Commit 04bab05a95fece32015d897d4058880bbb5c65eb
Committed by
Jiri Slaby
1 parent
788a2f69f9
Exists in
ti-linux-3.12.y
and in
2 other branches
memcg, vmscan: Fix forced scan of anonymous pages
commit 2ab051e11bfa3cbb7b24177f3d6aaed10a0d743e upstream. When memory cgoups are enabled, the code that decides to force to scan anonymous pages in get_scan_count() compares global values (free, high_watermark) to a value that is restricted to a memory cgroup (file). It make the code over-eager to force anon scan. For instance, it will force anon scan when scanning a memcg that is mainly populated by anonymous page, even when there is plenty of file pages to get rid of in others memcgs, even when swappiness == 0. It breaks user's expectation about swappiness and hurts performance. This patch makes sure that forced anon scan only happens when there not enough file pages for the all zone, not just in one random memcg. [hannes@cmpxchg.org: cleanups] Signed-off-by: Jerome Marchand <jmarchan@redhat.com> Acked-by: Michal Hocko <mhocko@suse.cz> Acked-by: Johannes Weiner <hannes@cmpxchg.org> Reviewed-by: Rik van Riel <riel@redhat.com> Cc: Mel Gorman <mgorman@suse.de> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Mel Gorman <mgorman@suse.de> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Showing 1 changed file with 15 additions and 8 deletions Inline Diff
mm/vmscan.c
1 | /* | 1 | /* |
2 | * linux/mm/vmscan.c | 2 | * linux/mm/vmscan.c |
3 | * | 3 | * |
4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds | 4 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
5 | * | 5 | * |
6 | * Swap reorganised 29.12.95, Stephen Tweedie. | 6 | * Swap reorganised 29.12.95, Stephen Tweedie. |
7 | * kswapd added: 7.1.96 sct | 7 | * kswapd added: 7.1.96 sct |
8 | * Removed kswapd_ctl limits, and swap out as many pages as needed | 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. | 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). | 10 | * Zone aware kswapd started 02/00, Kanoj Sarcar (kanoj@sgi.com). |
11 | * Multiqueue VM started 5.8.00, Rik van Riel. | 11 | * Multiqueue VM started 5.8.00, Rik van Riel. |
12 | */ | 12 | */ |
13 | 13 | ||
14 | #include <linux/mm.h> | 14 | #include <linux/mm.h> |
15 | #include <linux/module.h> | 15 | #include <linux/module.h> |
16 | #include <linux/gfp.h> | 16 | #include <linux/gfp.h> |
17 | #include <linux/kernel_stat.h> | 17 | #include <linux/kernel_stat.h> |
18 | #include <linux/swap.h> | 18 | #include <linux/swap.h> |
19 | #include <linux/pagemap.h> | 19 | #include <linux/pagemap.h> |
20 | #include <linux/init.h> | 20 | #include <linux/init.h> |
21 | #include <linux/highmem.h> | 21 | #include <linux/highmem.h> |
22 | #include <linux/vmpressure.h> | 22 | #include <linux/vmpressure.h> |
23 | #include <linux/vmstat.h> | 23 | #include <linux/vmstat.h> |
24 | #include <linux/file.h> | 24 | #include <linux/file.h> |
25 | #include <linux/writeback.h> | 25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | 26 | #include <linux/blkdev.h> |
27 | #include <linux/buffer_head.h> /* for try_to_release_page(), | 27 | #include <linux/buffer_head.h> /* for try_to_release_page(), |
28 | buffer_heads_over_limit */ | 28 | buffer_heads_over_limit */ |
29 | #include <linux/mm_inline.h> | 29 | #include <linux/mm_inline.h> |
30 | #include <linux/backing-dev.h> | 30 | #include <linux/backing-dev.h> |
31 | #include <linux/rmap.h> | 31 | #include <linux/rmap.h> |
32 | #include <linux/topology.h> | 32 | #include <linux/topology.h> |
33 | #include <linux/cpu.h> | 33 | #include <linux/cpu.h> |
34 | #include <linux/cpuset.h> | 34 | #include <linux/cpuset.h> |
35 | #include <linux/compaction.h> | 35 | #include <linux/compaction.h> |
36 | #include <linux/notifier.h> | 36 | #include <linux/notifier.h> |
37 | #include <linux/rwsem.h> | 37 | #include <linux/rwsem.h> |
38 | #include <linux/delay.h> | 38 | #include <linux/delay.h> |
39 | #include <linux/kthread.h> | 39 | #include <linux/kthread.h> |
40 | #include <linux/freezer.h> | 40 | #include <linux/freezer.h> |
41 | #include <linux/memcontrol.h> | 41 | #include <linux/memcontrol.h> |
42 | #include <linux/delayacct.h> | 42 | #include <linux/delayacct.h> |
43 | #include <linux/sysctl.h> | 43 | #include <linux/sysctl.h> |
44 | #include <linux/oom.h> | 44 | #include <linux/oom.h> |
45 | #include <linux/prefetch.h> | 45 | #include <linux/prefetch.h> |
46 | 46 | ||
47 | #include <asm/tlbflush.h> | 47 | #include <asm/tlbflush.h> |
48 | #include <asm/div64.h> | 48 | #include <asm/div64.h> |
49 | 49 | ||
50 | #include <linux/swapops.h> | 50 | #include <linux/swapops.h> |
51 | #include <linux/balloon_compaction.h> | 51 | #include <linux/balloon_compaction.h> |
52 | 52 | ||
53 | #include "internal.h" | 53 | #include "internal.h" |
54 | 54 | ||
55 | #define CREATE_TRACE_POINTS | 55 | #define CREATE_TRACE_POINTS |
56 | #include <trace/events/vmscan.h> | 56 | #include <trace/events/vmscan.h> |
57 | 57 | ||
58 | struct scan_control { | 58 | struct scan_control { |
59 | /* Incremented by the number of inactive pages that were scanned */ | 59 | /* Incremented by the number of inactive pages that were scanned */ |
60 | unsigned long nr_scanned; | 60 | unsigned long nr_scanned; |
61 | 61 | ||
62 | /* Number of pages freed so far during a call to shrink_zones() */ | 62 | /* Number of pages freed so far during a call to shrink_zones() */ |
63 | unsigned long nr_reclaimed; | 63 | unsigned long nr_reclaimed; |
64 | 64 | ||
65 | /* How many pages shrink_list() should reclaim */ | 65 | /* How many pages shrink_list() should reclaim */ |
66 | unsigned long nr_to_reclaim; | 66 | unsigned long nr_to_reclaim; |
67 | 67 | ||
68 | unsigned long hibernation_mode; | 68 | unsigned long hibernation_mode; |
69 | 69 | ||
70 | /* This context's GFP mask */ | 70 | /* This context's GFP mask */ |
71 | gfp_t gfp_mask; | 71 | gfp_t gfp_mask; |
72 | 72 | ||
73 | int may_writepage; | 73 | int may_writepage; |
74 | 74 | ||
75 | /* Can mapped pages be reclaimed? */ | 75 | /* Can mapped pages be reclaimed? */ |
76 | int may_unmap; | 76 | int may_unmap; |
77 | 77 | ||
78 | /* Can pages be swapped as part of reclaim? */ | 78 | /* Can pages be swapped as part of reclaim? */ |
79 | int may_swap; | 79 | int may_swap; |
80 | 80 | ||
81 | int order; | 81 | int order; |
82 | 82 | ||
83 | /* Scan (total_size >> priority) pages at once */ | 83 | /* Scan (total_size >> priority) pages at once */ |
84 | int priority; | 84 | int priority; |
85 | 85 | ||
86 | /* | 86 | /* |
87 | * The memory cgroup that hit its limit and as a result is the | 87 | * The memory cgroup that hit its limit and as a result is the |
88 | * primary target of this reclaim invocation. | 88 | * primary target of this reclaim invocation. |
89 | */ | 89 | */ |
90 | struct mem_cgroup *target_mem_cgroup; | 90 | struct mem_cgroup *target_mem_cgroup; |
91 | 91 | ||
92 | /* | 92 | /* |
93 | * Nodemask of nodes allowed by the caller. If NULL, all nodes | 93 | * Nodemask of nodes allowed by the caller. If NULL, all nodes |
94 | * are scanned. | 94 | * are scanned. |
95 | */ | 95 | */ |
96 | nodemask_t *nodemask; | 96 | nodemask_t *nodemask; |
97 | }; | 97 | }; |
98 | 98 | ||
99 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) | 99 | #define lru_to_page(_head) (list_entry((_head)->prev, struct page, lru)) |
100 | 100 | ||
101 | #ifdef ARCH_HAS_PREFETCH | 101 | #ifdef ARCH_HAS_PREFETCH |
102 | #define prefetch_prev_lru_page(_page, _base, _field) \ | 102 | #define prefetch_prev_lru_page(_page, _base, _field) \ |
103 | do { \ | 103 | do { \ |
104 | if ((_page)->lru.prev != _base) { \ | 104 | if ((_page)->lru.prev != _base) { \ |
105 | struct page *prev; \ | 105 | struct page *prev; \ |
106 | \ | 106 | \ |
107 | prev = lru_to_page(&(_page->lru)); \ | 107 | prev = lru_to_page(&(_page->lru)); \ |
108 | prefetch(&prev->_field); \ | 108 | prefetch(&prev->_field); \ |
109 | } \ | 109 | } \ |
110 | } while (0) | 110 | } while (0) |
111 | #else | 111 | #else |
112 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) | 112 | #define prefetch_prev_lru_page(_page, _base, _field) do { } while (0) |
113 | #endif | 113 | #endif |
114 | 114 | ||
115 | #ifdef ARCH_HAS_PREFETCHW | 115 | #ifdef ARCH_HAS_PREFETCHW |
116 | #define prefetchw_prev_lru_page(_page, _base, _field) \ | 116 | #define prefetchw_prev_lru_page(_page, _base, _field) \ |
117 | do { \ | 117 | do { \ |
118 | if ((_page)->lru.prev != _base) { \ | 118 | if ((_page)->lru.prev != _base) { \ |
119 | struct page *prev; \ | 119 | struct page *prev; \ |
120 | \ | 120 | \ |
121 | prev = lru_to_page(&(_page->lru)); \ | 121 | prev = lru_to_page(&(_page->lru)); \ |
122 | prefetchw(&prev->_field); \ | 122 | prefetchw(&prev->_field); \ |
123 | } \ | 123 | } \ |
124 | } while (0) | 124 | } while (0) |
125 | #else | 125 | #else |
126 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) | 126 | #define prefetchw_prev_lru_page(_page, _base, _field) do { } while (0) |
127 | #endif | 127 | #endif |
128 | 128 | ||
129 | /* | 129 | /* |
130 | * From 0 .. 100. Higher means more swappy. | 130 | * From 0 .. 100. Higher means more swappy. |
131 | */ | 131 | */ |
132 | int vm_swappiness = 60; | 132 | int vm_swappiness = 60; |
133 | unsigned long vm_total_pages; /* The total number of pages which the VM controls */ | 133 | unsigned long vm_total_pages; /* The total number of pages which the VM controls */ |
134 | 134 | ||
135 | static LIST_HEAD(shrinker_list); | 135 | static LIST_HEAD(shrinker_list); |
136 | static DECLARE_RWSEM(shrinker_rwsem); | 136 | static DECLARE_RWSEM(shrinker_rwsem); |
137 | 137 | ||
138 | #ifdef CONFIG_MEMCG | 138 | #ifdef CONFIG_MEMCG |
139 | static bool global_reclaim(struct scan_control *sc) | 139 | static bool global_reclaim(struct scan_control *sc) |
140 | { | 140 | { |
141 | return !sc->target_mem_cgroup; | 141 | return !sc->target_mem_cgroup; |
142 | } | 142 | } |
143 | #else | 143 | #else |
144 | static bool global_reclaim(struct scan_control *sc) | 144 | static bool global_reclaim(struct scan_control *sc) |
145 | { | 145 | { |
146 | return true; | 146 | return true; |
147 | } | 147 | } |
148 | #endif | 148 | #endif |
149 | 149 | ||
150 | static unsigned long zone_reclaimable_pages(struct zone *zone) | 150 | static unsigned long zone_reclaimable_pages(struct zone *zone) |
151 | { | 151 | { |
152 | int nr; | 152 | int nr; |
153 | 153 | ||
154 | nr = zone_page_state(zone, NR_ACTIVE_FILE) + | 154 | nr = zone_page_state(zone, NR_ACTIVE_FILE) + |
155 | zone_page_state(zone, NR_INACTIVE_FILE); | 155 | zone_page_state(zone, NR_INACTIVE_FILE); |
156 | 156 | ||
157 | if (get_nr_swap_pages() > 0) | 157 | if (get_nr_swap_pages() > 0) |
158 | nr += zone_page_state(zone, NR_ACTIVE_ANON) + | 158 | nr += zone_page_state(zone, NR_ACTIVE_ANON) + |
159 | zone_page_state(zone, NR_INACTIVE_ANON); | 159 | zone_page_state(zone, NR_INACTIVE_ANON); |
160 | 160 | ||
161 | return nr; | 161 | return nr; |
162 | } | 162 | } |
163 | 163 | ||
164 | bool zone_reclaimable(struct zone *zone) | 164 | bool zone_reclaimable(struct zone *zone) |
165 | { | 165 | { |
166 | return zone->pages_scanned < zone_reclaimable_pages(zone) * 6; | 166 | return zone->pages_scanned < zone_reclaimable_pages(zone) * 6; |
167 | } | 167 | } |
168 | 168 | ||
169 | static unsigned long get_lru_size(struct lruvec *lruvec, enum lru_list lru) | 169 | static unsigned long get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
170 | { | 170 | { |
171 | if (!mem_cgroup_disabled()) | 171 | if (!mem_cgroup_disabled()) |
172 | return mem_cgroup_get_lru_size(lruvec, lru); | 172 | return mem_cgroup_get_lru_size(lruvec, lru); |
173 | 173 | ||
174 | return zone_page_state(lruvec_zone(lruvec), NR_LRU_BASE + lru); | 174 | return zone_page_state(lruvec_zone(lruvec), NR_LRU_BASE + lru); |
175 | } | 175 | } |
176 | 176 | ||
177 | /* | 177 | /* |
178 | * Add a shrinker callback to be called from the vm. | 178 | * Add a shrinker callback to be called from the vm. |
179 | */ | 179 | */ |
180 | int register_shrinker(struct shrinker *shrinker) | 180 | int register_shrinker(struct shrinker *shrinker) |
181 | { | 181 | { |
182 | size_t size = sizeof(*shrinker->nr_deferred); | 182 | size_t size = sizeof(*shrinker->nr_deferred); |
183 | 183 | ||
184 | /* | 184 | /* |
185 | * If we only have one possible node in the system anyway, save | 185 | * If we only have one possible node in the system anyway, save |
186 | * ourselves the trouble and disable NUMA aware behavior. This way we | 186 | * ourselves the trouble and disable NUMA aware behavior. This way we |
187 | * will save memory and some small loop time later. | 187 | * will save memory and some small loop time later. |
188 | */ | 188 | */ |
189 | if (nr_node_ids == 1) | 189 | if (nr_node_ids == 1) |
190 | shrinker->flags &= ~SHRINKER_NUMA_AWARE; | 190 | shrinker->flags &= ~SHRINKER_NUMA_AWARE; |
191 | 191 | ||
192 | if (shrinker->flags & SHRINKER_NUMA_AWARE) | 192 | if (shrinker->flags & SHRINKER_NUMA_AWARE) |
193 | size *= nr_node_ids; | 193 | size *= nr_node_ids; |
194 | 194 | ||
195 | shrinker->nr_deferred = kzalloc(size, GFP_KERNEL); | 195 | shrinker->nr_deferred = kzalloc(size, GFP_KERNEL); |
196 | if (!shrinker->nr_deferred) | 196 | if (!shrinker->nr_deferred) |
197 | return -ENOMEM; | 197 | return -ENOMEM; |
198 | 198 | ||
199 | down_write(&shrinker_rwsem); | 199 | down_write(&shrinker_rwsem); |
200 | list_add_tail(&shrinker->list, &shrinker_list); | 200 | list_add_tail(&shrinker->list, &shrinker_list); |
201 | up_write(&shrinker_rwsem); | 201 | up_write(&shrinker_rwsem); |
202 | return 0; | 202 | return 0; |
203 | } | 203 | } |
204 | EXPORT_SYMBOL(register_shrinker); | 204 | EXPORT_SYMBOL(register_shrinker); |
205 | 205 | ||
206 | /* | 206 | /* |
207 | * Remove one | 207 | * Remove one |
208 | */ | 208 | */ |
209 | void unregister_shrinker(struct shrinker *shrinker) | 209 | void unregister_shrinker(struct shrinker *shrinker) |
210 | { | 210 | { |
211 | down_write(&shrinker_rwsem); | 211 | down_write(&shrinker_rwsem); |
212 | list_del(&shrinker->list); | 212 | list_del(&shrinker->list); |
213 | up_write(&shrinker_rwsem); | 213 | up_write(&shrinker_rwsem); |
214 | kfree(shrinker->nr_deferred); | 214 | kfree(shrinker->nr_deferred); |
215 | } | 215 | } |
216 | EXPORT_SYMBOL(unregister_shrinker); | 216 | EXPORT_SYMBOL(unregister_shrinker); |
217 | 217 | ||
218 | #define SHRINK_BATCH 128 | 218 | #define SHRINK_BATCH 128 |
219 | 219 | ||
220 | static unsigned long | 220 | static unsigned long |
221 | shrink_slab_node(struct shrink_control *shrinkctl, struct shrinker *shrinker, | 221 | shrink_slab_node(struct shrink_control *shrinkctl, struct shrinker *shrinker, |
222 | unsigned long nr_pages_scanned, unsigned long lru_pages) | 222 | unsigned long nr_pages_scanned, unsigned long lru_pages) |
223 | { | 223 | { |
224 | unsigned long freed = 0; | 224 | unsigned long freed = 0; |
225 | unsigned long long delta; | 225 | unsigned long long delta; |
226 | long total_scan; | 226 | long total_scan; |
227 | long freeable; | 227 | long freeable; |
228 | long nr; | 228 | long nr; |
229 | long new_nr; | 229 | long new_nr; |
230 | int nid = shrinkctl->nid; | 230 | int nid = shrinkctl->nid; |
231 | long batch_size = shrinker->batch ? shrinker->batch | 231 | long batch_size = shrinker->batch ? shrinker->batch |
232 | : SHRINK_BATCH; | 232 | : SHRINK_BATCH; |
233 | 233 | ||
234 | freeable = shrinker->count_objects(shrinker, shrinkctl); | 234 | freeable = shrinker->count_objects(shrinker, shrinkctl); |
235 | if (freeable == 0) | 235 | if (freeable == 0) |
236 | return 0; | 236 | return 0; |
237 | 237 | ||
238 | /* | 238 | /* |
239 | * copy the current shrinker scan count into a local variable | 239 | * copy the current shrinker scan count into a local variable |
240 | * and zero it so that other concurrent shrinker invocations | 240 | * and zero it so that other concurrent shrinker invocations |
241 | * don't also do this scanning work. | 241 | * don't also do this scanning work. |
242 | */ | 242 | */ |
243 | nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0); | 243 | nr = atomic_long_xchg(&shrinker->nr_deferred[nid], 0); |
244 | 244 | ||
245 | total_scan = nr; | 245 | total_scan = nr; |
246 | delta = (4 * nr_pages_scanned) / shrinker->seeks; | 246 | delta = (4 * nr_pages_scanned) / shrinker->seeks; |
247 | delta *= freeable; | 247 | delta *= freeable; |
248 | do_div(delta, lru_pages + 1); | 248 | do_div(delta, lru_pages + 1); |
249 | total_scan += delta; | 249 | total_scan += delta; |
250 | if (total_scan < 0) { | 250 | if (total_scan < 0) { |
251 | printk(KERN_ERR | 251 | printk(KERN_ERR |
252 | "shrink_slab: %pF negative objects to delete nr=%ld\n", | 252 | "shrink_slab: %pF negative objects to delete nr=%ld\n", |
253 | shrinker->scan_objects, total_scan); | 253 | shrinker->scan_objects, total_scan); |
254 | total_scan = freeable; | 254 | total_scan = freeable; |
255 | } | 255 | } |
256 | 256 | ||
257 | /* | 257 | /* |
258 | * We need to avoid excessive windup on filesystem shrinkers | 258 | * We need to avoid excessive windup on filesystem shrinkers |
259 | * due to large numbers of GFP_NOFS allocations causing the | 259 | * due to large numbers of GFP_NOFS allocations causing the |
260 | * shrinkers to return -1 all the time. This results in a large | 260 | * shrinkers to return -1 all the time. This results in a large |
261 | * nr being built up so when a shrink that can do some work | 261 | * nr being built up so when a shrink that can do some work |
262 | * comes along it empties the entire cache due to nr >>> | 262 | * comes along it empties the entire cache due to nr >>> |
263 | * freeable. This is bad for sustaining a working set in | 263 | * freeable. This is bad for sustaining a working set in |
264 | * memory. | 264 | * memory. |
265 | * | 265 | * |
266 | * Hence only allow the shrinker to scan the entire cache when | 266 | * Hence only allow the shrinker to scan the entire cache when |
267 | * a large delta change is calculated directly. | 267 | * a large delta change is calculated directly. |
268 | */ | 268 | */ |
269 | if (delta < freeable / 4) | 269 | if (delta < freeable / 4) |
270 | total_scan = min(total_scan, freeable / 2); | 270 | total_scan = min(total_scan, freeable / 2); |
271 | 271 | ||
272 | /* | 272 | /* |
273 | * Avoid risking looping forever due to too large nr value: | 273 | * Avoid risking looping forever due to too large nr value: |
274 | * never try to free more than twice the estimate number of | 274 | * never try to free more than twice the estimate number of |
275 | * freeable entries. | 275 | * freeable entries. |
276 | */ | 276 | */ |
277 | if (total_scan > freeable * 2) | 277 | if (total_scan > freeable * 2) |
278 | total_scan = freeable * 2; | 278 | total_scan = freeable * 2; |
279 | 279 | ||
280 | trace_mm_shrink_slab_start(shrinker, shrinkctl, nr, | 280 | trace_mm_shrink_slab_start(shrinker, shrinkctl, nr, |
281 | nr_pages_scanned, lru_pages, | 281 | nr_pages_scanned, lru_pages, |
282 | freeable, delta, total_scan); | 282 | freeable, delta, total_scan); |
283 | 283 | ||
284 | /* | 284 | /* |
285 | * Normally, we should not scan less than batch_size objects in one | 285 | * Normally, we should not scan less than batch_size objects in one |
286 | * pass to avoid too frequent shrinker calls, but if the slab has less | 286 | * pass to avoid too frequent shrinker calls, but if the slab has less |
287 | * than batch_size objects in total and we are really tight on memory, | 287 | * than batch_size objects in total and we are really tight on memory, |
288 | * we will try to reclaim all available objects, otherwise we can end | 288 | * we will try to reclaim all available objects, otherwise we can end |
289 | * up failing allocations although there are plenty of reclaimable | 289 | * up failing allocations although there are plenty of reclaimable |
290 | * objects spread over several slabs with usage less than the | 290 | * objects spread over several slabs with usage less than the |
291 | * batch_size. | 291 | * batch_size. |
292 | * | 292 | * |
293 | * We detect the "tight on memory" situations by looking at the total | 293 | * We detect the "tight on memory" situations by looking at the total |
294 | * number of objects we want to scan (total_scan). If it is greater | 294 | * number of objects we want to scan (total_scan). If it is greater |
295 | * than the total number of objects on slab (freeable), we must be | 295 | * than the total number of objects on slab (freeable), we must be |
296 | * scanning at high prio and therefore should try to reclaim as much as | 296 | * scanning at high prio and therefore should try to reclaim as much as |
297 | * possible. | 297 | * possible. |
298 | */ | 298 | */ |
299 | while (total_scan >= batch_size || | 299 | while (total_scan >= batch_size || |
300 | total_scan >= freeable) { | 300 | total_scan >= freeable) { |
301 | unsigned long ret; | 301 | unsigned long ret; |
302 | unsigned long nr_to_scan = min(batch_size, total_scan); | 302 | unsigned long nr_to_scan = min(batch_size, total_scan); |
303 | 303 | ||
304 | shrinkctl->nr_to_scan = nr_to_scan; | 304 | shrinkctl->nr_to_scan = nr_to_scan; |
305 | ret = shrinker->scan_objects(shrinker, shrinkctl); | 305 | ret = shrinker->scan_objects(shrinker, shrinkctl); |
306 | if (ret == SHRINK_STOP) | 306 | if (ret == SHRINK_STOP) |
307 | break; | 307 | break; |
308 | freed += ret; | 308 | freed += ret; |
309 | 309 | ||
310 | count_vm_events(SLABS_SCANNED, nr_to_scan); | 310 | count_vm_events(SLABS_SCANNED, nr_to_scan); |
311 | total_scan -= nr_to_scan; | 311 | total_scan -= nr_to_scan; |
312 | 312 | ||
313 | cond_resched(); | 313 | cond_resched(); |
314 | } | 314 | } |
315 | 315 | ||
316 | /* | 316 | /* |
317 | * move the unused scan count back into the shrinker in a | 317 | * move the unused scan count back into the shrinker in a |
318 | * manner that handles concurrent updates. If we exhausted the | 318 | * manner that handles concurrent updates. If we exhausted the |
319 | * scan, there is no need to do an update. | 319 | * scan, there is no need to do an update. |
320 | */ | 320 | */ |
321 | if (total_scan > 0) | 321 | if (total_scan > 0) |
322 | new_nr = atomic_long_add_return(total_scan, | 322 | new_nr = atomic_long_add_return(total_scan, |
323 | &shrinker->nr_deferred[nid]); | 323 | &shrinker->nr_deferred[nid]); |
324 | else | 324 | else |
325 | new_nr = atomic_long_read(&shrinker->nr_deferred[nid]); | 325 | new_nr = atomic_long_read(&shrinker->nr_deferred[nid]); |
326 | 326 | ||
327 | trace_mm_shrink_slab_end(shrinker, freed, nr, new_nr); | 327 | trace_mm_shrink_slab_end(shrinker, freed, nr, new_nr); |
328 | return freed; | 328 | return freed; |
329 | } | 329 | } |
330 | 330 | ||
331 | /* | 331 | /* |
332 | * Call the shrink functions to age shrinkable caches | 332 | * Call the shrink functions to age shrinkable caches |
333 | * | 333 | * |
334 | * Here we assume it costs one seek to replace a lru page and that it also | 334 | * Here we assume it costs one seek to replace a lru page and that it also |
335 | * takes a seek to recreate a cache object. With this in mind we age equal | 335 | * takes a seek to recreate a cache object. With this in mind we age equal |
336 | * percentages of the lru and ageable caches. This should balance the seeks | 336 | * percentages of the lru and ageable caches. This should balance the seeks |
337 | * generated by these structures. | 337 | * generated by these structures. |
338 | * | 338 | * |
339 | * If the vm encountered mapped pages on the LRU it increase the pressure on | 339 | * If the vm encountered mapped pages on the LRU it increase the pressure on |
340 | * slab to avoid swapping. | 340 | * slab to avoid swapping. |
341 | * | 341 | * |
342 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. | 342 | * We do weird things to avoid (scanned*seeks*entries) overflowing 32 bits. |
343 | * | 343 | * |
344 | * `lru_pages' represents the number of on-LRU pages in all the zones which | 344 | * `lru_pages' represents the number of on-LRU pages in all the zones which |
345 | * are eligible for the caller's allocation attempt. It is used for balancing | 345 | * are eligible for the caller's allocation attempt. It is used for balancing |
346 | * slab reclaim versus page reclaim. | 346 | * slab reclaim versus page reclaim. |
347 | * | 347 | * |
348 | * Returns the number of slab objects which we shrunk. | 348 | * Returns the number of slab objects which we shrunk. |
349 | */ | 349 | */ |
350 | unsigned long shrink_slab(struct shrink_control *shrinkctl, | 350 | unsigned long shrink_slab(struct shrink_control *shrinkctl, |
351 | unsigned long nr_pages_scanned, | 351 | unsigned long nr_pages_scanned, |
352 | unsigned long lru_pages) | 352 | unsigned long lru_pages) |
353 | { | 353 | { |
354 | struct shrinker *shrinker; | 354 | struct shrinker *shrinker; |
355 | unsigned long freed = 0; | 355 | unsigned long freed = 0; |
356 | 356 | ||
357 | if (nr_pages_scanned == 0) | 357 | if (nr_pages_scanned == 0) |
358 | nr_pages_scanned = SWAP_CLUSTER_MAX; | 358 | nr_pages_scanned = SWAP_CLUSTER_MAX; |
359 | 359 | ||
360 | if (!down_read_trylock(&shrinker_rwsem)) { | 360 | if (!down_read_trylock(&shrinker_rwsem)) { |
361 | /* | 361 | /* |
362 | * If we would return 0, our callers would understand that we | 362 | * If we would return 0, our callers would understand that we |
363 | * have nothing else to shrink and give up trying. By returning | 363 | * have nothing else to shrink and give up trying. By returning |
364 | * 1 we keep it going and assume we'll be able to shrink next | 364 | * 1 we keep it going and assume we'll be able to shrink next |
365 | * time. | 365 | * time. |
366 | */ | 366 | */ |
367 | freed = 1; | 367 | freed = 1; |
368 | goto out; | 368 | goto out; |
369 | } | 369 | } |
370 | 370 | ||
371 | list_for_each_entry(shrinker, &shrinker_list, list) { | 371 | list_for_each_entry(shrinker, &shrinker_list, list) { |
372 | if (!(shrinker->flags & SHRINKER_NUMA_AWARE)) { | 372 | if (!(shrinker->flags & SHRINKER_NUMA_AWARE)) { |
373 | shrinkctl->nid = 0; | 373 | shrinkctl->nid = 0; |
374 | freed += shrink_slab_node(shrinkctl, shrinker, | 374 | freed += shrink_slab_node(shrinkctl, shrinker, |
375 | nr_pages_scanned, lru_pages); | 375 | nr_pages_scanned, lru_pages); |
376 | continue; | 376 | continue; |
377 | } | 377 | } |
378 | 378 | ||
379 | for_each_node_mask(shrinkctl->nid, shrinkctl->nodes_to_scan) { | 379 | for_each_node_mask(shrinkctl->nid, shrinkctl->nodes_to_scan) { |
380 | if (node_online(shrinkctl->nid)) | 380 | if (node_online(shrinkctl->nid)) |
381 | freed += shrink_slab_node(shrinkctl, shrinker, | 381 | freed += shrink_slab_node(shrinkctl, shrinker, |
382 | nr_pages_scanned, lru_pages); | 382 | nr_pages_scanned, lru_pages); |
383 | 383 | ||
384 | } | 384 | } |
385 | } | 385 | } |
386 | up_read(&shrinker_rwsem); | 386 | up_read(&shrinker_rwsem); |
387 | out: | 387 | out: |
388 | cond_resched(); | 388 | cond_resched(); |
389 | return freed; | 389 | return freed; |
390 | } | 390 | } |
391 | 391 | ||
392 | static inline int is_page_cache_freeable(struct page *page) | 392 | static inline int is_page_cache_freeable(struct page *page) |
393 | { | 393 | { |
394 | /* | 394 | /* |
395 | * A freeable page cache page is referenced only by the caller | 395 | * A freeable page cache page is referenced only by the caller |
396 | * that isolated the page, the page cache radix tree and | 396 | * that isolated the page, the page cache radix tree and |
397 | * optional buffer heads at page->private. | 397 | * optional buffer heads at page->private. |
398 | */ | 398 | */ |
399 | return page_count(page) - page_has_private(page) == 2; | 399 | return page_count(page) - page_has_private(page) == 2; |
400 | } | 400 | } |
401 | 401 | ||
402 | static int may_write_to_queue(struct backing_dev_info *bdi, | 402 | static int may_write_to_queue(struct backing_dev_info *bdi, |
403 | struct scan_control *sc) | 403 | struct scan_control *sc) |
404 | { | 404 | { |
405 | if (current->flags & PF_SWAPWRITE) | 405 | if (current->flags & PF_SWAPWRITE) |
406 | return 1; | 406 | return 1; |
407 | if (!bdi_write_congested(bdi)) | 407 | if (!bdi_write_congested(bdi)) |
408 | return 1; | 408 | return 1; |
409 | if (bdi == current->backing_dev_info) | 409 | if (bdi == current->backing_dev_info) |
410 | return 1; | 410 | return 1; |
411 | return 0; | 411 | return 0; |
412 | } | 412 | } |
413 | 413 | ||
414 | /* | 414 | /* |
415 | * We detected a synchronous write error writing a page out. Probably | 415 | * We detected a synchronous write error writing a page out. Probably |
416 | * -ENOSPC. We need to propagate that into the address_space for a subsequent | 416 | * -ENOSPC. We need to propagate that into the address_space for a subsequent |
417 | * fsync(), msync() or close(). | 417 | * fsync(), msync() or close(). |
418 | * | 418 | * |
419 | * The tricky part is that after writepage we cannot touch the mapping: nothing | 419 | * The tricky part is that after writepage we cannot touch the mapping: nothing |
420 | * prevents it from being freed up. But we have a ref on the page and once | 420 | * prevents it from being freed up. But we have a ref on the page and once |
421 | * that page is locked, the mapping is pinned. | 421 | * that page is locked, the mapping is pinned. |
422 | * | 422 | * |
423 | * We're allowed to run sleeping lock_page() here because we know the caller has | 423 | * We're allowed to run sleeping lock_page() here because we know the caller has |
424 | * __GFP_FS. | 424 | * __GFP_FS. |
425 | */ | 425 | */ |
426 | static void handle_write_error(struct address_space *mapping, | 426 | static void handle_write_error(struct address_space *mapping, |
427 | struct page *page, int error) | 427 | struct page *page, int error) |
428 | { | 428 | { |
429 | lock_page(page); | 429 | lock_page(page); |
430 | if (page_mapping(page) == mapping) | 430 | if (page_mapping(page) == mapping) |
431 | mapping_set_error(mapping, error); | 431 | mapping_set_error(mapping, error); |
432 | unlock_page(page); | 432 | unlock_page(page); |
433 | } | 433 | } |
434 | 434 | ||
435 | /* possible outcome of pageout() */ | 435 | /* possible outcome of pageout() */ |
436 | typedef enum { | 436 | typedef enum { |
437 | /* failed to write page out, page is locked */ | 437 | /* failed to write page out, page is locked */ |
438 | PAGE_KEEP, | 438 | PAGE_KEEP, |
439 | /* move page to the active list, page is locked */ | 439 | /* move page to the active list, page is locked */ |
440 | PAGE_ACTIVATE, | 440 | PAGE_ACTIVATE, |
441 | /* page has been sent to the disk successfully, page is unlocked */ | 441 | /* page has been sent to the disk successfully, page is unlocked */ |
442 | PAGE_SUCCESS, | 442 | PAGE_SUCCESS, |
443 | /* page is clean and locked */ | 443 | /* page is clean and locked */ |
444 | PAGE_CLEAN, | 444 | PAGE_CLEAN, |
445 | } pageout_t; | 445 | } pageout_t; |
446 | 446 | ||
447 | /* | 447 | /* |
448 | * pageout is called by shrink_page_list() for each dirty page. | 448 | * pageout is called by shrink_page_list() for each dirty page. |
449 | * Calls ->writepage(). | 449 | * Calls ->writepage(). |
450 | */ | 450 | */ |
451 | static pageout_t pageout(struct page *page, struct address_space *mapping, | 451 | static pageout_t pageout(struct page *page, struct address_space *mapping, |
452 | struct scan_control *sc) | 452 | struct scan_control *sc) |
453 | { | 453 | { |
454 | /* | 454 | /* |
455 | * If the page is dirty, only perform writeback if that write | 455 | * If the page is dirty, only perform writeback if that write |
456 | * will be non-blocking. To prevent this allocation from being | 456 | * will be non-blocking. To prevent this allocation from being |
457 | * stalled by pagecache activity. But note that there may be | 457 | * stalled by pagecache activity. But note that there may be |
458 | * stalls if we need to run get_block(). We could test | 458 | * stalls if we need to run get_block(). We could test |
459 | * PagePrivate for that. | 459 | * PagePrivate for that. |
460 | * | 460 | * |
461 | * If this process is currently in __generic_file_aio_write() against | 461 | * If this process is currently in __generic_file_aio_write() against |
462 | * this page's queue, we can perform writeback even if that | 462 | * this page's queue, we can perform writeback even if that |
463 | * will block. | 463 | * will block. |
464 | * | 464 | * |
465 | * If the page is swapcache, write it back even if that would | 465 | * If the page is swapcache, write it back even if that would |
466 | * block, for some throttling. This happens by accident, because | 466 | * block, for some throttling. This happens by accident, because |
467 | * swap_backing_dev_info is bust: it doesn't reflect the | 467 | * swap_backing_dev_info is bust: it doesn't reflect the |
468 | * congestion state of the swapdevs. Easy to fix, if needed. | 468 | * congestion state of the swapdevs. Easy to fix, if needed. |
469 | */ | 469 | */ |
470 | if (!is_page_cache_freeable(page)) | 470 | if (!is_page_cache_freeable(page)) |
471 | return PAGE_KEEP; | 471 | return PAGE_KEEP; |
472 | if (!mapping) { | 472 | if (!mapping) { |
473 | /* | 473 | /* |
474 | * Some data journaling orphaned pages can have | 474 | * Some data journaling orphaned pages can have |
475 | * page->mapping == NULL while being dirty with clean buffers. | 475 | * page->mapping == NULL while being dirty with clean buffers. |
476 | */ | 476 | */ |
477 | if (page_has_private(page)) { | 477 | if (page_has_private(page)) { |
478 | if (try_to_free_buffers(page)) { | 478 | if (try_to_free_buffers(page)) { |
479 | ClearPageDirty(page); | 479 | ClearPageDirty(page); |
480 | printk("%s: orphaned page\n", __func__); | 480 | printk("%s: orphaned page\n", __func__); |
481 | return PAGE_CLEAN; | 481 | return PAGE_CLEAN; |
482 | } | 482 | } |
483 | } | 483 | } |
484 | return PAGE_KEEP; | 484 | return PAGE_KEEP; |
485 | } | 485 | } |
486 | if (mapping->a_ops->writepage == NULL) | 486 | if (mapping->a_ops->writepage == NULL) |
487 | return PAGE_ACTIVATE; | 487 | return PAGE_ACTIVATE; |
488 | if (!may_write_to_queue(mapping->backing_dev_info, sc)) | 488 | if (!may_write_to_queue(mapping->backing_dev_info, sc)) |
489 | return PAGE_KEEP; | 489 | return PAGE_KEEP; |
490 | 490 | ||
