Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
8cdea7c0 BS |
13 | * This program is free software; you can redistribute it and/or modify |
14 | * it under the terms of the GNU General Public License as published by | |
15 | * the Free Software Foundation; either version 2 of the License, or | |
16 | * (at your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, | |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | * GNU General Public License for more details. | |
22 | */ | |
23 | ||
24 | #include <linux/res_counter.h> | |
25 | #include <linux/memcontrol.h> | |
26 | #include <linux/cgroup.h> | |
78fb7466 | 27 | #include <linux/mm.h> |
4ffef5fe | 28 | #include <linux/hugetlb.h> |
d13d1443 | 29 | #include <linux/pagemap.h> |
d52aa412 | 30 | #include <linux/smp.h> |
8a9f3ccd | 31 | #include <linux/page-flags.h> |
66e1707b | 32 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
33 | #include <linux/bit_spinlock.h> |
34 | #include <linux/rcupdate.h> | |
e222432b | 35 | #include <linux/limits.h> |
8c7c6e34 | 36 | #include <linux/mutex.h> |
f64c3f54 | 37 | #include <linux/rbtree.h> |
072441e2 | 38 | #include <linux/shmem_fs.h> |
b6ac57d5 | 39 | #include <linux/slab.h> |
66e1707b | 40 | #include <linux/swap.h> |
02491447 | 41 | #include <linux/swapops.h> |
66e1707b | 42 | #include <linux/spinlock.h> |
2e72b634 KS |
43 | #include <linux/eventfd.h> |
44 | #include <linux/sort.h> | |
66e1707b | 45 | #include <linux/fs.h> |
d2ceb9b7 | 46 | #include <linux/seq_file.h> |
33327948 | 47 | #include <linux/vmalloc.h> |
b69408e8 | 48 | #include <linux/mm_inline.h> |
52d4b9ac | 49 | #include <linux/page_cgroup.h> |
cdec2e42 | 50 | #include <linux/cpu.h> |
158e0a2d | 51 | #include <linux/oom.h> |
08e552c6 | 52 | #include "internal.h" |
8cdea7c0 | 53 | |
8697d331 BS |
54 | #include <asm/uaccess.h> |
55 | ||
cc8e970c KM |
56 | #include <trace/events/vmscan.h> |
57 | ||
a181b0e8 | 58 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 59 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 60 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 61 | |
c077719b | 62 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 63 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 64 | int do_swap_account __read_mostly; |
a42c390c MH |
65 | |
66 | /* for remember boot option*/ | |
67 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED | |
68 | static int really_do_swap_account __initdata = 1; | |
69 | #else | |
70 | static int really_do_swap_account __initdata = 0; | |
71 | #endif | |
72 | ||
c077719b KH |
73 | #else |
74 | #define do_swap_account (0) | |
75 | #endif | |
76 | ||
77 | ||
d52aa412 KH |
78 | /* |
79 | * Statistics for memory cgroup. | |
80 | */ | |
81 | enum mem_cgroup_stat_index { | |
82 | /* | |
83 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
84 | */ | |
85 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 86 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 87 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
0c3e73e8 | 88 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
711d3d2c | 89 | MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ |
32047e2a | 90 | MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */ |
d52aa412 KH |
91 | MEM_CGROUP_STAT_NSTATS, |
92 | }; | |
93 | ||
e9f8974f JW |
94 | enum mem_cgroup_events_index { |
95 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
96 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
97 | MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */ | |
456f998e YH |
98 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
99 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
100 | MEM_CGROUP_EVENTS_NSTATS, |
101 | }; | |
7a159cc9 JW |
102 | /* |
103 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
104 | * it will be incremated by the number of pages. This counter is used for | |
105 | * for trigger some periodic events. This is straightforward and better | |
106 | * than using jiffies etc. to handle periodic memcg event. | |
107 | */ | |
108 | enum mem_cgroup_events_target { | |
109 | MEM_CGROUP_TARGET_THRESH, | |
110 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
111 | MEM_CGROUP_NTARGETS, | |
112 | }; | |
113 | #define THRESHOLDS_EVENTS_TARGET (128) | |
114 | #define SOFTLIMIT_EVENTS_TARGET (1024) | |
e9f8974f | 115 | |
d52aa412 | 116 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 117 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 118 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
7a159cc9 | 119 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
120 | }; |
121 | ||
6d12e2d8 KH |
122 | /* |
123 | * per-zone information in memory controller. | |
124 | */ | |
6d12e2d8 | 125 | struct mem_cgroup_per_zone { |
072c56c1 KH |
126 | /* |
127 | * spin_lock to protect the per cgroup LRU | |
128 | */ | |
b69408e8 CL |
129 | struct list_head lists[NR_LRU_LISTS]; |
130 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
131 | |
132 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
133 | struct rb_node tree_node; /* RB tree node */ |
134 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
135 | /* the soft limit is exceeded*/ | |
136 | bool on_tree; | |
4e416953 BS |
137 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
138 | /* use container_of */ | |
6d12e2d8 KH |
139 | }; |
140 | /* Macro for accessing counter */ | |
141 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
142 | ||
143 | struct mem_cgroup_per_node { | |
144 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
145 | }; | |
146 | ||
147 | struct mem_cgroup_lru_info { | |
148 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
149 | }; | |
150 | ||
f64c3f54 BS |
151 | /* |
152 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
153 | * their hierarchy representation | |
154 | */ | |
155 | ||
156 | struct mem_cgroup_tree_per_zone { | |
157 | struct rb_root rb_root; | |
158 | spinlock_t lock; | |
159 | }; | |
160 | ||
161 | struct mem_cgroup_tree_per_node { | |
162 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
163 | }; | |
164 | ||
165 | struct mem_cgroup_tree { | |
166 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
167 | }; | |
168 | ||
169 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
170 | ||
2e72b634 KS |
171 | struct mem_cgroup_threshold { |
172 | struct eventfd_ctx *eventfd; | |
173 | u64 threshold; | |
174 | }; | |
175 | ||
9490ff27 | 176 | /* For threshold */ |
2e72b634 KS |
177 | struct mem_cgroup_threshold_ary { |
178 | /* An array index points to threshold just below usage. */ | |
5407a562 | 179 | int current_threshold; |
2e72b634 KS |
180 | /* Size of entries[] */ |
181 | unsigned int size; | |
182 | /* Array of thresholds */ | |
183 | struct mem_cgroup_threshold entries[0]; | |
184 | }; | |
2c488db2 KS |
185 | |
186 | struct mem_cgroup_thresholds { | |
187 | /* Primary thresholds array */ | |
188 | struct mem_cgroup_threshold_ary *primary; | |
189 | /* | |
190 | * Spare threshold array. | |
191 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
192 | * It must be able to store at least primary->size - 1 entries. | |
193 | */ | |
194 | struct mem_cgroup_threshold_ary *spare; | |
195 | }; | |
196 | ||
9490ff27 KH |
197 | /* for OOM */ |
198 | struct mem_cgroup_eventfd_list { | |
199 | struct list_head list; | |
200 | struct eventfd_ctx *eventfd; | |
201 | }; | |
2e72b634 | 202 | |
2e72b634 | 203 | static void mem_cgroup_threshold(struct mem_cgroup *mem); |
9490ff27 | 204 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem); |
2e72b634 | 205 | |
8cdea7c0 BS |
206 | /* |
207 | * The memory controller data structure. The memory controller controls both | |
208 | * page cache and RSS per cgroup. We would eventually like to provide | |
209 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
210 | * to help the administrator determine what knobs to tune. | |
211 | * | |
212 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
213 | * we hit the water mark. May be even add a low water mark, such that |
214 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
215 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
216 | */ |
217 | struct mem_cgroup { | |
218 | struct cgroup_subsys_state css; | |
219 | /* | |
220 | * the counter to account for memory usage | |
221 | */ | |
222 | struct res_counter res; | |
8c7c6e34 KH |
223 | /* |
224 | * the counter to account for mem+swap usage. | |
225 | */ | |
226 | struct res_counter memsw; | |
78fb7466 PE |
227 | /* |
228 | * Per cgroup active and inactive list, similar to the | |
229 | * per zone LRU lists. | |
78fb7466 | 230 | */ |
6d12e2d8 | 231 | struct mem_cgroup_lru_info info; |
6d61ef40 | 232 | /* |
af901ca1 | 233 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 234 | * reclaimed from. |
6d61ef40 | 235 | */ |
04046e1a | 236 | int last_scanned_child; |
889976db YH |
237 | int last_scanned_node; |
238 | #if MAX_NUMNODES > 1 | |
239 | nodemask_t scan_nodes; | |
240 | unsigned long next_scan_node_update; | |
241 | #endif | |
18f59ea7 BS |
242 | /* |
243 | * Should the accounting and control be hierarchical, per subtree? | |
244 | */ | |
245 | bool use_hierarchy; | |
867578cb | 246 | atomic_t oom_lock; |
8c7c6e34 | 247 | atomic_t refcnt; |
14797e23 | 248 | |
a7885eb8 | 249 | unsigned int swappiness; |
3c11ecf4 KH |
250 | /* OOM-Killer disable */ |
251 | int oom_kill_disable; | |
a7885eb8 | 252 | |
22a668d7 KH |
253 | /* set when res.limit == memsw.limit */ |
254 | bool memsw_is_minimum; | |
255 | ||
2e72b634 KS |
256 | /* protect arrays of thresholds */ |
257 | struct mutex thresholds_lock; | |
258 | ||
259 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 260 | struct mem_cgroup_thresholds thresholds; |
907860ed | 261 | |
2e72b634 | 262 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 263 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 264 | |
9490ff27 KH |
265 | /* For oom notifier event fd */ |
266 | struct list_head oom_notify; | |
267 | ||
7dc74be0 DN |
268 | /* |
269 | * Should we move charges of a task when a task is moved into this | |
270 | * mem_cgroup ? And what type of charges should we move ? | |
271 | */ | |
272 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 273 | /* |
c62b1a3b | 274 | * percpu counter. |
d52aa412 | 275 | */ |
c62b1a3b | 276 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
277 | /* |
278 | * used when a cpu is offlined or other synchronizations | |
279 | * See mem_cgroup_read_stat(). | |
280 | */ | |
281 | struct mem_cgroup_stat_cpu nocpu_base; | |
282 | spinlock_t pcp_counter_lock; | |
8cdea7c0 BS |
283 | }; |
284 | ||
7dc74be0 DN |
285 | /* Stuffs for move charges at task migration. */ |
286 | /* | |
287 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
288 | * left-shifted bitmap of these types. | |
289 | */ | |
290 | enum move_type { | |
4ffef5fe | 291 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 292 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
293 | NR_MOVE_TYPE, |
294 | }; | |
295 | ||
4ffef5fe DN |
296 | /* "mc" and its members are protected by cgroup_mutex */ |
297 | static struct move_charge_struct { | |
b1dd693e | 298 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
299 | struct mem_cgroup *from; |
300 | struct mem_cgroup *to; | |
301 | unsigned long precharge; | |
854ffa8d | 302 | unsigned long moved_charge; |
483c30b5 | 303 | unsigned long moved_swap; |
8033b97c DN |
304 | struct task_struct *moving_task; /* a task moving charges */ |
305 | wait_queue_head_t waitq; /* a waitq for other context */ | |
306 | } mc = { | |
2bd9bb20 | 307 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
308 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
309 | }; | |
4ffef5fe | 310 | |
90254a65 DN |
311 | static bool move_anon(void) |
312 | { | |
313 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
314 | &mc.to->move_charge_at_immigrate); | |
315 | } | |
316 | ||
87946a72 DN |
317 | static bool move_file(void) |
318 | { | |
319 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
320 | &mc.to->move_charge_at_immigrate); | |
321 | } | |
322 | ||
4e416953 BS |
323 | /* |
324 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
325 | * limit reclaim to prevent infinite loops, if they ever occur. | |
326 | */ | |
327 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
328 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
329 | ||
217bc319 KH |
330 | enum charge_type { |
331 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
332 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 333 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 334 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 335 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 336 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
337 | NR_CHARGE_TYPE, |
338 | }; | |
339 | ||
8c7c6e34 KH |
340 | /* for encoding cft->private value on file */ |
341 | #define _MEM (0) | |
342 | #define _MEMSWAP (1) | |
9490ff27 | 343 | #define _OOM_TYPE (2) |
8c7c6e34 KH |
344 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
345 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
346 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
347 | /* Used for OOM nofiier */ |
348 | #define OOM_CONTROL (0) | |
8c7c6e34 | 349 | |
75822b44 BS |
350 | /* |
351 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
352 | */ | |
353 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
354 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
355 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
356 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
357 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
358 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 359 | |
8c7c6e34 KH |
360 | static void mem_cgroup_get(struct mem_cgroup *mem); |
361 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 362 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
26fe6168 | 363 | static void drain_all_stock_async(struct mem_cgroup *mem); |
8c7c6e34 | 364 | |
f64c3f54 BS |
365 | static struct mem_cgroup_per_zone * |
366 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
367 | { | |
368 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
369 | } | |
370 | ||
d324236b WF |
371 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
372 | { | |
373 | return &mem->css; | |
374 | } | |
375 | ||
f64c3f54 | 376 | static struct mem_cgroup_per_zone * |
97a6c37b | 377 | page_cgroup_zoneinfo(struct mem_cgroup *mem, struct page *page) |
f64c3f54 | 378 | { |
97a6c37b JW |
379 | int nid = page_to_nid(page); |
380 | int zid = page_zonenum(page); | |
f64c3f54 | 381 | |
f64c3f54 BS |
382 | return mem_cgroup_zoneinfo(mem, nid, zid); |
383 | } | |
384 | ||
385 | static struct mem_cgroup_tree_per_zone * | |
386 | soft_limit_tree_node_zone(int nid, int zid) | |
387 | { | |
388 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
389 | } | |
390 | ||
391 | static struct mem_cgroup_tree_per_zone * | |
392 | soft_limit_tree_from_page(struct page *page) | |
393 | { | |
394 | int nid = page_to_nid(page); | |
395 | int zid = page_zonenum(page); | |
396 | ||
397 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
398 | } | |
399 | ||
400 | static void | |
4e416953 | 401 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 402 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
403 | struct mem_cgroup_tree_per_zone *mctz, |
404 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
405 | { |
406 | struct rb_node **p = &mctz->rb_root.rb_node; | |
407 | struct rb_node *parent = NULL; | |
408 | struct mem_cgroup_per_zone *mz_node; | |
409 | ||
410 | if (mz->on_tree) | |
411 | return; | |
412 | ||
ef8745c1 KH |
413 | mz->usage_in_excess = new_usage_in_excess; |
414 | if (!mz->usage_in_excess) | |
415 | return; | |
f64c3f54 BS |
416 | while (*p) { |
417 | parent = *p; | |
418 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
419 | tree_node); | |
420 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
421 | p = &(*p)->rb_left; | |
422 | /* | |
423 | * We can't avoid mem cgroups that are over their soft | |
424 | * limit by the same amount | |
425 | */ | |
426 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
427 | p = &(*p)->rb_right; | |
428 | } | |
429 | rb_link_node(&mz->tree_node, parent, p); | |
430 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
431 | mz->on_tree = true; | |
4e416953 BS |
432 | } |
433 | ||
434 | static void | |
435 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
436 | struct mem_cgroup_per_zone *mz, | |
437 | struct mem_cgroup_tree_per_zone *mctz) | |
438 | { | |
439 | if (!mz->on_tree) | |
440 | return; | |
441 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
442 | mz->on_tree = false; | |
443 | } | |
444 | ||
f64c3f54 BS |
445 | static void |
446 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
447 | struct mem_cgroup_per_zone *mz, | |
448 | struct mem_cgroup_tree_per_zone *mctz) | |
449 | { | |
450 | spin_lock(&mctz->lock); | |
4e416953 | 451 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
452 | spin_unlock(&mctz->lock); |
453 | } | |
454 | ||
f64c3f54 BS |
455 | |
456 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
457 | { | |
ef8745c1 | 458 | unsigned long long excess; |
f64c3f54 BS |
459 | struct mem_cgroup_per_zone *mz; |
460 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
461 | int nid = page_to_nid(page); |
462 | int zid = page_zonenum(page); | |
f64c3f54 BS |
463 | mctz = soft_limit_tree_from_page(page); |
464 | ||
465 | /* | |
4e649152 KH |
466 | * Necessary to update all ancestors when hierarchy is used. |
467 | * because their event counter is not touched. | |
f64c3f54 | 468 | */ |
4e649152 KH |
469 | for (; mem; mem = parent_mem_cgroup(mem)) { |
470 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 471 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
472 | /* |
473 | * We have to update the tree if mz is on RB-tree or | |
474 | * mem is over its softlimit. | |
475 | */ | |
ef8745c1 | 476 | if (excess || mz->on_tree) { |
4e649152 KH |
477 | spin_lock(&mctz->lock); |
478 | /* if on-tree, remove it */ | |
479 | if (mz->on_tree) | |
480 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
481 | /* | |
ef8745c1 KH |
482 | * Insert again. mz->usage_in_excess will be updated. |
483 | * If excess is 0, no tree ops. | |
4e649152 | 484 | */ |
ef8745c1 | 485 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
486 | spin_unlock(&mctz->lock); |
487 | } | |
f64c3f54 BS |
488 | } |
489 | } | |
490 | ||
491 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
492 | { | |
493 | int node, zone; | |
494 | struct mem_cgroup_per_zone *mz; | |
495 | struct mem_cgroup_tree_per_zone *mctz; | |
496 | ||
497 | for_each_node_state(node, N_POSSIBLE) { | |
498 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
499 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
500 | mctz = soft_limit_tree_node_zone(node, zone); | |
501 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
502 | } | |
503 | } | |
504 | } | |
505 | ||
4e416953 BS |
506 | static struct mem_cgroup_per_zone * |
507 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
508 | { | |
509 | struct rb_node *rightmost = NULL; | |
26251eaf | 510 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
511 | |
512 | retry: | |
26251eaf | 513 | mz = NULL; |
4e416953 BS |
514 | rightmost = rb_last(&mctz->rb_root); |
515 | if (!rightmost) | |
516 | goto done; /* Nothing to reclaim from */ | |
517 | ||
518 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
519 | /* | |
520 | * Remove the node now but someone else can add it back, | |
521 | * we will to add it back at the end of reclaim to its correct | |
522 | * position in the tree. | |
523 | */ | |
524 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
525 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
526 | !css_tryget(&mz->mem->css)) | |
527 | goto retry; | |
528 | done: | |
529 | return mz; | |
530 | } | |
531 | ||
532 | static struct mem_cgroup_per_zone * | |
533 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
534 | { | |
535 | struct mem_cgroup_per_zone *mz; | |
536 | ||
537 | spin_lock(&mctz->lock); | |
538 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
539 | spin_unlock(&mctz->lock); | |
540 | return mz; | |
541 | } | |
542 | ||
711d3d2c KH |
543 | /* |
544 | * Implementation Note: reading percpu statistics for memcg. | |
545 | * | |
546 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
547 | * synchronization to implement "quick" read. There are trade-off between | |
548 | * reading cost and precision of value. Then, we may have a chance to implement | |
549 | * a periodic synchronizion of counter in memcg's counter. | |
550 | * | |
551 | * But this _read() function is used for user interface now. The user accounts | |
552 | * memory usage by memory cgroup and he _always_ requires exact value because | |
553 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
554 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
555 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
556 | * | |
557 | * If there are kernel internal actions which can make use of some not-exact | |
558 | * value, and reading all cpu value can be performance bottleneck in some | |
559 | * common workload, threashold and synchonization as vmstat[] should be | |
560 | * implemented. | |
561 | */ | |
7a159cc9 JW |
562 | static long mem_cgroup_read_stat(struct mem_cgroup *mem, |
563 | enum mem_cgroup_stat_index idx) | |
c62b1a3b | 564 | { |
7a159cc9 | 565 | long val = 0; |
c62b1a3b | 566 | int cpu; |
c62b1a3b | 567 | |
711d3d2c KH |
568 | get_online_cpus(); |
569 | for_each_online_cpu(cpu) | |
c62b1a3b | 570 | val += per_cpu(mem->stat->count[idx], cpu); |
711d3d2c KH |
571 | #ifdef CONFIG_HOTPLUG_CPU |
572 | spin_lock(&mem->pcp_counter_lock); | |
573 | val += mem->nocpu_base.count[idx]; | |
574 | spin_unlock(&mem->pcp_counter_lock); | |
575 | #endif | |
576 | put_online_cpus(); | |
c62b1a3b KH |
577 | return val; |
578 | } | |
579 | ||
7a159cc9 | 580 | static long mem_cgroup_local_usage(struct mem_cgroup *mem) |
c62b1a3b | 581 | { |
7a159cc9 | 582 | long ret; |
c62b1a3b KH |
583 | |
584 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); | |
585 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); | |
586 | return ret; | |
587 | } | |
588 | ||
0c3e73e8 BS |
589 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
590 | bool charge) | |
591 | { | |
592 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 593 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
594 | } |
595 | ||
456f998e YH |
596 | void mem_cgroup_pgfault(struct mem_cgroup *mem, int val) |
597 | { | |
598 | this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_PGFAULT], val); | |
599 | } | |
600 | ||
601 | void mem_cgroup_pgmajfault(struct mem_cgroup *mem, int val) | |
602 | { | |
603 | this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT], val); | |
604 | } | |
605 | ||
e9f8974f JW |
606 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *mem, |
607 | enum mem_cgroup_events_index idx) | |
608 | { | |
609 | unsigned long val = 0; | |
610 | int cpu; | |
611 | ||
612 | for_each_online_cpu(cpu) | |
613 | val += per_cpu(mem->stat->events[idx], cpu); | |
614 | #ifdef CONFIG_HOTPLUG_CPU | |
615 | spin_lock(&mem->pcp_counter_lock); | |
616 | val += mem->nocpu_base.events[idx]; | |
617 | spin_unlock(&mem->pcp_counter_lock); | |
618 | #endif | |
619 | return val; | |
620 | } | |
621 | ||
c05555b5 | 622 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
e401f176 | 623 | bool file, int nr_pages) |
d52aa412 | 624 | { |
c62b1a3b KH |
625 | preempt_disable(); |
626 | ||
e401f176 KH |
627 | if (file) |
628 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages); | |
d52aa412 | 629 | else |
e401f176 | 630 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages); |
55e462b0 | 631 | |
e401f176 KH |
632 | /* pagein of a big page is an event. So, ignore page size */ |
633 | if (nr_pages > 0) | |
e9f8974f | 634 | __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 635 | else { |
e9f8974f | 636 | __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
637 | nr_pages = -nr_pages; /* for event */ |
638 | } | |
e401f176 | 639 | |
e9f8974f | 640 | __this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
2e72b634 | 641 | |
c62b1a3b | 642 | preempt_enable(); |
6d12e2d8 KH |
643 | } |
644 | ||
889976db YH |
645 | static unsigned long |
646 | mem_cgroup_get_zonestat_node(struct mem_cgroup *mem, int nid, enum lru_list idx) | |
647 | { | |
648 | struct mem_cgroup_per_zone *mz; | |
649 | u64 total = 0; | |
650 | int zid; | |
651 | ||
652 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
653 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
654 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
655 | } | |
656 | return total; | |
657 | } | |
14067bb3 | 658 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 659 | enum lru_list idx) |
6d12e2d8 | 660 | { |
889976db | 661 | int nid; |
6d12e2d8 KH |
662 | u64 total = 0; |
663 | ||
664 | for_each_online_node(nid) | |
889976db | 665 | total += mem_cgroup_get_zonestat_node(mem, nid, idx); |
6d12e2d8 | 666 | return total; |
d52aa412 KH |
667 | } |
668 | ||
7a159cc9 JW |
669 | static bool __memcg_event_check(struct mem_cgroup *mem, int target) |
670 | { | |
671 | unsigned long val, next; | |
672 | ||
673 | val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]); | |
674 | next = this_cpu_read(mem->stat->targets[target]); | |
675 | /* from time_after() in jiffies.h */ | |
676 | return ((long)next - (long)val < 0); | |
677 | } | |
678 | ||
679 | static void __mem_cgroup_target_update(struct mem_cgroup *mem, int target) | |
