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> |
b6ac57d5 | 38 | #include <linux/slab.h> |
66e1707b | 39 | #include <linux/swap.h> |
02491447 | 40 | #include <linux/swapops.h> |
66e1707b | 41 | #include <linux/spinlock.h> |
2e72b634 KS |
42 | #include <linux/eventfd.h> |
43 | #include <linux/sort.h> | |
66e1707b | 44 | #include <linux/fs.h> |
d2ceb9b7 | 45 | #include <linux/seq_file.h> |
33327948 | 46 | #include <linux/vmalloc.h> |
b69408e8 | 47 | #include <linux/mm_inline.h> |
52d4b9ac | 48 | #include <linux/page_cgroup.h> |
cdec2e42 | 49 | #include <linux/cpu.h> |
08e552c6 | 50 | #include "internal.h" |
8cdea7c0 | 51 | |
8697d331 BS |
52 | #include <asm/uaccess.h> |
53 | ||
a181b0e8 | 54 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 55 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 56 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 57 | |
c077719b | 58 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 59 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b KH |
60 | int do_swap_account __read_mostly; |
61 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
62 | #else | |
63 | #define do_swap_account (0) | |
64 | #endif | |
65 | ||
d2265e6f KH |
66 | /* |
67 | * Per memcg event counter is incremented at every pagein/pageout. This counter | |
68 | * is used for trigger some periodic events. This is straightforward and better | |
69 | * than using jiffies etc. to handle periodic memcg event. | |
70 | * | |
71 | * These values will be used as !((event) & ((1 <<(thresh)) - 1)) | |
72 | */ | |
73 | #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */ | |
74 | #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */ | |
c077719b | 75 | |
d52aa412 KH |
76 | /* |
77 | * Statistics for memory cgroup. | |
78 | */ | |
79 | enum mem_cgroup_stat_index { | |
80 | /* | |
81 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
82 | */ | |
83 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 84 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 85 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
55e462b0 BR |
86 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
87 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
0c3e73e8 | 88 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
d2265e6f | 89 | MEM_CGROUP_EVENTS, /* incremented at every pagein/pageout */ |
d52aa412 KH |
90 | |
91 | MEM_CGROUP_STAT_NSTATS, | |
92 | }; | |
93 | ||
94 | struct mem_cgroup_stat_cpu { | |
95 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
d52aa412 KH |
96 | }; |
97 | ||
6d12e2d8 KH |
98 | /* |
99 | * per-zone information in memory controller. | |
100 | */ | |
6d12e2d8 | 101 | struct mem_cgroup_per_zone { |
072c56c1 KH |
102 | /* |
103 | * spin_lock to protect the per cgroup LRU | |
104 | */ | |
b69408e8 CL |
105 | struct list_head lists[NR_LRU_LISTS]; |
106 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
107 | |
108 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
109 | struct rb_node tree_node; /* RB tree node */ |
110 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
111 | /* the soft limit is exceeded*/ | |
112 | bool on_tree; | |
4e416953 BS |
113 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
114 | /* use container_of */ | |
6d12e2d8 KH |
115 | }; |
116 | /* Macro for accessing counter */ | |
117 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
118 | ||
119 | struct mem_cgroup_per_node { | |
120 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
121 | }; | |
122 | ||
123 | struct mem_cgroup_lru_info { | |
124 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
125 | }; | |
126 | ||
f64c3f54 BS |
127 | /* |
128 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
129 | * their hierarchy representation | |
130 | */ | |
131 | ||
132 | struct mem_cgroup_tree_per_zone { | |
133 | struct rb_root rb_root; | |
134 | spinlock_t lock; | |
135 | }; | |
136 | ||
137 | struct mem_cgroup_tree_per_node { | |
138 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
139 | }; | |
140 | ||
141 | struct mem_cgroup_tree { | |
142 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
143 | }; | |
144 | ||
145 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
146 | ||
2e72b634 KS |
147 | struct mem_cgroup_threshold { |
148 | struct eventfd_ctx *eventfd; | |
149 | u64 threshold; | |
150 | }; | |
151 | ||
9490ff27 | 152 | /* For threshold */ |
2e72b634 KS |
153 | struct mem_cgroup_threshold_ary { |
154 | /* An array index points to threshold just below usage. */ | |
5407a562 | 155 | int current_threshold; |
2e72b634 KS |
156 | /* Size of entries[] */ |
157 | unsigned int size; | |
158 | /* Array of thresholds */ | |
159 | struct mem_cgroup_threshold entries[0]; | |
160 | }; | |
2c488db2 KS |
161 | |
162 | struct mem_cgroup_thresholds { | |
163 | /* Primary thresholds array */ | |
164 | struct mem_cgroup_threshold_ary *primary; | |
165 | /* | |
166 | * Spare threshold array. | |
167 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
168 | * It must be able to store at least primary->size - 1 entries. | |
169 | */ | |
170 | struct mem_cgroup_threshold_ary *spare; | |
171 | }; | |
172 | ||
9490ff27 KH |
173 | /* for OOM */ |
174 | struct mem_cgroup_eventfd_list { | |
175 | struct list_head list; | |
176 | struct eventfd_ctx *eventfd; | |
177 | }; | |
2e72b634 | 178 | |
2e72b634 | 179 | static void mem_cgroup_threshold(struct mem_cgroup *mem); |
9490ff27 | 180 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem); |
2e72b634 | 181 | |
8cdea7c0 BS |
182 | /* |
183 | * The memory controller data structure. The memory controller controls both | |
184 | * page cache and RSS per cgroup. We would eventually like to provide | |
185 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
186 | * to help the administrator determine what knobs to tune. | |
187 | * | |
188 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
189 | * we hit the water mark. May be even add a low water mark, such that |
190 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
191 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
192 | */ |
193 | struct mem_cgroup { | |
194 | struct cgroup_subsys_state css; | |
195 | /* | |
196 | * the counter to account for memory usage | |
197 | */ | |
198 | struct res_counter res; | |
8c7c6e34 KH |
199 | /* |
200 | * the counter to account for mem+swap usage. | |
201 | */ | |
202 | struct res_counter memsw; | |
78fb7466 PE |
203 | /* |
204 | * Per cgroup active and inactive list, similar to the | |
205 | * per zone LRU lists. | |
78fb7466 | 206 | */ |
6d12e2d8 | 207 | struct mem_cgroup_lru_info info; |
072c56c1 | 208 | |
2733c06a KM |
209 | /* |
210 | protect against reclaim related member. | |
211 | */ | |
212 | spinlock_t reclaim_param_lock; | |
213 | ||
6d61ef40 | 214 | /* |
af901ca1 | 215 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 216 | * reclaimed from. |
6d61ef40 | 217 | */ |
04046e1a | 218 | int last_scanned_child; |
18f59ea7 BS |
219 | /* |
220 | * Should the accounting and control be hierarchical, per subtree? | |
221 | */ | |
222 | bool use_hierarchy; | |
867578cb | 223 | atomic_t oom_lock; |
8c7c6e34 | 224 | atomic_t refcnt; |
14797e23 | 225 | |
a7885eb8 | 226 | unsigned int swappiness; |
3c11ecf4 KH |
227 | /* OOM-Killer disable */ |
228 | int oom_kill_disable; | |
a7885eb8 | 229 | |
22a668d7 KH |
230 | /* set when res.limit == memsw.limit */ |
231 | bool memsw_is_minimum; | |
232 | ||
2e72b634 KS |
233 | /* protect arrays of thresholds */ |
234 | struct mutex thresholds_lock; | |
235 | ||
236 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 237 | struct mem_cgroup_thresholds thresholds; |
907860ed | 238 | |
2e72b634 | 239 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 240 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 241 | |
9490ff27 KH |
242 | /* For oom notifier event fd */ |
243 | struct list_head oom_notify; | |
244 | ||
7dc74be0 DN |
245 | /* |
246 | * Should we move charges of a task when a task is moved into this | |
247 | * mem_cgroup ? And what type of charges should we move ? | |
248 | */ | |
249 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 250 | /* |
c62b1a3b | 251 | * percpu counter. |
d52aa412 | 252 | */ |
c62b1a3b | 253 | struct mem_cgroup_stat_cpu *stat; |
8cdea7c0 BS |
254 | }; |
255 | ||
7dc74be0 DN |
256 | /* Stuffs for move charges at task migration. */ |
257 | /* | |
258 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
259 | * left-shifted bitmap of these types. | |
260 | */ | |
261 | enum move_type { | |
4ffef5fe | 262 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 263 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
264 | NR_MOVE_TYPE, |
265 | }; | |
266 | ||
4ffef5fe DN |
267 | /* "mc" and its members are protected by cgroup_mutex */ |
268 | static struct move_charge_struct { | |
269 | struct mem_cgroup *from; | |
270 | struct mem_cgroup *to; | |
271 | unsigned long precharge; | |
854ffa8d | 272 | unsigned long moved_charge; |
483c30b5 | 273 | unsigned long moved_swap; |
8033b97c DN |
274 | struct task_struct *moving_task; /* a task moving charges */ |
275 | wait_queue_head_t waitq; /* a waitq for other context */ | |
276 | } mc = { | |
277 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), | |
278 | }; | |
4ffef5fe | 279 | |
90254a65 DN |
280 | static bool move_anon(void) |
281 | { | |
282 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
283 | &mc.to->move_charge_at_immigrate); | |
284 | } | |
285 | ||
87946a72 DN |
286 | static bool move_file(void) |
287 | { | |
288 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
289 | &mc.to->move_charge_at_immigrate); | |
290 | } | |
291 | ||
4e416953 BS |
292 | /* |
293 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
294 | * limit reclaim to prevent infinite loops, if they ever occur. | |
295 | */ | |
296 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
297 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
298 | ||
217bc319 KH |
299 | enum charge_type { |
300 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
301 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 302 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 303 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 304 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 305 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
306 | NR_CHARGE_TYPE, |
307 | }; | |
308 | ||
52d4b9ac KH |
309 | /* only for here (for easy reading.) */ |
310 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
311 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 312 | #define PCGF_LOCK (1UL << PCG_LOCK) |
4b3bde4c BS |
313 | /* Not used, but added here for completeness */ |
314 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
217bc319 | 315 | |
8c7c6e34 KH |
316 | /* for encoding cft->private value on file */ |
317 | #define _MEM (0) | |
318 | #define _MEMSWAP (1) | |
9490ff27 | 319 | #define _OOM_TYPE (2) |
8c7c6e34 KH |
320 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
321 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
322 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
323 | /* Used for OOM nofiier */ |
324 | #define OOM_CONTROL (0) | |
8c7c6e34 | 325 | |
75822b44 BS |
326 | /* |
327 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
328 | */ | |
329 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
330 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
331 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
332 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
333 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
334 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 335 | |
8c7c6e34 KH |
336 | static void mem_cgroup_get(struct mem_cgroup *mem); |
337 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 338 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 339 | static void drain_all_stock_async(void); |
8c7c6e34 | 340 | |
f64c3f54 BS |
341 | static struct mem_cgroup_per_zone * |
342 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
343 | { | |
344 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
345 | } | |
346 | ||
d324236b WF |
347 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
348 | { | |
349 | return &mem->css; | |
350 | } | |
351 | ||
f64c3f54 BS |
352 | static struct mem_cgroup_per_zone * |
353 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
354 | { | |
355 | struct mem_cgroup *mem = pc->mem_cgroup; | |
356 | int nid = page_cgroup_nid(pc); | |
357 | int zid = page_cgroup_zid(pc); | |
358 | ||
359 | if (!mem) | |
360 | return NULL; | |
361 | ||
362 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
363 | } | |
364 | ||
365 | static struct mem_cgroup_tree_per_zone * | |
366 | soft_limit_tree_node_zone(int nid, int zid) | |
367 | { | |
368 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
369 | } | |
370 | ||
371 | static struct mem_cgroup_tree_per_zone * | |
372 | soft_limit_tree_from_page(struct page *page) | |
373 | { | |
374 | int nid = page_to_nid(page); | |