491 | if (clear_page_dirty_for_io(page)) { | 491 | if (clear_page_dirty_for_io(page)) { |
492 | int res; | 492 | int res; |
493 | struct writeback_control wbc = { | 493 | struct writeback_control wbc = { |
494 | .sync_mode = WB_SYNC_NONE, | 494 | .sync_mode = WB_SYNC_NONE, |
495 | .nr_to_write = SWAP_CLUSTER_MAX, | 495 | .nr_to_write = SWAP_CLUSTER_MAX, |
496 | .range_start = 0, | 496 | .range_start = 0, |
497 | .range_end = LLONG_MAX, | 497 | .range_end = LLONG_MAX, |
498 | .for_reclaim = 1, | 498 | .for_reclaim = 1, |
499 | }; | 499 | }; |
500 | 500 | ||
501 | SetPageReclaim(page); | 501 | SetPageReclaim(page); |
502 | res = mapping->a_ops->writepage(page, &wbc); | 502 | res = mapping->a_ops->writepage(page, &wbc); |
503 | if (res < 0) | 503 | if (res < 0) |
504 | handle_write_error(mapping, page, res); | 504 | handle_write_error(mapping, page, res); |
505 | if (res == AOP_WRITEPAGE_ACTIVATE) { | 505 | if (res == AOP_WRITEPAGE_ACTIVATE) { |
506 | ClearPageReclaim(page); | 506 | ClearPageReclaim(page); |
507 | return PAGE_ACTIVATE; | 507 | return PAGE_ACTIVATE; |
508 | } | 508 | } |
509 | 509 | ||
510 | if (!PageWriteback(page)) { | 510 | if (!PageWriteback(page)) { |
511 | /* synchronous write or broken a_ops? */ | 511 | /* synchronous write or broken a_ops? */ |
512 | ClearPageReclaim(page); | 512 | ClearPageReclaim(page); |
513 | } | 513 | } |
514 | trace_mm_vmscan_writepage(page, trace_reclaim_flags(page)); | 514 | trace_mm_vmscan_writepage(page, trace_reclaim_flags(page)); |
515 | inc_zone_page_state(page, NR_VMSCAN_WRITE); | 515 | inc_zone_page_state(page, NR_VMSCAN_WRITE); |
516 | return PAGE_SUCCESS; | 516 | return PAGE_SUCCESS; |
517 | } | 517 | } |
518 | 518 | ||
519 | return PAGE_CLEAN; | 519 | return PAGE_CLEAN; |
520 | } | 520 | } |
521 | 521 | ||
522 | /* | 522 | /* |
523 | * Same as remove_mapping, but if the page is removed from the mapping, it | 523 | * Same as remove_mapping, but if the page is removed from the mapping, it |
524 | * gets returned with a refcount of 0. | 524 | * gets returned with a refcount of 0. |
525 | */ | 525 | */ |
526 | static int __remove_mapping(struct address_space *mapping, struct page *page) | 526 | static int __remove_mapping(struct address_space *mapping, struct page *page) |
527 | { | 527 | { |
528 | BUG_ON(!PageLocked(page)); | 528 | BUG_ON(!PageLocked(page)); |
529 | BUG_ON(mapping != page_mapping(page)); | 529 | BUG_ON(mapping != page_mapping(page)); |
530 | 530 | ||
531 | spin_lock_irq(&mapping->tree_lock); | 531 | spin_lock_irq(&mapping->tree_lock); |
532 | /* | 532 | /* |
533 | * The non racy check for a busy page. | 533 | * The non racy check for a busy page. |
534 | * | 534 | * |
535 | * Must be careful with the order of the tests. When someone has | 535 | * Must be careful with the order of the tests. When someone has |
536 | * a ref to the page, it may be possible that they dirty it then | 536 | * a ref to the page, it may be possible that they dirty it then |
537 | * drop the reference. So if PageDirty is tested before page_count | 537 | * drop the reference. So if PageDirty is tested before page_count |
538 | * here, then the following race may occur: | 538 | * here, then the following race may occur: |
539 | * | 539 | * |
540 | * get_user_pages(&page); | 540 | * get_user_pages(&page); |
541 | * [user mapping goes away] | 541 | * [user mapping goes away] |
542 | * write_to(page); | 542 | * write_to(page); |
543 | * !PageDirty(page) [good] | 543 | * !PageDirty(page) [good] |
544 | * SetPageDirty(page); | 544 | * SetPageDirty(page); |
545 | * put_page(page); | 545 | * put_page(page); |
546 | * !page_count(page) [good, discard it] | 546 | * !page_count(page) [good, discard it] |
547 | * | 547 | * |
548 | * [oops, our write_to data is lost] | 548 | * [oops, our write_to data is lost] |
549 | * | 549 | * |
550 | * Reversing the order of the tests ensures such a situation cannot | 550 | * Reversing the order of the tests ensures such a situation cannot |
551 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags | 551 | * escape unnoticed. The smp_rmb is needed to ensure the page->flags |
552 | * load is not satisfied before that of page->_count. | 552 | * load is not satisfied before that of page->_count. |
553 | * | 553 | * |
554 | * Note that if SetPageDirty is always performed via set_page_dirty, | 554 | * Note that if SetPageDirty is always performed via set_page_dirty, |
555 | * and thus under tree_lock, then this ordering is not required. | 555 | * and thus under tree_lock, then this ordering is not required. |
556 | */ | 556 | */ |
557 | if (!page_freeze_refs(page, 2)) | 557 | if (!page_freeze_refs(page, 2)) |
558 | goto cannot_free; | 558 | goto cannot_free; |
559 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ | 559 | /* note: atomic_cmpxchg in page_freeze_refs provides the smp_rmb */ |
560 | if (unlikely(PageDirty(page))) { | 560 | if (unlikely(PageDirty(page))) { |
561 | page_unfreeze_refs(page, 2); | 561 | page_unfreeze_refs(page, 2); |
562 | goto cannot_free; | 562 | goto cannot_free; |
563 | } | 563 | } |
564 | 564 | ||
565 | if (PageSwapCache(page)) { | 565 | if (PageSwapCache(page)) { |
566 | swp_entry_t swap = { .val = page_private(page) }; | 566 | swp_entry_t swap = { .val = page_private(page) }; |
567 | __delete_from_swap_cache(page); | 567 | __delete_from_swap_cache(page); |
568 | spin_unlock_irq(&mapping->tree_lock); | 568 | spin_unlock_irq(&mapping->tree_lock); |
569 | swapcache_free(swap, page); | 569 | swapcache_free(swap, page); |
570 | } else { | 570 | } else { |
571 | void (*freepage)(struct page *); | 571 | void (*freepage)(struct page *); |
572 | 572 | ||
573 | freepage = mapping->a_ops->freepage; | 573 | freepage = mapping->a_ops->freepage; |
574 | 574 | ||
575 | __delete_from_page_cache(page); | 575 | __delete_from_page_cache(page); |
576 | spin_unlock_irq(&mapping->tree_lock); | 576 | spin_unlock_irq(&mapping->tree_lock); |
577 | mem_cgroup_uncharge_cache_page(page); | 577 | mem_cgroup_uncharge_cache_page(page); |
578 | 578 | ||
579 | if (freepage != NULL) | 579 | if (freepage != NULL) |
580 | freepage(page); | 580 | freepage(page); |
581 | } | 581 | } |
582 | 582 | ||
583 | return 1; | 583 | return 1; |
584 | 584 | ||
585 | cannot_free: | 585 | cannot_free: |
586 | spin_unlock_irq(&mapping->tree_lock); | 586 | spin_unlock_irq(&mapping->tree_lock); |
587 | return 0; | 587 | return 0; |
588 | } | 588 | } |
589 | 589 | ||
590 | /* | 590 | /* |
591 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if | 591 | * Attempt to detach a locked page from its ->mapping. If it is dirty or if |
592 | * someone else has a ref on the page, abort and return 0. If it was | 592 | * someone else has a ref on the page, abort and return 0. If it was |
593 | * successfully detached, return 1. Assumes the caller has a single ref on | 593 | * successfully detached, return 1. Assumes the caller has a single ref on |
594 | * this page. | 594 | * this page. |
595 | */ | 595 | */ |
596 | int remove_mapping(struct address_space *mapping, struct page *page) | 596 | int remove_mapping(struct address_space *mapping, struct page *page) |
597 | { | 597 | { |
598 | if (__remove_mapping(mapping, page)) { | 598 | if (__remove_mapping(mapping, page)) { |
599 | /* | 599 | /* |
600 | * Unfreezing the refcount with 1 rather than 2 effectively | 600 | * Unfreezing the refcount with 1 rather than 2 effectively |
601 | * drops the pagecache ref for us without requiring another | 601 | * drops the pagecache ref for us without requiring another |
602 | * atomic operation. | 602 | * atomic operation. |
603 | */ | 603 | */ |
604 | page_unfreeze_refs(page, 1); | 604 | page_unfreeze_refs(page, 1); |
605 | return 1; | 605 | return 1; |
606 | } | 606 | } |
607 | return 0; | 607 | return 0; |
608 | } | 608 | } |
609 | 609 | ||
610 | /** | 610 | /** |
611 | * putback_lru_page - put previously isolated page onto appropriate LRU list | 611 | * putback_lru_page - put previously isolated page onto appropriate LRU list |
612 | * @page: page to be put back to appropriate lru list | 612 | * @page: page to be put back to appropriate lru list |
613 | * | 613 | * |
614 | * Add previously isolated @page to appropriate LRU list. | 614 | * Add previously isolated @page to appropriate LRU list. |
615 | * Page may still be unevictable for other reasons. | 615 | * Page may still be unevictable for other reasons. |
616 | * | 616 | * |
617 | * lru_lock must not be held, interrupts must be enabled. | 617 | * lru_lock must not be held, interrupts must be enabled. |
618 | */ | 618 | */ |
619 | void putback_lru_page(struct page *page) | 619 | void putback_lru_page(struct page *page) |
620 | { | 620 | { |
621 | bool is_unevictable; | 621 | bool is_unevictable; |
622 | int was_unevictable = PageUnevictable(page); | 622 | int was_unevictable = PageUnevictable(page); |
623 | 623 | ||
624 | VM_BUG_ON(PageLRU(page)); | 624 | VM_BUG_ON(PageLRU(page)); |
625 | 625 | ||
626 | redo: | 626 | redo: |
627 | ClearPageUnevictable(page); | 627 | ClearPageUnevictable(page); |
628 | 628 | ||
629 | if (page_evictable(page)) { | 629 | if (page_evictable(page)) { |
630 | /* | 630 | /* |
631 | * For evictable pages, we can use the cache. | 631 | * For evictable pages, we can use the cache. |
632 | * In event of a race, worst case is we end up with an | 632 | * In event of a race, worst case is we end up with an |
633 | * unevictable page on [in]active list. | 633 | * unevictable page on [in]active list. |
634 | * We know how to handle that. | 634 | * We know how to handle that. |
635 | */ | 635 | */ |
636 | is_unevictable = false; | 636 | is_unevictable = false; |
637 | lru_cache_add(page); | 637 | lru_cache_add(page); |
638 | } else { | 638 | } else { |
639 | /* | 639 | /* |
640 | * Put unevictable pages directly on zone's unevictable | 640 | * Put unevictable pages directly on zone's unevictable |
641 | * list. | 641 | * list. |
642 | */ | 642 | */ |
643 | is_unevictable = true; | 643 | is_unevictable = true; |
644 | add_page_to_unevictable_list(page); | 644 | add_page_to_unevictable_list(page); |
645 | /* | 645 | /* |
646 | * When racing with an mlock or AS_UNEVICTABLE clearing | 646 | * When racing with an mlock or AS_UNEVICTABLE clearing |
647 | * (page is unlocked) make sure that if the other thread | 647 | * (page is unlocked) make sure that if the other thread |
648 | * does not observe our setting of PG_lru and fails | 648 | * does not observe our setting of PG_lru and fails |
649 | * isolation/check_move_unevictable_pages, | 649 | * isolation/check_move_unevictable_pages, |
650 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move | 650 | * we see PG_mlocked/AS_UNEVICTABLE cleared below and move |
651 | * the page back to the evictable list. | 651 | * the page back to the evictable list. |
652 | * | 652 | * |
653 | * The other side is TestClearPageMlocked() or shmem_lock(). | 653 | * The other side is TestClearPageMlocked() or shmem_lock(). |
654 | */ | 654 | */ |
655 | smp_mb(); | 655 | smp_mb(); |
656 | } | 656 | } |
657 | 657 | ||
658 | /* | 658 | /* |
659 | * page's status can change while we move it among lru. If an evictable | 659 | * page's status can change while we move it among lru. If an evictable |
660 | * page is on unevictable list, it never be freed. To avoid that, | 660 | * page is on unevictable list, it never be freed. To avoid that, |
661 | * check after we added it to the list, again. | 661 | * check after we added it to the list, again. |
662 | */ | 662 | */ |
663 | if (is_unevictable && page_evictable(page)) { | 663 | if (is_unevictable && page_evictable(page)) { |
664 | if (!isolate_lru_page(page)) { | 664 | if (!isolate_lru_page(page)) { |
665 | put_page(page); | 665 | put_page(page); |
666 | goto redo; | 666 | goto redo; |
667 | } | 667 | } |
668 | /* This means someone else dropped this page from LRU | 668 | /* This means someone else dropped this page from LRU |
669 | * So, it will be freed or putback to LRU again. There is | 669 | * So, it will be freed or putback to LRU again. There is |
670 | * nothing to do here. | 670 | * nothing to do here. |
671 | */ | 671 | */ |
672 | } | 672 | } |
673 | 673 | ||
674 | if (was_unevictable && !is_unevictable) | 674 | if (was_unevictable && !is_unevictable) |
675 | count_vm_event(UNEVICTABLE_PGRESCUED); | 675 | count_vm_event(UNEVICTABLE_PGRESCUED); |
676 | else if (!was_unevictable && is_unevictable) | 676 | else if (!was_unevictable && is_unevictable) |
677 | count_vm_event(UNEVICTABLE_PGCULLED); | 677 | count_vm_event(UNEVICTABLE_PGCULLED); |
678 | 678 | ||
679 | put_page(page); /* drop ref from isolate */ | 679 | put_page(page); /* drop ref from isolate */ |
680 | } | 680 | } |
681 | 681 | ||
682 | enum page_references { | 682 | enum page_references { |
683 | PAGEREF_RECLAIM, | 683 | PAGEREF_RECLAIM, |
684 | PAGEREF_RECLAIM_CLEAN, | 684 | PAGEREF_RECLAIM_CLEAN, |
685 | PAGEREF_KEEP, | 685 | PAGEREF_KEEP, |
686 | PAGEREF_ACTIVATE, | 686 | PAGEREF_ACTIVATE, |
687 | }; | 687 | }; |
688 | 688 | ||
689 | static enum page_references page_check_references(struct page *page, | 689 | static enum page_references page_check_references(struct page *page, |
690 | struct scan_control *sc) | 690 | struct scan_control *sc) |
691 | { | 691 | { |
692 | int referenced_ptes, referenced_page; | 692 | int referenced_ptes, referenced_page; |
693 | unsigned long vm_flags; | 693 | unsigned long vm_flags; |
694 | 694 | ||
695 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, | 695 | referenced_ptes = page_referenced(page, 1, sc->target_mem_cgroup, |
696 | &vm_flags); | 696 | &vm_flags); |
697 | referenced_page = TestClearPageReferenced(page); | 697 | referenced_page = TestClearPageReferenced(page); |
698 | 698 | ||
699 | /* | 699 | /* |
700 | * Mlock lost the isolation race with us. Let try_to_unmap() | 700 | * Mlock lost the isolation race with us. Let try_to_unmap() |
701 | * move the page to the unevictable list. | 701 | * move the page to the unevictable list. |
702 | */ | 702 | */ |
703 | if (vm_flags & VM_LOCKED) | 703 | if (vm_flags & VM_LOCKED) |
704 | return PAGEREF_RECLAIM; | 704 | return PAGEREF_RECLAIM; |
705 | 705 | ||
706 | if (referenced_ptes) { | 706 | if (referenced_ptes) { |
707 | if (PageSwapBacked(page)) | 707 | if (PageSwapBacked(page)) |
708 | return PAGEREF_ACTIVATE; | 708 | return PAGEREF_ACTIVATE; |
709 | /* | 709 | /* |
710 | * All mapped pages start out with page table | 710 | * All mapped pages start out with page table |
711 | * references from the instantiating fault, so we need | 711 | * references from the instantiating fault, so we need |
712 | * to look twice if a mapped file page is used more | 712 | * to look twice if a mapped file page is used more |
713 | * than once. | 713 | * than once. |
714 | * | 714 | * |
715 | * Mark it and spare it for another trip around the | 715 | * Mark it and spare it for another trip around the |
716 | * inactive list. Another page table reference will | 716 | * inactive list. Another page table reference will |
717 | * lead to its activation. | 717 | * lead to its activation. |
718 | * | 718 | * |
719 | * Note: the mark is set for activated pages as well | 719 | * Note: the mark is set for activated pages as well |
720 | * so that recently deactivated but used pages are | 720 | * so that recently deactivated but used pages are |
721 | * quickly recovered. | 721 | * quickly recovered. |
722 | */ | 722 | */ |
723 | SetPageReferenced(page); | 723 | SetPageReferenced(page); |
724 | 724 | ||
725 | if (referenced_page || referenced_ptes > 1) | 725 | if (referenced_page || referenced_ptes > 1) |
726 | return PAGEREF_ACTIVATE; | 726 | return PAGEREF_ACTIVATE; |
727 | 727 | ||
728 | /* | 728 | /* |
729 | * Activate file-backed executable pages after first usage. | 729 | * Activate file-backed executable pages after first usage. |
730 | */ | 730 | */ |
731 | if (vm_flags & VM_EXEC) | 731 | if (vm_flags & VM_EXEC) |
732 | return PAGEREF_ACTIVATE; | 732 | return PAGEREF_ACTIVATE; |
733 | 733 | ||
734 | return PAGEREF_KEEP; | 734 | return PAGEREF_KEEP; |
735 | } | 735 | } |
736 | 736 | ||
737 | /* Reclaim if clean, defer dirty pages to writeback */ | 737 | /* Reclaim if clean, defer dirty pages to writeback */ |
738 | if (referenced_page && !PageSwapBacked(page)) | 738 | if (referenced_page && !PageSwapBacked(page)) |
739 | return PAGEREF_RECLAIM_CLEAN; | 739 | return PAGEREF_RECLAIM_CLEAN; |
740 | 740 | ||
741 | return PAGEREF_RECLAIM; | 741 | return PAGEREF_RECLAIM; |
742 | } | 742 | } |
743 | 743 | ||
744 | /* Check if a page is dirty or under writeback */ | 744 | /* Check if a page is dirty or under writeback */ |
745 | static void page_check_dirty_writeback(struct page *page, | 745 | static void page_check_dirty_writeback(struct page *page, |
746 | bool *dirty, bool *writeback) | 746 | bool *dirty, bool *writeback) |
747 | { | 747 | { |
748 | struct address_space *mapping; | 748 | struct address_space *mapping; |
749 | 749 | ||
750 | /* | 750 | /* |
751 | * Anonymous pages are not handled by flushers and must be written | 751 | * Anonymous pages are not handled by flushers and must be written |
752 | * from reclaim context. Do not stall reclaim based on them | 752 | * from reclaim context. Do not stall reclaim based on them |
753 | */ | 753 | */ |
754 | if (!page_is_file_cache(page)) { | 754 | if (!page_is_file_cache(page)) { |
755 | *dirty = false; | 755 | *dirty = false; |
756 | *writeback = false; | 756 | *writeback = false; |
757 | return; | 757 | return; |
758 | } | 758 | } |
759 | 759 | ||
760 | /* By default assume that the page flags are accurate */ | 760 | /* By default assume that the page flags are accurate */ |
761 | *dirty = PageDirty(page); | 761 | *dirty = PageDirty(page); |
762 | *writeback = PageWriteback(page); | 762 | *writeback = PageWriteback(page); |
763 | 763 | ||
764 | /* Verify dirty/writeback state if the filesystem supports it */ | 764 | /* Verify dirty/writeback state if the filesystem supports it */ |
765 | if (!page_has_private(page)) | 765 | if (!page_has_private(page)) |
766 | return; | 766 | return; |
767 | 767 | ||
768 | mapping = page_mapping(page); | 768 | mapping = page_mapping(page); |
769 | if (mapping && mapping->a_ops->is_dirty_writeback) | 769 | if (mapping && mapping->a_ops->is_dirty_writeback) |
770 | mapping->a_ops->is_dirty_writeback(page, dirty, writeback); | 770 | mapping->a_ops->is_dirty_writeback(page, dirty, writeback); |
771 | } | 771 | } |
772 | 772 | ||
773 | /* | 773 | /* |
774 | * shrink_page_list() returns the number of reclaimed pages | 774 | * shrink_page_list() returns the number of reclaimed pages |
775 | */ | 775 | */ |
776 | static unsigned long shrink_page_list(struct list_head *page_list, | 776 | static unsigned long shrink_page_list(struct list_head *page_list, |
777 | struct zone *zone, | 777 | struct zone *zone, |
778 | struct scan_control *sc, | 778 | struct scan_control *sc, |
779 | enum ttu_flags ttu_flags, | 779 | enum ttu_flags ttu_flags, |
780 | unsigned long *ret_nr_dirty, | 780 | unsigned long *ret_nr_dirty, |
781 | unsigned long *ret_nr_unqueued_dirty, | 781 | unsigned long *ret_nr_unqueued_dirty, |
782 | unsigned long *ret_nr_congested, | 782 | unsigned long *ret_nr_congested, |
783 | unsigned long *ret_nr_writeback, | 783 | unsigned long *ret_nr_writeback, |
784 | unsigned long *ret_nr_immediate, | 784 | unsigned long *ret_nr_immediate, |
785 | bool force_reclaim) | 785 | bool force_reclaim) |
786 | { | 786 | { |
787 | LIST_HEAD(ret_pages); | 787 | LIST_HEAD(ret_pages); |
788 | LIST_HEAD(free_pages); | 788 | LIST_HEAD(free_pages); |
789 | int pgactivate = 0; | 789 | int pgactivate = 0; |
790 | unsigned long nr_unqueued_dirty = 0; | 790 | unsigned long nr_unqueued_dirty = 0; |
791 | unsigned long nr_dirty = 0; | 791 | unsigned long nr_dirty = 0; |
792 | unsigned long nr_congested = 0; | 792 | unsigned long nr_congested = 0; |
793 | unsigned long nr_reclaimed = 0; | 793 | unsigned long nr_reclaimed = 0; |
794 | unsigned long nr_writeback = 0; | 794 | unsigned long nr_writeback = 0; |
795 | unsigned long nr_immediate = 0; | 795 | unsigned long nr_immediate = 0; |
796 | 796 | ||
797 | cond_resched(); | 797 | cond_resched(); |
798 | 798 | ||
799 | mem_cgroup_uncharge_start(); | 799 | mem_cgroup_uncharge_start(); |
800 | while (!list_empty(page_list)) { | 800 | while (!list_empty(page_list)) { |
801 | struct address_space *mapping; | 801 | struct address_space *mapping; |
802 | struct page *page; | 802 | struct page *page; |
803 | int may_enter_fs; | 803 | int may_enter_fs; |
804 | enum page_references references = PAGEREF_RECLAIM_CLEAN; | 804 | enum page_references references = PAGEREF_RECLAIM_CLEAN; |
805 | bool dirty, writeback; | 805 | bool dirty, writeback; |
806 | 806 | ||
807 | cond_resched(); | 807 | cond_resched(); |
808 | 808 | ||
809 | page = lru_to_page(page_list); | 809 | page = lru_to_page(page_list); |
810 | list_del(&page->lru); | 810 | list_del(&page->lru); |
811 | 811 | ||
812 | if (!trylock_page(page)) | 812 | if (!trylock_page(page)) |
813 | goto keep; | 813 | goto keep; |
814 | 814 | ||
815 | VM_BUG_ON(PageActive(page)); | 815 | VM_BUG_ON(PageActive(page)); |
816 | VM_BUG_ON(page_zone(page) != zone); | 816 | VM_BUG_ON(page_zone(page) != zone); |
817 | 817 | ||
818 | sc->nr_scanned++; | 818 | sc->nr_scanned++; |
819 | 819 | ||
820 | if (unlikely(!page_evictable(page))) | 820 | if (unlikely(!page_evictable(page))) |
821 | goto cull_mlocked; | 821 | goto cull_mlocked; |
822 | 822 | ||
823 | if (!sc->may_unmap && page_mapped(page)) | 823 | if (!sc->may_unmap && page_mapped(page)) |
824 | goto keep_locked; | 824 | goto keep_locked; |
825 | 825 | ||
826 | /* Double the slab pressure for mapped and swapcache pages */ | 826 | /* Double the slab pressure for mapped and swapcache pages */ |
827 | if (page_mapped(page) || PageSwapCache(page)) | 827 | if (page_mapped(page) || PageSwapCache(page)) |
828 | sc->nr_scanned++; | 828 | sc->nr_scanned++; |
829 | 829 | ||
830 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || | 830 | may_enter_fs = (sc->gfp_mask & __GFP_FS) || |
831 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); | 831 | (PageSwapCache(page) && (sc->gfp_mask & __GFP_IO)); |
832 | 832 | ||
833 | /* | 833 | /* |
834 | * The number of dirty pages determines if a zone is marked | 834 | * The number of dirty pages determines if a zone is marked |
835 | * reclaim_congested which affects wait_iff_congested. kswapd | 835 | * reclaim_congested which affects wait_iff_congested. kswapd |
836 | * will stall and start writing pages if the tail of the LRU | 836 | * will stall and start writing pages if the tail of the LRU |
837 | * is all dirty unqueued pages. | 837 | * is all dirty unqueued pages. |
838 | */ | 838 | */ |
839 | page_check_dirty_writeback(page, &dirty, &writeback); | 839 | page_check_dirty_writeback(page, &dirty, &writeback); |
840 | if (dirty || writeback) | 840 | if (dirty || writeback) |
841 | nr_dirty++; | 841 | nr_dirty++; |
842 | 842 | ||
843 | if (dirty && !writeback) | 843 | if (dirty && !writeback) |
844 | nr_unqueued_dirty++; | 844 | nr_unqueued_dirty++; |
845 | 845 | ||
846 | /* | 846 | /* |
847 | * Treat this page as congested if the underlying BDI is or if | 847 | * Treat this page as congested if the underlying BDI is or if |
848 | * pages are cycling through the LRU so quickly that the | 848 | * pages are cycling through the LRU so quickly that the |
849 | * pages marked for immediate reclaim are making it to the | 849 | * pages marked for immediate reclaim are making it to the |
850 | * end of the LRU a second time. | 850 | * end of the LRU a second time. |
851 | */ | 851 | */ |
852 | mapping = page_mapping(page); | 852 | mapping = page_mapping(page); |
853 | if ((mapping && bdi_write_congested(mapping->backing_dev_info)) || | 853 | if ((mapping && bdi_write_congested(mapping->backing_dev_info)) || |
854 | (writeback && PageReclaim(page))) | 854 | (writeback && PageReclaim(page))) |
855 | nr_congested++; | 855 | nr_congested++; |
856 | 856 | ||
857 | /* | 857 | /* |
858 | * If a page at the tail of the LRU is under writeback, there | 858 | * If a page at the tail of the LRU is under writeback, there |
859 | * are three cases to consider. | 859 | * are three cases to consider. |
860 | * | 860 | * |
861 | * 1) If reclaim is encountering an excessive number of pages | 861 | * 1) If reclaim is encountering an excessive number of pages |
862 | * under writeback and this page is both under writeback and | 862 | * under writeback and this page is both under writeback and |
863 | * PageReclaim then it indicates that pages are being queued | 863 | * PageReclaim then it indicates that pages are being queued |
864 | * for IO but are being recycled through the LRU before the | 864 | * for IO but are being recycled through the LRU before the |
865 | * IO can complete. Waiting on the page itself risks an | 865 | * IO can complete. Waiting on the page itself risks an |
866 | * indefinite stall if it is impossible to writeback the | 866 | * indefinite stall if it is impossible to writeback the |
867 | * page due to IO error or disconnected storage so instead | 867 | * page due to IO error or disconnected storage so instead |
868 | * note that the LRU is being scanned too quickly and the | 868 | * note that the LRU is being scanned too quickly and the |
869 | * caller can stall after page list has been processed. | 869 | * caller can stall after page list has been processed. |
870 | * | 870 | * |
871 | * 2) Global reclaim encounters a page, memcg encounters a | 871 | * 2) Global reclaim encounters a page, memcg encounters a |
872 | * page that is not marked for immediate reclaim or | 872 | * page that is not marked for immediate reclaim or |
873 | * the caller does not have __GFP_IO. In this case mark | 873 | * the caller does not have __GFP_IO. In this case mark |
874 | * the page for immediate reclaim and continue scanning. | 874 | * the page for immediate reclaim and continue scanning. |
875 | * | 875 | * |
876 | * __GFP_IO is checked because a loop driver thread might | 876 | * __GFP_IO is checked because a loop driver thread might |
877 | * enter reclaim, and deadlock if it waits on a page for | 877 | * enter reclaim, and deadlock if it waits on a page for |
878 | * which it is needed to do the write (loop masks off | 878 | * which it is needed to do the write (loop masks off |
879 | * __GFP_IO|__GFP_FS for this reason); but more thought | 879 | * __GFP_IO|__GFP_FS for this reason); but more thought |
880 | * would probably show more reasons. | 880 | * would probably show more reasons. |
881 | * | 881 | * |
882 | * Don't require __GFP_FS, since we're not going into the | 882 | * Don't require __GFP_FS, since we're not going into the |
883 | * FS, just waiting on its writeback completion. Worryingly, | 883 | * FS, just waiting on its writeback completion. Worryingly, |
884 | * ext4 gfs2 and xfs allocate pages with | 884 | * ext4 gfs2 and xfs allocate pages with |
885 | * grab_cache_page_write_begin(,,AOP_FLAG_NOFS), so testing | 885 | * grab_cache_page_write_begin(,,AOP_FLAG_NOFS), so testing |
886 | * may_enter_fs here is liable to OOM on them. | 886 | * may_enter_fs here is liable to OOM on them. |
887 | * | 887 | * |
888 | * 3) memcg encounters a page that is not already marked | 888 | * 3) memcg encounters a page that is not already marked |
889 | * PageReclaim. memcg does not have any dirty pages | 889 | * PageReclaim. memcg does not have any dirty pages |
890 | * throttling so we could easily OOM just because too many | 890 | * throttling so we could easily OOM just because too many |
891 | * pages are in writeback and there is nothing else to | 891 | * pages are in writeback and there is nothing else to |
892 | * reclaim. Wait for the writeback to complete. | 892 | * reclaim. Wait for the writeback to complete. |
893 | */ | 893 | */ |
894 | if (PageWriteback(page)) { | 894 | if (PageWriteback(page)) { |
895 | /* Case 1 above */ | 895 | /* Case 1 above */ |
896 | if (current_is_kswapd() && | 896 | if (current_is_kswapd() && |
897 | PageReclaim(page) && | 897 | PageReclaim(page) && |
898 | zone_is_reclaim_writeback(zone)) { | 898 | zone_is_reclaim_writeback(zone)) { |
899 | nr_immediate++; | 899 | nr_immediate++; |
900 | goto keep_locked; | 900 | goto keep_locked; |
901 | 901 | ||
902 | /* Case 2 above */ | 902 | /* Case 2 above */ |
903 | } else if (global_reclaim(sc) || | 903 | } else if (global_reclaim(sc) || |
904 | !PageReclaim(page) || !(sc->gfp_mask & __GFP_IO)) { | 904 | !PageReclaim(page) || !(sc->gfp_mask & __GFP_IO)) { |
905 | /* | 905 | /* |
906 | * This is slightly racy - end_page_writeback() | 906 | * This is slightly racy - end_page_writeback() |
907 | * might have just cleared PageReclaim, then | 907 | * might have just cleared PageReclaim, then |
908 | * setting PageReclaim here end up interpreted | 908 | * setting PageReclaim here end up interpreted |
909 | * as PageReadahead - but that does not matter | 909 | * as PageReadahead - but that does not matter |
910 | * enough to care. What we do want is for this | 910 | * enough to care. What we do want is for this |
911 | * page to have PageReclaim set next time memcg | 911 | * page to have PageReclaim set next time memcg |
912 | * reclaim reaches the tests above, so it will | 912 | * reclaim reaches the tests above, so it will |
913 | * then wait_on_page_writeback() to avoid OOM; | 913 | * then wait_on_page_writeback() to avoid OOM; |
914 | * and it's also appropriate in global reclaim. | 914 | * and it's also appropriate in global reclaim. |
915 | */ | 915 | */ |
916 | SetPageReclaim(page); | 916 | SetPageReclaim(page); |
917 | nr_writeback++; | 917 | nr_writeback++; |
918 | 918 | ||
919 | goto keep_locked; | 919 | goto keep_locked; |
920 | 920 | ||
921 | /* Case 3 above */ | 921 | /* Case 3 above */ |
922 | } else { | 922 | } else { |
923 | wait_on_page_writeback(page); | 923 | wait_on_page_writeback(page); |
924 | } | 924 | } |
925 | } | 925 | } |
926 | 926 | ||
927 | if (!force_reclaim) | 927 | if (!force_reclaim) |
928 | references = page_check_references(page, sc); | 928 | references = page_check_references(page, sc); |
929 | 929 | ||
930 | switch (references) { | 930 | switch (references) { |
931 | case PAGEREF_ACTIVATE: | 931 | case PAGEREF_ACTIVATE: |
932 | goto activate_locked; | 932 | goto activate_locked; |
933 | case PAGEREF_KEEP: | 933 | case PAGEREF_KEEP: |
934 | goto keep_locked; | 934 | goto keep_locked; |
935 | case PAGEREF_RECLAIM: | 935 | case PAGEREF_RECLAIM: |
936 | case PAGEREF_RECLAIM_CLEAN: | 936 | case PAGEREF_RECLAIM_CLEAN: |
937 | ; /* try to reclaim the page below */ | 937 | ; /* try to reclaim the page below */ |
938 | } | 938 | } |
939 | 939 | ||
940 | /* | 940 | /* |
941 | * Anonymous process memory has backing store? | 941 | * Anonymous process memory has backing store? |
942 | * Try to allocate it some swap space here. | 942 | * Try to allocate it some swap space here. |
943 | */ | 943 | */ |
944 | if (PageAnon(page) && !PageSwapCache(page)) { | 944 | if (PageAnon(page) && !PageSwapCache(page)) { |
945 | if (!(sc->gfp_mask & __GFP_IO)) | 945 | if (!(sc->gfp_mask & __GFP_IO)) |
946 | goto keep_locked; | 946 | goto keep_locked; |
947 | if (!add_to_swap(page, page_list)) | 947 | if (!add_to_swap(page, page_list)) |
948 | goto activate_locked; | 948 | goto activate_locked; |
949 | may_enter_fs = 1; | 949 | may_enter_fs = 1; |
950 | 950 | ||
951 | /* Adding to swap updated mapping */ | 951 | /* Adding to swap updated mapping */ |
952 | mapping = page_mapping(page); | 952 | mapping = page_mapping(page); |
953 | } | 953 | } |
954 | 954 | ||
955 | /* | 955 | /* |
956 | * The page is mapped into the page tables of one or more | 956 | * The page is mapped into the page tables of one or more |
957 | * processes. Try to unmap it here. | 957 | * processes. Try to unmap it here. |
958 | */ | 958 | */ |
959 | if (page_mapped(page) && mapping) { | 959 | if (page_mapped(page) && mapping) { |
960 | switch (try_to_unmap(page, ttu_flags)) { | 960 | switch (try_to_unmap(page, ttu_flags)) { |
961 | case SWAP_FAIL: | 961 | case SWAP_FAIL: |
962 | goto activate_locked; | 962 | goto activate_locked; |
963 | case SWAP_AGAIN: | 963 | case SWAP_AGAIN: |
964 | goto keep_locked; | 964 | goto keep_locked; |
965 | case SWAP_MLOCK: | 965 | case SWAP_MLOCK: |
966 | goto cull_mlocked; | 966 | goto cull_mlocked; |
967 | case SWAP_SUCCESS: | 967 | case SWAP_SUCCESS: |
968 | ; /* try to free the page below */ | 968 | ; /* try to free the page below */ |
969 | } | 969 | } |
970 | } | 970 | } |
971 | 971 | ||
972 | if (PageDirty(page)) { | 972 | if (PageDirty(page)) { |
973 | /* | 973 | /* |
974 | * Only kswapd can writeback filesystem pages to | 974 | * Only kswapd can writeback filesystem pages to |
975 | * avoid risk of stack overflow but only writeback | 975 | * avoid risk of stack overflow but only writeback |
976 | * if many dirty pages have been encountered. | 976 | * if many dirty pages have been encountered. |
977 | */ | 977 | */ |
978 | if (page_is_file_cache(page) && | 978 | if (page_is_file_cache(page) && |
979 | (!current_is_kswapd() || | 979 | (!current_is_kswapd() || |
980 | !zone_is_reclaim_dirty(zone))) { | 980 | !zone_is_reclaim_dirty(zone))) { |
981 | /* | 981 | /* |
982 | * Immediately reclaim when written back. | 982 | * Immediately reclaim when written back. |
983 | * Similar in principal to deactivate_page() | 983 | * Similar in principal to deactivate_page() |
984 | * except we already have the page isolated | 984 | * except we already have the page isolated |
985 | * and know it's dirty | 985 | * and know it's dirty |
986 | */ | 986 | */ |
987 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); | 987 | inc_zone_page_state(page, NR_VMSCAN_IMMEDIATE); |
988 | SetPageReclaim(page); | 988 | SetPageReclaim(page); |
989 | 989 | ||
990 | goto keep_locked; | 990 | goto keep_locked; |
991 | } | 991 | } |
992 | 992 | ||
993 | if (references == PAGEREF_RECLAIM_CLEAN) | 993 | if (references == PAGEREF_RECLAIM_CLEAN) |
994 | goto keep_locked; | 994 | goto keep_locked; |
995 | if (!may_enter_fs) | 995 | if (!may_enter_fs) |
996 | goto keep_locked; | 996 | goto keep_locked; |
997 | if (!sc->may_writepage) | 997 | if (!sc->may_writepage) |
998 | goto keep_locked; | 998 | goto keep_locked; |