d2265e6f | 680 | { |
7a159cc9 | 681 | unsigned long val, next; |
d2265e6f | 682 | |
e9f8974f | 683 | val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
d2265e6f | 684 | |
7a159cc9 JW |
685 | switch (target) { |
686 | case MEM_CGROUP_TARGET_THRESH: | |
687 | next = val + THRESHOLDS_EVENTS_TARGET; | |
688 | break; | |
689 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
690 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
691 | break; | |
692 | default: | |
693 | return; | |
694 | } | |
695 | ||
696 | this_cpu_write(mem->stat->targets[target], next); | |
d2265e6f KH |
697 | } |
698 | ||
699 | /* | |
700 | * Check events in order. | |
701 | * | |
702 | */ | |
703 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) | |
704 | { | |
705 | /* threshold event is triggered in finer grain than soft limit */ | |
7a159cc9 | 706 | if (unlikely(__memcg_event_check(mem, MEM_CGROUP_TARGET_THRESH))) { |
d2265e6f | 707 | mem_cgroup_threshold(mem); |
7a159cc9 JW |
708 | __mem_cgroup_target_update(mem, MEM_CGROUP_TARGET_THRESH); |
709 | if (unlikely(__memcg_event_check(mem, | |
710 | MEM_CGROUP_TARGET_SOFTLIMIT))){ | |
d2265e6f | 711 | mem_cgroup_update_tree(mem, page); |
7a159cc9 JW |
712 | __mem_cgroup_target_update(mem, |
713 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
714 | } | |
d2265e6f KH |
715 | } |
716 | } | |
717 | ||
d5b69e38 | 718 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
719 | { |
720 | return container_of(cgroup_subsys_state(cont, | |
721 | mem_cgroup_subsys_id), struct mem_cgroup, | |
722 | css); | |
723 | } | |
724 | ||
cf475ad2 | 725 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 726 | { |
31a78f23 BS |
727 | /* |
728 | * mm_update_next_owner() may clear mm->owner to NULL | |
729 | * if it races with swapoff, page migration, etc. | |
730 | * So this can be called with p == NULL. | |
731 | */ | |
732 | if (unlikely(!p)) | |
733 | return NULL; | |
734 | ||
78fb7466 PE |
735 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
736 | struct mem_cgroup, css); | |
737 | } | |
738 | ||
a433658c | 739 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 KH |
740 | { |
741 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
742 | |
743 | if (!mm) | |
744 | return NULL; | |
54595fe2 KH |
745 | /* |
746 | * Because we have no locks, mm->owner's may be being moved to other | |
747 | * cgroup. We use css_tryget() here even if this looks | |
748 | * pessimistic (rather than adding locks here). | |
749 | */ | |
750 | rcu_read_lock(); | |
751 | do { | |
752 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
753 | if (unlikely(!mem)) | |
754 | break; | |
755 | } while (!css_tryget(&mem->css)); | |
756 | rcu_read_unlock(); | |
757 | return mem; | |
758 | } | |
759 | ||
7d74b06f KH |
760 | /* The caller has to guarantee "mem" exists before calling this */ |
761 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem) | |
14067bb3 | 762 | { |
711d3d2c KH |
763 | struct cgroup_subsys_state *css; |
764 | int found; | |
765 | ||
766 | if (!mem) /* ROOT cgroup has the smallest ID */ | |
767 | return root_mem_cgroup; /*css_put/get against root is ignored*/ | |
768 | if (!mem->use_hierarchy) { | |
769 | if (css_tryget(&mem->css)) | |
770 | return mem; | |
771 | return NULL; | |
772 | } | |
773 | rcu_read_lock(); | |
774 | /* | |
775 | * searching a memory cgroup which has the smallest ID under given | |
776 | * ROOT cgroup. (ID >= 1) | |
777 | */ | |
778 | css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found); | |
779 | if (css && css_tryget(css)) | |
780 | mem = container_of(css, struct mem_cgroup, css); | |
781 | else | |
782 | mem = NULL; | |
783 | rcu_read_unlock(); | |
784 | return mem; | |
7d74b06f KH |
785 | } |
786 | ||
787 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, | |
788 | struct mem_cgroup *root, | |
789 | bool cond) | |
790 | { | |
791 | int nextid = css_id(&iter->css) + 1; | |
792 | int found; | |
793 | int hierarchy_used; | |
14067bb3 | 794 | struct cgroup_subsys_state *css; |
14067bb3 | 795 | |
7d74b06f | 796 | hierarchy_used = iter->use_hierarchy; |
14067bb3 | 797 | |
7d74b06f | 798 | css_put(&iter->css); |
711d3d2c KH |
799 | /* If no ROOT, walk all, ignore hierarchy */ |
800 | if (!cond || (root && !hierarchy_used)) | |
7d74b06f | 801 | return NULL; |
14067bb3 | 802 | |
711d3d2c KH |
803 | if (!root) |
804 | root = root_mem_cgroup; | |
805 | ||
7d74b06f KH |
806 | do { |
807 | iter = NULL; | |
14067bb3 | 808 | rcu_read_lock(); |
7d74b06f KH |
809 | |
810 | css = css_get_next(&mem_cgroup_subsys, nextid, | |
811 | &root->css, &found); | |
14067bb3 | 812 | if (css && css_tryget(css)) |
7d74b06f | 813 | iter = container_of(css, struct mem_cgroup, css); |
14067bb3 | 814 | rcu_read_unlock(); |
7d74b06f | 815 | /* If css is NULL, no more cgroups will be found */ |
14067bb3 | 816 | nextid = found + 1; |
7d74b06f | 817 | } while (css && !iter); |
14067bb3 | 818 | |
7d74b06f | 819 | return iter; |
14067bb3 | 820 | } |
7d74b06f KH |
821 | /* |
822 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please | |
823 | * be careful that "break" loop is not allowed. We have reference count. | |
824 | * Instead of that modify "cond" to be false and "continue" to exit the loop. | |
825 | */ | |
826 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ | |
827 | for (iter = mem_cgroup_start_loop(root);\ | |
828 | iter != NULL;\ | |
829 | iter = mem_cgroup_get_next(iter, root, cond)) | |
830 | ||
831 | #define for_each_mem_cgroup_tree(iter, root) \ | |
832 | for_each_mem_cgroup_tree_cond(iter, root, true) | |
833 | ||
711d3d2c KH |
834 | #define for_each_mem_cgroup_all(iter) \ |
835 | for_each_mem_cgroup_tree_cond(iter, NULL, true) | |
836 | ||
14067bb3 | 837 | |
4b3bde4c BS |
838 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
839 | { | |
840 | return (mem == root_mem_cgroup); | |
841 | } | |
842 | ||
456f998e YH |
843 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
844 | { | |
845 | struct mem_cgroup *mem; | |
846 | ||
847 | if (!mm) | |
848 | return; | |
849 | ||
850 | rcu_read_lock(); | |
851 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
852 | if (unlikely(!mem)) | |
853 | goto out; | |
854 | ||
855 | switch (idx) { | |
856 | case PGMAJFAULT: | |
857 | mem_cgroup_pgmajfault(mem, 1); | |
858 | break; | |
859 | case PGFAULT: | |
860 | mem_cgroup_pgfault(mem, 1); | |
861 | break; | |
862 | default: | |
863 | BUG(); | |
864 | } | |
865 | out: | |
866 | rcu_read_unlock(); | |
867 | } | |
868 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
869 | ||
08e552c6 KH |
870 | /* |
871 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
872 | * Operations are called by routine of global LRU independently from memcg. | |
873 | * What we have to take care of here is validness of pc->mem_cgroup. | |
874 | * | |
875 | * Changes to pc->mem_cgroup happens when | |
876 | * 1. charge | |
877 | * 2. moving account | |
878 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
879 | * It is added to LRU before charge. | |
880 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
881 | * When moving account, the page is not on LRU. It's isolated. | |
882 | */ | |
4f98a2fe | 883 | |
08e552c6 KH |
884 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
885 | { | |
886 | struct page_cgroup *pc; | |
08e552c6 | 887 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 888 | |
f8d66542 | 889 | if (mem_cgroup_disabled()) |
08e552c6 KH |
890 | return; |
891 | pc = lookup_page_cgroup(page); | |
892 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 893 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 894 | return; |
4b3bde4c | 895 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
896 | /* |
897 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
898 | * removed from global LRU. | |
899 | */ | |
97a6c37b | 900 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
901 | /* huge page split is done under lru_lock. so, we have no races. */ |
902 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); | |
4b3bde4c BS |
903 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
904 | return; | |
905 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 | 906 | list_del_init(&pc->lru); |
6d12e2d8 KH |
907 | } |
908 | ||
08e552c6 | 909 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 910 | { |
08e552c6 KH |
911 | mem_cgroup_del_lru_list(page, page_lru(page)); |
912 | } | |
b69408e8 | 913 | |
3f58a829 MK |
914 | /* |
915 | * Writeback is about to end against a page which has been marked for immediate | |
916 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
917 | * inactive list. | |
918 | */ | |
919 | void mem_cgroup_rotate_reclaimable_page(struct page *page) | |
920 | { | |
921 | struct mem_cgroup_per_zone *mz; | |
922 | struct page_cgroup *pc; | |
923 | enum lru_list lru = page_lru(page); | |
924 | ||
925 | if (mem_cgroup_disabled()) | |
926 | return; | |
927 | ||
928 | pc = lookup_page_cgroup(page); | |
929 | /* unused or root page is not rotated. */ | |
930 | if (!PageCgroupUsed(pc)) | |
931 | return; | |
932 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
933 | smp_rmb(); | |
934 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
935 | return; | |
97a6c37b | 936 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3f58a829 MK |
937 | list_move_tail(&pc->lru, &mz->lists[lru]); |
938 | } | |
939 | ||
08e552c6 KH |
940 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
941 | { | |
942 | struct mem_cgroup_per_zone *mz; | |
943 | struct page_cgroup *pc; | |
b69408e8 | 944 | |
f8d66542 | 945 | if (mem_cgroup_disabled()) |
08e552c6 | 946 | return; |
6d12e2d8 | 947 | |
08e552c6 | 948 | pc = lookup_page_cgroup(page); |
4b3bde4c | 949 | /* unused or root page is not rotated. */ |
713735b4 JW |
950 | if (!PageCgroupUsed(pc)) |
951 | return; | |
952 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
953 | smp_rmb(); | |
954 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 | 955 | return; |
97a6c37b | 956 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
08e552c6 | 957 | list_move(&pc->lru, &mz->lists[lru]); |
6d12e2d8 KH |
958 | } |
959 | ||
08e552c6 | 960 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 961 | { |
08e552c6 KH |
962 | struct page_cgroup *pc; |
963 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 964 | |
f8d66542 | 965 | if (mem_cgroup_disabled()) |
08e552c6 KH |
966 | return; |
967 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 968 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
08e552c6 | 969 | if (!PageCgroupUsed(pc)) |
894bc310 | 970 | return; |
713735b4 JW |
971 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
972 | smp_rmb(); | |
97a6c37b | 973 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
974 | /* huge page split is done under lru_lock. so, we have no races. */ |
975 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); | |
4b3bde4c BS |
976 | SetPageCgroupAcctLRU(pc); |
977 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
978 | return; | |
08e552c6 KH |
979 | list_add(&pc->lru, &mz->lists[lru]); |
980 | } | |
544122e5 | 981 | |
08e552c6 | 982 | /* |
5a6475a4 KH |
983 | * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed |
984 | * while it's linked to lru because the page may be reused after it's fully | |
985 | * uncharged. To handle that, unlink page_cgroup from LRU when charge it again. | |
986 | * It's done under lock_page and expected that zone->lru_lock isnever held. | |
08e552c6 | 987 | */ |
5a6475a4 | 988 | static void mem_cgroup_lru_del_before_commit(struct page *page) |
08e552c6 | 989 | { |
544122e5 KH |
990 | unsigned long flags; |
991 | struct zone *zone = page_zone(page); | |
992 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
993 | ||
5a6475a4 KH |
994 | /* |
995 | * Doing this check without taking ->lru_lock seems wrong but this | |
996 | * is safe. Because if page_cgroup's USED bit is unset, the page | |
997 | * will not be added to any memcg's LRU. If page_cgroup's USED bit is | |
998 | * set, the commit after this will fail, anyway. | |
999 | * This all charge/uncharge is done under some mutual execustion. | |
1000 | * So, we don't need to taking care of changes in USED bit. | |
1001 | */ | |
1002 | if (likely(!PageLRU(page))) | |
1003 | return; | |
1004 | ||
544122e5 KH |
1005 | spin_lock_irqsave(&zone->lru_lock, flags); |
1006 | /* | |
1007 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
1008 | * is guarded by lock_page() because the page is SwapCache. | |
1009 | */ | |
1010 | if (!PageCgroupUsed(pc)) | |
1011 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
1012 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
1013 | } |
1014 | ||
5a6475a4 | 1015 | static void mem_cgroup_lru_add_after_commit(struct page *page) |
544122e5 KH |
1016 | { |
1017 | unsigned long flags; | |
1018 | struct zone *zone = page_zone(page); | |
1019 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1020 | ||
5a6475a4 KH |
1021 | /* taking care of that the page is added to LRU while we commit it */ |
1022 | if (likely(!PageLRU(page))) | |
1023 | return; | |
544122e5 KH |
1024 | spin_lock_irqsave(&zone->lru_lock, flags); |
1025 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 1026 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
1027 | mem_cgroup_add_lru_list(page, page_lru(page)); |
1028 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
1029 | } | |
1030 | ||
1031 | ||
08e552c6 KH |
1032 | void mem_cgroup_move_lists(struct page *page, |
1033 | enum lru_list from, enum lru_list to) | |
1034 | { | |
f8d66542 | 1035 | if (mem_cgroup_disabled()) |
08e552c6 KH |
1036 | return; |
1037 | mem_cgroup_del_lru_list(page, from); | |
1038 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
1039 | } |
1040 | ||
4c4a2214 DR |
1041 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
1042 | { | |
1043 | int ret; | |
0b7f569e | 1044 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1045 | struct task_struct *p; |
4c4a2214 | 1046 | |
158e0a2d KH |
1047 | p = find_lock_task_mm(task); |
1048 | if (!p) | |
1049 | return 0; | |
1050 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1051 | task_unlock(p); | |
0b7f569e KH |
1052 | if (!curr) |
1053 | return 0; | |
d31f56db DN |
1054 | /* |
1055 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
1056 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
1057 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
1058 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
1059 | */ | |
1060 | if (mem->use_hierarchy) | |
0b7f569e KH |
1061 | ret = css_is_ancestor(&curr->css, &mem->css); |
1062 | else | |
1063 | ret = (curr == mem); | |
1064 | css_put(&curr->css); | |
4c4a2214 DR |
1065 | return ret; |
1066 | } | |
1067 | ||
c772be93 | 1068 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
1069 | { |
1070 | unsigned long active; | |
1071 | unsigned long inactive; | |
c772be93 KM |
1072 | unsigned long gb; |
1073 | unsigned long inactive_ratio; | |
14797e23 | 1074 | |
14067bb3 KH |
1075 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
1076 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 1077 | |
c772be93 KM |
1078 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1079 | if (gb) | |
1080 | inactive_ratio = int_sqrt(10 * gb); | |
1081 | else | |
1082 | inactive_ratio = 1; | |
1083 | ||
1084 | if (present_pages) { | |
1085 | present_pages[0] = inactive; | |
1086 | present_pages[1] = active; | |
1087 | } | |
1088 | ||
1089 | return inactive_ratio; | |
1090 | } | |
1091 | ||
1092 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
1093 | { | |
1094 | unsigned long active; | |
1095 | unsigned long inactive; | |
1096 | unsigned long present_pages[2]; | |
1097 | unsigned long inactive_ratio; | |
1098 | ||
1099 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
1100 | ||
1101 | inactive = present_pages[0]; | |
1102 | active = present_pages[1]; | |
1103 | ||
1104 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
1105 | return 1; |
1106 | ||
1107 | return 0; | |
1108 | } | |
1109 | ||
56e49d21 RR |
1110 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
1111 | { | |
1112 | unsigned long active; | |
1113 | unsigned long inactive; | |
1114 | ||
1115 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
1116 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
1117 | ||
1118 | return (active > inactive); | |
1119 | } | |
1120 | ||
1bac180b YH |
1121 | unsigned long mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, |
1122 | struct zone *zone, | |
1123 | enum lru_list lru) | |
a3d8e054 | 1124 | { |
13d7e3a2 | 1125 | int nid = zone_to_nid(zone); |
a3d8e054 KM |
1126 | int zid = zone_idx(zone); |
1127 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1128 | ||
1129 | return MEM_CGROUP_ZSTAT(mz, lru); | |
1130 | } | |
1131 | ||
406eb0c9 YH |
1132 | #ifdef CONFIG_NUMA |
1133 | static unsigned long mem_cgroup_node_nr_file_lru_pages(struct mem_cgroup *memcg, | |
1134 | int nid) | |
1135 | { | |
1136 | unsigned long ret; | |
1137 | ||
1138 | ret = mem_cgroup_get_zonestat_node(memcg, nid, LRU_INACTIVE_FILE) + | |
1139 | mem_cgroup_get_zonestat_node(memcg, nid, LRU_ACTIVE_FILE); | |
1140 | ||
1141 | return ret; | |
1142 | } | |
1143 | ||
1144 | static unsigned long mem_cgroup_nr_file_lru_pages(struct mem_cgroup *memcg) | |
1145 | { | |
1146 | u64 total = 0; | |
1147 | int nid; | |
1148 | ||
1149 | for_each_node_state(nid, N_HIGH_MEMORY) | |
1150 | total += mem_cgroup_node_nr_file_lru_pages(memcg, nid); | |
1151 | ||
1152 | return total; | |
1153 | } | |
1154 | ||
1155 | static unsigned long mem_cgroup_node_nr_anon_lru_pages(struct mem_cgroup *memcg, | |
1156 | int nid) | |
1157 | { | |
1158 | unsigned long ret; | |
1159 | ||
1160 | ret = mem_cgroup_get_zonestat_node(memcg, nid, LRU_INACTIVE_ANON) + | |
1161 | mem_cgroup_get_zonestat_node(memcg, nid, LRU_ACTIVE_ANON); | |
1162 | ||
1163 | return ret; | |
1164 | } | |
1165 | ||
1166 | static unsigned long mem_cgroup_nr_anon_lru_pages(struct mem_cgroup *memcg) | |
1167 | { | |
1168 | u64 total = 0; | |
1169 | int nid; | |
1170 | ||
1171 | for_each_node_state(nid, N_HIGH_MEMORY) | |
1172 | total += mem_cgroup_node_nr_anon_lru_pages(memcg, nid); | |
1173 | ||
1174 | return total; | |
1175 | } | |
1176 | ||
1177 | static unsigned long | |
1178 | mem_cgroup_node_nr_unevictable_lru_pages(struct mem_cgroup *memcg, int nid) | |
1179 | { | |
1180 | return mem_cgroup_get_zonestat_node(memcg, nid, LRU_UNEVICTABLE); | |
1181 | } | |
1182 | ||
1183 | static unsigned long | |
1184 | mem_cgroup_nr_unevictable_lru_pages(struct mem_cgroup *memcg) | |
1185 | { | |
1186 | u64 total = 0; | |
1187 | int nid; | |
1188 | ||
1189 | for_each_node_state(nid, N_HIGH_MEMORY) | |
1190 | total += mem_cgroup_node_nr_unevictable_lru_pages(memcg, nid); | |
1191 | ||
1192 | return total; | |
1193 | } | |
1194 | ||
1195 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
1196 | int nid) | |
1197 | { | |
1198 | enum lru_list l; | |
1199 | u64 total = 0; | |
1200 | ||
1201 | for_each_lru(l) | |
1202 | total += mem_cgroup_get_zonestat_node(memcg, nid, l); | |
1203 | ||
1204 | return total; | |
1205 | } | |
1206 | ||
1207 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg) | |
1208 | { | |
1209 | u64 total = 0; | |
1210 | int nid; | |
1211 | ||
1212 | for_each_node_state(nid, N_HIGH_MEMORY) | |
1213 | total += mem_cgroup_node_nr_lru_pages(memcg, nid); | |
1214 | ||
1215 | return total; | |
1216 | } | |
1217 | #endif /* CONFIG_NUMA */ | |
1218 | ||
3e2f41f1 KM |
1219 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1220 | struct zone *zone) | |
1221 | { | |
13d7e3a2 | 1222 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1223 | int zid = zone_idx(zone); |
1224 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1225 | ||
1226 | return &mz->reclaim_stat; | |
1227 | } | |
1228 | ||
1229 | struct zone_reclaim_stat * | |
1230 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1231 | { | |
1232 | struct page_cgroup *pc; | |
1233 | struct mem_cgroup_per_zone *mz; | |
1234 | ||
1235 | if (mem_cgroup_disabled()) | |
1236 | return NULL; | |
1237 | ||
1238 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1239 | if (!PageCgroupUsed(pc)) |
1240 | return NULL; | |
713735b4 JW |
1241 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1242 | smp_rmb(); | |
97a6c37b | 1243 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3e2f41f1 KM |
1244 | return &mz->reclaim_stat; |
1245 | } | |
1246 | ||
66e1707b BS |
1247 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
1248 | struct list_head *dst, | |
1249 | unsigned long *scanned, int order, | |
1250 | int mode, struct zone *z, | |
1251 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 1252 | int active, int file) |
66e1707b BS |
1253 | { |
1254 | unsigned long nr_taken = 0; | |
1255 | struct page *page; | |
1256 | unsigned long scan; | |
1257 | LIST_HEAD(pc_list); | |
1258 | struct list_head *src; | |
ff7283fa | 1259 | struct page_cgroup *pc, *tmp; |
13d7e3a2 | 1260 | int nid = zone_to_nid(z); |
1ecaab2b KH |
1261 | int zid = zone_idx(z); |
1262 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 1263 | int lru = LRU_FILE * file + active; |
2ffebca6 | 1264 | int ret; |
66e1707b | 1265 | |
cf475ad2 | 1266 | BUG_ON(!mem_cont); |
1ecaab2b | 1267 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 1268 | src = &mz->lists[lru]; |
66e1707b | 1269 | |
ff7283fa KH |
1270 | scan = 0; |
1271 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 1272 | if (scan >= nr_to_scan) |
ff7283fa | 1273 | break; |
08e552c6 | 1274 | |
52d4b9ac KH |
1275 | if (unlikely(!PageCgroupUsed(pc))) |
1276 | continue; | |
5564e88b | 1277 | |
6b3ae58e | 1278 | page = lookup_cgroup_page(pc); |
5564e88b | 1279 | |
436c6541 | 1280 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1281 | continue; |
ff7283fa | 1282 | |
436c6541 | 1283 | scan++; |
2ffebca6 KH |
1284 | ret = __isolate_lru_page(page, mode, file); |
1285 | switch (ret) { | |
1286 | case 0: | |
66e1707b | 1287 | list_move(&page->lru, dst); |
2ffebca6 | 1288 | mem_cgroup_del_lru(page); |
2c888cfb | 1289 | nr_taken += hpage_nr_pages(page); |
2ffebca6 KH |
1290 | break; |
1291 | case -EBUSY: | |
1292 | /* we don't affect global LRU but rotate in our LRU */ | |
1293 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1294 | break; | |
1295 | default: | |
1296 | break; | |
66e1707b BS |
1297 | } |
1298 | } | |
1299 | ||
66e1707b | 1300 | *scanned = scan; |
cc8e970c KM |
1301 | |
1302 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, | |
1303 | 0, 0, 0, mode); | |
1304 | ||
66e1707b BS |
1305 | return nr_taken; |
1306 | } | |
1307 | ||
6d61ef40 BS |
1308 | #define mem_cgroup_from_res_counter(counter, member) \ |
1309 | container_of(counter, struct mem_cgroup, member) | |
1310 | ||
19942822 | 1311 | /** |
9d11ea9f JW |
1312 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1313 | * @mem: the memory cgroup | |
19942822 | 1314 | * |
9d11ea9f | 1315 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1316 | * pages. |
19942822 | 1317 | */ |
7ec99d62 | 1318 | static unsigned long mem_cgroup_margin(struct mem_cgroup *mem) |
19942822 | 1319 | { |
9d11ea9f JW |
1320 | unsigned long long margin; |
1321 | ||
1322 | margin = res_counter_margin(&mem->res); | |
1323 | if (do_swap_account) | |
1324 | margin = min(margin, res_counter_margin(&mem->memsw)); | |
7ec99d62 | 1325 | return margin >> PAGE_SHIFT; |
19942822 JW |
1326 | } |
1327 | ||
a7885eb8 KM |
1328 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1329 | { | |
1330 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1331 | |
1332 | /* root ? */ | |
1333 | if (cgrp->parent == NULL) | |
1334 | return vm_swappiness; | |
1335 | ||
bf1ff263 | 1336 | return memcg->swappiness; |
a7885eb8 KM |
1337 | } |
1338 | ||
32047e2a KH |
1339 | static void mem_cgroup_start_move(struct mem_cgroup *mem) |
1340 | { | |
1341 | int cpu; | |
1489ebad KH |
1342 | |
1343 | get_online_cpus(); | |
1344 | spin_lock(&mem->pcp_counter_lock); | |
1345 | for_each_online_cpu(cpu) | |
32047e2a | 1346 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1489ebad KH |
1347 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; |
1348 | spin_unlock(&mem->pcp_counter_lock); | |
1349 | put_online_cpus(); | |
32047e2a KH |
1350 | |
1351 | synchronize_rcu(); | |
1352 | } | |
1353 | ||
1354 | static void mem_cgroup_end_move(struct mem_cgroup *mem) | |
1355 | { | |
1356 | int cpu; | |
1357 | ||
1358 | if (!mem) | |
1359 | return; | |
1489ebad KH |
1360 | get_online_cpus(); |
1361 | spin_lock(&mem->pcp_counter_lock); | |
1362 | for_each_online_cpu(cpu) | |
32047e2a | 1363 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1489ebad KH |
1364 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; |
1365 | spin_unlock(&mem->pcp_counter_lock); | |
1366 | put_online_cpus(); | |
32047e2a KH |
1367 | } |
1368 | /* | |
1369 | * 2 routines for checking "mem" is under move_account() or not. | |
1370 | * | |
1371 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1372 | * for avoiding race in accounting. If true, | |
1373 | * pc->mem_cgroup may be overwritten. | |
1374 | * | |
1375 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1376 | * under hierarchy of moving cgroups. This is for | |
1377 | * waiting at hith-memory prressure caused by "move". | |
1378 | */ | |
1379 | ||
1380 | static bool mem_cgroup_stealed(struct mem_cgroup *mem) | |
1381 | { | |
1382 | VM_BUG_ON(!rcu_read_lock_held()); | |
1383 | return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0; | |
1384 | } | |
4b534334 KH |
1385 | |
1386 | static bool mem_cgroup_under_move(struct mem_cgroup *mem) | |
1387 | { | |
2bd9bb20 KH |