375 | int zid = page_zonenum(page); | |
376 | ||
377 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
378 | } | |
379 | ||
380 | static void | |
4e416953 | 381 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 382 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
383 | struct mem_cgroup_tree_per_zone *mctz, |
384 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
385 | { |
386 | struct rb_node **p = &mctz->rb_root.rb_node; | |
387 | struct rb_node *parent = NULL; | |
388 | struct mem_cgroup_per_zone *mz_node; | |
389 | ||
390 | if (mz->on_tree) | |
391 | return; | |
392 | ||
ef8745c1 KH |
393 | mz->usage_in_excess = new_usage_in_excess; |
394 | if (!mz->usage_in_excess) | |
395 | return; | |
f64c3f54 BS |
396 | while (*p) { |
397 | parent = *p; | |
398 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
399 | tree_node); | |
400 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
401 | p = &(*p)->rb_left; | |
402 | /* | |
403 | * We can't avoid mem cgroups that are over their soft | |
404 | * limit by the same amount | |
405 | */ | |
406 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
407 | p = &(*p)->rb_right; | |
408 | } | |
409 | rb_link_node(&mz->tree_node, parent, p); | |
410 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
411 | mz->on_tree = true; | |
4e416953 BS |
412 | } |
413 | ||
414 | static void | |
415 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
416 | struct mem_cgroup_per_zone *mz, | |
417 | struct mem_cgroup_tree_per_zone *mctz) | |
418 | { | |
419 | if (!mz->on_tree) | |
420 | return; | |
421 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
422 | mz->on_tree = false; | |
423 | } | |
424 | ||
f64c3f54 BS |
425 | static void |
426 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
427 | struct mem_cgroup_per_zone *mz, | |
428 | struct mem_cgroup_tree_per_zone *mctz) | |
429 | { | |
430 | spin_lock(&mctz->lock); | |
4e416953 | 431 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
432 | spin_unlock(&mctz->lock); |
433 | } | |
434 | ||
f64c3f54 BS |
435 | |
436 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
437 | { | |
ef8745c1 | 438 | unsigned long long excess; |
f64c3f54 BS |
439 | struct mem_cgroup_per_zone *mz; |
440 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
441 | int nid = page_to_nid(page); |
442 | int zid = page_zonenum(page); | |
f64c3f54 BS |
443 | mctz = soft_limit_tree_from_page(page); |
444 | ||
445 | /* | |
4e649152 KH |
446 | * Necessary to update all ancestors when hierarchy is used. |
447 | * because their event counter is not touched. | |
f64c3f54 | 448 | */ |
4e649152 KH |
449 | for (; mem; mem = parent_mem_cgroup(mem)) { |
450 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 451 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
452 | /* |
453 | * We have to update the tree if mz is on RB-tree or | |
454 | * mem is over its softlimit. | |
455 | */ | |
ef8745c1 | 456 | if (excess || mz->on_tree) { |
4e649152 KH |
457 | spin_lock(&mctz->lock); |
458 | /* if on-tree, remove it */ | |
459 | if (mz->on_tree) | |
460 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
461 | /* | |
ef8745c1 KH |
462 | * Insert again. mz->usage_in_excess will be updated. |
463 | * If excess is 0, no tree ops. | |
4e649152 | 464 | */ |
ef8745c1 | 465 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
466 | spin_unlock(&mctz->lock); |
467 | } | |
f64c3f54 BS |
468 | } |
469 | } | |
470 | ||
471 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
472 | { | |
473 | int node, zone; | |
474 | struct mem_cgroup_per_zone *mz; | |
475 | struct mem_cgroup_tree_per_zone *mctz; | |
476 | ||
477 | for_each_node_state(node, N_POSSIBLE) { | |
478 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
479 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
480 | mctz = soft_limit_tree_node_zone(node, zone); | |
481 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
482 | } | |
483 | } | |
484 | } | |
485 | ||
4e416953 BS |
486 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
487 | { | |
488 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; | |
489 | } | |
490 | ||
491 | static struct mem_cgroup_per_zone * | |
492 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
493 | { | |
494 | struct rb_node *rightmost = NULL; | |
26251eaf | 495 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
496 | |
497 | retry: | |
26251eaf | 498 | mz = NULL; |
4e416953 BS |
499 | rightmost = rb_last(&mctz->rb_root); |
500 | if (!rightmost) | |
501 | goto done; /* Nothing to reclaim from */ | |
502 | ||
503 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
504 | /* | |
505 | * Remove the node now but someone else can add it back, | |
506 | * we will to add it back at the end of reclaim to its correct | |
507 | * position in the tree. | |
508 | */ | |
509 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
510 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
511 | !css_tryget(&mz->mem->css)) | |
512 | goto retry; | |
513 | done: | |
514 | return mz; | |
515 | } | |
516 | ||
517 | static struct mem_cgroup_per_zone * | |
518 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
519 | { | |
520 | struct mem_cgroup_per_zone *mz; | |
521 | ||
522 | spin_lock(&mctz->lock); | |
523 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
524 | spin_unlock(&mctz->lock); | |
525 | return mz; | |
526 | } | |
527 | ||
c62b1a3b KH |
528 | static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, |
529 | enum mem_cgroup_stat_index idx) | |
530 | { | |
531 | int cpu; | |
532 | s64 val = 0; | |
533 | ||
534 | for_each_possible_cpu(cpu) | |
535 | val += per_cpu(mem->stat->count[idx], cpu); | |
536 | return val; | |
537 | } | |
538 | ||
539 | static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) | |
540 | { | |
541 | s64 ret; | |
542 | ||
543 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); | |
544 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); | |
545 | return ret; | |
546 | } | |
547 | ||
0c3e73e8 BS |
548 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
549 | bool charge) | |
550 | { | |
551 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 552 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
553 | } |
554 | ||
c05555b5 KH |
555 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
556 | struct page_cgroup *pc, | |
557 | bool charge) | |
d52aa412 | 558 | { |
0c3e73e8 | 559 | int val = (charge) ? 1 : -1; |
d52aa412 | 560 | |
c62b1a3b KH |
561 | preempt_disable(); |
562 | ||
c05555b5 | 563 | if (PageCgroupCache(pc)) |
c62b1a3b | 564 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val); |
d52aa412 | 565 | else |
c62b1a3b | 566 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val); |
55e462b0 BR |
567 | |
568 | if (charge) | |
c62b1a3b | 569 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); |
55e462b0 | 570 | else |
c62b1a3b | 571 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); |
d2265e6f | 572 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]); |
2e72b634 | 573 | |
c62b1a3b | 574 | preempt_enable(); |
6d12e2d8 KH |
575 | } |
576 | ||
14067bb3 | 577 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 578 | enum lru_list idx) |
6d12e2d8 KH |
579 | { |
580 | int nid, zid; | |
581 | struct mem_cgroup_per_zone *mz; | |
582 | u64 total = 0; | |
583 | ||
584 | for_each_online_node(nid) | |
585 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
586 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
587 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
588 | } | |
589 | return total; | |
d52aa412 KH |
590 | } |
591 | ||
d2265e6f KH |
592 | static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift) |
593 | { | |
594 | s64 val; | |
595 | ||
596 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]); | |
597 | ||
598 | return !(val & ((1 << event_mask_shift) - 1)); | |
599 | } | |
600 | ||
601 | /* | |
602 | * Check events in order. | |
603 | * | |
604 | */ | |
605 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) | |
606 | { | |
607 | /* threshold event is triggered in finer grain than soft limit */ | |
608 | if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) { | |
609 | mem_cgroup_threshold(mem); | |
610 | if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH))) | |
611 | mem_cgroup_update_tree(mem, page); | |
612 | } | |
613 | } | |
614 | ||
d5b69e38 | 615 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
616 | { |
617 | return container_of(cgroup_subsys_state(cont, | |
618 | mem_cgroup_subsys_id), struct mem_cgroup, | |
619 | css); | |
620 | } | |
621 | ||
cf475ad2 | 622 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 623 | { |
31a78f23 BS |
624 | /* |
625 | * mm_update_next_owner() may clear mm->owner to NULL | |
626 | * if it races with swapoff, page migration, etc. | |
627 | * So this can be called with p == NULL. | |
628 | */ | |
629 | if (unlikely(!p)) | |
630 | return NULL; | |
631 | ||
78fb7466 PE |
632 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
633 | struct mem_cgroup, css); | |
634 | } | |
635 | ||
54595fe2 KH |
636 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
637 | { | |
638 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
639 | |
640 | if (!mm) | |
641 | return NULL; | |
54595fe2 KH |
642 | /* |
643 | * Because we have no locks, mm->owner's may be being moved to other | |
644 | * cgroup. We use css_tryget() here even if this looks | |
645 | * pessimistic (rather than adding locks here). | |
646 | */ | |
647 | rcu_read_lock(); | |
648 | do { | |
649 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
650 | if (unlikely(!mem)) | |
651 | break; | |
652 | } while (!css_tryget(&mem->css)); | |
653 | rcu_read_unlock(); | |
654 | return mem; | |
655 | } | |
656 | ||
14067bb3 KH |
657 | /* |
658 | * Call callback function against all cgroup under hierarchy tree. | |
659 | */ | |
660 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
661 | int (*func)(struct mem_cgroup *, void *)) | |
662 | { | |
663 | int found, ret, nextid; | |
664 | struct cgroup_subsys_state *css; | |
665 | struct mem_cgroup *mem; | |
666 | ||
667 | if (!root->use_hierarchy) | |
668 | return (*func)(root, data); | |
669 | ||
670 | nextid = 1; | |
671 | do { | |
672 | ret = 0; | |
673 | mem = NULL; | |
674 | ||
675 | rcu_read_lock(); | |
676 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
677 | &found); | |
678 | if (css && css_tryget(css)) | |
679 | mem = container_of(css, struct mem_cgroup, css); | |
680 | rcu_read_unlock(); | |
681 | ||
682 | if (mem) { | |
683 | ret = (*func)(mem, data); | |
684 | css_put(&mem->css); | |
685 | } | |
686 | nextid = found + 1; | |
687 | } while (!ret && css); | |
688 | ||
689 | return ret; | |
690 | } | |
691 | ||
4b3bde4c BS |
692 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
693 | { | |
694 | return (mem == root_mem_cgroup); | |
695 | } | |
696 | ||
08e552c6 KH |
697 | /* |
698 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
699 | * Operations are called by routine of global LRU independently from memcg. | |
700 | * What we have to take care of here is validness of pc->mem_cgroup. | |
701 | * | |
702 | * Changes to pc->mem_cgroup happens when | |
703 | * 1. charge | |
704 | * 2. moving account | |
705 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
706 | * It is added to LRU before charge. | |
707 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
708 | * When moving account, the page is not on LRU. It's isolated. | |
709 | */ | |
4f98a2fe | 710 | |
08e552c6 KH |
711 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
712 | { | |
713 | struct page_cgroup *pc; | |
08e552c6 | 714 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 715 | |
f8d66542 | 716 | if (mem_cgroup_disabled()) |
08e552c6 KH |
717 | return; |
718 | pc = lookup_page_cgroup(page); | |
719 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 720 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 721 | return; |
4b3bde4c | 722 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
723 | /* |
724 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
725 | * removed from global LRU. | |
726 | */ | |
08e552c6 | 727 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 728 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
4b3bde4c BS |
729 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
730 | return; | |
731 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 KH |
732 | list_del_init(&pc->lru); |
733 | return; | |
6d12e2d8 KH |
734 | } |
735 | ||
08e552c6 | 736 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 737 | { |
08e552c6 KH |
738 | mem_cgroup_del_lru_list(page, page_lru(page)); |
739 | } | |
b69408e8 | 740 | |
08e552c6 KH |
741 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
742 | { | |
743 | struct mem_cgroup_per_zone *mz; | |
744 | struct page_cgroup *pc; | |
b69408e8 | 745 | |
f8d66542 | 746 | if (mem_cgroup_disabled()) |
08e552c6 | 747 | return; |
6d12e2d8 | 748 | |
08e552c6 | 749 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