999 | 999 | ||
1000 | /* Page is dirty, try to write it out here */ | 1000 | /* Page is dirty, try to write it out here */ |
1001 | switch (pageout(page, mapping, sc)) { | 1001 | switch (pageout(page, mapping, sc)) { |
1002 | case PAGE_KEEP: | 1002 | case PAGE_KEEP: |
1003 | goto keep_locked; | 1003 | goto keep_locked; |
1004 | case PAGE_ACTIVATE: | 1004 | case PAGE_ACTIVATE: |
1005 | goto activate_locked; | 1005 | goto activate_locked; |
1006 | case PAGE_SUCCESS: | 1006 | case PAGE_SUCCESS: |
1007 | if (PageWriteback(page)) | 1007 | if (PageWriteback(page)) |
1008 | goto keep; | 1008 | goto keep; |
1009 | if (PageDirty(page)) | 1009 | if (PageDirty(page)) |
1010 | goto keep; | 1010 | goto keep; |
1011 | 1011 | ||
1012 | /* | 1012 | /* |
1013 | * A synchronous write - probably a ramdisk. Go | 1013 | * A synchronous write - probably a ramdisk. Go |
1014 | * ahead and try to reclaim the page. | 1014 | * ahead and try to reclaim the page. |
1015 | */ | 1015 | */ |
1016 | if (!trylock_page(page)) | 1016 | if (!trylock_page(page)) |
1017 | goto keep; | 1017 | goto keep; |
1018 | if (PageDirty(page) || PageWriteback(page)) | 1018 | if (PageDirty(page) || PageWriteback(page)) |
1019 | goto keep_locked; | 1019 | goto keep_locked; |
1020 | mapping = page_mapping(page); | 1020 | mapping = page_mapping(page); |
1021 | case PAGE_CLEAN: | 1021 | case PAGE_CLEAN: |
1022 | ; /* try to free the page below */ | 1022 | ; /* try to free the page below */ |
1023 | } | 1023 | } |
1024 | } | 1024 | } |
1025 | 1025 | ||
1026 | /* | 1026 | /* |
1027 | * If the page has buffers, try to free the buffer mappings | 1027 | * If the page has buffers, try to free the buffer mappings |
1028 | * associated with this page. If we succeed we try to free | 1028 | * associated with this page. If we succeed we try to free |
1029 | * the page as well. | 1029 | * the page as well. |
1030 | * | 1030 | * |
1031 | * We do this even if the page is PageDirty(). | 1031 | * We do this even if the page is PageDirty(). |
1032 | * try_to_release_page() does not perform I/O, but it is | 1032 | * try_to_release_page() does not perform I/O, but it is |
1033 | * possible for a page to have PageDirty set, but it is actually | 1033 | * possible for a page to have PageDirty set, but it is actually |
1034 | * clean (all its buffers are clean). This happens if the | 1034 | * clean (all its buffers are clean). This happens if the |
1035 | * buffers were written out directly, with submit_bh(). ext3 | 1035 | * buffers were written out directly, with submit_bh(). ext3 |
1036 | * will do this, as well as the blockdev mapping. | 1036 | * will do this, as well as the blockdev mapping. |
1037 | * try_to_release_page() will discover that cleanness and will | 1037 | * try_to_release_page() will discover that cleanness and will |
1038 | * drop the buffers and mark the page clean - it can be freed. | 1038 | * drop the buffers and mark the page clean - it can be freed. |
1039 | * | 1039 | * |
1040 | * Rarely, pages can have buffers and no ->mapping. These are | 1040 | * Rarely, pages can have buffers and no ->mapping. These are |
1041 | * the pages which were not successfully invalidated in | 1041 | * the pages which were not successfully invalidated in |
1042 | * truncate_complete_page(). We try to drop those buffers here | 1042 | * truncate_complete_page(). We try to drop those buffers here |
1043 | * and if that worked, and the page is no longer mapped into | 1043 | * and if that worked, and the page is no longer mapped into |
1044 | * process address space (page_count == 1) it can be freed. | 1044 | * process address space (page_count == 1) it can be freed. |
1045 | * Otherwise, leave the page on the LRU so it is swappable. | 1045 | * Otherwise, leave the page on the LRU so it is swappable. |
1046 | */ | 1046 | */ |
1047 | if (page_has_private(page)) { | 1047 | if (page_has_private(page)) { |
1048 | if (!try_to_release_page(page, sc->gfp_mask)) | 1048 | if (!try_to_release_page(page, sc->gfp_mask)) |
1049 | goto activate_locked; | 1049 | goto activate_locked; |
1050 | if (!mapping && page_count(page) == 1) { | 1050 | if (!mapping && page_count(page) == 1) { |
1051 | unlock_page(page); | 1051 | unlock_page(page); |
1052 | if (put_page_testzero(page)) | 1052 | if (put_page_testzero(page)) |
1053 | goto free_it; | 1053 | goto free_it; |
1054 | else { | 1054 | else { |
1055 | /* | 1055 | /* |
1056 | * rare race with speculative reference. | 1056 | * rare race with speculative reference. |
1057 | * the speculative reference will free | 1057 | * the speculative reference will free |
1058 | * this page shortly, so we may | 1058 | * this page shortly, so we may |
1059 | * increment nr_reclaimed here (and | 1059 | * increment nr_reclaimed here (and |
1060 | * leave it off the LRU). | 1060 | * leave it off the LRU). |
1061 | */ | 1061 | */ |
1062 | nr_reclaimed++; | 1062 | nr_reclaimed++; |
1063 | continue; | 1063 | continue; |
1064 | } | 1064 | } |
1065 | } | 1065 | } |
1066 | } | 1066 | } |
1067 | 1067 | ||
1068 | if (!mapping || !__remove_mapping(mapping, page)) | 1068 | if (!mapping || !__remove_mapping(mapping, page)) |
1069 | goto keep_locked; | 1069 | goto keep_locked; |
1070 | 1070 | ||
1071 | /* | 1071 | /* |
1072 | * At this point, we have no other references and there is | 1072 | * At this point, we have no other references and there is |
1073 | * no way to pick any more up (removed from LRU, removed | 1073 | * no way to pick any more up (removed from LRU, removed |
1074 | * from pagecache). Can use non-atomic bitops now (and | 1074 | * from pagecache). Can use non-atomic bitops now (and |
1075 | * we obviously don't have to worry about waking up a process | 1075 | * we obviously don't have to worry about waking up a process |
1076 | * waiting on the page lock, because there are no references. | 1076 | * waiting on the page lock, because there are no references. |
1077 | */ | 1077 | */ |
1078 | __clear_page_locked(page); | 1078 | __clear_page_locked(page); |
1079 | free_it: | 1079 | free_it: |
1080 | nr_reclaimed++; | 1080 | nr_reclaimed++; |
1081 | 1081 | ||
1082 | /* | 1082 | /* |
1083 | * Is there need to periodically free_page_list? It would | 1083 | * Is there need to periodically free_page_list? It would |
1084 | * appear not as the counts should be low | 1084 | * appear not as the counts should be low |
1085 | */ | 1085 | */ |
1086 | list_add(&page->lru, &free_pages); | 1086 | list_add(&page->lru, &free_pages); |
1087 | continue; | 1087 | continue; |
1088 | 1088 | ||
1089 | cull_mlocked: | 1089 | cull_mlocked: |
1090 | if (PageSwapCache(page)) | 1090 | if (PageSwapCache(page)) |
1091 | try_to_free_swap(page); | 1091 | try_to_free_swap(page); |
1092 | unlock_page(page); | 1092 | unlock_page(page); |
1093 | putback_lru_page(page); | 1093 | putback_lru_page(page); |
1094 | continue; | 1094 | continue; |
1095 | 1095 | ||
1096 | activate_locked: | 1096 | activate_locked: |
1097 | /* Not a candidate for swapping, so reclaim swap space. */ | 1097 | /* Not a candidate for swapping, so reclaim swap space. */ |
1098 | if (PageSwapCache(page) && vm_swap_full()) | 1098 | if (PageSwapCache(page) && vm_swap_full()) |
1099 | try_to_free_swap(page); | 1099 | try_to_free_swap(page); |
1100 | VM_BUG_ON(PageActive(page)); | 1100 | VM_BUG_ON(PageActive(page)); |
1101 | SetPageActive(page); | 1101 | SetPageActive(page); |
1102 | pgactivate++; | 1102 | pgactivate++; |
1103 | keep_locked: | 1103 | keep_locked: |
1104 | unlock_page(page); | 1104 | unlock_page(page); |
1105 | keep: | 1105 | keep: |
1106 | list_add(&page->lru, &ret_pages); | 1106 | list_add(&page->lru, &ret_pages); |
1107 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); | 1107 | VM_BUG_ON(PageLRU(page) || PageUnevictable(page)); |
1108 | } | 1108 | } |
1109 | 1109 | ||
1110 | free_hot_cold_page_list(&free_pages, true); | 1110 | free_hot_cold_page_list(&free_pages, true); |
1111 | 1111 | ||
1112 | list_splice(&ret_pages, page_list); | 1112 | list_splice(&ret_pages, page_list); |
1113 | count_vm_events(PGACTIVATE, pgactivate); | 1113 | count_vm_events(PGACTIVATE, pgactivate); |
1114 | mem_cgroup_uncharge_end(); | 1114 | mem_cgroup_uncharge_end(); |
1115 | *ret_nr_dirty += nr_dirty; | 1115 | *ret_nr_dirty += nr_dirty; |
1116 | *ret_nr_congested += nr_congested; | 1116 | *ret_nr_congested += nr_congested; |
1117 | *ret_nr_unqueued_dirty += nr_unqueued_dirty; | 1117 | *ret_nr_unqueued_dirty += nr_unqueued_dirty; |
1118 | *ret_nr_writeback += nr_writeback; | 1118 | *ret_nr_writeback += nr_writeback; |
1119 | *ret_nr_immediate += nr_immediate; | 1119 | *ret_nr_immediate += nr_immediate; |
1120 | return nr_reclaimed; | 1120 | return nr_reclaimed; |
1121 | } | 1121 | } |
1122 | 1122 | ||
1123 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, | 1123 | unsigned long reclaim_clean_pages_from_list(struct zone *zone, |
1124 | struct list_head *page_list) | 1124 | struct list_head *page_list) |
1125 | { | 1125 | { |
1126 | struct scan_control sc = { | 1126 | struct scan_control sc = { |
1127 | .gfp_mask = GFP_KERNEL, | 1127 | .gfp_mask = GFP_KERNEL, |
1128 | .priority = DEF_PRIORITY, | 1128 | .priority = DEF_PRIORITY, |
1129 | .may_unmap = 1, | 1129 | .may_unmap = 1, |
1130 | }; | 1130 | }; |
1131 | unsigned long ret, dummy1, dummy2, dummy3, dummy4, dummy5; | 1131 | unsigned long ret, dummy1, dummy2, dummy3, dummy4, dummy5; |
1132 | struct page *page, *next; | 1132 | struct page *page, *next; |
1133 | LIST_HEAD(clean_pages); | 1133 | LIST_HEAD(clean_pages); |
1134 | 1134 | ||
1135 | list_for_each_entry_safe(page, next, page_list, lru) { | 1135 | list_for_each_entry_safe(page, next, page_list, lru) { |
1136 | if (page_is_file_cache(page) && !PageDirty(page) && | 1136 | if (page_is_file_cache(page) && !PageDirty(page) && |
1137 | !isolated_balloon_page(page)) { | 1137 | !isolated_balloon_page(page)) { |
1138 | ClearPageActive(page); | 1138 | ClearPageActive(page); |
1139 | list_move(&page->lru, &clean_pages); | 1139 | list_move(&page->lru, &clean_pages); |
1140 | } | 1140 | } |
1141 | } | 1141 | } |
1142 | 1142 | ||
1143 | ret = shrink_page_list(&clean_pages, zone, &sc, | 1143 | ret = shrink_page_list(&clean_pages, zone, &sc, |
1144 | TTU_UNMAP|TTU_IGNORE_ACCESS, | 1144 | TTU_UNMAP|TTU_IGNORE_ACCESS, |
1145 | &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, true); | 1145 | &dummy1, &dummy2, &dummy3, &dummy4, &dummy5, true); |
1146 | list_splice(&clean_pages, page_list); | 1146 | list_splice(&clean_pages, page_list); |
1147 | mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret); | 1147 | mod_zone_page_state(zone, NR_ISOLATED_FILE, -ret); |
1148 | return ret; | 1148 | return ret; |
1149 | } | 1149 | } |
1150 | 1150 | ||
1151 | /* | 1151 | /* |
1152 | * Attempt to remove the specified page from its LRU. Only take this page | 1152 | * Attempt to remove the specified page from its LRU. Only take this page |
1153 | * if it is of the appropriate PageActive status. Pages which are being | 1153 | * if it is of the appropriate PageActive status. Pages which are being |
1154 | * freed elsewhere are also ignored. | 1154 | * freed elsewhere are also ignored. |
1155 | * | 1155 | * |
1156 | * page: page to consider | 1156 | * page: page to consider |
1157 | * mode: one of the LRU isolation modes defined above | 1157 | * mode: one of the LRU isolation modes defined above |
1158 | * | 1158 | * |
1159 | * returns 0 on success, -ve errno on failure. | 1159 | * returns 0 on success, -ve errno on failure. |
1160 | */ | 1160 | */ |
1161 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) | 1161 | int __isolate_lru_page(struct page *page, isolate_mode_t mode) |
1162 | { | 1162 | { |
1163 | int ret = -EINVAL; | 1163 | int ret = -EINVAL; |
1164 | 1164 | ||
1165 | /* Only take pages on the LRU. */ | 1165 | /* Only take pages on the LRU. */ |
1166 | if (!PageLRU(page)) | 1166 | if (!PageLRU(page)) |
1167 | return ret; | 1167 | return ret; |
1168 | 1168 | ||
1169 | /* Compaction should not handle unevictable pages but CMA can do so */ | 1169 | /* Compaction should not handle unevictable pages but CMA can do so */ |
1170 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) | 1170 | if (PageUnevictable(page) && !(mode & ISOLATE_UNEVICTABLE)) |
1171 | return ret; | 1171 | return ret; |
1172 | 1172 | ||
1173 | ret = -EBUSY; | 1173 | ret = -EBUSY; |
1174 | 1174 | ||
1175 | /* | 1175 | /* |
1176 | * To minimise LRU disruption, the caller can indicate that it only | 1176 | * To minimise LRU disruption, the caller can indicate that it only |
1177 | * wants to isolate pages it will be able to operate on without | 1177 | * wants to isolate pages it will be able to operate on without |
1178 | * blocking - clean pages for the most part. | 1178 | * blocking - clean pages for the most part. |
1179 | * | 1179 | * |
1180 | * ISOLATE_CLEAN means that only clean pages should be isolated. This | 1180 | * ISOLATE_CLEAN means that only clean pages should be isolated. This |
1181 | * is used by reclaim when it is cannot write to backing storage | 1181 | * is used by reclaim when it is cannot write to backing storage |
1182 | * | 1182 | * |
1183 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages | 1183 | * ISOLATE_ASYNC_MIGRATE is used to indicate that it only wants to pages |
1184 | * that it is possible to migrate without blocking | 1184 | * that it is possible to migrate without blocking |
1185 | */ | 1185 | */ |
1186 | if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) { | 1186 | if (mode & (ISOLATE_CLEAN|ISOLATE_ASYNC_MIGRATE)) { |
1187 | /* All the caller can do on PageWriteback is block */ | 1187 | /* All the caller can do on PageWriteback is block */ |
1188 | if (PageWriteback(page)) | 1188 | if (PageWriteback(page)) |
1189 | return ret; | 1189 | return ret; |
1190 | 1190 | ||
1191 | if (PageDirty(page)) { | 1191 | if (PageDirty(page)) { |
1192 | struct address_space *mapping; | 1192 | struct address_space *mapping; |
1193 | 1193 | ||
1194 | /* ISOLATE_CLEAN means only clean pages */ | 1194 | /* ISOLATE_CLEAN means only clean pages */ |
1195 | if (mode & ISOLATE_CLEAN) | 1195 | if (mode & ISOLATE_CLEAN) |
1196 | return ret; | 1196 | return ret; |
1197 | 1197 | ||
1198 | /* | 1198 | /* |
1199 | * Only pages without mappings or that have a | 1199 | * Only pages without mappings or that have a |
1200 | * ->migratepage callback are possible to migrate | 1200 | * ->migratepage callback are possible to migrate |
1201 | * without blocking | 1201 | * without blocking |
1202 | */ | 1202 | */ |
1203 | mapping = page_mapping(page); | 1203 | mapping = page_mapping(page); |
1204 | if (mapping && !mapping->a_ops->migratepage) | 1204 | if (mapping && !mapping->a_ops->migratepage) |
1205 | return ret; | 1205 | return ret; |
1206 | } | 1206 | } |
1207 | } | 1207 | } |
1208 | 1208 | ||
1209 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) | 1209 | if ((mode & ISOLATE_UNMAPPED) && page_mapped(page)) |
1210 | return ret; | 1210 | return ret; |
1211 | 1211 | ||
1212 | if (likely(get_page_unless_zero(page))) { | 1212 | if (likely(get_page_unless_zero(page))) { |
1213 | /* | 1213 | /* |
1214 | * Be careful not to clear PageLRU until after we're | 1214 | * Be careful not to clear PageLRU until after we're |
1215 | * sure the page is not being freed elsewhere -- the | 1215 | * sure the page is not being freed elsewhere -- the |
1216 | * page release code relies on it. | 1216 | * page release code relies on it. |
1217 | */ | 1217 | */ |
1218 | ClearPageLRU(page); | 1218 | ClearPageLRU(page); |
1219 | ret = 0; | 1219 | ret = 0; |
1220 | } | 1220 | } |
1221 | 1221 | ||
1222 | return ret; | 1222 | return ret; |
1223 | } | 1223 | } |
1224 | 1224 | ||
1225 | /* | 1225 | /* |
1226 | * zone->lru_lock is heavily contended. Some of the functions that | 1226 | * zone->lru_lock is heavily contended. Some of the functions that |
1227 | * shrink the lists perform better by taking out a batch of pages | 1227 | * shrink the lists perform better by taking out a batch of pages |
1228 | * and working on them outside the LRU lock. | 1228 | * and working on them outside the LRU lock. |
1229 | * | 1229 | * |
1230 | * For pagecache intensive workloads, this function is the hottest | 1230 | * For pagecache intensive workloads, this function is the hottest |
1231 | * spot in the kernel (apart from copy_*_user functions). | 1231 | * spot in the kernel (apart from copy_*_user functions). |
1232 | * | 1232 | * |
1233 | * Appropriate locks must be held before calling this function. | 1233 | * Appropriate locks must be held before calling this function. |
1234 | * | 1234 | * |
1235 | * @nr_to_scan: The number of pages to look through on the list. | 1235 | * @nr_to_scan: The number of pages to look through on the list. |
1236 | * @lruvec: The LRU vector to pull pages from. | 1236 | * @lruvec: The LRU vector to pull pages from. |
1237 | * @dst: The temp list to put pages on to. | 1237 | * @dst: The temp list to put pages on to. |
1238 | * @nr_scanned: The number of pages that were scanned. | 1238 | * @nr_scanned: The number of pages that were scanned. |
1239 | * @sc: The scan_control struct for this reclaim session | 1239 | * @sc: The scan_control struct for this reclaim session |
1240 | * @mode: One of the LRU isolation modes | 1240 | * @mode: One of the LRU isolation modes |
1241 | * @lru: LRU list id for isolating | 1241 | * @lru: LRU list id for isolating |
1242 | * | 1242 | * |
1243 | * returns how many pages were moved onto *@dst. | 1243 | * returns how many pages were moved onto *@dst. |
1244 | */ | 1244 | */ |
1245 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, | 1245 | static unsigned long isolate_lru_pages(unsigned long nr_to_scan, |
1246 | struct lruvec *lruvec, struct list_head *dst, | 1246 | struct lruvec *lruvec, struct list_head *dst, |
1247 | unsigned long *nr_scanned, struct scan_control *sc, | 1247 | unsigned long *nr_scanned, struct scan_control *sc, |
1248 | isolate_mode_t mode, enum lru_list lru) | 1248 | isolate_mode_t mode, enum lru_list lru) |
1249 | { | 1249 | { |
1250 | struct list_head *src = &lruvec->lists[lru]; | 1250 | struct list_head *src = &lruvec->lists[lru]; |
1251 | unsigned long nr_taken = 0; | 1251 | unsigned long nr_taken = 0; |
1252 | unsigned long scan; | 1252 | unsigned long scan; |
1253 | 1253 | ||
1254 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { | 1254 | for (scan = 0; scan < nr_to_scan && !list_empty(src); scan++) { |
1255 | struct page *page; | 1255 | struct page *page; |
1256 | int nr_pages; | 1256 | int nr_pages; |
1257 | 1257 | ||
1258 | page = lru_to_page(src); | 1258 | page = lru_to_page(src); |
1259 | prefetchw_prev_lru_page(page, src, flags); | 1259 | prefetchw_prev_lru_page(page, src, flags); |
1260 | 1260 | ||
1261 | VM_BUG_ON(!PageLRU(page)); | 1261 | VM_BUG_ON(!PageLRU(page)); |
1262 | 1262 | ||
1263 | switch (__isolate_lru_page(page, mode)) { | 1263 | switch (__isolate_lru_page(page, mode)) { |
1264 | case 0: | 1264 | case 0: |
1265 | nr_pages = hpage_nr_pages(page); | 1265 | nr_pages = hpage_nr_pages(page); |
1266 | mem_cgroup_update_lru_size(lruvec, lru, -nr_pages); | 1266 | mem_cgroup_update_lru_size(lruvec, lru, -nr_pages); |
1267 | list_move(&page->lru, dst); | 1267 | list_move(&page->lru, dst); |
1268 | nr_taken += nr_pages; | 1268 | nr_taken += nr_pages; |
1269 | break; | 1269 | break; |
1270 | 1270 | ||
1271 | case -EBUSY: | 1271 | case -EBUSY: |
1272 | /* else it is being freed elsewhere */ | 1272 | /* else it is being freed elsewhere */ |
1273 | list_move(&page->lru, src); | 1273 | list_move(&page->lru, src); |
1274 | continue; | 1274 | continue; |
1275 | 1275 | ||
1276 | default: | 1276 | default: |
1277 | BUG(); | 1277 | BUG(); |
1278 | } | 1278 | } |
1279 | } | 1279 | } |
1280 | 1280 | ||
1281 | *nr_scanned = scan; | 1281 | *nr_scanned = scan; |
1282 | trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan, | 1282 | trace_mm_vmscan_lru_isolate(sc->order, nr_to_scan, scan, |
1283 | nr_taken, mode, is_file_lru(lru)); | 1283 | nr_taken, mode, is_file_lru(lru)); |
1284 | return nr_taken; | 1284 | return nr_taken; |
1285 | } | 1285 | } |
1286 | 1286 | ||
1287 | /** | 1287 | /** |
1288 | * isolate_lru_page - tries to isolate a page from its LRU list | 1288 | * isolate_lru_page - tries to isolate a page from its LRU list |
1289 | * @page: page to isolate from its LRU list | 1289 | * @page: page to isolate from its LRU list |
1290 | * | 1290 | * |
1291 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the | 1291 | * Isolates a @page from an LRU list, clears PageLRU and adjusts the |
1292 | * vmstat statistic corresponding to whatever LRU list the page was on. | 1292 | * vmstat statistic corresponding to whatever LRU list the page was on. |
1293 | * | 1293 | * |
1294 | * Returns 0 if the page was removed from an LRU list. | 1294 | * Returns 0 if the page was removed from an LRU list. |
1295 | * Returns -EBUSY if the page was not on an LRU list. | 1295 | * Returns -EBUSY if the page was not on an LRU list. |
1296 | * | 1296 | * |
1297 | * The returned page will have PageLRU() cleared. If it was found on | 1297 | * The returned page will have PageLRU() cleared. If it was found on |
1298 | * the active list, it will have PageActive set. If it was found on | 1298 | * the active list, it will have PageActive set. If it was found on |
1299 | * the unevictable list, it will have the PageUnevictable bit set. That flag | 1299 | * the unevictable list, it will have the PageUnevictable bit set. That flag |
1300 | * may need to be cleared by the caller before letting the page go. | 1300 | * may need to be cleared by the caller before letting the page go. |
1301 | * | 1301 | * |
1302 | * The vmstat statistic corresponding to the list on which the page was | 1302 | * The vmstat statistic corresponding to the list on which the page was |
1303 | * found will be decremented. | 1303 | * found will be decremented. |
1304 | * | 1304 | * |
1305 | * Restrictions: | 1305 | * Restrictions: |
1306 | * (1) Must be called with an elevated refcount on the page. This is a | 1306 | * (1) Must be called with an elevated refcount on the page. This is a |
1307 | * fundamentnal difference from isolate_lru_pages (which is called | 1307 | * fundamentnal difference from isolate_lru_pages (which is called |
1308 | * without a stable reference). | 1308 | * without a stable reference). |
1309 | * (2) the lru_lock must not be held. | 1309 | * (2) the lru_lock must not be held. |
1310 | * (3) interrupts must be enabled. | 1310 | * (3) interrupts must be enabled. |
1311 | */ | 1311 | */ |
1312 | int isolate_lru_page(struct page *page) | 1312 | int isolate_lru_page(struct page *page) |
1313 | { | 1313 | { |
1314 | int ret = -EBUSY; | 1314 | int ret = -EBUSY; |
1315 | 1315 | ||
1316 | VM_BUG_ON(!page_count(page)); | 1316 | VM_BUG_ON(!page_count(page)); |
1317 | 1317 | ||
1318 | if (PageLRU(page)) { | 1318 | if (PageLRU(page)) { |
1319 | struct zone *zone = page_zone(page); | 1319 | struct zone *zone = page_zone(page); |
1320 | struct lruvec *lruvec; | 1320 | struct lruvec *lruvec; |
1321 | 1321 | ||
1322 | spin_lock_irq(&zone->lru_lock); | 1322 | spin_lock_irq(&zone->lru_lock); |
1323 | lruvec = mem_cgroup_page_lruvec(page, zone); | 1323 | lruvec = mem_cgroup_page_lruvec(page, zone); |
1324 | if (PageLRU(page)) { | 1324 | if (PageLRU(page)) { |
1325 | int lru = page_lru(page); | 1325 | int lru = page_lru(page); |
1326 | get_page(page); | 1326 | get_page(page); |
1327 | ClearPageLRU(page); | 1327 | ClearPageLRU(page); |
1328 | del_page_from_lru_list(page, lruvec, lru); | 1328 | del_page_from_lru_list(page, lruvec, lru); |
1329 | ret = 0; | 1329 | ret = 0; |
1330 | } | 1330 | } |
1331 | spin_unlock_irq(&zone->lru_lock); | 1331 | spin_unlock_irq(&zone->lru_lock); |
1332 | } | 1332 | } |
1333 | return ret; | 1333 | return ret; |
1334 | } | 1334 | } |
1335 | 1335 | ||
1336 | /* | 1336 | /* |
1337 | * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and | 1337 | * A direct reclaimer may isolate SWAP_CLUSTER_MAX pages from the LRU list and |
1338 | * then get resheduled. When there are massive number of tasks doing page | 1338 | * then get resheduled. When there are massive number of tasks doing page |
1339 | * allocation, such sleeping direct reclaimers may keep piling up on each CPU, | 1339 | * allocation, such sleeping direct reclaimers may keep piling up on each CPU, |
1340 | * the LRU list will go small and be scanned faster than necessary, leading to | 1340 | * the LRU list will go small and be scanned faster than necessary, leading to |
1341 | * unnecessary swapping, thrashing and OOM. | 1341 | * unnecessary swapping, thrashing and OOM. |
1342 | */ | 1342 | */ |
1343 | static int too_many_isolated(struct zone *zone, int file, | 1343 | static int too_many_isolated(struct zone *zone, int file, |
1344 | struct scan_control *sc) | 1344 | struct scan_control *sc) |
1345 | { | 1345 | { |
1346 | unsigned long inactive, isolated; | 1346 | unsigned long inactive, isolated; |
1347 | 1347 | ||
1348 | if (current_is_kswapd()) | 1348 | if (current_is_kswapd()) |
1349 | return 0; | 1349 | return 0; |
1350 | 1350 | ||
1351 | if (!global_reclaim(sc)) | 1351 | if (!global_reclaim(sc)) |
1352 | return 0; | 1352 | return 0; |
1353 | 1353 | ||
1354 | if (file) { | 1354 | if (file) { |
1355 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); | 1355 | inactive = zone_page_state(zone, NR_INACTIVE_FILE); |
1356 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); | 1356 | isolated = zone_page_state(zone, NR_ISOLATED_FILE); |
1357 | } else { | 1357 | } else { |
1358 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | 1358 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
1359 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); | 1359 | isolated = zone_page_state(zone, NR_ISOLATED_ANON); |
1360 | } | 1360 | } |
1361 | 1361 | ||
1362 | /* | 1362 | /* |
1363 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they | 1363 | * GFP_NOIO/GFP_NOFS callers are allowed to isolate more pages, so they |
1364 | * won't get blocked by normal direct-reclaimers, forming a circular | 1364 | * won't get blocked by normal direct-reclaimers, forming a circular |
1365 | * deadlock. | 1365 | * deadlock. |
1366 | */ | 1366 | */ |
1367 | if ((sc->gfp_mask & GFP_IOFS) == GFP_IOFS) | 1367 | if ((sc->gfp_mask & GFP_IOFS) == GFP_IOFS) |
1368 | inactive >>= 3; | 1368 | inactive >>= 3; |
1369 | 1369 | ||
1370 | return isolated > inactive; | 1370 | return isolated > inactive; |
1371 | } | 1371 | } |
1372 | 1372 | ||
1373 | static noinline_for_stack void | 1373 | static noinline_for_stack void |
1374 | putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list) | 1374 | putback_inactive_pages(struct lruvec *lruvec, struct list_head *page_list) |
1375 | { | 1375 | { |
1376 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; | 1376 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1377 | struct zone *zone = lruvec_zone(lruvec); | 1377 | struct zone *zone = lruvec_zone(lruvec); |
1378 | LIST_HEAD(pages_to_free); | 1378 | LIST_HEAD(pages_to_free); |
1379 | 1379 | ||
1380 | /* | 1380 | /* |
1381 | * Put back any unfreeable pages. | 1381 | * Put back any unfreeable pages. |
1382 | */ | 1382 | */ |
1383 | while (!list_empty(page_list)) { | 1383 | while (!list_empty(page_list)) { |
1384 | struct page *page = lru_to_page(page_list); | 1384 | struct page *page = lru_to_page(page_list); |
1385 | int lru; | 1385 | int lru; |
1386 | 1386 | ||
1387 | VM_BUG_ON(PageLRU(page)); | 1387 | VM_BUG_ON(PageLRU(page)); |
1388 | list_del(&page->lru); | 1388 | list_del(&page->lru); |
1389 | if (unlikely(!page_evictable(page))) { | 1389 | if (unlikely(!page_evictable(page))) { |
1390 | spin_unlock_irq(&zone->lru_lock); | 1390 | spin_unlock_irq(&zone->lru_lock); |
1391 | putback_lru_page(page); | 1391 | putback_lru_page(page); |
1392 | spin_lock_irq(&zone->lru_lock); | 1392 | spin_lock_irq(&zone->lru_lock); |
1393 | continue; | 1393 | continue; |
1394 | } | 1394 | } |
1395 | 1395 | ||
1396 | lruvec = mem_cgroup_page_lruvec(page, zone); | 1396 | lruvec = mem_cgroup_page_lruvec(page, zone); |
1397 | 1397 | ||
1398 | SetPageLRU(page); | 1398 | SetPageLRU(page); |
1399 | lru = page_lru(page); | 1399 | lru = page_lru(page); |
1400 | add_page_to_lru_list(page, lruvec, lru); | 1400 | add_page_to_lru_list(page, lruvec, lru); |
1401 | 1401 | ||
1402 | if (is_active_lru(lru)) { | 1402 | if (is_active_lru(lru)) { |
1403 | int file = is_file_lru(lru); | 1403 | int file = is_file_lru(lru); |
1404 | int numpages = hpage_nr_pages(page); | 1404 | int numpages = hpage_nr_pages(page); |
1405 | reclaim_stat->recent_rotated[file] += numpages; | 1405 | reclaim_stat->recent_rotated[file] += numpages; |
1406 | } | 1406 | } |
1407 | if (put_page_testzero(page)) { | 1407 | if (put_page_testzero(page)) { |
1408 | __ClearPageLRU(page); | 1408 | __ClearPageLRU(page); |
1409 | __ClearPageActive(page); | 1409 | __ClearPageActive(page); |
1410 | del_page_from_lru_list(page, lruvec, lru); | 1410 | del_page_from_lru_list(page, lruvec, lru); |
1411 | 1411 | ||
1412 | if (unlikely(PageCompound(page))) { | 1412 | if (unlikely(PageCompound(page))) { |
1413 | spin_unlock_irq(&zone->lru_lock); | 1413 | spin_unlock_irq(&zone->lru_lock); |
1414 | (*get_compound_page_dtor(page))(page); | 1414 | (*get_compound_page_dtor(page))(page); |
1415 | spin_lock_irq(&zone->lru_lock); | 1415 | spin_lock_irq(&zone->lru_lock); |
1416 | } else | 1416 | } else |
1417 | list_add(&page->lru, &pages_to_free); | 1417 | list_add(&page->lru, &pages_to_free); |
1418 | } | 1418 | } |
1419 | } | 1419 | } |
1420 | 1420 | ||
1421 | /* | 1421 | /* |
1422 | * To save our caller's stack, now use input list for pages to free. | 1422 | * To save our caller's stack, now use input list for pages to free. |
1423 | */ | 1423 | */ |
1424 | list_splice(&pages_to_free, page_list); | 1424 | list_splice(&pages_to_free, page_list); |
1425 | } | 1425 | } |
1426 | 1426 | ||
1427 | /* | 1427 | /* |
1428 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number | 1428 | * shrink_inactive_list() is a helper for shrink_zone(). It returns the number |
1429 | * of reclaimed pages | 1429 | * of reclaimed pages |
1430 | */ | 1430 | */ |
1431 | static noinline_for_stack unsigned long | 1431 | static noinline_for_stack unsigned long |
1432 | shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec, | 1432 | shrink_inactive_list(unsigned long nr_to_scan, struct lruvec *lruvec, |
1433 | struct scan_control *sc, enum lru_list lru) | 1433 | struct scan_control *sc, enum lru_list lru) |
1434 | { | 1434 | { |
1435 | LIST_HEAD(page_list); | 1435 | LIST_HEAD(page_list); |
1436 | unsigned long nr_scanned; | 1436 | unsigned long nr_scanned; |
1437 | unsigned long nr_reclaimed = 0; | 1437 | unsigned long nr_reclaimed = 0; |
1438 | unsigned long nr_taken; | 1438 | unsigned long nr_taken; |
1439 | unsigned long nr_dirty = 0; | 1439 | unsigned long nr_dirty = 0; |
1440 | unsigned long nr_congested = 0; | 1440 | unsigned long nr_congested = 0; |
1441 | unsigned long nr_unqueued_dirty = 0; | 1441 | unsigned long nr_unqueued_dirty = 0; |
1442 | unsigned long nr_writeback = 0; | 1442 | unsigned long nr_writeback = 0; |
1443 | unsigned long nr_immediate = 0; | 1443 | unsigned long nr_immediate = 0; |
1444 | isolate_mode_t isolate_mode = 0; | 1444 | isolate_mode_t isolate_mode = 0; |
1445 | int file = is_file_lru(lru); | 1445 | int file = is_file_lru(lru); |
1446 | struct zone *zone = lruvec_zone(lruvec); | 1446 | struct zone *zone = lruvec_zone(lruvec); |
1447 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; | 1447 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1448 | 1448 | ||
1449 | while (unlikely(too_many_isolated(zone, file, sc))) { | 1449 | while (unlikely(too_many_isolated(zone, file, sc))) { |
1450 | congestion_wait(BLK_RW_ASYNC, HZ/10); | 1450 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
1451 | 1451 | ||
1452 | /* We are about to die and free our memory. Return now. */ | 1452 | /* We are about to die and free our memory. Return now. */ |
1453 | if (fatal_signal_pending(current)) | 1453 | if (fatal_signal_pending(current)) |
1454 | return SWAP_CLUSTER_MAX; | 1454 | return SWAP_CLUSTER_MAX; |
1455 | } | 1455 | } |
1456 | 1456 | ||
1457 | lru_add_drain(); | 1457 | lru_add_drain(); |
1458 | 1458 | ||
1459 | if (!sc->may_unmap) | 1459 | if (!sc->may_unmap) |
1460 | isolate_mode |= ISOLATE_UNMAPPED; | 1460 | isolate_mode |= ISOLATE_UNMAPPED; |
1461 | if (!sc->may_writepage) | 1461 | if (!sc->may_writepage) |
1462 | isolate_mode |= ISOLATE_CLEAN; | 1462 | isolate_mode |= ISOLATE_CLEAN; |
1463 | 1463 | ||
1464 | spin_lock_irq(&zone->lru_lock); | 1464 | spin_lock_irq(&zone->lru_lock); |
1465 | 1465 | ||
1466 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, | 1466 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &page_list, |
1467 | &nr_scanned, sc, isolate_mode, lru); | 1467 | &nr_scanned, sc, isolate_mode, lru); |
1468 | 1468 | ||
1469 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); | 1469 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
1470 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); | 1470 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
1471 | 1471 | ||
1472 | if (global_reclaim(sc)) { | 1472 | if (global_reclaim(sc)) { |
1473 | zone->pages_scanned += nr_scanned; | 1473 | zone->pages_scanned += nr_scanned; |
1474 | if (current_is_kswapd()) | 1474 | if (current_is_kswapd()) |
1475 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned); | 1475 | __count_zone_vm_events(PGSCAN_KSWAPD, zone, nr_scanned); |
1476 | else | 1476 | else |
1477 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned); | 1477 | __count_zone_vm_events(PGSCAN_DIRECT, zone, nr_scanned); |
1478 | } | 1478 | } |
1479 | spin_unlock_irq(&zone->lru_lock); | 1479 | spin_unlock_irq(&zone->lru_lock); |
1480 | 1480 | ||
1481 | if (nr_taken == 0) | 1481 | if (nr_taken == 0) |
1482 | return 0; | 1482 | return 0; |
1483 | 1483 | ||
1484 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP, | 1484 | nr_reclaimed = shrink_page_list(&page_list, zone, sc, TTU_UNMAP, |
1485 | &nr_dirty, &nr_unqueued_dirty, &nr_congested, | 1485 | &nr_dirty, &nr_unqueued_dirty, &nr_congested, |
1486 | &nr_writeback, &nr_immediate, | 1486 | &nr_writeback, &nr_immediate, |
