - nr_reclaimed) |
2468 | reclaimable = true; | |
2469 | ||
9b4f98cd JW |
2470 | } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed, |
2471 | sc->nr_scanned - nr_scanned, sc)); | |
2344d7e4 JW |
2472 | |
2473 | return reclaimable; | |
f16015fb JW |
2474 | } |
2475 | ||
53853e2d VB |
2476 | /* |
2477 | * Returns true if compaction should go ahead for a high-order request, or | |
2478 | * the high-order allocation would succeed without compaction. | |
2479 | */ | |
0b06496a | 2480 | static inline bool compaction_ready(struct zone *zone, int order) |
fe4b1b24 MG |
2481 | { |
2482 | unsigned long balance_gap, watermark; | |
2483 | bool watermark_ok; | |
2484 | ||
fe4b1b24 MG |
2485 | /* |
2486 | * Compaction takes time to run and there are potentially other | |
2487 | * callers using the pages just freed. Continue reclaiming until | |
2488 | * there is a buffer of free pages available to give compaction | |
2489 | * a reasonable chance of completing and allocating the page | |
2490 | */ | |
4be89a34 JZ |
2491 | balance_gap = min(low_wmark_pages(zone), DIV_ROUND_UP( |
2492 | zone->managed_pages, KSWAPD_ZONE_BALANCE_GAP_RATIO)); | |
0b06496a | 2493 | watermark = high_wmark_pages(zone) + balance_gap + (2UL << order); |
e2b19197 | 2494 | watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0); |
fe4b1b24 MG |
2495 | |
2496 | /* | |
2497 | * If compaction is deferred, reclaim up to a point where | |
2498 | * compaction will have a chance of success when re-enabled | |
2499 | */ | |
0b06496a | 2500 | if (compaction_deferred(zone, order)) |
fe4b1b24 MG |
2501 | return watermark_ok; |
2502 | ||
53853e2d VB |
2503 | /* |
2504 | * If compaction is not ready to start and allocation is not likely | |
2505 | * to succeed without it, then keep reclaiming. | |
2506 | */ | |
ebff3980 | 2507 | if (compaction_suitable(zone, order, 0, 0) == COMPACT_SKIPPED) |
fe4b1b24 MG |
2508 | return false; |
2509 | ||
2510 | return watermark_ok; | |
2511 | } | |
2512 | ||
1da177e4 LT |
2513 | /* |
2514 | * This is the direct reclaim path, for page-allocating processes. We only | |
2515 | * try to reclaim pages from zones which will satisfy the caller's allocation | |
2516 | * request. | |
2517 | * | |
41858966 MG |
2518 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). |
2519 | * Because: | |
1da177e4 LT |
2520 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
2521 | * allocation or | |
41858966 MG |
2522 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones |
2523 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' | |
2524 | * zone defense algorithm. | |
1da177e4 | 2525 | * |
1da177e4 LT |
2526 | * If a zone is deemed to be full of pinned pages then just give it a light |
2527 | * scan then give up on it. | |
2344d7e4 JW |
2528 | * |
2529 | * Returns true if a zone was reclaimable. | |
1da177e4 | 2530 | */ |
2344d7e4 | 2531 | static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc) |
1da177e4 | 2532 | { |
dd1a239f | 2533 | struct zoneref *z; |
54a6eb5c | 2534 | struct zone *zone; |
0608f43d AM |
2535 | unsigned long nr_soft_reclaimed; |
2536 | unsigned long nr_soft_scanned; | |
619d0d76 | 2537 | gfp_t orig_mask; |
9bbc04ee | 2538 | enum zone_type requested_highidx = gfp_zone(sc->gfp_mask); |
2344d7e4 | 2539 | bool reclaimable = false; |
1cfb419b | 2540 | |
cc715d99 MG |
2541 | /* |
2542 | * If the number of buffer_heads in the machine exceeds the maximum | |
2543 | * allowed level, force direct reclaim to scan the highmem zone as | |
2544 | * highmem pages could be pinning lowmem pages storing buffer_heads | |
2545 | */ | |
619d0d76 | 2546 | orig_mask = sc->gfp_mask; |
cc715d99 MG |
2547 | if (buffer_heads_over_limit) |
2548 | sc->gfp_mask |= __GFP_HIGHMEM; | |
2549 | ||
d4debc66 | 2550 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
6b4f7799 JW |
2551 | requested_highidx, sc->nodemask) { |
2552 | enum zone_type classzone_idx; | |
2553 | ||
f3fe6512 | 2554 | if (!populated_zone(zone)) |
1da177e4 | 2555 | continue; |
6b4f7799 JW |
2556 | |
2557 | classzone_idx = requested_highidx; | |
2558 | while (!populated_zone(zone->zone_pgdat->node_zones + | |
2559 | classzone_idx)) | |
2560 | classzone_idx--; | |
2561 | ||
1cfb419b KH |
2562 | /* |
2563 | * Take care memory controller reclaiming has small influence | |
2564 | * to global LRU. | |
2565 | */ | |
89b5fae5 | 2566 | if (global_reclaim(sc)) { |
344736f2 VD |
2567 | if (!cpuset_zone_allowed(zone, |
2568 | GFP_KERNEL | __GFP_HARDWALL)) | |
1cfb419b | 2569 | continue; |
65ec02cb | 2570 | |
6e543d57 LD |
2571 | if (sc->priority != DEF_PRIORITY && |
2572 | !zone_reclaimable(zone)) | |
1cfb419b | 2573 | continue; /* Let kswapd poll it */ |
0b06496a JW |
2574 | |
2575 | /* | |
2576 | * If we already have plenty of memory free for | |
2577 | * compaction in this zone, don't free any more. | |
2578 | * Even though compaction is invoked for any | |
2579 | * non-zero order, only frequent costly order | |
2580 | * reclamation is disruptive enough to become a | |
2581 | * noticeable problem, like transparent huge | |
2582 | * page allocations. | |
2583 | */ | |
2584 | if (IS_ENABLED(CONFIG_COMPACTION) && | |
2585 | sc->order > PAGE_ALLOC_COSTLY_ORDER && | |
2586 | zonelist_zone_idx(z) <= requested_highidx && | |
2587 | compaction_ready(zone, sc->order)) { | |
2588 | sc->compaction_ready = true; | |
2589 | continue; | |
e0887c19 | 2590 | } |
0b06496a | 2591 | |
0608f43d AM |
2592 | /* |
2593 | * This steals pages from memory cgroups over softlimit | |
2594 | * and returns the number of reclaimed pages and | |
2595 | * scanned pages. This works for global memory pressure | |
2596 | * and balancing, not for a memcg's limit. | |
2597 | */ | |
2598 | nr_soft_scanned = 0; | |
2599 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | |
2600 | sc->order, sc->gfp_mask, | |
2601 | &nr_soft_scanned); | |
2602 | sc->nr_reclaimed += nr_soft_reclaimed; | |
2603 | sc->nr_scanned += nr_soft_scanned; | |
2344d7e4 JW |
2604 | if (nr_soft_reclaimed) |
2605 | reclaimable = true; | |
ac34a1a3 | 2606 | /* need some check for avoid more shrink_zone() */ |
1cfb419b | 2607 | } |
408d8544 | 2608 | |
6b4f7799 | 2609 | if (shrink_zone(zone, sc, zone_idx(zone) == classzone_idx)) |
2344d7e4 JW |
2610 | reclaimable = true; |
2611 | ||
2612 | if (global_reclaim(sc) && | |
2613 | !reclaimable && zone_reclaimable(zone)) | |
2614 | reclaimable = true; | |
1da177e4 | 2615 | } |
e0c23279 | 2616 | |
619d0d76 WY |
2617 | /* |
2618 | * Restore to original mask to avoid the impact on the caller if we | |
2619 | * promoted it to __GFP_HIGHMEM. | |
2620 | */ | |
2621 | sc->gfp_mask = orig_mask; | |
d1908362 | 2622 | |
2344d7e4 | 2623 | return reclaimable; |
1da177e4 | 2624 | } |
4f98a2fe | 2625 | |
1da177e4 LT |
2626 | /* |
2627 | * This is the main entry point to direct page reclaim. | |
2628 | * | |
2629 | * If a full scan of the inactive list fails to free enough memory then we | |
2630 | * are "out of memory" and something needs to be killed. | |
2631 | * | |
2632 | * If the caller is !__GFP_FS then the probability of a failure is reasonably | |
2633 | * high - the zone may be full of dirty or under-writeback pages, which this | |
5b0830cb JA |
2634 | * caller can't do much about. We kick the writeback threads and take explicit |
2635 | * naps in the hope that some of these pages can be written. But if the | |
2636 | * allocating task holds filesystem locks which prevent writeout this might not | |
2637 | * work, and the allocation attempt will fail. | |
a41f24ea NA |
2638 | * |
2639 | * returns: 0, if no pages reclaimed | |
2640 | * else, the number of pages reclaimed | |
1da177e4 | 2641 | */ |
dac1d27b | 2642 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