750 | /* |
751 | * Used bit is set without atomic ops but after smp_wmb(). | |
752 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
753 | */ | |
08e552c6 | 754 | smp_rmb(); |
4b3bde4c BS |
755 | /* unused or root page is not rotated. */ |
756 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 KH |
757 | return; |
758 | mz = page_cgroup_zoneinfo(pc); | |
759 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
760 | } |
761 | ||
08e552c6 | 762 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 763 | { |
08e552c6 KH |
764 | struct page_cgroup *pc; |
765 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 766 | |
f8d66542 | 767 | if (mem_cgroup_disabled()) |
08e552c6 KH |
768 | return; |
769 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 770 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
bd112db8 DN |
771 | /* |
772 | * Used bit is set without atomic ops but after smp_wmb(). | |
773 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
774 | */ | |
08e552c6 KH |
775 | smp_rmb(); |
776 | if (!PageCgroupUsed(pc)) | |
894bc310 | 777 | return; |
b69408e8 | 778 | |
08e552c6 | 779 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 780 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
4b3bde4c BS |
781 | SetPageCgroupAcctLRU(pc); |
782 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
783 | return; | |
08e552c6 KH |
784 | list_add(&pc->lru, &mz->lists[lru]); |
785 | } | |
544122e5 | 786 | |
08e552c6 | 787 | /* |
544122e5 KH |
788 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
789 | * lru because the page may.be reused after it's fully uncharged (because of | |
790 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
791 | * it again. This function is only used to charge SwapCache. It's done under | |
792 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 793 | */ |
544122e5 | 794 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 795 | { |
544122e5 KH |
796 | unsigned long flags; |
797 | struct zone *zone = page_zone(page); | |
798 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
799 | ||
800 | spin_lock_irqsave(&zone->lru_lock, flags); | |
801 | /* | |
802 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
803 | * is guarded by lock_page() because the page is SwapCache. | |
804 | */ | |
805 | if (!PageCgroupUsed(pc)) | |
806 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
807 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
808 | } |
809 | ||
544122e5 KH |
810 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
811 | { | |
812 | unsigned long flags; | |
813 | struct zone *zone = page_zone(page); | |
814 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
815 | ||
816 | spin_lock_irqsave(&zone->lru_lock, flags); | |
817 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 818 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
819 | mem_cgroup_add_lru_list(page, page_lru(page)); |
820 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
821 | } | |
822 | ||
823 | ||
08e552c6 KH |
824 | void mem_cgroup_move_lists(struct page *page, |
825 | enum lru_list from, enum lru_list to) | |
826 | { | |
f8d66542 | 827 | if (mem_cgroup_disabled()) |
08e552c6 KH |
828 | return; |
829 | mem_cgroup_del_lru_list(page, from); | |
830 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
831 | } |
832 | ||
4c4a2214 DR |
833 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
834 | { | |
835 | int ret; | |
0b7f569e | 836 | struct mem_cgroup *curr = NULL; |
4c4a2214 DR |
837 | |
838 | task_lock(task); | |
0b7f569e KH |
839 | rcu_read_lock(); |
840 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
841 | rcu_read_unlock(); | |
4c4a2214 | 842 | task_unlock(task); |
0b7f569e KH |
843 | if (!curr) |
844 | return 0; | |
d31f56db DN |
845 | /* |
846 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
847 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
848 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
849 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
850 | */ | |
851 | if (mem->use_hierarchy) | |
0b7f569e KH |
852 | ret = css_is_ancestor(&curr->css, &mem->css); |
853 | else | |
854 | ret = (curr == mem); | |
855 | css_put(&curr->css); | |
4c4a2214 DR |
856 | return ret; |
857 | } | |
858 | ||
c772be93 | 859 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
860 | { |
861 | unsigned long active; | |
862 | unsigned long inactive; | |
c772be93 KM |
863 | unsigned long gb; |
864 | unsigned long inactive_ratio; | |
14797e23 | 865 | |
14067bb3 KH |
866 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
867 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 868 | |
c772be93 KM |
869 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
870 | if (gb) | |
871 | inactive_ratio = int_sqrt(10 * gb); | |
872 | else | |
873 | inactive_ratio = 1; | |
874 | ||
875 | if (present_pages) { | |
876 | present_pages[0] = inactive; | |
877 | present_pages[1] = active; | |
878 | } | |
879 | ||
880 | return inactive_ratio; | |
881 | } | |
882 | ||
883 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
884 | { | |
885 | unsigned long active; | |
886 | unsigned long inactive; | |
887 | unsigned long present_pages[2]; | |
888 | unsigned long inactive_ratio; | |
889 | ||
890 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
891 | ||
892 | inactive = present_pages[0]; | |
893 | active = present_pages[1]; | |
894 | ||
895 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
896 | return 1; |
897 | ||
898 | return 0; | |
899 | } | |
900 | ||
56e49d21 RR |
901 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
902 | { | |
903 | unsigned long active; | |
904 | unsigned long inactive; | |
905 | ||
906 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
907 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
908 | ||
909 | return (active > inactive); | |
910 | } | |
911 | ||
a3d8e054 KM |
912 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
913 | struct zone *zone, | |
914 | enum lru_list lru) | |
915 | { | |
916 | int nid = zone->zone_pgdat->node_id; | |
917 | int zid = zone_idx(zone); | |
918 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
919 | ||
920 | return MEM_CGROUP_ZSTAT(mz, lru); | |
921 | } | |
922 | ||
3e2f41f1 KM |
923 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
924 | struct zone *zone) | |
925 | { | |
926 | int nid = zone->zone_pgdat->node_id; | |
927 | int zid = zone_idx(zone); | |
928 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
929 | ||
930 | return &mz->reclaim_stat; | |
931 | } | |
932 | ||
933 | struct zone_reclaim_stat * | |
934 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
935 | { | |
936 | struct page_cgroup *pc; | |
937 | struct mem_cgroup_per_zone *mz; | |
938 | ||
939 | if (mem_cgroup_disabled()) | |
940 | return NULL; | |
941 | ||
942 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
943 | /* |
944 | * Used bit is set without atomic ops but after smp_wmb(). | |
945 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
946 | */ | |
947 | smp_rmb(); | |
948 | if (!PageCgroupUsed(pc)) | |
949 | return NULL; | |
950 | ||
3e2f41f1 KM |
951 | mz = page_cgroup_zoneinfo(pc); |
952 | if (!mz) | |
953 | return NULL; | |
954 | ||
955 | return &mz->reclaim_stat; | |
956 | } | |
957 | ||
66e1707b BS |
958 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
959 | struct list_head *dst, | |
960 | unsigned long *scanned, int order, | |
961 | int mode, struct zone *z, | |
962 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 963 | int active, int file) |
66e1707b BS |
964 | { |
965 | unsigned long nr_taken = 0; | |
966 | struct page *page; | |
967 | unsigned long scan; | |
968 | LIST_HEAD(pc_list); | |
969 | struct list_head *src; | |
ff7283fa | 970 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
971 | int nid = z->zone_pgdat->node_id; |
972 | int zid = zone_idx(z); | |
973 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 974 | int lru = LRU_FILE * file + active; |
2ffebca6 | 975 | int ret; |
66e1707b | 976 | |
cf475ad2 | 977 | BUG_ON(!mem_cont); |
1ecaab2b | 978 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 979 | src = &mz->lists[lru]; |
66e1707b | 980 | |
ff7283fa KH |
981 | scan = 0; |
982 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 983 | if (scan >= nr_to_scan) |
ff7283fa | 984 | break; |
08e552c6 KH |
985 | |
986 | page = pc->page; | |
52d4b9ac KH |
987 | if (unlikely(!PageCgroupUsed(pc))) |
988 | continue; | |
436c6541 | 989 | if (unlikely(!PageLRU(page))) |
ff7283fa | 990 | continue; |
ff7283fa | 991 | |
436c6541 | 992 | scan++; |
2ffebca6 KH |
993 | ret = __isolate_lru_page(page, mode, file); |
994 | switch (ret) { | |
995 | case 0: | |
66e1707b | 996 | list_move(&page->lru, dst); |
2ffebca6 | 997 | mem_cgroup_del_lru(page); |
66e1707b | 998 | nr_taken++; |
2ffebca6 KH |
999 | break; |
1000 | case -EBUSY: | |
1001 | /* we don't affect global LRU but rotate in our LRU */ | |
1002 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1003 | break; | |
1004 | default: | |
1005 | break; | |
66e1707b BS |
1006 | } |
1007 | } | |
1008 | ||
66e1707b BS |
1009 | *scanned = scan; |
1010 | return nr_taken; | |
1011 | } | |
1012 | ||
6d61ef40 BS |
1013 | #define mem_cgroup_from_res_counter(counter, member) \ |
1014 | container_of(counter, struct mem_cgroup, member) | |
1015 | ||
b85a96c0 DN |
1016 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
1017 | { | |
1018 | if (do_swap_account) { | |
1019 | if (res_counter_check_under_limit(&mem->res) && | |
1020 | res_counter_check_under_limit(&mem->memsw)) | |
1021 | return true; | |
1022 | } else | |
1023 | if (res_counter_check_under_limit(&mem->res)) | |
1024 | return true; | |
1025 | return false; | |
1026 | } | |
1027 | ||
a7885eb8 KM |
1028 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1029 | { | |
1030 | struct cgroup *cgrp = memcg->css.cgroup; | |
1031 | unsigned int swappiness; | |
1032 | ||
1033 | /* root ? */ | |
1034 | if (cgrp->parent == NULL) | |
1035 | return vm_swappiness; | |
1036 | ||
1037 | spin_lock(&memcg->reclaim_param_lock); | |
1038 | swappiness = memcg->swappiness; | |
1039 | spin_unlock(&memcg->reclaim_param_lock); | |
1040 | ||
1041 | return swappiness; | |
1042 | } | |
1043 | ||
81d39c20 KH |
1044 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
1045 | { | |
1046 | int *val = data; | |
1047 | (*val)++; | |
1048 | return 0; | |
1049 | } | |
e222432b BS |
1050 | |
1051 | /** | |
6a6135b6 | 1052 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1053 | * @memcg: The memory cgroup that went over limit |
1054 | * @p: Task that is going to be killed | |
1055 | * | |
1056 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1057 | * enabled | |
1058 | */ | |
1059 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1060 | { | |
1061 | struct cgroup *task_cgrp; | |
1062 | struct cgroup *mem_cgrp; | |
1063 | /* | |
1064 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1065 | * on the assumption that OOM is serialized for memory controller. | |
1066 | * If this assumption is broken, revisit this code. | |
1067 | */ | |
1068 | static char memcg_name[PATH_MAX]; | |
1069 | int ret; | |
1070 | ||
d31f56db | 1071 | if (!memcg || !p) |
e222432b BS |
1072 | return; |
1073 | ||
1074 | ||
1075 | rcu_read_lock(); | |
1076 | ||
1077 | mem_cgrp = memcg->css.cgroup; | |
1078 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1079 | ||
1080 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1081 | if (ret < 0) { | |
1082 | /* | |
1083 | * Unfortunately, we are unable to convert to a useful name | |
1084 | * But we'll still print out the usage information | |
1085 | */ | |
1086 | rcu_read_unlock(); | |
1087 | goto done; | |
1088 | } | |
1089 | rcu_read_unlock(); | |
1090 | ||
1091 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1092 | ||
1093 | rcu_read_lock(); | |
1094 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1095 | if (ret < 0) { | |
1096 | rcu_read_unlock(); | |
1097 | goto done; | |
1098 | } | |
1099 | rcu_read_unlock(); | |
1100 | ||
1101 | /* | |
1102 | * Continues from above, so we don't need an KERN_ level | |
1103 | */ | |
1104 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1105 | done: | |
1106 | ||
1107 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1108 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1109 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1110 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1111 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1112 | "failcnt %llu\n", | |
1113 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1114 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1115 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1116 | } | |
1117 | ||
81d39c20 KH |
1118 | /* |
1119 | * This function returns the number of memcg under hierarchy tree. Returns | |
1120 | * 1(self count) if no children. | |
1121 | */ | |
1122 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1123 | { | |
1124 | int num = 0; | |