1487 | false); | 1487 | false); |
1488 | 1488 | ||
1489 | spin_lock_irq(&zone->lru_lock); | 1489 | spin_lock_irq(&zone->lru_lock); |
1490 | 1490 | ||
1491 | reclaim_stat->recent_scanned[file] += nr_taken; | 1491 | reclaim_stat->recent_scanned[file] += nr_taken; |
1492 | 1492 | ||
1493 | if (global_reclaim(sc)) { | 1493 | if (global_reclaim(sc)) { |
1494 | if (current_is_kswapd()) | 1494 | if (current_is_kswapd()) |
1495 | __count_zone_vm_events(PGSTEAL_KSWAPD, zone, | 1495 | __count_zone_vm_events(PGSTEAL_KSWAPD, zone, |
1496 | nr_reclaimed); | 1496 | nr_reclaimed); |
1497 | else | 1497 | else |
1498 | __count_zone_vm_events(PGSTEAL_DIRECT, zone, | 1498 | __count_zone_vm_events(PGSTEAL_DIRECT, zone, |
1499 | nr_reclaimed); | 1499 | nr_reclaimed); |
1500 | } | 1500 | } |
1501 | 1501 | ||
1502 | putback_inactive_pages(lruvec, &page_list); | 1502 | putback_inactive_pages(lruvec, &page_list); |
1503 | 1503 | ||
1504 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); | 1504 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
1505 | 1505 | ||
1506 | spin_unlock_irq(&zone->lru_lock); | 1506 | spin_unlock_irq(&zone->lru_lock); |
1507 | 1507 | ||
1508 | free_hot_cold_page_list(&page_list, true); | 1508 | free_hot_cold_page_list(&page_list, true); |
1509 | 1509 | ||
1510 | /* | 1510 | /* |
1511 | * If reclaim is isolating dirty pages under writeback, it implies | 1511 | * If reclaim is isolating dirty pages under writeback, it implies |
1512 | * that the long-lived page allocation rate is exceeding the page | 1512 | * that the long-lived page allocation rate is exceeding the page |
1513 | * laundering rate. Either the global limits are not being effective | 1513 | * laundering rate. Either the global limits are not being effective |
1514 | * at throttling processes due to the page distribution throughout | 1514 | * at throttling processes due to the page distribution throughout |
1515 | * zones or there is heavy usage of a slow backing device. The | 1515 | * zones or there is heavy usage of a slow backing device. The |
1516 | * only option is to throttle from reclaim context which is not ideal | 1516 | * only option is to throttle from reclaim context which is not ideal |
1517 | * as there is no guarantee the dirtying process is throttled in the | 1517 | * as there is no guarantee the dirtying process is throttled in the |
1518 | * same way balance_dirty_pages() manages. | 1518 | * same way balance_dirty_pages() manages. |
1519 | * | 1519 | * |
1520 | * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number | 1520 | * Once a zone is flagged ZONE_WRITEBACK, kswapd will count the number |
1521 | * of pages under pages flagged for immediate reclaim and stall if any | 1521 | * of pages under pages flagged for immediate reclaim and stall if any |
1522 | * are encountered in the nr_immediate check below. | 1522 | * are encountered in the nr_immediate check below. |
1523 | */ | 1523 | */ |
1524 | if (nr_writeback && nr_writeback == nr_taken) | 1524 | if (nr_writeback && nr_writeback == nr_taken) |
1525 | zone_set_flag(zone, ZONE_WRITEBACK); | 1525 | zone_set_flag(zone, ZONE_WRITEBACK); |
1526 | 1526 | ||
1527 | /* | 1527 | /* |
1528 | * memcg will stall in page writeback so only consider forcibly | 1528 | * memcg will stall in page writeback so only consider forcibly |
1529 | * stalling for global reclaim | 1529 | * stalling for global reclaim |
1530 | */ | 1530 | */ |
1531 | if (global_reclaim(sc)) { | 1531 | if (global_reclaim(sc)) { |
1532 | /* | 1532 | /* |
1533 | * Tag a zone as congested if all the dirty pages scanned were | 1533 | * Tag a zone as congested if all the dirty pages scanned were |
1534 | * backed by a congested BDI and wait_iff_congested will stall. | 1534 | * backed by a congested BDI and wait_iff_congested will stall. |
1535 | */ | 1535 | */ |
1536 | if (nr_dirty && nr_dirty == nr_congested) | 1536 | if (nr_dirty && nr_dirty == nr_congested) |
1537 | zone_set_flag(zone, ZONE_CONGESTED); | 1537 | zone_set_flag(zone, ZONE_CONGESTED); |
1538 | 1538 | ||
1539 | /* | 1539 | /* |
1540 | * If dirty pages are scanned that are not queued for IO, it | 1540 | * If dirty pages are scanned that are not queued for IO, it |
1541 | * implies that flushers are not keeping up. In this case, flag | 1541 | * implies that flushers are not keeping up. In this case, flag |
1542 | * the zone ZONE_TAIL_LRU_DIRTY and kswapd will start writing | 1542 | * the zone ZONE_TAIL_LRU_DIRTY and kswapd will start writing |
1543 | * pages from reclaim context. | 1543 | * pages from reclaim context. |
1544 | */ | 1544 | */ |
1545 | if (nr_unqueued_dirty == nr_taken) | 1545 | if (nr_unqueued_dirty == nr_taken) |
1546 | zone_set_flag(zone, ZONE_TAIL_LRU_DIRTY); | 1546 | zone_set_flag(zone, ZONE_TAIL_LRU_DIRTY); |
1547 | 1547 | ||
1548 | /* | 1548 | /* |
1549 | * If kswapd scans pages marked marked for immediate | 1549 | * If kswapd scans pages marked marked for immediate |
1550 | * reclaim and under writeback (nr_immediate), it implies | 1550 | * reclaim and under writeback (nr_immediate), it implies |
1551 | * that pages are cycling through the LRU faster than | 1551 | * that pages are cycling through the LRU faster than |
1552 | * they are written so also forcibly stall. | 1552 | * they are written so also forcibly stall. |
1553 | */ | 1553 | */ |
1554 | if (nr_immediate) | 1554 | if (nr_immediate) |
1555 | congestion_wait(BLK_RW_ASYNC, HZ/10); | 1555 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
1556 | } | 1556 | } |
1557 | 1557 | ||
1558 | /* | 1558 | /* |
1559 | * Stall direct reclaim for IO completions if underlying BDIs or zone | 1559 | * Stall direct reclaim for IO completions if underlying BDIs or zone |
1560 | * is congested. Allow kswapd to continue until it starts encountering | 1560 | * is congested. Allow kswapd to continue until it starts encountering |
1561 | * unqueued dirty pages or cycling through the LRU too quickly. | 1561 | * unqueued dirty pages or cycling through the LRU too quickly. |
1562 | */ | 1562 | */ |
1563 | if (!sc->hibernation_mode && !current_is_kswapd()) | 1563 | if (!sc->hibernation_mode && !current_is_kswapd()) |
1564 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); | 1564 | wait_iff_congested(zone, BLK_RW_ASYNC, HZ/10); |
1565 | 1565 | ||
1566 | trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id, | 1566 | trace_mm_vmscan_lru_shrink_inactive(zone->zone_pgdat->node_id, |
1567 | zone_idx(zone), | 1567 | zone_idx(zone), |
1568 | nr_scanned, nr_reclaimed, | 1568 | nr_scanned, nr_reclaimed, |
1569 | sc->priority, | 1569 | sc->priority, |
1570 | trace_shrink_flags(file)); | 1570 | trace_shrink_flags(file)); |
1571 | return nr_reclaimed; | 1571 | return nr_reclaimed; |
1572 | } | 1572 | } |
1573 | 1573 | ||
1574 | /* | 1574 | /* |
1575 | * This moves pages from the active list to the inactive list. | 1575 | * This moves pages from the active list to the inactive list. |
1576 | * | 1576 | * |
1577 | * We move them the other way if the page is referenced by one or more | 1577 | * We move them the other way if the page is referenced by one or more |
1578 | * processes, from rmap. | 1578 | * processes, from rmap. |
1579 | * | 1579 | * |
1580 | * If the pages are mostly unmapped, the processing is fast and it is | 1580 | * If the pages are mostly unmapped, the processing is fast and it is |
1581 | * appropriate to hold zone->lru_lock across the whole operation. But if | 1581 | * appropriate to hold zone->lru_lock across the whole operation. But if |
1582 | * the pages are mapped, the processing is slow (page_referenced()) so we | 1582 | * the pages are mapped, the processing is slow (page_referenced()) so we |
1583 | * should drop zone->lru_lock around each page. It's impossible to balance | 1583 | * should drop zone->lru_lock around each page. It's impossible to balance |
1584 | * this, so instead we remove the pages from the LRU while processing them. | 1584 | * this, so instead we remove the pages from the LRU while processing them. |
1585 | * It is safe to rely on PG_active against the non-LRU pages in here because | 1585 | * It is safe to rely on PG_active against the non-LRU pages in here because |
1586 | * nobody will play with that bit on a non-LRU page. | 1586 | * nobody will play with that bit on a non-LRU page. |
1587 | * | 1587 | * |
1588 | * The downside is that we have to touch page->_count against each page. | 1588 | * The downside is that we have to touch page->_count against each page. |
1589 | * But we had to alter page->flags anyway. | 1589 | * But we had to alter page->flags anyway. |
1590 | */ | 1590 | */ |
1591 | 1591 | ||
1592 | static void move_active_pages_to_lru(struct lruvec *lruvec, | 1592 | static void move_active_pages_to_lru(struct lruvec *lruvec, |
1593 | struct list_head *list, | 1593 | struct list_head *list, |
1594 | struct list_head *pages_to_free, | 1594 | struct list_head *pages_to_free, |
1595 | enum lru_list lru) | 1595 | enum lru_list lru) |
1596 | { | 1596 | { |
1597 | struct zone *zone = lruvec_zone(lruvec); | 1597 | struct zone *zone = lruvec_zone(lruvec); |
1598 | unsigned long pgmoved = 0; | 1598 | unsigned long pgmoved = 0; |
1599 | struct page *page; | 1599 | struct page *page; |
1600 | int nr_pages; | 1600 | int nr_pages; |
1601 | 1601 | ||
1602 | while (!list_empty(list)) { | 1602 | while (!list_empty(list)) { |
1603 | page = lru_to_page(list); | 1603 | page = lru_to_page(list); |
1604 | lruvec = mem_cgroup_page_lruvec(page, zone); | 1604 | lruvec = mem_cgroup_page_lruvec(page, zone); |
1605 | 1605 | ||
1606 | VM_BUG_ON(PageLRU(page)); | 1606 | VM_BUG_ON(PageLRU(page)); |
1607 | SetPageLRU(page); | 1607 | SetPageLRU(page); |
1608 | 1608 | ||
1609 | nr_pages = hpage_nr_pages(page); | 1609 | nr_pages = hpage_nr_pages(page); |
1610 | mem_cgroup_update_lru_size(lruvec, lru, nr_pages); | 1610 | mem_cgroup_update_lru_size(lruvec, lru, nr_pages); |
1611 | list_move(&page->lru, &lruvec->lists[lru]); | 1611 | list_move(&page->lru, &lruvec->lists[lru]); |
1612 | pgmoved += nr_pages; | 1612 | pgmoved += nr_pages; |
1613 | 1613 | ||
1614 | if (put_page_testzero(page)) { | 1614 | if (put_page_testzero(page)) { |
1615 | __ClearPageLRU(page); | 1615 | __ClearPageLRU(page); |
1616 | __ClearPageActive(page); | 1616 | __ClearPageActive(page); |
1617 | del_page_from_lru_list(page, lruvec, lru); | 1617 | del_page_from_lru_list(page, lruvec, lru); |
1618 | 1618 | ||
1619 | if (unlikely(PageCompound(page))) { | 1619 | if (unlikely(PageCompound(page))) { |
1620 | spin_unlock_irq(&zone->lru_lock); | 1620 | spin_unlock_irq(&zone->lru_lock); |
1621 | (*get_compound_page_dtor(page))(page); | 1621 | (*get_compound_page_dtor(page))(page); |
1622 | spin_lock_irq(&zone->lru_lock); | 1622 | spin_lock_irq(&zone->lru_lock); |
1623 | } else | 1623 | } else |
1624 | list_add(&page->lru, pages_to_free); | 1624 | list_add(&page->lru, pages_to_free); |
1625 | } | 1625 | } |
1626 | } | 1626 | } |
1627 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); | 1627 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, pgmoved); |
1628 | if (!is_active_lru(lru)) | 1628 | if (!is_active_lru(lru)) |
1629 | __count_vm_events(PGDEACTIVATE, pgmoved); | 1629 | __count_vm_events(PGDEACTIVATE, pgmoved); |
1630 | } | 1630 | } |
1631 | 1631 | ||
1632 | static void shrink_active_list(unsigned long nr_to_scan, | 1632 | static void shrink_active_list(unsigned long nr_to_scan, |
1633 | struct lruvec *lruvec, | 1633 | struct lruvec *lruvec, |
1634 | struct scan_control *sc, | 1634 | struct scan_control *sc, |
1635 | enum lru_list lru) | 1635 | enum lru_list lru) |
1636 | { | 1636 | { |
1637 | unsigned long nr_taken; | 1637 | unsigned long nr_taken; |
1638 | unsigned long nr_scanned; | 1638 | unsigned long nr_scanned; |
1639 | unsigned long vm_flags; | 1639 | unsigned long vm_flags; |
1640 | LIST_HEAD(l_hold); /* The pages which were snipped off */ | 1640 | LIST_HEAD(l_hold); /* The pages which were snipped off */ |
1641 | LIST_HEAD(l_active); | 1641 | LIST_HEAD(l_active); |
1642 | LIST_HEAD(l_inactive); | 1642 | LIST_HEAD(l_inactive); |
1643 | struct page *page; | 1643 | struct page *page; |
1644 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; | 1644 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1645 | unsigned long nr_rotated = 0; | 1645 | unsigned long nr_rotated = 0; |
1646 | isolate_mode_t isolate_mode = 0; | 1646 | isolate_mode_t isolate_mode = 0; |
1647 | int file = is_file_lru(lru); | 1647 | int file = is_file_lru(lru); |
1648 | struct zone *zone = lruvec_zone(lruvec); | 1648 | struct zone *zone = lruvec_zone(lruvec); |
1649 | 1649 | ||
1650 | lru_add_drain(); | 1650 | lru_add_drain(); |
1651 | 1651 | ||
1652 | if (!sc->may_unmap) | 1652 | if (!sc->may_unmap) |
1653 | isolate_mode |= ISOLATE_UNMAPPED; | 1653 | isolate_mode |= ISOLATE_UNMAPPED; |
1654 | if (!sc->may_writepage) | 1654 | if (!sc->may_writepage) |
1655 | isolate_mode |= ISOLATE_CLEAN; | 1655 | isolate_mode |= ISOLATE_CLEAN; |
1656 | 1656 | ||
1657 | spin_lock_irq(&zone->lru_lock); | 1657 | spin_lock_irq(&zone->lru_lock); |
1658 | 1658 | ||
1659 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, | 1659 | nr_taken = isolate_lru_pages(nr_to_scan, lruvec, &l_hold, |
1660 | &nr_scanned, sc, isolate_mode, lru); | 1660 | &nr_scanned, sc, isolate_mode, lru); |
1661 | if (global_reclaim(sc)) | 1661 | if (global_reclaim(sc)) |
1662 | zone->pages_scanned += nr_scanned; | 1662 | zone->pages_scanned += nr_scanned; |
1663 | 1663 | ||
1664 | reclaim_stat->recent_scanned[file] += nr_taken; | 1664 | reclaim_stat->recent_scanned[file] += nr_taken; |
1665 | 1665 | ||
1666 | __count_zone_vm_events(PGREFILL, zone, nr_scanned); | 1666 | __count_zone_vm_events(PGREFILL, zone, nr_scanned); |
1667 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); | 1667 | __mod_zone_page_state(zone, NR_LRU_BASE + lru, -nr_taken); |
1668 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); | 1668 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, nr_taken); |
1669 | spin_unlock_irq(&zone->lru_lock); | 1669 | spin_unlock_irq(&zone->lru_lock); |
1670 | 1670 | ||
1671 | while (!list_empty(&l_hold)) { | 1671 | while (!list_empty(&l_hold)) { |
1672 | cond_resched(); | 1672 | cond_resched(); |
1673 | page = lru_to_page(&l_hold); | 1673 | page = lru_to_page(&l_hold); |
1674 | list_del(&page->lru); | 1674 | list_del(&page->lru); |
1675 | 1675 | ||
1676 | if (unlikely(!page_evictable(page))) { | 1676 | if (unlikely(!page_evictable(page))) { |
1677 | putback_lru_page(page); | 1677 | putback_lru_page(page); |
1678 | continue; | 1678 | continue; |
1679 | } | 1679 | } |
1680 | 1680 | ||
1681 | if (unlikely(buffer_heads_over_limit)) { | 1681 | if (unlikely(buffer_heads_over_limit)) { |
1682 | if (page_has_private(page) && trylock_page(page)) { | 1682 | if (page_has_private(page) && trylock_page(page)) { |
1683 | if (page_has_private(page)) | 1683 | if (page_has_private(page)) |
1684 | try_to_release_page(page, 0); | 1684 | try_to_release_page(page, 0); |
1685 | unlock_page(page); | 1685 | unlock_page(page); |
1686 | } | 1686 | } |
1687 | } | 1687 | } |
1688 | 1688 | ||
1689 | if (page_referenced(page, 0, sc->target_mem_cgroup, | 1689 | if (page_referenced(page, 0, sc->target_mem_cgroup, |
1690 | &vm_flags)) { | 1690 | &vm_flags)) { |
1691 | nr_rotated += hpage_nr_pages(page); | 1691 | nr_rotated += hpage_nr_pages(page); |
1692 | /* | 1692 | /* |
1693 | * Identify referenced, file-backed active pages and | 1693 | * Identify referenced, file-backed active pages and |
1694 | * give them one more trip around the active list. So | 1694 | * give them one more trip around the active list. So |
1695 | * that executable code get better chances to stay in | 1695 | * that executable code get better chances to stay in |
1696 | * memory under moderate memory pressure. Anon pages | 1696 | * memory under moderate memory pressure. Anon pages |
1697 | * are not likely to be evicted by use-once streaming | 1697 | * are not likely to be evicted by use-once streaming |
1698 | * IO, plus JVM can create lots of anon VM_EXEC pages, | 1698 | * IO, plus JVM can create lots of anon VM_EXEC pages, |
1699 | * so we ignore them here. | 1699 | * so we ignore them here. |
1700 | */ | 1700 | */ |
1701 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { | 1701 | if ((vm_flags & VM_EXEC) && page_is_file_cache(page)) { |
1702 | list_add(&page->lru, &l_active); | 1702 | list_add(&page->lru, &l_active); |
1703 | continue; | 1703 | continue; |
1704 | } | 1704 | } |
1705 | } | 1705 | } |
1706 | 1706 | ||
1707 | ClearPageActive(page); /* we are de-activating */ | 1707 | ClearPageActive(page); /* we are de-activating */ |
1708 | list_add(&page->lru, &l_inactive); | 1708 | list_add(&page->lru, &l_inactive); |
1709 | } | 1709 | } |
1710 | 1710 | ||
1711 | /* | 1711 | /* |
1712 | * Move pages back to the lru list. | 1712 | * Move pages back to the lru list. |
1713 | */ | 1713 | */ |
1714 | spin_lock_irq(&zone->lru_lock); | 1714 | spin_lock_irq(&zone->lru_lock); |
1715 | /* | 1715 | /* |
1716 | * Count referenced pages from currently used mappings as rotated, | 1716 | * Count referenced pages from currently used mappings as rotated, |
1717 | * even though only some of them are actually re-activated. This | 1717 | * even though only some of them are actually re-activated. This |
1718 | * helps balance scan pressure between file and anonymous pages in | 1718 | * helps balance scan pressure between file and anonymous pages in |
1719 | * get_scan_ratio. | 1719 | * get_scan_ratio. |
1720 | */ | 1720 | */ |
1721 | reclaim_stat->recent_rotated[file] += nr_rotated; | 1721 | reclaim_stat->recent_rotated[file] += nr_rotated; |
1722 | 1722 | ||
1723 | move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru); | 1723 | move_active_pages_to_lru(lruvec, &l_active, &l_hold, lru); |
1724 | move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE); | 1724 | move_active_pages_to_lru(lruvec, &l_inactive, &l_hold, lru - LRU_ACTIVE); |
1725 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); | 1725 | __mod_zone_page_state(zone, NR_ISOLATED_ANON + file, -nr_taken); |
1726 | spin_unlock_irq(&zone->lru_lock); | 1726 | spin_unlock_irq(&zone->lru_lock); |
1727 | 1727 | ||
1728 | free_hot_cold_page_list(&l_hold, true); | 1728 | free_hot_cold_page_list(&l_hold, true); |
1729 | } | 1729 | } |
1730 | 1730 | ||
1731 | #ifdef CONFIG_SWAP | 1731 | #ifdef CONFIG_SWAP |
1732 | static int inactive_anon_is_low_global(struct zone *zone) | 1732 | static int inactive_anon_is_low_global(struct zone *zone) |
1733 | { | 1733 | { |
1734 | unsigned long active, inactive; | 1734 | unsigned long active, inactive; |
1735 | 1735 | ||
1736 | active = zone_page_state(zone, NR_ACTIVE_ANON); | 1736 | active = zone_page_state(zone, NR_ACTIVE_ANON); |
1737 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); | 1737 | inactive = zone_page_state(zone, NR_INACTIVE_ANON); |
1738 | 1738 | ||
1739 | if (inactive * zone->inactive_ratio < active) | 1739 | if (inactive * zone->inactive_ratio < active) |
1740 | return 1; | 1740 | return 1; |
1741 | 1741 | ||
1742 | return 0; | 1742 | return 0; |
1743 | } | 1743 | } |
1744 | 1744 | ||
1745 | /** | 1745 | /** |
1746 | * inactive_anon_is_low - check if anonymous pages need to be deactivated | 1746 | * inactive_anon_is_low - check if anonymous pages need to be deactivated |
1747 | * @lruvec: LRU vector to check | 1747 | * @lruvec: LRU vector to check |
1748 | * | 1748 | * |
1749 | * Returns true if the zone does not have enough inactive anon pages, | 1749 | * Returns true if the zone does not have enough inactive anon pages, |
1750 | * meaning some active anon pages need to be deactivated. | 1750 | * meaning some active anon pages need to be deactivated. |
1751 | */ | 1751 | */ |
1752 | static int inactive_anon_is_low(struct lruvec *lruvec) | 1752 | static int inactive_anon_is_low(struct lruvec *lruvec) |
1753 | { | 1753 | { |
1754 | /* | 1754 | /* |
1755 | * If we don't have swap space, anonymous page deactivation | 1755 | * If we don't have swap space, anonymous page deactivation |
1756 | * is pointless. | 1756 | * is pointless. |
1757 | */ | 1757 | */ |
1758 | if (!total_swap_pages) | 1758 | if (!total_swap_pages) |
1759 | return 0; | 1759 | return 0; |
1760 | 1760 | ||
1761 | if (!mem_cgroup_disabled()) | 1761 | if (!mem_cgroup_disabled()) |
1762 | return mem_cgroup_inactive_anon_is_low(lruvec); | 1762 | return mem_cgroup_inactive_anon_is_low(lruvec); |
1763 | 1763 | ||
1764 | return inactive_anon_is_low_global(lruvec_zone(lruvec)); | 1764 | return inactive_anon_is_low_global(lruvec_zone(lruvec)); |
1765 | } | 1765 | } |
1766 | #else | 1766 | #else |
1767 | static inline int inactive_anon_is_low(struct lruvec *lruvec) | 1767 | static inline int inactive_anon_is_low(struct lruvec *lruvec) |
1768 | { | 1768 | { |
1769 | return 0; | 1769 | return 0; |
1770 | } | 1770 | } |
1771 | #endif | 1771 | #endif |
1772 | 1772 | ||
1773 | /** | 1773 | /** |
1774 | * inactive_file_is_low - check if file pages need to be deactivated | 1774 | * inactive_file_is_low - check if file pages need to be deactivated |
1775 | * @lruvec: LRU vector to check | 1775 | * @lruvec: LRU vector to check |
1776 | * | 1776 | * |
1777 | * When the system is doing streaming IO, memory pressure here | 1777 | * When the system is doing streaming IO, memory pressure here |
1778 | * ensures that active file pages get deactivated, until more | 1778 | * ensures that active file pages get deactivated, until more |
1779 | * than half of the file pages are on the inactive list. | 1779 | * than half of the file pages are on the inactive list. |
1780 | * | 1780 | * |
1781 | * Once we get to that situation, protect the system's working | 1781 | * Once we get to that situation, protect the system's working |
1782 | * set from being evicted by disabling active file page aging. | 1782 | * set from being evicted by disabling active file page aging. |
1783 | * | 1783 | * |
1784 | * This uses a different ratio than the anonymous pages, because | 1784 | * This uses a different ratio than the anonymous pages, because |
1785 | * the page cache uses a use-once replacement algorithm. | 1785 | * the page cache uses a use-once replacement algorithm. |
1786 | */ | 1786 | */ |
1787 | static int inactive_file_is_low(struct lruvec *lruvec) | 1787 | static int inactive_file_is_low(struct lruvec *lruvec) |
1788 | { | 1788 | { |
1789 | unsigned long inactive; | 1789 | unsigned long inactive; |
1790 | unsigned long active; | 1790 | unsigned long active; |
1791 | 1791 | ||
1792 | inactive = get_lru_size(lruvec, LRU_INACTIVE_FILE); | 1792 | inactive = get_lru_size(lruvec, LRU_INACTIVE_FILE); |
1793 | active = get_lru_size(lruvec, LRU_ACTIVE_FILE); | 1793 | active = get_lru_size(lruvec, LRU_ACTIVE_FILE); |
1794 | 1794 | ||
1795 | return active > inactive; | 1795 | return active > inactive; |
1796 | } | 1796 | } |
1797 | 1797 | ||
1798 | static int inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru) | 1798 | static int inactive_list_is_low(struct lruvec *lruvec, enum lru_list lru) |
1799 | { | 1799 | { |
1800 | if (is_file_lru(lru)) | 1800 | if (is_file_lru(lru)) |
1801 | return inactive_file_is_low(lruvec); | 1801 | return inactive_file_is_low(lruvec); |
1802 | else | 1802 | else |
1803 | return inactive_anon_is_low(lruvec); | 1803 | return inactive_anon_is_low(lruvec); |
1804 | } | 1804 | } |
1805 | 1805 | ||
1806 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, | 1806 | static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan, |
1807 | struct lruvec *lruvec, struct scan_control *sc) | 1807 | struct lruvec *lruvec, struct scan_control *sc) |
1808 | { | 1808 | { |
1809 | if (is_active_lru(lru)) { | 1809 | if (is_active_lru(lru)) { |
1810 | if (inactive_list_is_low(lruvec, lru)) | 1810 | if (inactive_list_is_low(lruvec, lru)) |
1811 | shrink_active_list(nr_to_scan, lruvec, sc, lru); | 1811 | shrink_active_list(nr_to_scan, lruvec, sc, lru); |
1812 | return 0; | 1812 | return 0; |
1813 | } | 1813 | } |
1814 | 1814 | ||
1815 | return shrink_inactive_list(nr_to_scan, lruvec, sc, lru); | 1815 | return shrink_inactive_list(nr_to_scan, lruvec, sc, lru); |
1816 | } | 1816 | } |
1817 | 1817 | ||
1818 | static int vmscan_swappiness(struct scan_control *sc) | 1818 | static int vmscan_swappiness(struct scan_control *sc) |
1819 | { | 1819 | { |
1820 | if (global_reclaim(sc)) | 1820 | if (global_reclaim(sc)) |
1821 | return vm_swappiness; | 1821 | return vm_swappiness; |
1822 | return mem_cgroup_swappiness(sc->target_mem_cgroup); | 1822 | return mem_cgroup_swappiness(sc->target_mem_cgroup); |
1823 | } | 1823 | } |
1824 | 1824 | ||
1825 | enum scan_balance { | 1825 | enum scan_balance { |
1826 | SCAN_EQUAL, | 1826 | SCAN_EQUAL, |
1827 | SCAN_FRACT, | 1827 | SCAN_FRACT, |
1828 | SCAN_ANON, | 1828 | SCAN_ANON, |
1829 | SCAN_FILE, | 1829 | SCAN_FILE, |
1830 | }; | 1830 | }; |
1831 | 1831 | ||
1832 | /* | 1832 | /* |
1833 | * Determine how aggressively the anon and file LRU lists should be | 1833 | * Determine how aggressively the anon and file LRU lists should be |
1834 | * scanned. The relative value of each set of LRU lists is determined | 1834 | * scanned. The relative value of each set of LRU lists is determined |
1835 | * by looking at the fraction of the pages scanned we did rotate back | 1835 | * by looking at the fraction of the pages scanned we did rotate back |
1836 | * onto the active list instead of evict. | 1836 | * onto the active list instead of evict. |
1837 | * | 1837 | * |
1838 | * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan | 1838 | * nr[0] = anon inactive pages to scan; nr[1] = anon active pages to scan |
1839 | * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan | 1839 | * nr[2] = file inactive pages to scan; nr[3] = file active pages to scan |
1840 | */ | 1840 | */ |
1841 | static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc, | 1841 | static void get_scan_count(struct lruvec *lruvec, struct scan_control *sc, |
1842 | unsigned long *nr) | 1842 | unsigned long *nr) |
1843 | { | 1843 | { |
1844 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; | 1844 | struct zone_reclaim_stat *reclaim_stat = &lruvec->reclaim_stat; |
1845 | u64 fraction[2]; | 1845 | u64 fraction[2]; |
1846 | u64 denominator = 0; /* gcc */ | 1846 | u64 denominator = 0; /* gcc */ |
1847 | struct zone *zone = lruvec_zone(lruvec); | 1847 | struct zone *zone = lruvec_zone(lruvec); |
1848 | unsigned long anon_prio, file_prio; | 1848 | unsigned long anon_prio, file_prio; |
1849 | enum scan_balance scan_balance; | 1849 | enum scan_balance scan_balance; |
1850 | unsigned long anon, file, free; | 1850 | unsigned long anon, file; |
1851 | bool force_scan = false; | 1851 | bool force_scan = false; |
1852 | unsigned long ap, fp; | 1852 | unsigned long ap, fp; |
1853 | enum lru_list lru; | 1853 | enum lru_list lru; |
1854 | 1854 | ||
1855 | /* | 1855 | /* |
1856 | * If the zone or memcg is small, nr[l] can be 0. This | 1856 | * If the zone or memcg is small, nr[l] can be 0. This |
1857 | * results in no scanning on this priority and a potential | 1857 | * results in no scanning on this priority and a potential |
1858 | * priority drop. Global direct reclaim can go to the next | 1858 | * priority drop. Global direct reclaim can go to the next |
1859 | * zone and tends to have no problems. Global kswapd is for | 1859 | * zone and tends to have no problems. Global kswapd is for |
1860 | * zone balancing and it needs to scan a minimum amount. When | 1860 | * zone balancing and it needs to scan a minimum amount. When |
1861 | * reclaiming for a memcg, a priority drop can cause high | 1861 | * reclaiming for a memcg, a priority drop can cause high |
1862 | * latencies, so it's better to scan a minimum amount there as | 1862 | * latencies, so it's better to scan a minimum amount there as |
1863 | * well. | 1863 | * well. |
1864 | */ | 1864 | */ |
1865 | if (current_is_kswapd() && !zone_reclaimable(zone)) | 1865 | if (current_is_kswapd() && !zone_reclaimable(zone)) |
1866 | force_scan = true; | 1866 | force_scan = true; |
1867 | if (!global_reclaim(sc)) | 1867 | if (!global_reclaim(sc)) |
1868 | force_scan = true; | 1868 | force_scan = true; |
1869 | 1869 | ||
1870 | /* If we have no swap space, do not bother scanning anon pages. */ | 1870 | /* If we have no swap space, do not bother scanning anon pages. */ |
1871 | if (!sc->may_swap || (get_nr_swap_pages() <= 0)) { | 1871 | if (!sc->may_swap || (get_nr_swap_pages() <= 0)) { |
1872 | scan_balance = SCAN_FILE; | 1872 | scan_balance = SCAN_FILE; |
1873 | goto out; | 1873 | goto out; |
1874 | } | 1874 | } |
1875 | 1875 | ||
1876 | /* | 1876 | /* |
1877 | * Global reclaim will swap to prevent OOM even with no | 1877 | * Global reclaim will swap to prevent OOM even with no |
1878 | * swappiness, but memcg users want to use this knob to | 1878 | * swappiness, but memcg users want to use this knob to |
1879 | * disable swapping for individual groups completely when | 1879 | * disable swapping for individual groups completely when |
1880 | * using the memory controller's swap limit feature would be | 1880 | * using the memory controller's swap limit feature would be |
1881 | * too expensive. | 1881 | * too expensive. |
1882 | */ | 1882 | */ |
1883 | if (!global_reclaim(sc) && !vmscan_swappiness(sc)) { | 1883 | if (!global_reclaim(sc) && !vmscan_swappiness(sc)) { |
1884 | scan_balance = SCAN_FILE; | 1884 | scan_balance = SCAN_FILE; |
1885 | goto out; | 1885 | goto out; |
1886 | } | 1886 | } |
1887 | 1887 | ||
1888 | /* | 1888 | /* |
1889 | * Do not apply any pressure balancing cleverness when the | 1889 | * Do not apply any pressure balancing cleverness when the |
1890 | * system is close to OOM, scan both anon and file equally | 1890 | * system is close to OOM, scan both anon and file equally |
1891 | * (unless the swappiness setting disagrees with swapping). | 1891 | * (unless the swappiness setting disagrees with swapping). |
1892 | */ | 1892 | */ |
1893 | if (!sc->priority && vmscan_swappiness(sc)) { | 1893 | if (!sc->priority && vmscan_swappiness(sc)) { |
1894 | scan_balance = SCAN_EQUAL; | 1894 | scan_balance = SCAN_EQUAL; |
1895 | goto out; | 1895 | goto out; |
1896 | } | 1896 | } |
1897 | 1897 | ||
1898 | anon = get_lru_size(lruvec, LRU_ACTIVE_ANON) + | ||
1899 | get_lru_size(lruvec, LRU_INACTIVE_ANON); | ||
1900 | file = get_lru_size(lruvec, LRU_ACTIVE_FILE) + | ||
1901 | get_lru_size(lruvec, LRU_INACTIVE_FILE); | ||
1902 | |||
1903 | /* | 1898 | /* |
1904 | * If it's foreseeable that reclaiming the file cache won't be | 1899 | * If it's foreseeable that reclaiming the file cache won't be |
1905 | * enough to get the zone back into a desirable shape, we have | 1900 | * enough to get the zone back into a desirable shape, we have |
1906 | * to swap. Better start now and leave the - probably heavily | 1901 | * to swap. Better start now and leave the - probably heavily |
1907 | * thrashing - remaining file pages alone. | 1902 | * thrashing - remaining file pages alone. |
1908 | */ | 1903 | */ |
1909 | if (global_reclaim(sc)) { | 1904 | if (global_reclaim(sc)) { |
1910 | free = zone_page_state(zone, NR_FREE_PAGES); | 1905 | unsigned long zonefile; |
1911 | if (unlikely(file + free <= high_wmark_pages(zone))) { | 1906 | unsigned long zonefree; |
1907 | |||
1908 | zonefree = zone_page_state(zone, NR_FREE_PAGES); | ||
1909 | zonefile = zone_page_state(zone, NR_ACTIVE_FILE) + | ||
1910 | zone_page_state(zone, NR_INACTIVE_FILE); | ||
1911 | |||
1912 | if (unlikely(zonefile + zonefree <= high_wmark_pages(zone))) { | ||
1912 | scan_balance = SCAN_ANON; | 1913 | scan_balance = SCAN_ANON; |
1913 | goto out; | 1914 | goto out; |
1914 | } | 1915 | } |
1915 | } | 1916 | } |
1916 | 1917 | ||
1917 | /* | 1918 | /* |
1918 | * There is enough inactive page cache, do not reclaim | 1919 | * There is enough inactive page cache, do not reclaim |
1919 | * anything from the anonymous working set right now. | 1920 | * anything from the anonymous working set right now. |
1920 | */ | 1921 | */ |
1921 | if (!inactive_file_is_low(lruvec)) { | 1922 | if (!inactive_file_is_low(lruvec)) { |
1922 | scan_balance = SCAN_FILE; | 1923 | scan_balance = SCAN_FILE; |
1923 | goto out; | 1924 | goto out; |
1924 | } | 1925 | } |
1925 | 1926 | ||
1926 | scan_balance = SCAN_FRACT; | 1927 | scan_balance = SCAN_FRACT; |
1927 | 1928 | ||
1928 | /* | 1929 | /* |
1929 | * With swappiness at 100, anonymous and file have the same priority. | 1930 | * With swappiness at 100, anonymous and file have the same priority. |
1930 | * This scanning priority is essentially the inverse of IO cost. | 1931 | * This scanning priority is essentially the inverse of IO cost. |
1931 | */ | 1932 | */ |
1932 | anon_prio = vmscan_swappiness(sc); | 1933 | anon_prio = vmscan_swappiness(sc); |
1933 | file_prio = 200 - anon_prio; | 1934 | file_prio = 200 - anon_prio; |
1934 | 1935 | ||
1935 | /* | 1936 | /* |
1936 | * OK, so we have swap space and a fair amount of page cache | 1937 | * OK, so we have swap space and a fair amount of page cache |
1937 | * pages. We use the recently rotated / recently scanned | 1938 | * pages. We use the recently rotated / recently scanned |
1938 | * ratios to determine how valuable each cache is. | 1939 | * ratios to determine how valuable each cache is. |
1939 | * | 1940 | * |
1940 | * Because workloads change over time (and to avoid overflow) | 1941 | * Because workloads change over time (and to avoid overflow) |
1941 | * we keep these statistics as a floating average, which ends | 1942 | * we keep these statistics as a floating average, which ends |
1942 | * up weighing recent references more than old ones. | 1943 | * up weighing recent references more than old ones. |
1943 | * | 1944 | * |
1944 | * anon in [0], file in [1] | 1945 | * anon in [0], file in [1] |
1945 | */ | 1946 | */ |
1947 | |||
1948 | anon = get_lru_size(lruvec, LRU_ACTIVE_ANON) + | ||
1949 | get_lru_size(lruvec, LRU_INACTIVE_ANON); | ||
1950 | file = get_lru_size(lruvec, LRU_ACTIVE_FILE) + | ||
1951 | get_lru_size(lruvec, LRU_INACTIVE_FILE); | ||
1952 | |||
1946 | spin_lock_irq(&zone->lru_lock); | 1953 | spin_lock_irq(&zone->lru_lock); |
1947 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { | 1954 | if (unlikely(reclaim_stat->recent_scanned[0] > anon / 4)) { |