3115cd91 | 2643 | struct scan_control *sc) |
1da177e4 | 2644 | { |
241994ed | 2645 | int initial_priority = sc->priority; |
69e05944 | 2646 | unsigned long total_scanned = 0; |
22fba335 | 2647 | unsigned long writeback_threshold; |
2344d7e4 | 2648 | bool zones_reclaimable; |
241994ed | 2649 | retry: |
873b4771 KK |
2650 | delayacct_freepages_start(); |
2651 | ||
89b5fae5 | 2652 | if (global_reclaim(sc)) |
1cfb419b | 2653 | count_vm_event(ALLOCSTALL); |
1da177e4 | 2654 | |
9e3b2f8c | 2655 | do { |
70ddf637 AV |
2656 | vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup, |
2657 | sc->priority); | |
66e1707b | 2658 | sc->nr_scanned = 0; |
2344d7e4 | 2659 | zones_reclaimable = shrink_zones(zonelist, sc); |
c6a8a8c5 | 2660 | |
66e1707b | 2661 | total_scanned += sc->nr_scanned; |
bb21c7ce | 2662 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
0b06496a JW |
2663 | break; |
2664 | ||
2665 | if (sc->compaction_ready) | |
2666 | break; | |
1da177e4 | 2667 | |
0e50ce3b MK |
2668 | /* |
2669 | * If we're getting trouble reclaiming, start doing | |
2670 | * writepage even in laptop mode. | |
2671 | */ | |
2672 | if (sc->priority < DEF_PRIORITY - 2) | |
2673 | sc->may_writepage = 1; | |
2674 | ||
1da177e4 LT |
2675 | /* |
2676 | * Try to write back as many pages as we just scanned. This | |
2677 | * tends to cause slow streaming writers to write data to the | |
2678 | * disk smoothly, at the dirtying rate, which is nice. But | |
2679 | * that's undesirable in laptop mode, where we *want* lumpy | |
2680 | * writeout. So in laptop mode, write out the whole world. | |
2681 | */ | |
22fba335 KM |
2682 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; |
2683 | if (total_scanned > writeback_threshold) { | |
0e175a18 CW |
2684 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, |
2685 | WB_REASON_TRY_TO_FREE_PAGES); | |
66e1707b | 2686 | sc->may_writepage = 1; |
1da177e4 | 2687 | } |
0b06496a | 2688 | } while (--sc->priority >= 0); |
bb21c7ce | 2689 | |
873b4771 KK |
2690 | delayacct_freepages_end(); |
2691 | ||
bb21c7ce KM |
2692 | if (sc->nr_reclaimed) |
2693 | return sc->nr_reclaimed; | |
2694 | ||
0cee34fd | 2695 | /* Aborted reclaim to try compaction? don't OOM, then */ |
0b06496a | 2696 | if (sc->compaction_ready) |
7335084d MG |
2697 | return 1; |
2698 | ||
241994ed JW |
2699 | /* Untapped cgroup reserves? Don't OOM, retry. */ |
2700 | if (!sc->may_thrash) { | |
2701 | sc->priority = initial_priority; | |
2702 | sc->may_thrash = 1; | |
2703 | goto retry; | |
2704 | } | |
2705 | ||
2344d7e4 JW |
2706 | /* Any of the zones still reclaimable? Don't OOM. */ |
2707 | if (zones_reclaimable) | |
bb21c7ce KM |
2708 | return 1; |
2709 | ||
2710 | return 0; | |
1da177e4 LT |
2711 | } |
2712 | ||
5515061d MG |
2713 | static bool pfmemalloc_watermark_ok(pg_data_t *pgdat) |
2714 | { | |
2715 | struct zone *zone; | |
2716 | unsigned long pfmemalloc_reserve = 0; | |
2717 | unsigned long free_pages = 0; | |
2718 | int i; | |
2719 | bool wmark_ok; | |
2720 | ||
2721 | for (i = 0; i <= ZONE_NORMAL; i++) { | |
2722 | zone = &pgdat->node_zones[i]; | |
f012a84a NA |
2723 | if (!populated_zone(zone) || |
2724 | zone_reclaimable_pages(zone) == 0) | |
675becce MG |
2725 | continue; |
2726 | ||
5515061d MG |
2727 | pfmemalloc_reserve += min_wmark_pages(zone); |
2728 | free_pages += zone_page_state(zone, NR_FREE_PAGES); | |
2729 | } | |
2730 | ||
675becce MG |
2731 | /* If there are no reserves (unexpected config) then do not throttle */ |
2732 | if (!pfmemalloc_reserve) | |
2733 | return true; | |
2734 | ||
5515061d MG |
2735 | wmark_ok = free_pages > pfmemalloc_reserve / 2; |
2736 | ||
2737 | /* kswapd must be awake if processes are being throttled */ | |
2738 | if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) { | |
2739 | pgdat->classzone_idx = min(pgdat->classzone_idx, | |
2740 | (enum zone_type)ZONE_NORMAL); | |
2741 | wake_up_interruptible(&pgdat->kswapd_wait); | |
2742 | } | |
2743 | ||
2744 | return wmark_ok; | |
2745 | } | |
2746 | ||
2747 | /* | |
2748 | * Throttle direct reclaimers if backing storage is backed by the network | |
2749 | * and the PFMEMALLOC reserve for the preferred node is getting dangerously | |
2750 | * depleted. kswapd will continue to make progress and wake the processes | |
50694c28 MG |
2751 | * when the low watermark is reached. |
2752 | * | |
2753 | * Returns true if a fatal signal was delivered during throttling. If this | |
2754 | * happens, the page allocator should not consider triggering the OOM killer. | |
5515061d | 2755 | */ |
50694c28 | 2756 | static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist, |
5515061d MG |
2757 | nodemask_t *nodemask) |
2758 | { | |
675becce | 2759 | struct zoneref *z; |
5515061d | 2760 | struct zone *zone; |
675becce | 2761 | pg_data_t *pgdat = NULL; |
5515061d MG |
2762 | |
2763 | /* | |
2764 | * Kernel threads should not be throttled as they may be indirectly | |
2765 | * responsible for cleaning pages necessary for reclaim to make forward | |
2766 | * progress. kjournald for example may enter direct reclaim while | |
2767 | * committing a transaction where throttling it could forcing other | |
2768 | * processes to block on log_wait_commit(). | |
2769 | */ | |
2770 | if (current->flags & PF_KTHREAD) | |
50694c28 MG |
2771 | goto out; |
2772 | ||
2773 | /* | |
2774 | * If a fatal signal is pending, this process should not throttle. | |
2775 | * It should return quickly so it can exit and free its memory | |
2776 | */ | |
2777 | if (fatal_signal_pending(current)) | |
2778 | goto out; | |
5515061d | 2779 | |
675becce MG |
2780 | /* |
2781 | * Check if the pfmemalloc reserves are ok by finding the first node | |
2782 | * with a usable ZONE_NORMAL or lower zone. The expectation is that | |
2783 | * GFP_KERNEL will be required for allocating network buffers when | |
2784 | * swapping over the network so ZONE_HIGHMEM is unusable. | |
2785 | * | |
2786 | * Throttling is based on the first usable node and throttled processes | |
2787 | * wait on a queue until kswapd makes progress and wakes them. There | |
2788 | * is an affinity then between processes waking up and where reclaim | |
2789 | * progress has been made assuming the process wakes on the same node. | |
2790 | * More importantly, processes running on remote nodes will not compete | |
2791 | * for remote pfmemalloc reserves and processes on different nodes | |
2792 | * should make reasonable progress. | |
2793 | */ | |
2794 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
17636faa | 2795 | gfp_zone(gfp_mask), nodemask) { |
675becce MG |
2796 | if (zone_idx(zone) > ZONE_NORMAL) |
2797 | continue; | |
2798 | ||
2799 | /* Throttle based on the first usable node */ | |
2800 | pgdat = zone->zone_pgdat; | |
2801 | if (pfmemalloc_watermark_ok(pgdat)) | |
2802 | goto out; | |
2803 | break; | |
2804 | } | |
2805 | ||
2806 | /* If no zone was usable by the allocation flags then do not throttle */ | |
2807 | if (!pgdat) | |
50694c28 | 2808 | goto out; |
5515061d | 2809 | |
68243e76 MG |
2810 | /* Account for the throttling */ |
2811 | count_vm_event(PGSCAN_DIRECT_THROTTLE); | |
2812 | ||
5515061d MG |
2813 | /* |
2814 | * If the caller cannot enter the filesystem, it's possible that it | |
2815 | * is due to the caller holding an FS lock or performing a journal | |
2816 | * transaction in the case of a filesystem like ext[3|4]. In this case, | |
2817 | * it is not safe to block on pfmemalloc_wait as kswapd could be | |
2818 | * blocked waiting on the same lock. Instead, throttle for up to a | |
2819 | * second before continuing. | |
2820 | */ | |
2821 | if (!(gfp_mask & __GFP_FS)) { | |