1125 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
1126 | return num; | |
1127 | } | |
1128 | ||
6d61ef40 | 1129 | /* |
04046e1a KH |
1130 | * Visit the first child (need not be the first child as per the ordering |
1131 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1132 | * that to reclaim free pages from. | |
1133 | */ | |
1134 | static struct mem_cgroup * | |
1135 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1136 | { | |
1137 | struct mem_cgroup *ret = NULL; | |
1138 | struct cgroup_subsys_state *css; | |
1139 | int nextid, found; | |
1140 | ||
1141 | if (!root_mem->use_hierarchy) { | |
1142 | css_get(&root_mem->css); | |
1143 | ret = root_mem; | |
1144 | } | |
1145 | ||
1146 | while (!ret) { | |
1147 | rcu_read_lock(); | |
1148 | nextid = root_mem->last_scanned_child + 1; | |
1149 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1150 | &found); | |
1151 | if (css && css_tryget(css)) | |
1152 | ret = container_of(css, struct mem_cgroup, css); | |
1153 | ||
1154 | rcu_read_unlock(); | |
1155 | /* Updates scanning parameter */ | |
1156 | spin_lock(&root_mem->reclaim_param_lock); | |
1157 | if (!css) { | |
1158 | /* this means start scan from ID:1 */ | |
1159 | root_mem->last_scanned_child = 0; | |
1160 | } else | |
1161 | root_mem->last_scanned_child = found; | |
1162 | spin_unlock(&root_mem->reclaim_param_lock); | |
1163 | } | |
1164 | ||
1165 | return ret; | |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1170 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1171 | * based on its position in the children list. | |
6d61ef40 BS |
1172 | * |
1173 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1174 | * |
1175 | * We give up and return to the caller when we visit root_mem twice. | |
1176 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1177 | * |
1178 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1179 | */ |
1180 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1181 | struct zone *zone, |
75822b44 BS |
1182 | gfp_t gfp_mask, |
1183 | unsigned long reclaim_options) | |
6d61ef40 | 1184 | { |
04046e1a KH |
1185 | struct mem_cgroup *victim; |
1186 | int ret, total = 0; | |
1187 | int loop = 0; | |
75822b44 BS |
1188 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1189 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 BS |
1190 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
1191 | unsigned long excess = mem_cgroup_get_excess(root_mem); | |
04046e1a | 1192 | |
22a668d7 KH |
1193 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1194 | if (root_mem->memsw_is_minimum) | |
1195 | noswap = true; | |
1196 | ||
4e416953 | 1197 | while (1) { |
04046e1a | 1198 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1199 | if (victim == root_mem) { |
04046e1a | 1200 | loop++; |
cdec2e42 KH |
1201 | if (loop >= 1) |
1202 | drain_all_stock_async(); | |
4e416953 BS |
1203 | if (loop >= 2) { |
1204 | /* | |
1205 | * If we have not been able to reclaim | |
1206 | * anything, it might because there are | |
1207 | * no reclaimable pages under this hierarchy | |
1208 | */ | |
1209 | if (!check_soft || !total) { | |
1210 | css_put(&victim->css); | |
1211 | break; | |
1212 | } | |
1213 | /* | |
1214 | * We want to do more targetted reclaim. | |
1215 | * excess >> 2 is not to excessive so as to | |
1216 | * reclaim too much, nor too less that we keep | |
1217 | * coming back to reclaim from this cgroup | |
1218 | */ | |
1219 | if (total >= (excess >> 2) || | |
1220 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1221 | css_put(&victim->css); | |
1222 | break; | |
1223 | } | |
1224 | } | |
1225 | } | |
c62b1a3b | 1226 | if (!mem_cgroup_local_usage(victim)) { |
04046e1a KH |
1227 | /* this cgroup's local usage == 0 */ |
1228 | css_put(&victim->css); | |
6d61ef40 BS |
1229 | continue; |
1230 | } | |
04046e1a | 1231 | /* we use swappiness of local cgroup */ |
4e416953 BS |
1232 | if (check_soft) |
1233 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
1234 | noswap, get_swappiness(victim), zone, | |
1235 | zone->zone_pgdat->node_id); | |
1236 | else | |
1237 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1238 | noswap, get_swappiness(victim)); | |
04046e1a | 1239 | css_put(&victim->css); |
81d39c20 KH |
1240 | /* |
1241 | * At shrinking usage, we can't check we should stop here or | |
1242 | * reclaim more. It's depends on callers. last_scanned_child | |
1243 | * will work enough for keeping fairness under tree. | |
1244 | */ | |
1245 | if (shrink) | |
1246 | return ret; | |
04046e1a | 1247 | total += ret; |
4e416953 BS |
1248 | if (check_soft) { |
1249 | if (res_counter_check_under_soft_limit(&root_mem->res)) | |
1250 | return total; | |
1251 | } else if (mem_cgroup_check_under_limit(root_mem)) | |
04046e1a | 1252 | return 1 + total; |
6d61ef40 | 1253 | } |
04046e1a | 1254 | return total; |
6d61ef40 BS |
1255 | } |
1256 | ||
867578cb | 1257 | static int mem_cgroup_oom_lock_cb(struct mem_cgroup *mem, void *data) |
a636b327 | 1258 | { |
867578cb KH |
1259 | int *val = (int *)data; |
1260 | int x; | |
1261 | /* | |
1262 | * Logically, we can stop scanning immediately when we find | |
1263 | * a memcg is already locked. But condidering unlock ops and | |
1264 | * creation/removal of memcg, scan-all is simple operation. | |
1265 | */ | |
1266 | x = atomic_inc_return(&mem->oom_lock); | |
1267 | *val = max(x, *val); | |
1268 | return 0; | |
1269 | } | |
1270 | /* | |
1271 | * Check OOM-Killer is already running under our hierarchy. | |
1272 | * If someone is running, return false. | |
1273 | */ | |
1274 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) | |
1275 | { | |
1276 | int lock_count = 0; | |
a636b327 | 1277 | |
867578cb KH |
1278 | mem_cgroup_walk_tree(mem, &lock_count, mem_cgroup_oom_lock_cb); |
1279 | ||
1280 | if (lock_count == 1) | |
1281 | return true; | |
1282 | return false; | |
a636b327 | 1283 | } |
0b7f569e | 1284 | |
867578cb | 1285 | static int mem_cgroup_oom_unlock_cb(struct mem_cgroup *mem, void *data) |
0b7f569e | 1286 | { |
867578cb KH |
1287 | /* |
1288 | * When a new child is created while the hierarchy is under oom, | |
1289 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1290 | * atomic_add_unless() here. | |
1291 | */ | |
1292 | atomic_add_unless(&mem->oom_lock, -1, 0); | |
0b7f569e KH |
1293 | return 0; |
1294 | } | |
1295 | ||
867578cb KH |
1296 | static void mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
1297 | { | |
1298 | mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_unlock_cb); | |
1299 | } | |
1300 | ||
1301 | static DEFINE_MUTEX(memcg_oom_mutex); | |
1302 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); | |
1303 | ||
dc98df5a KH |
1304 | struct oom_wait_info { |
1305 | struct mem_cgroup *mem; | |
1306 | wait_queue_t wait; | |
1307 | }; | |
1308 | ||
1309 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1310 | unsigned mode, int sync, void *arg) | |
1311 | { | |
1312 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; | |
1313 | struct oom_wait_info *oom_wait_info; | |
1314 | ||
1315 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
1316 | ||
1317 | if (oom_wait_info->mem == wake_mem) | |
1318 | goto wakeup; | |
1319 | /* if no hierarchy, no match */ | |
1320 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) | |
1321 | return 0; | |
1322 | /* | |
1323 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1324 | * Then we can use css_is_ancestor without taking care of RCU. | |
1325 | */ | |
1326 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && | |
1327 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) | |
1328 | return 0; | |
1329 | ||
1330 | wakeup: | |
1331 | return autoremove_wake_function(wait, mode, sync, arg); | |
1332 | } | |
1333 | ||
1334 | static void memcg_wakeup_oom(struct mem_cgroup *mem) | |
1335 | { | |
1336 | /* for filtering, pass "mem" as argument. */ | |
1337 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); | |
1338 | } | |
1339 | ||
3c11ecf4 KH |
1340 | static void memcg_oom_recover(struct mem_cgroup *mem) |
1341 | { | |
4d845ebf | 1342 | if (atomic_read(&mem->oom_lock)) |
3c11ecf4 KH |
1343 | memcg_wakeup_oom(mem); |
1344 | } | |
1345 | ||
867578cb KH |
1346 | /* |
1347 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1348 | */ | |
1349 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) | |
0b7f569e | 1350 | { |
dc98df5a | 1351 | struct oom_wait_info owait; |
3c11ecf4 | 1352 | bool locked, need_to_kill; |
867578cb | 1353 | |
dc98df5a KH |
1354 | owait.mem = mem; |
1355 | owait.wait.flags = 0; | |
1356 | owait.wait.func = memcg_oom_wake_function; | |
1357 | owait.wait.private = current; | |
1358 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1359 | need_to_kill = true; |
867578cb KH |
1360 | /* At first, try to OOM lock hierarchy under mem.*/ |
1361 | mutex_lock(&memcg_oom_mutex); | |
1362 | locked = mem_cgroup_oom_lock(mem); | |
1363 | /* | |
1364 | * Even if signal_pending(), we can't quit charge() loop without | |
1365 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1366 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1367 | */ | |
3c11ecf4 KH |
1368 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
1369 | if (!locked || mem->oom_kill_disable) | |
1370 | need_to_kill = false; | |
1371 | if (locked) | |
9490ff27 | 1372 | mem_cgroup_oom_notify(mem); |
867578cb KH |
1373 | mutex_unlock(&memcg_oom_mutex); |
1374 | ||
3c11ecf4 KH |
1375 | if (need_to_kill) { |
1376 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
867578cb | 1377 | mem_cgroup_out_of_memory(mem, mask); |
3c11ecf4 | 1378 | } else { |
867578cb | 1379 | schedule(); |
dc98df5a | 1380 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb KH |
1381 | } |
1382 | mutex_lock(&memcg_oom_mutex); | |
1383 | mem_cgroup_oom_unlock(mem); | |
dc98df5a | 1384 | memcg_wakeup_oom(mem); |
867578cb KH |
1385 | mutex_unlock(&memcg_oom_mutex); |
1386 | ||
1387 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) | |
1388 | return false; | |
1389 | /* Give chance to dying process */ | |
1390 | schedule_timeout(1); | |
1391 | return true; | |
0b7f569e KH |
1392 | } |
1393 | ||
d69b042f BS |
1394 | /* |
1395 | * Currently used to update mapped file statistics, but the routine can be | |
1396 | * generalized to update other statistics as well. | |
1397 | */ | |
d8046582 | 1398 | void mem_cgroup_update_file_mapped(struct page *page, int val) |
d69b042f BS |
1399 | { |
1400 | struct mem_cgroup *mem; | |
d69b042f BS |
1401 | struct page_cgroup *pc; |
1402 | ||
d69b042f BS |
1403 | pc = lookup_page_cgroup(page); |
1404 | if (unlikely(!pc)) | |
1405 | return; | |
1406 | ||
1407 | lock_page_cgroup(pc); | |
1408 | mem = pc->mem_cgroup; | |
8725d541 | 1409 | if (!mem || !PageCgroupUsed(pc)) |
d69b042f BS |
1410 | goto done; |
1411 | ||
1412 | /* | |
c62b1a3b | 1413 | * Preemption is already disabled. We can use __this_cpu_xxx |
d69b042f | 1414 | */ |
8725d541 KH |
1415 | if (val > 0) { |
1416 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1417 | SetPageCgroupFileMapped(pc); | |
1418 | } else { | |
1419 | __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1420 | ClearPageCgroupFileMapped(pc); | |
1421 | } | |
d69b042f | 1422 | |
d69b042f BS |
1423 | done: |
1424 | unlock_page_cgroup(pc); | |
1425 | } | |
0b7f569e | 1426 | |
cdec2e42 KH |
1427 | /* |
1428 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1429 | * TODO: maybe necessary to use big numbers in big irons. | |
1430 | */ | |
1431 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1432 | struct memcg_stock_pcp { | |
1433 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1434 | int charge; | |
1435 | struct work_struct work; | |
1436 | }; | |
1437 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1438 | static atomic_t memcg_drain_count; | |
1439 | ||
1440 | /* | |
1441 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1442 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1443 | * cgroup which is not current target, returns false. This stock will be | |
1444 | * refilled. | |
1445 | */ | |
1446 | static bool consume_stock(struct mem_cgroup *mem) | |
1447 | { | |
1448 | struct memcg_stock_pcp *stock; | |
1449 | bool ret = true; | |
1450 | ||
1451 | stock = &get_cpu_var(memcg_stock); | |
1452 | if (mem == stock->cached && stock->charge) | |
1453 | stock->charge -= PAGE_SIZE; | |
1454 | else /* need to call res_counter_charge */ | |
1455 | ret = false; | |
1456 | put_cpu_var(memcg_stock); | |
1457 | return ret; | |
1458 | } | |
1459 | ||
1460 | /* | |
1461 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1462 | */ | |
1463 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1464 | { | |
1465 | struct mem_cgroup *old = stock->cached; | |
1466 | ||
1467 | if (stock->charge) { | |
1468 | res_counter_uncharge(&old->res, stock->charge); | |
1469 | if (do_swap_account) | |
1470 | res_counter_uncharge(&old->memsw, stock->charge); | |
1471 | } | |
1472 | stock->cached = NULL; | |