1948 | reclaim_stat->recent_scanned[0] /= 2; | 1955 | reclaim_stat->recent_scanned[0] /= 2; |
1949 | reclaim_stat->recent_rotated[0] /= 2; | 1956 | reclaim_stat->recent_rotated[0] /= 2; |
1950 | } | 1957 | } |
1951 | 1958 | ||
1952 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { | 1959 | if (unlikely(reclaim_stat->recent_scanned[1] > file / 4)) { |
1953 | reclaim_stat->recent_scanned[1] /= 2; | 1960 | reclaim_stat->recent_scanned[1] /= 2; |
1954 | reclaim_stat->recent_rotated[1] /= 2; | 1961 | reclaim_stat->recent_rotated[1] /= 2; |
1955 | } | 1962 | } |
1956 | 1963 | ||
1957 | /* | 1964 | /* |
1958 | * The amount of pressure on anon vs file pages is inversely | 1965 | * The amount of pressure on anon vs file pages is inversely |
1959 | * proportional to the fraction of recently scanned pages on | 1966 | * proportional to the fraction of recently scanned pages on |
1960 | * each list that were recently referenced and in active use. | 1967 | * each list that were recently referenced and in active use. |
1961 | */ | 1968 | */ |
1962 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); | 1969 | ap = anon_prio * (reclaim_stat->recent_scanned[0] + 1); |
1963 | ap /= reclaim_stat->recent_rotated[0] + 1; | 1970 | ap /= reclaim_stat->recent_rotated[0] + 1; |
1964 | 1971 | ||
1965 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); | 1972 | fp = file_prio * (reclaim_stat->recent_scanned[1] + 1); |
1966 | fp /= reclaim_stat->recent_rotated[1] + 1; | 1973 | fp /= reclaim_stat->recent_rotated[1] + 1; |
1967 | spin_unlock_irq(&zone->lru_lock); | 1974 | spin_unlock_irq(&zone->lru_lock); |
1968 | 1975 | ||
1969 | fraction[0] = ap; | 1976 | fraction[0] = ap; |
1970 | fraction[1] = fp; | 1977 | fraction[1] = fp; |
1971 | denominator = ap + fp + 1; | 1978 | denominator = ap + fp + 1; |
1972 | out: | 1979 | out: |
1973 | for_each_evictable_lru(lru) { | 1980 | for_each_evictable_lru(lru) { |
1974 | int file = is_file_lru(lru); | 1981 | int file = is_file_lru(lru); |
1975 | unsigned long size; | 1982 | unsigned long size; |
1976 | unsigned long scan; | 1983 | unsigned long scan; |
1977 | 1984 | ||
1978 | size = get_lru_size(lruvec, lru); | 1985 | size = get_lru_size(lruvec, lru); |
1979 | scan = size >> sc->priority; | 1986 | scan = size >> sc->priority; |
1980 | 1987 | ||
1981 | if (!scan && force_scan) | 1988 | if (!scan && force_scan) |
1982 | scan = min(size, SWAP_CLUSTER_MAX); | 1989 | scan = min(size, SWAP_CLUSTER_MAX); |
1983 | 1990 | ||
1984 | switch (scan_balance) { | 1991 | switch (scan_balance) { |
1985 | case SCAN_EQUAL: | 1992 | case SCAN_EQUAL: |
1986 | /* Scan lists relative to size */ | 1993 | /* Scan lists relative to size */ |
1987 | break; | 1994 | break; |
1988 | case SCAN_FRACT: | 1995 | case SCAN_FRACT: |
1989 | /* | 1996 | /* |
1990 | * Scan types proportional to swappiness and | 1997 | * Scan types proportional to swappiness and |
1991 | * their relative recent reclaim efficiency. | 1998 | * their relative recent reclaim efficiency. |
1992 | */ | 1999 | */ |
1993 | scan = div64_u64(scan * fraction[file], denominator); | 2000 | scan = div64_u64(scan * fraction[file], denominator); |
1994 | break; | 2001 | break; |
1995 | case SCAN_FILE: | 2002 | case SCAN_FILE: |
1996 | case SCAN_ANON: | 2003 | case SCAN_ANON: |
1997 | /* Scan one type exclusively */ | 2004 | /* Scan one type exclusively */ |
1998 | if ((scan_balance == SCAN_FILE) != file) | 2005 | if ((scan_balance == SCAN_FILE) != file) |
1999 | scan = 0; | 2006 | scan = 0; |
2000 | break; | 2007 | break; |
2001 | default: | 2008 | default: |
2002 | /* Look ma, no brain */ | 2009 | /* Look ma, no brain */ |
2003 | BUG(); | 2010 | BUG(); |
2004 | } | 2011 | } |
2005 | nr[lru] = scan; | 2012 | nr[lru] = scan; |
2006 | } | 2013 | } |
2007 | } | 2014 | } |
2008 | 2015 | ||
2009 | /* | 2016 | /* |
2010 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. | 2017 | * This is a basic per-zone page freer. Used by both kswapd and direct reclaim. |
2011 | */ | 2018 | */ |
2012 | static void shrink_lruvec(struct lruvec *lruvec, struct scan_control *sc) | 2019 | static void shrink_lruvec(struct lruvec *lruvec, struct scan_control *sc) |
2013 | { | 2020 | { |
2014 | unsigned long nr[NR_LRU_LISTS]; | 2021 | unsigned long nr[NR_LRU_LISTS]; |
2015 | unsigned long targets[NR_LRU_LISTS]; | 2022 | unsigned long targets[NR_LRU_LISTS]; |
2016 | unsigned long nr_to_scan; | 2023 | unsigned long nr_to_scan; |
2017 | enum lru_list lru; | 2024 | enum lru_list lru; |
2018 | unsigned long nr_reclaimed = 0; | 2025 | unsigned long nr_reclaimed = 0; |
2019 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; | 2026 | unsigned long nr_to_reclaim = sc->nr_to_reclaim; |
2020 | struct blk_plug plug; | 2027 | struct blk_plug plug; |
2021 | bool scan_adjusted; | 2028 | bool scan_adjusted; |
2022 | 2029 | ||
2023 | get_scan_count(lruvec, sc, nr); | 2030 | get_scan_count(lruvec, sc, nr); |
2024 | 2031 | ||
2025 | /* Record the original scan target for proportional adjustments later */ | 2032 | /* Record the original scan target for proportional adjustments later */ |
2026 | memcpy(targets, nr, sizeof(nr)); | 2033 | memcpy(targets, nr, sizeof(nr)); |
2027 | 2034 | ||
2028 | /* | 2035 | /* |
2029 | * Global reclaiming within direct reclaim at DEF_PRIORITY is a normal | 2036 | * Global reclaiming within direct reclaim at DEF_PRIORITY is a normal |
2030 | * event that can occur when there is little memory pressure e.g. | 2037 | * event that can occur when there is little memory pressure e.g. |
2031 | * multiple streaming readers/writers. Hence, we do not abort scanning | 2038 | * multiple streaming readers/writers. Hence, we do not abort scanning |
2032 | * when the requested number of pages are reclaimed when scanning at | 2039 | * when the requested number of pages are reclaimed when scanning at |
2033 | * DEF_PRIORITY on the assumption that the fact we are direct | 2040 | * DEF_PRIORITY on the assumption that the fact we are direct |
2034 | * reclaiming implies that kswapd is not keeping up and it is best to | 2041 | * reclaiming implies that kswapd is not keeping up and it is best to |
2035 | * do a batch of work at once. For memcg reclaim one check is made to | 2042 | * do a batch of work at once. For memcg reclaim one check is made to |
2036 | * abort proportional reclaim if either the file or anon lru has already | 2043 | * abort proportional reclaim if either the file or anon lru has already |
2037 | * dropped to zero at the first pass. | 2044 | * dropped to zero at the first pass. |
2038 | */ | 2045 | */ |
2039 | scan_adjusted = (global_reclaim(sc) && !current_is_kswapd() && | 2046 | scan_adjusted = (global_reclaim(sc) && !current_is_kswapd() && |
2040 | sc->priority == DEF_PRIORITY); | 2047 | sc->priority == DEF_PRIORITY); |
2041 | 2048 | ||
2042 | blk_start_plug(&plug); | 2049 | blk_start_plug(&plug); |
2043 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || | 2050 | while (nr[LRU_INACTIVE_ANON] || nr[LRU_ACTIVE_FILE] || |
2044 | nr[LRU_INACTIVE_FILE]) { | 2051 | nr[LRU_INACTIVE_FILE]) { |
2045 | unsigned long nr_anon, nr_file, percentage; | 2052 | unsigned long nr_anon, nr_file, percentage; |
2046 | unsigned long nr_scanned; | 2053 | unsigned long nr_scanned; |
2047 | 2054 | ||
2048 | for_each_evictable_lru(lru) { | 2055 | for_each_evictable_lru(lru) { |
2049 | if (nr[lru]) { | 2056 | if (nr[lru]) { |
2050 | nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX); | 2057 | nr_to_scan = min(nr[lru], SWAP_CLUSTER_MAX); |
2051 | nr[lru] -= nr_to_scan; | 2058 | nr[lru] -= nr_to_scan; |
2052 | 2059 | ||
2053 | nr_reclaimed += shrink_list(lru, nr_to_scan, | 2060 | nr_reclaimed += shrink_list(lru, nr_to_scan, |
2054 | lruvec, sc); | 2061 | lruvec, sc); |
2055 | } | 2062 | } |
2056 | } | 2063 | } |
2057 | 2064 | ||
2058 | if (nr_reclaimed < nr_to_reclaim || scan_adjusted) | 2065 | if (nr_reclaimed < nr_to_reclaim || scan_adjusted) |
2059 | continue; | 2066 | continue; |
2060 | 2067 | ||
2061 | /* | 2068 | /* |
2062 | * For kswapd and memcg, reclaim at least the number of pages | 2069 | * For kswapd and memcg, reclaim at least the number of pages |
2063 | * requested. Ensure that the anon and file LRUs are scanned | 2070 | * requested. Ensure that the anon and file LRUs are scanned |
2064 | * proportionally what was requested by get_scan_count(). We | 2071 | * proportionally what was requested by get_scan_count(). We |
2065 | * stop reclaiming one LRU and reduce the amount scanning | 2072 | * stop reclaiming one LRU and reduce the amount scanning |
2066 | * proportional to the original scan target. | 2073 | * proportional to the original scan target. |
2067 | */ | 2074 | */ |
2068 | nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE]; | 2075 | nr_file = nr[LRU_INACTIVE_FILE] + nr[LRU_ACTIVE_FILE]; |
2069 | nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON]; | 2076 | nr_anon = nr[LRU_INACTIVE_ANON] + nr[LRU_ACTIVE_ANON]; |
2070 | 2077 | ||
2071 | /* | 2078 | /* |
2072 | * It's just vindictive to attack the larger once the smaller | 2079 | * It's just vindictive to attack the larger once the smaller |
2073 | * has gone to zero. And given the way we stop scanning the | 2080 | * has gone to zero. And given the way we stop scanning the |
2074 | * smaller below, this makes sure that we only make one nudge | 2081 | * smaller below, this makes sure that we only make one nudge |
2075 | * towards proportionality once we've got nr_to_reclaim. | 2082 | * towards proportionality once we've got nr_to_reclaim. |
2076 | */ | 2083 | */ |
2077 | if (!nr_file || !nr_anon) | 2084 | if (!nr_file || !nr_anon) |
2078 | break; | 2085 | break; |
2079 | 2086 | ||
2080 | if (nr_file > nr_anon) { | 2087 | if (nr_file > nr_anon) { |
2081 | unsigned long scan_target = targets[LRU_INACTIVE_ANON] + | 2088 | unsigned long scan_target = targets[LRU_INACTIVE_ANON] + |
2082 | targets[LRU_ACTIVE_ANON] + 1; | 2089 | targets[LRU_ACTIVE_ANON] + 1; |
2083 | lru = LRU_BASE; | 2090 | lru = LRU_BASE; |
2084 | percentage = nr_anon * 100 / scan_target; | 2091 | percentage = nr_anon * 100 / scan_target; |
2085 | } else { | 2092 | } else { |
2086 | unsigned long scan_target = targets[LRU_INACTIVE_FILE] + | 2093 | unsigned long scan_target = targets[LRU_INACTIVE_FILE] + |
2087 | targets[LRU_ACTIVE_FILE] + 1; | 2094 | targets[LRU_ACTIVE_FILE] + 1; |
2088 | lru = LRU_FILE; | 2095 | lru = LRU_FILE; |
2089 | percentage = nr_file * 100 / scan_target; | 2096 | percentage = nr_file * 100 / scan_target; |
2090 | } | 2097 | } |
2091 | 2098 | ||
2092 | /* Stop scanning the smaller of the LRU */ | 2099 | /* Stop scanning the smaller of the LRU */ |
2093 | nr[lru] = 0; | 2100 | nr[lru] = 0; |
2094 | nr[lru + LRU_ACTIVE] = 0; | 2101 | nr[lru + LRU_ACTIVE] = 0; |
2095 | 2102 | ||
2096 | /* | 2103 | /* |
2097 | * Recalculate the other LRU scan count based on its original | 2104 | * Recalculate the other LRU scan count based on its original |
2098 | * scan target and the percentage scanning already complete | 2105 | * scan target and the percentage scanning already complete |
2099 | */ | 2106 | */ |
2100 | lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE; | 2107 | lru = (lru == LRU_FILE) ? LRU_BASE : LRU_FILE; |
2101 | nr_scanned = targets[lru] - nr[lru]; | 2108 | nr_scanned = targets[lru] - nr[lru]; |
2102 | nr[lru] = targets[lru] * (100 - percentage) / 100; | 2109 | nr[lru] = targets[lru] * (100 - percentage) / 100; |
2103 | nr[lru] -= min(nr[lru], nr_scanned); | 2110 | nr[lru] -= min(nr[lru], nr_scanned); |
2104 | 2111 | ||
2105 | lru += LRU_ACTIVE; | 2112 | lru += LRU_ACTIVE; |
2106 | nr_scanned = targets[lru] - nr[lru]; | 2113 | nr_scanned = targets[lru] - nr[lru]; |
2107 | nr[lru] = targets[lru] * (100 - percentage) / 100; | 2114 | nr[lru] = targets[lru] * (100 - percentage) / 100; |
2108 | nr[lru] -= min(nr[lru], nr_scanned); | 2115 | nr[lru] -= min(nr[lru], nr_scanned); |
2109 | 2116 | ||
2110 | scan_adjusted = true; | 2117 | scan_adjusted = true; |
2111 | } | 2118 | } |
2112 | blk_finish_plug(&plug); | 2119 | blk_finish_plug(&plug); |
2113 | sc->nr_reclaimed += nr_reclaimed; | 2120 | sc->nr_reclaimed += nr_reclaimed; |
2114 | 2121 | ||
2115 | /* | 2122 | /* |
2116 | * Even if we did not try to evict anon pages at all, we want to | 2123 | * Even if we did not try to evict anon pages at all, we want to |
2117 | * rebalance the anon lru active/inactive ratio. | 2124 | * rebalance the anon lru active/inactive ratio. |
2118 | */ | 2125 | */ |
2119 | if (inactive_anon_is_low(lruvec)) | 2126 | if (inactive_anon_is_low(lruvec)) |
2120 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, | 2127 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
2121 | sc, LRU_ACTIVE_ANON); | 2128 | sc, LRU_ACTIVE_ANON); |
2122 | 2129 | ||
2123 | throttle_vm_writeout(sc->gfp_mask); | 2130 | throttle_vm_writeout(sc->gfp_mask); |
2124 | } | 2131 | } |
2125 | 2132 | ||
2126 | /* Use reclaim/compaction for costly allocs or under memory pressure */ | 2133 | /* Use reclaim/compaction for costly allocs or under memory pressure */ |
2127 | static bool in_reclaim_compaction(struct scan_control *sc) | 2134 | static bool in_reclaim_compaction(struct scan_control *sc) |
2128 | { | 2135 | { |
2129 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && | 2136 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && |
2130 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || | 2137 | (sc->order > PAGE_ALLOC_COSTLY_ORDER || |
2131 | sc->priority < DEF_PRIORITY - 2)) | 2138 | sc->priority < DEF_PRIORITY - 2)) |
2132 | return true; | 2139 | return true; |
2133 | 2140 | ||
2134 | return false; | 2141 | return false; |
2135 | } | 2142 | } |
2136 | 2143 | ||
2137 | /* | 2144 | /* |
2138 | * Reclaim/compaction is used for high-order allocation requests. It reclaims | 2145 | * Reclaim/compaction is used for high-order allocation requests. It reclaims |
2139 | * order-0 pages before compacting the zone. should_continue_reclaim() returns | 2146 | * order-0 pages before compacting the zone. should_continue_reclaim() returns |
2140 | * true if more pages should be reclaimed such that when the page allocator | 2147 | * true if more pages should be reclaimed such that when the page allocator |
2141 | * calls try_to_compact_zone() that it will have enough free pages to succeed. | 2148 | * calls try_to_compact_zone() that it will have enough free pages to succeed. |
2142 | * It will give up earlier than that if there is difficulty reclaiming pages. | 2149 | * It will give up earlier than that if there is difficulty reclaiming pages. |
2143 | */ | 2150 | */ |
2144 | static inline bool should_continue_reclaim(struct zone *zone, | 2151 | static inline bool should_continue_reclaim(struct zone *zone, |
2145 | unsigned long nr_reclaimed, | 2152 | unsigned long nr_reclaimed, |
2146 | unsigned long nr_scanned, | 2153 | unsigned long nr_scanned, |
2147 | struct scan_control *sc) | 2154 | struct scan_control *sc) |
2148 | { | 2155 | { |
2149 | unsigned long pages_for_compaction; | 2156 | unsigned long pages_for_compaction; |
2150 | unsigned long inactive_lru_pages; | 2157 | unsigned long inactive_lru_pages; |
2151 | 2158 | ||
2152 | /* If not in reclaim/compaction mode, stop */ | 2159 | /* If not in reclaim/compaction mode, stop */ |
2153 | if (!in_reclaim_compaction(sc)) | 2160 | if (!in_reclaim_compaction(sc)) |
2154 | return false; | 2161 | return false; |
2155 | 2162 | ||
2156 | /* Consider stopping depending on scan and reclaim activity */ | 2163 | /* Consider stopping depending on scan and reclaim activity */ |
2157 | if (sc->gfp_mask & __GFP_REPEAT) { | 2164 | if (sc->gfp_mask & __GFP_REPEAT) { |
2158 | /* | 2165 | /* |
2159 | * For __GFP_REPEAT allocations, stop reclaiming if the | 2166 | * For __GFP_REPEAT allocations, stop reclaiming if the |
2160 | * full LRU list has been scanned and we are still failing | 2167 | * full LRU list has been scanned and we are still failing |
2161 | * to reclaim pages. This full LRU scan is potentially | 2168 | * to reclaim pages. This full LRU scan is potentially |
2162 | * expensive but a __GFP_REPEAT caller really wants to succeed | 2169 | * expensive but a __GFP_REPEAT caller really wants to succeed |
2163 | */ | 2170 | */ |
2164 | if (!nr_reclaimed && !nr_scanned) | 2171 | if (!nr_reclaimed && !nr_scanned) |
2165 | return false; | 2172 | return false; |
2166 | } else { | 2173 | } else { |
2167 | /* | 2174 | /* |
2168 | * For non-__GFP_REPEAT allocations which can presumably | 2175 | * For non-__GFP_REPEAT allocations which can presumably |
2169 | * fail without consequence, stop if we failed to reclaim | 2176 | * fail without consequence, stop if we failed to reclaim |
2170 | * any pages from the last SWAP_CLUSTER_MAX number of | 2177 | * any pages from the last SWAP_CLUSTER_MAX number of |
2171 | * pages that were scanned. This will return to the | 2178 | * pages that were scanned. This will return to the |
2172 | * caller faster at the risk reclaim/compaction and | 2179 | * caller faster at the risk reclaim/compaction and |
2173 | * the resulting allocation attempt fails | 2180 | * the resulting allocation attempt fails |
2174 | */ | 2181 | */ |
2175 | if (!nr_reclaimed) | 2182 | if (!nr_reclaimed) |
2176 | return false; | 2183 | return false; |
2177 | } | 2184 | } |
2178 | 2185 | ||
2179 | /* | 2186 | /* |
2180 | * If we have not reclaimed enough pages for compaction and the | 2187 | * If we have not reclaimed enough pages for compaction and the |
2181 | * inactive lists are large enough, continue reclaiming | 2188 | * inactive lists are large enough, continue reclaiming |
2182 | */ | 2189 | */ |
2183 | pages_for_compaction = (2UL << sc->order); | 2190 | pages_for_compaction = (2UL << sc->order); |
2184 | inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE); | 2191 | inactive_lru_pages = zone_page_state(zone, NR_INACTIVE_FILE); |
2185 | if (get_nr_swap_pages() > 0) | 2192 | if (get_nr_swap_pages() > 0) |
2186 | inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON); | 2193 | inactive_lru_pages += zone_page_state(zone, NR_INACTIVE_ANON); |
2187 | if (sc->nr_reclaimed < pages_for_compaction && | 2194 | if (sc->nr_reclaimed < pages_for_compaction && |
2188 | inactive_lru_pages > pages_for_compaction) | 2195 | inactive_lru_pages > pages_for_compaction) |
2189 | return true; | 2196 | return true; |
2190 | 2197 | ||
2191 | /* If compaction would go ahead or the allocation would succeed, stop */ | 2198 | /* If compaction would go ahead or the allocation would succeed, stop */ |
2192 | switch (compaction_suitable(zone, sc->order)) { | 2199 | switch (compaction_suitable(zone, sc->order)) { |
2193 | case COMPACT_PARTIAL: | 2200 | case COMPACT_PARTIAL: |
2194 | case COMPACT_CONTINUE: | 2201 | case COMPACT_CONTINUE: |
2195 | return false; | 2202 | return false; |
2196 | default: | 2203 | default: |
2197 | return true; | 2204 | return true; |
2198 | } | 2205 | } |
2199 | } | 2206 | } |
2200 | 2207 | ||
2201 | static void shrink_zone(struct zone *zone, struct scan_control *sc) | 2208 | static void shrink_zone(struct zone *zone, struct scan_control *sc) |
2202 | { | 2209 | { |
2203 | unsigned long nr_reclaimed, nr_scanned; | 2210 | unsigned long nr_reclaimed, nr_scanned; |
2204 | 2211 | ||
2205 | do { | 2212 | do { |
2206 | struct mem_cgroup *root = sc->target_mem_cgroup; | 2213 | struct mem_cgroup *root = sc->target_mem_cgroup; |
2207 | struct mem_cgroup_reclaim_cookie reclaim = { | 2214 | struct mem_cgroup_reclaim_cookie reclaim = { |
2208 | .zone = zone, | 2215 | .zone = zone, |
2209 | .priority = sc->priority, | 2216 | .priority = sc->priority, |
2210 | }; | 2217 | }; |
2211 | struct mem_cgroup *memcg; | 2218 | struct mem_cgroup *memcg; |
2212 | 2219 | ||
2213 | nr_reclaimed = sc->nr_reclaimed; | 2220 | nr_reclaimed = sc->nr_reclaimed; |
2214 | nr_scanned = sc->nr_scanned; | 2221 | nr_scanned = sc->nr_scanned; |
2215 | 2222 | ||
2216 | memcg = mem_cgroup_iter(root, NULL, &reclaim); | 2223 | memcg = mem_cgroup_iter(root, NULL, &reclaim); |
2217 | do { | 2224 | do { |
2218 | struct lruvec *lruvec; | 2225 | struct lruvec *lruvec; |
2219 | 2226 | ||
2220 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); | 2227 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
2221 | 2228 | ||
2222 | shrink_lruvec(lruvec, sc); | 2229 | shrink_lruvec(lruvec, sc); |
2223 | 2230 | ||
2224 | /* | 2231 | /* |
2225 | * Direct reclaim and kswapd have to scan all memory | 2232 | * Direct reclaim and kswapd have to scan all memory |
2226 | * cgroups to fulfill the overall scan target for the | 2233 | * cgroups to fulfill the overall scan target for the |
2227 | * zone. | 2234 | * zone. |
2228 | * | 2235 | * |
2229 | * Limit reclaim, on the other hand, only cares about | 2236 | * Limit reclaim, on the other hand, only cares about |
2230 | * nr_to_reclaim pages to be reclaimed and it will | 2237 | * nr_to_reclaim pages to be reclaimed and it will |
2231 | * retry with decreasing priority if one round over the | 2238 | * retry with decreasing priority if one round over the |
2232 | * whole hierarchy is not sufficient. | 2239 | * whole hierarchy is not sufficient. |
2233 | */ | 2240 | */ |
2234 | if (!global_reclaim(sc) && | 2241 | if (!global_reclaim(sc) && |
2235 | sc->nr_reclaimed >= sc->nr_to_reclaim) { | 2242 | sc->nr_reclaimed >= sc->nr_to_reclaim) { |
2236 | mem_cgroup_iter_break(root, memcg); | 2243 | mem_cgroup_iter_break(root, memcg); |
2237 | break; | 2244 | break; |
2238 | } | 2245 | } |
2239 | memcg = mem_cgroup_iter(root, memcg, &reclaim); | 2246 | memcg = mem_cgroup_iter(root, memcg, &reclaim); |
2240 | } while (memcg); | 2247 | } while (memcg); |
2241 | 2248 | ||
2242 | vmpressure(sc->gfp_mask, sc->target_mem_cgroup, | 2249 | vmpressure(sc->gfp_mask, sc->target_mem_cgroup, |
2243 | sc->nr_scanned - nr_scanned, | 2250 | sc->nr_scanned - nr_scanned, |
2244 | sc->nr_reclaimed - nr_reclaimed); | 2251 | sc->nr_reclaimed - nr_reclaimed); |
2245 | 2252 | ||
2246 | } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed, | 2253 | } while (should_continue_reclaim(zone, sc->nr_reclaimed - nr_reclaimed, |
2247 | sc->nr_scanned - nr_scanned, sc)); | 2254 | sc->nr_scanned - nr_scanned, sc)); |
2248 | } | 2255 | } |
2249 | 2256 | ||
2250 | /* Returns true if compaction should go ahead for a high-order request */ | 2257 | /* Returns true if compaction should go ahead for a high-order request */ |
2251 | static inline bool compaction_ready(struct zone *zone, struct scan_control *sc) | 2258 | static inline bool compaction_ready(struct zone *zone, struct scan_control *sc) |
2252 | { | 2259 | { |
2253 | unsigned long balance_gap, watermark; | 2260 | unsigned long balance_gap, watermark; |
2254 | bool watermark_ok; | 2261 | bool watermark_ok; |
2255 | 2262 | ||
2256 | /* Do not consider compaction for orders reclaim is meant to satisfy */ | 2263 | /* Do not consider compaction for orders reclaim is meant to satisfy */ |
2257 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER) | 2264 | if (sc->order <= PAGE_ALLOC_COSTLY_ORDER) |
2258 | return false; | 2265 | return false; |
2259 | 2266 | ||
2260 | /* | 2267 | /* |
2261 | * Compaction takes time to run and there are potentially other | 2268 | * Compaction takes time to run and there are potentially other |
2262 | * callers using the pages just freed. Continue reclaiming until | 2269 | * callers using the pages just freed. Continue reclaiming until |
2263 | * there is a buffer of free pages available to give compaction | 2270 | * there is a buffer of free pages available to give compaction |
2264 | * a reasonable chance of completing and allocating the page | 2271 | * a reasonable chance of completing and allocating the page |
2265 | */ | 2272 | */ |
2266 | balance_gap = min(low_wmark_pages(zone), | 2273 | balance_gap = min(low_wmark_pages(zone), |
2267 | (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / | 2274 | (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / |
2268 | KSWAPD_ZONE_BALANCE_GAP_RATIO); | 2275 | KSWAPD_ZONE_BALANCE_GAP_RATIO); |
2269 | watermark = high_wmark_pages(zone) + balance_gap + (2UL << sc->order); | 2276 | watermark = high_wmark_pages(zone) + balance_gap + (2UL << sc->order); |
2270 | watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0); | 2277 | watermark_ok = zone_watermark_ok_safe(zone, 0, watermark, 0, 0); |
2271 | 2278 | ||
2272 | /* | 2279 | /* |
2273 | * If compaction is deferred, reclaim up to a point where | 2280 | * If compaction is deferred, reclaim up to a point where |
2274 | * compaction will have a chance of success when re-enabled | 2281 | * compaction will have a chance of success when re-enabled |
2275 | */ | 2282 | */ |
2276 | if (compaction_deferred(zone, sc->order)) | 2283 | if (compaction_deferred(zone, sc->order)) |
2277 | return watermark_ok; | 2284 | return watermark_ok; |
2278 | 2285 | ||
2279 | /* If compaction is not ready to start, keep reclaiming */ | 2286 | /* If compaction is not ready to start, keep reclaiming */ |
2280 | if (!compaction_suitable(zone, sc->order)) | 2287 | if (!compaction_suitable(zone, sc->order)) |
2281 | return false; | 2288 | return false; |
2282 | 2289 | ||
2283 | return watermark_ok; | 2290 | return watermark_ok; |
2284 | } | 2291 | } |
2285 | 2292 | ||
2286 | /* | 2293 | /* |
2287 | * This is the direct reclaim path, for page-allocating processes. We only | 2294 | * This is the direct reclaim path, for page-allocating processes. We only |
2288 | * try to reclaim pages from zones which will satisfy the caller's allocation | 2295 | * try to reclaim pages from zones which will satisfy the caller's allocation |
2289 | * request. | 2296 | * request. |
2290 | * | 2297 | * |
2291 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). | 2298 | * We reclaim from a zone even if that zone is over high_wmark_pages(zone). |
2292 | * Because: | 2299 | * Because: |
2293 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order | 2300 | * a) The caller may be trying to free *extra* pages to satisfy a higher-order |
2294 | * allocation or | 2301 | * allocation or |
2295 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones | 2302 | * b) The target zone may be at high_wmark_pages(zone) but the lower zones |
2296 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' | 2303 | * must go *over* high_wmark_pages(zone) to satisfy the `incremental min' |
2297 | * zone defense algorithm. | 2304 | * zone defense algorithm. |
2298 | * | 2305 | * |
2299 | * If a zone is deemed to be full of pinned pages then just give it a light | 2306 | * If a zone is deemed to be full of pinned pages then just give it a light |
2300 | * scan then give up on it. | 2307 | * scan then give up on it. |
2301 | * | 2308 | * |
2302 | * This function returns true if a zone is being reclaimed for a costly | 2309 | * This function returns true if a zone is being reclaimed for a costly |
2303 | * high-order allocation and compaction is ready to begin. This indicates to | 2310 | * high-order allocation and compaction is ready to begin. This indicates to |
2304 | * the caller that it should consider retrying the allocation instead of | 2311 | * the caller that it should consider retrying the allocation instead of |
2305 | * further reclaim. | 2312 | * further reclaim. |
2306 | */ | 2313 | */ |
2307 | static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc) | 2314 | static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc) |
2308 | { | 2315 | { |
2309 | struct zoneref *z; | 2316 | struct zoneref *z; |
2310 | struct zone *zone; | 2317 | struct zone *zone; |
2311 | unsigned long nr_soft_reclaimed; | 2318 | unsigned long nr_soft_reclaimed; |
2312 | unsigned long nr_soft_scanned; | 2319 | unsigned long nr_soft_scanned; |
2313 | bool aborted_reclaim = false; | 2320 | bool aborted_reclaim = false; |
2314 | 2321 | ||
2315 | /* | 2322 | /* |
2316 | * If the number of buffer_heads in the machine exceeds the maximum | 2323 | * If the number of buffer_heads in the machine exceeds the maximum |
2317 | * allowed level, force direct reclaim to scan the highmem zone as | 2324 | * allowed level, force direct reclaim to scan the highmem zone as |
2318 | * highmem pages could be pinning lowmem pages storing buffer_heads | 2325 | * highmem pages could be pinning lowmem pages storing buffer_heads |
2319 | */ | 2326 | */ |
2320 | if (buffer_heads_over_limit) | 2327 | if (buffer_heads_over_limit) |
2321 | sc->gfp_mask |= __GFP_HIGHMEM; | 2328 | sc->gfp_mask |= __GFP_HIGHMEM; |
2322 | 2329 | ||
2323 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | 2330 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
2324 | gfp_zone(sc->gfp_mask), sc->nodemask) { | 2331 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
2325 | if (!populated_zone(zone)) | 2332 | if (!populated_zone(zone)) |
2326 | continue; | 2333 | continue; |
2327 | /* | 2334 | /* |
2328 | * Take care memory controller reclaiming has small influence | 2335 | * Take care memory controller reclaiming has small influence |
2329 | * to global LRU. | 2336 | * to global LRU. |
2330 | */ | 2337 | */ |
2331 | if (global_reclaim(sc)) { | 2338 | if (global_reclaim(sc)) { |
2332 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) | 2339 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
2333 | continue; | 2340 | continue; |
2334 | if (sc->priority != DEF_PRIORITY && | 2341 | if (sc->priority != DEF_PRIORITY && |
2335 | !zone_reclaimable(zone)) | 2342 | !zone_reclaimable(zone)) |
2336 | continue; /* Let kswapd poll it */ | 2343 | continue; /* Let kswapd poll it */ |
2337 | if (IS_ENABLED(CONFIG_COMPACTION)) { | 2344 | if (IS_ENABLED(CONFIG_COMPACTION)) { |
2338 | /* | 2345 | /* |
2339 | * If we already have plenty of memory free for | 2346 | * If we already have plenty of memory free for |
2340 | * compaction in this zone, don't free any more. | 2347 | * compaction in this zone, don't free any more. |
2341 | * Even though compaction is invoked for any | 2348 | * Even though compaction is invoked for any |
2342 | * non-zero order, only frequent costly order | 2349 | * non-zero order, only frequent costly order |
2343 | * reclamation is disruptive enough to become a | 2350 | * reclamation is disruptive enough to become a |
2344 | * noticeable problem, like transparent huge | 2351 | * noticeable problem, like transparent huge |
2345 | * page allocations. | 2352 | * page allocations. |
2346 | */ | 2353 | */ |
2347 | if (compaction_ready(zone, sc)) { | 2354 | if (compaction_ready(zone, sc)) { |
2348 | aborted_reclaim = true; | 2355 | aborted_reclaim = true; |
2349 | continue; | 2356 | continue; |
2350 | } | 2357 | } |
2351 | } | 2358 | } |
2352 | /* | 2359 | /* |
2353 | * This steals pages from memory cgroups over softlimit | 2360 | * This steals pages from memory cgroups over softlimit |
2354 | * and returns the number of reclaimed pages and | 2361 | * and returns the number of reclaimed pages and |
2355 | * scanned pages. This works for global memory pressure | 2362 | * scanned pages. This works for global memory pressure |
2356 | * and balancing, not for a memcg's limit. | 2363 | * and balancing, not for a memcg's limit. |
2357 | */ | 2364 | */ |
2358 | nr_soft_scanned = 0; | 2365 | nr_soft_scanned = 0; |
2359 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | 2366 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
2360 | sc->order, sc->gfp_mask, | 2367 | sc->order, sc->gfp_mask, |
2361 | &nr_soft_scanned); | 2368 | &nr_soft_scanned); |
2362 | sc->nr_reclaimed += nr_soft_reclaimed; | 2369 | sc->nr_reclaimed += nr_soft_reclaimed; |
2363 | sc->nr_scanned += nr_soft_scanned; | 2370 | sc->nr_scanned += nr_soft_scanned; |
2364 | /* need some check for avoid more shrink_zone() */ | 2371 | /* need some check for avoid more shrink_zone() */ |
2365 | } | 2372 | } |
2366 | 2373 | ||
2367 | shrink_zone(zone, sc); | 2374 | shrink_zone(zone, sc); |
2368 | } | 2375 | } |
2369 | 2376 | ||
2370 | return aborted_reclaim; | 2377 | return aborted_reclaim; |
2371 | } | 2378 | } |
2372 | 2379 | ||
2373 | /* All zones in zonelist are unreclaimable? */ | 2380 | /* All zones in zonelist are unreclaimable? */ |
2374 | static bool all_unreclaimable(struct zonelist *zonelist, | 2381 | static bool all_unreclaimable(struct zonelist *zonelist, |
2375 | struct scan_control *sc) | 2382 | struct scan_control *sc) |
2376 | { | 2383 | { |
2377 | struct zoneref *z; | 2384 | struct zoneref *z; |
2378 | struct zone *zone; | 2385 | struct zone *zone; |
2379 | 2386 | ||
2380 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | 2387 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
2381 | gfp_zone(sc->gfp_mask), sc->nodemask) { | 2388 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
2382 | if (!populated_zone(zone)) | 2389 | if (!populated_zone(zone)) |
2383 | continue; | 2390 | continue; |
2384 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) | 2391 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
2385 | continue; | 2392 | continue; |
2386 | if (zone_reclaimable(zone)) | 2393 | if (zone_reclaimable(zone)) |
2387 | return false; | 2394 | return false; |
2388 | } | 2395 | } |
2389 | 2396 | ||
2390 | return true; | 2397 | return true; |
2391 | } | 2398 | } |
2392 | 2399 | ||
2393 | /* | 2400 | /* |
2394 | * This is the main entry point to direct page reclaim. | 2401 | * This is the main entry point to direct page reclaim. |
2395 | * | 2402 | * |
2396 | * If a full scan of the inactive list fails to free enough memory then we | 2403 | * If a full scan of the inactive list fails to free enough memory then we |
2397 | * are "out of memory" and something needs to be killed. | 2404 | * are "out of memory" and something needs to be killed. |
2398 | * | 2405 | * |
2399 | * If the caller is !__GFP_FS then the probability of a failure is reasonably | 2406 | * If the caller is !__GFP_FS then the probability of a failure is reasonably |
2400 | * high - the zone may be full of dirty or under-writeback pages, which this | 2407 | * high - the zone may be full of dirty or under-writeback pages, which this |
2401 | * caller can't do much about. We kick the writeback threads and take explicit | 2408 | * caller can't do much about. We kick the writeback threads and take explicit |
2402 | * naps in the hope that some of these pages can be written. But if the | 2409 | * naps in the hope that some of these pages can be written. But if the |