2822 | wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, | |
2823 | pfmemalloc_watermark_ok(pgdat), HZ); | |
50694c28 MG |
2824 | |
2825 | goto check_pending; | |
5515061d MG |
2826 | } |
2827 | ||
2828 | /* Throttle until kswapd wakes the process */ | |
2829 | wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, | |
2830 | pfmemalloc_watermark_ok(pgdat)); | |
50694c28 MG |
2831 | |
2832 | check_pending: | |
2833 | if (fatal_signal_pending(current)) | |
2834 | return true; | |
2835 | ||
2836 | out: | |
2837 | return false; | |
5515061d MG |
2838 | } |
2839 | ||
dac1d27b | 2840 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
327c0e96 | 2841 | gfp_t gfp_mask, nodemask_t *nodemask) |
66e1707b | 2842 | { |
33906bc5 | 2843 | unsigned long nr_reclaimed; |
66e1707b | 2844 | struct scan_control sc = { |
ee814fe2 | 2845 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
21caf2fc | 2846 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
ee814fe2 JW |
2847 | .order = order, |
2848 | .nodemask = nodemask, | |
2849 | .priority = DEF_PRIORITY, | |
66e1707b | 2850 | .may_writepage = !laptop_mode, |
a6dc60f8 | 2851 | .may_unmap = 1, |
2e2e4259 | 2852 | .may_swap = 1, |
66e1707b BS |
2853 | }; |
2854 | ||
5515061d | 2855 | /* |
50694c28 MG |
2856 | * Do not enter reclaim if fatal signal was delivered while throttled. |
2857 | * 1 is returned so that the page allocator does not OOM kill at this | |
2858 | * point. | |
5515061d | 2859 | */ |
50694c28 | 2860 | if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask)) |
5515061d MG |
2861 | return 1; |
2862 | ||
33906bc5 MG |
2863 | trace_mm_vmscan_direct_reclaim_begin(order, |
2864 | sc.may_writepage, | |
2865 | gfp_mask); | |
2866 | ||
3115cd91 | 2867 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
33906bc5 MG |
2868 | |
2869 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); | |
2870 | ||
2871 | return nr_reclaimed; | |
66e1707b BS |
2872 | } |
2873 | ||
c255a458 | 2874 | #ifdef CONFIG_MEMCG |
66e1707b | 2875 | |
72835c86 | 2876 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg, |
4e416953 | 2877 | gfp_t gfp_mask, bool noswap, |
0ae5e89c YH |
2878 | struct zone *zone, |
2879 | unsigned long *nr_scanned) | |
4e416953 BS |
2880 | { |
2881 | struct scan_control sc = { | |
b8f5c566 | 2882 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
ee814fe2 | 2883 | .target_mem_cgroup = memcg, |
4e416953 BS |
2884 | .may_writepage = !laptop_mode, |
2885 | .may_unmap = 1, | |
2886 | .may_swap = !noswap, | |
4e416953 | 2887 | }; |
6b4f7799 | 2888 | unsigned long lru_pages; |
0ae5e89c | 2889 | |
4e416953 BS |
2890 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2891 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); | |
bdce6d9e | 2892 | |
9e3b2f8c | 2893 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, |
bdce6d9e KM |
2894 | sc.may_writepage, |
2895 | sc.gfp_mask); | |
2896 | ||
4e416953 BS |
2897 | /* |
2898 | * NOTE: Although we can get the priority field, using it | |
2899 | * here is not a good idea, since it limits the pages we can scan. | |
2900 | * if we don't reclaim here, the shrink_zone from balance_pgdat | |
2901 | * will pick up pages from other mem cgroup's as well. We hack | |
2902 | * the priority and make it zero. | |
2903 | */ | |
33377678 | 2904 | shrink_zone_memcg(zone, memcg, &sc, &lru_pages); |
bdce6d9e KM |
2905 | |
2906 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); | |
2907 | ||
0ae5e89c | 2908 | *nr_scanned = sc.nr_scanned; |
4e416953 BS |
2909 | return sc.nr_reclaimed; |
2910 | } | |
2911 | ||
72835c86 | 2912 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, |
b70a2a21 | 2913 | unsigned long nr_pages, |
a7885eb8 | 2914 | gfp_t gfp_mask, |
b70a2a21 | 2915 | bool may_swap) |
66e1707b | 2916 | { |
4e416953 | 2917 | struct zonelist *zonelist; |
bdce6d9e | 2918 | unsigned long nr_reclaimed; |
889976db | 2919 | int nid; |
66e1707b | 2920 | struct scan_control sc = { |
b70a2a21 | 2921 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
a09ed5e0 YH |
2922 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2923 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), | |
ee814fe2 JW |
2924 | .target_mem_cgroup = memcg, |
2925 | .priority = DEF_PRIORITY, | |
2926 | .may_writepage = !laptop_mode, | |
2927 | .may_unmap = 1, | |
b70a2a21 | 2928 | .may_swap = may_swap, |
a09ed5e0 | 2929 | }; |
66e1707b | 2930 | |
889976db YH |
2931 | /* |
2932 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't | |
2933 | * take care of from where we get pages. So the node where we start the | |
2934 | * scan does not need to be the current node. | |
2935 | */ | |
72835c86 | 2936 | nid = mem_cgroup_select_victim_node(memcg); |
889976db YH |
2937 | |
2938 | zonelist = NODE_DATA(nid)->node_zonelists; | |
bdce6d9e KM |
2939 | |
2940 | trace_mm_vmscan_memcg_reclaim_begin(0, | |
2941 | sc.may_writepage, | |
2942 | sc.gfp_mask); | |
2943 | ||
3115cd91 | 2944 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
bdce6d9e KM |
2945 | |
2946 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); | |
2947 | ||
2948 | return nr_reclaimed; | |
66e1707b BS |
2949 | } |
2950 | #endif | |
2951 | ||
9e3b2f8c | 2952 | static void age_active_anon(struct zone *zone, struct scan_control *sc) |
f16015fb | 2953 | { |
b95a2f2d | 2954 | struct mem_cgroup *memcg; |
f16015fb | 2955 | |
b95a2f2d JW |
2956 | if (!total_swap_pages) |
2957 | return; | |
2958 | ||
2959 | memcg = mem_cgroup_iter(NULL, NULL, NULL); | |
2960 | do { | |
c56d5c7d | 2961 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
b95a2f2d | 2962 | |
c56d5c7d | 2963 | if (inactive_anon_is_low(lruvec)) |
1a93be0e | 2964 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
9e3b2f8c | 2965 | sc, LRU_ACTIVE_ANON); |
b95a2f2d JW |
2966 | |
2967 | memcg = mem_cgroup_iter(NULL, memcg, NULL); | |
2968 | } while (memcg); | |
f16015fb JW |
2969 | } |
2970 | ||
60cefed4 JW |
2971 | static bool zone_balanced(struct zone *zone, int order, |
2972 | unsigned long balance_gap, int classzone_idx) | |
2973 | { | |
2974 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) + | |
e2b19197 | 2975 | balance_gap, classzone_idx)) |
60cefed4 JW |
2976 | return false; |
2977 | ||
ebff3980 VB |
2978 | if (IS_ENABLED(CONFIG_COMPACTION) && order && compaction_suitable(zone, |
2979 | order, 0, classzone_idx) == COMPACT_SKIPPED) | |
60cefed4 JW |
2980 | return false; |
2981 | ||
2982 | return true; | |
2983 | } | |
2984 | ||
1741c877 | 2985 | /* |
4ae0a48b ZC |
2986 | * pgdat_balanced() is used when checking if a node is balanced. |
2987 | * | |
2988 | * For order-0, all zones must be balanced! | |
2989 | * | |
2990 | * For high-order allocations only zones that meet watermarks and are in a | |
2991 | * zone allowed by the callers classzone_idx are added to balanced_pages. The | |
2992 | * total of balanced pages must be at least 25% of the zones allowed by | |
2993 | * classzone_idx for the node to be considered balanced. Forcing all zones to | |
2994 | * be balanced for high orders can cause excessive reclaim when there are | |
2995 | * imbalanced zones. | |
1741c877 MG |
2996 | * The choice of 25% is due to |
2997 | * o a 16M DMA zone that is balanced will not balance a zone on any | |
2998 | * reasonable sized machine | |
2999 | * o On all other machines, the top zone must be at least a reasonable | |
25985edc | 3000 | * percentage of the middle zones. For example, on 32-bit x86, highmem |
1741c877 MG |
3001 | * would need to be at least 256M for it to be balance a whole node. |
3002 | * Similarly, on x86-64 the Normal zone would need to be at least 1G | |
3003 | * to balance a node on its own. These seemed like reasonable ratios. | |
3004 | */ | |