1388 | struct mem_cgroup *from; |
1389 | struct mem_cgroup *to; | |
4b534334 | 1390 | bool ret = false; |
2bd9bb20 KH |
1391 | /* |
1392 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1393 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1394 | */ | |
1395 | spin_lock(&mc.lock); | |
1396 | from = mc.from; | |
1397 | to = mc.to; | |
1398 | if (!from) | |
1399 | goto unlock; | |
1400 | if (from == mem || to == mem | |
1401 | || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css)) | |
1402 | || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css))) | |
1403 | ret = true; | |
1404 | unlock: | |
1405 | spin_unlock(&mc.lock); | |
4b534334 KH |
1406 | return ret; |
1407 | } | |
1408 | ||
1409 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem) | |
1410 | { | |
1411 | if (mc.moving_task && current != mc.moving_task) { | |
1412 | if (mem_cgroup_under_move(mem)) { | |
1413 | DEFINE_WAIT(wait); | |
1414 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1415 | /* moving charge context might have finished. */ | |
1416 | if (mc.moving_task) | |
1417 | schedule(); | |
1418 | finish_wait(&mc.waitq, &wait); | |
1419 | return true; | |
1420 | } | |
1421 | } | |
1422 | return false; | |
1423 | } | |
1424 | ||
e222432b | 1425 | /** |
6a6135b6 | 1426 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1427 | * @memcg: The memory cgroup that went over limit |
1428 | * @p: Task that is going to be killed | |
1429 | * | |
1430 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1431 | * enabled | |
1432 | */ | |
1433 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1434 | { | |
1435 | struct cgroup *task_cgrp; | |
1436 | struct cgroup *mem_cgrp; | |
1437 | /* | |
1438 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1439 | * on the assumption that OOM is serialized for memory controller. | |
1440 | * If this assumption is broken, revisit this code. | |
1441 | */ | |
1442 | static char memcg_name[PATH_MAX]; | |
1443 | int ret; | |
1444 | ||
d31f56db | 1445 | if (!memcg || !p) |
e222432b BS |
1446 | return; |
1447 | ||
1448 | ||
1449 | rcu_read_lock(); | |
1450 | ||
1451 | mem_cgrp = memcg->css.cgroup; | |
1452 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1453 | ||
1454 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1455 | if (ret < 0) { | |
1456 | /* | |
1457 | * Unfortunately, we are unable to convert to a useful name | |
1458 | * But we'll still print out the usage information | |
1459 | */ | |
1460 | rcu_read_unlock(); | |
1461 | goto done; | |
1462 | } | |
1463 | rcu_read_unlock(); | |
1464 | ||
1465 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1466 | ||
1467 | rcu_read_lock(); | |
1468 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1469 | if (ret < 0) { | |
1470 | rcu_read_unlock(); | |
1471 | goto done; | |
1472 | } | |
1473 | rcu_read_unlock(); | |
1474 | ||
1475 | /* | |
1476 | * Continues from above, so we don't need an KERN_ level | |
1477 | */ | |
1478 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1479 | done: | |
1480 | ||
1481 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1482 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1483 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1484 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1485 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1486 | "failcnt %llu\n", | |
1487 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1488 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1489 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1490 | } | |
1491 | ||
81d39c20 KH |
1492 | /* |
1493 | * This function returns the number of memcg under hierarchy tree. Returns | |
1494 | * 1(self count) if no children. | |
1495 | */ | |
1496 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1497 | { | |
1498 | int num = 0; | |
7d74b06f KH |
1499 | struct mem_cgroup *iter; |
1500 | ||
1501 | for_each_mem_cgroup_tree(iter, mem) | |
1502 | num++; | |
81d39c20 KH |
1503 | return num; |
1504 | } | |
1505 | ||
a63d83f4 DR |
1506 | /* |
1507 | * Return the memory (and swap, if configured) limit for a memcg. | |
1508 | */ | |
1509 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1510 | { | |
1511 | u64 limit; | |
1512 | u64 memsw; | |
1513 | ||
f3e8eb70 JW |
1514 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1515 | limit += total_swap_pages << PAGE_SHIFT; | |
1516 | ||
a63d83f4 DR |
1517 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1518 | /* | |
1519 | * If memsw is finite and limits the amount of swap space available | |
1520 | * to this memcg, return that limit. | |
1521 | */ | |
1522 | return min(limit, memsw); | |
1523 | } | |
1524 | ||
6d61ef40 | 1525 | /* |
04046e1a KH |
1526 | * Visit the first child (need not be the first child as per the ordering |
1527 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1528 | * that to reclaim free pages from. | |
1529 | */ | |
1530 | static struct mem_cgroup * | |
1531 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1532 | { | |
1533 | struct mem_cgroup *ret = NULL; | |
1534 | struct cgroup_subsys_state *css; | |
1535 | int nextid, found; | |
1536 | ||
1537 | if (!root_mem->use_hierarchy) { | |
1538 | css_get(&root_mem->css); | |
1539 | ret = root_mem; | |
1540 | } | |
1541 | ||
1542 | while (!ret) { | |
1543 | rcu_read_lock(); | |
1544 | nextid = root_mem->last_scanned_child + 1; | |
1545 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1546 | &found); | |
1547 | if (css && css_tryget(css)) | |
1548 | ret = container_of(css, struct mem_cgroup, css); | |
1549 | ||
1550 | rcu_read_unlock(); | |
1551 | /* Updates scanning parameter */ | |
04046e1a KH |
1552 | if (!css) { |
1553 | /* this means start scan from ID:1 */ | |
1554 | root_mem->last_scanned_child = 0; | |
1555 | } else | |
1556 | root_mem->last_scanned_child = found; | |
04046e1a KH |
1557 | } |
1558 | ||
1559 | return ret; | |
1560 | } | |
1561 | ||
889976db YH |
1562 | #if MAX_NUMNODES > 1 |
1563 | ||
1564 | /* | |
1565 | * Always updating the nodemask is not very good - even if we have an empty | |
1566 | * list or the wrong list here, we can start from some node and traverse all | |
1567 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1568 | * | |
1569 | */ | |
1570 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *mem) | |
1571 | { | |
1572 | int nid; | |
1573 | ||
1574 | if (time_after(mem->next_scan_node_update, jiffies)) | |
1575 | return; | |
1576 | ||
1577 | mem->next_scan_node_update = jiffies + 10*HZ; | |
1578 | /* make a nodemask where this memcg uses memory from */ | |
1579 | mem->scan_nodes = node_states[N_HIGH_MEMORY]; | |
1580 | ||
1581 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1582 | ||
1583 | if (mem_cgroup_get_zonestat_node(mem, nid, LRU_INACTIVE_FILE) || | |
1584 | mem_cgroup_get_zonestat_node(mem, nid, LRU_ACTIVE_FILE)) | |
1585 | continue; | |
1586 | ||
1587 | if (total_swap_pages && | |
1588 | (mem_cgroup_get_zonestat_node(mem, nid, LRU_INACTIVE_ANON) || | |
1589 | mem_cgroup_get_zonestat_node(mem, nid, LRU_ACTIVE_ANON))) | |
1590 | continue; | |
1591 | node_clear(nid, mem->scan_nodes); | |
1592 | } | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * Selecting a node where we start reclaim from. Because what we need is just | |
1597 | * reducing usage counter, start from anywhere is O,K. Considering | |
1598 | * memory reclaim from current node, there are pros. and cons. | |
1599 | * | |
1600 | * Freeing memory from current node means freeing memory from a node which | |
1601 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1602 | * hit limits, it will see a contention on a node. But freeing from remote | |
1603 | * node means more costs for memory reclaim because of memory latency. | |
1604 | * | |
1605 | * Now, we use round-robin. Better algorithm is welcomed. | |
1606 | */ | |
1607 | int mem_cgroup_select_victim_node(struct mem_cgroup *mem) | |
1608 | { | |
1609 | int node; | |
1610 | ||
1611 | mem_cgroup_may_update_nodemask(mem); | |
1612 | node = mem->last_scanned_node; | |
1613 | ||
1614 | node = next_node(node, mem->scan_nodes); | |
1615 | if (node == MAX_NUMNODES) | |
1616 | node = first_node(mem->scan_nodes); | |
1617 | /* | |
1618 | * We call this when we hit limit, not when pages are added to LRU. | |
1619 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1620 | * memcg is too small and all pages are not on LRU. In that case, | |
1621 | * we use curret node. | |
1622 | */ | |
1623 | if (unlikely(node == MAX_NUMNODES)) | |
1624 | node = numa_node_id(); | |
1625 | ||
1626 | mem->last_scanned_node = node; | |
1627 | return node; | |
1628 | } | |
1629 | ||
1630 | #else | |
1631 | int mem_cgroup_select_victim_node(struct mem_cgroup *mem) | |
1632 | { | |
1633 | return 0; | |
1634 | } | |
1635 | #endif | |
1636 | ||
04046e1a KH |
1637 | /* |
1638 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1639 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1640 | * based on its position in the children list. | |
6d61ef40 BS |
1641 | * |
1642 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1643 | * |
1644 | * We give up and return to the caller when we visit root_mem twice. | |
1645 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1646 | * |
1647 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1648 | */ |
1649 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1650 | struct zone *zone, |
75822b44 | 1651 | gfp_t gfp_mask, |
0ae5e89c YH |
1652 | unsigned long reclaim_options, |
1653 | unsigned long *total_scanned) | |
6d61ef40 | 1654 | { |
04046e1a KH |
1655 | struct mem_cgroup *victim; |
1656 | int ret, total = 0; | |
1657 | int loop = 0; | |
75822b44 BS |
1658 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1659 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 | 1660 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
9d11ea9f | 1661 | unsigned long excess; |
0ae5e89c | 1662 | unsigned long nr_scanned; |
9d11ea9f JW |
1663 | |
1664 | excess = res_counter_soft_limit_excess(&root_mem->res) >> PAGE_SHIFT; | |
04046e1a | 1665 | |
22a668d7 | 1666 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
7ae534d0 | 1667 | if (!check_soft && root_mem->memsw_is_minimum) |
22a668d7 KH |
1668 | noswap = true; |
1669 | ||
4e416953 | 1670 | while (1) { |
04046e1a | 1671 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1672 | if (victim == root_mem) { |
04046e1a | 1673 | loop++; |
fbc29a25 KH |
1674 | /* |
1675 | * We are not draining per cpu cached charges during | |
1676 | * soft limit reclaim because global reclaim doesn't | |
1677 | * care about charges. It tries to free some memory and | |
1678 | * charges will not give any. | |
1679 | */ | |
1680 | if (!check_soft && loop >= 1) | |
26fe6168 | 1681 | drain_all_stock_async(root_mem); |
4e416953 BS |
1682 | if (loop >= 2) { |
1683 | /* | |
1684 | * If we have not been able to reclaim | |
1685 | * anything, it might because there are | |
1686 | * no reclaimable pages under this hierarchy | |
1687 | */ | |
1688 | if (!check_soft || !total) { | |
1689 | css_put(&victim->css); | |
1690 | break; | |
1691 | } | |
1692 | /* | |
25985edc | 1693 | * We want to do more targeted reclaim. |
4e416953 BS |
1694 | * excess >> 2 is not to excessive so as to |
1695 | * reclaim too much, nor too less that we keep | |
1696 | * coming back to reclaim from this cgroup | |
1697 | */ | |
1698 | if (total >= (excess >> 2) || | |
1699 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1700 | css_put(&victim->css); | |
1701 | break; | |
1702 | } | |
1703 | } | |
1704 | } | |
c62b1a3b | 1705 | if (!mem_cgroup_local_usage(victim)) { |
04046e1a KH |
1706 | /* this cgroup's local usage == 0 */ |
1707 | css_put(&victim->css); | |
6d61ef40 BS |
1708 | continue; |
1709 | } | |
04046e1a | 1710 | /* we use swappiness of local cgroup */ |
0ae5e89c | 1711 | if (check_soft) { |
4e416953 | 1712 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, |
0ae5e89c YH |
1713 | noswap, get_swappiness(victim), zone, |
1714 | &nr_scanned); | |
1715 | *total_scanned += nr_scanned; | |
1716 | } else | |
4e416953 BS |
1717 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, |
1718 | noswap, get_swappiness(victim)); | |
04046e1a | 1719 | css_put(&victim->css); |
81d39c20 KH |
1720 | /* |
1721 | * At shrinking usage, we can't check we should stop here or | |
1722 | * reclaim more. It's depends on callers. last_scanned_child | |
1723 | * will work enough for keeping fairness under tree. | |
1724 | */ | |
1725 | if (shrink) | |
1726 | return ret; | |
04046e1a | 1727 | total += ret; |
4e416953 | 1728 | if (check_soft) { |
9d11ea9f | 1729 | if (!res_counter_soft_limit_excess(&root_mem->res)) |
4e416953 | 1730 | return total; |
9d11ea9f | 1731 | } else if (mem_cgroup_margin(root_mem)) |
4fd14ebf | 1732 | return total; |
6d61ef40 | 1733 | } |
04046e1a | 1734 | return total; |
6d61ef40 BS |
1735 | } |
1736 | ||
867578cb KH |
1737 | /* |
1738 | * Check OOM-Killer is already running under our hierarchy. | |
1739 | * If someone is running, return false. | |
1740 | */ | |
1741 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) | |
1742 | { | |
7d74b06f KH |
1743 | int x, lock_count = 0; |
1744 | struct mem_cgroup *iter; | |
a636b327 | 1745 | |
7d74b06f KH |
1746 | for_each_mem_cgroup_tree(iter, mem) { |
1747 | x = atomic_inc_return(&iter->oom_lock); | |
1748 | lock_count = max(x, lock_count); | |
1749 | } | |
867578cb KH |
1750 | |
1751 | if (lock_count == 1) | |
1752 | return true; | |
1753 | return false; | |
a636b327 | 1754 | } |
0b7f569e | 1755 | |
7d74b06f | 1756 | static int mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
0b7f569e | 1757 | { |
7d74b06f KH |
1758 | struct mem_cgroup *iter; |
1759 | ||
867578cb KH |
1760 | /* |
1761 | * When a new child is created while the hierarchy is under oom, | |
1762 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1763 | * atomic_add_unless() here. | |
1764 | */ | |
7d74b06f KH |
1765 | for_each_mem_cgroup_tree(iter, mem) |
1766 | atomic_add_unless(&iter->oom_lock, -1, 0); | |
0b7f569e KH |
1767 | return 0; |
1768 | } | |
1769 | ||
867578cb KH |
1770 | |
1771 | static DEFINE_MUTEX(memcg_oom_mutex); | |
1772 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); | |
1773 | ||
dc98df5a KH |
1774 | struct oom_wait_info { |
1775 | struct mem_cgroup *mem; | |
1776 | wait_queue_t wait; | |
1777 | }; | |
1778 | ||
1779 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1780 | unsigned mode, int sync, void *arg) | |
1781 | { | |
1782 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; | |
1783 | struct oom_wait_info *oom_wait_info; | |
1784 | ||
1785 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
1786 | ||
1787 | if (oom_wait_info->mem == wake_mem) | |
1788 | goto wakeup; | |
1789 | /* if no hierarchy, no match */ | |
1790 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) | |
1791 | return 0; | |
1792 | /* | |
1793 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1794 | * Then we can use css_is_ancestor without taking care of RCU. | |
1795 | */ | |
1796 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && | |
1797 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) | |
1798 | return 0; | |
1799 | ||
1800 | wakeup: | |
1801 | return autoremove_wake_function(wait, mode, sync, arg); | |
1802 | } | |
1803 | ||
1804 | static void memcg_wakeup_oom(struct mem_cgroup *mem) | |
1805 | { | |
1806 | /* for filtering, pass "mem" as argument. */ | |
1807 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); | |
1808 | } | |
1809 | ||
3c11ecf4 KH |
1810 | static void memcg_oom_recover(struct mem_cgroup *mem) |
1811 | { | |
2bd9bb20 | 1812 | if (mem && atomic_read(&mem->oom_lock)) |
3c11ecf4 KH |
1813 | memcg_wakeup_oom(mem); |
1814 | } | |
1815 | ||
867578cb KH |
1816 | /* |
1817 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1818 | */ | |
1819 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) | |
0b7f569e | 1820 | { |
dc98df5a | 1821 | struct oom_wait_info owait; |
3c11ecf4 | 1822 | bool locked, need_to_kill; |
867578cb | 1823 | |
dc98df5a KH |
1824 | owait.mem = mem; |
1825 | owait.wait.flags = 0; | |
1826 | owait.wait.func = memcg_oom_wake_function; | |
1827 | owait.wait.private = current; | |
1828 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1829 | need_to_kill = true; |
867578cb KH |
1830 | /* At first, try to OOM lock hierarchy under mem.*/ |
1831 | mutex_lock(&memcg_oom_mutex); | |
1832 | locked = mem_cgroup_oom_lock(mem); | |
1833 | /* | |
1834 | * Even if signal_pending(), we can't quit charge() loop without | |
1835 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1836 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1837 | */ | |
3c11ecf4 KH |
1838 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
1839 | if (!locked || mem->oom_kill_disable) | |
1840 | need_to_kill = false; | |
1841 | if (locked) | |
9490ff27 | 1842 | mem_cgroup_oom_notify(mem); |
867578cb KH |
1843 | mutex_unlock(&memcg_oom_mutex); |
1844 | ||
3c11ecf4 KH |
1845 | if (need_to_kill) { |
1846 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
867578cb | 1847 | mem_cgroup_out_of_memory(mem, mask); |
3c11ecf4 | 1848 | } else { |
867578cb | 1849 | schedule(); |
dc98df5a | 1850 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb KH |
1851 | } |
1852 | mutex_lock(&memcg_oom_mutex); | |
1853 | mem_cgroup_oom_unlock(mem); | |
dc98df5a | 1854 | memcg_wakeup_oom(mem); |
867578cb KH |
1855 | mutex_unlock(&memcg_oom_mutex); |
1856 | ||
1857 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) | |
1858 | return false; | |
1859 | /* Give chance to dying process */ | |
1860 | schedule_timeout(1); | |
1861 | return true; | |
0b7f569e KH |
1862 | } |
1863 | ||
d69b042f BS |
1864 | /* |
1865 | * Currently used to update mapped file statistics, but the routine can be | |
1866 | * generalized to update other statistics as well. | |
32047e2a KH |
1867 | * |
1868 | * Notes: Race condition | |
1869 | * | |
1870 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1871 | * it tends to be costly. But considering some conditions, we doesn't need | |
1872 | * to do so _always_. | |
1873 | * | |
1874 | * Considering "charge", lock_page_cgroup() is not required because all | |
1875 | * file-stat operations happen after a page is attached to radix-tree. There | |
1876 | * are no race with "charge". | |
1877 | * | |
1878 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1879 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1880 | * if there are race with "uncharge". Statistics itself is properly handled | |
1881 | * by flags. | |
1882 | * | |
1883 | * Considering "move", this is an only case we see a race. To make the race | |
1884 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
1885 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 1886 | */ |
26174efd | 1887 | |
2a7106f2 GT |
1888 | void mem_cgroup_update_page_stat(struct page *page, |
1889 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f BS |
1890 | { |
1891 | struct mem_cgroup *mem; | |
32047e2a KH |
1892 | struct page_cgroup *pc = lookup_page_cgroup(page); |
1893 | bool need_unlock = false; | |
dbd4ea78 | 1894 | unsigned long uninitialized_var(flags); |
d69b042f | 1895 | |
d69b042f BS |
1896 | if (unlikely(!pc)) |
1897 | return; | |
1898 | ||
32047e2a | 1899 | rcu_read_lock(); |
d69b042f | 1900 | mem = pc->mem_cgroup; |
32047e2a KH |
1901 | if (unlikely(!mem || !PageCgroupUsed(pc))) |
1902 | goto out; | |
1903 | /* pc->mem_cgroup is unstable ? */ | |
ca3e0214 | 1904 | if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) { |
32047e2a | 1905 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 1906 | move_lock_page_cgroup(pc, &flags); |
32047e2a KH |
1907 | need_unlock = true; |
1908 | mem = pc->mem_cgroup; | |
1909 | if (!mem || !PageCgroupUsed(pc)) | |
1910 | goto out; | |
1911 | } | |
26174efd | 1912 | |
26174efd | 1913 | switch (idx) { |
2a7106f2 | 1914 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
1915 | if (val > 0) |
1916 | SetPageCgroupFileMapped(pc); | |
1917 | else if (!page_mapped(page)) | |
0c270f8f | 1918 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 1919 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1920 | break; |
1921 | default: | |
1922 | BUG(); | |
8725d541 | 1923 | } |
d69b042f | 1924 | |
2a7106f2 GT |
1925 | this_cpu_add(mem->stat->count[idx], val); |
1926 | ||
32047e2a KH |
1927 | out: |
1928 | if (unlikely(need_unlock)) | |
dbd4ea78 | 1929 | move_unlock_page_cgroup(pc, &flags); |
32047e2a KH |
1930 | rcu_read_unlock(); |
1931 | return; | |
d69b042f | 1932 | } |
2a7106f2 | 1933 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
26174efd | 1934 | |
cdec2e42 KH |
1935 | /* |
1936 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1937 | * TODO: maybe necessary to use big numbers in big irons. | |
1938 | */ | |
7ec99d62 | 1939 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1940 | struct memcg_stock_pcp { |
1941 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1942 | unsigned int nr_pages; |
cdec2e42 | 1943 | struct work_struct work; |
26fe6168 KH |
1944 | unsigned long flags; |
1945 | #define FLUSHING_CACHED_CHARGE (0) | |
cdec2e42 KH |
1946 | }; |
1947 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
26fe6168 | 1948 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
1949 | |
1950 | /* | |
11c9ea4e | 1951 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
1952 | * from local stock and true is returned. If the stock is 0 or charges from a |
1953 | * cgroup which is not current target, returns false. This stock will be | |
1954 | * refilled. | |
1955 | */ | |
1956 | static bool consume_stock(struct mem_cgroup *mem) | |
1957 | { | |
1958 | struct memcg_stock_pcp *stock; | |
1959 | bool ret = true; | |
1960 | ||
1961 | stock = &get_cpu_var(memcg_stock); | |
11c9ea4e JW |
1962 | if (mem == stock->cached && stock->nr_pages) |
1963 | stock->nr_pages--; | |
cdec2e42 KH |
1964 | else /* need to call res_counter_charge */ |
1965 | ret = false; | |
1966 | put_cpu_var(memcg_stock); | |
1967 | return ret; | |
1968 | } | |
1969 | ||
1970 | /* | |
1971 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1972 | */ | |
1973 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1974 | { | |
1975 | struct mem_cgroup *old = stock->cached; | |
1976 | ||
11c9ea4e JW |
1977 | if (stock->nr_pages) { |
1978 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
1979 | ||
1980 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 1981 | if (do_swap_account) |
11c9ea4e JW |
1982 | res_counter_uncharge(&old->memsw, bytes); |
1983 | stock->nr_pages = 0; | |
cdec2e42 KH |
1984 | } |
1985 | stock->cached = NULL; | |
cdec2e42 KH |
1986 | } |
1987 | ||
1988 | /* | |
1989 | * This must be called under preempt disabled or must be called by | |
1990 | * a thread which is pinned to local cpu. | |
1991 | */ | |
1992 | static void drain_local_stock(struct work_struct *dummy) | |
1993 | { | |
1994 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1995 | drain_stock(stock); | |
26fe6168 | 1996 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
1997 | } |
1998 | ||
1999 | /* | |
2000 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2001 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2002 | */ |
11c9ea4e | 2003 | static void refill_stock(struct mem_cgroup *mem, unsigned int nr_pages) |
cdec2e42 KH |
2004 | { |
2005 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2006 | ||
2007 | if (stock->cached != mem) { /* reset if necessary */ | |
2008 | drain_stock(stock); | |
2009 | stock->cached = mem; | |
2010 | } | |
11c9ea4e | 2011 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2012 | put_cpu_var(memcg_stock); |
2013 | } | |
2014 | ||
2015 | /* | |
2016 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2017 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2018 | * expects some charges will be back to res_counter later but cannot wait for | |
2019 | * it. | |
2020 | */ | |
26fe6168 | 2021 | static void drain_all_stock_async(struct mem_cgroup *root_mem) |
cdec2e42 | 2022 | { |
26fe6168 KH |
2023 | int cpu, curcpu; |
2024 | /* | |
2025 | * If someone calls draining, avoid adding more kworker runs. | |
cdec2e42 | 2026 | */ |
26fe6168 | 2027 | if (!mutex_trylock(&percpu_charge_mutex)) |
cdec2e42 KH |
2028 | return; |
2029 | /* Notify other cpus that system-wide "drain" is running */ | |
cdec2e42 | 2030 | get_online_cpus(); |
26fe6168 KH |
2031 | /* |
2032 | * Get a hint for avoiding draining charges on the current cpu, | |
2033 | * which must be exhausted by our charging. It is not required that | |
2034 | * this be a precise check, so we use raw_smp_processor_id() instead of | |
2035 | * getcpu()/putcpu(). | |
2036 | */ | |
2037 | curcpu = raw_smp_processor_id(); | |
cdec2e42 KH |
2038 | for_each_online_cpu(cpu) { |
2039 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
26fe6168 KH |
2040 | struct mem_cgroup *mem; |
2041 | ||
2042 | if (cpu == curcpu) | |
2043 | continue; | |
2044 | ||
2045 | mem = stock->cached; | |
2046 | if (!mem) | |
2047 | continue; | |
2048 | if (mem != root_mem) { | |
2049 | if (!root_mem->use_hierarchy) | |
2050 | continue; | |
2051 | /* check whether "mem" is under tree of "root_mem" */ | |
2052 | if (!css_is_ancestor(&mem->css, &root_mem->css)) | |
2053 | continue; | |
2054 | } | |
2055 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) | |
2056 | schedule_work_on(cpu, &stock->work); | |
cdec2e42 KH |
2057 | } |
2058 | put_online_cpus(); | |
26fe6168 | 2059 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2060 | /* We don't wait for flush_work */ |
2061 | } | |
2062 | ||
2063 | /* This is a synchronous drain interface. */ | |