2403 | * allocating task holds filesystem locks which prevent writeout this might not | 2410 | * allocating task holds filesystem locks which prevent writeout this might not |
2404 | * work, and the allocation attempt will fail. | 2411 | * work, and the allocation attempt will fail. |
2405 | * | 2412 | * |
2406 | * returns: 0, if no pages reclaimed | 2413 | * returns: 0, if no pages reclaimed |
2407 | * else, the number of pages reclaimed | 2414 | * else, the number of pages reclaimed |
2408 | */ | 2415 | */ |
2409 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, | 2416 | static unsigned long do_try_to_free_pages(struct zonelist *zonelist, |
2410 | struct scan_control *sc, | 2417 | struct scan_control *sc, |
2411 | struct shrink_control *shrink) | 2418 | struct shrink_control *shrink) |
2412 | { | 2419 | { |
2413 | unsigned long total_scanned = 0; | 2420 | unsigned long total_scanned = 0; |
2414 | struct reclaim_state *reclaim_state = current->reclaim_state; | 2421 | struct reclaim_state *reclaim_state = current->reclaim_state; |
2415 | struct zoneref *z; | 2422 | struct zoneref *z; |
2416 | struct zone *zone; | 2423 | struct zone *zone; |
2417 | unsigned long writeback_threshold; | 2424 | unsigned long writeback_threshold; |
2418 | bool aborted_reclaim; | 2425 | bool aborted_reclaim; |
2419 | 2426 | ||
2420 | delayacct_freepages_start(); | 2427 | delayacct_freepages_start(); |
2421 | 2428 | ||
2422 | if (global_reclaim(sc)) | 2429 | if (global_reclaim(sc)) |
2423 | count_vm_event(ALLOCSTALL); | 2430 | count_vm_event(ALLOCSTALL); |
2424 | 2431 | ||
2425 | do { | 2432 | do { |
2426 | vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup, | 2433 | vmpressure_prio(sc->gfp_mask, sc->target_mem_cgroup, |
2427 | sc->priority); | 2434 | sc->priority); |
2428 | sc->nr_scanned = 0; | 2435 | sc->nr_scanned = 0; |
2429 | aborted_reclaim = shrink_zones(zonelist, sc); | 2436 | aborted_reclaim = shrink_zones(zonelist, sc); |
2430 | 2437 | ||
2431 | /* | 2438 | /* |
2432 | * Don't shrink slabs when reclaiming memory from over limit | 2439 | * Don't shrink slabs when reclaiming memory from over limit |
2433 | * cgroups but do shrink slab at least once when aborting | 2440 | * cgroups but do shrink slab at least once when aborting |
2434 | * reclaim for compaction to avoid unevenly scanning file/anon | 2441 | * reclaim for compaction to avoid unevenly scanning file/anon |
2435 | * LRU pages over slab pages. | 2442 | * LRU pages over slab pages. |
2436 | */ | 2443 | */ |
2437 | if (global_reclaim(sc)) { | 2444 | if (global_reclaim(sc)) { |
2438 | unsigned long lru_pages = 0; | 2445 | unsigned long lru_pages = 0; |
2439 | 2446 | ||
2440 | nodes_clear(shrink->nodes_to_scan); | 2447 | nodes_clear(shrink->nodes_to_scan); |
2441 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | 2448 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
2442 | gfp_zone(sc->gfp_mask), sc->nodemask) { | 2449 | gfp_zone(sc->gfp_mask), sc->nodemask) { |
2443 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) | 2450 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
2444 | continue; | 2451 | continue; |
2445 | 2452 | ||
2446 | lru_pages += zone_reclaimable_pages(zone); | 2453 | lru_pages += zone_reclaimable_pages(zone); |
2447 | node_set(zone_to_nid(zone), | 2454 | node_set(zone_to_nid(zone), |
2448 | shrink->nodes_to_scan); | 2455 | shrink->nodes_to_scan); |
2449 | } | 2456 | } |
2450 | 2457 | ||
2451 | shrink_slab(shrink, sc->nr_scanned, lru_pages); | 2458 | shrink_slab(shrink, sc->nr_scanned, lru_pages); |
2452 | if (reclaim_state) { | 2459 | if (reclaim_state) { |
2453 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; | 2460 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
2454 | reclaim_state->reclaimed_slab = 0; | 2461 | reclaim_state->reclaimed_slab = 0; |
2455 | } | 2462 | } |
2456 | } | 2463 | } |
2457 | total_scanned += sc->nr_scanned; | 2464 | total_scanned += sc->nr_scanned; |
2458 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) | 2465 | if (sc->nr_reclaimed >= sc->nr_to_reclaim) |
2459 | goto out; | 2466 | goto out; |
2460 | 2467 | ||
2461 | /* | 2468 | /* |
2462 | * If we're getting trouble reclaiming, start doing | 2469 | * If we're getting trouble reclaiming, start doing |
2463 | * writepage even in laptop mode. | 2470 | * writepage even in laptop mode. |
2464 | */ | 2471 | */ |
2465 | if (sc->priority < DEF_PRIORITY - 2) | 2472 | if (sc->priority < DEF_PRIORITY - 2) |
2466 | sc->may_writepage = 1; | 2473 | sc->may_writepage = 1; |
2467 | 2474 | ||
2468 | /* | 2475 | /* |
2469 | * Try to write back as many pages as we just scanned. This | 2476 | * Try to write back as many pages as we just scanned. This |
2470 | * tends to cause slow streaming writers to write data to the | 2477 | * tends to cause slow streaming writers to write data to the |
2471 | * disk smoothly, at the dirtying rate, which is nice. But | 2478 | * disk smoothly, at the dirtying rate, which is nice. But |
2472 | * that's undesirable in laptop mode, where we *want* lumpy | 2479 | * that's undesirable in laptop mode, where we *want* lumpy |
2473 | * writeout. So in laptop mode, write out the whole world. | 2480 | * writeout. So in laptop mode, write out the whole world. |
2474 | */ | 2481 | */ |
2475 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; | 2482 | writeback_threshold = sc->nr_to_reclaim + sc->nr_to_reclaim / 2; |
2476 | if (total_scanned > writeback_threshold) { | 2483 | if (total_scanned > writeback_threshold) { |
2477 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, | 2484 | wakeup_flusher_threads(laptop_mode ? 0 : total_scanned, |
2478 | WB_REASON_TRY_TO_FREE_PAGES); | 2485 | WB_REASON_TRY_TO_FREE_PAGES); |
2479 | sc->may_writepage = 1; | 2486 | sc->may_writepage = 1; |
2480 | } | 2487 | } |
2481 | } while (--sc->priority >= 0 && !aborted_reclaim); | 2488 | } while (--sc->priority >= 0 && !aborted_reclaim); |
2482 | 2489 | ||
2483 | out: | 2490 | out: |
2484 | delayacct_freepages_end(); | 2491 | delayacct_freepages_end(); |
2485 | 2492 | ||
2486 | if (sc->nr_reclaimed) | 2493 | if (sc->nr_reclaimed) |
2487 | return sc->nr_reclaimed; | 2494 | return sc->nr_reclaimed; |
2488 | 2495 | ||
2489 | /* | 2496 | /* |
2490 | * As hibernation is going on, kswapd is freezed so that it can't mark | 2497 | * As hibernation is going on, kswapd is freezed so that it can't mark |
2491 | * the zone into all_unreclaimable. Thus bypassing all_unreclaimable | 2498 | * the zone into all_unreclaimable. Thus bypassing all_unreclaimable |
2492 | * check. | 2499 | * check. |
2493 | */ | 2500 | */ |
2494 | if (oom_killer_disabled) | 2501 | if (oom_killer_disabled) |
2495 | return 0; | 2502 | return 0; |
2496 | 2503 | ||
2497 | /* Aborted reclaim to try compaction? don't OOM, then */ | 2504 | /* Aborted reclaim to try compaction? don't OOM, then */ |
2498 | if (aborted_reclaim) | 2505 | if (aborted_reclaim) |
2499 | return 1; | 2506 | return 1; |
2500 | 2507 | ||
2501 | /* top priority shrink_zones still had more to do? don't OOM, then */ | 2508 | /* top priority shrink_zones still had more to do? don't OOM, then */ |
2502 | if (global_reclaim(sc) && !all_unreclaimable(zonelist, sc)) | 2509 | if (global_reclaim(sc) && !all_unreclaimable(zonelist, sc)) |
2503 | return 1; | 2510 | return 1; |
2504 | 2511 | ||
2505 | return 0; | 2512 | return 0; |
2506 | } | 2513 | } |
2507 | 2514 | ||
2508 | static bool pfmemalloc_watermark_ok(pg_data_t *pgdat) | 2515 | static bool pfmemalloc_watermark_ok(pg_data_t *pgdat) |
2509 | { | 2516 | { |
2510 | struct zone *zone; | 2517 | struct zone *zone; |
2511 | unsigned long pfmemalloc_reserve = 0; | 2518 | unsigned long pfmemalloc_reserve = 0; |
2512 | unsigned long free_pages = 0; | 2519 | unsigned long free_pages = 0; |
2513 | int i; | 2520 | int i; |
2514 | bool wmark_ok; | 2521 | bool wmark_ok; |
2515 | 2522 | ||
2516 | for (i = 0; i <= ZONE_NORMAL; i++) { | 2523 | for (i = 0; i <= ZONE_NORMAL; i++) { |
2517 | zone = &pgdat->node_zones[i]; | 2524 | zone = &pgdat->node_zones[i]; |
2518 | if (!populated_zone(zone)) | 2525 | if (!populated_zone(zone)) |
2519 | continue; | 2526 | continue; |
2520 | 2527 | ||
2521 | pfmemalloc_reserve += min_wmark_pages(zone); | 2528 | pfmemalloc_reserve += min_wmark_pages(zone); |
2522 | free_pages += zone_page_state(zone, NR_FREE_PAGES); | 2529 | free_pages += zone_page_state(zone, NR_FREE_PAGES); |
2523 | } | 2530 | } |
2524 | 2531 | ||
2525 | /* If there are no reserves (unexpected config) then do not throttle */ | 2532 | /* If there are no reserves (unexpected config) then do not throttle */ |
2526 | if (!pfmemalloc_reserve) | 2533 | if (!pfmemalloc_reserve) |
2527 | return true; | 2534 | return true; |
2528 | 2535 | ||
2529 | wmark_ok = free_pages > pfmemalloc_reserve / 2; | 2536 | wmark_ok = free_pages > pfmemalloc_reserve / 2; |
2530 | 2537 | ||
2531 | /* kswapd must be awake if processes are being throttled */ | 2538 | /* kswapd must be awake if processes are being throttled */ |
2532 | if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) { | 2539 | if (!wmark_ok && waitqueue_active(&pgdat->kswapd_wait)) { |
2533 | pgdat->classzone_idx = min(pgdat->classzone_idx, | 2540 | pgdat->classzone_idx = min(pgdat->classzone_idx, |
2534 | (enum zone_type)ZONE_NORMAL); | 2541 | (enum zone_type)ZONE_NORMAL); |
2535 | wake_up_interruptible(&pgdat->kswapd_wait); | 2542 | wake_up_interruptible(&pgdat->kswapd_wait); |
2536 | } | 2543 | } |
2537 | 2544 | ||
2538 | return wmark_ok; | 2545 | return wmark_ok; |
2539 | } | 2546 | } |
2540 | 2547 | ||
2541 | /* | 2548 | /* |
2542 | * Throttle direct reclaimers if backing storage is backed by the network | 2549 | * Throttle direct reclaimers if backing storage is backed by the network |
2543 | * and the PFMEMALLOC reserve for the preferred node is getting dangerously | 2550 | * and the PFMEMALLOC reserve for the preferred node is getting dangerously |
2544 | * depleted. kswapd will continue to make progress and wake the processes | 2551 | * depleted. kswapd will continue to make progress and wake the processes |
2545 | * when the low watermark is reached. | 2552 | * when the low watermark is reached. |
2546 | * | 2553 | * |
2547 | * Returns true if a fatal signal was delivered during throttling. If this | 2554 | * Returns true if a fatal signal was delivered during throttling. If this |
2548 | * happens, the page allocator should not consider triggering the OOM killer. | 2555 | * happens, the page allocator should not consider triggering the OOM killer. |
2549 | */ | 2556 | */ |
2550 | static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist, | 2557 | static bool throttle_direct_reclaim(gfp_t gfp_mask, struct zonelist *zonelist, |
2551 | nodemask_t *nodemask) | 2558 | nodemask_t *nodemask) |
2552 | { | 2559 | { |
2553 | struct zoneref *z; | 2560 | struct zoneref *z; |
2554 | struct zone *zone; | 2561 | struct zone *zone; |
2555 | pg_data_t *pgdat = NULL; | 2562 | pg_data_t *pgdat = NULL; |
2556 | 2563 | ||
2557 | /* | 2564 | /* |
2558 | * Kernel threads should not be throttled as they may be indirectly | 2565 | * Kernel threads should not be throttled as they may be indirectly |
2559 | * responsible for cleaning pages necessary for reclaim to make forward | 2566 | * responsible for cleaning pages necessary for reclaim to make forward |
2560 | * progress. kjournald for example may enter direct reclaim while | 2567 | * progress. kjournald for example may enter direct reclaim while |
2561 | * committing a transaction where throttling it could forcing other | 2568 | * committing a transaction where throttling it could forcing other |
2562 | * processes to block on log_wait_commit(). | 2569 | * processes to block on log_wait_commit(). |
2563 | */ | 2570 | */ |
2564 | if (current->flags & PF_KTHREAD) | 2571 | if (current->flags & PF_KTHREAD) |
2565 | goto out; | 2572 | goto out; |
2566 | 2573 | ||
2567 | /* | 2574 | /* |
2568 | * If a fatal signal is pending, this process should not throttle. | 2575 | * If a fatal signal is pending, this process should not throttle. |
2569 | * It should return quickly so it can exit and free its memory | 2576 | * It should return quickly so it can exit and free its memory |
2570 | */ | 2577 | */ |
2571 | if (fatal_signal_pending(current)) | 2578 | if (fatal_signal_pending(current)) |
2572 | goto out; | 2579 | goto out; |
2573 | 2580 | ||
2574 | /* | 2581 | /* |
2575 | * Check if the pfmemalloc reserves are ok by finding the first node | 2582 | * Check if the pfmemalloc reserves are ok by finding the first node |
2576 | * with a usable ZONE_NORMAL or lower zone. The expectation is that | 2583 | * with a usable ZONE_NORMAL or lower zone. The expectation is that |
2577 | * GFP_KERNEL will be required for allocating network buffers when | 2584 | * GFP_KERNEL will be required for allocating network buffers when |
2578 | * swapping over the network so ZONE_HIGHMEM is unusable. | 2585 | * swapping over the network so ZONE_HIGHMEM is unusable. |
2579 | * | 2586 | * |
2580 | * Throttling is based on the first usable node and throttled processes | 2587 | * Throttling is based on the first usable node and throttled processes |
2581 | * wait on a queue until kswapd makes progress and wakes them. There | 2588 | * wait on a queue until kswapd makes progress and wakes them. There |
2582 | * is an affinity then between processes waking up and where reclaim | 2589 | * is an affinity then between processes waking up and where reclaim |
2583 | * progress has been made assuming the process wakes on the same node. | 2590 | * progress has been made assuming the process wakes on the same node. |
2584 | * More importantly, processes running on remote nodes will not compete | 2591 | * More importantly, processes running on remote nodes will not compete |
2585 | * for remote pfmemalloc reserves and processes on different nodes | 2592 | * for remote pfmemalloc reserves and processes on different nodes |
2586 | * should make reasonable progress. | 2593 | * should make reasonable progress. |
2587 | */ | 2594 | */ |
2588 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | 2595 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
2589 | gfp_mask, nodemask) { | 2596 | gfp_mask, nodemask) { |
2590 | if (zone_idx(zone) > ZONE_NORMAL) | 2597 | if (zone_idx(zone) > ZONE_NORMAL) |
2591 | continue; | 2598 | continue; |
2592 | 2599 | ||
2593 | /* Throttle based on the first usable node */ | 2600 | /* Throttle based on the first usable node */ |
2594 | pgdat = zone->zone_pgdat; | 2601 | pgdat = zone->zone_pgdat; |
2595 | if (pfmemalloc_watermark_ok(pgdat)) | 2602 | if (pfmemalloc_watermark_ok(pgdat)) |
2596 | goto out; | 2603 | goto out; |
2597 | break; | 2604 | break; |
2598 | } | 2605 | } |
2599 | 2606 | ||
2600 | /* If no zone was usable by the allocation flags then do not throttle */ | 2607 | /* If no zone was usable by the allocation flags then do not throttle */ |
2601 | if (!pgdat) | 2608 | if (!pgdat) |
2602 | goto out; | 2609 | goto out; |
2603 | 2610 | ||
2604 | /* Account for the throttling */ | 2611 | /* Account for the throttling */ |
2605 | count_vm_event(PGSCAN_DIRECT_THROTTLE); | 2612 | count_vm_event(PGSCAN_DIRECT_THROTTLE); |
2606 | 2613 | ||
2607 | /* | 2614 | /* |
2608 | * If the caller cannot enter the filesystem, it's possible that it | 2615 | * If the caller cannot enter the filesystem, it's possible that it |
2609 | * is due to the caller holding an FS lock or performing a journal | 2616 | * is due to the caller holding an FS lock or performing a journal |
2610 | * transaction in the case of a filesystem like ext[3|4]. In this case, | 2617 | * transaction in the case of a filesystem like ext[3|4]. In this case, |
2611 | * it is not safe to block on pfmemalloc_wait as kswapd could be | 2618 | * it is not safe to block on pfmemalloc_wait as kswapd could be |
2612 | * blocked waiting on the same lock. Instead, throttle for up to a | 2619 | * blocked waiting on the same lock. Instead, throttle for up to a |
2613 | * second before continuing. | 2620 | * second before continuing. |
2614 | */ | 2621 | */ |
2615 | if (!(gfp_mask & __GFP_FS)) { | 2622 | if (!(gfp_mask & __GFP_FS)) { |
2616 | wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, | 2623 | wait_event_interruptible_timeout(pgdat->pfmemalloc_wait, |
2617 | pfmemalloc_watermark_ok(pgdat), HZ); | 2624 | pfmemalloc_watermark_ok(pgdat), HZ); |
2618 | 2625 | ||
2619 | goto check_pending; | 2626 | goto check_pending; |
2620 | } | 2627 | } |
2621 | 2628 | ||
2622 | /* Throttle until kswapd wakes the process */ | 2629 | /* Throttle until kswapd wakes the process */ |
2623 | wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, | 2630 | wait_event_killable(zone->zone_pgdat->pfmemalloc_wait, |
2624 | pfmemalloc_watermark_ok(pgdat)); | 2631 | pfmemalloc_watermark_ok(pgdat)); |
2625 | 2632 | ||
2626 | check_pending: | 2633 | check_pending: |
2627 | if (fatal_signal_pending(current)) | 2634 | if (fatal_signal_pending(current)) |
2628 | return true; | 2635 | return true; |
2629 | 2636 | ||
2630 | out: | 2637 | out: |
2631 | return false; | 2638 | return false; |
2632 | } | 2639 | } |
2633 | 2640 | ||
2634 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, | 2641 | unsigned long try_to_free_pages(struct zonelist *zonelist, int order, |
2635 | gfp_t gfp_mask, nodemask_t *nodemask) | 2642 | gfp_t gfp_mask, nodemask_t *nodemask) |
2636 | { | 2643 | { |
2637 | unsigned long nr_reclaimed; | 2644 | unsigned long nr_reclaimed; |
2638 | struct scan_control sc = { | 2645 | struct scan_control sc = { |
2639 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), | 2646 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
2640 | .may_writepage = !laptop_mode, | 2647 | .may_writepage = !laptop_mode, |
2641 | .nr_to_reclaim = SWAP_CLUSTER_MAX, | 2648 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
2642 | .may_unmap = 1, | 2649 | .may_unmap = 1, |
2643 | .may_swap = 1, | 2650 | .may_swap = 1, |
2644 | .order = order, | 2651 | .order = order, |
2645 | .priority = DEF_PRIORITY, | 2652 | .priority = DEF_PRIORITY, |
2646 | .target_mem_cgroup = NULL, | 2653 | .target_mem_cgroup = NULL, |
2647 | .nodemask = nodemask, | 2654 | .nodemask = nodemask, |
2648 | }; | 2655 | }; |
2649 | struct shrink_control shrink = { | 2656 | struct shrink_control shrink = { |
2650 | .gfp_mask = sc.gfp_mask, | 2657 | .gfp_mask = sc.gfp_mask, |
2651 | }; | 2658 | }; |
2652 | 2659 | ||
2653 | /* | 2660 | /* |
2654 | * Do not enter reclaim if fatal signal was delivered while throttled. | 2661 | * Do not enter reclaim if fatal signal was delivered while throttled. |
2655 | * 1 is returned so that the page allocator does not OOM kill at this | 2662 | * 1 is returned so that the page allocator does not OOM kill at this |
2656 | * point. | 2663 | * point. |
2657 | */ | 2664 | */ |
2658 | if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask)) | 2665 | if (throttle_direct_reclaim(gfp_mask, zonelist, nodemask)) |
2659 | return 1; | 2666 | return 1; |
2660 | 2667 | ||
2661 | trace_mm_vmscan_direct_reclaim_begin(order, | 2668 | trace_mm_vmscan_direct_reclaim_begin(order, |
2662 | sc.may_writepage, | 2669 | sc.may_writepage, |
2663 | gfp_mask); | 2670 | gfp_mask); |
2664 | 2671 | ||
2665 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); | 2672 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
2666 | 2673 | ||
2667 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); | 2674 | trace_mm_vmscan_direct_reclaim_end(nr_reclaimed); |
2668 | 2675 | ||
2669 | return nr_reclaimed; | 2676 | return nr_reclaimed; |
2670 | } | 2677 | } |
2671 | 2678 | ||
2672 | #ifdef CONFIG_MEMCG | 2679 | #ifdef CONFIG_MEMCG |
2673 | 2680 | ||
2674 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg, | 2681 | unsigned long mem_cgroup_shrink_node_zone(struct mem_cgroup *memcg, |
2675 | gfp_t gfp_mask, bool noswap, | 2682 | gfp_t gfp_mask, bool noswap, |
2676 | struct zone *zone, | 2683 | struct zone *zone, |
2677 | unsigned long *nr_scanned) | 2684 | unsigned long *nr_scanned) |
2678 | { | 2685 | { |
2679 | struct scan_control sc = { | 2686 | struct scan_control sc = { |
2680 | .nr_scanned = 0, | 2687 | .nr_scanned = 0, |
2681 | .nr_to_reclaim = SWAP_CLUSTER_MAX, | 2688 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
2682 | .may_writepage = !laptop_mode, | 2689 | .may_writepage = !laptop_mode, |
2683 | .may_unmap = 1, | 2690 | .may_unmap = 1, |
2684 | .may_swap = !noswap, | 2691 | .may_swap = !noswap, |
2685 | .order = 0, | 2692 | .order = 0, |
2686 | .priority = 0, | 2693 | .priority = 0, |
2687 | .target_mem_cgroup = memcg, | 2694 | .target_mem_cgroup = memcg, |
2688 | }; | 2695 | }; |
2689 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); | 2696 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
2690 | 2697 | ||
2691 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | | 2698 | sc.gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2692 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); | 2699 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK); |
2693 | 2700 | ||
2694 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, | 2701 | trace_mm_vmscan_memcg_softlimit_reclaim_begin(sc.order, |
2695 | sc.may_writepage, | 2702 | sc.may_writepage, |
2696 | sc.gfp_mask); | 2703 | sc.gfp_mask); |
2697 | 2704 | ||
2698 | /* | 2705 | /* |
2699 | * NOTE: Although we can get the priority field, using it | 2706 | * NOTE: Although we can get the priority field, using it |
2700 | * here is not a good idea, since it limits the pages we can scan. | 2707 | * here is not a good idea, since it limits the pages we can scan. |
2701 | * if we don't reclaim here, the shrink_zone from balance_pgdat | 2708 | * if we don't reclaim here, the shrink_zone from balance_pgdat |
2702 | * will pick up pages from other mem cgroup's as well. We hack | 2709 | * will pick up pages from other mem cgroup's as well. We hack |
2703 | * the priority and make it zero. | 2710 | * the priority and make it zero. |
2704 | */ | 2711 | */ |
2705 | shrink_lruvec(lruvec, &sc); | 2712 | shrink_lruvec(lruvec, &sc); |
2706 | 2713 | ||
2707 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); | 2714 | trace_mm_vmscan_memcg_softlimit_reclaim_end(sc.nr_reclaimed); |
2708 | 2715 | ||
2709 | *nr_scanned = sc.nr_scanned; | 2716 | *nr_scanned = sc.nr_scanned; |
2710 | return sc.nr_reclaimed; | 2717 | return sc.nr_reclaimed; |
2711 | } | 2718 | } |
2712 | 2719 | ||
2713 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, | 2720 | unsigned long try_to_free_mem_cgroup_pages(struct mem_cgroup *memcg, |
2714 | gfp_t gfp_mask, | 2721 | gfp_t gfp_mask, |
2715 | bool noswap) | 2722 | bool noswap) |
2716 | { | 2723 | { |
2717 | struct zonelist *zonelist; | 2724 | struct zonelist *zonelist; |
2718 | unsigned long nr_reclaimed; | 2725 | unsigned long nr_reclaimed; |
2719 | int nid; | 2726 | int nid; |
2720 | struct scan_control sc = { | 2727 | struct scan_control sc = { |
2721 | .may_writepage = !laptop_mode, | 2728 | .may_writepage = !laptop_mode, |
2722 | .may_unmap = 1, | 2729 | .may_unmap = 1, |
2723 | .may_swap = !noswap, | 2730 | .may_swap = !noswap, |
2724 | .nr_to_reclaim = SWAP_CLUSTER_MAX, | 2731 | .nr_to_reclaim = SWAP_CLUSTER_MAX, |
2725 | .order = 0, | 2732 | .order = 0, |
2726 | .priority = DEF_PRIORITY, | 2733 | .priority = DEF_PRIORITY, |
2727 | .target_mem_cgroup = memcg, | 2734 | .target_mem_cgroup = memcg, |
2728 | .nodemask = NULL, /* we don't care the placement */ | 2735 | .nodemask = NULL, /* we don't care the placement */ |
2729 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | | 2736 | .gfp_mask = (gfp_mask & GFP_RECLAIM_MASK) | |
2730 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), | 2737 | (GFP_HIGHUSER_MOVABLE & ~GFP_RECLAIM_MASK), |
2731 | }; | 2738 | }; |
2732 | struct shrink_control shrink = { | 2739 | struct shrink_control shrink = { |
2733 | .gfp_mask = sc.gfp_mask, | 2740 | .gfp_mask = sc.gfp_mask, |
2734 | }; | 2741 | }; |
2735 | 2742 | ||
2736 | /* | 2743 | /* |
2737 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't | 2744 | * Unlike direct reclaim via alloc_pages(), memcg's reclaim doesn't |
2738 | * take care of from where we get pages. So the node where we start the | 2745 | * take care of from where we get pages. So the node where we start the |
2739 | * scan does not need to be the current node. | 2746 | * scan does not need to be the current node. |
2740 | */ | 2747 | */ |
2741 | nid = mem_cgroup_select_victim_node(memcg); | 2748 | nid = mem_cgroup_select_victim_node(memcg); |
2742 | 2749 | ||
2743 | zonelist = NODE_DATA(nid)->node_zonelists; | 2750 | zonelist = NODE_DATA(nid)->node_zonelists; |
2744 | 2751 | ||
2745 | trace_mm_vmscan_memcg_reclaim_begin(0, | 2752 | trace_mm_vmscan_memcg_reclaim_begin(0, |
2746 | sc.may_writepage, | 2753 | sc.may_writepage, |
2747 | sc.gfp_mask); | 2754 | sc.gfp_mask); |
2748 | 2755 | ||
2749 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); | 2756 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
2750 | 2757 | ||
2751 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); | 2758 | trace_mm_vmscan_memcg_reclaim_end(nr_reclaimed); |
2752 | 2759 | ||
2753 | return nr_reclaimed; | 2760 | return nr_reclaimed; |
2754 | } | 2761 | } |
2755 | #endif | 2762 | #endif |
2756 | 2763 | ||
2757 | static void age_active_anon(struct zone *zone, struct scan_control *sc) | 2764 | static void age_active_anon(struct zone *zone, struct scan_control *sc) |
2758 | { | 2765 | { |
2759 | struct mem_cgroup *memcg; | 2766 | struct mem_cgroup *memcg; |
2760 | 2767 | ||
2761 | if (!total_swap_pages) | 2768 | if (!total_swap_pages) |
2762 | return; | 2769 | return; |
2763 | 2770 | ||
2764 | memcg = mem_cgroup_iter(NULL, NULL, NULL); | 2771 | memcg = mem_cgroup_iter(NULL, NULL, NULL); |
2765 | do { | 2772 | do { |
2766 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); | 2773 | struct lruvec *lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
2767 | 2774 | ||
2768 | if (inactive_anon_is_low(lruvec)) | 2775 | if (inactive_anon_is_low(lruvec)) |
2769 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, | 2776 | shrink_active_list(SWAP_CLUSTER_MAX, lruvec, |
2770 | sc, LRU_ACTIVE_ANON); | 2777 | sc, LRU_ACTIVE_ANON); |
2771 | 2778 | ||
2772 | memcg = mem_cgroup_iter(NULL, memcg, NULL); | 2779 | memcg = mem_cgroup_iter(NULL, memcg, NULL); |
2773 | } while (memcg); | 2780 | } while (memcg); |
2774 | } | 2781 | } |
2775 | 2782 | ||
2776 | static bool zone_balanced(struct zone *zone, int order, | 2783 | static bool zone_balanced(struct zone *zone, int order, |
2777 | unsigned long balance_gap, int classzone_idx) | 2784 | unsigned long balance_gap, int classzone_idx) |
2778 | { | 2785 | { |
2779 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) + | 2786 | if (!zone_watermark_ok_safe(zone, order, high_wmark_pages(zone) + |
2780 | balance_gap, classzone_idx, 0)) | 2787 | balance_gap, classzone_idx, 0)) |
2781 | return false; | 2788 | return false; |
2782 | 2789 | ||
2783 | if (IS_ENABLED(CONFIG_COMPACTION) && order && | 2790 | if (IS_ENABLED(CONFIG_COMPACTION) && order && |
2784 | !compaction_suitable(zone, order)) | 2791 | !compaction_suitable(zone, order)) |
2785 | return false; | 2792 | return false; |
2786 | 2793 | ||
2787 | return true; | 2794 | return true; |
2788 | } | 2795 | } |
2789 | 2796 | ||
2790 | /* | 2797 | /* |
2791 | * pgdat_balanced() is used when checking if a node is balanced. | 2798 | * pgdat_balanced() is used when checking if a node is balanced. |
2792 | * | 2799 | * |
2793 | * For order-0, all zones must be balanced! | 2800 | * For order-0, all zones must be balanced! |
2794 | * | 2801 | * |
2795 | * For high-order allocations only zones that meet watermarks and are in a | 2802 | * For high-order allocations only zones that meet watermarks and are in a |
2796 | * zone allowed by the callers classzone_idx are added to balanced_pages. The | 2803 | * zone allowed by the callers classzone_idx are added to balanced_pages. The |
2797 | * total of balanced pages must be at least 25% of the zones allowed by | 2804 | * total of balanced pages must be at least 25% of the zones allowed by |
2798 | * classzone_idx for the node to be considered balanced. Forcing all zones to | 2805 | * classzone_idx for the node to be considered balanced. Forcing all zones to |
2799 | * be balanced for high orders can cause excessive reclaim when there are | 2806 | * be balanced for high orders can cause excessive reclaim when there are |
2800 | * imbalanced zones. | 2807 | * imbalanced zones. |
2801 | * The choice of 25% is due to | 2808 | * The choice of 25% is due to |
2802 | * o a 16M DMA zone that is balanced will not balance a zone on any | 2809 | * o a 16M DMA zone that is balanced will not balance a zone on any |
2803 | * reasonable sized machine | 2810 | * reasonable sized machine |
2804 | * o On all other machines, the top zone must be at least a reasonable | 2811 | * o On all other machines, the top zone must be at least a reasonable |
2805 | * percentage of the middle zones. For example, on 32-bit x86, highmem | 2812 | * percentage of the middle zones. For example, on 32-bit x86, highmem |
2806 | * would need to be at least 256M for it to be balance a whole node. | 2813 | * would need to be at least 256M for it to be balance a whole node. |
2807 | * Similarly, on x86-64 the Normal zone would need to be at least 1G | 2814 | * Similarly, on x86-64 the Normal zone would need to be at least 1G |
2808 | * to balance a node on its own. These seemed like reasonable ratios. | 2815 | * to balance a node on its own. These seemed like reasonable ratios. |
2809 | */ | 2816 | */ |
2810 | static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx) | 2817 | static bool pgdat_balanced(pg_data_t *pgdat, int order, int classzone_idx) |
2811 | { | 2818 | { |
2812 | unsigned long managed_pages = 0; | 2819 | unsigned long managed_pages = 0; |
2813 | unsigned long balanced_pages = 0; | 2820 | unsigned long balanced_pages = 0; |
2814 | int i; | 2821 | int i; |
2815 | 2822 | ||
2816 | /* Check the watermark levels */ | 2823 | /* Check the watermark levels */ |
2817 | for (i = 0; i <= classzone_idx; i++) { | 2824 | for (i = 0; i <= classzone_idx; i++) { |
2818 | struct zone *zone = pgdat->node_zones + i; | 2825 | struct zone *zone = pgdat->node_zones + i; |
2819 | 2826 | ||
2820 | if (!populated_zone(zone)) | 2827 | if (!populated_zone(zone)) |
2821 | continue; | 2828 | continue; |
2822 | 2829 | ||
2823 | managed_pages += zone->managed_pages; | 2830 | managed_pages += zone->managed_pages; |
2824 | 2831 | ||
2825 | /* | 2832 | /* |
2826 | * A special case here: | 2833 | * A special case here: |
2827 | * | 2834 | * |
2828 | * balance_pgdat() skips over all_unreclaimable after | 2835 | * balance_pgdat() skips over all_unreclaimable after |
2829 | * DEF_PRIORITY. Effectively, it considers them balanced so | 2836 | * DEF_PRIORITY. Effectively, it considers them balanced so |
2830 | * they must be considered balanced here as well! | 2837 | * they must be considered balanced here as well! |
2831 | */ | 2838 | */ |
2832 | if (!zone_reclaimable(zone)) { | 2839 | if (!zone_reclaimable(zone)) { |
2833 | balanced_pages += zone->managed_pages; | 2840 | balanced_pages += zone->managed_pages; |
2834 | continue; | 2841 | continue; |
2835 | } | 2842 | } |
2836 | 2843 | ||
2837 | if (zone_balanced(zone, order, 0, i)) | 2844 | if (zone_balanced(zone, order, 0, i)) |
2838 | balanced_pages += zone->managed_pages; | 2845 | balanced_pages += zone->managed_pages; |
2839 | else if (!order) | 2846 | else if (!order) |
2840 | return false; | 2847 | return false; |
2841 | } | 2848 | } |
2842 | 2849 | ||
2843 | if (order) | 2850 | if (order) |
2844 | return balanced_pages >= (managed_pages >> 2); | 2851 | return balanced_pages >= (managed_pages >> 2); |
2845 | else | 2852 | else |
2846 | return true; | 2853 | return true; |
2847 | } | 2854 | } |
2848 | 2855 | ||
2849 | /* | 2856 | /* |
2850 | * Prepare kswapd for sleeping. This verifies that there are no processes | 2857 | * Prepare kswapd for sleeping. This verifies that there are no processes |
2851 | * waiting in throttle_direct_reclaim() and that watermarks have been met. | 2858 | * waiting in throttle_direct_reclaim() and that watermarks have been met. |
2852 | * | 2859 | * |
2853 | * Returns true if kswapd is ready to sleep | 2860 | * Returns true if kswapd is ready to sleep |
2854 | */ | 2861 | */ |
2855 | static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining, | 2862 | static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, long remaining, |
2856 | int classzone_idx) | 2863 | int classzone_idx) |
2857 | { | 2864 | { |
2858 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ | 2865 | /* If a direct reclaimer woke kswapd within HZ/10, it's premature */ |
2859 | if (remaining) | 2866 | if (remaining) |
2860 | return false; | 2867 | return false; |
2861 | 2868 | ||
2862 | /* | 2869 | /* |
2863 | * There is a potential race between when kswapd checks its watermarks | 2870 | * There is a potential race between when kswapd checks its watermarks |
2864 | * and a process gets throttled. There is also a potential race if | 2871 | * and a process gets throttled. There is also a potential race if |
2865 | * processes get throttled, kswapd wakes, a large process exits therby | 2872 | * processes get throttled, kswapd wakes, a large process exits therby |
2866 | * balancing the zones that causes kswapd to miss a wakeup. If kswapd | 2873 | * balancing the zones that causes kswapd to miss a wakeup. If kswapd |
2867 | * is going to sleep, no process should be sleeping on pfmemalloc_wait | 2874 | * is going to sleep, no process should be sleeping on pfmemalloc_wait |
2868 | * so wake them now if necessary. If necessary, processes will wake | 2875 | * so wake them now if necessary. If necessary, processes will wake |
2869 | * kswapd and get throttled again | 2876 | * kswapd and get throttled again |
2870 | */ | 2877 | */ |
2871 | if (waitqueue_active(&pgdat->pfmemalloc_wait)) { | 2878 | if (waitqueue_active(&pgdat->pfmemalloc_wait)) { |
2872 | wake_up(&pgdat->pfmemalloc_wait); | 2879 | wake_up(&pgdat->pfmemalloc_wait); |
2873 | return false; | 2880 | return false; |