4ae0a48b | 3005 | static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx) |
1741c877 | 3006 | { |
b40da049 | 3007 | unsigned long managed_pages = 0; |
4ae0a48b | 3008 | unsigned long balanced_pages = 0; |
1741c877 MG |
3009 | int i; |
3010 | ||
4ae0a48b ZC |
3011 | /* Check the watermark levels */ |
3012 | for (i = 0; i <= classzone_idx; i++) { | |
3013 | struct zone *zone = pgdat->node_zones + i; | |
1741c877 | 3014 | |
4ae0a48b ZC |
3015 | if (!populated_zone(zone)) |
3016 | continue; | |
3017 | ||
b40da049 | 3018 | managed_pages += zone->managed_pages; |
4ae0a48b ZC |
3019 | |
3020 | /* | |
3021 | * A special case here: | |
3022 | * | |
3023 | * balance_pgdat() skips over all_unreclaimable after | |
3024 | * DEF_PRIORITY. Effectively, it considers them balanced so | |
3025 | * they must be considered balanced here as well! | |
3026 | */ | |
6e543d57 | 3027 | if (!zone_reclaimable(zone)) { |
b40da049 | 3028 | balanced_pages += zone->managed_pages; |
4ae0a48b ZC |
3029 | continue; |
3030 | } | |
3031 | ||
3032 | if (zone_balanced(zone, order, 0, i)) | |
b40da049 | 3033 | balanced_pages += zone->managed_pages; |
4ae0a48b ZC |
3034 | else if (!order) |
3035 | return false; | |
3036 | } | |
3037 | ||
3038 | if (order) | |
b40da049 | 3039 | return balanced_pages >= (managed_pages >> 2); |
4ae0a48b ZC |
3040 | else |
3041 | return true; | |
1741c877 MG |
3042 | } |
3043 | ||
5515061d MG |
3044 | /* |
3045 | * Prepare kswapd for sleeping. This verifies that there are no processes | |
3046 | * waiting in throttle_direct_reclaim() and that watermarks have been met. | |
3047 | * | |
3048 | * Returns true if kswapd is ready to sleep | |
3049 | */ | |
3050 | static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining, | |
dc83edd9 | 3051 | int classzone_idx) |
f50de2d3 | 3052 | { |
f50de2d3 MG |
3053 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ |
3054 | if (remaining) | |
5515061d MG |
3055 | return false; |
3056 | ||
3057 | /* | |
9e5e3661 VB |
3058 | * The throttled processes are normally woken up in balance_pgdat() as |
3059 | * soon as pfmemalloc_watermark_ok() is true. But there is a potential | |
3060 | * race between when kswapd checks the watermarks and a process gets | |
3061 | * throttled. There is also a potential race if processes get | |
3062 | * throttled, kswapd wakes, a large process exits thereby balancing the | |
3063 | * zones, which causes kswapd to exit balance_pgdat() before reaching | |
3064 | * the wake up checks. If kswapd is going to sleep, no process should | |
3065 | * be sleeping on pfmemalloc_wait, so wake them now if necessary. If | |
3066 | * the wake up is premature, processes will wake kswapd and get | |
3067 | * throttled again. The difference from wake ups in balance_pgdat() is | |
3068 | * that here we are under prepare_to_wait(). | |
5515061d | 3069 | */ |
9e5e3661 VB |
3070 | if (waitqueue_active(&pgdat->pfmemalloc_wait)) |
3071 | wake_up_all(&pgdat->pfmemalloc_wait); | |
f50de2d3 | 3072 | |
4ae0a48b | 3073 | return pgdat_balanced(pgdat, order, classzone_idx); |
f50de2d3 MG |
3074 | } |
3075 | ||
75485363 MG |
3076 | /* |
3077 | * kswapd shrinks the zone by the number of pages required to reach | |
3078 | * the high watermark. | |
b8e83b94 MG |
3079 | * |
3080 | * Returns true if kswapd scanned at least the requested number of pages to | |
283aba9f MG |
3081 | * reclaim or if the lack of progress was due to pages under writeback. |
3082 | * This is used to determine if the scanning priority needs to be raised. | |
75485363 | 3083 | */ |
b8e83b94 | 3084 | static bool kswapd_shrink_zone(struct zone *zone, |
7c954f6d | 3085 | int classzone_idx, |
75485363 | 3086 | struct scan_control *sc, |
2ab44f43 | 3087 | unsigned long *nr_attempted) |
75485363 | 3088 | { |
7c954f6d MG |
3089 | int testorder = sc->order; |
3090 | unsigned long balance_gap; | |
7c954f6d | 3091 | bool lowmem_pressure; |
75485363 MG |
3092 | |
3093 | /* Reclaim above the high watermark. */ | |
3094 | sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone)); | |
7c954f6d MG |
3095 | |
3096 | /* | |
3097 | * Kswapd reclaims only single pages with compaction enabled. Trying | |
3098 | * too hard to reclaim until contiguous free pages have become | |
3099 | * available can hurt performance by evicting too much useful data | |
3100 | * from memory. Do not reclaim more than needed for compaction. | |
3101 | */ | |
3102 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && | |
ebff3980 VB |
3103 | compaction_suitable(zone, sc->order, 0, classzone_idx) |
3104 | != COMPACT_SKIPPED) | |
7c954f6d MG |
3105 | testorder = 0; |
3106 | ||
3107 | /* | |
3108 | * We put equal pressure on every zone, unless one zone has way too | |
3109 | * many pages free already. The "too many pages" is defined as the | |
3110 | * high wmark plus a "gap" where the gap is either the low | |
3111 | * watermark or 1% of the zone, whichever is smaller. | |
3112 | */ | |
4be89a34 JZ |
3113 | balance_gap = min(low_wmark_pages(zone), DIV_ROUND_UP( |
3114 | zone->managed_pages, KSWAPD_ZONE_BALANCE_GAP_RATIO)); | |
7c954f6d MG |
3115 | |
3116 | /* | |
3117 | * If there is no low memory pressure or the zone is balanced then no | |
3118 | * reclaim is necessary | |
3119 | */ | |
3120 | lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone)); | |
3121 | if (!lowmem_pressure && zone_balanced(zone, testorder, | |
3122 | balance_gap, classzone_idx)) | |
3123 | return true; | |
3124 | ||
6b4f7799 | 3125 | shrink_zone(zone, sc, zone_idx(zone) == classzone_idx); |
75485363 | 3126 | |
2ab44f43 MG |
3127 | /* Account for the number of pages attempted to reclaim */ |
3128 | *nr_attempted += sc->nr_to_reclaim; | |
3129 | ||
57054651 | 3130 | clear_bit(ZONE_WRITEBACK, &zone->flags); |
283aba9f | 3131 | |
7c954f6d MG |
3132 | /* |
3133 | * If a zone reaches its high watermark, consider it to be no longer | |
3134 | * congested. It's possible there are dirty pages backed by congested | |
3135 | * BDIs but as pressure is relieved, speculatively avoid congestion | |
3136 | * waits. | |
3137 | */ | |
6e543d57 | 3138 | if (zone_reclaimable(zone) && |
7c954f6d | 3139 | zone_balanced(zone, testorder, 0, classzone_idx)) { |
57054651 JW |
3140 | clear_bit(ZONE_CONGESTED, &zone->flags); |
3141 | clear_bit(ZONE_DIRTY, &zone->flags); | |
7c954f6d MG |
3142 | } |
3143 | ||
b8e83b94 | 3144 | return sc->nr_scanned >= sc->nr_to_reclaim; |
75485363 MG |
3145 | } |
3146 | ||
1da177e4 LT |
3147 | /* |
3148 | * For kswapd, balance_pgdat() will work across all this node's zones until | |
41858966 | 3149 | * they are all at high_wmark_pages(zone). |
1da177e4 | 3150 | * |
0abdee2b | 3151 | * Returns the final order kswapd was reclaiming at |
1da177e4 LT |
3152 | * |
3153 | * There is special handling here for zones which are full of pinned pages. | |
3154 | * This can happen if the pages are all mlocked, or if they are all used by | |
3155 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. | |
3156 | * What we do is to detect the case where all pages in the zone have been | |
3157 | * scanned twice and there has been zero successful reclaim. Mark the zone as | |
3158 | * dead and from now on, only perform a short scan. Basically we're polling | |
3159 | * the zone for when the problem goes away. | |
3160 | * | |
3161 | * kswapd scans the zones in the highmem->normal->dma direction. It skips | |
41858966 MG |
3162 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
3163 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the | |
3164 | * lower zones regardless of the number of free pages in the lower zones. This | |
3165 | * interoperates with the page allocator fallback scheme to ensure that aging | |