2064 | static void drain_all_stock_sync(void) | |
2065 | { | |
2066 | /* called when force_empty is called */ | |
26fe6168 | 2067 | mutex_lock(&percpu_charge_mutex); |
cdec2e42 | 2068 | schedule_on_each_cpu(drain_local_stock); |
26fe6168 | 2069 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2070 | } |
2071 | ||
711d3d2c KH |
2072 | /* |
2073 | * This function drains percpu counter value from DEAD cpu and | |
2074 | * move it to local cpu. Note that this function can be preempted. | |
2075 | */ | |
2076 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu) | |
2077 | { | |
2078 | int i; | |
2079 | ||
2080 | spin_lock(&mem->pcp_counter_lock); | |
2081 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { | |
7a159cc9 | 2082 | long x = per_cpu(mem->stat->count[i], cpu); |
711d3d2c KH |
2083 | |
2084 | per_cpu(mem->stat->count[i], cpu) = 0; | |
2085 | mem->nocpu_base.count[i] += x; | |
2086 | } | |
e9f8974f JW |
2087 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
2088 | unsigned long x = per_cpu(mem->stat->events[i], cpu); | |
2089 | ||
2090 | per_cpu(mem->stat->events[i], cpu) = 0; | |
2091 | mem->nocpu_base.events[i] += x; | |
2092 | } | |
1489ebad KH |
2093 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
2094 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; | |
2095 | spin_unlock(&mem->pcp_counter_lock); | |
2096 | } | |
2097 | ||
2098 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu) | |
2099 | { | |
2100 | int idx = MEM_CGROUP_ON_MOVE; | |
2101 | ||
2102 | spin_lock(&mem->pcp_counter_lock); | |
2103 | per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx]; | |
711d3d2c KH |
2104 | spin_unlock(&mem->pcp_counter_lock); |
2105 | } | |
2106 | ||
2107 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2108 | unsigned long action, |
2109 | void *hcpu) | |
2110 | { | |
2111 | int cpu = (unsigned long)hcpu; | |
2112 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2113 | struct mem_cgroup *iter; |
cdec2e42 | 2114 | |
1489ebad KH |
2115 | if ((action == CPU_ONLINE)) { |
2116 | for_each_mem_cgroup_all(iter) | |
2117 | synchronize_mem_cgroup_on_move(iter, cpu); | |
2118 | return NOTIFY_OK; | |
2119 | } | |
2120 | ||
711d3d2c | 2121 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 2122 | return NOTIFY_OK; |
711d3d2c KH |
2123 | |
2124 | for_each_mem_cgroup_all(iter) | |
2125 | mem_cgroup_drain_pcp_counter(iter, cpu); | |
2126 | ||
cdec2e42 KH |
2127 | stock = &per_cpu(memcg_stock, cpu); |
2128 | drain_stock(stock); | |
2129 | return NOTIFY_OK; | |
2130 | } | |
2131 | ||
4b534334 KH |
2132 | |
2133 | /* See __mem_cgroup_try_charge() for details */ | |
2134 | enum { | |
2135 | CHARGE_OK, /* success */ | |
2136 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2137 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2138 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2139 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2140 | }; | |
2141 | ||
7ec99d62 JW |
2142 | static int mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask, |
2143 | unsigned int nr_pages, bool oom_check) | |
4b534334 | 2144 | { |
7ec99d62 | 2145 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2146 | struct mem_cgroup *mem_over_limit; |
2147 | struct res_counter *fail_res; | |
2148 | unsigned long flags = 0; | |
2149 | int ret; | |
2150 | ||
2151 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
2152 | ||
2153 | if (likely(!ret)) { | |
2154 | if (!do_swap_account) | |
2155 | return CHARGE_OK; | |
2156 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); | |
2157 | if (likely(!ret)) | |
2158 | return CHARGE_OK; | |
2159 | ||
01c88e2d | 2160 | res_counter_uncharge(&mem->res, csize); |
4b534334 KH |
2161 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2162 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2163 | } else | |
2164 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2165 | /* |
7ec99d62 JW |
2166 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2167 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2168 | * |
2169 | * Never reclaim on behalf of optional batching, retry with a | |
2170 | * single page instead. | |
2171 | */ | |
7ec99d62 | 2172 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2173 | return CHARGE_RETRY; |
2174 | ||
2175 | if (!(gfp_mask & __GFP_WAIT)) | |
2176 | return CHARGE_WOULDBLOCK; | |
2177 | ||
2178 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, | |
0ae5e89c | 2179 | gfp_mask, flags, NULL); |
7ec99d62 | 2180 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2181 | return CHARGE_RETRY; |
4b534334 | 2182 | /* |
19942822 JW |
2183 | * Even though the limit is exceeded at this point, reclaim |
2184 | * may have been able to free some pages. Retry the charge | |
2185 | * before killing the task. | |
2186 | * | |
2187 | * Only for regular pages, though: huge pages are rather | |
2188 | * unlikely to succeed so close to the limit, and we fall back | |
2189 | * to regular pages anyway in case of failure. | |
4b534334 | 2190 | */ |
7ec99d62 | 2191 | if (nr_pages == 1 && ret) |
4b534334 KH |
2192 | return CHARGE_RETRY; |
2193 | ||
2194 | /* | |
2195 | * At task move, charge accounts can be doubly counted. So, it's | |
2196 | * better to wait until the end of task_move if something is going on. | |
2197 | */ | |
2198 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2199 | return CHARGE_RETRY; | |
2200 | ||
2201 | /* If we don't need to call oom-killer at el, return immediately */ | |
2202 | if (!oom_check) | |
2203 | return CHARGE_NOMEM; | |
2204 | /* check OOM */ | |
2205 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) | |
2206 | return CHARGE_OOM_DIE; | |
2207 | ||
2208 | return CHARGE_RETRY; | |
2209 | } | |
2210 | ||
f817ed48 KH |
2211 | /* |
2212 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
2213 | * oom-killer can be invoked. | |
8a9f3ccd | 2214 | */ |
f817ed48 | 2215 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2216 | gfp_t gfp_mask, |
7ec99d62 JW |
2217 | unsigned int nr_pages, |
2218 | struct mem_cgroup **memcg, | |
2219 | bool oom) | |
8a9f3ccd | 2220 | { |
7ec99d62 | 2221 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 KH |
2222 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
2223 | struct mem_cgroup *mem = NULL; | |
2224 | int ret; | |
a636b327 | 2225 | |
867578cb KH |
2226 | /* |
2227 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2228 | * in system level. So, allow to go ahead dying process in addition to | |
2229 | * MEMDIE process. | |
2230 | */ | |
2231 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2232 | || fatal_signal_pending(current))) | |
2233 | goto bypass; | |
a636b327 | 2234 | |
8a9f3ccd | 2235 | /* |
3be91277 HD |
2236 | * We always charge the cgroup the mm_struct belongs to. |
2237 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2238 | * thread group leader migrates. It's possible that mm is not |
2239 | * set, if so charge the init_mm (happens for pagecache usage). | |
2240 | */ | |
f75ca962 KH |
2241 | if (!*memcg && !mm) |
2242 | goto bypass; | |
2243 | again: | |
2244 | if (*memcg) { /* css should be a valid one */ | |
4b534334 | 2245 | mem = *memcg; |
f75ca962 KH |
2246 | VM_BUG_ON(css_is_removed(&mem->css)); |
2247 | if (mem_cgroup_is_root(mem)) | |
2248 | goto done; | |
7ec99d62 | 2249 | if (nr_pages == 1 && consume_stock(mem)) |
f75ca962 | 2250 | goto done; |
4b534334 KH |
2251 | css_get(&mem->css); |
2252 | } else { | |
f75ca962 | 2253 | struct task_struct *p; |
54595fe2 | 2254 | |
f75ca962 KH |
2255 | rcu_read_lock(); |
2256 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2257 | /* |
ebb76ce1 KH |
2258 | * Because we don't have task_lock(), "p" can exit. |
2259 | * In that case, "mem" can point to root or p can be NULL with | |
2260 | * race with swapoff. Then, we have small risk of mis-accouning. | |
2261 | * But such kind of mis-account by race always happens because | |
2262 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2263 | * small race, here. | |
2264 | * (*) swapoff at el will charge against mm-struct not against | |
2265 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 KH |
2266 | */ |
2267 | mem = mem_cgroup_from_task(p); | |
ebb76ce1 | 2268 | if (!mem || mem_cgroup_is_root(mem)) { |
f75ca962 KH |
2269 | rcu_read_unlock(); |
2270 | goto done; | |
2271 | } | |
7ec99d62 | 2272 | if (nr_pages == 1 && consume_stock(mem)) { |
f75ca962 KH |
2273 | /* |
2274 | * It seems dagerous to access memcg without css_get(). | |
2275 | * But considering how consume_stok works, it's not | |
2276 | * necessary. If consume_stock success, some charges | |
2277 | * from this memcg are cached on this cpu. So, we | |
2278 | * don't need to call css_get()/css_tryget() before | |
2279 | * calling consume_stock(). | |
2280 | */ | |
2281 | rcu_read_unlock(); | |
2282 | goto done; | |
2283 | } | |
2284 | /* after here, we may be blocked. we need to get refcnt */ | |
2285 | if (!css_tryget(&mem->css)) { | |
2286 | rcu_read_unlock(); | |
2287 | goto again; | |
2288 | } | |
2289 | rcu_read_unlock(); | |
2290 | } | |
8a9f3ccd | 2291 | |
4b534334 KH |
2292 | do { |
2293 | bool oom_check; | |
7a81b88c | 2294 | |
4b534334 | 2295 | /* If killed, bypass charge */ |
f75ca962 KH |
2296 | if (fatal_signal_pending(current)) { |
2297 | css_put(&mem->css); | |
4b534334 | 2298 | goto bypass; |
f75ca962 | 2299 | } |
6d61ef40 | 2300 | |
4b534334 KH |
2301 | oom_check = false; |
2302 | if (oom && !nr_oom_retries) { | |
2303 | oom_check = true; | |
2304 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2305 | } |
66e1707b | 2306 | |
7ec99d62 | 2307 | ret = mem_cgroup_do_charge(mem, gfp_mask, batch, oom_check); |
4b534334 KH |
2308 | switch (ret) { |
2309 | case CHARGE_OK: | |
2310 | break; | |
2311 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2312 | batch = nr_pages; |
f75ca962 KH |
2313 | css_put(&mem->css); |
2314 | mem = NULL; | |
2315 | goto again; | |
4b534334 | 2316 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
f75ca962 | 2317 | css_put(&mem->css); |
4b534334 KH |
2318 | goto nomem; |
2319 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 KH |
2320 | if (!oom) { |
2321 | css_put(&mem->css); | |
867578cb | 2322 | goto nomem; |
f75ca962 | 2323 | } |
4b534334 KH |
2324 | /* If oom, we never return -ENOMEM */ |
2325 | nr_oom_retries--; | |
2326 | break; | |
2327 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
f75ca962 | 2328 | css_put(&mem->css); |
867578cb | 2329 | goto bypass; |
66e1707b | 2330 | } |
4b534334 KH |
2331 | } while (ret != CHARGE_OK); |
2332 | ||
7ec99d62 JW |
2333 | if (batch > nr_pages) |
2334 | refill_stock(mem, batch - nr_pages); | |
f75ca962 | 2335 | css_put(&mem->css); |
0c3e73e8 | 2336 | done: |
f75ca962 | 2337 | *memcg = mem; |
7a81b88c KH |
2338 | return 0; |
2339 | nomem: | |
f75ca962 | 2340 | *memcg = NULL; |
7a81b88c | 2341 | return -ENOMEM; |
867578cb KH |
2342 | bypass: |
2343 | *memcg = NULL; | |
2344 | return 0; | |
7a81b88c | 2345 | } |
8a9f3ccd | 2346 | |
a3032a2c DN |
2347 | /* |
2348 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2349 | * This function is for that and do uncharge, put css's refcnt. | |
2350 | * gotten by try_charge(). | |
2351 | */ | |
854ffa8d | 2352 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
e7018b8d | 2353 | unsigned int nr_pages) |
a3032a2c DN |
2354 | { |
2355 | if (!mem_cgroup_is_root(mem)) { | |
e7018b8d JW |
2356 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2357 | ||
2358 | res_counter_uncharge(&mem->res, bytes); | |
a3032a2c | 2359 | if (do_swap_account) |
e7018b8d | 2360 | res_counter_uncharge(&mem->memsw, bytes); |
a3032a2c | 2361 | } |
854ffa8d DN |
2362 | } |
2363 | ||
a3b2d692 KH |
2364 | /* |
2365 | * A helper function to get mem_cgroup from ID. must be called under | |
2366 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2367 | * it's concern. (dropping refcnt from swap can be called against removed | |
2368 | * memcg.) | |
2369 | */ | |
2370 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2371 | { | |
2372 | struct cgroup_subsys_state *css; | |
2373 | ||
2374 | /* ID 0 is unused ID */ | |
2375 | if (!id) | |
2376 | return NULL; | |
2377 | css = css_lookup(&mem_cgroup_subsys, id); | |
2378 | if (!css) | |
2379 | return NULL; | |
2380 | return container_of(css, struct mem_cgroup, css); | |
2381 | } | |
2382 | ||
e42d9d5d | 2383 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2384 | { |
e42d9d5d | 2385 | struct mem_cgroup *mem = NULL; |
3c776e64 | 2386 | struct page_cgroup *pc; |
a3b2d692 | 2387 | unsigned short id; |
b5a84319 KH |
2388 | swp_entry_t ent; |
2389 | ||
3c776e64 DN |
2390 | VM_BUG_ON(!PageLocked(page)); |
2391 | ||
3c776e64 | 2392 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2393 | lock_page_cgroup(pc); |
a3b2d692 | 2394 | if (PageCgroupUsed(pc)) { |
3c776e64 | 2395 | mem = pc->mem_cgroup; |
a3b2d692 KH |
2396 | if (mem && !css_tryget(&mem->css)) |
2397 | mem = NULL; | |
e42d9d5d | 2398 | } else if (PageSwapCache(page)) { |
3c776e64 | 2399 | ent.val = page_private(page); |
a3b2d692 KH |
2400 | id = lookup_swap_cgroup(ent); |
2401 | rcu_read_lock(); | |
2402 | mem = mem_cgroup_lookup(id); | |
2403 | if (mem && !css_tryget(&mem->css)) | |
2404 | mem = NULL; | |
2405 | rcu_read_unlock(); | |
3c776e64 | 2406 | } |
c0bd3f63 | 2407 | unlock_page_cgroup(pc); |
b5a84319 KH |
2408 | return mem; |
2409 | } | |
2410 | ||
ca3e0214 | 2411 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, |
5564e88b | 2412 | struct page *page, |
7ec99d62 | 2413 | unsigned int nr_pages, |
ca3e0214 | 2414 | struct page_cgroup *pc, |
7ec99d62 | 2415 | enum charge_type ctype) |
7a81b88c | 2416 | { |
ca3e0214 KH |
2417 | lock_page_cgroup(pc); |
2418 | if (unlikely(PageCgroupUsed(pc))) { | |
2419 | unlock_page_cgroup(pc); | |
e7018b8d | 2420 | __mem_cgroup_cancel_charge(mem, nr_pages); |
ca3e0214 KH |
2421 | return; |
2422 | } | |
2423 | /* | |
2424 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2425 | * accessed by any other context at this point. | |
2426 | */ | |
8a9f3ccd | 2427 | pc->mem_cgroup = mem; |
261fb61a KH |
2428 | /* |
2429 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2430 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2431 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2432 | * before USED bit, we need memory barrier here. | |
2433 | * See mem_cgroup_add_lru_list(), etc. | |
2434 | */ | |
08e552c6 | 2435 | smp_wmb(); |
4b3bde4c BS |
2436 | switch (ctype) { |
2437 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
2438 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
2439 | SetPageCgroupCache(pc); | |
2440 | SetPageCgroupUsed(pc); | |
2441 | break; | |
2442 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2443 | ClearPageCgroupCache(pc); | |
2444 | SetPageCgroupUsed(pc); | |
2445 | break; | |
2446 | default: | |
2447 | break; | |
2448 | } | |
3be91277 | 2449 | |
ca3e0214 | 2450 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages); |
52d4b9ac | 2451 | unlock_page_cgroup(pc); |
430e4863 KH |
2452 | /* |
2453 | * "charge_statistics" updated event counter. Then, check it. | |
2454 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2455 | * if they exceeds softlimit. | |
2456 | */ | |
5564e88b | 2457 | memcg_check_events(mem, page); |
7a81b88c | 2458 | } |
66e1707b | 2459 | |
ca3e0214 KH |
2460 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2461 | ||
2462 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
2463 | (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION)) | |
2464 | /* | |
2465 | * Because tail pages are not marked as "used", set it. We're under | |
2466 | * zone->lru_lock, 'splitting on pmd' and compund_lock. | |
2467 | */ | |
2468 | void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail) | |
2469 | { | |
2470 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
2471 | struct page_cgroup *tail_pc = lookup_page_cgroup(tail); | |
2472 | unsigned long flags; | |
2473 | ||
3d37c4a9 KH |
2474 | if (mem_cgroup_disabled()) |
2475 | return; | |
ca3e0214 | 2476 | /* |
ece35ca8 | 2477 | * We have no races with charge/uncharge but will have races with |
ca3e0214 KH |
2478 | * page state accounting. |
2479 | */ | |
2480 | move_lock_page_cgroup(head_pc, &flags); | |
2481 | ||
2482 | tail_pc->mem_cgroup = head_pc->mem_cgroup; | |
2483 | smp_wmb(); /* see __commit_charge() */ | |
ece35ca8 KH |
2484 | if (PageCgroupAcctLRU(head_pc)) { |
2485 | enum lru_list lru; | |
2486 | struct mem_cgroup_per_zone *mz; | |
2487 | ||
2488 | /* | |
2489 | * LRU flags cannot be copied because we need to add tail | |
2490 | *.page to LRU by generic call and our hook will be called. | |
2491 | * We hold lru_lock, then, reduce counter directly. | |
2492 | */ | |
2493 | lru = page_lru(head); | |
97a6c37b | 2494 | mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head); |
ece35ca8 KH |
2495 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
2496 | } | |
ca3e0214 KH |
2497 | tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2498 | move_unlock_page_cgroup(head_pc, &flags); | |
2499 | } | |
2500 | #endif | |
2501 | ||
f817ed48 | 2502 | /** |
de3638d9 | 2503 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2504 | * @page: the page |
7ec99d62 | 2505 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2506 | * @pc: page_cgroup of the page. |
2507 | * @from: mem_cgroup which the page is moved from. | |
2508 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2509 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2510 | * |
2511 | * The caller must confirm following. | |
08e552c6 | 2512 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2513 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2514 | * |
854ffa8d | 2515 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
25985edc | 2516 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
854ffa8d DN |
2517 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
2518 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2519 | */ |
7ec99d62 JW |
2520 | static int mem_cgroup_move_account(struct page *page, |
2521 | unsigned int nr_pages, | |
2522 | struct page_cgroup *pc, | |
2523 | struct mem_cgroup *from, | |
2524 | struct mem_cgroup *to, | |
2525 | bool uncharge) | |
f817ed48 | 2526 | { |
de3638d9 JW |
2527 | unsigned long flags; |
2528 | int ret; | |
987eba66 | 2529 | |
f817ed48 | 2530 | VM_BUG_ON(from == to); |
5564e88b | 2531 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2532 | /* |
2533 | * The page is isolated from LRU. So, collapse function | |
2534 | * will not handle this page. But page splitting can happen. | |
2535 | * Do this check under compound_page_lock(). The caller should | |
2536 | * hold it. | |
2537 | */ | |
2538 | ret = -EBUSY; | |
7ec99d62 | 2539 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2540 | goto out; |
2541 | ||
2542 | lock_page_cgroup(pc); | |
2543 | ||
2544 | ret = -EINVAL; | |
2545 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2546 | goto unlock; | |
2547 | ||
2548 | move_lock_page_cgroup(pc, &flags); | |
f817ed48 | 2549 | |
8725d541 | 2550 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2551 | /* Update mapped_file data for mem_cgroup */ |
2552 | preempt_disable(); | |
2553 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2554 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2555 | preempt_enable(); | |
d69b042f | 2556 | } |
987eba66 | 2557 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages); |
854ffa8d DN |
2558 | if (uncharge) |
2559 | /* This is not "cancel", but cancel_charge does all we need. */ | |
e7018b8d | 2560 | __mem_cgroup_cancel_charge(from, nr_pages); |
d69b042f | 2561 | |
854ffa8d | 2562 | /* caller should have done css_get */ |
08e552c6 | 2563 | pc->mem_cgroup = to; |
987eba66 | 2564 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages); |
88703267 KH |
2565 | /* |
2566 | * We charges against "to" which may not have any tasks. Then, "to" | |
2567 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2568 | * this function is just force_empty() and move charge, so it's |
25985edc | 2569 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2570 | * status here. |
88703267 | 2571 | */ |
de3638d9 JW |
2572 | move_unlock_page_cgroup(pc, &flags); |
2573 | ret = 0; | |
2574 | unlock: | |
57f9fd7d | 2575 | unlock_page_cgroup(pc); |
d2265e6f KH |
2576 | /* |
2577 | * check events | |
2578 | */ | |
5564e88b JW |
2579 | memcg_check_events(to, page); |
2580 | memcg_check_events(from, page); | |
de3638d9 | 2581 | out: |
f817ed48 KH |
2582 | return ret; |
2583 | } | |
2584 | ||
2585 | /* | |
2586 | * move charges to its parent. | |
2587 | */ | |
2588 | ||
5564e88b JW |
2589 | static int mem_cgroup_move_parent(struct page *page, |
2590 | struct page_cgroup *pc, | |
f817ed48 KH |
2591 | struct mem_cgroup *child, |
2592 | gfp_t gfp_mask) | |
2593 | { | |
2594 | struct cgroup *cg = child->css.cgroup; | |
2595 | struct cgroup *pcg = cg->parent; | |
2596 | struct mem_cgroup *parent; | |
7ec99d62 | 2597 | unsigned int nr_pages; |
4be4489f | 2598 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2599 | int ret; |
2600 | ||
2601 | /* Is ROOT ? */ | |
2602 | if (!pcg) | |
2603 | return -EINVAL; | |
2604 | ||
57f9fd7d DN |
2605 | ret = -EBUSY; |
2606 | if (!get_page_unless_zero(page)) | |
2607 | goto out; | |
2608 | if (isolate_lru_page(page)) | |
2609 | goto put; | |
52dbb905 | 2610 | |
7ec99d62 | 2611 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2612 | |
f817ed48 | 2613 | parent = mem_cgroup_from_cont(pcg); |
7ec99d62 | 2614 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
a636b327 | 2615 | if (ret || !parent) |
57f9fd7d | 2616 | goto put_back; |
f817ed48 | 2617 | |
7ec99d62 | 2618 | if (nr_pages > 1) |
987eba66 KH |
2619 | flags = compound_lock_irqsave(page); |
2620 | ||
7ec99d62 | 2621 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
854ffa8d | 2622 | if (ret) |
7ec99d62 | 2623 | __mem_cgroup_cancel_charge(parent, nr_pages); |
8dba474f | 2624 | |
7ec99d62 | 2625 | if (nr_pages > 1) |
987eba66 | 2626 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2627 | put_back: |
08e552c6 | 2628 | putback_lru_page(page); |
57f9fd7d | 2629 | put: |
40d58138 | 2630 | put_page(page); |
57f9fd7d | 2631 | out: |
f817ed48 KH |
2632 | return ret; |
2633 | } | |
2634 | ||
7a81b88c KH |
2635 | /* |
2636 | * Charge the memory controller for page usage. | |
2637 | * Return | |
2638 | * 0 if the charge was successful | |
2639 | * < 0 if the cgroup is over its limit | |
2640 | */ | |
2641 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2642 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2643 | { |
73045c47 | 2644 | struct mem_cgroup *mem = NULL; |
7ec99d62 | 2645 | unsigned int nr_pages = 1; |
7a81b88c | 2646 | struct page_cgroup *pc; |
8493ae43 | 2647 | bool oom = true; |
7a81b88c | 2648 | int ret; |
ec168510 | 2649 | |
37c2ac78 | 2650 | if (PageTransHuge(page)) { |
7ec99d62 | 2651 | nr_pages <<= compound_order(page); |
37c2ac78 | 2652 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2653 | /* |
2654 | * Never OOM-kill a process for a huge page. The | |
2655 | * fault handler will fall back to regular pages. | |
2656 | */ | |
2657 | oom = false; | |
37c2ac78 | 2658 | } |
7a81b88c KH |
2659 | |
2660 | pc = lookup_page_cgroup(page); | |
af4a6621 | 2661 | BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */ |
7a81b88c | 2662 | |
7ec99d62 | 2663 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &mem, oom); |
a636b327 | 2664 | if (ret || !mem) |
7a81b88c KH |
2665 | return ret; |
2666 | ||
7ec99d62 | 2667 | __mem_cgroup_commit_charge(mem, page, nr_pages, pc, ctype); |
8a9f3ccd | 2668 | return 0; |
8a9f3ccd BS |
2669 | } |
2670 | ||
7a81b88c KH |
2671 | int mem_cgroup_newpage_charge(struct page *page, |
2672 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2673 | { |
f8d66542 | 2674 | if (mem_cgroup_disabled()) |
cede86ac | 2675 | return 0; |
69029cd5 KH |
2676 | /* |
2677 | * If already mapped, we don't have to account. | |
2678 | * If page cache, page->mapping has address_space. | |
2679 | * But page->mapping may have out-of-use anon_vma pointer, | |
2680 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
2681 | * is NULL. | |
2682 | */ | |
2683 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2684 | return 0; | |
2685 | if (unlikely(!mm)) | |
2686 | mm = &init_mm; | |
217bc319 | 2687 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
73045c47 | 2688 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2689 | } |
2690 | ||
83aae4c7 DN |
2691 | static void |
2692 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2693 | enum charge_type ctype); | |
2694 | ||
5a6475a4 KH |
2695 | static void |
2696 | __mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *mem, | |
2697 | enum charge_type ctype) | |
2698 | { | |
2699 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2700 | /* | |
2701 | * In some case, SwapCache, FUSE(splice_buf->radixtree), the page | |
2702 | * is already on LRU. It means the page may on some other page_cgroup's | |
2703 | * LRU. Take care of it. | |
2704 | */ | |
2705 | mem_cgroup_lru_del_before_commit(page); | |
2706 | __mem_cgroup_commit_charge(mem, page, 1, pc, ctype); | |
2707 | mem_cgroup_lru_add_after_commit(page); | |
2708 | return; | |
2709 | } | |
2710 | ||
e1a1cd59 BS |
2711 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2712 | gfp_t gfp_mask) | |
8697d331 | 2713 | { |
5a6475a4 | 2714 | struct mem_cgroup *mem = NULL; |
b5a84319 KH |
2715 | int ret; |
2716 | ||
f8d66542 | 2717 | if (mem_cgroup_disabled()) |
cede86ac | 2718 | return 0; |
52d4b9ac KH |
2719 | if (PageCompound(page)) |