2874 | } | 2881 | } |
2875 | 2882 | ||
2876 | return pgdat_balanced(pgdat, order, classzone_idx); | 2883 | return pgdat_balanced(pgdat, order, classzone_idx); |
2877 | } | 2884 | } |
2878 | 2885 | ||
2879 | /* | 2886 | /* |
2880 | * kswapd shrinks the zone by the number of pages required to reach | 2887 | * kswapd shrinks the zone by the number of pages required to reach |
2881 | * the high watermark. | 2888 | * the high watermark. |
2882 | * | 2889 | * |
2883 | * Returns true if kswapd scanned at least the requested number of pages to | 2890 | * Returns true if kswapd scanned at least the requested number of pages to |
2884 | * reclaim or if the lack of progress was due to pages under writeback. | 2891 | * reclaim or if the lack of progress was due to pages under writeback. |
2885 | * This is used to determine if the scanning priority needs to be raised. | 2892 | * This is used to determine if the scanning priority needs to be raised. |
2886 | */ | 2893 | */ |
2887 | static bool kswapd_shrink_zone(struct zone *zone, | 2894 | static bool kswapd_shrink_zone(struct zone *zone, |
2888 | int classzone_idx, | 2895 | int classzone_idx, |
2889 | struct scan_control *sc, | 2896 | struct scan_control *sc, |
2890 | unsigned long lru_pages, | 2897 | unsigned long lru_pages, |
2891 | unsigned long *nr_attempted) | 2898 | unsigned long *nr_attempted) |
2892 | { | 2899 | { |
2893 | int testorder = sc->order; | 2900 | int testorder = sc->order; |
2894 | unsigned long balance_gap; | 2901 | unsigned long balance_gap; |
2895 | struct reclaim_state *reclaim_state = current->reclaim_state; | 2902 | struct reclaim_state *reclaim_state = current->reclaim_state; |
2896 | struct shrink_control shrink = { | 2903 | struct shrink_control shrink = { |
2897 | .gfp_mask = sc->gfp_mask, | 2904 | .gfp_mask = sc->gfp_mask, |
2898 | }; | 2905 | }; |
2899 | bool lowmem_pressure; | 2906 | bool lowmem_pressure; |
2900 | 2907 | ||
2901 | /* Reclaim above the high watermark. */ | 2908 | /* Reclaim above the high watermark. */ |
2902 | sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone)); | 2909 | sc->nr_to_reclaim = max(SWAP_CLUSTER_MAX, high_wmark_pages(zone)); |
2903 | 2910 | ||
2904 | /* | 2911 | /* |
2905 | * Kswapd reclaims only single pages with compaction enabled. Trying | 2912 | * Kswapd reclaims only single pages with compaction enabled. Trying |
2906 | * too hard to reclaim until contiguous free pages have become | 2913 | * too hard to reclaim until contiguous free pages have become |
2907 | * available can hurt performance by evicting too much useful data | 2914 | * available can hurt performance by evicting too much useful data |
2908 | * from memory. Do not reclaim more than needed for compaction. | 2915 | * from memory. Do not reclaim more than needed for compaction. |
2909 | */ | 2916 | */ |
2910 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && | 2917 | if (IS_ENABLED(CONFIG_COMPACTION) && sc->order && |
2911 | compaction_suitable(zone, sc->order) != | 2918 | compaction_suitable(zone, sc->order) != |
2912 | COMPACT_SKIPPED) | 2919 | COMPACT_SKIPPED) |
2913 | testorder = 0; | 2920 | testorder = 0; |
2914 | 2921 | ||
2915 | /* | 2922 | /* |
2916 | * We put equal pressure on every zone, unless one zone has way too | 2923 | * We put equal pressure on every zone, unless one zone has way too |
2917 | * many pages free already. The "too many pages" is defined as the | 2924 | * many pages free already. The "too many pages" is defined as the |
2918 | * high wmark plus a "gap" where the gap is either the low | 2925 | * high wmark plus a "gap" where the gap is either the low |
2919 | * watermark or 1% of the zone, whichever is smaller. | 2926 | * watermark or 1% of the zone, whichever is smaller. |
2920 | */ | 2927 | */ |
2921 | balance_gap = min(low_wmark_pages(zone), | 2928 | balance_gap = min(low_wmark_pages(zone), |
2922 | (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / | 2929 | (zone->managed_pages + KSWAPD_ZONE_BALANCE_GAP_RATIO-1) / |
2923 | KSWAPD_ZONE_BALANCE_GAP_RATIO); | 2930 | KSWAPD_ZONE_BALANCE_GAP_RATIO); |
2924 | 2931 | ||
2925 | /* | 2932 | /* |
2926 | * If there is no low memory pressure or the zone is balanced then no | 2933 | * If there is no low memory pressure or the zone is balanced then no |
2927 | * reclaim is necessary | 2934 | * reclaim is necessary |
2928 | */ | 2935 | */ |
2929 | lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone)); | 2936 | lowmem_pressure = (buffer_heads_over_limit && is_highmem(zone)); |
2930 | if (!lowmem_pressure && zone_balanced(zone, testorder, | 2937 | if (!lowmem_pressure && zone_balanced(zone, testorder, |
2931 | balance_gap, classzone_idx)) | 2938 | balance_gap, classzone_idx)) |
2932 | return true; | 2939 | return true; |
2933 | 2940 | ||
2934 | shrink_zone(zone, sc); | 2941 | shrink_zone(zone, sc); |
2935 | nodes_clear(shrink.nodes_to_scan); | 2942 | nodes_clear(shrink.nodes_to_scan); |
2936 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); | 2943 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); |
2937 | 2944 | ||
2938 | reclaim_state->reclaimed_slab = 0; | 2945 | reclaim_state->reclaimed_slab = 0; |
2939 | shrink_slab(&shrink, sc->nr_scanned, lru_pages); | 2946 | shrink_slab(&shrink, sc->nr_scanned, lru_pages); |
2940 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; | 2947 | sc->nr_reclaimed += reclaim_state->reclaimed_slab; |
2941 | 2948 | ||
2942 | /* Account for the number of pages attempted to reclaim */ | 2949 | /* Account for the number of pages attempted to reclaim */ |
2943 | *nr_attempted += sc->nr_to_reclaim; | 2950 | *nr_attempted += sc->nr_to_reclaim; |
2944 | 2951 | ||
2945 | zone_clear_flag(zone, ZONE_WRITEBACK); | 2952 | zone_clear_flag(zone, ZONE_WRITEBACK); |
2946 | 2953 | ||
2947 | /* | 2954 | /* |
2948 | * If a zone reaches its high watermark, consider it to be no longer | 2955 | * If a zone reaches its high watermark, consider it to be no longer |
2949 | * congested. It's possible there are dirty pages backed by congested | 2956 | * congested. It's possible there are dirty pages backed by congested |
2950 | * BDIs but as pressure is relieved, speculatively avoid congestion | 2957 | * BDIs but as pressure is relieved, speculatively avoid congestion |
2951 | * waits. | 2958 | * waits. |
2952 | */ | 2959 | */ |
2953 | if (zone_reclaimable(zone) && | 2960 | if (zone_reclaimable(zone) && |
2954 | zone_balanced(zone, testorder, 0, classzone_idx)) { | 2961 | zone_balanced(zone, testorder, 0, classzone_idx)) { |
2955 | zone_clear_flag(zone, ZONE_CONGESTED); | 2962 | zone_clear_flag(zone, ZONE_CONGESTED); |
2956 | zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY); | 2963 | zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY); |
2957 | } | 2964 | } |
2958 | 2965 | ||
2959 | return sc->nr_scanned >= sc->nr_to_reclaim; | 2966 | return sc->nr_scanned >= sc->nr_to_reclaim; |
2960 | } | 2967 | } |
2961 | 2968 | ||
2962 | /* | 2969 | /* |
2963 | * For kswapd, balance_pgdat() will work across all this node's zones until | 2970 | * For kswapd, balance_pgdat() will work across all this node's zones until |
2964 | * they are all at high_wmark_pages(zone). | 2971 | * they are all at high_wmark_pages(zone). |
2965 | * | 2972 | * |
2966 | * Returns the final order kswapd was reclaiming at | 2973 | * Returns the final order kswapd was reclaiming at |
2967 | * | 2974 | * |
2968 | * There is special handling here for zones which are full of pinned pages. | 2975 | * There is special handling here for zones which are full of pinned pages. |
2969 | * This can happen if the pages are all mlocked, or if they are all used by | 2976 | * This can happen if the pages are all mlocked, or if they are all used by |
2970 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. | 2977 | * device drivers (say, ZONE_DMA). Or if they are all in use by hugetlb. |
2971 | * What we do is to detect the case where all pages in the zone have been | 2978 | * What we do is to detect the case where all pages in the zone have been |
2972 | * scanned twice and there has been zero successful reclaim. Mark the zone as | 2979 | * scanned twice and there has been zero successful reclaim. Mark the zone as |
2973 | * dead and from now on, only perform a short scan. Basically we're polling | 2980 | * dead and from now on, only perform a short scan. Basically we're polling |
2974 | * the zone for when the problem goes away. | 2981 | * the zone for when the problem goes away. |
2975 | * | 2982 | * |
2976 | * kswapd scans the zones in the highmem->normal->dma direction. It skips | 2983 | * kswapd scans the zones in the highmem->normal->dma direction. It skips |
2977 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is | 2984 | * zones which have free_pages > high_wmark_pages(zone), but once a zone is |
2978 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the | 2985 | * found to have free_pages <= high_wmark_pages(zone), we scan that zone and the |
2979 | * lower zones regardless of the number of free pages in the lower zones. This | 2986 | * lower zones regardless of the number of free pages in the lower zones. This |
2980 | * interoperates with the page allocator fallback scheme to ensure that aging | 2987 | * interoperates with the page allocator fallback scheme to ensure that aging |
2981 | * of pages is balanced across the zones. | 2988 | * of pages is balanced across the zones. |
2982 | */ | 2989 | */ |
2983 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, | 2990 | static unsigned long balance_pgdat(pg_data_t *pgdat, int order, |
2984 | int *classzone_idx) | 2991 | int *classzone_idx) |
2985 | { | 2992 | { |
2986 | int i; | 2993 | int i; |
2987 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ | 2994 | int end_zone = 0; /* Inclusive. 0 = ZONE_DMA */ |
2988 | unsigned long nr_soft_reclaimed; | 2995 | unsigned long nr_soft_reclaimed; |
2989 | unsigned long nr_soft_scanned; | 2996 | unsigned long nr_soft_scanned; |
2990 | struct scan_control sc = { | 2997 | struct scan_control sc = { |
2991 | .gfp_mask = GFP_KERNEL, | 2998 | .gfp_mask = GFP_KERNEL, |
2992 | .priority = DEF_PRIORITY, | 2999 | .priority = DEF_PRIORITY, |
2993 | .may_unmap = 1, | 3000 | .may_unmap = 1, |
2994 | .may_swap = 1, | 3001 | .may_swap = 1, |
2995 | .may_writepage = !laptop_mode, | 3002 | .may_writepage = !laptop_mode, |
2996 | .order = order, | 3003 | .order = order, |
2997 | .target_mem_cgroup = NULL, | 3004 | .target_mem_cgroup = NULL, |
2998 | }; | 3005 | }; |
2999 | count_vm_event(PAGEOUTRUN); | 3006 | count_vm_event(PAGEOUTRUN); |
3000 | 3007 | ||
3001 | do { | 3008 | do { |
3002 | unsigned long lru_pages = 0; | 3009 | unsigned long lru_pages = 0; |
3003 | unsigned long nr_attempted = 0; | 3010 | unsigned long nr_attempted = 0; |
3004 | bool raise_priority = true; | 3011 | bool raise_priority = true; |
3005 | bool pgdat_needs_compaction = (order > 0); | 3012 | bool pgdat_needs_compaction = (order > 0); |
3006 | 3013 | ||
3007 | sc.nr_reclaimed = 0; | 3014 | sc.nr_reclaimed = 0; |
3008 | 3015 | ||
3009 | /* | 3016 | /* |
3010 | * Scan in the highmem->dma direction for the highest | 3017 | * Scan in the highmem->dma direction for the highest |
3011 | * zone which needs scanning | 3018 | * zone which needs scanning |
3012 | */ | 3019 | */ |
3013 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { | 3020 | for (i = pgdat->nr_zones - 1; i >= 0; i--) { |
3014 | struct zone *zone = pgdat->node_zones + i; | 3021 | struct zone *zone = pgdat->node_zones + i; |
3015 | 3022 | ||
3016 | if (!populated_zone(zone)) | 3023 | if (!populated_zone(zone)) |
3017 | continue; | 3024 | continue; |
3018 | 3025 | ||
3019 | if (sc.priority != DEF_PRIORITY && | 3026 | if (sc.priority != DEF_PRIORITY && |
3020 | !zone_reclaimable(zone)) | 3027 | !zone_reclaimable(zone)) |
3021 | continue; | 3028 | continue; |
3022 | 3029 | ||
3023 | /* | 3030 | /* |
3024 | * Do some background aging of the anon list, to give | 3031 | * Do some background aging of the anon list, to give |
3025 | * pages a chance to be referenced before reclaiming. | 3032 | * pages a chance to be referenced before reclaiming. |
3026 | */ | 3033 | */ |
3027 | age_active_anon(zone, &sc); | 3034 | age_active_anon(zone, &sc); |
3028 | 3035 | ||
3029 | /* | 3036 | /* |
3030 | * If the number of buffer_heads in the machine | 3037 | * If the number of buffer_heads in the machine |
3031 | * exceeds the maximum allowed level and this node | 3038 | * exceeds the maximum allowed level and this node |
3032 | * has a highmem zone, force kswapd to reclaim from | 3039 | * has a highmem zone, force kswapd to reclaim from |
3033 | * it to relieve lowmem pressure. | 3040 | * it to relieve lowmem pressure. |
3034 | */ | 3041 | */ |
3035 | if (buffer_heads_over_limit && is_highmem_idx(i)) { | 3042 | if (buffer_heads_over_limit && is_highmem_idx(i)) { |
3036 | end_zone = i; | 3043 | end_zone = i; |
3037 | break; | 3044 | break; |
3038 | } | 3045 | } |
3039 | 3046 | ||
3040 | if (!zone_balanced(zone, order, 0, 0)) { | 3047 | if (!zone_balanced(zone, order, 0, 0)) { |
3041 | end_zone = i; | 3048 | end_zone = i; |
3042 | break; | 3049 | break; |
3043 | } else { | 3050 | } else { |
3044 | /* | 3051 | /* |
3045 | * If balanced, clear the dirty and congested | 3052 | * If balanced, clear the dirty and congested |
3046 | * flags | 3053 | * flags |
3047 | */ | 3054 | */ |
3048 | zone_clear_flag(zone, ZONE_CONGESTED); | 3055 | zone_clear_flag(zone, ZONE_CONGESTED); |
3049 | zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY); | 3056 | zone_clear_flag(zone, ZONE_TAIL_LRU_DIRTY); |
3050 | } | 3057 | } |
3051 | } | 3058 | } |
3052 | 3059 | ||
3053 | if (i < 0) | 3060 | if (i < 0) |
3054 | goto out; | 3061 | goto out; |
3055 | 3062 | ||
3056 | for (i = 0; i <= end_zone; i++) { | 3063 | for (i = 0; i <= end_zone; i++) { |
3057 | struct zone *zone = pgdat->node_zones + i; | 3064 | struct zone *zone = pgdat->node_zones + i; |
3058 | 3065 | ||
3059 | if (!populated_zone(zone)) | 3066 | if (!populated_zone(zone)) |
3060 | continue; | 3067 | continue; |
3061 | 3068 | ||
3062 | lru_pages += zone_reclaimable_pages(zone); | 3069 | lru_pages += zone_reclaimable_pages(zone); |
3063 | 3070 | ||
3064 | /* | 3071 | /* |
3065 | * If any zone is currently balanced then kswapd will | 3072 | * If any zone is currently balanced then kswapd will |
3066 | * not call compaction as it is expected that the | 3073 | * not call compaction as it is expected that the |
3067 | * necessary pages are already available. | 3074 | * necessary pages are already available. |
3068 | */ | 3075 | */ |
3069 | if (pgdat_needs_compaction && | 3076 | if (pgdat_needs_compaction && |
3070 | zone_watermark_ok(zone, order, | 3077 | zone_watermark_ok(zone, order, |
3071 | low_wmark_pages(zone), | 3078 | low_wmark_pages(zone), |
3072 | *classzone_idx, 0)) | 3079 | *classzone_idx, 0)) |
3073 | pgdat_needs_compaction = false; | 3080 | pgdat_needs_compaction = false; |
3074 | } | 3081 | } |
3075 | 3082 | ||
3076 | /* | 3083 | /* |
3077 | * If we're getting trouble reclaiming, start doing writepage | 3084 | * If we're getting trouble reclaiming, start doing writepage |
3078 | * even in laptop mode. | 3085 | * even in laptop mode. |
3079 | */ | 3086 | */ |
3080 | if (sc.priority < DEF_PRIORITY - 2) | 3087 | if (sc.priority < DEF_PRIORITY - 2) |
3081 | sc.may_writepage = 1; | 3088 | sc.may_writepage = 1; |
3082 | 3089 | ||
3083 | /* | 3090 | /* |
3084 | * Now scan the zone in the dma->highmem direction, stopping | 3091 | * Now scan the zone in the dma->highmem direction, stopping |
3085 | * at the last zone which needs scanning. | 3092 | * at the last zone which needs scanning. |
3086 | * | 3093 | * |
3087 | * We do this because the page allocator works in the opposite | 3094 | * We do this because the page allocator works in the opposite |
3088 | * direction. This prevents the page allocator from allocating | 3095 | * direction. This prevents the page allocator from allocating |
3089 | * pages behind kswapd's direction of progress, which would | 3096 | * pages behind kswapd's direction of progress, which would |
3090 | * cause too much scanning of the lower zones. | 3097 | * cause too much scanning of the lower zones. |
3091 | */ | 3098 | */ |
3092 | for (i = 0; i <= end_zone; i++) { | 3099 | for (i = 0; i <= end_zone; i++) { |
3093 | struct zone *zone = pgdat->node_zones + i; | 3100 | struct zone *zone = pgdat->node_zones + i; |
3094 | 3101 | ||
3095 | if (!populated_zone(zone)) | 3102 | if (!populated_zone(zone)) |
3096 | continue; | 3103 | continue; |
3097 | 3104 | ||
3098 | if (sc.priority != DEF_PRIORITY && | 3105 | if (sc.priority != DEF_PRIORITY && |
3099 | !zone_reclaimable(zone)) | 3106 | !zone_reclaimable(zone)) |
3100 | continue; | 3107 | continue; |
3101 | 3108 | ||
3102 | sc.nr_scanned = 0; | 3109 | sc.nr_scanned = 0; |
3103 | 3110 | ||
3104 | nr_soft_scanned = 0; | 3111 | nr_soft_scanned = 0; |
3105 | /* | 3112 | /* |
3106 | * Call soft limit reclaim before calling shrink_zone. | 3113 | * Call soft limit reclaim before calling shrink_zone. |
3107 | */ | 3114 | */ |
3108 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, | 3115 | nr_soft_reclaimed = mem_cgroup_soft_limit_reclaim(zone, |
3109 | order, sc.gfp_mask, | 3116 | order, sc.gfp_mask, |
3110 | &nr_soft_scanned); | 3117 | &nr_soft_scanned); |
3111 | sc.nr_reclaimed += nr_soft_reclaimed; | 3118 | sc.nr_reclaimed += nr_soft_reclaimed; |
3112 | 3119 | ||
3113 | /* | 3120 | /* |
3114 | * There should be no need to raise the scanning | 3121 | * There should be no need to raise the scanning |
3115 | * priority if enough pages are already being scanned | 3122 | * priority if enough pages are already being scanned |
3116 | * that that high watermark would be met at 100% | 3123 | * that that high watermark would be met at 100% |
3117 | * efficiency. | 3124 | * efficiency. |
3118 | */ | 3125 | */ |
3119 | if (kswapd_shrink_zone(zone, end_zone, &sc, | 3126 | if (kswapd_shrink_zone(zone, end_zone, &sc, |
3120 | lru_pages, &nr_attempted)) | 3127 | lru_pages, &nr_attempted)) |
3121 | raise_priority = false; | 3128 | raise_priority = false; |
3122 | } | 3129 | } |
3123 | 3130 | ||
3124 | /* | 3131 | /* |
3125 | * If the low watermark is met there is no need for processes | 3132 | * If the low watermark is met there is no need for processes |
3126 | * to be throttled on pfmemalloc_wait as they should not be | 3133 | * to be throttled on pfmemalloc_wait as they should not be |
3127 | * able to safely make forward progress. Wake them | 3134 | * able to safely make forward progress. Wake them |
3128 | */ | 3135 | */ |
3129 | if (waitqueue_active(&pgdat->pfmemalloc_wait) && | 3136 | if (waitqueue_active(&pgdat->pfmemalloc_wait) && |
3130 | pfmemalloc_watermark_ok(pgdat)) | 3137 | pfmemalloc_watermark_ok(pgdat)) |
3131 | wake_up(&pgdat->pfmemalloc_wait); | 3138 | wake_up(&pgdat->pfmemalloc_wait); |
3132 | 3139 | ||
3133 | /* | 3140 | /* |
3134 | * Fragmentation may mean that the system cannot be rebalanced | 3141 | * Fragmentation may mean that the system cannot be rebalanced |
3135 | * for high-order allocations in all zones. If twice the | 3142 | * for high-order allocations in all zones. If twice the |
3136 | * allocation size has been reclaimed and the zones are still | 3143 | * allocation size has been reclaimed and the zones are still |
3137 | * not balanced then recheck the watermarks at order-0 to | 3144 | * not balanced then recheck the watermarks at order-0 to |
3138 | * prevent kswapd reclaiming excessively. Assume that a | 3145 | * prevent kswapd reclaiming excessively. Assume that a |
3139 | * process requested a high-order can direct reclaim/compact. | 3146 | * process requested a high-order can direct reclaim/compact. |
3140 | */ | 3147 | */ |
3141 | if (order && sc.nr_reclaimed >= 2UL << order) | 3148 | if (order && sc.nr_reclaimed >= 2UL << order) |
3142 | order = sc.order = 0; | 3149 | order = sc.order = 0; |
3143 | 3150 | ||
3144 | /* Check if kswapd should be suspending */ | 3151 | /* Check if kswapd should be suspending */ |
3145 | if (try_to_freeze() || kthread_should_stop()) | 3152 | if (try_to_freeze() || kthread_should_stop()) |
3146 | break; | 3153 | break; |
3147 | 3154 | ||
3148 | /* | 3155 | /* |
3149 | * Compact if necessary and kswapd is reclaiming at least the | 3156 | * Compact if necessary and kswapd is reclaiming at least the |
3150 | * high watermark number of pages as requsted | 3157 | * high watermark number of pages as requsted |
3151 | */ | 3158 | */ |
3152 | if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted) | 3159 | if (pgdat_needs_compaction && sc.nr_reclaimed > nr_attempted) |
3153 | compact_pgdat(pgdat, order); | 3160 | compact_pgdat(pgdat, order); |
3154 | 3161 | ||
3155 | /* | 3162 | /* |
3156 | * Raise priority if scanning rate is too low or there was no | 3163 | * Raise priority if scanning rate is too low or there was no |
3157 | * progress in reclaiming pages | 3164 | * progress in reclaiming pages |
3158 | */ | 3165 | */ |
3159 | if (raise_priority || !sc.nr_reclaimed) | 3166 | if (raise_priority || !sc.nr_reclaimed) |
3160 | sc.priority--; | 3167 | sc.priority--; |
3161 | } while (sc.priority >= 1 && | 3168 | } while (sc.priority >= 1 && |
3162 | !pgdat_balanced(pgdat, order, *classzone_idx)); | 3169 | !pgdat_balanced(pgdat, order, *classzone_idx)); |
3163 | 3170 | ||
3164 | out: | 3171 | out: |
3165 | /* | 3172 | /* |
3166 | * Return the order we were reclaiming at so prepare_kswapd_sleep() | 3173 | * Return the order we were reclaiming at so prepare_kswapd_sleep() |
3167 | * makes a decision on the order we were last reclaiming at. However, | 3174 | * makes a decision on the order we were last reclaiming at. However, |
3168 | * if another caller entered the allocator slow path while kswapd | 3175 | * if another caller entered the allocator slow path while kswapd |
3169 | * was awake, order will remain at the higher level | 3176 | * was awake, order will remain at the higher level |
3170 | */ | 3177 | */ |
3171 | *classzone_idx = end_zone; | 3178 | *classzone_idx = end_zone; |
3172 | return order; | 3179 | return order; |
3173 | } | 3180 | } |
3174 | 3181 | ||
3175 | static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx) | 3182 | static void kswapd_try_to_sleep(pg_data_t *pgdat, int order, int classzone_idx) |
3176 | { | 3183 | { |
3177 | long remaining = 0; | 3184 | long remaining = 0; |
3178 | DEFINE_WAIT(wait); | 3185 | DEFINE_WAIT(wait); |
3179 | 3186 | ||
3180 | if (freezing(current) || kthread_should_stop()) | 3187 | if (freezing(current) || kthread_should_stop()) |
3181 | return; | 3188 | return; |
3182 | 3189 | ||
3183 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | 3190 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
3184 | 3191 | ||
3185 | /* Try to sleep for a short interval */ | 3192 | /* Try to sleep for a short interval */ |
3186 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { | 3193 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
3187 | remaining = schedule_timeout(HZ/10); | 3194 | remaining = schedule_timeout(HZ/10); |
3188 | finish_wait(&pgdat->kswapd_wait, &wait); | 3195 | finish_wait(&pgdat->kswapd_wait, &wait); |
3189 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); | 3196 | prepare_to_wait(&pgdat->kswapd_wait, &wait, TASK_INTERRUPTIBLE); |
3190 | } | 3197 | } |
3191 | 3198 | ||
3192 | /* | 3199 | /* |
3193 | * After a short sleep, check if it was a premature sleep. If not, then | 3200 | * After a short sleep, check if it was a premature sleep. If not, then |
3194 | * go fully to sleep until explicitly woken up. | 3201 | * go fully to sleep until explicitly woken up. |
3195 | */ | 3202 | */ |
3196 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { | 3203 | if (prepare_kswapd_sleep(pgdat, order, remaining, classzone_idx)) { |
3197 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); | 3204 | trace_mm_vmscan_kswapd_sleep(pgdat->node_id); |
3198 | 3205 | ||
3199 | /* | 3206 | /* |
3200 | * vmstat counters are not perfectly accurate and the estimated | 3207 | * vmstat counters are not perfectly accurate and the estimated |
3201 | * value for counters such as NR_FREE_PAGES can deviate from the | 3208 | * value for counters such as NR_FREE_PAGES can deviate from the |
3202 | * true value by nr_online_cpus * threshold. To avoid the zone | 3209 | * true value by nr_online_cpus * threshold. To avoid the zone |
3203 | * watermarks being breached while under pressure, we reduce the | 3210 | * watermarks being breached while under pressure, we reduce the |
3204 | * per-cpu vmstat threshold while kswapd is awake and restore | 3211 | * per-cpu vmstat threshold while kswapd is awake and restore |
3205 | * them before going back to sleep. | 3212 | * them before going back to sleep. |
3206 | */ | 3213 | */ |
3207 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); | 3214 | set_pgdat_percpu_threshold(pgdat, calculate_normal_threshold); |
3208 | 3215 | ||
3209 | /* | 3216 | /* |
3210 | * Compaction records what page blocks it recently failed to | 3217 | * Compaction records what page blocks it recently failed to |
3211 | * isolate pages from and skips them in the future scanning. | 3218 | * isolate pages from and skips them in the future scanning. |
3212 | * When kswapd is going to sleep, it is reasonable to assume | 3219 | * When kswapd is going to sleep, it is reasonable to assume |
3213 | * that pages and compaction may succeed so reset the cache. | 3220 | * that pages and compaction may succeed so reset the cache. |
3214 | */ | 3221 | */ |
3215 | reset_isolation_suitable(pgdat); | 3222 | reset_isolation_suitable(pgdat); |
3216 | 3223 | ||
3217 | if (!kthread_should_stop()) | 3224 | if (!kthread_should_stop()) |
3218 | schedule(); | 3225 | schedule(); |
3219 | 3226 | ||
3220 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); | 3227 | set_pgdat_percpu_threshold(pgdat, calculate_pressure_threshold); |
3221 | } else { | 3228 | } else { |
3222 | if (remaining) | 3229 | if (remaining) |
3223 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); | 3230 | count_vm_event(KSWAPD_LOW_WMARK_HIT_QUICKLY); |
3224 | else | 3231 | else |
3225 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); | 3232 | count_vm_event(KSWAPD_HIGH_WMARK_HIT_QUICKLY); |
3226 | } | 3233 | } |
3227 | finish_wait(&pgdat->kswapd_wait, &wait); | 3234 | finish_wait(&pgdat->kswapd_wait, &wait); |
3228 | } | 3235 | } |
3229 | 3236 | ||
3230 | /* | 3237 | /* |
3231 | * The background pageout daemon, started as a kernel thread | 3238 | * The background pageout daemon, started as a kernel thread |
3232 | * from the init process. | 3239 | * from the init process. |
3233 | * | 3240 | * |
3234 | * This basically trickles out pages so that we have _some_ | 3241 | * This basically trickles out pages so that we have _some_ |
3235 | * free memory available even if there is no other activity | 3242 | * free memory available even if there is no other activity |
3236 | * that frees anything up. This is needed for things like routing | 3243 | * that frees anything up. This is needed for things like routing |
3237 | * etc, where we otherwise might have all activity going on in | 3244 | * etc, where we otherwise might have all activity going on in |
3238 | * asynchronous contexts that cannot page things out. | 3245 | * asynchronous contexts that cannot page things out. |
3239 | * | 3246 | * |
3240 | * If there are applications that are active memory-allocators | 3247 | * If there are applications that are active memory-allocators |
3241 | * (most normal use), this basically shouldn't matter. | 3248 | * (most normal use), this basically shouldn't matter. |
3242 | */ | 3249 | */ |
3243 | static int kswapd(void *p) | 3250 | static int kswapd(void *p) |
3244 | { | 3251 | { |
3245 | unsigned long order, new_order; | 3252 | unsigned long order, new_order; |
3246 | unsigned balanced_order; | 3253 | unsigned balanced_order; |
3247 | int classzone_idx, new_classzone_idx; | 3254 | int classzone_idx, new_classzone_idx; |
3248 | int balanced_classzone_idx; | 3255 | int balanced_classzone_idx; |
3249 | pg_data_t *pgdat = (pg_data_t*)p; | 3256 | pg_data_t *pgdat = (pg_data_t*)p; |
3250 | struct task_struct *tsk = current; | 3257 | struct task_struct *tsk = current; |
3251 | 3258 | ||
3252 | struct reclaim_state reclaim_state = { | 3259 | struct reclaim_state reclaim_state = { |
3253 | .reclaimed_slab = 0, | 3260 | .reclaimed_slab = 0, |
3254 | }; | 3261 | }; |
3255 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); | 3262 | const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id); |
3256 | 3263 | ||
3257 | lockdep_set_current_reclaim_state(GFP_KERNEL); | 3264 | lockdep_set_current_reclaim_state(GFP_KERNEL); |
3258 | 3265 | ||
3259 | if (!cpumask_empty(cpumask)) | 3266 | if (!cpumask_empty(cpumask)) |
3260 | set_cpus_allowed_ptr(tsk, cpumask); | 3267 | set_cpus_allowed_ptr(tsk, cpumask); |
3261 | current->reclaim_state = &reclaim_state; | 3268 | current->reclaim_state = &reclaim_state; |
3262 | 3269 | ||
3263 | /* | 3270 | /* |
3264 | * Tell the memory management that we're a "memory allocator", | 3271 | * Tell the memory management that we're a "memory allocator", |
3265 | * and that if we need more memory we should get access to it | 3272 | * and that if we need more memory we should get access to it |
3266 | * regardless (see "__alloc_pages()"). "kswapd" should | 3273 | * regardless (see "__alloc_pages()"). "kswapd" should |
3267 | * never get caught in the normal page freeing logic. | 3274 | * never get caught in the normal page freeing logic. |
3268 | * | 3275 | * |
3269 | * (Kswapd normally doesn't need memory anyway, but sometimes | 3276 | * (Kswapd normally doesn't need memory anyway, but sometimes |
3270 | * you need a small amount of memory in order to be able to | 3277 | * you need a small amount of memory in order to be able to |
3271 | * page out something else, and this flag essentially protects | 3278 | * page out something else, and this flag essentially protects |
3272 | * us from recursively trying to free more memory as we're | 3279 | * us from recursively trying to free more memory as we're |
3273 | * trying to free the first piece of memory in the first place). | 3280 | * trying to free the first piece of memory in the first place). |
3274 | */ | 3281 | */ |
3275 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; | 3282 | tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD; |
3276 | set_freezable(); | 3283 | set_freezable(); |
3277 | 3284 | ||
3278 | order = new_order = 0; | 3285 | order = new_order = 0; |
3279 | balanced_order = 0; | 3286 | balanced_order = 0; |
3280 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; | 3287 | classzone_idx = new_classzone_idx = pgdat->nr_zones - 1; |
3281 | balanced_classzone_idx = classzone_idx; | 3288 | balanced_classzone_idx = classzone_idx; |
3282 | for ( ; ; ) { | 3289 | for ( ; ; ) { |
3283 | bool ret; | 3290 | bool ret; |
3284 | 3291 | ||
3285 | /* | 3292 | /* |
3286 | * If the last balance_pgdat was unsuccessful it's unlikely a | 3293 | * If the last balance_pgdat was unsuccessful it's unlikely a |
3287 | * new request of a similar or harder type will succeed soon | 3294 | * new request of a similar or harder type will succeed soon |
3288 | * so consider going to sleep on the basis we reclaimed at | 3295 | * so consider going to sleep on the basis we reclaimed at |
3289 | */ | 3296 | */ |
3290 | if (balanced_classzone_idx >= new_classzone_idx && | 3297 | if (balanced_classzone_idx >= new_classzone_idx && |
3291 | balanced_order == new_order) { | 3298 | balanced_order == new_order) { |
3292 | new_order = pgdat->kswapd_max_order; | 3299 | new_order = pgdat->kswapd_max_order; |
3293 | new_classzone_idx = pgdat->classzone_idx; | 3300 | new_classzone_idx = pgdat->classzone_idx; |
3294 | pgdat->kswapd_max_order = 0; | 3301 | pgdat->kswapd_max_order = 0; |
3295 | pgdat->classzone_idx = pgdat->nr_zones - 1; | 3302 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
3296 | } | 3303 | } |
3297 | 3304 | ||
3298 | if (order < new_order || classzone_idx > new_classzone_idx) { | 3305 | if (order < new_order || classzone_idx > new_classzone_idx) { |
3299 | /* | 3306 | /* |
3300 | * Don't sleep if someone wants a larger 'order' | 3307 | * Don't sleep if someone wants a larger 'order' |
3301 | * allocation or has tigher zone constraints | 3308 | * allocation or has tigher zone constraints |
3302 | */ | 3309 | */ |
3303 | order = new_order; | 3310 | order = new_order; |
3304 | classzone_idx = new_classzone_idx; | 3311 | classzone_idx = new_classzone_idx; |
3305 | } else { | 3312 | } else { |
3306 | kswapd_try_to_sleep(pgdat, balanced_order, | 3313 | kswapd_try_to_sleep(pgdat, balanced_order, |
3307 | balanced_classzone_idx); | 3314 | balanced_classzone_idx); |
3308 | order = pgdat->kswapd_max_order; | 3315 | order = pgdat->kswapd_max_order; |
3309 | classzone_idx = pgdat->classzone_idx; | 3316 | classzone_idx = pgdat->classzone_idx; |
3310 | new_order = order; | 3317 | new_order = order; |
3311 | new_classzone_idx = classzone_idx; | 3318 | new_classzone_idx = classzone_idx; |
3312 | pgdat->kswapd_max_order = 0; | 3319 | pgdat->kswapd_max_order = 0; |
3313 | pgdat->classzone_idx = pgdat->nr_zones - 1; | 3320 | pgdat->classzone_idx = pgdat->nr_zones - 1; |
3314 | } | 3321 | } |
3315 | 3322 | ||
3316 | ret = try_to_freeze(); | 3323 | ret = try_to_freeze(); |
3317 | if (kthread_should_stop()) | 3324 | if (kthread_should_stop()) |
3318 | break; | 3325 | break; |
3319 | 3326 | ||
3320 | /* | 3327 | /* |
3321 | * We can speed up thawing tasks if we don't call balance_pgdat | 3328 | * We can speed up thawing tasks if we don't call balance_pgdat |
3322 | * after returning from the refrigerator | 3329 | * after returning from the refrigerator |
3323 | */ | 3330 | */ |
3324 | if (!ret) { | 3331 | if (!ret) { |
3325 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); | 3332 | trace_mm_vmscan_kswapd_wake(pgdat->node_id, order); |
3326 | balanced_classzone_idx = classzone_idx; | 3333 | balanced_classzone_idx = classzone_idx; |
3327 | balanced_order = balance_pgdat(pgdat, order, | 3334 | balanced_order = balance_pgdat(pgdat, order, |
3328 | &balanced_classzone_idx); | 3335 | &balanced_classzone_idx); |