3166 | * of pages is balanced across the zones. | |
1da177e4 | 3167 | */ |
99504748 | 3168 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, |
dc83edd9 | 3169 | int *classzone_idx) |
1da177e4 | 3170 | { |
1da177e4 | 3171 | int i; |
99504748 | 3172 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
0608f43d AM |
3173 | unsigned long nr_soft_reclaimed; |
3174 | unsigned long nr_soft_scanned; | |
179e9639 AM |
3175 | struct scan_control sc = { |
3176 | .gfp_mask = GFP_KERNEL, | |
ee814fe2 | 3177 | .order = order, |
b8e83b94 | 3178 | .priority = DEF_PRIORITY, |
ee814fe2 | 3179 | .may_writepage = !laptop_mode, |
a6dc60f8 | 3180 | .may_unmap = 1, |
2e2e4259 | 3181 | .may_swap = 1, |
179e9639 | 3182 | }; |
f8891e5e | 3183 | count_vm_event(PAGEOUTRUN); |
1da177e4 | 3184 | |
9e3b2f8c | 3185 | do { |
2ab44f43 | 3186 | unsigned long nr_attempted = 0; |
b8e83b94 | 3187 | bool raise_priority = true; |
2ab44f43 | 3188 | bool pgdat_needs_compaction = (order > 0); |
b8e83b94 MG |
3189 | |
3190 | sc.nr_reclaimed = 0; | |
1da177e4 | 3191 | |
d6277db4 RW |
3192 | /* |
3193 | * Scan in the highmem->dma direction for the highest | |
3194 | * zone which needs scanning | |
3195 | */ | |
3196 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { | |
3197 | struct zone *zone = pgdat->node_zones + i; | |
1da177e4 | 3198 | |
d6277db4 RW |
3199 | if (!populated_zone(zone)) |
3200 | continue; | |
1da177e4 | 3201 | |
6e543d57 LD |
3202 | if (sc.priority != DEF_PRIORITY && |
3203 | !zone_reclaimable(zone)) | |
d6277db4 | 3204 | continue; |
1da177e4 | 3205 | |
556adecb RR |
3206 | /* |
3207 | * Do some background aging of the anon list, to give | |
3208 | * pages a chance to be referenced before reclaiming. | |
3209 | */ | |
9e3b2f8c | 3210 | age_active_anon(zone, &sc); |
556adecb | 3211 | |
cc715d99 MG |
3212 | /* |
3213 | * If the number of buffer_heads in the machine | |
3214 | * exceeds the maximum allowed level and this node | |
3215 | * has a highmem zone, force kswapd to reclaim from | |
3216 | * it to relieve lowmem pressure. | |
3217 | */ | |
3218 | if (buffer_heads_over_limit && is_highmem_idx(i)) { | |
3219 | end_zone = i; | |
3220 | break; | |
3221 | } | |
3222 | ||
60cefed4 | 3223 | if (!zone_balanced(zone, order, 0, 0)) { |
d6277db4 | 3224 | end_zone = i; |
e1dbeda6 | 3225 | break; |
439423f6 | 3226 | } else { |
d43006d5 MG |
3227 | /* |
3228 | * If balanced, clear the dirty and congested | |
3229 | * flags | |
3230 | */ | |
57054651 JW |
3231 | clear_bit(ZONE_CONGESTED, &zone->flags); |
3232 | clear_bit(ZONE_DIRTY, &zone->flags); | |
1da177e4 | 3233 | } |
1da177e4 | 3234 | } |
dafcb73e | 3235 | |
b8e83b94 | 3236 | if (i < 0) |
e1dbeda6 AM |
3237 | goto out; |
3238 | ||
1da177e4 LT |
3239 | for (i = 0; i <= end_zone; i++) { |
3240 | struct zone *zone = pgdat->node_zones + i; | |
3241 | ||
2ab44f43 MG |
3242 | if (!populated_zone(zone)) |
3243 | continue; | |
3244 | ||
2ab44f43 MG |
3245 | /* |
3246 | * If any zone is currently balanced then kswapd will | |
3247 | * not call compaction as it is expected that the | |
3248 | * necessary pages are already available. | |
3249 | */ | |
3250 | if (pgdat_needs_compaction && | |
3251 | zone_watermark_ok(zone, order, | |
3252 | low_wmark_pages(zone), | |
3253 | *classzone_idx, 0)) | |
3254 | pgdat_needs_compaction = false; | |
1da177e4 LT |
3255 | } |
3256 | ||
b7ea3c41 MG |
3257 | /* |
3258 | * If we're getting trouble reclaiming, start doing writepage | |
3259 | * even in laptop mode. | |
3260 | */ | |
3261 | if (sc.priority < DEF_PRIORITY - 2) | |
3262 | sc.may_writepage = 1; | |
3263 | ||
1da177e4 LT |
3264 | /* |
3265 | * Now scan the zone in the dma->highmem direction, stopping | |
3266 | * at the last zone which needs scanning. | |
3267 | * | |
3268 | * We do this because the page allocator works in the opposite | |
3269 | * direction. This prevents the page allocator from allocating | |
3270 | * pages behind kswapd's direction of progress, which would | |
3271 | * cause too much scanning of the lower zones. | |
3272 | */ | |
3273 | for (i = 0; i <= end_zone; i++) { | |
3274 | struct zone *zone = pgdat->node_zones + i; | |
3275 | ||
f3fe6512 | 3276 | if (!populated_zone(zone)) |
1da177e4 LT |
3277 | continue; |
3278 | ||
6e543d57 LD |
3279 | if (sc.priority != DEF_PRIORITY && |
3280 | !zone_reclaimable(zone)) | |
1da177e4 LT |
3281 | continue; |
3282 | ||
1da177e4 | 3283 | sc.nr_scanned = 0; |
4e416953 | 3284 | |
0608f43d AM |
3285 | nr_soft_scanned = 0; |
3286 | /* | |
3287 | * Call soft limit reclaim before calling shrink_zone. | |
3288 | */ | |
3289 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | |
3290 | order, sc.gfp_mask, | |
3291 | &nr_soft_scanned); | |
3292 | sc.nr_reclaimed += nr_soft_reclaimed; | |
3293 | ||
32a4330d | 3294 | /* |
7c954f6d MG |
3295 | * There should be no need to raise the scanning |
3296 | * priority if enough pages are already being scanned | |
3297 | * that that high watermark would be met at 100% | |
3298 | * efficiency. | |
fe2c2a10 | 3299 | */ |
6b4f7799 JW |
3300 | if (kswapd_shrink_zone(zone, end_zone, |
3301 | &sc, &nr_attempted)) | |
7c954f6d | 3302 | raise_priority = false; |
1da177e4 | 3303 | } |
5515061d MG |
3304 | |
3305 | /* | |
3306 | * If the low watermark is met there is no need for processes | |
3307 | * to be throttled on pfmemalloc_wait as they should not be | |
3308 | * able to safely make forward progress. Wake them | |
3309 | */ | |
3310 | if (waitqueue_active(&pgdat->pfmemalloc_wait) && | |
3311 | pfmemalloc_watermark_ok(pgdat)) | |
cfc51155 | 3312 | wake_up_all(&pgdat->pfmemalloc_wait); |
5515061d | 3313 | |
1da177e4 | 3314 | /* |
b8e83b94 MG |
3315 | * Fragmentation may mean that the system cannot be rebalanced |
3316 | * for high-order allocations in all zones. If twice the | |
3317 | * allocation size has been reclaimed and the zones are still | |
3318 | * not balanced then recheck the watermarks at order-0 to | |
3319 | * prevent kswapd reclaiming excessively. Assume that a | |
3320 | * process requested a high-order can direct reclaim/compact. | |
1da177e4 | 3321 | */ |
b8e83b94 MG |
3322 | if (order && sc.nr_reclaimed >= 2UL << order) |
3323 | order = sc.order = 0; | |
8357376d | 3324 | |
b8e83b94 MG |
3325 | /* Check if kswapd should be suspending */ |
3326 | if (try_to_freeze() || kthread_should_stop()) | |
3327 | break; | |
8357376d | 3328 | |
2ab44f43 MG |
3329 | /* |
3330 | * Compact if necessary and kswapd is reclaiming at least the | |
3331 | * high watermark number of pages as requsted | |
3332 | */ | |
3333 | if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted) | |
3334 | compact_pgdat(pgdat, order); | |
3335 | ||
73ce02e9 | 3336 | /* |
b8e83b94 MG |
3337 | * Raise priority if scanning rate is too low or there was no |
3338 | * progress in reclaiming pages | |
73ce02e9 | 3339 | */ |
b8e83b94 MG |
3340 | if (raise_priority || !sc.nr_reclaimed) |
3341 | sc.priority--; | |
9aa41348 | 3342 | } while (sc.priority >= 1 && |
b8e83b94 | 3343 | !pgdat_balanced(pgdat, order, *classzone_idx)); |
1da177e4 | 3344 | |
b8e83b94 | 3345 | out: |
0abdee2b | 3346 | /* |
5515061d | 3347 | * Return the order we were reclaiming at so prepare_kswapd_sleep() |
0abdee2b MG |
3348 | * makes a decision on the order we were last reclaiming at. However, |
3349 | * if another caller entered the allocator slow path while kswapd | |
3350 | * was awake, order will remain at the higher level | |
3351 | */ | |
dc83edd9 | 3352 | *classzone_idx = end_zone; |
0abdee2b | 3353 | return order; |
1da177e4 LT |
3354 | } |
3355 | ||
dc83edd9 | 3356 | static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx) |
f0bc0a60 KM |
3357 | { |
3358 | long remaining = 0; | |