1473 | stock->charge = 0; | |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * This must be called under preempt disabled or must be called by | |
1478 | * a thread which is pinned to local cpu. | |
1479 | */ | |
1480 | static void drain_local_stock(struct work_struct *dummy) | |
1481 | { | |
1482 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1483 | drain_stock(stock); | |
1484 | } | |
1485 | ||
1486 | /* | |
1487 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 1488 | * This will be consumed by consume_stock() function, later. |
cdec2e42 KH |
1489 | */ |
1490 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1491 | { | |
1492 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1493 | ||
1494 | if (stock->cached != mem) { /* reset if necessary */ | |
1495 | drain_stock(stock); | |
1496 | stock->cached = mem; | |
1497 | } | |
1498 | stock->charge += val; | |
1499 | put_cpu_var(memcg_stock); | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1504 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1505 | * expects some charges will be back to res_counter later but cannot wait for | |
1506 | * it. | |
1507 | */ | |
1508 | static void drain_all_stock_async(void) | |
1509 | { | |
1510 | int cpu; | |
1511 | /* This function is for scheduling "drain" in asynchronous way. | |
1512 | * The result of "drain" is not directly handled by callers. Then, | |
1513 | * if someone is calling drain, we don't have to call drain more. | |
1514 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1515 | * there is a race. We just do loose check here. | |
1516 | */ | |
1517 | if (atomic_read(&memcg_drain_count)) | |
1518 | return; | |
1519 | /* Notify other cpus that system-wide "drain" is running */ | |
1520 | atomic_inc(&memcg_drain_count); | |
1521 | get_online_cpus(); | |
1522 | for_each_online_cpu(cpu) { | |
1523 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1524 | schedule_work_on(cpu, &stock->work); | |
1525 | } | |
1526 | put_online_cpus(); | |
1527 | atomic_dec(&memcg_drain_count); | |
1528 | /* We don't wait for flush_work */ | |
1529 | } | |
1530 | ||
1531 | /* This is a synchronous drain interface. */ | |
1532 | static void drain_all_stock_sync(void) | |
1533 | { | |
1534 | /* called when force_empty is called */ | |
1535 | atomic_inc(&memcg_drain_count); | |
1536 | schedule_on_each_cpu(drain_local_stock); | |
1537 | atomic_dec(&memcg_drain_count); | |
1538 | } | |
1539 | ||
1540 | static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb, | |
1541 | unsigned long action, | |
1542 | void *hcpu) | |
1543 | { | |
1544 | int cpu = (unsigned long)hcpu; | |
1545 | struct memcg_stock_pcp *stock; | |
1546 | ||
1547 | if (action != CPU_DEAD) | |
1548 | return NOTIFY_OK; | |
1549 | stock = &per_cpu(memcg_stock, cpu); | |
1550 | drain_stock(stock); | |
1551 | return NOTIFY_OK; | |
1552 | } | |
1553 | ||
f817ed48 KH |
1554 | /* |
1555 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1556 | * oom-killer can be invoked. | |
8a9f3ccd | 1557 | */ |
f817ed48 | 1558 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
430e4863 | 1559 | gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom) |
8a9f3ccd | 1560 | { |
4e649152 | 1561 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 1562 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
4e649152 | 1563 | struct res_counter *fail_res; |
cdec2e42 | 1564 | int csize = CHARGE_SIZE; |
a636b327 | 1565 | |
867578cb KH |
1566 | /* |
1567 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
1568 | * in system level. So, allow to go ahead dying process in addition to | |
1569 | * MEMDIE process. | |
1570 | */ | |
1571 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
1572 | || fatal_signal_pending(current))) | |
1573 | goto bypass; | |
a636b327 | 1574 | |
8a9f3ccd | 1575 | /* |
3be91277 HD |
1576 | * We always charge the cgroup the mm_struct belongs to. |
1577 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
1578 | * thread group leader migrates. It's possible that mm is not |
1579 | * set, if so charge the init_mm (happens for pagecache usage). | |
1580 | */ | |
54595fe2 KH |
1581 | mem = *memcg; |
1582 | if (likely(!mem)) { | |
1583 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 1584 | *memcg = mem; |
e8589cc1 | 1585 | } else { |
7a81b88c | 1586 | css_get(&mem->css); |
e8589cc1 | 1587 | } |
54595fe2 KH |
1588 | if (unlikely(!mem)) |
1589 | return 0; | |
1590 | ||
46f7e602 | 1591 | VM_BUG_ON(css_is_removed(&mem->css)); |
cdec2e42 KH |
1592 | if (mem_cgroup_is_root(mem)) |
1593 | goto done; | |
8a9f3ccd | 1594 | |
8c7c6e34 | 1595 | while (1) { |
0c3e73e8 | 1596 | int ret = 0; |
75822b44 | 1597 | unsigned long flags = 0; |
7a81b88c | 1598 | |
cdec2e42 | 1599 | if (consume_stock(mem)) |
430e4863 | 1600 | goto done; |
cdec2e42 KH |
1601 | |
1602 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
8c7c6e34 KH |
1603 | if (likely(!ret)) { |
1604 | if (!do_swap_account) | |
1605 | break; | |
cdec2e42 | 1606 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); |
8c7c6e34 KH |
1607 | if (likely(!ret)) |
1608 | break; | |
1609 | /* mem+swap counter fails */ | |
cdec2e42 | 1610 | res_counter_uncharge(&mem->res, csize); |
75822b44 | 1611 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
6d61ef40 BS |
1612 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1613 | memsw); | |
1614 | } else | |
1615 | /* mem counter fails */ | |
1616 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1617 | res); | |
1618 | ||
cdec2e42 KH |
1619 | /* reduce request size and retry */ |
1620 | if (csize > PAGE_SIZE) { | |
1621 | csize = PAGE_SIZE; | |
1622 | continue; | |
1623 | } | |
3be91277 | 1624 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 1625 | goto nomem; |
e1a1cd59 | 1626 | |
4e416953 BS |
1627 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
1628 | gfp_mask, flags); | |
4d1c6273 DN |
1629 | if (ret) |
1630 | continue; | |
66e1707b BS |
1631 | |
1632 | /* | |
8869b8f6 HD |
1633 | * try_to_free_mem_cgroup_pages() might not give us a full |
1634 | * picture of reclaim. Some pages are reclaimed and might be | |
1635 | * moved to swap cache or just unmapped from the cgroup. | |
1636 | * Check the limit again to see if the reclaim reduced the | |
1637 | * current usage of the cgroup before giving up | |
8c7c6e34 | 1638 | * |
8869b8f6 | 1639 | */ |
b85a96c0 DN |
1640 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
1641 | continue; | |
3be91277 | 1642 | |
8033b97c DN |
1643 | /* try to avoid oom while someone is moving charge */ |
1644 | if (mc.moving_task && current != mc.moving_task) { | |
1645 | struct mem_cgroup *from, *to; | |
1646 | bool do_continue = false; | |
1647 | /* | |
1648 | * There is a small race that "from" or "to" can be | |
1649 | * freed by rmdir, so we use css_tryget(). | |
1650 | */ | |
8033b97c DN |
1651 | from = mc.from; |
1652 | to = mc.to; | |
1653 | if (from && css_tryget(&from->css)) { | |
1654 | if (mem_over_limit->use_hierarchy) | |
1655 | do_continue = css_is_ancestor( | |
1656 | &from->css, | |
1657 | &mem_over_limit->css); | |
1658 | else | |
1659 | do_continue = (from == mem_over_limit); | |
1660 | css_put(&from->css); | |
1661 | } | |
1662 | if (!do_continue && to && css_tryget(&to->css)) { | |
1663 | if (mem_over_limit->use_hierarchy) | |
1664 | do_continue = css_is_ancestor( | |
1665 | &to->css, | |
1666 | &mem_over_limit->css); | |
1667 | else | |
1668 | do_continue = (to == mem_over_limit); | |
1669 | css_put(&to->css); | |
1670 | } | |
8033b97c DN |
1671 | if (do_continue) { |
1672 | DEFINE_WAIT(wait); | |
1673 | prepare_to_wait(&mc.waitq, &wait, | |
1674 | TASK_INTERRUPTIBLE); | |
1675 | /* moving charge context might have finished. */ | |
1676 | if (mc.moving_task) | |
1677 | schedule(); | |
1678 | finish_wait(&mc.waitq, &wait); | |
1679 | continue; | |
1680 | } | |
1681 | } | |
1682 | ||
3be91277 | 1683 | if (!nr_retries--) { |
867578cb KH |
1684 | if (!oom) |
1685 | goto nomem; | |
1686 | if (mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) { | |
1687 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
1688 | continue; | |
a636b327 | 1689 | } |
867578cb KH |
1690 | /* When we reach here, current task is dying .*/ |
1691 | css_put(&mem->css); | |
1692 | goto bypass; | |
66e1707b | 1693 | } |
8a9f3ccd | 1694 | } |
cdec2e42 KH |
1695 | if (csize > PAGE_SIZE) |
1696 | refill_stock(mem, csize - PAGE_SIZE); | |
0c3e73e8 | 1697 | done: |
7a81b88c KH |
1698 | return 0; |
1699 | nomem: | |
1700 | css_put(&mem->css); | |
1701 | return -ENOMEM; | |
867578cb KH |
1702 | bypass: |
1703 | *memcg = NULL; | |
1704 | return 0; | |
7a81b88c | 1705 | } |
8a9f3ccd | 1706 | |
a3032a2c DN |
1707 | /* |
1708 | * Somemtimes we have to undo a charge we got by try_charge(). | |
1709 | * This function is for that and do uncharge, put css's refcnt. | |
1710 | * gotten by try_charge(). | |
1711 | */ | |
854ffa8d DN |
1712 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
1713 | unsigned long count) | |
a3032a2c DN |
1714 | { |
1715 | if (!mem_cgroup_is_root(mem)) { | |
854ffa8d | 1716 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
a3032a2c | 1717 | if (do_swap_account) |
854ffa8d DN |
1718 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
1719 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
1720 | WARN_ON_ONCE(count > INT_MAX); | |
1721 | __css_put(&mem->css, (int)count); | |
a3032a2c | 1722 | } |
854ffa8d DN |
1723 | /* we don't need css_put for root */ |
1724 | } | |
1725 | ||
1726 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem) | |
1727 | { | |
1728 | __mem_cgroup_cancel_charge(mem, 1); | |
a3032a2c DN |
1729 | } |
1730 | ||
a3b2d692 KH |
1731 | /* |
1732 | * A helper function to get mem_cgroup from ID. must be called under | |
1733 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
1734 | * it's concern. (dropping refcnt from swap can be called against removed | |
1735 | * memcg.) | |
1736 | */ | |
1737 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
1738 | { | |
1739 | struct cgroup_subsys_state *css; | |
1740 | ||
1741 | /* ID 0 is unused ID */ | |
1742 | if (!id) | |
1743 | return NULL; | |
1744 | css = css_lookup(&mem_cgroup_subsys, id); | |
1745 | if (!css) | |
1746 | return NULL; | |
1747 | return container_of(css, struct mem_cgroup, css); | |
1748 | } | |
1749 | ||
e42d9d5d | 1750 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 1751 | { |
e42d9d5d | 1752 | struct mem_cgroup *mem = NULL; |
3c776e64 | 1753 | struct page_cgroup *pc; |
a3b2d692 | 1754 | unsigned short id; |
b5a84319 KH |
1755 | swp_entry_t ent; |
1756 | ||
3c776e64 DN |
1757 | VM_BUG_ON(!PageLocked(page)); |
1758 | ||
3c776e64 | 1759 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 1760 | lock_page_cgroup(pc); |
a3b2d692 | 1761 | if (PageCgroupUsed(pc)) { |
3c776e64 | 1762 | mem = pc->mem_cgroup; |
a3b2d692 KH |
1763 | if (mem && !css_tryget(&mem->css)) |
1764 | mem = NULL; | |
e42d9d5d | 1765 | } else if (PageSwapCache(page)) { |
3c776e64 | 1766 | ent.val = page_private(page); |
a3b2d692 KH |
1767 | id = lookup_swap_cgroup(ent); |
1768 | rcu_read_lock(); | |
1769 | mem = mem_cgroup_lookup(id); | |
1770 | if (mem && !css_tryget(&mem->css)) | |
1771 | mem = NULL; | |
1772 | rcu_read_unlock(); | |
3c776e64 | 1773 | } |
c0bd3f63 | 1774 | unlock_page_cgroup(pc); |
b5a84319 KH |
1775 | return mem; |
1776 | } | |
1777 | ||
7a81b88c | 1778 | /* |
a5e924f5 | 1779 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
1780 | * USED state. If already USED, uncharge and return. |
1781 | */ | |
1782 | ||
1783 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
1784 | struct page_cgroup *pc, | |
1785 | enum charge_type ctype) | |
1786 | { | |
7a81b88c KH |
1787 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
1788 | if (!mem) | |
1789 | return; | |
52d4b9ac KH |
1790 | |
1791 | lock_page_cgroup(pc); | |
1792 | if (unlikely(PageCgroupUsed(pc))) { | |
1793 | unlock_page_cgroup(pc); | |
a3032a2c | 1794 | mem_cgroup_cancel_charge(mem); |
7a81b88c | 1795 | return; |
52d4b9ac | 1796 | } |
4b3bde4c | 1797 | |
8a9f3ccd | 1798 | pc->mem_cgroup = mem; |
261fb61a KH |
1799 | /* |
1800 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
1801 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
1802 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
1803 | * before USED bit, we need memory barrier here. | |
1804 | * See mem_cgroup_add_lru_list(), etc. | |
1805 | */ | |
08e552c6 | 1806 | smp_wmb(); |
4b3bde4c BS |
1807 | switch (ctype) { |
1808 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
1809 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
1810 | SetPageCgroupCache(pc); | |
1811 | SetPageCgroupUsed(pc); | |
1812 | break; | |
1813 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1814 | ClearPageCgroupCache(pc); | |
1815 | SetPageCgroupUsed(pc); | |
1816 | break; | |
1817 | default: | |
1818 | break; | |
1819 | } | |
3be91277 | 1820 | |
08e552c6 | 1821 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 1822 | |
52d4b9ac | 1823 | unlock_page_cgroup(pc); |