2720 | return 0; | |
accf163e KH |
2721 | /* |
2722 | * Corner case handling. This is called from add_to_page_cache() | |
2723 | * in usual. But some FS (shmem) precharges this page before calling it | |
2724 | * and call add_to_page_cache() with GFP_NOWAIT. | |
2725 | * | |
2726 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
2727 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
2728 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
2729 | * And when the page is SwapCache, it should take swap information |
2730 | * into account. This is under lock_page() now. | |
accf163e KH |
2731 | */ |
2732 | if (!(gfp_mask & __GFP_WAIT)) { | |
2733 | struct page_cgroup *pc; | |
2734 | ||
52d4b9ac KH |
2735 | pc = lookup_page_cgroup(page); |
2736 | if (!pc) | |
2737 | return 0; | |
2738 | lock_page_cgroup(pc); | |
2739 | if (PageCgroupUsed(pc)) { | |
2740 | unlock_page_cgroup(pc); | |
accf163e KH |
2741 | return 0; |
2742 | } | |
52d4b9ac | 2743 | unlock_page_cgroup(pc); |
accf163e KH |
2744 | } |
2745 | ||
73045c47 | 2746 | if (unlikely(!mm)) |
8697d331 | 2747 | mm = &init_mm; |
accf163e | 2748 | |
5a6475a4 KH |
2749 | if (page_is_file_cache(page)) { |
2750 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &mem, true); | |
2751 | if (ret || !mem) | |
2752 | return ret; | |
b5a84319 | 2753 | |
5a6475a4 KH |
2754 | /* |
2755 | * FUSE reuses pages without going through the final | |
2756 | * put that would remove them from the LRU list, make | |
2757 | * sure that they get relinked properly. | |
2758 | */ | |
2759 | __mem_cgroup_commit_charge_lrucare(page, mem, | |
2760 | MEM_CGROUP_CHARGE_TYPE_CACHE); | |
2761 | return ret; | |
2762 | } | |
83aae4c7 DN |
2763 | /* shmem */ |
2764 | if (PageSwapCache(page)) { | |
2765 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
2766 | if (!ret) | |
2767 | __mem_cgroup_commit_charge_swapin(page, mem, | |
2768 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
2769 | } else | |
2770 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2771 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
b5a84319 | 2772 | |
b5a84319 | 2773 | return ret; |
e8589cc1 KH |
2774 | } |
2775 | ||
54595fe2 KH |
2776 | /* |
2777 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2778 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2779 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2780 | * "commit()" or removed by "cancel()" |
2781 | */ | |
8c7c6e34 KH |
2782 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2783 | struct page *page, | |
2784 | gfp_t mask, struct mem_cgroup **ptr) | |
2785 | { | |
2786 | struct mem_cgroup *mem; | |
54595fe2 | 2787 | int ret; |
8c7c6e34 | 2788 | |
56039efa KH |
2789 | *ptr = NULL; |
2790 | ||
f8d66542 | 2791 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2792 | return 0; |
2793 | ||
2794 | if (!do_swap_account) | |
2795 | goto charge_cur_mm; | |
8c7c6e34 KH |
2796 | /* |
2797 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2798 | * the pte, and even removed page from swap cache: in those cases |
2799 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2800 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2801 | */ |
2802 | if (!PageSwapCache(page)) | |
407f9c8b | 2803 | goto charge_cur_mm; |
e42d9d5d | 2804 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
2805 | if (!mem) |
2806 | goto charge_cur_mm; | |
8c7c6e34 | 2807 | *ptr = mem; |
7ec99d62 | 2808 | ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true); |
54595fe2 KH |
2809 | css_put(&mem->css); |
2810 | return ret; | |
8c7c6e34 KH |
2811 | charge_cur_mm: |
2812 | if (unlikely(!mm)) | |
2813 | mm = &init_mm; | |
7ec99d62 | 2814 | return __mem_cgroup_try_charge(mm, mask, 1, ptr, true); |
8c7c6e34 KH |
2815 | } |
2816 | ||
83aae4c7 DN |
2817 | static void |
2818 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2819 | enum charge_type ctype) | |
7a81b88c | 2820 | { |
f8d66542 | 2821 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2822 | return; |
2823 | if (!ptr) | |
2824 | return; | |
88703267 | 2825 | cgroup_exclude_rmdir(&ptr->css); |
5a6475a4 KH |
2826 | |
2827 | __mem_cgroup_commit_charge_lrucare(page, ptr, ctype); | |
8c7c6e34 KH |
2828 | /* |
2829 | * Now swap is on-memory. This means this page may be | |
2830 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2831 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2832 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2833 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2834 | */ |
03f3c433 | 2835 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2836 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2837 | unsigned short id; |
8c7c6e34 | 2838 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2839 | |
2840 | id = swap_cgroup_record(ent, 0); | |
2841 | rcu_read_lock(); | |
2842 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2843 | if (memcg) { |
a3b2d692 KH |
2844 | /* |
2845 | * This recorded memcg can be obsolete one. So, avoid | |
2846 | * calling css_tryget | |
2847 | */ | |
0c3e73e8 | 2848 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2849 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2850 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2851 | mem_cgroup_put(memcg); |
2852 | } | |
a3b2d692 | 2853 | rcu_read_unlock(); |
8c7c6e34 | 2854 | } |
88703267 KH |
2855 | /* |
2856 | * At swapin, we may charge account against cgroup which has no tasks. | |
2857 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2858 | * In that case, we need to call pre_destroy() again. check it here. | |
2859 | */ | |
2860 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2861 | } |
2862 | ||
83aae4c7 DN |
2863 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2864 | { | |
2865 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2866 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2867 | } | |
2868 | ||
7a81b88c KH |
2869 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2870 | { | |
f8d66542 | 2871 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2872 | return; |
2873 | if (!mem) | |
2874 | return; | |
e7018b8d | 2875 | __mem_cgroup_cancel_charge(mem, 1); |
7a81b88c KH |
2876 | } |
2877 | ||
7ec99d62 JW |
2878 | static void mem_cgroup_do_uncharge(struct mem_cgroup *mem, |
2879 | unsigned int nr_pages, | |
2880 | const enum charge_type ctype) | |
569b846d KH |
2881 | { |
2882 | struct memcg_batch_info *batch = NULL; | |
2883 | bool uncharge_memsw = true; | |
7ec99d62 | 2884 | |
569b846d KH |
2885 | /* If swapout, usage of swap doesn't decrease */ |
2886 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2887 | uncharge_memsw = false; | |
569b846d KH |
2888 | |
2889 | batch = ¤t->memcg_batch; | |
2890 | /* | |
2891 | * In usual, we do css_get() when we remember memcg pointer. | |
2892 | * But in this case, we keep res->usage until end of a series of | |
2893 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2894 | */ | |
2895 | if (!batch->memcg) | |
2896 | batch->memcg = mem; | |
3c11ecf4 KH |
2897 | /* |
2898 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2899 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2900 | * the same cgroup and we have chance to coalesce uncharges. |
2901 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2902 | * because we want to do uncharge as soon as possible. | |
2903 | */ | |
2904 | ||
2905 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2906 | goto direct_uncharge; | |
2907 | ||
7ec99d62 | 2908 | if (nr_pages > 1) |
ec168510 AA |
2909 | goto direct_uncharge; |
2910 | ||
569b846d KH |
2911 | /* |
2912 | * In typical case, batch->memcg == mem. This means we can | |
2913 | * merge a series of uncharges to an uncharge of res_counter. | |
2914 | * If not, we uncharge res_counter ony by one. | |
2915 | */ | |
2916 | if (batch->memcg != mem) | |
2917 | goto direct_uncharge; | |
2918 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2919 | batch->nr_pages++; |
569b846d | 2920 | if (uncharge_memsw) |
7ffd4ca7 | 2921 | batch->memsw_nr_pages++; |
569b846d KH |
2922 | return; |
2923 | direct_uncharge: | |
7ec99d62 | 2924 | res_counter_uncharge(&mem->res, nr_pages * PAGE_SIZE); |
569b846d | 2925 | if (uncharge_memsw) |
7ec99d62 | 2926 | res_counter_uncharge(&mem->memsw, nr_pages * PAGE_SIZE); |
3c11ecf4 KH |
2927 | if (unlikely(batch->memcg != mem)) |
2928 | memcg_oom_recover(mem); | |
569b846d KH |
2929 | return; |
2930 | } | |
7a81b88c | 2931 | |
8a9f3ccd | 2932 | /* |
69029cd5 | 2933 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2934 | */ |
8c7c6e34 | 2935 | static struct mem_cgroup * |
69029cd5 | 2936 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2937 | { |
8c7c6e34 | 2938 | struct mem_cgroup *mem = NULL; |
7ec99d62 JW |
2939 | unsigned int nr_pages = 1; |
2940 | struct page_cgroup *pc; | |
8a9f3ccd | 2941 | |
f8d66542 | 2942 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2943 | return NULL; |
4077960e | 2944 | |
d13d1443 | 2945 | if (PageSwapCache(page)) |
8c7c6e34 | 2946 | return NULL; |
d13d1443 | 2947 | |
37c2ac78 | 2948 | if (PageTransHuge(page)) { |
7ec99d62 | 2949 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2950 | VM_BUG_ON(!PageTransHuge(page)); |
2951 | } | |
8697d331 | 2952 | /* |
3c541e14 | 2953 | * Check if our page_cgroup is valid |
8697d331 | 2954 | */ |
52d4b9ac KH |
2955 | pc = lookup_page_cgroup(page); |
2956 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2957 | return NULL; |
b9c565d5 | 2958 | |
52d4b9ac | 2959 | lock_page_cgroup(pc); |
d13d1443 | 2960 | |
8c7c6e34 KH |
2961 | mem = pc->mem_cgroup; |
2962 | ||
d13d1443 KH |
2963 | if (!PageCgroupUsed(pc)) |
2964 | goto unlock_out; | |
2965 | ||
2966 | switch (ctype) { | |
2967 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2968 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2969 | /* See mem_cgroup_prepare_migration() */ |
2970 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2971 | goto unlock_out; |
2972 | break; | |
2973 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2974 | if (!PageAnon(page)) { /* Shared memory */ | |
2975 | if (page->mapping && !page_is_file_cache(page)) | |
2976 | goto unlock_out; | |
2977 | } else if (page_mapped(page)) /* Anon */ | |
2978 | goto unlock_out; | |
2979 | break; | |
2980 | default: | |
2981 | break; | |
52d4b9ac | 2982 | } |
d13d1443 | 2983 | |
7ec99d62 | 2984 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -nr_pages); |
04046e1a | 2985 | |
52d4b9ac | 2986 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2987 | /* |
2988 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2989 | * freed from LRU. This is safe because uncharged page is expected not | |
2990 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2991 | * special functions. | |
2992 | */ | |
b9c565d5 | 2993 | |
52d4b9ac | 2994 | unlock_page_cgroup(pc); |
f75ca962 KH |
2995 | /* |
2996 | * even after unlock, we have mem->res.usage here and this memcg | |
2997 | * will never be freed. | |
2998 | */ | |
d2265e6f | 2999 | memcg_check_events(mem, page); |
f75ca962 KH |
3000 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
3001 | mem_cgroup_swap_statistics(mem, true); | |
3002 | mem_cgroup_get(mem); | |
3003 | } | |
3004 | if (!mem_cgroup_is_root(mem)) | |
7ec99d62 | 3005 | mem_cgroup_do_uncharge(mem, nr_pages, ctype); |
6d12e2d8 | 3006 | |
8c7c6e34 | 3007 | return mem; |
d13d1443 KH |
3008 | |
3009 | unlock_out: | |
3010 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3011 | return NULL; |
3c541e14 BS |
3012 | } |
3013 | ||
69029cd5 KH |
3014 | void mem_cgroup_uncharge_page(struct page *page) |
3015 | { | |
52d4b9ac KH |
3016 | /* early check. */ |
3017 | if (page_mapped(page)) | |
3018 | return; | |
3019 | if (page->mapping && !PageAnon(page)) | |
3020 | return; | |
69029cd5 KH |
3021 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
3022 | } | |
3023 | ||
3024 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3025 | { | |
3026 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3027 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
3028 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
3029 | } | |
3030 | ||
569b846d KH |
3031 | /* |
3032 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3033 | * In that cases, pages are freed continuously and we can expect pages | |
3034 | * are in the same memcg. All these calls itself limits the number of | |
3035 | * pages freed at once, then uncharge_start/end() is called properly. | |
3036 | * This may be called prural(2) times in a context, | |
3037 | */ | |
3038 | ||
3039 | void mem_cgroup_uncharge_start(void) | |
3040 | { | |
3041 | current->memcg_batch.do_batch++; | |
3042 | /* We can do nest. */ | |
3043 | if (current->memcg_batch.do_batch == 1) { | |
3044 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3045 | current->memcg_batch.nr_pages = 0; |
3046 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3047 | } |
3048 | } | |
3049 | ||
3050 | void mem_cgroup_uncharge_end(void) | |
3051 | { | |
3052 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3053 | ||
3054 | if (!batch->do_batch) | |
3055 | return; | |
3056 | ||
3057 | batch->do_batch--; | |
3058 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3059 | return; | |
3060 | ||
3061 | if (!batch->memcg) | |
3062 | return; | |
3063 | /* | |
3064 | * This "batch->memcg" is valid without any css_get/put etc... | |
3065 | * bacause we hide charges behind us. | |
3066 | */ | |
7ffd4ca7 JW |
3067 | if (batch->nr_pages) |
3068 | res_counter_uncharge(&batch->memcg->res, | |
3069 | batch->nr_pages * PAGE_SIZE); | |
3070 | if (batch->memsw_nr_pages) | |
3071 | res_counter_uncharge(&batch->memcg->memsw, | |
3072 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3073 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3074 | /* forget this pointer (for sanity check) */ |
3075 | batch->memcg = NULL; | |
3076 | } | |
3077 | ||
e767e056 | 3078 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3079 | /* |
e767e056 | 3080 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3081 | * memcg information is recorded to swap_cgroup of "ent" |
3082 | */ | |
8a9478ca KH |
3083 | void |
3084 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3085 | { |
3086 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3087 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3088 | ||
3089 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3090 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3091 | ||
3092 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3093 | |
f75ca962 KH |
3094 | /* |
3095 | * record memcg information, if swapout && memcg != NULL, | |
3096 | * mem_cgroup_get() was called in uncharge(). | |
3097 | */ | |
3098 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3099 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3100 | } |
e767e056 | 3101 | #endif |
8c7c6e34 KH |
3102 | |
3103 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3104 | /* | |
3105 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3106 | * uncharge "memsw" account. | |
3107 | */ | |
3108 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3109 | { |
8c7c6e34 | 3110 | struct mem_cgroup *memcg; |
a3b2d692 | 3111 | unsigned short id; |
8c7c6e34 KH |
3112 | |
3113 | if (!do_swap_account) | |
3114 | return; | |
3115 | ||
a3b2d692 KH |
3116 | id = swap_cgroup_record(ent, 0); |
3117 | rcu_read_lock(); | |
3118 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3119 | if (memcg) { |
a3b2d692 KH |
3120 | /* |
3121 | * We uncharge this because swap is freed. | |
3122 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3123 | */ | |
0c3e73e8 | 3124 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3125 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3126 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3127 | mem_cgroup_put(memcg); |
3128 | } | |
a3b2d692 | 3129 | rcu_read_unlock(); |
d13d1443 | 3130 | } |
02491447 DN |
3131 | |
3132 | /** | |
3133 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3134 | * @entry: swap entry to be moved | |
3135 | * @from: mem_cgroup which the entry is moved from | |
3136 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 3137 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
3138 | * |
3139 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3140 | * as the mem_cgroup's id of @from. | |
3141 | * | |
3142 | * Returns 0 on success, -EINVAL on failure. | |
3143 | * | |
3144 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3145 | * both res and memsw, and called css_get(). | |
3146 | */ | |
3147 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3148 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3149 | { |
3150 | unsigned short old_id, new_id; | |
3151 | ||
3152 | old_id = css_id(&from->css); | |
3153 | new_id = css_id(&to->css); | |
3154 | ||
3155 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3156 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3157 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3158 | /* |
483c30b5 DN |
3159 | * This function is only called from task migration context now. |
3160 | * It postpones res_counter and refcount handling till the end | |
3161 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3162 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3163 | * because if the process that has been moved to @to does | |
3164 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3165 | */ |
02491447 | 3166 | mem_cgroup_get(to); |
483c30b5 DN |
3167 | if (need_fixup) { |
3168 | if (!mem_cgroup_is_root(from)) | |
3169 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
3170 | mem_cgroup_put(from); | |
3171 | /* | |
3172 | * we charged both to->res and to->memsw, so we should | |
3173 | * uncharge to->res. | |
3174 | */ | |
3175 | if (!mem_cgroup_is_root(to)) | |
3176 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 3177 | } |
02491447 DN |
3178 | return 0; |
3179 | } | |
3180 | return -EINVAL; | |
3181 | } | |
3182 | #else | |
3183 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3184 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3185 | { |
3186 | return -EINVAL; | |
3187 | } | |
8c7c6e34 | 3188 | #endif |
d13d1443 | 3189 | |
ae41be37 | 3190 | /* |
01b1ae63 KH |
3191 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3192 | * page belongs to. | |
ae41be37 | 3193 | */ |
ac39cf8c | 3194 | int mem_cgroup_prepare_migration(struct page *page, |
ef6a3c63 | 3195 | struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask) |
ae41be37 | 3196 | { |
e8589cc1 | 3197 | struct mem_cgroup *mem = NULL; |
7ec99d62 | 3198 | struct page_cgroup *pc; |
ac39cf8c | 3199 | enum charge_type ctype; |
e8589cc1 | 3200 | int ret = 0; |
8869b8f6 | 3201 | |
56039efa KH |
3202 | *ptr = NULL; |
3203 | ||
ec168510 | 3204 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3205 | if (mem_cgroup_disabled()) |
4077960e BS |
3206 | return 0; |
3207 | ||
52d4b9ac KH |
3208 | pc = lookup_page_cgroup(page); |
3209 | lock_page_cgroup(pc); | |
3210 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
3211 | mem = pc->mem_cgroup; |
3212 | css_get(&mem->css); | |
ac39cf8c | 3213 | /* |
3214 | * At migrating an anonymous page, its mapcount goes down | |
3215 | * to 0 and uncharge() will be called. But, even if it's fully | |
3216 | * unmapped, migration may fail and this page has to be | |
3217 | * charged again. We set MIGRATION flag here and delay uncharge | |
3218 | * until end_migration() is called | |
3219 | * | |
3220 | * Corner Case Thinking | |
3221 | * A) | |
3222 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3223 | * while migration was ongoing. | |
3224 | * If unmap finds the old page, uncharge() of it will be delayed | |
3225 | * until end_migration(). If unmap finds a new page, it's | |
3226 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3227 | * finds swap_migration_entry, the new page will not be mapped | |
3228 | * and end_migration() will find it(mapcount==0). | |
3229 | * | |
3230 | * B) | |
3231 | * When the old page was mapped but migraion fails, the kernel | |
3232 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3233 | * if mapcount goes down to 0. We can do remap successfully | |
3234 | * without charging it again. | |
3235 | * | |
3236 | * C) | |
3237 | * The "old" page is under lock_page() until the end of | |
3238 | * migration, so, the old page itself will not be swapped-out. | |
3239 | * If the new page is swapped out before end_migraton, our | |
3240 | * hook to usual swap-out path will catch the event. | |
3241 | */ | |
3242 | if (PageAnon(page)) | |
3243 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3244 | } |
52d4b9ac | 3245 | unlock_page_cgroup(pc); |
ac39cf8c | 3246 | /* |
3247 | * If the page is not charged at this point, | |
3248 | * we return here. | |
3249 | */ | |
3250 | if (!mem) | |
3251 | return 0; | |
01b1ae63 | 3252 | |
93d5c9be | 3253 | *ptr = mem; |
7ec99d62 | 3254 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false); |
ac39cf8c | 3255 | css_put(&mem->css);/* drop extra refcnt */ |
3256 | if (ret || *ptr == NULL) { | |
3257 | if (PageAnon(page)) { | |
3258 | lock_page_cgroup(pc); | |
3259 | ClearPageCgroupMigration(pc); | |
3260 | unlock_page_cgroup(pc); | |
3261 | /* | |
3262 | * The old page may be fully unmapped while we kept it. | |
3263 | */ | |
3264 | mem_cgroup_uncharge_page(page); | |
3265 | } | |
3266 | return -ENOMEM; | |
e8589cc1 | 3267 | } |
ac39cf8c | 3268 | /* |
3269 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3270 | * is called before end_migration, we can catch all events on this new | |
3271 | * page. In the case new page is migrated but not remapped, new page's | |
3272 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3273 | */ | |
3274 | pc = lookup_page_cgroup(newpage); | |
3275 | if (PageAnon(page)) | |
3276 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3277 | else if (page_is_file_cache(page)) | |
3278 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3279 | else | |
3280 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
7ec99d62 | 3281 | __mem_cgroup_commit_charge(mem, page, 1, pc, ctype); |
e8589cc1 | 3282 | return ret; |
ae41be37 | 3283 | } |
8869b8f6 | 3284 | |
69029cd5 | 3285 | /* remove redundant charge if migration failed*/ |
01b1ae63 | 3286 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
50de1dd9 | 3287 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3288 | { |
ac39cf8c | 3289 | struct page *used, *unused; |
01b1ae63 | 3290 | struct page_cgroup *pc; |
01b1ae63 KH |
3291 | |
3292 | if (!mem) | |
3293 | return; | |
ac39cf8c | 3294 | /* blocks rmdir() */ |
88703267 | 3295 | cgroup_exclude_rmdir(&mem->css); |
50de1dd9 | 3296 | if (!migration_ok) { |
ac39cf8c | 3297 | used = oldpage; |
3298 | unused = newpage; | |
01b1ae63 | 3299 | } else { |
ac39cf8c | 3300 | used = newpage; |
01b1ae63 KH |
3301 | unused = oldpage; |
3302 | } | |
69029cd5 | 3303 | /* |
ac39cf8c | 3304 | * We disallowed uncharge of pages under migration because mapcount |
3305 | * of the page goes down to zero, temporarly. | |
3306 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3307 | */ |
ac39cf8c | 3308 | pc = lookup_page_cgroup(oldpage); |
3309 | lock_page_cgroup(pc); | |
3310 | ClearPageCgroupMigration(pc); | |
3311 | unlock_page_cgroup(pc); | |
01b1ae63 | 3312 | |
ac39cf8c | 3313 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
3314 | ||
01b1ae63 | 3315 | /* |
ac39cf8c | 3316 | * If a page is a file cache, radix-tree replacement is very atomic |
3317 | * and we can skip this check. When it was an Anon page, its mapcount | |
3318 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3319 | * uncharged yet. There are several case but page->mapcount check | |
3320 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3321 | * check. (see prepare_charge() also) | |
69029cd5 | 3322 | */ |
ac39cf8c | 3323 | if (PageAnon(used)) |
3324 | mem_cgroup_uncharge_page(used); | |
88703267 | 3325 | /* |
ac39cf8c | 3326 | * At migration, we may charge account against cgroup which has no |
3327 | * tasks. | |
88703267 KH |
3328 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3329 | * In that case, we need to call pre_destroy() again. check it here. | |
3330 | */ | |
3331 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 3332 | } |
78fb7466 | 3333 | |
c9b0ed51 | 3334 | /* |
ae3abae6 DN |
3335 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
3336 | * Calling hierarchical_reclaim is not enough because we should update | |
3337 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
3338 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
3339 | * not from the memcg which this page would be charged to. | |
3340 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 3341 | */ |
ae3abae6 | 3342 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
3343 | struct mm_struct *mm, |
3344 | gfp_t gfp_mask) | |
c9b0ed51 | 3345 | { |
56039efa | 3346 | struct mem_cgroup *mem; |
ae3abae6 | 3347 | int ret; |
c9b0ed51 | 3348 | |
f8d66542 | 3349 | if (mem_cgroup_disabled()) |
cede86ac | 3350 | return 0; |
c9b0ed51 | 3351 | |
ae3abae6 DN |
3352 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
3353 | if (!ret) | |
3354 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 3355 | |
ae3abae6 | 3356 | return ret; |
c9b0ed51 KH |
3357 | } |
3358 | ||
f212ad7c DN |
3359 | #ifdef CONFIG_DEBUG_VM |
3360 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3361 | { | |
3362 | struct page_cgroup *pc; | |
3363 | ||
3364 | pc = lookup_page_cgroup(page); | |
3365 | if (likely(pc) && PageCgroupUsed(pc)) | |
3366 | return pc; | |
3367 | return NULL; | |
3368 | } | |
3369 | ||