3359 | DEFINE_WAIT(wait); | |
3360 | ||
3361 | if (freezing(current) || kthread_should_stop()) | |
3362 | return; | |
3363 | ||
3364 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | |
3365 | ||
3366 | /* Try to sleep for a short interval */ | |
5515061d | 3367 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
f0bc0a60 KM |
3368 | remaining = schedule_timeout(HZ/10); |
3369 | finish_wait(&pgdat->kswapd_wait, &wait); | |
3370 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | |
3371 | } | |
3372 | ||
3373 | /* | |
3374 | * After a short sleep, check if it was a premature sleep. If not, then | |
3375 | * go fully to sleep until explicitly woken up. | |
3376 | */ | |
5515061d | 3377 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
f0bc0a60 KM |
3378 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
3379 | ||
3380 | /* | |
3381 | * vmstat counters are not perfectly accurate and the estimated | |
3382 | * value for counters such as NR_FREE_PAGES can deviate from the | |
3383 | * true value by nr_online_cpus * threshold. To avoid the zone | |
3384 | * watermarks being breached while under pressure, we reduce the | |
3385 | * per-cpu vmstat threshold while kswapd is awake and restore | |
3386 | * them before going back to sleep. | |
3387 | */ | |
3388 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); | |
1c7e7f6c | 3389 | |
62997027 MG |
3390 | /* |
3391 | * Compaction records what page blocks it recently failed to | |
3392 | * isolate pages from and skips them in the future scanning. | |
3393 | * When kswapd is going to sleep, it is reasonable to assume | |
3394 | * that pages and compaction may succeed so reset the cache. | |
3395 | */ | |
3396 | reset_isolation_suitable(pgdat); | |
3397 | ||
1c7e7f6c AK |
3398 | if (!kthread_should_stop()) |
3399 | schedule(); | |
3400 | ||
f0bc0a60 KM |
3401 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
3402 | } else { | |
3403 | if (remaining) | |
3404 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); | |
3405 | else | |
3406 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); | |
3407 | } | |
3408 | finish_wait(&pgdat->kswapd_wait, &wait); | |
3409 | } | |
3410 | ||
1da177e4 LT |
3411 | /* |
3412 | * The background pageout daemon, started as a kernel thread | |
4f98a2fe | 3413 | * from the init process. |
1da177e4 LT |
3414 | * |
3415 | * This basically trickles out pages so that we have _some_ | |
3416 | * free memory available even if there is no other activity | |
3417 | * that frees anything up. This is needed for things like routing | |
3418 | * etc, where we otherwise might have all activity going on in | |
3419 | * asynchronous contexts that cannot page things out. | |
3420 | * | |
3421 | * If there are applications that are active memory-allocators | |
3422 | * (most normal use), this basically shouldn't matter. | |
3423 | */ | |
3424 | static int kswapd(void *p) | |
3425 | { | |
215ddd66 | 3426 | unsigned long order, new_order; |
d2ebd0f6 | 3427 | unsigned balanced_order; |
215ddd66 | 3428 | int classzone_idx, new_classzone_idx; |
d2ebd0f6 | 3429 | int balanced_classzone_idx; |
1da177e4 LT |
3430 | pg_data_t *pgdat = (pg_data_t*)p; |
3431 | struct task_struct *tsk = current; | |
f0bc0a60 | 3432 | |
1da177e4 LT |
3433 | struct reclaim_state reclaim_state = { |
3434 | .reclaimed_slab = 0, | |
3435 | }; | |
a70f7302 | 3436 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
1da177e4 | 3437 | |
cf40bd16 NP |
3438 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
3439 | ||
174596a0 | 3440 | if (!cpumask_empty(cpumask)) |
c5f59f08 | 3441 | set_cpus_allowed_ptr(tsk, cpumask); |
1da177e4 LT |
3442 | current->reclaim_state = &reclaim_state; |
3443 | ||
3444 | /* | |
3445 | * Tell the memory management that we're a "memory allocator", | |
3446 | * and that if we need more memory we should get access to it | |
3447 | * regardless (see "__alloc_pages()"). "kswapd" should | |
3448 | * never get caught in the normal page freeing logic. | |
3449 | * | |
3450 | * (Kswapd normally doesn't need memory anyway, but sometimes | |
3451 | * you need a small amount of memory in order to be able to | |
3452 | * page out something else, and this flag essentially protects | |
3453 | * us from recursively trying to free more memory as we're | |
3454 | * trying to free the first piece of memory in the first place). | |
3455 | */ | |
930d9152 | 3456 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
83144186 | 3457 | set_freezable(); |
1da177e4 | 3458 | |
215ddd66 | 3459 | order = new_order = 0; |
d2ebd0f6 | 3460 | balanced_order = 0; |
215ddd66 | 3461 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; |
d2ebd0f6 | 3462 | balanced_classzone_idx = classzone_idx; |
1da177e4 | 3463 | for ( ; ; ) { |
6f6313d4 | 3464 | bool ret; |
3e1d1d28 | 3465 | |
215ddd66 MG |
3466 | /* |
3467 | * If the last balance_pgdat was unsuccessful it's unlikely a | |
3468 | * new request of a similar or harder type will succeed soon | |
3469 | * so consider going to sleep on the basis we reclaimed at | |
3470 | */ | |
81c5857b | 3471 | if (balanced_order == new_order) { |
215ddd66 MG |
3472 | new_order = pgdat->kswapd_max_order; |
3473 | new_classzone_idx = pgdat->classzone_idx; | |
3474 | pgdat->kswapd_max_order = 0; | |
3475 | pgdat->classzone_idx = pgdat->nr_zones - 1; | |
3476 | } | |
3477 | ||
99504748 | 3478 | if (order < new_order || classzone_idx > new_classzone_idx) { |
1da177e4 LT |
3479 | /* |
3480 | * Don't sleep if someone wants a larger 'order' | |
99504748 | 3481 | * allocation or has tigher zone constraints |
1da177e4 LT |
3482 | */ |
3483 | order = new_order; | |
99504748 | 3484 | classzone_idx = new_classzone_idx; |
1da177e4 | 3485 | } else { |
d2ebd0f6 AS |
3486 | kswapd_try_to_sleep(pgdat, balanced_order, |
3487 | balanced_classzone_idx); | |
1da177e4 | 3488 | order = pgdat->kswapd_max_order; |
99504748 | 3489 | classzone_idx = pgdat->classzone_idx; |
f0dfcde0 AS |
3490 | new_order = order; |
3491 | new_classzone_idx = classzone_idx; | |
4d40502e | 3492 | pgdat->kswapd_max_order = 0; |
215ddd66 | 3493 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
1da177e4 | 3494 | } |
1da177e4 | 3495 | |
8fe23e05 DR |
3496 | ret = try_to_freeze(); |
3497 | if (kthread_should_stop()) | |
3498 | break; | |
3499 | ||
3500 | /* | |
3501 | * We can speed up thawing tasks if we don't call balance_pgdat | |
3502 | * after returning from the refrigerator | |
3503 | */ | |
33906bc5 MG |
3504 | if (!ret) { |
3505 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); | |
d2ebd0f6 AS |
3506 | balanced_classzone_idx = classzone_idx; |
3507 | balanced_order = balance_pgdat(pgdat, order, | |
3508 | &balanced_classzone_idx); | |
33906bc5 | 3509 | } |
1da177e4 | 3510 | } |
b0a8cc58 | 3511 | |
71abdc15 | 3512 | tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD); |
b0a8cc58 | 3513 | current->reclaim_state = NULL; |
71abdc15 JW |
3514 | lockdep_clear_current_reclaim_state(); |
3515 | ||
1da177e4 LT |
3516 | return 0; |
3517 | } | |
3518 | ||
3519 | /* | |
3520 | * A zone is low on free memory, so wake its kswapd task to service it. | |
3521 | */ | |
99504748 | 3522 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx) |
1da177e4 LT |
3523 | { |
3524 | pg_data_t *pgdat; | |
3525 | ||
f3fe6512 | 3526 | if (!populated_zone(zone)) |
1da177e4 LT |
3527 | return; |
3528 | ||
344736f2 | 3529 | if (!cpuset_zone_allowed(zone, GFP_KERNEL | __GFP_HARDWALL)) |
1da177e4 | 3530 | return; |
88f5acf8 | 3531 | pgdat = zone->zone_pgdat; |
99504748 | 3532 | if (pgdat->kswapd_max_order < order) { |
1da177e4 | 3533 | pgdat->kswapd_max_order = order; |
99504748 MG |
3534 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); |