430e4863 KH |
1824 | /* |
1825 | * "charge_statistics" updated event counter. Then, check it. | |
1826 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
1827 | * if they exceeds softlimit. | |
1828 | */ | |
d2265e6f | 1829 | memcg_check_events(mem, pc->page); |
7a81b88c | 1830 | } |
66e1707b | 1831 | |
f817ed48 | 1832 | /** |
57f9fd7d | 1833 | * __mem_cgroup_move_account - move account of the page |
f817ed48 KH |
1834 | * @pc: page_cgroup of the page. |
1835 | * @from: mem_cgroup which the page is moved from. | |
1836 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 1837 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
1838 | * |
1839 | * The caller must confirm following. | |
08e552c6 | 1840 | * - page is not on LRU (isolate_page() is useful.) |
57f9fd7d | 1841 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
f817ed48 | 1842 | * |
854ffa8d DN |
1843 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
1844 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is | |
1845 | * true, this function does "uncharge" from old cgroup, but it doesn't if | |
1846 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 KH |
1847 | */ |
1848 | ||
57f9fd7d | 1849 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
854ffa8d | 1850 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
f817ed48 | 1851 | { |
f817ed48 | 1852 | VM_BUG_ON(from == to); |
08e552c6 | 1853 | VM_BUG_ON(PageLRU(pc->page)); |
57f9fd7d DN |
1854 | VM_BUG_ON(!PageCgroupLocked(pc)); |
1855 | VM_BUG_ON(!PageCgroupUsed(pc)); | |
1856 | VM_BUG_ON(pc->mem_cgroup != from); | |
f817ed48 | 1857 | |
8725d541 | 1858 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
1859 | /* Update mapped_file data for mem_cgroup */ |
1860 | preempt_disable(); | |
1861 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1862 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1863 | preempt_enable(); | |
d69b042f | 1864 | } |
854ffa8d DN |
1865 | mem_cgroup_charge_statistics(from, pc, false); |
1866 | if (uncharge) | |
1867 | /* This is not "cancel", but cancel_charge does all we need. */ | |
1868 | mem_cgroup_cancel_charge(from); | |
d69b042f | 1869 | |
854ffa8d | 1870 | /* caller should have done css_get */ |
08e552c6 KH |
1871 | pc->mem_cgroup = to; |
1872 | mem_cgroup_charge_statistics(to, pc, true); | |
88703267 KH |
1873 | /* |
1874 | * We charges against "to" which may not have any tasks. Then, "to" | |
1875 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe DN |
1876 | * this function is just force_empty() and move charge, so it's |
1877 | * garanteed that "to" is never removed. So, we don't check rmdir | |
1878 | * status here. | |
88703267 | 1879 | */ |
57f9fd7d DN |
1880 | } |
1881 | ||
1882 | /* | |
1883 | * check whether the @pc is valid for moving account and call | |
1884 | * __mem_cgroup_move_account() | |
1885 | */ | |
1886 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
854ffa8d | 1887 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
57f9fd7d DN |
1888 | { |
1889 | int ret = -EINVAL; | |
1890 | lock_page_cgroup(pc); | |
1891 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { | |
854ffa8d | 1892 | __mem_cgroup_move_account(pc, from, to, uncharge); |
57f9fd7d DN |
1893 | ret = 0; |
1894 | } | |
1895 | unlock_page_cgroup(pc); | |
d2265e6f KH |
1896 | /* |
1897 | * check events | |
1898 | */ | |
1899 | memcg_check_events(to, pc->page); | |
1900 | memcg_check_events(from, pc->page); | |
f817ed48 KH |
1901 | return ret; |
1902 | } | |
1903 | ||
1904 | /* | |
1905 | * move charges to its parent. | |
1906 | */ | |
1907 | ||
1908 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1909 | struct mem_cgroup *child, | |
1910 | gfp_t gfp_mask) | |
1911 | { | |
08e552c6 | 1912 | struct page *page = pc->page; |
f817ed48 KH |
1913 | struct cgroup *cg = child->css.cgroup; |
1914 | struct cgroup *pcg = cg->parent; | |
1915 | struct mem_cgroup *parent; | |
f817ed48 KH |
1916 | int ret; |
1917 | ||
1918 | /* Is ROOT ? */ | |
1919 | if (!pcg) | |
1920 | return -EINVAL; | |
1921 | ||
57f9fd7d DN |
1922 | ret = -EBUSY; |
1923 | if (!get_page_unless_zero(page)) | |
1924 | goto out; | |
1925 | if (isolate_lru_page(page)) | |
1926 | goto put; | |
08e552c6 | 1927 | |
f817ed48 | 1928 | parent = mem_cgroup_from_cont(pcg); |
430e4863 | 1929 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 1930 | if (ret || !parent) |
57f9fd7d | 1931 | goto put_back; |
f817ed48 | 1932 | |
854ffa8d DN |
1933 | ret = mem_cgroup_move_account(pc, child, parent, true); |
1934 | if (ret) | |
1935 | mem_cgroup_cancel_charge(parent); | |
57f9fd7d | 1936 | put_back: |
08e552c6 | 1937 | putback_lru_page(page); |
57f9fd7d | 1938 | put: |
40d58138 | 1939 | put_page(page); |
57f9fd7d | 1940 | out: |
f817ed48 KH |
1941 | return ret; |
1942 | } | |
1943 | ||
7a81b88c KH |
1944 | /* |
1945 | * Charge the memory controller for page usage. | |
1946 | * Return | |
1947 | * 0 if the charge was successful | |
1948 | * < 0 if the cgroup is over its limit | |
1949 | */ | |
1950 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1951 | gfp_t gfp_mask, enum charge_type ctype, | |
1952 | struct mem_cgroup *memcg) | |
1953 | { | |
1954 | struct mem_cgroup *mem; | |
1955 | struct page_cgroup *pc; | |
1956 | int ret; | |
1957 | ||
1958 | pc = lookup_page_cgroup(page); | |
1959 | /* can happen at boot */ | |
1960 | if (unlikely(!pc)) | |
1961 | return 0; | |
1962 | prefetchw(pc); | |
1963 | ||
1964 | mem = memcg; | |
430e4863 | 1965 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 1966 | if (ret || !mem) |
7a81b88c KH |
1967 | return ret; |
1968 | ||
1969 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1970 | return 0; |
8a9f3ccd BS |
1971 | } |
1972 | ||
7a81b88c KH |
1973 | int mem_cgroup_newpage_charge(struct page *page, |
1974 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1975 | { |
f8d66542 | 1976 | if (mem_cgroup_disabled()) |
cede86ac | 1977 | return 0; |
52d4b9ac KH |
1978 | if (PageCompound(page)) |
1979 | return 0; | |
69029cd5 KH |
1980 | /* |
1981 | * If already mapped, we don't have to account. | |
1982 | * If page cache, page->mapping has address_space. | |
1983 | * But page->mapping may have out-of-use anon_vma pointer, | |
1984 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1985 | * is NULL. | |
1986 | */ | |
1987 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1988 | return 0; | |
1989 | if (unlikely(!mm)) | |
1990 | mm = &init_mm; | |
217bc319 | 1991 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 1992 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
1993 | } |
1994 | ||
83aae4c7 DN |
1995 | static void |
1996 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1997 | enum charge_type ctype); | |
1998 | ||
e1a1cd59 BS |
1999 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2000 | gfp_t gfp_mask) | |
8697d331 | 2001 | { |
b5a84319 KH |
2002 | struct mem_cgroup *mem = NULL; |
2003 | int ret; | |
2004 | ||
f8d66542 | 2005 | if (mem_cgroup_disabled()) |
cede86ac | 2006 | return 0; |
52d4b9ac KH |
2007 | if (PageCompound(page)) |
2008 | return 0; | |
accf163e KH |
2009 | /* |
2010 | * Corner case handling. This is called from add_to_page_cache() | |
2011 | * in usual. But some FS (shmem) precharges this page before calling it | |
2012 | * and call add_to_page_cache() with GFP_NOWAIT. | |
2013 | * | |
2014 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
2015 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
2016 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
2017 | * And when the page is SwapCache, it should take swap information |
2018 | * into account. This is under lock_page() now. | |
accf163e KH |
2019 | */ |
2020 | if (!(gfp_mask & __GFP_WAIT)) { | |
2021 | struct page_cgroup *pc; | |
2022 | ||
52d4b9ac KH |
2023 | |
2024 | pc = lookup_page_cgroup(page); | |
2025 | if (!pc) | |
2026 | return 0; | |
2027 | lock_page_cgroup(pc); | |
2028 | if (PageCgroupUsed(pc)) { | |
2029 | unlock_page_cgroup(pc); | |
accf163e KH |
2030 | return 0; |
2031 | } | |
52d4b9ac | 2032 | unlock_page_cgroup(pc); |
accf163e KH |
2033 | } |
2034 | ||
b5a84319 | 2035 | if (unlikely(!mm && !mem)) |
8697d331 | 2036 | mm = &init_mm; |
accf163e | 2037 | |
c05555b5 KH |
2038 | if (page_is_file_cache(page)) |
2039 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 2040 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 | 2041 | |
83aae4c7 DN |
2042 | /* shmem */ |
2043 | if (PageSwapCache(page)) { | |
2044 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
2045 | if (!ret) | |
2046 | __mem_cgroup_commit_charge_swapin(page, mem, | |
2047 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
2048 | } else | |
2049 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
2050 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
b5a84319 | 2051 | |
b5a84319 | 2052 | return ret; |
e8589cc1 KH |
2053 | } |
2054 | ||
54595fe2 KH |
2055 | /* |
2056 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2057 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2058 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2059 | * "commit()" or removed by "cancel()" |
2060 | */ | |
8c7c6e34 KH |
2061 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2062 | struct page *page, | |
2063 | gfp_t mask, struct mem_cgroup **ptr) | |
2064 | { | |
2065 | struct mem_cgroup *mem; | |
54595fe2 | 2066 | int ret; |
8c7c6e34 | 2067 | |
f8d66542 | 2068 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2069 | return 0; |
2070 | ||
2071 | if (!do_swap_account) | |
2072 | goto charge_cur_mm; | |
8c7c6e34 KH |
2073 | /* |
2074 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2075 | * the pte, and even removed page from swap cache: in those cases |
2076 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2077 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2078 | */ |
2079 | if (!PageSwapCache(page)) | |
407f9c8b | 2080 | goto charge_cur_mm; |
e42d9d5d | 2081 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
2082 | if (!mem) |
2083 | goto charge_cur_mm; | |
8c7c6e34 | 2084 | *ptr = mem; |
430e4863 | 2085 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); |
54595fe2 KH |
2086 | /* drop extra refcnt from tryget */ |
2087 | css_put(&mem->css); | |
2088 | return ret; | |
8c7c6e34 KH |
2089 | charge_cur_mm: |
2090 | if (unlikely(!mm)) | |
2091 | mm = &init_mm; | |
430e4863 | 2092 | return __mem_cgroup_try_charge(mm, mask, ptr, true); |
8c7c6e34 KH |
2093 | } |
2094 | ||
83aae4c7 DN |
2095 | static void |
2096 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2097 | enum charge_type ctype) | |
7a81b88c KH |
2098 | { |
2099 | struct page_cgroup *pc; | |
2100 | ||
f8d66542 | 2101 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2102 | return; |
2103 | if (!ptr) | |
2104 | return; | |
88703267 | 2105 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 2106 | pc = lookup_page_cgroup(page); |
544122e5 | 2107 | mem_cgroup_lru_del_before_commit_swapcache(page); |
83aae4c7 | 2108 | __mem_cgroup_commit_charge(ptr, pc, ctype); |
544122e5 | 2109 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
2110 | /* |
2111 | * Now swap is on-memory. This means this page may be | |
2112 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2113 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2114 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2115 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2116 | */ |
03f3c433 | 2117 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2118 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2119 | unsigned short id; |
8c7c6e34 | 2120 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2121 | |
2122 | id = swap_cgroup_record(ent, 0); | |
2123 | rcu_read_lock(); | |
2124 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2125 | if (memcg) { |
a3b2d692 KH |
2126 | /* |
2127 | * This recorded memcg can be obsolete one. So, avoid | |
2128 | * calling css_tryget | |
2129 | */ | |
0c3e73e8 | 2130 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2131 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2132 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2133 | mem_cgroup_put(memcg); |
2134 | } | |
a3b2d692 | 2135 | rcu_read_unlock(); |
8c7c6e34 | 2136 | } |
88703267 KH |
2137 | /* |
2138 | * At swapin, we may charge account against cgroup which has no tasks. | |
2139 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2140 | * In that case, we need to call pre_destroy() again. check it here. | |
2141 | */ | |
2142 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2143 | } |
2144 | ||
83aae4c7 DN |
2145 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2146 | { | |
2147 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2148 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2149 | } | |
2150 | ||
7a81b88c KH |
2151 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2152 | { | |
f8d66542 | 2153 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2154 | return; |
2155 | if (!mem) | |
2156 | return; | |