3370 | bool mem_cgroup_bad_page_check(struct page *page) | |
3371 | { | |
3372 | if (mem_cgroup_disabled()) | |
3373 | return false; | |
3374 | ||
3375 | return lookup_page_cgroup_used(page) != NULL; | |
3376 | } | |
3377 | ||
3378 | void mem_cgroup_print_bad_page(struct page *page) | |
3379 | { | |
3380 | struct page_cgroup *pc; | |
3381 | ||
3382 | pc = lookup_page_cgroup_used(page); | |
3383 | if (pc) { | |
3384 | int ret = -1; | |
3385 | char *path; | |
3386 | ||
3387 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p", | |
3388 | pc, pc->flags, pc->mem_cgroup); | |
3389 | ||
3390 | path = kmalloc(PATH_MAX, GFP_KERNEL); | |
3391 | if (path) { | |
3392 | rcu_read_lock(); | |
3393 | ret = cgroup_path(pc->mem_cgroup->css.cgroup, | |
3394 | path, PATH_MAX); | |
3395 | rcu_read_unlock(); | |
3396 | } | |
3397 | ||
3398 | printk(KERN_CONT "(%s)\n", | |
3399 | (ret < 0) ? "cannot get the path" : path); | |
3400 | kfree(path); | |
3401 | } | |
3402 | } | |
3403 | #endif | |
3404 | ||
8c7c6e34 KH |
3405 | static DEFINE_MUTEX(set_limit_mutex); |
3406 | ||
d38d2a75 | 3407 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3408 | unsigned long long val) |
628f4235 | 3409 | { |
81d39c20 | 3410 | int retry_count; |
3c11ecf4 | 3411 | u64 memswlimit, memlimit; |
628f4235 | 3412 | int ret = 0; |
81d39c20 KH |
3413 | int children = mem_cgroup_count_children(memcg); |
3414 | u64 curusage, oldusage; | |
3c11ecf4 | 3415 | int enlarge; |
81d39c20 KH |
3416 | |
3417 | /* | |
3418 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3419 | * is depends on callers. We set our retry-count to be function | |
3420 | * of # of children which we should visit in this loop. | |
3421 | */ | |
3422 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3423 | ||
3424 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3425 | |
3c11ecf4 | 3426 | enlarge = 0; |
8c7c6e34 | 3427 | while (retry_count) { |
628f4235 KH |
3428 | if (signal_pending(current)) { |
3429 | ret = -EINTR; | |
3430 | break; | |
3431 | } | |
8c7c6e34 KH |
3432 | /* |
3433 | * Rather than hide all in some function, I do this in | |
3434 | * open coded manner. You see what this really does. | |
3435 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3436 | */ | |
3437 | mutex_lock(&set_limit_mutex); | |
3438 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3439 | if (memswlimit < val) { | |
3440 | ret = -EINVAL; | |
3441 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3442 | break; |
3443 | } | |
3c11ecf4 KH |
3444 | |
3445 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3446 | if (memlimit < val) | |
3447 | enlarge = 1; | |
3448 | ||
8c7c6e34 | 3449 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3450 | if (!ret) { |
3451 | if (memswlimit == val) | |
3452 | memcg->memsw_is_minimum = true; | |
3453 | else | |
3454 | memcg->memsw_is_minimum = false; | |
3455 | } | |
8c7c6e34 KH |
3456 | mutex_unlock(&set_limit_mutex); |
3457 | ||
3458 | if (!ret) | |
3459 | break; | |
3460 | ||
aa20d489 | 3461 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
0ae5e89c YH |
3462 | MEM_CGROUP_RECLAIM_SHRINK, |
3463 | NULL); | |
81d39c20 KH |
3464 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3465 | /* Usage is reduced ? */ | |
3466 | if (curusage >= oldusage) | |
3467 | retry_count--; | |
3468 | else | |
3469 | oldusage = curusage; | |
8c7c6e34 | 3470 | } |
3c11ecf4 KH |
3471 | if (!ret && enlarge) |
3472 | memcg_oom_recover(memcg); | |
14797e23 | 3473 | |
8c7c6e34 KH |
3474 | return ret; |
3475 | } | |
3476 | ||
338c8431 LZ |
3477 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3478 | unsigned long long val) | |
8c7c6e34 | 3479 | { |
81d39c20 | 3480 | int retry_count; |
3c11ecf4 | 3481 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3482 | int children = mem_cgroup_count_children(memcg); |
3483 | int ret = -EBUSY; | |
3c11ecf4 | 3484 | int enlarge = 0; |
8c7c6e34 | 3485 | |
81d39c20 KH |
3486 | /* see mem_cgroup_resize_res_limit */ |
3487 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3488 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3489 | while (retry_count) { |
3490 | if (signal_pending(current)) { | |
3491 | ret = -EINTR; | |
3492 | break; | |
3493 | } | |
3494 | /* | |
3495 | * Rather than hide all in some function, I do this in | |
3496 | * open coded manner. You see what this really does. | |
3497 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3498 | */ | |
3499 | mutex_lock(&set_limit_mutex); | |
3500 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3501 | if (memlimit > val) { | |
3502 | ret = -EINVAL; | |
3503 | mutex_unlock(&set_limit_mutex); | |
3504 | break; | |
3505 | } | |
3c11ecf4 KH |
3506 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3507 | if (memswlimit < val) | |
3508 | enlarge = 1; | |
8c7c6e34 | 3509 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3510 | if (!ret) { |
3511 | if (memlimit == val) | |
3512 | memcg->memsw_is_minimum = true; | |
3513 | else | |
3514 | memcg->memsw_is_minimum = false; | |
3515 | } | |
8c7c6e34 KH |
3516 | mutex_unlock(&set_limit_mutex); |
3517 | ||
3518 | if (!ret) | |
3519 | break; | |
3520 | ||
4e416953 | 3521 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 | 3522 | MEM_CGROUP_RECLAIM_NOSWAP | |
0ae5e89c YH |
3523 | MEM_CGROUP_RECLAIM_SHRINK, |
3524 | NULL); | |
8c7c6e34 | 3525 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3526 | /* Usage is reduced ? */ |
8c7c6e34 | 3527 | if (curusage >= oldusage) |
628f4235 | 3528 | retry_count--; |
81d39c20 KH |
3529 | else |
3530 | oldusage = curusage; | |
628f4235 | 3531 | } |
3c11ecf4 KH |
3532 | if (!ret && enlarge) |
3533 | memcg_oom_recover(memcg); | |
628f4235 KH |
3534 | return ret; |
3535 | } | |
3536 | ||
4e416953 | 3537 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3538 | gfp_t gfp_mask, |
3539 | unsigned long *total_scanned) | |
4e416953 BS |
3540 | { |
3541 | unsigned long nr_reclaimed = 0; | |
3542 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3543 | unsigned long reclaimed; | |
3544 | int loop = 0; | |
3545 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3546 | unsigned long long excess; |
0ae5e89c | 3547 | unsigned long nr_scanned; |
4e416953 BS |
3548 | |
3549 | if (order > 0) | |
3550 | return 0; | |
3551 | ||
00918b6a | 3552 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3553 | /* |
3554 | * This loop can run a while, specially if mem_cgroup's continuously | |
3555 | * keep exceeding their soft limit and putting the system under | |
3556 | * pressure | |
3557 | */ | |
3558 | do { | |
3559 | if (next_mz) | |
3560 | mz = next_mz; | |
3561 | else | |
3562 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3563 | if (!mz) | |
3564 | break; | |
3565 | ||
0ae5e89c | 3566 | nr_scanned = 0; |
4e416953 BS |
3567 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, |
3568 | gfp_mask, | |
0ae5e89c YH |
3569 | MEM_CGROUP_RECLAIM_SOFT, |
3570 | &nr_scanned); | |
4e416953 | 3571 | nr_reclaimed += reclaimed; |
0ae5e89c | 3572 | *total_scanned += nr_scanned; |
4e416953 BS |
3573 | spin_lock(&mctz->lock); |
3574 | ||
3575 | /* | |
3576 | * If we failed to reclaim anything from this memory cgroup | |
3577 | * it is time to move on to the next cgroup | |
3578 | */ | |
3579 | next_mz = NULL; | |
3580 | if (!reclaimed) { | |
3581 | do { | |
3582 | /* | |
3583 | * Loop until we find yet another one. | |
3584 | * | |
3585 | * By the time we get the soft_limit lock | |
3586 | * again, someone might have aded the | |
3587 | * group back on the RB tree. Iterate to | |
3588 | * make sure we get a different mem. | |
3589 | * mem_cgroup_largest_soft_limit_node returns | |
3590 | * NULL if no other cgroup is present on | |
3591 | * the tree | |
3592 | */ | |
3593 | next_mz = | |
3594 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3595 | if (next_mz == mz) |
4e416953 | 3596 | css_put(&next_mz->mem->css); |
39cc98f1 | 3597 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3598 | break; |
3599 | } while (1); | |
3600 | } | |
4e416953 | 3601 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 3602 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
3603 | /* |
3604 | * One school of thought says that we should not add | |
3605 | * back the node to the tree if reclaim returns 0. | |
3606 | * But our reclaim could return 0, simply because due | |
3607 | * to priority we are exposing a smaller subset of | |
3608 | * memory to reclaim from. Consider this as a longer | |
3609 | * term TODO. | |
3610 | */ | |
ef8745c1 KH |
3611 | /* If excess == 0, no tree ops */ |
3612 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
3613 | spin_unlock(&mctz->lock); |
3614 | css_put(&mz->mem->css); | |
3615 | loop++; | |
3616 | /* | |
3617 | * Could not reclaim anything and there are no more | |
3618 | * mem cgroups to try or we seem to be looping without | |
3619 | * reclaiming anything. | |
3620 | */ | |
3621 | if (!nr_reclaimed && | |
3622 | (next_mz == NULL || | |
3623 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3624 | break; | |
3625 | } while (!nr_reclaimed); | |
3626 | if (next_mz) | |
3627 | css_put(&next_mz->mem->css); | |
3628 | return nr_reclaimed; | |
3629 | } | |
3630 | ||
cc847582 KH |
3631 | /* |
3632 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3633 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3634 | */ | |
f817ed48 | 3635 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 3636 | int node, int zid, enum lru_list lru) |
cc847582 | 3637 | { |
08e552c6 KH |
3638 | struct zone *zone; |
3639 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 3640 | struct page_cgroup *pc, *busy; |
08e552c6 | 3641 | unsigned long flags, loop; |
072c56c1 | 3642 | struct list_head *list; |
f817ed48 | 3643 | int ret = 0; |
072c56c1 | 3644 | |
08e552c6 KH |
3645 | zone = &NODE_DATA(node)->node_zones[zid]; |
3646 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 3647 | list = &mz->lists[lru]; |
cc847582 | 3648 | |
f817ed48 KH |
3649 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
3650 | /* give some margin against EBUSY etc...*/ | |
3651 | loop += 256; | |
3652 | busy = NULL; | |
3653 | while (loop--) { | |
5564e88b JW |
3654 | struct page *page; |
3655 | ||
f817ed48 | 3656 | ret = 0; |
08e552c6 | 3657 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3658 | if (list_empty(list)) { |
08e552c6 | 3659 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3660 | break; |
f817ed48 KH |
3661 | } |
3662 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
3663 | if (busy == pc) { | |
3664 | list_move(&pc->lru, list); | |
648bcc77 | 3665 | busy = NULL; |
08e552c6 | 3666 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3667 | continue; |
3668 | } | |
08e552c6 | 3669 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3670 | |
6b3ae58e | 3671 | page = lookup_cgroup_page(pc); |
5564e88b JW |
3672 | |
3673 | ret = mem_cgroup_move_parent(page, pc, mem, GFP_KERNEL); | |
f817ed48 | 3674 | if (ret == -ENOMEM) |
52d4b9ac | 3675 | break; |
f817ed48 KH |
3676 | |
3677 | if (ret == -EBUSY || ret == -EINVAL) { | |
3678 | /* found lock contention or "pc" is obsolete. */ | |
3679 | busy = pc; | |
3680 | cond_resched(); | |
3681 | } else | |
3682 | busy = NULL; | |
cc847582 | 3683 | } |
08e552c6 | 3684 | |
f817ed48 KH |
3685 | if (!ret && !list_empty(list)) |
3686 | return -EBUSY; | |
3687 | return ret; | |
cc847582 KH |
3688 | } |
3689 | ||
3690 | /* | |
3691 | * make mem_cgroup's charge to be 0 if there is no task. | |
3692 | * This enables deleting this mem_cgroup. | |
3693 | */ | |
c1e862c1 | 3694 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 3695 | { |
f817ed48 KH |
3696 | int ret; |
3697 | int node, zid, shrink; | |
3698 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 3699 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 3700 | |
cc847582 | 3701 | css_get(&mem->css); |
f817ed48 KH |
3702 | |
3703 | shrink = 0; | |
c1e862c1 KH |
3704 | /* should free all ? */ |
3705 | if (free_all) | |
3706 | goto try_to_free; | |
f817ed48 | 3707 | move_account: |
fce66477 | 3708 | do { |
f817ed48 | 3709 | ret = -EBUSY; |
c1e862c1 KH |
3710 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3711 | goto out; | |
3712 | ret = -EINTR; | |
3713 | if (signal_pending(current)) | |
cc847582 | 3714 | goto out; |
52d4b9ac KH |
3715 | /* This is for making all *used* pages to be on LRU. */ |
3716 | lru_add_drain_all(); | |
cdec2e42 | 3717 | drain_all_stock_sync(); |
f817ed48 | 3718 | ret = 0; |
32047e2a | 3719 | mem_cgroup_start_move(mem); |
299b4eaa | 3720 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3721 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 3722 | enum lru_list l; |
f817ed48 KH |
3723 | for_each_lru(l) { |
3724 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 3725 | node, zid, l); |
f817ed48 KH |
3726 | if (ret) |
3727 | break; | |
3728 | } | |
1ecaab2b | 3729 | } |
f817ed48 KH |
3730 | if (ret) |
3731 | break; | |
3732 | } | |
32047e2a | 3733 | mem_cgroup_end_move(mem); |
3c11ecf4 | 3734 | memcg_oom_recover(mem); |
f817ed48 KH |
3735 | /* it seems parent cgroup doesn't have enough mem */ |
3736 | if (ret == -ENOMEM) | |
3737 | goto try_to_free; | |
52d4b9ac | 3738 | cond_resched(); |
fce66477 DN |
3739 | /* "ret" should also be checked to ensure all lists are empty. */ |
3740 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
3741 | out: |
3742 | css_put(&mem->css); | |
3743 | return ret; | |
f817ed48 KH |
3744 | |
3745 | try_to_free: | |
c1e862c1 KH |
3746 | /* returns EBUSY if there is a task or if we come here twice. */ |
3747 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3748 | ret = -EBUSY; |
3749 | goto out; | |
3750 | } | |
c1e862c1 KH |
3751 | /* we call try-to-free pages for make this cgroup empty */ |
3752 | lru_add_drain_all(); | |
f817ed48 KH |
3753 | /* try to free all pages in this cgroup */ |
3754 | shrink = 1; | |
3755 | while (nr_retries && mem->res.usage > 0) { | |
3756 | int progress; | |
c1e862c1 KH |
3757 | |
3758 | if (signal_pending(current)) { | |
3759 | ret = -EINTR; | |
3760 | goto out; | |
3761 | } | |
a7885eb8 KM |
3762 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
3763 | false, get_swappiness(mem)); | |
c1e862c1 | 3764 | if (!progress) { |
f817ed48 | 3765 | nr_retries--; |
c1e862c1 | 3766 | /* maybe some writeback is necessary */ |
8aa7e847 | 3767 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3768 | } |
f817ed48 KH |
3769 | |
3770 | } | |
08e552c6 | 3771 | lru_add_drain(); |
f817ed48 | 3772 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3773 | goto move_account; |
cc847582 KH |
3774 | } |
3775 | ||
c1e862c1 KH |
3776 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3777 | { | |
3778 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3779 | } | |
3780 | ||
3781 | ||
18f59ea7 BS |
3782 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3783 | { | |
3784 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3785 | } | |
3786 | ||
3787 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3788 | u64 val) | |
3789 | { | |
3790 | int retval = 0; | |
3791 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3792 | struct cgroup *parent = cont->parent; | |
3793 | struct mem_cgroup *parent_mem = NULL; | |
3794 | ||
3795 | if (parent) | |
3796 | parent_mem = mem_cgroup_from_cont(parent); | |
3797 | ||
3798 | cgroup_lock(); | |
3799 | /* | |
af901ca1 | 3800 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3801 | * in the child subtrees. If it is unset, then the change can |
3802 | * occur, provided the current cgroup has no children. | |
3803 | * | |
3804 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3805 | * set if there are no children. | |
3806 | */ | |
3807 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
3808 | (val == 1 || val == 0)) { | |
3809 | if (list_empty(&cont->children)) | |
3810 | mem->use_hierarchy = val; | |
3811 | else | |
3812 | retval = -EBUSY; | |
3813 | } else | |
3814 | retval = -EINVAL; | |
3815 | cgroup_unlock(); | |
3816 | ||
3817 | return retval; | |
3818 | } | |
3819 | ||
0c3e73e8 | 3820 | |
7a159cc9 JW |
3821 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *mem, |
3822 | enum mem_cgroup_stat_index idx) | |
0c3e73e8 | 3823 | { |
7d74b06f | 3824 | struct mem_cgroup *iter; |
7a159cc9 | 3825 | long val = 0; |
0c3e73e8 | 3826 | |
7a159cc9 | 3827 | /* Per-cpu values can be negative, use a signed accumulator */ |
7d74b06f KH |
3828 | for_each_mem_cgroup_tree(iter, mem) |
3829 | val += mem_cgroup_read_stat(iter, idx); | |
3830 | ||
3831 | if (val < 0) /* race ? */ | |
3832 | val = 0; | |
3833 | return val; | |
0c3e73e8 BS |
3834 | } |
3835 | ||
104f3928 KS |
3836 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
3837 | { | |
7d74b06f | 3838 | u64 val; |
104f3928 KS |
3839 | |
3840 | if (!mem_cgroup_is_root(mem)) { | |
3841 | if (!swap) | |
3842 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
3843 | else | |
3844 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
3845 | } | |
3846 | ||
7a159cc9 JW |
3847 | val = mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_CACHE); |
3848 | val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3849 | |
7d74b06f | 3850 | if (swap) |
7a159cc9 | 3851 | val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3852 | |
3853 | return val << PAGE_SHIFT; | |
3854 | } | |
3855 | ||
2c3daa72 | 3856 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3857 | { |
8c7c6e34 | 3858 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 3859 | u64 val; |
8c7c6e34 KH |
3860 | int type, name; |
3861 | ||
3862 | type = MEMFILE_TYPE(cft->private); | |
3863 | name = MEMFILE_ATTR(cft->private); | |
3864 | switch (type) { | |
3865 | case _MEM: | |
104f3928 KS |
3866 | if (name == RES_USAGE) |
3867 | val = mem_cgroup_usage(mem, false); | |
3868 | else | |
0c3e73e8 | 3869 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
3870 | break; |
3871 | case _MEMSWAP: | |
104f3928 KS |
3872 | if (name == RES_USAGE) |
3873 | val = mem_cgroup_usage(mem, true); | |
3874 | else | |
0c3e73e8 | 3875 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
3876 | break; |
3877 | default: | |
3878 | BUG(); | |
3879 | break; | |
3880 | } | |
3881 | return val; | |
8cdea7c0 | 3882 | } |
628f4235 KH |
3883 | /* |
3884 | * The user of this function is... | |
3885 | * RES_LIMIT. | |
3886 | */ | |
856c13aa PM |
3887 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3888 | const char *buffer) | |
8cdea7c0 | 3889 | { |
628f4235 | 3890 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3891 | int type, name; |
628f4235 KH |
3892 | unsigned long long val; |
3893 | int ret; | |
3894 | ||
8c7c6e34 KH |
3895 | type = MEMFILE_TYPE(cft->private); |
3896 | name = MEMFILE_ATTR(cft->private); | |
3897 | switch (name) { | |
628f4235 | 3898 | case RES_LIMIT: |
4b3bde4c BS |
3899 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3900 | ret = -EINVAL; | |
3901 | break; | |
3902 | } | |
628f4235 KH |
3903 | /* This function does all necessary parse...reuse it */ |
3904 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3905 | if (ret) |
3906 | break; | |
3907 | if (type == _MEM) | |
628f4235 | 3908 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3909 | else |
3910 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3911 | break; |
296c81d8 BS |
3912 | case RES_SOFT_LIMIT: |
3913 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3914 | if (ret) | |
3915 | break; | |
3916 | /* | |
3917 | * For memsw, soft limits are hard to implement in terms | |
3918 | * of semantics, for now, we support soft limits for | |
3919 | * control without swap | |
3920 | */ | |
3921 | if (type == _MEM) | |
3922 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3923 | else | |
3924 | ret = -EINVAL; | |
3925 | break; | |
628f4235 KH |
3926 | default: |
3927 | ret = -EINVAL; /* should be BUG() ? */ | |
3928 | break; | |
3929 | } | |
3930 | return ret; | |
8cdea7c0 BS |
3931 | } |
3932 | ||
fee7b548 KH |
3933 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3934 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3935 | { | |
3936 | struct cgroup *cgroup; | |
3937 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3938 | ||
3939 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3940 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3941 | cgroup = memcg->css.cgroup; | |
3942 | if (!memcg->use_hierarchy) | |
3943 | goto out; | |
3944 | ||
3945 | while (cgroup->parent) { | |
3946 | cgroup = cgroup->parent; | |
3947 | memcg = mem_cgroup_from_cont(cgroup); | |
3948 | if (!memcg->use_hierarchy) | |
3949 | break; | |
3950 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3951 | min_limit = min(min_limit, tmp); | |
3952 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3953 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3954 | } | |
3955 | out: | |
3956 | *mem_limit = min_limit; | |
3957 | *memsw_limit = min_memsw_limit; | |
3958 | return; | |
3959 | } | |
3960 | ||
29f2a4da | 3961 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
3962 | { |
3963 | struct mem_cgroup *mem; | |
8c7c6e34 | 3964 | int type, name; |
c84872e1 PE |
3965 | |
3966 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
3967 | type = MEMFILE_TYPE(event); |
3968 | name = MEMFILE_ATTR(event); | |
3969 | switch (name) { | |
29f2a4da | 3970 | case RES_MAX_USAGE: |
8c7c6e34 KH |
3971 | if (type == _MEM) |
3972 | res_counter_reset_max(&mem->res); | |
3973 | else | |
3974 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
3975 | break; |
3976 | case RES_FAILCNT: | |
8c7c6e34 KH |
3977 | if (type == _MEM) |
3978 | res_counter_reset_failcnt(&mem->res); | |
3979 | else | |
3980 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
3981 | break; |
3982 | } | |
f64c3f54 | 3983 | |
85cc59db | 3984 | return 0; |
c84872e1 PE |
3985 | } |
3986 | ||
7dc74be0 DN |
3987 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3988 | struct cftype *cft) | |
3989 | { | |
3990 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3991 | } | |
3992 | ||
02491447 | 3993 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3994 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3995 | struct cftype *cft, u64 val) | |
3996 | { | |
3997 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3998 | ||
3999 | if (val >= (1 << NR_MOVE_TYPE)) | |
4000 | return -EINVAL; | |
4001 | /* | |
4002 | * We check this value several times in both in can_attach() and | |
4003 | * attach(), so we need cgroup lock to prevent this value from being | |
4004 | * inconsistent. | |
4005 | */ | |
4006 | cgroup_lock(); | |
4007 | mem->move_charge_at_immigrate = val; | |
4008 | cgroup_unlock(); | |
4009 | ||
4010 | return 0; | |
4011 | } | |
02491447 DN |
4012 | #else |
4013 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4014 | struct cftype *cft, u64 val) | |
4015 | { | |
4016 | return -ENOSYS; | |
4017 | } | |
4018 | #endif | |
7dc74be0 | 4019 | |
14067bb3 KH |
4020 | |
4021 | /* For read statistics */ | |
4022 | enum { | |
4023 | MCS_CACHE, | |
4024 | MCS_RSS, | |
d8046582 | 4025 | MCS_FILE_MAPPED, |
14067bb3 KH |
4026 | MCS_PGPGIN, |
4027 | MCS_PGPGOUT, | |
1dd3a273 | 4028 | MCS_SWAP, |
456f998e YH |
4029 | MCS_PGFAULT, |
4030 | MCS_PGMAJFAULT, | |
14067bb3 KH |
4031 | MCS_INACTIVE_ANON, |
4032 | MCS_ACTIVE_ANON, | |
4033 | MCS_INACTIVE_FILE, | |
4034 | MCS_ACTIVE_FILE, | |
4035 | MCS_UNEVICTABLE, | |
4036 | NR_MCS_STAT, | |
4037 | }; | |
4038 | ||
4039 | struct mcs_total_stat { | |
4040 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
4041 | }; |
4042 | ||
14067bb3 KH |
4043 | struct { |
4044 | char *local_name; | |
4045 | char *total_name; | |
4046 | } memcg_stat_strings[NR_MCS_STAT] = { | |
4047 | {"cache", "total_cache"}, | |
4048 | {"rss", "total_rss"}, | |
d69b042f | 4049 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
4050 | {"pgpgin", "total_pgpgin"}, |
4051 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 4052 | {"swap", "total_swap"}, |
456f998e YH |
4053 | {"pgfault", "total_pgfault"}, |
4054 | {"pgmajfault", "total_pgmajfault"}, | |
14067bb3 KH |
4055 | {"inactive_anon", "total_inactive_anon"}, |
4056 | {"active_anon", "total_active_anon"}, | |
4057 | {"inactive_file", "total_inactive_file"}, | |
4058 | {"active_file", "total_active_file"}, | |
4059 | {"unevictable", "total_unevictable"} | |
4060 | }; | |
4061 | ||
4062 | ||
7d74b06f KH |
4063 | static void |
4064 | mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
14067bb3 | 4065 | { |
14067bb3 KH |
4066 | s64 val; |
4067 | ||
4068 | /* per cpu stat */ | |
c62b1a3b | 4069 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 4070 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 4071 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 4072 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 4073 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 4074 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
e9f8974f | 4075 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGPGIN); |
14067bb3 | 4076 | s->stat[MCS_PGPGIN] += val; |
e9f8974f | 4077 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGPGOUT); |
14067bb3 | 4078 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 4079 | if (do_swap_account) { |
c62b1a3b | 4080 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
4081 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
4082 | } | |
456f998e YH |
4083 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGFAULT); |
4084 | s->stat[MCS_PGFAULT] += val; | |
4085 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGMAJFAULT); | |
4086 | s->stat[MCS_PGMAJFAULT] += val; | |
14067bb3 KH |
4087 | |
4088 | /* per zone stat */ | |
4089 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