3535 | } | |
8d0986e2 | 3536 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
1da177e4 | 3537 | return; |
892f795d | 3538 | if (zone_balanced(zone, order, 0, 0)) |
88f5acf8 MG |
3539 | return; |
3540 | ||
3541 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); | |
8d0986e2 | 3542 | wake_up_interruptible(&pgdat->kswapd_wait); |
1da177e4 LT |
3543 | } |
3544 | ||
c6f37f12 | 3545 | #ifdef CONFIG_HIBERNATION |
1da177e4 | 3546 | /* |
7b51755c | 3547 | * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of |
d6277db4 RW |
3548 | * freed pages. |
3549 | * | |
3550 | * Rather than trying to age LRUs the aim is to preserve the overall | |
3551 | * LRU order by reclaiming preferentially | |
3552 | * inactive > active > active referenced > active mapped | |
1da177e4 | 3553 | */ |
7b51755c | 3554 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
1da177e4 | 3555 | { |
d6277db4 | 3556 | struct reclaim_state reclaim_state; |
d6277db4 | 3557 | struct scan_control sc = { |
ee814fe2 | 3558 | .nr_to_reclaim = nr_to_reclaim, |
7b51755c | 3559 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
ee814fe2 | 3560 | .priority = DEF_PRIORITY, |
d6277db4 | 3561 | .may_writepage = 1, |
ee814fe2 JW |
3562 | .may_unmap = 1, |
3563 | .may_swap = 1, | |
7b51755c | 3564 | .hibernation_mode = 1, |
1da177e4 | 3565 | }; |
a09ed5e0 | 3566 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); |
7b51755c KM |
3567 | struct task_struct *p = current; |
3568 | unsigned long nr_reclaimed; | |
1da177e4 | 3569 | |
7b51755c KM |
3570 | p->flags |= PF_MEMALLOC; |
3571 | lockdep_set_current_reclaim_state(sc.gfp_mask); | |
3572 | reclaim_state.reclaimed_slab = 0; | |
3573 | p->reclaim_state = &reclaim_state; | |
d6277db4 | 3574 | |
3115cd91 | 3575 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc); |
d979677c | 3576 | |
7b51755c KM |
3577 | p->reclaim_state = NULL; |
3578 | lockdep_clear_current_reclaim_state(); | |
3579 | p->flags &= ~PF_MEMALLOC; | |
d6277db4 | 3580 | |
7b51755c | 3581 | return nr_reclaimed; |
1da177e4 | 3582 | } |
c6f37f12 | 3583 | #endif /* CONFIG_HIBERNATION */ |
1da177e4 | 3584 | |
1da177e4 LT |
3585 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
3586 | not required for correctness. So if the last cpu in a node goes | |
3587 | away, we get changed to run anywhere: as the first one comes back, | |
3588 | restore their cpu bindings. */ | |
fcb35a9b GKH |
3589 | static int cpu_callback(struct notifier_block *nfb, unsigned long action, |
3590 | void *hcpu) | |
1da177e4 | 3591 | { |
58c0a4a7 | 3592 | int nid; |
1da177e4 | 3593 | |
8bb78442 | 3594 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
48fb2e24 | 3595 | for_each_node_state(nid, N_MEMORY) { |
c5f59f08 | 3596 | pg_data_t *pgdat = NODE_DATA(nid); |
a70f7302 RR |
3597 | const struct cpumask *mask; |
3598 | ||
3599 | mask = cpumask_of_node(pgdat->node_id); | |
c5f59f08 | 3600 | |
3e597945 | 3601 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
1da177e4 | 3602 | /* One of our CPUs online: restore mask */ |
c5f59f08 | 3603 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
1da177e4 LT |
3604 | } |
3605 | } | |
3606 | return NOTIFY_OK; | |
3607 | } | |
1da177e4 | 3608 | |
3218ae14 YG |
3609 | /* |
3610 | * This kswapd start function will be called by init and node-hot-add. | |
3611 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. | |
3612 | */ | |
3613 | int kswapd_run(int nid) | |
3614 | { | |
3615 | pg_data_t *pgdat = NODE_DATA(nid); | |
3616 | int ret = 0; | |
3617 | ||
3618 | if (pgdat->kswapd) | |
3619 | return 0; | |
3620 | ||
3621 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); | |
3622 | if (IS_ERR(pgdat->kswapd)) { | |
3623 | /* failure at boot is fatal */ | |
3624 | BUG_ON(system_state == SYSTEM_BOOTING); | |
d5dc0ad9 GS |
3625 | pr_err("Failed to start kswapd on node %d\n", nid); |
3626 | ret = PTR_ERR(pgdat->kswapd); | |
d72515b8 | 3627 | pgdat->kswapd = NULL; |
3218ae14 YG |
3628 | } |
3629 | return ret; | |
3630 | } | |
3631 | ||
8fe23e05 | 3632 | /* |
d8adde17 | 3633 | * Called by memory hotplug when all memory in a node is offlined. Caller must |
bfc8c901 | 3634 | * hold mem_hotplug_begin/end(). |
8fe23e05 DR |
3635 | */ |
3636 | void kswapd_stop(int nid) | |
3637 | { | |
3638 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; | |
3639 | ||
d8adde17 | 3640 | if (kswapd) { |
8fe23e05 | 3641 | kthread_stop(kswapd); |
d8adde17 JL |
3642 | NODE_DATA(nid)->kswapd = NULL; |
3643 | } | |
8fe23e05 DR |
3644 | } |
3645 | ||
1da177e4 LT |
3646 | static int __init kswapd_init(void) |
3647 | { | |
3218ae14 | 3648 | int nid; |
69e05944 | 3649 | |
1da177e4 | 3650 | swap_setup(); |
48fb2e24 | 3651 | for_each_node_state(nid, N_MEMORY) |
3218ae14 | 3652 | kswapd_run(nid); |
1da177e4 LT |
3653 | hotcpu_notifier(cpu_callback, 0); |
3654 | return 0; | |
3655 | } | |
3656 | ||
3657 | module_init(kswapd_init) | |
9eeff239 CL |
3658 | |
3659 | #ifdef CONFIG_NUMA | |
3660 | /* | |
3661 | * Zone reclaim mode | |
3662 | * | |
3663 | * If non-zero call zone_reclaim when the number of free pages falls below | |
3664 | * the watermarks. | |
9eeff239 CL |
3665 | */ |
3666 | int zone_reclaim_mode __read_mostly; | |
3667 | ||
1b2ffb78 | 3668 | #define RECLAIM_OFF 0 |
7d03431c | 3669 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
1b2ffb78 | 3670 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
95bbc0c7 | 3671 | #define RECLAIM_UNMAP (1<<2) /* Unmap pages during reclaim */ |
1b2ffb78 | 3672 | |
a92f7126 CL |
3673 | /* |
3674 | * Priority for ZONE_RECLAIM. This determines the fraction of pages | |
3675 | * of a node considered for each zone_reclaim. 4 scans 1/16th of | |
3676 | * a zone. | |
3677 | */ | |
3678 | #define ZONE_RECLAIM_PRIORITY 4 | |
3679 | ||
9614634f CL |
3680 | /* |
3681 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to | |
3682 | * occur. | |
3683 | */ | |
3684 | int sysctl_min_unmapped_ratio = 1; | |
3685 | ||
0ff38490 CL |
3686 | /* |
3687 | * If the number of slab pages in a zone grows beyond this percentage then | |
3688 | * slab reclaim needs to occur. | |
3689 | */ | |
3690 | int sysctl_min_slab_ratio = 5; | |
3691 | ||
90afa5de MG |
3692 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) |
3693 | { | |
3694 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); | |
3695 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + | |
3696 | zone_page_state(zone, NR_ACTIVE_FILE); | |
3697 | ||
3698 | /* | |
3699 | * It's possible for there to be more file mapped pages than | |
3700 | * accounted for by the pages on the file LRU lists because | |
3701 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED | |
3702 | */ | |
3703 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; | |
3704 | } | |
3705 | ||
3706 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ | |
d031a157 | 3707 | static unsigned long zone_pagecache_reclaimable(struct zone *zone) |
90afa5de | 3708 | { |
d031a157 AM |
3709 | unsigned long nr_pagecache_reclaimable; |
3710 | unsigned long delta = 0; | |
90afa5de MG |
3711 | |
3712 | /* | |
95bbc0c7 | 3713 | * If RECLAIM_UNMAP is set, then all file pages are considered |
90afa5de MG |
3714 | * potentially reclaimable. Otherwise, we have to worry about |
3715 | * pages like swapcache and zone_unmapped_file_pages() provides | |
3716 | * a better estimate | |
3717 | */ | |
95bbc0c7 | 3718 | if (zone_reclaim_mode & RECLAIM_UNMAP) |
90afa5de MG |
3719 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); |
3720 | else | |
3721 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); | |
3722 | ||
3723 | /* If we can't clean pages, remove dirty pages from consideration */ | |
3724 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) | |