a3032a2c | 2157 | mem_cgroup_cancel_charge(mem); |
7a81b88c KH |
2158 | } |
2159 | ||
569b846d KH |
2160 | static void |
2161 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype) | |
2162 | { | |
2163 | struct memcg_batch_info *batch = NULL; | |
2164 | bool uncharge_memsw = true; | |
2165 | /* If swapout, usage of swap doesn't decrease */ | |
2166 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2167 | uncharge_memsw = false; | |
569b846d KH |
2168 | |
2169 | batch = ¤t->memcg_batch; | |
2170 | /* | |
2171 | * In usual, we do css_get() when we remember memcg pointer. | |
2172 | * But in this case, we keep res->usage until end of a series of | |
2173 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2174 | */ | |
2175 | if (!batch->memcg) | |
2176 | batch->memcg = mem; | |
3c11ecf4 KH |
2177 | /* |
2178 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
2179 | * In those cases, all pages freed continously can be expected to be in | |
2180 | * the same cgroup and we have chance to coalesce uncharges. | |
2181 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2182 | * because we want to do uncharge as soon as possible. | |
2183 | */ | |
2184 | ||
2185 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2186 | goto direct_uncharge; | |
2187 | ||
569b846d KH |
2188 | /* |
2189 | * In typical case, batch->memcg == mem. This means we can | |
2190 | * merge a series of uncharges to an uncharge of res_counter. | |
2191 | * If not, we uncharge res_counter ony by one. | |
2192 | */ | |
2193 | if (batch->memcg != mem) | |
2194 | goto direct_uncharge; | |
2195 | /* remember freed charge and uncharge it later */ | |
2196 | batch->bytes += PAGE_SIZE; | |
2197 | if (uncharge_memsw) | |
2198 | batch->memsw_bytes += PAGE_SIZE; | |
2199 | return; | |
2200 | direct_uncharge: | |
2201 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
2202 | if (uncharge_memsw) | |
2203 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
3c11ecf4 KH |
2204 | if (unlikely(batch->memcg != mem)) |
2205 | memcg_oom_recover(mem); | |
569b846d KH |
2206 | return; |
2207 | } | |
7a81b88c | 2208 | |
8a9f3ccd | 2209 | /* |
69029cd5 | 2210 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2211 | */ |
8c7c6e34 | 2212 | static struct mem_cgroup * |
69029cd5 | 2213 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2214 | { |
8289546e | 2215 | struct page_cgroup *pc; |
8c7c6e34 | 2216 | struct mem_cgroup *mem = NULL; |
072c56c1 | 2217 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 2218 | |
f8d66542 | 2219 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2220 | return NULL; |
4077960e | 2221 | |
d13d1443 | 2222 | if (PageSwapCache(page)) |
8c7c6e34 | 2223 | return NULL; |
d13d1443 | 2224 | |
8697d331 | 2225 | /* |
3c541e14 | 2226 | * Check if our page_cgroup is valid |
8697d331 | 2227 | */ |
52d4b9ac KH |
2228 | pc = lookup_page_cgroup(page); |
2229 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2230 | return NULL; |
b9c565d5 | 2231 | |
52d4b9ac | 2232 | lock_page_cgroup(pc); |
d13d1443 | 2233 | |
8c7c6e34 KH |
2234 | mem = pc->mem_cgroup; |
2235 | ||
d13d1443 KH |
2236 | if (!PageCgroupUsed(pc)) |
2237 | goto unlock_out; | |
2238 | ||
2239 | switch (ctype) { | |
2240 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2241 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2242 | /* See mem_cgroup_prepare_migration() */ |
2243 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2244 | goto unlock_out; |
2245 | break; | |
2246 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2247 | if (!PageAnon(page)) { /* Shared memory */ | |
2248 | if (page->mapping && !page_is_file_cache(page)) | |
2249 | goto unlock_out; | |
2250 | } else if (page_mapped(page)) /* Anon */ | |
2251 | goto unlock_out; | |
2252 | break; | |
2253 | default: | |
2254 | break; | |
52d4b9ac | 2255 | } |
d13d1443 | 2256 | |
569b846d KH |
2257 | if (!mem_cgroup_is_root(mem)) |
2258 | __do_uncharge(mem, ctype); | |
0c3e73e8 BS |
2259 | if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
2260 | mem_cgroup_swap_statistics(mem, true); | |
08e552c6 | 2261 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 2262 | |
52d4b9ac | 2263 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2264 | /* |
2265 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2266 | * freed from LRU. This is safe because uncharged page is expected not | |
2267 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2268 | * special functions. | |
2269 | */ | |
b9c565d5 | 2270 | |
69029cd5 | 2271 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 2272 | unlock_page_cgroup(pc); |
fb59e9f1 | 2273 | |
d2265e6f | 2274 | memcg_check_events(mem, page); |
a7fe942e KH |
2275 | /* at swapout, this memcg will be accessed to record to swap */ |
2276 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2277 | css_put(&mem->css); | |
6d12e2d8 | 2278 | |
8c7c6e34 | 2279 | return mem; |
d13d1443 KH |
2280 | |
2281 | unlock_out: | |
2282 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2283 | return NULL; |
3c541e14 BS |
2284 | } |
2285 | ||
69029cd5 KH |
2286 | void mem_cgroup_uncharge_page(struct page *page) |
2287 | { | |
52d4b9ac KH |
2288 | /* early check. */ |
2289 | if (page_mapped(page)) | |
2290 | return; | |
2291 | if (page->mapping && !PageAnon(page)) | |
2292 | return; | |
69029cd5 KH |
2293 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2294 | } | |
2295 | ||
2296 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2297 | { | |
2298 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2299 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2300 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2301 | } | |
2302 | ||
569b846d KH |
2303 | /* |
2304 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2305 | * In that cases, pages are freed continuously and we can expect pages | |
2306 | * are in the same memcg. All these calls itself limits the number of | |
2307 | * pages freed at once, then uncharge_start/end() is called properly. | |
2308 | * This may be called prural(2) times in a context, | |
2309 | */ | |
2310 | ||
2311 | void mem_cgroup_uncharge_start(void) | |
2312 | { | |
2313 | current->memcg_batch.do_batch++; | |
2314 | /* We can do nest. */ | |
2315 | if (current->memcg_batch.do_batch == 1) { | |
2316 | current->memcg_batch.memcg = NULL; | |
2317 | current->memcg_batch.bytes = 0; | |
2318 | current->memcg_batch.memsw_bytes = 0; | |
2319 | } | |
2320 | } | |
2321 | ||
2322 | void mem_cgroup_uncharge_end(void) | |
2323 | { | |
2324 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2325 | ||
2326 | if (!batch->do_batch) | |
2327 | return; | |
2328 | ||
2329 | batch->do_batch--; | |
2330 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2331 | return; | |
2332 | ||
2333 | if (!batch->memcg) | |
2334 | return; | |
2335 | /* | |
2336 | * This "batch->memcg" is valid without any css_get/put etc... | |
2337 | * bacause we hide charges behind us. | |
2338 | */ | |
2339 | if (batch->bytes) | |
2340 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2341 | if (batch->memsw_bytes) | |
2342 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
3c11ecf4 | 2343 | memcg_oom_recover(batch->memcg); |
569b846d KH |
2344 | /* forget this pointer (for sanity check) */ |
2345 | batch->memcg = NULL; | |
2346 | } | |
2347 | ||
e767e056 | 2348 | #ifdef CONFIG_SWAP |
8c7c6e34 | 2349 | /* |
e767e056 | 2350 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
2351 | * memcg information is recorded to swap_cgroup of "ent" |
2352 | */ | |
8a9478ca KH |
2353 | void |
2354 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
2355 | { |
2356 | struct mem_cgroup *memcg; | |
8a9478ca KH |
2357 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
2358 | ||
2359 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2360 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2361 | ||
2362 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 2363 | |
8c7c6e34 | 2364 | /* record memcg information */ |
8a9478ca | 2365 | if (do_swap_account && swapout && memcg) { |
a3b2d692 | 2366 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 KH |
2367 | mem_cgroup_get(memcg); |
2368 | } | |
8a9478ca | 2369 | if (swapout && memcg) |
a7fe942e | 2370 | css_put(&memcg->css); |
8c7c6e34 | 2371 | } |
e767e056 | 2372 | #endif |
8c7c6e34 KH |
2373 | |
2374 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2375 | /* | |
2376 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2377 | * uncharge "memsw" account. | |
2378 | */ | |
2379 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 2380 | { |
8c7c6e34 | 2381 | struct mem_cgroup *memcg; |
a3b2d692 | 2382 | unsigned short id; |
8c7c6e34 KH |
2383 | |
2384 | if (!do_swap_account) | |
2385 | return; | |
2386 | ||
a3b2d692 KH |
2387 | id = swap_cgroup_record(ent, 0); |
2388 | rcu_read_lock(); | |
2389 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2390 | if (memcg) { |
a3b2d692 KH |
2391 | /* |
2392 | * We uncharge this because swap is freed. | |
2393 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2394 | */ | |
0c3e73e8 | 2395 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2396 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2397 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2398 | mem_cgroup_put(memcg); |
2399 | } | |
a3b2d692 | 2400 | rcu_read_unlock(); |
d13d1443 | 2401 | } |
02491447 DN |
2402 | |
2403 | /** | |
2404 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2405 | * @entry: swap entry to be moved | |
2406 | * @from: mem_cgroup which the entry is moved from | |
2407 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 2408 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
2409 | * |
2410 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2411 | * as the mem_cgroup's id of @from. | |
2412 | * | |
2413 | * Returns 0 on success, -EINVAL on failure. | |
2414 | * | |
2415 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
2416 | * both res and memsw, and called css_get(). | |
2417 | */ | |
2418 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2419 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2420 | { |
2421 | unsigned short old_id, new_id; | |
2422 | ||
2423 | old_id = css_id(&from->css); | |
2424 | new_id = css_id(&to->css); | |
2425 | ||
2426 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2427 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2428 | mem_cgroup_swap_statistics(to, true); |
02491447 | 2429 | /* |
483c30b5 DN |
2430 | * This function is only called from task migration context now. |
2431 | * It postpones res_counter and refcount handling till the end | |
2432 | * of task migration(mem_cgroup_clear_mc()) for performance | |
2433 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
2434 | * because if the process that has been moved to @to does | |
2435 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 2436 | */ |
02491447 | 2437 | mem_cgroup_get(to); |
483c30b5 DN |
2438 | if (need_fixup) { |
2439 | if (!mem_cgroup_is_root(from)) | |
2440 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
2441 | mem_cgroup_put(from); | |
2442 | /* | |
2443 | * we charged both to->res and to->memsw, so we should | |
2444 | * uncharge to->res. | |
2445 | */ | |
2446 | if (!mem_cgroup_is_root(to)) | |
2447 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
2448 | css_put(&to->css); | |
2449 | } | |
02491447 DN |
2450 | return 0; |
2451 | } | |
2452 | return -EINVAL; | |
2453 | } | |
2454 | #else | |
2455 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2456 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2457 | { |
2458 | return -EINVAL; | |
2459 | } | |
8c7c6e34 | 2460 | #endif |
d13d1443 | 2461 | |
ae41be37 | 2462 | /* |
01b1ae63 KH |
2463 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
2464 | * page belongs to. | |
ae41be37 | 2465 | */ |
ac39cf8c | 2466 | int mem_cgroup_prepare_migration(struct page *page, |
2467 | struct page *newpage, struct mem_cgroup **ptr) | |
ae41be37 KH |
2468 | { |
2469 | struct page_cgroup *pc; | |
e8589cc1 | 2470 | struct mem_cgroup *mem = NULL; |
ac39cf8c | 2471 | enum charge_type ctype; |
e8589cc1 | 2472 | int ret = 0; |
8869b8f6 | 2473 | |
f8d66542 | 2474 | if (mem_cgroup_disabled()) |
4077960e BS |
2475 | return 0; |
2476 | ||
52d4b9ac KH |
2477 | pc = lookup_page_cgroup(page); |
2478 | lock_page_cgroup(pc); | |
2479 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
2480 | mem = pc->mem_cgroup; |
2481 | css_get(&mem->css); | |
ac39cf8c | 2482 | /* |
2483 | * At migrating an anonymous page, its mapcount goes down | |
2484 | * to 0 and uncharge() will be called. But, even if it's fully | |
2485 | * unmapped, migration may fail and this page has to be | |
2486 | * charged again. We set MIGRATION flag here and delay uncharge | |
2487 | * until end_migration() is called | |
2488 | * | |
2489 | * Corner Case Thinking | |
2490 | * A) | |
2491 | * When the old page was mapped as Anon and it's unmap-and-freed | |
2492 | * while migration was ongoing. | |