3329 | } | 3336 | } |
3330 | } | 3337 | } |
3331 | 3338 | ||
3332 | tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD); | 3339 | tsk->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD); |
3333 | current->reclaim_state = NULL; | 3340 | current->reclaim_state = NULL; |
3334 | lockdep_clear_current_reclaim_state(); | 3341 | lockdep_clear_current_reclaim_state(); |
3335 | 3342 | ||
3336 | return 0; | 3343 | return 0; |
3337 | } | 3344 | } |
3338 | 3345 | ||
3339 | /* | 3346 | /* |
3340 | * A zone is low on free memory, so wake its kswapd task to service it. | 3347 | * A zone is low on free memory, so wake its kswapd task to service it. |
3341 | */ | 3348 | */ |
3342 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx) | 3349 | void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx) |
3343 | { | 3350 | { |
3344 | pg_data_t *pgdat; | 3351 | pg_data_t *pgdat; |
3345 | 3352 | ||
3346 | if (!populated_zone(zone)) | 3353 | if (!populated_zone(zone)) |
3347 | return; | 3354 | return; |
3348 | 3355 | ||
3349 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) | 3356 | if (!cpuset_zone_allowed_hardwall(zone, GFP_KERNEL)) |
3350 | return; | 3357 | return; |
3351 | pgdat = zone->zone_pgdat; | 3358 | pgdat = zone->zone_pgdat; |
3352 | if (pgdat->kswapd_max_order < order) { | 3359 | if (pgdat->kswapd_max_order < order) { |
3353 | pgdat->kswapd_max_order = order; | 3360 | pgdat->kswapd_max_order = order; |
3354 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); | 3361 | pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx); |
3355 | } | 3362 | } |
3356 | if (!waitqueue_active(&pgdat->kswapd_wait)) | 3363 | if (!waitqueue_active(&pgdat->kswapd_wait)) |
3357 | return; | 3364 | return; |
3358 | if (zone_balanced(zone, order, 0, 0)) | 3365 | if (zone_balanced(zone, order, 0, 0)) |
3359 | return; | 3366 | return; |
3360 | 3367 | ||
3361 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); | 3368 | trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order); |
3362 | wake_up_interruptible(&pgdat->kswapd_wait); | 3369 | wake_up_interruptible(&pgdat->kswapd_wait); |
3363 | } | 3370 | } |
3364 | 3371 | ||
3365 | #ifdef CONFIG_HIBERNATION | 3372 | #ifdef CONFIG_HIBERNATION |
3366 | /* | 3373 | /* |
3367 | * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of | 3374 | * Try to free `nr_to_reclaim' of memory, system-wide, and return the number of |
3368 | * freed pages. | 3375 | * freed pages. |
3369 | * | 3376 | * |
3370 | * Rather than trying to age LRUs the aim is to preserve the overall | 3377 | * Rather than trying to age LRUs the aim is to preserve the overall |
3371 | * LRU order by reclaiming preferentially | 3378 | * LRU order by reclaiming preferentially |
3372 | * inactive > active > active referenced > active mapped | 3379 | * inactive > active > active referenced > active mapped |
3373 | */ | 3380 | */ |
3374 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) | 3381 | unsigned long shrink_all_memory(unsigned long nr_to_reclaim) |
3375 | { | 3382 | { |
3376 | struct reclaim_state reclaim_state; | 3383 | struct reclaim_state reclaim_state; |
3377 | struct scan_control sc = { | 3384 | struct scan_control sc = { |
3378 | .gfp_mask = GFP_HIGHUSER_MOVABLE, | 3385 | .gfp_mask = GFP_HIGHUSER_MOVABLE, |
3379 | .may_swap = 1, | 3386 | .may_swap = 1, |
3380 | .may_unmap = 1, | 3387 | .may_unmap = 1, |
3381 | .may_writepage = 1, | 3388 | .may_writepage = 1, |
3382 | .nr_to_reclaim = nr_to_reclaim, | 3389 | .nr_to_reclaim = nr_to_reclaim, |
3383 | .hibernation_mode = 1, | 3390 | .hibernation_mode = 1, |
3384 | .order = 0, | 3391 | .order = 0, |
3385 | .priority = DEF_PRIORITY, | 3392 | .priority = DEF_PRIORITY, |
3386 | }; | 3393 | }; |
3387 | struct shrink_control shrink = { | 3394 | struct shrink_control shrink = { |
3388 | .gfp_mask = sc.gfp_mask, | 3395 | .gfp_mask = sc.gfp_mask, |
3389 | }; | 3396 | }; |
3390 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); | 3397 | struct zonelist *zonelist = node_zonelist(numa_node_id(), sc.gfp_mask); |
3391 | struct task_struct *p = current; | 3398 | struct task_struct *p = current; |
3392 | unsigned long nr_reclaimed; | 3399 | unsigned long nr_reclaimed; |
3393 | 3400 | ||
3394 | p->flags |= PF_MEMALLOC; | 3401 | p->flags |= PF_MEMALLOC; |
3395 | lockdep_set_current_reclaim_state(sc.gfp_mask); | 3402 | lockdep_set_current_reclaim_state(sc.gfp_mask); |
3396 | reclaim_state.reclaimed_slab = 0; | 3403 | reclaim_state.reclaimed_slab = 0; |
3397 | p->reclaim_state = &reclaim_state; | 3404 | p->reclaim_state = &reclaim_state; |
3398 | 3405 | ||
3399 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); | 3406 | nr_reclaimed = do_try_to_free_pages(zonelist, &sc, &shrink); |
3400 | 3407 | ||
3401 | p->reclaim_state = NULL; | 3408 | p->reclaim_state = NULL; |
3402 | lockdep_clear_current_reclaim_state(); | 3409 | lockdep_clear_current_reclaim_state(); |
3403 | p->flags &= ~PF_MEMALLOC; | 3410 | p->flags &= ~PF_MEMALLOC; |
3404 | 3411 | ||
3405 | return nr_reclaimed; | 3412 | return nr_reclaimed; |
3406 | } | 3413 | } |
3407 | #endif /* CONFIG_HIBERNATION */ | 3414 | #endif /* CONFIG_HIBERNATION */ |
3408 | 3415 | ||
3409 | /* It's optimal to keep kswapds on the same CPUs as their memory, but | 3416 | /* It's optimal to keep kswapds on the same CPUs as their memory, but |
3410 | not required for correctness. So if the last cpu in a node goes | 3417 | not required for correctness. So if the last cpu in a node goes |
3411 | away, we get changed to run anywhere: as the first one comes back, | 3418 | away, we get changed to run anywhere: as the first one comes back, |
3412 | restore their cpu bindings. */ | 3419 | restore their cpu bindings. */ |
3413 | static int cpu_callback(struct notifier_block *nfb, unsigned long action, | 3420 | static int cpu_callback(struct notifier_block *nfb, unsigned long action, |
3414 | void *hcpu) | 3421 | void *hcpu) |
3415 | { | 3422 | { |
3416 | int nid; | 3423 | int nid; |
3417 | 3424 | ||
3418 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { | 3425 | if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN) { |
3419 | for_each_node_state(nid, N_MEMORY) { | 3426 | for_each_node_state(nid, N_MEMORY) { |
3420 | pg_data_t *pgdat = NODE_DATA(nid); | 3427 | pg_data_t *pgdat = NODE_DATA(nid); |
3421 | const struct cpumask *mask; | 3428 | const struct cpumask *mask; |
3422 | 3429 | ||
3423 | mask = cpumask_of_node(pgdat->node_id); | 3430 | mask = cpumask_of_node(pgdat->node_id); |
3424 | 3431 | ||
3425 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) | 3432 | if (cpumask_any_and(cpu_online_mask, mask) < nr_cpu_ids) |
3426 | /* One of our CPUs online: restore mask */ | 3433 | /* One of our CPUs online: restore mask */ |
3427 | set_cpus_allowed_ptr(pgdat->kswapd, mask); | 3434 | set_cpus_allowed_ptr(pgdat->kswapd, mask); |
3428 | } | 3435 | } |
3429 | } | 3436 | } |
3430 | return NOTIFY_OK; | 3437 | return NOTIFY_OK; |
3431 | } | 3438 | } |
3432 | 3439 | ||
3433 | /* | 3440 | /* |
3434 | * This kswapd start function will be called by init and node-hot-add. | 3441 | * This kswapd start function will be called by init and node-hot-add. |
3435 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. | 3442 | * On node-hot-add, kswapd will moved to proper cpus if cpus are hot-added. |
3436 | */ | 3443 | */ |
3437 | int kswapd_run(int nid) | 3444 | int kswapd_run(int nid) |
3438 | { | 3445 | { |
3439 | pg_data_t *pgdat = NODE_DATA(nid); | 3446 | pg_data_t *pgdat = NODE_DATA(nid); |
3440 | int ret = 0; | 3447 | int ret = 0; |
3441 | 3448 | ||
3442 | if (pgdat->kswapd) | 3449 | if (pgdat->kswapd) |
3443 | return 0; | 3450 | return 0; |
3444 | 3451 | ||
3445 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); | 3452 | pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid); |
3446 | if (IS_ERR(pgdat->kswapd)) { | 3453 | if (IS_ERR(pgdat->kswapd)) { |
3447 | /* failure at boot is fatal */ | 3454 | /* failure at boot is fatal */ |
3448 | BUG_ON(system_state == SYSTEM_BOOTING); | 3455 | BUG_ON(system_state == SYSTEM_BOOTING); |
3449 | pr_err("Failed to start kswapd on node %d\n", nid); | 3456 | pr_err("Failed to start kswapd on node %d\n", nid); |
3450 | ret = PTR_ERR(pgdat->kswapd); | 3457 | ret = PTR_ERR(pgdat->kswapd); |
3451 | pgdat->kswapd = NULL; | 3458 | pgdat->kswapd = NULL; |
3452 | } | 3459 | } |
3453 | return ret; | 3460 | return ret; |
3454 | } | 3461 | } |
3455 | 3462 | ||
3456 | /* | 3463 | /* |
3457 | * Called by memory hotplug when all memory in a node is offlined. Caller must | 3464 | * Called by memory hotplug when all memory in a node is offlined. Caller must |
3458 | * hold lock_memory_hotplug(). | 3465 | * hold lock_memory_hotplug(). |
3459 | */ | 3466 | */ |
3460 | void kswapd_stop(int nid) | 3467 | void kswapd_stop(int nid) |
3461 | { | 3468 | { |
3462 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; | 3469 | struct task_struct *kswapd = NODE_DATA(nid)->kswapd; |
3463 | 3470 | ||
3464 | if (kswapd) { | 3471 | if (kswapd) { |
3465 | kthread_stop(kswapd); | 3472 | kthread_stop(kswapd); |
3466 | NODE_DATA(nid)->kswapd = NULL; | 3473 | NODE_DATA(nid)->kswapd = NULL; |
3467 | } | 3474 | } |
3468 | } | 3475 | } |
3469 | 3476 | ||
3470 | static int __init kswapd_init(void) | 3477 | static int __init kswapd_init(void) |
3471 | { | 3478 | { |
3472 | int nid; | 3479 | int nid; |
3473 | 3480 | ||
3474 | swap_setup(); | 3481 | swap_setup(); |
3475 | for_each_node_state(nid, N_MEMORY) | 3482 | for_each_node_state(nid, N_MEMORY) |
3476 | kswapd_run(nid); | 3483 | kswapd_run(nid); |
3477 | hotcpu_notifier(cpu_callback, 0); | 3484 | hotcpu_notifier(cpu_callback, 0); |
3478 | return 0; | 3485 | return 0; |
3479 | } | 3486 | } |
3480 | 3487 | ||
3481 | module_init(kswapd_init) | 3488 | module_init(kswapd_init) |
3482 | 3489 | ||
3483 | #ifdef CONFIG_NUMA | 3490 | #ifdef CONFIG_NUMA |
3484 | /* | 3491 | /* |
3485 | * Zone reclaim mode | 3492 | * Zone reclaim mode |
3486 | * | 3493 | * |
3487 | * If non-zero call zone_reclaim when the number of free pages falls below | 3494 | * If non-zero call zone_reclaim when the number of free pages falls below |
3488 | * the watermarks. | 3495 | * the watermarks. |
3489 | */ | 3496 | */ |
3490 | int zone_reclaim_mode __read_mostly; | 3497 | int zone_reclaim_mode __read_mostly; |
3491 | 3498 | ||
3492 | #define RECLAIM_OFF 0 | 3499 | #define RECLAIM_OFF 0 |
3493 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ | 3500 | #define RECLAIM_ZONE (1<<0) /* Run shrink_inactive_list on the zone */ |
3494 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ | 3501 | #define RECLAIM_WRITE (1<<1) /* Writeout pages during reclaim */ |
3495 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ | 3502 | #define RECLAIM_SWAP (1<<2) /* Swap pages out during reclaim */ |
3496 | 3503 | ||
3497 | /* | 3504 | /* |
3498 | * Priority for ZONE_RECLAIM. This determines the fraction of pages | 3505 | * Priority for ZONE_RECLAIM. This determines the fraction of pages |
3499 | * of a node considered for each zone_reclaim. 4 scans 1/16th of | 3506 | * of a node considered for each zone_reclaim. 4 scans 1/16th of |
3500 | * a zone. | 3507 | * a zone. |
3501 | */ | 3508 | */ |
3502 | #define ZONE_RECLAIM_PRIORITY 4 | 3509 | #define ZONE_RECLAIM_PRIORITY 4 |
3503 | 3510 | ||
3504 | /* | 3511 | /* |
3505 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to | 3512 | * Percentage of pages in a zone that must be unmapped for zone_reclaim to |
3506 | * occur. | 3513 | * occur. |
3507 | */ | 3514 | */ |
3508 | int sysctl_min_unmapped_ratio = 1; | 3515 | int sysctl_min_unmapped_ratio = 1; |
3509 | 3516 | ||
3510 | /* | 3517 | /* |
3511 | * If the number of slab pages in a zone grows beyond this percentage then | 3518 | * If the number of slab pages in a zone grows beyond this percentage then |
3512 | * slab reclaim needs to occur. | 3519 | * slab reclaim needs to occur. |
3513 | */ | 3520 | */ |
3514 | int sysctl_min_slab_ratio = 5; | 3521 | int sysctl_min_slab_ratio = 5; |
3515 | 3522 | ||
3516 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) | 3523 | static inline unsigned long zone_unmapped_file_pages(struct zone *zone) |
3517 | { | 3524 | { |
3518 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); | 3525 | unsigned long file_mapped = zone_page_state(zone, NR_FILE_MAPPED); |
3519 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + | 3526 | unsigned long file_lru = zone_page_state(zone, NR_INACTIVE_FILE) + |
3520 | zone_page_state(zone, NR_ACTIVE_FILE); | 3527 | zone_page_state(zone, NR_ACTIVE_FILE); |
3521 | 3528 | ||
3522 | /* | 3529 | /* |
3523 | * It's possible for there to be more file mapped pages than | 3530 | * It's possible for there to be more file mapped pages than |
3524 | * accounted for by the pages on the file LRU lists because | 3531 | * accounted for by the pages on the file LRU lists because |
3525 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED | 3532 | * tmpfs pages accounted for as ANON can also be FILE_MAPPED |
3526 | */ | 3533 | */ |
3527 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; | 3534 | return (file_lru > file_mapped) ? (file_lru - file_mapped) : 0; |
3528 | } | 3535 | } |
3529 | 3536 | ||
3530 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ | 3537 | /* Work out how many page cache pages we can reclaim in this reclaim_mode */ |
3531 | static long zone_pagecache_reclaimable(struct zone *zone) | 3538 | static long zone_pagecache_reclaimable(struct zone *zone) |
3532 | { | 3539 | { |
3533 | long nr_pagecache_reclaimable; | 3540 | long nr_pagecache_reclaimable; |
3534 | long delta = 0; | 3541 | long delta = 0; |
3535 | 3542 | ||
3536 | /* | 3543 | /* |
3537 | * If RECLAIM_SWAP is set, then all file pages are considered | 3544 | * If RECLAIM_SWAP is set, then all file pages are considered |
3538 | * potentially reclaimable. Otherwise, we have to worry about | 3545 | * potentially reclaimable. Otherwise, we have to worry about |
3539 | * pages like swapcache and zone_unmapped_file_pages() provides | 3546 | * pages like swapcache and zone_unmapped_file_pages() provides |
3540 | * a better estimate | 3547 | * a better estimate |
3541 | */ | 3548 | */ |
3542 | if (zone_reclaim_mode & RECLAIM_SWAP) | 3549 | if (zone_reclaim_mode & RECLAIM_SWAP) |
3543 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); | 3550 | nr_pagecache_reclaimable = zone_page_state(zone, NR_FILE_PAGES); |
3544 | else | 3551 | else |
3545 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); | 3552 | nr_pagecache_reclaimable = zone_unmapped_file_pages(zone); |
3546 | 3553 | ||
3547 | /* If we can't clean pages, remove dirty pages from consideration */ | 3554 | /* If we can't clean pages, remove dirty pages from consideration */ |
3548 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) | 3555 | if (!(zone_reclaim_mode & RECLAIM_WRITE)) |
3549 | delta += zone_page_state(zone, NR_FILE_DIRTY); | 3556 | delta += zone_page_state(zone, NR_FILE_DIRTY); |
3550 | 3557 | ||
3551 | /* Watch for any possible underflows due to delta */ | 3558 | /* Watch for any possible underflows due to delta */ |
3552 | if (unlikely(delta > nr_pagecache_reclaimable)) | 3559 | if (unlikely(delta > nr_pagecache_reclaimable)) |
3553 | delta = nr_pagecache_reclaimable; | 3560 | delta = nr_pagecache_reclaimable; |
3554 | 3561 | ||
3555 | return nr_pagecache_reclaimable - delta; | 3562 | return nr_pagecache_reclaimable - delta; |
3556 | } | 3563 | } |
3557 | 3564 | ||
3558 | /* | 3565 | /* |
3559 | * Try to free up some pages from this zone through reclaim. | 3566 | * Try to free up some pages from this zone through reclaim. |
3560 | */ | 3567 | */ |
3561 | static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | 3568 | static int __zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
3562 | { | 3569 | { |
3563 | /* Minimum pages needed in order to stay on node */ | 3570 | /* Minimum pages needed in order to stay on node */ |
3564 | const unsigned long nr_pages = 1 << order; | 3571 | const unsigned long nr_pages = 1 << order; |
3565 | struct task_struct *p = current; | 3572 | struct task_struct *p = current; |
3566 | struct reclaim_state reclaim_state; | 3573 | struct reclaim_state reclaim_state; |
3567 | struct scan_control sc = { | 3574 | struct scan_control sc = { |
3568 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), | 3575 | .may_writepage = !!(zone_reclaim_mode & RECLAIM_WRITE), |
3569 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP), | 3576 | .may_unmap = !!(zone_reclaim_mode & RECLAIM_SWAP), |
3570 | .may_swap = 1, | 3577 | .may_swap = 1, |
3571 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), | 3578 | .nr_to_reclaim = max(nr_pages, SWAP_CLUSTER_MAX), |
3572 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), | 3579 | .gfp_mask = (gfp_mask = memalloc_noio_flags(gfp_mask)), |
3573 | .order = order, | 3580 | .order = order, |
3574 | .priority = ZONE_RECLAIM_PRIORITY, | 3581 | .priority = ZONE_RECLAIM_PRIORITY, |
3575 | }; | 3582 | }; |
3576 | struct shrink_control shrink = { | 3583 | struct shrink_control shrink = { |
3577 | .gfp_mask = sc.gfp_mask, | 3584 | .gfp_mask = sc.gfp_mask, |
3578 | }; | 3585 | }; |
3579 | unsigned long nr_slab_pages0, nr_slab_pages1; | 3586 | unsigned long nr_slab_pages0, nr_slab_pages1; |
3580 | 3587 | ||
3581 | cond_resched(); | 3588 | cond_resched(); |
3582 | /* | 3589 | /* |
3583 | * We need to be able to allocate from the reserves for RECLAIM_SWAP | 3590 | * We need to be able to allocate from the reserves for RECLAIM_SWAP |
3584 | * and we also need to be able to write out pages for RECLAIM_WRITE | 3591 | * and we also need to be able to write out pages for RECLAIM_WRITE |
3585 | * and RECLAIM_SWAP. | 3592 | * and RECLAIM_SWAP. |
3586 | */ | 3593 | */ |
3587 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; | 3594 | p->flags |= PF_MEMALLOC | PF_SWAPWRITE; |
3588 | lockdep_set_current_reclaim_state(gfp_mask); | 3595 | lockdep_set_current_reclaim_state(gfp_mask); |
3589 | reclaim_state.reclaimed_slab = 0; | 3596 | reclaim_state.reclaimed_slab = 0; |
3590 | p->reclaim_state = &reclaim_state; | 3597 | p->reclaim_state = &reclaim_state; |
3591 | 3598 | ||
3592 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { | 3599 | if (zone_pagecache_reclaimable(zone) > zone->min_unmapped_pages) { |
3593 | /* | 3600 | /* |
3594 | * Free memory by calling shrink zone with increasing | 3601 | * Free memory by calling shrink zone with increasing |
3595 | * priorities until we have enough memory freed. | 3602 | * priorities until we have enough memory freed. |
3596 | */ | 3603 | */ |
3597 | do { | 3604 | do { |
3598 | shrink_zone(zone, &sc); | 3605 | shrink_zone(zone, &sc); |
3599 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); | 3606 | } while (sc.nr_reclaimed < nr_pages && --sc.priority >= 0); |
3600 | } | 3607 | } |
3601 | 3608 | ||
3602 | nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); | 3609 | nr_slab_pages0 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
3603 | if (nr_slab_pages0 > zone->min_slab_pages) { | 3610 | if (nr_slab_pages0 > zone->min_slab_pages) { |
3604 | /* | 3611 | /* |
3605 | * shrink_slab() does not currently allow us to determine how | 3612 | * shrink_slab() does not currently allow us to determine how |
3606 | * many pages were freed in this zone. So we take the current | 3613 | * many pages were freed in this zone. So we take the current |
3607 | * number of slab pages and shake the slab until it is reduced | 3614 | * number of slab pages and shake the slab until it is reduced |
3608 | * by the same nr_pages that we used for reclaiming unmapped | 3615 | * by the same nr_pages that we used for reclaiming unmapped |
3609 | * pages. | 3616 | * pages. |
3610 | */ | 3617 | */ |
3611 | nodes_clear(shrink.nodes_to_scan); | 3618 | nodes_clear(shrink.nodes_to_scan); |
3612 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); | 3619 | node_set(zone_to_nid(zone), shrink.nodes_to_scan); |
3613 | for (;;) { | 3620 | for (;;) { |
3614 | unsigned long lru_pages = zone_reclaimable_pages(zone); | 3621 | unsigned long lru_pages = zone_reclaimable_pages(zone); |
3615 | 3622 | ||
3616 | /* No reclaimable slab or very low memory pressure */ | 3623 | /* No reclaimable slab or very low memory pressure */ |
3617 | if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages)) | 3624 | if (!shrink_slab(&shrink, sc.nr_scanned, lru_pages)) |
3618 | break; | 3625 | break; |
3619 | 3626 | ||
3620 | /* Freed enough memory */ | 3627 | /* Freed enough memory */ |
3621 | nr_slab_pages1 = zone_page_state(zone, | 3628 | nr_slab_pages1 = zone_page_state(zone, |
3622 | NR_SLAB_RECLAIMABLE); | 3629 | NR_SLAB_RECLAIMABLE); |
3623 | if (nr_slab_pages1 + nr_pages <= nr_slab_pages0) | 3630 | if (nr_slab_pages1 + nr_pages <= nr_slab_pages0) |
3624 | break; | 3631 | break; |
3625 | } | 3632 | } |
3626 | 3633 | ||
3627 | /* | 3634 | /* |
3628 | * Update nr_reclaimed by the number of slab pages we | 3635 | * Update nr_reclaimed by the number of slab pages we |
3629 | * reclaimed from this zone. | 3636 | * reclaimed from this zone. |
3630 | */ | 3637 | */ |
3631 | nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); | 3638 | nr_slab_pages1 = zone_page_state(zone, NR_SLAB_RECLAIMABLE); |
3632 | if (nr_slab_pages1 < nr_slab_pages0) | 3639 | if (nr_slab_pages1 < nr_slab_pages0) |
3633 | sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1; | 3640 | sc.nr_reclaimed += nr_slab_pages0 - nr_slab_pages1; |
3634 | } | 3641 | } |
3635 | 3642 | ||
3636 | p->reclaim_state = NULL; | 3643 | p->reclaim_state = NULL; |
3637 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); | 3644 | current->flags &= ~(PF_MEMALLOC | PF_SWAPWRITE); |
3638 | lockdep_clear_current_reclaim_state(); | 3645 | lockdep_clear_current_reclaim_state(); |
3639 | return sc.nr_reclaimed >= nr_pages; | 3646 | return sc.nr_reclaimed >= nr_pages; |
3640 | } | 3647 | } |
3641 | 3648 | ||
3642 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) | 3649 | int zone_reclaim(struct zone *zone, gfp_t gfp_mask, unsigned int order) |
3643 | { | 3650 | { |
3644 | int node_id; | 3651 | int node_id; |
3645 | int ret; | 3652 | int ret; |
3646 | 3653 | ||
3647 | /* | 3654 | /* |
3648 | * Zone reclaim reclaims unmapped file backed pages and | 3655 | * Zone reclaim reclaims unmapped file backed pages and |
3649 | * slab pages if we are over the defined limits. | 3656 | * slab pages if we are over the defined limits. |
3650 | * | 3657 | * |
3651 | * A small portion of unmapped file backed pages is needed for | 3658 | * A small portion of unmapped file backed pages is needed for |
3652 | * file I/O otherwise pages read by file I/O will be immediately | 3659 | * file I/O otherwise pages read by file I/O will be immediately |
3653 | * thrown out if the zone is overallocated. So we do not reclaim | 3660 | * thrown out if the zone is overallocated. So we do not reclaim |
3654 | * if less than a specified percentage of the zone is used by | 3661 | * if less than a specified percentage of the zone is used by |
3655 | * unmapped file backed pages. | 3662 | * unmapped file backed pages. |
3656 | */ | 3663 | */ |
3657 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && | 3664 | if (zone_pagecache_reclaimable(zone) <= zone->min_unmapped_pages && |
3658 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) | 3665 | zone_page_state(zone, NR_SLAB_RECLAIMABLE) <= zone->min_slab_pages) |
3659 | return ZONE_RECLAIM_FULL; | 3666 | return ZONE_RECLAIM_FULL; |
3660 | 3667 | ||
3661 | if (!zone_reclaimable(zone)) | 3668 | if (!zone_reclaimable(zone)) |
3662 | return ZONE_RECLAIM_FULL; | 3669 | return ZONE_RECLAIM_FULL; |
3663 | 3670 | ||
3664 | /* | 3671 | /* |
3665 | * Do not scan if the allocation should not be delayed. | 3672 | * Do not scan if the allocation should not be delayed. |
3666 | */ | 3673 | */ |
3667 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) | 3674 | if (!(gfp_mask & __GFP_WAIT) || (current->flags & PF_MEMALLOC)) |
3668 | return ZONE_RECLAIM_NOSCAN; | 3675 | return ZONE_RECLAIM_NOSCAN; |
3669 | 3676 | ||
3670 | /* | 3677 | /* |
3671 | * Only run zone reclaim on the local zone or on zones that do not | 3678 | * Only run zone reclaim on the local zone or on zones that do not |
3672 | * have associated processors. This will favor the local processor | 3679 | * have associated processors. This will favor the local processor |
3673 | * over remote processors and spread off node memory allocations | 3680 | * over remote processors and spread off node memory allocations |
3674 | * as wide as possible. | 3681 | * as wide as possible. |
3675 | */ | 3682 | */ |
3676 | node_id = zone_to_nid(zone); | 3683 | node_id = zone_to_nid(zone); |
3677 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) | 3684 | if (node_state(node_id, N_CPU) && node_id != numa_node_id()) |
3678 | return ZONE_RECLAIM_NOSCAN; | 3685 | return ZONE_RECLAIM_NOSCAN; |
3679 | 3686 | ||
3680 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) | 3687 | if (zone_test_and_set_flag(zone, ZONE_RECLAIM_LOCKED)) |
3681 | return ZONE_RECLAIM_NOSCAN; | 3688 | return ZONE_RECLAIM_NOSCAN; |
3682 | 3689 | ||
3683 | ret = __zone_reclaim(zone, gfp_mask, order); | 3690 | ret = __zone_reclaim(zone, gfp_mask, order); |
3684 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); | 3691 | zone_clear_flag(zone, ZONE_RECLAIM_LOCKED); |
3685 | 3692 | ||
3686 | if (!ret) | 3693 | if (!ret) |
3687 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); | 3694 | count_vm_event(PGSCAN_ZONE_RECLAIM_FAILED); |
3688 | 3695 | ||
3689 | return ret; | 3696 | return ret; |
3690 | } | 3697 | } |
3691 | #endif | 3698 | #endif |
3692 | 3699 | ||
3693 | /* | 3700 | /* |
3694 | * page_evictable - test whether a page is evictable | 3701 | * page_evictable - test whether a page is evictable |
3695 | * @page: the page to test | 3702 | * @page: the page to test |
3696 | * | 3703 | * |
3697 | * Test whether page is evictable--i.e., should be placed on active/inactive | 3704 | * Test whether page is evictable--i.e., should be placed on active/inactive |
3698 | * lists vs unevictable list. | 3705 | * lists vs unevictable list. |
3699 | * | 3706 | * |
3700 | * Reasons page might not be evictable: | 3707 | * Reasons page might not be evictable: |
3701 | * (1) page's mapping marked unevictable | 3708 | * (1) page's mapping marked unevictable |
3702 | * (2) page is part of an mlocked VMA | 3709 | * (2) page is part of an mlocked VMA |
3703 | * | 3710 | * |
3704 | */ | 3711 | */ |
3705 | int page_evictable(struct page *page) | 3712 | int page_evictable(struct page *page) |
3706 | { | 3713 | { |
3707 | return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); | 3714 | return !mapping_unevictable(page_mapping(page)) && !PageMlocked(page); |
3708 | } | 3715 | } |
3709 | 3716 | ||
3710 | #ifdef CONFIG_SHMEM | 3717 | #ifdef CONFIG_SHMEM |
3711 | /** | 3718 | /** |
3712 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list | 3719 | * check_move_unevictable_pages - check pages for evictability and move to appropriate zone lru list |
3713 | * @pages: array of pages to check | 3720 | * @pages: array of pages to check |
3714 | * @nr_pages: number of pages to check | 3721 | * @nr_pages: number of pages to check |
3715 | * | 3722 | * |
3716 | * Checks pages for evictability and moves them to the appropriate lru list. | 3723 | * Checks pages for evictability and moves them to the appropriate lru list. |
3717 | * | 3724 | * |
3718 | * This function is only used for SysV IPC SHM_UNLOCK. | 3725 | * This function is only used for SysV IPC SHM_UNLOCK. |
3719 | */ | 3726 | */ |
3720 | void check_move_unevictable_pages(struct page **pages, int nr_pages) | 3727 | void check_move_unevictable_pages(struct page **pages, int nr_pages) |
3721 | { | 3728 | { |
3722 | struct lruvec *lruvec; | 3729 | struct lruvec *lruvec; |
3723 | struct zone *zone = NULL; | 3730 | struct zone *zone = NULL; |
3724 | int pgscanned = 0; | 3731 | int pgscanned = 0; |
3725 | int pgrescued = 0; | 3732 | int pgrescued = 0; |
3726 | int i; | 3733 | int i; |
3727 | 3734 | ||
3728 | for (i = 0; i < nr_pages; i++) { | 3735 | for (i = 0; i < nr_pages; i++) { |
3729 | struct page *page = pages[i]; | 3736 | struct page *page = pages[i]; |
3730 | struct zone *pagezone; | 3737 | struct zone *pagezone; |
3731 | 3738 | ||
3732 | pgscanned++; | 3739 | pgscanned++; |
3733 | pagezone = page_zone(page); | 3740 | pagezone = page_zone(page); |
3734 | if (pagezone != zone) { | 3741 | if (pagezone != zone) { |
3735 | if (zone) | 3742 | if (zone) |
3736 | spin_unlock_irq(&zone->lru_lock); | 3743 | spin_unlock_irq(&zone->lru_lock); |
3737 | zone = pagezone; | 3744 | zone = pagezone; |
3738 | spin_lock_irq(&zone->lru_lock); | 3745 | spin_lock_irq(&zone->lru_lock); |
3739 | } | 3746 | } |
3740 | lruvec = mem_cgroup_page_lruvec(page, zone); | 3747 | lruvec = mem_cgroup_page_lruvec(page, zone); |
3741 | 3748 | ||
3742 | if (!PageLRU(page) || !PageUnevictable(page)) | 3749 | if (!PageLRU(page) || !PageUnevictable(page)) |
3743 | continue; | 3750 | continue; |
3744 | 3751 | ||
3745 | if (page_evictable(page)) { | 3752 | if (page_evictable(page)) { |
3746 | enum lru_list lru = page_lru_base_type(page); | 3753 | enum lru_list lru = page_lru_base_type(page); |
3747 | 3754 | ||
3748 | VM_BUG_ON(PageActive(page)); | 3755 | VM_BUG_ON(PageActive(page)); |
3749 | ClearPageUnevictable(page); | 3756 | ClearPageUnevictable(page); |
3750 | del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE); | 3757 | del_page_from_lru_list(page, lruvec, LRU_UNEVICTABLE); |
3751 | add_page_to_lru_list(page, lruvec, lru); | 3758 | add_page_to_lru_list(page, lruvec, lru); |
3752 | pgrescued++; | 3759 | pgrescued++; |
3753 | } | 3760 | } |
3754 | } | 3761 | } |
3755 | 3762 | ||
3756 | if (zone) { | 3763 | if (zone) { |
3757 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); | 3764 | __count_vm_events(UNEVICTABLE_PGRESCUED, pgrescued); |
3758 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); | 3765 | __count_vm_events(UNEVICTABLE_PGSCANNED, pgscanned); |
3759 | spin_unlock_irq(&zone->lru_lock); | 3766 | spin_unlock_irq(&zone->lru_lock); |
3760 | } | 3767 | } |
3761 | } | 3768 | } |
3762 | #endif /* CONFIG_SHMEM */ | 3769 | #endif /* CONFIG_SHMEM */ |
3763 | 3770 | ||
3764 | static void warn_scan_unevictable_pages(void) | 3771 | static void warn_scan_unevictable_pages(void) |
3765 | { | 3772 | { |
3766 | printk_once(KERN_WARNING | 3773 | printk_once(KERN_WARNING |
3767 | "%s: The scan_unevictable_pages sysctl/node-interface has been " | 3774 | "%s: The scan_unevictable_pages sysctl/node-interface has been " |
3768 | "disabled for lack of a legitimate use case. If you have " | 3775 | "disabled for lack of a legitimate use case. If you have " |
3769 | "one, please send an email to linux-mm@kvack.org.\n", | 3776 | "one, please send an email to linux-mm@kvack.org.\n", |
3770 | current->comm); | 3777 | current->comm); |
3771 | } | 3778 | } |
3772 | 3779 | ||
3773 | /* | 3780 | /* |
3774 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of | 3781 | * scan_unevictable_pages [vm] sysctl handler. On demand re-scan of |
3775 | * all nodes' unevictable lists for evictable pages | 3782 | * all nodes' unevictable lists for evictable pages |
3776 | */ | 3783 | */ |
3777 | unsigned long scan_unevictable_pages; | 3784 | unsigned long scan_unevictable_pages; |
3778 | 3785 | ||
3779 | int scan_unevictable_handler(struct ctl_table *table, int write, | 3786 | int scan_unevictable_handler(struct ctl_table *table, int write, |
3780 | void __user *buffer, | 3787 | void __user *buffer, |
3781 | size_t *length, loff_t *ppos) | 3788 | size_t *length, loff_t *ppos) |
3782 | { | 3789 | { |
3783 | warn_scan_unevictable_pages(); | 3790 | warn_scan_unevictable_pages(); |
3784 | proc_doulongvec_minmax(table, write, buffer, length, ppos); | 3791 | proc_doulongvec_minmax(table, write, buffer, length, ppos); |
3785 | scan_unevictable_pages = 0; | 3792 | scan_unevictable_pages = 0; |
3786 | return 0; | 3793 | return 0; |
3787 | } | 3794 | } |
3788 | 3795 | ||
3789 | #ifdef CONFIG_NUMA | 3796 | #ifdef CONFIG_NUMA |
3790 | /* | 3797 | /* |
3791 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of | 3798 | * per node 'scan_unevictable_pages' attribute. On demand re-scan of |
3792 | * a specified node's per zone unevictable lists for evictable pages. | 3799 | * a specified node's per zone unevictable lists for evictable pages. |
3793 | */ | 3800 | */ |
3794 | 3801 | ||
3795 | static ssize_t read_scan_unevictable_node(struct device *dev, | 3802 | static ssize_t read_scan_unevictable_node(struct device *dev, |
3796 | struct device_attribute *attr, | 3803 | struct device_attribute *attr, |
3797 | char *buf) | 3804 | char *buf) |
3798 | { | 3805 | { |
3799 | warn_scan_unevictable_pages(); | 3806 | warn_scan_unevictable_pages(); |
3800 | return sprintf(buf, "0\n"); /* always zero; should fit... */ | 3807 | return sprintf(buf, "0\n"); /* always zero; should fit... */ |
3801 | } | 3808 | } |
3802 | 3809 | ||
3803 | static ssize_t write_scan_unevictable_node(struct device *dev, | 3810 | static ssize_t write_scan_unevictable_node(struct device *dev, |
3804 | struct device_attribute *attr, | 3811 | struct device_attribute *attr, |
3805 | const char *buf, size_t count) | 3812 | const char *buf, size_t count) |
3806 | { | 3813 | { |
3807 | warn_scan_unevictable_pages(); | 3814 | warn_scan_unevictable_pages(); |
3808 | return 1; | 3815 | return 1; |
3809 | } | 3816 | } |
3810 | 3817 | ||
3811 | 3818 | ||
3812 | static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, | 3819 | static DEVICE_ATTR(scan_unevictable_pages, S_IRUGO | S_IWUSR, |
3813 | read_scan_unevictable_node, | 3820 | read_scan_unevictable_node, |
3814 | write_scan_unevictable_node); | 3821 | write_scan_unevictable_node); |
3815 | 3822 | ||
3816 | int scan_unevictable_register_node(struct node *node) | 3823 | int scan_unevictable_register_node(struct node *node) |
3817 | { | 3824 | { |
3818 | return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages); | 3825 | return device_create_file(&node->dev, &dev_attr_scan_unevictable_pages); |
3819 | } | 3826 | } |
3820 | 3827 | ||
3821 | void scan_unevictable_unregister_node(struct node *node) | 3828 | void scan_unevictable_unregister_node(struct node *node) |