3725 | delta += zone_page_state(zone, NR_FILE_DIRTY); | |
3726 | ||
3727 | /* Watch for any possible underflows due to delta */ | |
3728 | if (unlikely(delta > nr_pagecache_reclaimable)) | |
3729 | delta = nr_pagecache_reclaimable; | |
3730 | ||
3731 | return nr_pagecache_reclaimable - delta; | |
3732 | } | |
3733 | ||
9eeff239 CL |
3734 | /* |
3735 | * Try to free up some pages from this zone through reclaim. | |
3736 | */ | |
179e9639 | 3737 | static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
9eeff239 | 3738 | { |
7fb2d46d | 3739 | /* Minimum pages needed in order to stay on node */ |
69e05944 | 3740 | const unsigned long nr_pages = 1 << order; |
9eeff239 CL |
3741 | struct task_struct *p = current; |
3742 | struct reclaim_state reclaim_state; | |
179e9639 | 3743 | struct scan_control sc = { |
62b726c1 | 3744 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
21caf2fc | 3745 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
bd2f6199 | 3746 | .order = order, |
9e3b2f8c | 3747 | .priority = ZONE_RECLAIM_PRIORITY, |
ee814fe2 | 3748 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
95bbc0c7 | 3749 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_UNMAP), |
ee814fe2 | 3750 | .may_swap = 1, |
179e9639 | 3751 | }; |
9eeff239 | 3752 | |
9eeff239 | 3753 | cond_resched(); |
d4f7796e | 3754 | /* |
95bbc0c7 | 3755 | * We need to be able to allocate from the reserves for RECLAIM_UNMAP |
d4f7796e | 3756 | * and we also need to be able to write out pages for RECLAIM_WRITE |
95bbc0c7 | 3757 | * and RECLAIM_UNMAP. |
d4f7796e CL |
3758 | */ |
3759 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; | |
76ca542d | 3760 | lockdep_set_current_reclaim_state(gfp_mask); |
9eeff239 CL |
3761 | reclaim_state.reclaimed_slab = 0; |
3762 | p->reclaim_state = &reclaim_state; | |
c84db23c | 3763 | |
90afa5de | 3764 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { |
0ff38490 CL |
3765 | /* |
3766 | * Free memory by calling shrink zone with increasing | |
3767 | * priorities until we have enough memory freed. | |
3768 | */ | |
0ff38490 | 3769 | do { |
6b4f7799 | 3770 | shrink_zone(zone, &sc, true); |
9e3b2f8c | 3771 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); |
0ff38490 | 3772 | } |
c84db23c | 3773 | |
9eeff239 | 3774 | p->reclaim_state = NULL; |
d4f7796e | 3775 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
76ca542d | 3776 | lockdep_clear_current_reclaim_state(); |
a79311c1 | 3777 | return sc.nr_reclaimed >= nr_pages; |
9eeff239 | 3778 | } |
179e9639 AM |
3779 | |
3780 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | |
3781 | { | |
179e9639 | 3782 | int node_id; |
d773ed6b | 3783 | int ret; |
179e9639 AM |
3784 | |
3785 | /* | |
0ff38490 CL |
3786 | * Zone reclaim reclaims unmapped file backed pages and |
3787 | * slab pages if we are over the defined limits. | |
34aa1330 | 3788 | * |
9614634f CL |
3789 | * A small portion of unmapped file backed pages is needed for |
3790 | * file I/O otherwise pages read by file I/O will be immediately | |
3791 | * thrown out if the zone is overallocated. So we do not reclaim | |
3792 | * if less than a specified percentage of the zone is used by | |
3793 | * unmapped file backed pages. | |
179e9639 | 3794 | */ |
90afa5de MG |
3795 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && |
3796 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) | |
fa5e084e | 3797 | return ZONE_RECLAIM_FULL; |
179e9639 | 3798 | |
6e543d57 | 3799 | if (!zone_reclaimable(zone)) |
fa5e084e | 3800 | return ZONE_RECLAIM_FULL; |
d773ed6b | 3801 | |
179e9639 | 3802 | /* |
d773ed6b | 3803 | * Do not scan if the allocation should not be delayed. |
179e9639 | 3804 | */ |
d0164adc | 3805 | if (!gfpflags_allow_blocking(gfp_mask) || (current->flags & PF_MEMALLOC)) |
fa5e084e | 3806 | return ZONE_RECLAIM_NOSCAN; |
179e9639 AM |
3807 | |
3808 | /* | |
3809 | * Only run zone reclaim on the local zone or on zones that do not | |
3810 | * have associated processors. This will favor the local processor | |
3811 | * over remote processors and spread off node memory allocations | |
3812 | * as wide as possible. | |
3813 | */ | |
89fa3024 | 3814 | node_id = zone_to_nid(zone); |
37c0708d | 3815 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
fa5e084e | 3816 | return ZONE_RECLAIM_NOSCAN; |
d773ed6b | 3817 | |
57054651 | 3818 | if (test_and_set_bit(ZONE_RECLAIM_LOCKED, &zone->flags)) |
fa5e084e MG |
3819 | return ZONE_RECLAIM_NOSCAN; |
3820 | ||
d773ed6b | 3821 | ret = __zone_reclaim(zone, gfp_mask, order); |
57054651 | 3822 | clear_bit(ZONE_RECLAIM_LOCKED, &zone->flags); |
d773ed6b | 3823 | |
24cf7251 MG |
3824 | if (!ret) |
3825 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); | |
3826 | ||
d773ed6b | 3827 | return ret; |
179e9639 | 3828 | } |
9eeff239 | 3829 | #endif |
894bc310 | 3830 | |
894bc310 LS |
3831 | /* |
3832 | * page_evictable - test whether a page is evictable | |
3833 | * @page: the page to test | |
894bc310 LS |
3834 | * |
3835 | * Test whether page is evictable--i.e., should be placed on active/inactive | |
39b5f29a | 3836 | * lists vs unevictable list. |
894bc310 LS |
3837 | * |
3838 | * Reasons page might not be evictable: | |
ba9ddf49 | 3839 | * (1) page's mapping marked unevictable |
b291f000 | 3840 | * (2) page is part of an mlocked VMA |
ba9ddf49 | 3841 | * |
894bc310 | 3842 | */ |
39b5f29a | 3843 | int page_evictable(struct page *page) |
894bc310 | 3844 | { |
39b5f29a | 3845 | return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); |
894bc310 | 3846 | } |
89e004ea | 3847 | |
85046579 | 3848 | #ifdef CONFIG_SHMEM |
89e004ea | 3849 | /** |
24513264 HD |
3850 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list |
3851 | * @pages: array of pages to check | |
3852 | * @nr_pages: number of pages to check | |
89e004ea | 3853 | * |
24513264 | 3854 | * Checks pages for evictability and moves them to the appropriate lru list. |
85046579 HD |
3855 | * |
3856 | * This function is only used for SysV IPC SHM_UNLOCK. | |
89e004ea | 3857 | */ |
24513264 | 3858 | void check_move_unevictable_pages(struct page **pages, int nr_pages) |
89e004ea | 3859 | { |
925b7673 | 3860 | struct lruvec *lruvec; |
24513264 HD |
3861 | struct zone *zone = NULL; |
3862 | int pgscanned = 0; | |
3863 | int pgrescued = 0; | |
3864 | int i; | |
89e004ea | 3865 | |
24513264 HD |
3866 | for (i = 0; i < nr_pages; i++) { |
3867 | struct page *page = pages[i]; | |
3868 | struct zone *pagezone; | |
89e004ea | 3869 | |
24513264 HD |
3870 | pgscanned++; |
3871 | pagezone = page_zone(page); | |
3872 | if (pagezone != zone) { | |
3873 | if (zone) | |
3874 | spin_unlock_irq(&zone->lru_lock); | |
3875 | zone = pagezone; | |
3876 | spin_lock_irq(&zone->lru_lock); | |
3877 | } | |
fa9add64 | 3878 | lruvec = mem_cgroup_page_lruvec(page, zone); |
89e004ea | 3879 | |
24513264 HD |
3880 | if (!PageLRU(page) || !PageUnevictable(page)) |
3881 | continue; | |
89e004ea | 3882 | |
39b5f29a | 3883 | if (page_evictable(page)) { |
24513264 HD |
3884 | enum lru_list lru = page_lru_base_type(page); |
3885 | ||
309381fe | 3886 | VM_BUG_ON_PAGE(PageActive(page), page); |
24513264 | 3887 | ClearPageUnevictable(page); |
fa9add64 HD |
3888 | del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE); |
3889 | add_page_to_lru_list(page, lruvec, lru); | |
24513264 | 3890 | pgrescued++; |
89e004ea | 3891 | } |
24513264 | 3892 | } |
89e004ea | 3893 | |
24513264 HD |
3894 | if (zone) { |
3895 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | |
3896 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); | |
3897 | spin_unlock_irq(&zone->lru_lock); | |
89e004ea | 3898 | } |
89e004ea | 3899 | } |
85046579 | 3900 | #endif /* CONFIG_SHMEM */ |