4090 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
4091 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
4092 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
4093 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
4094 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
4095 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
4096 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
4097 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
4098 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
14067bb3 KH |
4099 | } |
4100 | ||
4101 | static void | |
4102 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
4103 | { | |
7d74b06f KH |
4104 | struct mem_cgroup *iter; |
4105 | ||
4106 | for_each_mem_cgroup_tree(iter, mem) | |
4107 | mem_cgroup_get_local_stat(iter, s); | |
14067bb3 KH |
4108 | } |
4109 | ||
406eb0c9 YH |
4110 | #ifdef CONFIG_NUMA |
4111 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) | |
4112 | { | |
4113 | int nid; | |
4114 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4115 | unsigned long node_nr; | |
4116 | struct cgroup *cont = m->private; | |
4117 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
4118 | ||
4119 | total_nr = mem_cgroup_nr_lru_pages(mem_cont); | |
4120 | seq_printf(m, "total=%lu", total_nr); | |
4121 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4122 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid); | |
4123 | seq_printf(m, " N%d=%lu", nid, node_nr); | |
4124 | } | |
4125 | seq_putc(m, '\n'); | |
4126 | ||
4127 | file_nr = mem_cgroup_nr_file_lru_pages(mem_cont); | |
4128 | seq_printf(m, "file=%lu", file_nr); | |
4129 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4130 | node_nr = mem_cgroup_node_nr_file_lru_pages(mem_cont, nid); | |
4131 | seq_printf(m, " N%d=%lu", nid, node_nr); | |
4132 | } | |
4133 | seq_putc(m, '\n'); | |
4134 | ||
4135 | anon_nr = mem_cgroup_nr_anon_lru_pages(mem_cont); | |
4136 | seq_printf(m, "anon=%lu", anon_nr); | |
4137 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4138 | node_nr = mem_cgroup_node_nr_anon_lru_pages(mem_cont, nid); | |
4139 | seq_printf(m, " N%d=%lu", nid, node_nr); | |
4140 | } | |
4141 | seq_putc(m, '\n'); | |
4142 | ||
4143 | unevictable_nr = mem_cgroup_nr_unevictable_lru_pages(mem_cont); | |
4144 | seq_printf(m, "unevictable=%lu", unevictable_nr); | |
4145 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4146 | node_nr = mem_cgroup_node_nr_unevictable_lru_pages(mem_cont, | |
4147 | nid); | |
4148 | seq_printf(m, " N%d=%lu", nid, node_nr); | |
4149 | } | |
4150 | seq_putc(m, '\n'); | |
4151 | return 0; | |
4152 | } | |
4153 | #endif /* CONFIG_NUMA */ | |
4154 | ||
c64745cf PM |
4155 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
4156 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 4157 | { |
d2ceb9b7 | 4158 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 4159 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
4160 | int i; |
4161 | ||
14067bb3 KH |
4162 | memset(&mystat, 0, sizeof(mystat)); |
4163 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 4164 | |
406eb0c9 | 4165 | |
1dd3a273 DN |
4166 | for (i = 0; i < NR_MCS_STAT; i++) { |
4167 | if (i == MCS_SWAP && !do_swap_account) | |
4168 | continue; | |
14067bb3 | 4169 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 4170 | } |
7b854121 | 4171 | |
14067bb3 | 4172 | /* Hierarchical information */ |
fee7b548 KH |
4173 | { |
4174 | unsigned long long limit, memsw_limit; | |
4175 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
4176 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
4177 | if (do_swap_account) | |
4178 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
4179 | } | |
7f016ee8 | 4180 | |
14067bb3 KH |
4181 | memset(&mystat, 0, sizeof(mystat)); |
4182 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
4183 | for (i = 0; i < NR_MCS_STAT; i++) { |
4184 | if (i == MCS_SWAP && !do_swap_account) | |
4185 | continue; | |
14067bb3 | 4186 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 4187 | } |
14067bb3 | 4188 | |
7f016ee8 | 4189 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 4190 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
4191 | |
4192 | { | |
4193 | int nid, zid; | |
4194 | struct mem_cgroup_per_zone *mz; | |
4195 | unsigned long recent_rotated[2] = {0, 0}; | |
4196 | unsigned long recent_scanned[2] = {0, 0}; | |
4197 | ||
4198 | for_each_online_node(nid) | |
4199 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
4200 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
4201 | ||
4202 | recent_rotated[0] += | |
4203 | mz->reclaim_stat.recent_rotated[0]; | |
4204 | recent_rotated[1] += | |
4205 | mz->reclaim_stat.recent_rotated[1]; | |
4206 | recent_scanned[0] += | |
4207 | mz->reclaim_stat.recent_scanned[0]; | |
4208 | recent_scanned[1] += | |
4209 | mz->reclaim_stat.recent_scanned[1]; | |
4210 | } | |
4211 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
4212 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
4213 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
4214 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
4215 | } | |
4216 | #endif | |
4217 | ||
d2ceb9b7 KH |
4218 | return 0; |
4219 | } | |
4220 | ||
a7885eb8 KM |
4221 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4222 | { | |
4223 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4224 | ||
4225 | return get_swappiness(memcg); | |
4226 | } | |
4227 | ||
4228 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4229 | u64 val) | |
4230 | { | |
4231 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4232 | struct mem_cgroup *parent; | |
068b38c1 | 4233 | |
a7885eb8 KM |
4234 | if (val > 100) |
4235 | return -EINVAL; | |
4236 | ||
4237 | if (cgrp->parent == NULL) | |
4238 | return -EINVAL; | |
4239 | ||
4240 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4241 | |
4242 | cgroup_lock(); | |
4243 | ||
a7885eb8 KM |
4244 | /* If under hierarchy, only empty-root can set this value */ |
4245 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4246 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4247 | cgroup_unlock(); | |
a7885eb8 | 4248 | return -EINVAL; |
068b38c1 | 4249 | } |
a7885eb8 | 4250 | |
a7885eb8 | 4251 | memcg->swappiness = val; |
a7885eb8 | 4252 | |
068b38c1 LZ |
4253 | cgroup_unlock(); |
4254 | ||
a7885eb8 KM |
4255 | return 0; |
4256 | } | |
4257 | ||
2e72b634 KS |
4258 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4259 | { | |
4260 | struct mem_cgroup_threshold_ary *t; | |
4261 | u64 usage; | |
4262 | int i; | |
4263 | ||
4264 | rcu_read_lock(); | |
4265 | if (!swap) | |
2c488db2 | 4266 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4267 | else |
2c488db2 | 4268 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4269 | |
4270 | if (!t) | |
4271 | goto unlock; | |
4272 | ||
4273 | usage = mem_cgroup_usage(memcg, swap); | |
4274 | ||
4275 | /* | |
4276 | * current_threshold points to threshold just below usage. | |
4277 | * If it's not true, a threshold was crossed after last | |
4278 | * call of __mem_cgroup_threshold(). | |
4279 | */ | |
5407a562 | 4280 | i = t->current_threshold; |
2e72b634 KS |
4281 | |
4282 | /* | |
4283 | * Iterate backward over array of thresholds starting from | |
4284 | * current_threshold and check if a threshold is crossed. | |
4285 | * If none of thresholds below usage is crossed, we read | |
4286 | * only one element of the array here. | |
4287 | */ | |
4288 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4289 | eventfd_signal(t->entries[i].eventfd, 1); | |
4290 | ||
4291 | /* i = current_threshold + 1 */ | |
4292 | i++; | |
4293 | ||
4294 | /* | |
4295 | * Iterate forward over array of thresholds starting from | |
4296 | * current_threshold+1 and check if a threshold is crossed. | |
4297 | * If none of thresholds above usage is crossed, we read | |
4298 | * only one element of the array here. | |
4299 | */ | |
4300 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4301 | eventfd_signal(t->entries[i].eventfd, 1); | |
4302 | ||
4303 | /* Update current_threshold */ | |
5407a562 | 4304 | t->current_threshold = i - 1; |
2e72b634 KS |
4305 | unlock: |
4306 | rcu_read_unlock(); | |
4307 | } | |
4308 | ||
4309 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4310 | { | |
ad4ca5f4 KS |
4311 | while (memcg) { |
4312 | __mem_cgroup_threshold(memcg, false); | |
4313 | if (do_swap_account) | |
4314 | __mem_cgroup_threshold(memcg, true); | |
4315 | ||
4316 | memcg = parent_mem_cgroup(memcg); | |
4317 | } | |
2e72b634 KS |
4318 | } |
4319 | ||
4320 | static int compare_thresholds(const void *a, const void *b) | |
4321 | { | |
4322 | const struct mem_cgroup_threshold *_a = a; | |
4323 | const struct mem_cgroup_threshold *_b = b; | |
4324 | ||
4325 | return _a->threshold - _b->threshold; | |
4326 | } | |
4327 | ||
7d74b06f | 4328 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem) |
9490ff27 KH |
4329 | { |
4330 | struct mem_cgroup_eventfd_list *ev; | |
4331 | ||
4332 | list_for_each_entry(ev, &mem->oom_notify, list) | |
4333 | eventfd_signal(ev->eventfd, 1); | |
4334 | return 0; | |
4335 | } | |
4336 | ||
4337 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) | |
4338 | { | |
7d74b06f KH |
4339 | struct mem_cgroup *iter; |
4340 | ||
4341 | for_each_mem_cgroup_tree(iter, mem) | |
4342 | mem_cgroup_oom_notify_cb(iter); | |
9490ff27 KH |
4343 | } |
4344 | ||
4345 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4346 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4347 | { |
4348 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4349 | struct mem_cgroup_thresholds *thresholds; |
4350 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4351 | int type = MEMFILE_TYPE(cft->private); |
4352 | u64 threshold, usage; | |
2c488db2 | 4353 | int i, size, ret; |
2e72b634 KS |
4354 | |
4355 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4356 | if (ret) | |
4357 | return ret; | |
4358 | ||
4359 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4360 | |
2e72b634 | 4361 | if (type == _MEM) |
2c488db2 | 4362 | thresholds = &memcg->thresholds; |
2e72b634 | 4363 | else if (type == _MEMSWAP) |
2c488db2 | 4364 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4365 | else |
4366 | BUG(); | |
4367 | ||
4368 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4369 | ||
4370 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4371 | if (thresholds->primary) |
2e72b634 KS |
4372 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4373 | ||
2c488db2 | 4374 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4375 | |
4376 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4377 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4378 | GFP_KERNEL); |
2c488db2 | 4379 | if (!new) { |
2e72b634 KS |
4380 | ret = -ENOMEM; |
4381 | goto unlock; | |
4382 | } | |
2c488db2 | 4383 | new->size = size; |
2e72b634 KS |
4384 | |
4385 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4386 | if (thresholds->primary) { |
4387 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4388 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4389 | } |
4390 | ||
2e72b634 | 4391 | /* Add new threshold */ |
2c488db2 KS |
4392 | new->entries[size - 1].eventfd = eventfd; |
4393 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4394 | |
4395 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4396 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4397 | compare_thresholds, NULL); |
4398 | ||
4399 | /* Find current threshold */ | |
2c488db2 | 4400 | new->current_threshold = -1; |
2e72b634 | 4401 | for (i = 0; i < size; i++) { |
2c488db2 | 4402 | if (new->entries[i].threshold < usage) { |
2e72b634 | 4403 | /* |
2c488db2 KS |
4404 | * new->current_threshold will not be used until |
4405 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4406 | * it here. |
4407 | */ | |
2c488db2 | 4408 | ++new->current_threshold; |
2e72b634 KS |
4409 | } |
4410 | } | |
4411 | ||
2c488db2 KS |
4412 | /* Free old spare buffer and save old primary buffer as spare */ |
4413 | kfree(thresholds->spare); | |
4414 | thresholds->spare = thresholds->primary; | |
4415 | ||
4416 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4417 | |
907860ed | 4418 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4419 | synchronize_rcu(); |
4420 | ||
2e72b634 KS |
4421 | unlock: |
4422 | mutex_unlock(&memcg->thresholds_lock); | |
4423 | ||
4424 | return ret; | |
4425 | } | |
4426 | ||
907860ed | 4427 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4428 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4429 | { |
4430 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4431 | struct mem_cgroup_thresholds *thresholds; |
4432 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4433 | int type = MEMFILE_TYPE(cft->private); |
4434 | u64 usage; | |
2c488db2 | 4435 | int i, j, size; |
2e72b634 KS |
4436 | |
4437 | mutex_lock(&memcg->thresholds_lock); | |
4438 | if (type == _MEM) | |
2c488db2 | 4439 | thresholds = &memcg->thresholds; |
2e72b634 | 4440 | else if (type == _MEMSWAP) |
2c488db2 | 4441 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4442 | else |
4443 | BUG(); | |
4444 | ||
4445 | /* | |
4446 | * Something went wrong if we trying to unregister a threshold | |
4447 | * if we don't have thresholds | |
4448 | */ | |
4449 | BUG_ON(!thresholds); | |
4450 | ||
4451 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4452 | ||
4453 | /* Check if a threshold crossed before removing */ | |
4454 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4455 | ||
4456 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4457 | size = 0; |
4458 | for (i = 0; i < thresholds->primary->size; i++) { | |
4459 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4460 | size++; |
4461 | } | |
4462 | ||
2c488db2 | 4463 | new = thresholds->spare; |
907860ed | 4464 | |
2e72b634 KS |
4465 | /* Set thresholds array to NULL if we don't have thresholds */ |
4466 | if (!size) { | |
2c488db2 KS |
4467 | kfree(new); |
4468 | new = NULL; | |
907860ed | 4469 | goto swap_buffers; |
2e72b634 KS |
4470 | } |
4471 | ||
2c488db2 | 4472 | new->size = size; |
2e72b634 KS |
4473 | |
4474 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4475 | new->current_threshold = -1; |
4476 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4477 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4478 | continue; |
4479 | ||
2c488db2 KS |
4480 | new->entries[j] = thresholds->primary->entries[i]; |
4481 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4482 | /* |
2c488db2 | 4483 | * new->current_threshold will not be used |
2e72b634 KS |
4484 | * until rcu_assign_pointer(), so it's safe to increment |
4485 | * it here. | |
4486 | */ | |
2c488db2 | 4487 | ++new->current_threshold; |
2e72b634 KS |
4488 | } |
4489 | j++; | |
4490 | } | |
4491 | ||
907860ed | 4492 | swap_buffers: |
2c488db2 KS |
4493 | /* Swap primary and spare array */ |
4494 | thresholds->spare = thresholds->primary; | |
4495 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4496 | |
907860ed | 4497 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4498 | synchronize_rcu(); |
4499 | ||
2e72b634 | 4500 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4501 | } |
c1e862c1 | 4502 | |
9490ff27 KH |
4503 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4504 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4505 | { | |
4506 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4507 | struct mem_cgroup_eventfd_list *event; | |
4508 | int type = MEMFILE_TYPE(cft->private); | |
4509 | ||
4510 | BUG_ON(type != _OOM_TYPE); | |
4511 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4512 | if (!event) | |
4513 | return -ENOMEM; | |
4514 | ||
4515 | mutex_lock(&memcg_oom_mutex); | |
4516 | ||
4517 | event->eventfd = eventfd; | |
4518 | list_add(&event->list, &memcg->oom_notify); | |
4519 | ||
4520 | /* already in OOM ? */ | |
4521 | if (atomic_read(&memcg->oom_lock)) | |
4522 | eventfd_signal(eventfd, 1); | |
4523 | mutex_unlock(&memcg_oom_mutex); | |
4524 | ||
4525 | return 0; | |
4526 | } | |
4527 | ||
907860ed | 4528 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4529 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4530 | { | |
4531 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4532 | struct mem_cgroup_eventfd_list *ev, *tmp; | |
4533 | int type = MEMFILE_TYPE(cft->private); | |
4534 | ||
4535 | BUG_ON(type != _OOM_TYPE); | |
4536 | ||
4537 | mutex_lock(&memcg_oom_mutex); | |
4538 | ||
4539 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { | |
4540 | if (ev->eventfd == eventfd) { | |
4541 | list_del(&ev->list); | |
4542 | kfree(ev); | |
4543 | } | |
4544 | } | |
4545 | ||
4546 | mutex_unlock(&memcg_oom_mutex); | |
9490ff27 KH |
4547 | } |
4548 | ||
3c11ecf4 KH |
4549 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4550 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4551 | { | |
4552 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4553 | ||
4554 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); | |
4555 | ||
4556 | if (atomic_read(&mem->oom_lock)) | |
4557 | cb->fill(cb, "under_oom", 1); | |
4558 | else | |
4559 | cb->fill(cb, "under_oom", 0); | |
4560 | return 0; | |
4561 | } | |
4562 | ||
3c11ecf4 KH |
4563 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4564 | struct cftype *cft, u64 val) | |
4565 | { | |
4566 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4567 | struct mem_cgroup *parent; | |
4568 | ||
4569 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4570 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4571 | return -EINVAL; | |
4572 | ||
4573 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4574 | ||
4575 | cgroup_lock(); | |
4576 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4577 | if ((parent->use_hierarchy) || | |
4578 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { | |
4579 | cgroup_unlock(); | |
4580 | return -EINVAL; | |
4581 | } | |
4582 | mem->oom_kill_disable = val; | |
4d845ebf KH |
4583 | if (!val) |
4584 | memcg_oom_recover(mem); | |
3c11ecf4 KH |
4585 | cgroup_unlock(); |
4586 | return 0; | |
4587 | } | |
4588 | ||
406eb0c9 YH |
4589 | #ifdef CONFIG_NUMA |
4590 | static const struct file_operations mem_control_numa_stat_file_operations = { | |
4591 | .read = seq_read, | |
4592 | .llseek = seq_lseek, | |
4593 | .release = single_release, | |
4594 | }; | |
4595 | ||
4596 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) | |
4597 | { | |
4598 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
4599 | ||
4600 | file->f_op = &mem_control_numa_stat_file_operations; | |
4601 | return single_open(file, mem_control_numa_stat_show, cont); | |
4602 | } | |
4603 | #endif /* CONFIG_NUMA */ | |
4604 | ||
8cdea7c0 BS |
4605 | static struct cftype mem_cgroup_files[] = { |
4606 | { | |
0eea1030 | 4607 | .name = "usage_in_bytes", |
8c7c6e34 | 4608 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4609 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4610 | .register_event = mem_cgroup_usage_register_event, |
4611 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4612 | }, |
c84872e1 PE |
4613 | { |
4614 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4615 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4616 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4617 | .read_u64 = mem_cgroup_read, |
4618 | }, | |
8cdea7c0 | 4619 | { |
0eea1030 | 4620 | .name = "limit_in_bytes", |
8c7c6e34 | 4621 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4622 | .write_string = mem_cgroup_write, |
2c3daa72 | 4623 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4624 | }, |
296c81d8 BS |
4625 | { |
4626 | .name = "soft_limit_in_bytes", | |
4627 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4628 | .write_string = mem_cgroup_write, | |
4629 | .read_u64 = mem_cgroup_read, | |
4630 | }, | |
8cdea7c0 BS |
4631 | { |
4632 | .name = "failcnt", | |
8c7c6e34 | 4633 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4634 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4635 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4636 | }, |
d2ceb9b7 KH |
4637 | { |
4638 | .name = "stat", | |
c64745cf | 4639 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4640 | }, |
c1e862c1 KH |
4641 | { |
4642 | .name = "force_empty", | |
4643 | .trigger = mem_cgroup_force_empty_write, | |
4644 | }, | |
18f59ea7 BS |
4645 | { |
4646 | .name = "use_hierarchy", | |
4647 | .write_u64 = mem_cgroup_hierarchy_write, | |
4648 | .read_u64 = mem_cgroup_hierarchy_read, | |
4649 | }, | |
a7885eb8 KM |
4650 | { |
4651 | .name = "swappiness", | |
4652 | .read_u64 = mem_cgroup_swappiness_read, | |
4653 | .write_u64 = mem_cgroup_swappiness_write, | |
4654 | }, | |
7dc74be0 DN |
4655 | { |
4656 | .name = "move_charge_at_immigrate", | |
4657 | .read_u64 = mem_cgroup_move_charge_read, | |
4658 | .write_u64 = mem_cgroup_move_charge_write, | |
4659 | }, | |
9490ff27 KH |
4660 | { |
4661 | .name = "oom_control", | |
3c11ecf4 KH |
4662 | .read_map = mem_cgroup_oom_control_read, |
4663 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4664 | .register_event = mem_cgroup_oom_register_event, |
4665 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4666 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4667 | }, | |
406eb0c9 YH |
4668 | #ifdef CONFIG_NUMA |
4669 | { | |
4670 | .name = "numa_stat", | |
4671 | .open = mem_control_numa_stat_open, | |
89577127 | 4672 | .mode = S_IRUGO, |
406eb0c9 YH |
4673 | }, |
4674 | #endif | |
8cdea7c0 BS |
4675 | }; |
4676 | ||
8c7c6e34 KH |
4677 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4678 | static struct cftype memsw_cgroup_files[] = { | |
4679 | { | |
4680 | .name = "memsw.usage_in_bytes", | |
4681 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4682 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4683 | .register_event = mem_cgroup_usage_register_event, |
4684 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4685 | }, |
4686 | { | |
4687 | .name = "memsw.max_usage_in_bytes", | |
4688 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4689 | .trigger = mem_cgroup_reset, | |
4690 | .read_u64 = mem_cgroup_read, | |
4691 | }, | |
4692 | { | |
4693 | .name = "memsw.limit_in_bytes", | |
4694 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4695 | .write_string = mem_cgroup_write, | |
4696 | .read_u64 = mem_cgroup_read, | |
4697 | }, | |
4698 | { | |
4699 | .name = "memsw.failcnt", | |
4700 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4701 | .trigger = mem_cgroup_reset, | |
4702 | .read_u64 = mem_cgroup_read, | |
4703 | }, | |
4704 | }; | |
4705 | ||
4706 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4707 | { | |
4708 | if (!do_swap_account) | |
4709 | return 0; | |
4710 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4711 | ARRAY_SIZE(memsw_cgroup_files)); | |
4712 | }; | |
4713 | #else | |
4714 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4715 | { | |
4716 | return 0; | |
4717 | } | |
4718 | #endif | |
4719 | ||
6d12e2d8 KH |
4720 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4721 | { | |
4722 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4723 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 4724 | enum lru_list l; |
41e3355d | 4725 | int zone, tmp = node; |
1ecaab2b KH |
4726 | /* |
4727 | * This routine is called against possible nodes. | |
4728 | * But it's BUG to call kmalloc() against offline node. | |
4729 | * | |
4730 | * TODO: this routine can waste much memory for nodes which will | |
4731 | * never be onlined. It's better to use memory hotplug callback | |
4732 | * function. | |
4733 | */ | |
41e3355d KH |
4734 | if (!node_state(node, N_NORMAL_MEMORY)) |
4735 | tmp = -1; | |
17295c88 | 4736 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4737 | if (!pn) |
4738 | return 1; | |
1ecaab2b | 4739 | |
6d12e2d8 | 4740 | mem->info.nodeinfo[node] = pn; |
1ecaab2b KH |
4741 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4742 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
4743 | for_each_lru(l) |
4744 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 4745 | mz->usage_in_excess = 0; |
4e416953 BS |
4746 | mz->on_tree = false; |
4747 | mz->mem = mem; | |
1ecaab2b | 4748 | } |
6d12e2d8 KH |
4749 | return 0; |
4750 | } | |
4751 | ||
1ecaab2b KH |
4752 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4753 | { | |
4754 | kfree(mem->info.nodeinfo[node]); | |
4755 | } | |
4756 | ||
33327948 KH |
4757 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4758 | { | |
4759 | struct mem_cgroup *mem; | |
c62b1a3b | 4760 | int size = sizeof(struct mem_cgroup); |
33327948 | 4761 | |
c62b1a3b | 4762 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4763 | if (size < PAGE_SIZE) |
17295c88 | 4764 | mem = kzalloc(size, GFP_KERNEL); |
33327948 | 4765 | else |
17295c88 | 4766 | mem = vzalloc(size); |
33327948 | 4767 | |
e7bbcdf3 DC |
4768 | if (!mem) |
4769 | return NULL; | |
4770 | ||
c62b1a3b | 4771 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
d2e61b8d DC |
4772 | if (!mem->stat) |
4773 | goto out_free; | |
711d3d2c | 4774 | spin_lock_init(&mem->pcp_counter_lock); |
33327948 | 4775 | return mem; |
d2e61b8d DC |
4776 | |
4777 | out_free: | |
4778 | if (size < PAGE_SIZE) | |
4779 | kfree(mem); | |
4780 | else | |
4781 | vfree(mem); | |
4782 | return NULL; | |
33327948 KH |
4783 | } |
4784 | ||
8c7c6e34 KH |
4785 | /* |
4786 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4787 | * (scanning all at force_empty is too costly...) | |
4788 | * | |
4789 | * Instead of clearing all references at force_empty, we remember | |
4790 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4791 | * it goes down to 0. | |
4792 | * | |
8c7c6e34 KH |
4793 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4794 | */ | |
4795 | ||
a7ba0eef | 4796 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 4797 | { |
08e552c6 KH |
4798 | int node; |
4799 | ||
f64c3f54 | 4800 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
4801 | free_css_id(&mem_cgroup_subsys, &mem->css); |
4802 | ||
08e552c6 KH |
4803 | for_each_node_state(node, N_POSSIBLE) |
4804 | free_mem_cgroup_per_zone_info(mem, node); | |
4805 | ||
c62b1a3b KH |
4806 | free_percpu(mem->stat); |
4807 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
4808 | kfree(mem); |
4809 | else | |
4810 | vfree(mem); | |
4811 | } | |
4812 | ||
8c7c6e34 KH |
4813 | static void mem_cgroup_get(struct mem_cgroup *mem) |
4814 | { | |
4815 | atomic_inc(&mem->refcnt); | |
4816 | } | |
4817 | ||
483c30b5 | 4818 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 4819 | { |
483c30b5 | 4820 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 4821 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 4822 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
4823 | if (parent) |
4824 | mem_cgroup_put(parent); | |
4825 | } | |
8c7c6e34 KH |
4826 | } |
4827 | ||
483c30b5 DN |
4828 | static void mem_cgroup_put(struct mem_cgroup *mem) |
4829 | { | |
4830 | __mem_cgroup_put(mem, 1); | |
4831 | } | |
4832 | ||
7bcc1bb1 DN |
4833 | /* |
4834 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4835 | */ | |
4836 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
4837 | { | |
4838 | if (!mem->res.parent) | |
4839 | return NULL; | |
4840 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
4841 | } | |
33327948 | 4842 | |
c077719b KH |
4843 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4844 | static void __init enable_swap_cgroup(void) | |
4845 | { | |
f8d66542 | 4846 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4847 | do_swap_account = 1; |
4848 | } | |
4849 | #else | |
4850 | static void __init enable_swap_cgroup(void) | |
4851 | { | |
4852 | } | |
4853 | #endif | |
4854 | ||
f64c3f54 BS |
4855 | static int mem_cgroup_soft_limit_tree_init(void) |
4856 | { | |
4857 | struct mem_cgroup_tree_per_node *rtpn; | |
4858 | struct mem_cgroup_tree_per_zone *rtpz; | |
4859 | int tmp, node, zone; | |
4860 | ||
4861 | for_each_node_state(node, N_POSSIBLE) { | |
4862 | tmp = node; | |
4863 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4864 | tmp = -1; | |
4865 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4866 | if (!rtpn) | |
4867 | return 1; | |
4868 | ||
4869 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4870 | ||
4871 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4872 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4873 | rtpz->rb_root = RB_ROOT; | |
4874 | spin_lock_init(&rtpz->lock); | |
4875 | } | |
4876 | } | |
4877 | return 0; | |
4878 | } | |
4879 | ||
0eb253e2 | 4880 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4881 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4882 | { | |
28dbc4b6 | 4883 | struct mem_cgroup *mem, *parent; |
04046e1a | 4884 | long error = -ENOMEM; |
6d12e2d8 | 4885 | int node; |
8cdea7c0 | 4886 | |
c8dad2bb JB |
4887 | mem = mem_cgroup_alloc(); |
4888 | if (!mem) | |
04046e1a | 4889 | return ERR_PTR(error); |
78fb7466 | 4890 | |
6d12e2d8 KH |
4891 | for_each_node_state(node, N_POSSIBLE) |
4892 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
4893 | goto free_out; | |
f64c3f54 | 4894 | |
c077719b | 4895 | /* root ? */ |
28dbc4b6 | 4896 | if (cont->parent == NULL) { |
cdec2e42 | 4897 | int cpu; |
c077719b | 4898 | enable_swap_cgroup(); |
28dbc4b6 | 4899 | parent = NULL; |
4b3bde4c | 4900 | root_mem_cgroup = mem; |
f64c3f54 BS |
4901 | if (mem_cgroup_soft_limit_tree_init()) |
4902 | goto free_out; | |
cdec2e42 KH |
4903 | for_each_possible_cpu(cpu) { |
4904 | struct memcg_stock_pcp *stock = | |
4905 | &per_cpu(memcg_stock, cpu); | |
4906 | INIT_WORK(&stock->work, drain_local_stock); | |
4907 | } | |
711d3d2c | 4908 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4909 | } else { |
28dbc4b6 | 4910 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 | 4911 | mem->use_hierarchy = parent->use_hierarchy; |
3c11ecf4 | 4912 | mem->oom_kill_disable = parent->oom_kill_disable; |
18f59ea7 | 4913 | } |
28dbc4b6 | 4914 | |
18f59ea7 BS |
4915 | if (parent && parent->use_hierarchy) { |
4916 | res_counter_init(&mem->res, &parent->res); | |
4917 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4918 | /* |
4919 | * We increment refcnt of the parent to ensure that we can | |
4920 | * safely access it on res_counter_charge/uncharge. | |
4921 | * This refcnt will be decremented when freeing this | |
4922 | * mem_cgroup(see mem_cgroup_put). | |
4923 | */ | |
4924 | mem_cgroup_get(parent); | |
18f59ea7 BS |
4925 | } else { |
4926 | res_counter_init(&mem->res, NULL); | |
4927 | res_counter_init(&mem->memsw, NULL); | |
4928 | } | |
04046e1a | 4929 | mem->last_scanned_child = 0; |
889976db | 4930 | mem->last_scanned_node = MAX_NUMNODES; |
9490ff27 | 4931 | INIT_LIST_HEAD(&mem->oom_notify); |
6d61ef40 | 4932 | |
a7885eb8 KM |
4933 | if (parent) |
4934 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 4935 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 4936 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 4937 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 4938 | return &mem->css; |
6d12e2d8 | 4939 | free_out: |
a7ba0eef | 4940 | __mem_cgroup_free(mem); |
4b3bde4c | 4941 | root_mem_cgroup = NULL; |
04046e1a | 4942 | return ERR_PTR(error); |
8cdea7c0 BS |
4943 | } |
4944 | ||
ec64f515 | 4945 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4946 | struct cgroup *cont) |
4947 | { | |
4948 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
4949 | |
4950 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
4951 | } |
4952 | ||
8cdea7c0 BS |
4953 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4954 | struct cgroup *cont) | |
4955 | { | |
c268e994 | 4956 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 4957 | |
c268e994 | 4958 | mem_cgroup_put(mem); |
8cdea7c0 BS |
4959 | } |
4960 | ||
4961 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4962 | struct cgroup *cont) | |
4963 | { | |
8c7c6e34 KH |
4964 | int ret; |
4965 | ||
4966 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4967 | ARRAY_SIZE(mem_cgroup_files)); | |
4968 | ||
4969 | if (!ret) | |
4970 | ret = register_memsw_files(cont, ss); | |
4971 | return ret; | |
8cdea7c0 BS |
4972 | } |
4973 | ||
02491447 | 4974 | #ifdef CONFIG_MMU |
7dc74be0 | 4975 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4976 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4977 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4978 | { |
854ffa8d DN |
4979 | int ret = 0; |
4980 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
4981 | struct mem_cgroup *mem = mc.to; |
4982 | ||
854ffa8d DN |
4983 | if (mem_cgroup_is_root(mem)) { |
4984 | mc.precharge += count; | |
4985 | /* we don't need css_get for root */ | |
4986 | return ret; | |
4987 | } | |
4988 | /* try to charge at once */ | |
4989 | if (count > 1) { | |
4990 | struct res_counter *dummy; | |
4991 | /* | |
4992 | * "mem" cannot be under rmdir() because we've already checked | |
4993 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
4994 | * are still under the same cgroup_mutex. So we can postpone | |
4995 | * css_get(). | |
4996 | */ | |
4997 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
4998 | goto one_by_one; | |
4999 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
5000 | PAGE_SIZE * count, &dummy)) { | |
5001 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
5002 | goto one_by_one; | |
5003 | } | |
5004 | mc.precharge += count; | |
854ffa8d DN |
5005 | return ret; |
5006 | } | |
5007 | one_by_one: | |
5008 | /* fall back to one by one charge */ | |
5009 | while (count--) { | |
5010 | if (signal_pending(current)) { | |
5011 | ret = -EINTR; | |
5012 | break; | |
5013 | } | |
5014 | if (!batch_count--) { | |
5015 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5016 | cond_resched(); | |
5017 | } | |
7ec99d62 | 5018 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, 1, &mem, false); |
854ffa8d DN |
5019 | if (ret || !mem) |
5020 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
5021 | return -ENOMEM; | |
5022 | mc.precharge++; | |
5023 | } | |
4ffef5fe DN |
5024 | return ret; |
5025 | } | |
5026 | ||
5027 | /** | |
5028 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
5029 | * @vma: the vma the pte to be checked belongs | |
5030 | * @addr: the address corresponding to the pte to be checked | |
5031 | * @ptent: the pte to be checked | |
02491447 | 5032 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5033 | * |
5034 | * Returns | |
5035 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5036 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5037 | * move charge. if @target is not NULL, the page is stored in target->page | |
5038 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5039 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5040 | * target for charge migration. if @target is not NULL, the entry is stored | |
5041 | * in target->ent. | |
4ffef5fe DN |
5042 | * |
5043 | * Called with pte lock held. | |
5044 | */ | |
4ffef5fe DN |
5045 | union mc_target { |
5046 | struct page *page; | |
02491447 | 5047 | swp_entry_t ent; |
4ffef5fe DN |
5048 | }; |
5049 | ||
4ffef5fe DN |
5050 | enum mc_target_type { |
5051 | MC_TARGET_NONE, /* not used */ | |
5052 | MC_TARGET_PAGE, | |
02491447 | 5053 | MC_TARGET_SWAP, |
4ffef5fe DN |
5054 | }; |
5055 | ||
90254a65 DN |
5056 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5057 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5058 | { |
90254a65 | 5059 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5060 | |
90254a65 DN |
5061 | if (!page || !page_mapped(page)) |
5062 | return NULL; | |
5063 | if (PageAnon(page)) { | |
5064 | /* we don't move shared anon */ | |
5065 | if (!move_anon() || page_mapcount(page) > 2) | |
5066 | return NULL; | |
87946a72 DN |
5067 | } else if (!move_file()) |
5068 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5069 | return NULL; |
5070 | if (!get_page_unless_zero(page)) | |
5071 | return NULL; | |
5072 | ||
5073 | return page; | |
5074 | } | |
5075 | ||
5076 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5077 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5078 | { | |
5079 | int usage_count; | |
5080 | struct page *page = NULL; | |
5081 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5082 | ||
5083 | if (!move_anon() || non_swap_entry(ent)) | |
5084 | return NULL; | |
5085 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
5086 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
5087 | if (page) |
5088 | put_page(page); | |
90254a65 | 5089 | return NULL; |
02491447 | 5090 | } |
90254a65 DN |
5091 | if (do_swap_account) |
5092 | entry->val = ent.val; | |
5093 | ||
5094 | return page; | |
5095 | } | |
5096 | ||
87946a72 DN |
5097 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5098 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5099 | { | |
5100 | struct page *page = NULL; | |
5101 | struct inode *inode; | |
5102 | struct address_space *mapping; | |
5103 | pgoff_t pgoff; | |
5104 | ||
5105 | if (!vma->vm_file) /* anonymous vma */ | |
5106 | return NULL; | |
5107 | if (!move_file()) | |
5108 | return NULL; | |
5109 | ||
5110 | inode = vma->vm_file->f_path.dentry->d_inode; | |
5111 | mapping = vma->vm_file->f_mapping; | |
5112 | if (pte_none(ptent)) | |
5113 | pgoff = linear_page_index(vma, addr); | |
5114 | else /* pte_file(ptent) is true */ | |
5115 | pgoff = pte_to_pgoff(ptent); | |
5116 | ||
5117 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
5118 | if (!mapping_cap_swap_backed(mapping)) { /* normal file */ | |
5119 | page = find_get_page(mapping, pgoff); | |
5120 | } else { /* shmem/tmpfs file. we should take account of swap too. */ | |
5121 | swp_entry_t ent; | |
5122 | mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent); | |
5123 | if (do_swap_account) | |
5124 | entry->val = ent.val; | |
5125 | } | |
5126 | ||
5127 | return page; | |
5128 | } | |
5129 | ||
90254a65 DN |
5130 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
5131 | unsigned long addr, pte_t ptent, union mc_target *target) | |
5132 | { | |
5133 | struct page *page = NULL; | |
5134 | struct page_cgroup *pc; | |
5135 | int ret = 0; | |
5136 | swp_entry_t ent = { .val = 0 }; | |
5137 | ||
5138 | if (pte_present(ptent)) | |
5139 | page = mc_handle_present_pte(vma, addr, ptent); | |
5140 | else if (is_swap_pte(ptent)) | |
5141 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5142 | else if (pte_none(ptent) || pte_file(ptent)) |
5143 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5144 | |
5145 | if (!page && !ent.val) | |
5146 | return 0; | |
02491447 DN |
5147 | if (page) { |
5148 | pc = lookup_page_cgroup(page); | |
5149 | /* | |
5150 | * Do only loose check w/o page_cgroup lock. | |
5151 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5152 | * the lock. | |
5153 | */ | |
5154 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5155 | ret = MC_TARGET_PAGE; | |
5156 | if (target) | |
5157 | target->page = page; | |
5158 | } | |
5159 | if (!ret || !target) | |
5160 | put_page(page); | |
5161 | } | |
90254a65 DN |
5162 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5163 | if (ent.val && !ret && | |
7f0f1546 KH |
5164 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
5165 | ret = MC_TARGET_SWAP; | |
5166 | if (target) | |
5167 | target->ent = ent; | |
4ffef5fe | 5168 | } |
4ffef5fe DN |
5169 | return ret; |
5170 | } | |
5171 | ||
5172 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
5173 | unsigned long addr, unsigned long end, | |
5174 | struct mm_walk *walk) | |
5175 | { | |
5176 | struct vm_area_struct *vma = walk->private; | |
5177 | pte_t *pte; | |
5178 | spinlock_t *ptl; | |
5179 | ||
03319327 DH |
5180 | split_huge_page_pmd(walk->mm, pmd); |
5181 | ||
4ffef5fe DN |
5182 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5183 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
5184 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
5185 | mc.precharge++; /* increment precharge temporarily */ | |
5186 | pte_unmap_unlock(pte - 1, ptl); | |
5187 | cond_resched(); | |
5188 | ||
7dc74be0 DN |
5189 | return 0; |
5190 | } | |
5191 | ||
4ffef5fe DN |
5192 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5193 | { | |
5194 | unsigned long precharge; | |
5195 | struct vm_area_struct *vma; | |
5196 | ||
dfe076b0 | 5197 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5198 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5199 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5200 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5201 | .mm = mm, | |
5202 | .private = vma, | |
5203 | }; | |
5204 | if (is_vm_hugetlb_page(vma)) | |
5205 | continue; | |
4ffef5fe DN |
5206 | walk_page_range(vma->vm_start, vma->vm_end, |
5207 | &mem_cgroup_count_precharge_walk); | |
5208 | } | |
dfe076b0 | 5209 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5210 | |
5211 | precharge = mc.precharge; | |
5212 | mc.precharge = 0; | |
5213 | ||
5214 | return precharge; | |
5215 | } | |
5216 | ||
4ffef5fe DN |
5217 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5218 | { | |
dfe076b0 DN |
5219 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5220 | ||
5221 | VM_BUG_ON(mc.moving_task); | |
5222 | mc.moving_task = current; | |
5223 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5224 | } |
5225 | ||
dfe076b0 DN |
5226 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5227 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5228 | { |
2bd9bb20 KH |
5229 | struct mem_cgroup *from = mc.from; |
5230 | struct mem_cgroup *to = mc.to; | |
5231 | ||
4ffef5fe | 5232 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5233 | if (mc.precharge) { |
5234 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5235 | mc.precharge = 0; | |
5236 | } | |
5237 | /* | |
5238 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5239 | * we must uncharge here. | |
5240 | */ | |
5241 | if (mc.moved_charge) { | |
5242 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5243 | mc.moved_charge = 0; | |
4ffef5fe | 5244 | } |
483c30b5 DN |
5245 | /* we must fixup refcnts and charges */ |
5246 | if (mc.moved_swap) { | |
483c30b5 DN |
5247 | /* uncharge swap account from the old cgroup */ |
5248 | if (!mem_cgroup_is_root(mc.from)) | |
5249 | res_counter_uncharge(&mc.from->memsw, | |
5250 | PAGE_SIZE * mc.moved_swap); | |
5251 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5252 | ||
5253 | if (!mem_cgroup_is_root(mc.to)) { | |
5254 | /* | |
5255 | * we charged both to->res and to->memsw, so we should | |
5256 | * uncharge to->res. | |
5257 | */ | |
5258 | res_counter_uncharge(&mc.to->res, | |
5259 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5260 | } |
5261 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5262 | mc.moved_swap = 0; |
5263 | } | |
dfe076b0 DN |
5264 | memcg_oom_recover(from); |
5265 | memcg_oom_recover(to); | |
5266 | wake_up_all(&mc.waitq); | |
5267 | } | |
5268 | ||
5269 | static void mem_cgroup_clear_mc(void) | |
5270 | { | |
5271 | struct mem_cgroup *from = mc.from; | |
5272 | ||
5273 | /* | |
5274 | * we must clear moving_task before waking up waiters at the end of | |
5275 | * task migration. | |
5276 | */ | |
5277 | mc.moving_task = NULL; | |
5278 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5279 | spin_lock(&mc.lock); |
4ffef5fe DN |
5280 | mc.from = NULL; |
5281 | mc.to = NULL; | |
2bd9bb20 | 5282 | spin_unlock(&mc.lock); |
32047e2a | 5283 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5284 | } |
5285 | ||
7dc74be0 DN |
5286 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
5287 | struct cgroup *cgroup, | |
f780bdb7 | 5288 | struct task_struct *p) |
7dc74be0 DN |
5289 | { |
5290 | int ret = 0; | |
5291 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
5292 | ||
5293 | if (mem->move_charge_at_immigrate) { | |
5294 | struct mm_struct *mm; | |
5295 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5296 | ||
5297 | VM_BUG_ON(from == mem); | |
5298 | ||
5299 | mm = get_task_mm(p); | |
5300 | if (!mm) | |
5301 | return 0; | |
7dc74be0 | 5302 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5303 | if (mm->owner == p) { |
5304 | VM_BUG_ON(mc.from); | |
5305 | VM_BUG_ON(mc.to); | |
5306 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5307 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5308 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5309 | mem_cgroup_start_move(from); |
2bd9bb20 | 5310 | spin_lock(&mc.lock); |
4ffef5fe DN |
5311 | mc.from = from; |
5312 | mc.to = mem; | |
2bd9bb20 | 5313 | spin_unlock(&mc.lock); |
dfe076b0 | 5314 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5315 | |
5316 | ret = mem_cgroup_precharge_mc(mm); | |
5317 | if (ret) | |
5318 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5319 | } |
5320 | mmput(mm); | |
7dc74be0 DN |
5321 | } |
5322 | return ret; | |
5323 | } | |
5324 | ||
5325 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5326 | struct cgroup *cgroup, | |
f780bdb7 | 5327 | struct task_struct *p) |
7dc74be0 | 5328 | { |
4ffef5fe | 5329 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5330 | } |
5331 | ||
4ffef5fe DN |
5332 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5333 | unsigned long addr, unsigned long end, | |
5334 | struct mm_walk *walk) | |
7dc74be0 | 5335 | { |
4ffef5fe DN |
5336 | int ret = 0; |
5337 | struct vm_area_struct *vma = walk->private; | |
5338 | pte_t *pte; | |
5339 | spinlock_t *ptl; | |
5340 | ||
03319327 | 5341 | split_huge_page_pmd(walk->mm, pmd); |
4ffef5fe DN |
5342 | retry: |
5343 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5344 | for (; addr != end; addr += PAGE_SIZE) { | |
5345 | pte_t ptent = *(pte++); | |
5346 | union mc_target target; | |
5347 | int type; | |
5348 | struct page *page; | |
5349 | struct page_cgroup *pc; | |
02491447 | 5350 | swp_entry_t ent; |
4ffef5fe DN |
5351 | |
5352 | if (!mc.precharge) | |
5353 | break; | |
5354 | ||
5355 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
5356 | switch (type) { | |
5357 | case MC_TARGET_PAGE: | |
5358 | page = target.page; | |
5359 | if (isolate_lru_page(page)) | |
5360 | goto put; | |
5361 | pc = lookup_page_cgroup(page); | |
7ec99d62 JW |
5362 | if (!mem_cgroup_move_account(page, 1, pc, |
5363 | mc.from, mc.to, false)) { | |
4ffef5fe | 5364 | mc.precharge--; |
854ffa8d DN |
5365 | /* we uncharge from mc.from later. */ |
5366 | mc.moved_charge++; | |
4ffef5fe DN |
5367 | } |
5368 | putback_lru_page(page); | |
5369 | put: /* is_target_pte_for_mc() gets the page */ | |
5370 | put_page(page); | |
5371 | break; | |
02491447 DN |
5372 | case MC_TARGET_SWAP: |
5373 | ent = target.ent; | |
483c30b5 DN |
5374 | if (!mem_cgroup_move_swap_account(ent, |
5375 | mc.from, mc.to, false)) { | |
02491447 | 5376 | mc.precharge--; |
483c30b5 DN |
5377 | /* we fixup refcnts and charges later. */ |
5378 | mc.moved_swap++; | |
5379 | } | |
02491447 | 5380 | break; |
4ffef5fe DN |
5381 | default: |
5382 | break; | |
5383 | } | |
5384 | } | |
5385 | pte_unmap_unlock(pte - 1, ptl); | |
5386 | cond_resched(); | |
5387 | ||
5388 | if (addr != end) { | |
5389 | /* | |
5390 | * We have consumed all precharges we got in can_attach(). | |
5391 | * We try charge one by one, but don't do any additional | |
5392 | * charges to mc.to if we have failed in charge once in attach() | |
5393 | * phase. | |
5394 | */ | |
854ffa8d | 5395 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5396 | if (!ret) |
5397 | goto retry; | |
5398 | } | |
5399 | ||
5400 | return ret; | |
5401 | } | |
5402 | ||
5403 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5404 | { | |
5405 | struct vm_area_struct *vma; | |
5406 | ||
5407 | lru_add_drain_all(); | |
dfe076b0 DN |
5408 | retry: |
5409 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5410 | /* | |
5411 | * Someone who are holding the mmap_sem might be waiting in | |
5412 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5413 | * and retry. Because we cancel precharges, we might not be able | |
5414 | * to move enough charges, but moving charge is a best-effort | |
5415 | * feature anyway, so it wouldn't be a big problem. | |
5416 | */ | |
5417 | __mem_cgroup_clear_mc(); | |
5418 | cond_resched(); | |
5419 | goto retry; | |
5420 | } | |
4ffef5fe DN |
5421 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5422 | int ret; | |
5423 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5424 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5425 | .mm = mm, | |
5426 | .private = vma, | |
5427 | }; | |
5428 | if (is_vm_hugetlb_page(vma)) | |
5429 | continue; | |
4ffef5fe DN |
5430 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5431 | &mem_cgroup_move_charge_walk); | |
5432 | if (ret) | |
5433 | /* | |
5434 | * means we have consumed all precharges and failed in | |
5435 | * doing additional charge. Just abandon here. | |
5436 | */ | |
5437 | break; | |
5438 | } | |
dfe076b0 | 5439 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5440 | } |
5441 | ||
67e465a7 BS |
5442 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
5443 | struct cgroup *cont, | |
5444 | struct cgroup *old_cont, | |
f780bdb7 | 5445 | struct task_struct *p) |
67e465a7 | 5446 | { |
a433658c | 5447 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5448 | |
dfe076b0 | 5449 | if (mm) { |
a433658c KM |
5450 | if (mc.to) |
5451 | mem_cgroup_move_charge(mm); | |
5452 | put_swap_token(mm); | |
dfe076b0 DN |
5453 | mmput(mm); |
5454 | } | |
a433658c KM |
5455 | if (mc.to) |
5456 | mem_cgroup_clear_mc(); | |
67e465a7 | 5457 | } |
5cfb80a7 DN |
5458 | #else /* !CONFIG_MMU */ |
5459 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
5460 | struct cgroup *cgroup, | |
f780bdb7 | 5461 | struct task_struct *p) |
5cfb80a7 DN |
5462 | { |
5463 | return 0; | |
5464 | } | |
5465 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5466 | struct cgroup *cgroup, | |
f780bdb7 | 5467 | struct task_struct *p) |
5cfb80a7 DN |
5468 | { |
5469 | } | |
5470 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
5471 | struct cgroup *cont, | |
5472 | struct cgroup *old_cont, | |
f780bdb7 | 5473 | struct task_struct *p) |
5cfb80a7 DN |
5474 | { |
5475 | } | |
5476 | #endif | |
67e465a7 | 5477 | |
8cdea7c0 BS |
5478 | struct cgroup_subsys mem_cgroup_subsys = { |
5479 | .name = "memory", | |
5480 | .subsys_id = mem_cgroup_subsys_id, | |
5481 | .create = mem_cgroup_create, | |
df878fb0 | 5482 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
5483 | .destroy = mem_cgroup_destroy, |
5484 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
5485 | .can_attach = mem_cgroup_can_attach, |
5486 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5487 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5488 | .early_init = 0, |
04046e1a | 5489 | .use_id = 1, |
8cdea7c0 | 5490 | }; |
c077719b KH |
5491 | |
5492 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5493 | static int __init enable_swap_account(char *s) |
5494 | { | |
5495 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5496 | if (!strcmp(s, "1")) |
a42c390c | 5497 | really_do_swap_account = 1; |
a2c8990a | 5498 | else if (!strcmp(s, "0")) |
a42c390c MH |
5499 | really_do_swap_account = 0; |
5500 | return 1; | |
5501 | } | |
a2c8990a | 5502 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5503 | |
c077719b | 5504 | #endif |