2493 | * If unmap finds the old page, uncharge() of it will be delayed | |
2494 | * until end_migration(). If unmap finds a new page, it's | |
2495 | * uncharged when it make mapcount to be 1->0. If unmap code | |
2496 | * finds swap_migration_entry, the new page will not be mapped | |
2497 | * and end_migration() will find it(mapcount==0). | |
2498 | * | |
2499 | * B) | |
2500 | * When the old page was mapped but migraion fails, the kernel | |
2501 | * remaps it. A charge for it is kept by MIGRATION flag even | |
2502 | * if mapcount goes down to 0. We can do remap successfully | |
2503 | * without charging it again. | |
2504 | * | |
2505 | * C) | |
2506 | * The "old" page is under lock_page() until the end of | |
2507 | * migration, so, the old page itself will not be swapped-out. | |
2508 | * If the new page is swapped out before end_migraton, our | |
2509 | * hook to usual swap-out path will catch the event. | |
2510 | */ | |
2511 | if (PageAnon(page)) | |
2512 | SetPageCgroupMigration(pc); | |
e8589cc1 | 2513 | } |
52d4b9ac | 2514 | unlock_page_cgroup(pc); |
ac39cf8c | 2515 | /* |
2516 | * If the page is not charged at this point, | |
2517 | * we return here. | |
2518 | */ | |
2519 | if (!mem) | |
2520 | return 0; | |
01b1ae63 | 2521 | |
93d5c9be | 2522 | *ptr = mem; |
ac39cf8c | 2523 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false); |
2524 | css_put(&mem->css);/* drop extra refcnt */ | |
2525 | if (ret || *ptr == NULL) { | |
2526 | if (PageAnon(page)) { | |
2527 | lock_page_cgroup(pc); | |
2528 | ClearPageCgroupMigration(pc); | |
2529 | unlock_page_cgroup(pc); | |
2530 | /* | |
2531 | * The old page may be fully unmapped while we kept it. | |
2532 | */ | |
2533 | mem_cgroup_uncharge_page(page); | |
2534 | } | |
2535 | return -ENOMEM; | |
e8589cc1 | 2536 | } |
ac39cf8c | 2537 | /* |
2538 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
2539 | * is called before end_migration, we can catch all events on this new | |
2540 | * page. In the case new page is migrated but not remapped, new page's | |
2541 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
2542 | */ | |
2543 | pc = lookup_page_cgroup(newpage); | |
2544 | if (PageAnon(page)) | |
2545 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2546 | else if (page_is_file_cache(page)) | |
2547 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2548 | else | |
2549 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
2550 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
e8589cc1 | 2551 | return ret; |
ae41be37 | 2552 | } |
8869b8f6 | 2553 | |
69029cd5 | 2554 | /* remove redundant charge if migration failed*/ |
01b1ae63 | 2555 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
ac39cf8c | 2556 | struct page *oldpage, struct page *newpage) |
ae41be37 | 2557 | { |
ac39cf8c | 2558 | struct page *used, *unused; |
01b1ae63 | 2559 | struct page_cgroup *pc; |
01b1ae63 KH |
2560 | |
2561 | if (!mem) | |
2562 | return; | |
ac39cf8c | 2563 | /* blocks rmdir() */ |
88703267 | 2564 | cgroup_exclude_rmdir(&mem->css); |
01b1ae63 KH |
2565 | /* at migration success, oldpage->mapping is NULL. */ |
2566 | if (oldpage->mapping) { | |
ac39cf8c | 2567 | used = oldpage; |
2568 | unused = newpage; | |
01b1ae63 | 2569 | } else { |
ac39cf8c | 2570 | used = newpage; |
01b1ae63 KH |
2571 | unused = oldpage; |
2572 | } | |
69029cd5 | 2573 | /* |
ac39cf8c | 2574 | * We disallowed uncharge of pages under migration because mapcount |
2575 | * of the page goes down to zero, temporarly. | |
2576 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 2577 | */ |
ac39cf8c | 2578 | pc = lookup_page_cgroup(oldpage); |
2579 | lock_page_cgroup(pc); | |
2580 | ClearPageCgroupMigration(pc); | |
2581 | unlock_page_cgroup(pc); | |
01b1ae63 | 2582 | |
ac39cf8c | 2583 | if (unused != oldpage) |
2584 | pc = lookup_page_cgroup(unused); | |
2585 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); | |
2586 | ||
2587 | pc = lookup_page_cgroup(used); | |
01b1ae63 | 2588 | /* |
ac39cf8c | 2589 | * If a page is a file cache, radix-tree replacement is very atomic |
2590 | * and we can skip this check. When it was an Anon page, its mapcount | |
2591 | * goes down to 0. But because we added MIGRATION flage, it's not | |
2592 | * uncharged yet. There are several case but page->mapcount check | |
2593 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
2594 | * check. (see prepare_charge() also) | |
69029cd5 | 2595 | */ |
ac39cf8c | 2596 | if (PageAnon(used)) |
2597 | mem_cgroup_uncharge_page(used); | |
88703267 | 2598 | /* |
ac39cf8c | 2599 | * At migration, we may charge account against cgroup which has no |
2600 | * tasks. | |
88703267 KH |
2601 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
2602 | * In that case, we need to call pre_destroy() again. check it here. | |
2603 | */ | |
2604 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 2605 | } |
78fb7466 | 2606 | |
c9b0ed51 | 2607 | /* |
ae3abae6 DN |
2608 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
2609 | * Calling hierarchical_reclaim is not enough because we should update | |
2610 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
2611 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
2612 | * not from the memcg which this page would be charged to. | |
2613 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 2614 | */ |
ae3abae6 | 2615 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
2616 | struct mm_struct *mm, |
2617 | gfp_t gfp_mask) | |
c9b0ed51 | 2618 | { |
b5a84319 | 2619 | struct mem_cgroup *mem = NULL; |
ae3abae6 | 2620 | int ret; |
c9b0ed51 | 2621 | |
f8d66542 | 2622 | if (mem_cgroup_disabled()) |
cede86ac | 2623 | return 0; |
c9b0ed51 | 2624 | |
ae3abae6 DN |
2625 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2626 | if (!ret) | |
2627 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 2628 | |
ae3abae6 | 2629 | return ret; |
c9b0ed51 KH |
2630 | } |
2631 | ||
8c7c6e34 KH |
2632 | static DEFINE_MUTEX(set_limit_mutex); |
2633 | ||
d38d2a75 | 2634 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 2635 | unsigned long long val) |
628f4235 | 2636 | { |
81d39c20 | 2637 | int retry_count; |
3c11ecf4 | 2638 | u64 memswlimit, memlimit; |
628f4235 | 2639 | int ret = 0; |
81d39c20 KH |
2640 | int children = mem_cgroup_count_children(memcg); |
2641 | u64 curusage, oldusage; | |
3c11ecf4 | 2642 | int enlarge; |
81d39c20 KH |
2643 | |
2644 | /* | |
2645 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2646 | * is depends on callers. We set our retry-count to be function | |
2647 | * of # of children which we should visit in this loop. | |
2648 | */ | |
2649 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
2650 | ||
2651 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 2652 | |
3c11ecf4 | 2653 | enlarge = 0; |
8c7c6e34 | 2654 | while (retry_count) { |
628f4235 KH |
2655 | if (signal_pending(current)) { |
2656 | ret = -EINTR; | |
2657 | break; | |
2658 | } | |
8c7c6e34 KH |
2659 | /* |
2660 | * Rather than hide all in some function, I do this in | |
2661 | * open coded manner. You see what this really does. | |
2662 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2663 | */ | |
2664 | mutex_lock(&set_limit_mutex); | |
2665 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2666 | if (memswlimit < val) { | |
2667 | ret = -EINVAL; | |
2668 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
2669 | break; |
2670 | } | |
3c11ecf4 KH |
2671 | |
2672 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2673 | if (memlimit < val) | |
2674 | enlarge = 1; | |
2675 | ||
8c7c6e34 | 2676 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
2677 | if (!ret) { |
2678 | if (memswlimit == val) | |
2679 | memcg->memsw_is_minimum = true; | |
2680 | else | |
2681 | memcg->memsw_is_minimum = false; | |
2682 | } | |
8c7c6e34 KH |
2683 | mutex_unlock(&set_limit_mutex); |
2684 | ||
2685 | if (!ret) | |
2686 | break; | |
2687 | ||
aa20d489 | 2688 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
4e416953 | 2689 | MEM_CGROUP_RECLAIM_SHRINK); |
81d39c20 KH |
2690 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
2691 | /* Usage is reduced ? */ | |
2692 | if (curusage >= oldusage) | |
2693 | retry_count--; | |
2694 | else | |
2695 | oldusage = curusage; | |
8c7c6e34 | 2696 | } |
3c11ecf4 KH |
2697 | if (!ret && enlarge) |
2698 | memcg_oom_recover(memcg); | |
14797e23 | 2699 | |
8c7c6e34 KH |
2700 | return ret; |
2701 | } | |
2702 | ||
338c8431 LZ |
2703 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
2704 | unsigned long long val) | |
8c7c6e34 | 2705 | { |
81d39c20 | 2706 | int retry_count; |
3c11ecf4 | 2707 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
2708 | int children = mem_cgroup_count_children(memcg); |
2709 | int ret = -EBUSY; | |
3c11ecf4 | 2710 | int enlarge = 0; |
8c7c6e34 | 2711 | |
81d39c20 KH |
2712 | /* see mem_cgroup_resize_res_limit */ |
2713 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
2714 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
2715 | while (retry_count) { |
2716 | if (signal_pending(current)) { | |
2717 | ret = -EINTR; | |
2718 | break; | |
2719 | } | |
2720 | /* | |
2721 | * Rather than hide all in some function, I do this in | |
2722 | * open coded manner. You see what this really does. | |
2723 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2724 | */ | |
2725 | mutex_lock(&set_limit_mutex); | |
2726 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2727 | if (memlimit > val) { | |
2728 | ret = -EINVAL; | |
2729 | mutex_unlock(&set_limit_mutex); | |
2730 | break; | |
2731 | } | |
3c11ecf4 KH |
2732 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
2733 | if (memswlimit < val) | |
2734 | enlarge = 1; | |
8c7c6e34 | 2735 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
2736 | if (!ret) { |
2737 | if (memlimit == val) | |
2738 | memcg->memsw_is_minimum = true; | |
2739 | else | |
2740 | memcg->memsw_is_minimum = false; | |
2741 | } | |
8c7c6e34 KH |
2742 | mutex_unlock(&set_limit_mutex); |
2743 | ||
2744 | if (!ret) | |
2745 | break; | |
2746 | ||
4e416953 | 2747 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 BS |
2748 | MEM_CGROUP_RECLAIM_NOSWAP | |
2749 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 2750 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 2751 | /* Usage is reduced ? */ |
8c7c6e34 | 2752 | if (curusage >= oldusage) |
628f4235 | 2753 | retry_count--; |
81d39c20 KH |
2754 | else |
2755 | oldusage = curusage; | |
628f4235 | 2756 | } |
3c11ecf4 KH |
2757 | if (!ret && enlarge) |
2758 | memcg_oom_recover(memcg); | |
628f4235 KH |
2759 | return ret; |
2760 | } | |
2761 | ||
4e416953 BS |
2762 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2763 | gfp_t gfp_mask, int nid, | |
2764 | int zid) | |
2765 | { | |
2766 | unsigned long nr_reclaimed = 0; | |
2767 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2768 | unsigned long reclaimed; | |
2769 | int loop = 0; | |
2770 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 2771 | unsigned long long excess; |
4e416953 BS |
2772 | |
2773 | if (order > 0) | |
2774 | return 0; | |
2775 | ||
2776 | mctz = soft_limit_tree_node_zone(nid, zid); | |
2777 | /* | |
2778 | * This loop can run a while, specially if mem_cgroup's continuously | |
2779 | * keep exceeding their soft limit and putting the system under | |
2780 | * pressure | |
2781 | */ | |
2782 | do { | |
2783 | if (next_mz) | |
2784 | mz = next_mz; | |
2785 | else | |
2786 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2787 | if (!mz) | |
2788 | break; | |
2789 | ||
2790 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
2791 | gfp_mask, | |
2792 | MEM_CGROUP_RECLAIM_SOFT); | |
2793 | nr_reclaimed += reclaimed; | |
2794 | spin_lock(&mctz->lock); | |
2795 | ||
2796 | /* | |
2797 | * If we failed to reclaim anything from this memory cgroup | |
2798 | * it is time to move on to the next cgroup | |
2799 | */ | |
2800 | next_mz = NULL; | |
2801 | if (!reclaimed) { | |
2802 | do { | |
2803 | /* | |
2804 | * Loop until we find yet another one. | |
2805 | * | |
2806 | * By the time we get the soft_limit lock | |
2807 | * again, someone might have aded the | |
2808 | * group back on the RB tree. Iterate to | |
2809 | * make sure we get a different mem. | |
2810 | * mem_cgroup_largest_soft_limit_node returns | |
2811 | * NULL if no other cgroup is present on | |
2812 | * the tree | |
2813 | */ | |
2814 | next_mz = | |
2815 | __mem_cgroup_largest_soft_limit_node(mctz); | |
2816 | if (next_mz == mz) { | |
2817 | css_put(&next_mz->mem->css); | |
2818 | next_mz = NULL; | |
2819 | } else /* next_mz == NULL or other memcg */ | |
2820 | break; | |
2821 | } while (1); | |
2822 | } | |
4e416953 | 2823 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 2824 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
2825 | /* |
2826 | * One school of thought says that we should not add | |
2827 | * back the node to the tree if reclaim returns 0. | |
2828 | * But our reclaim could return 0, simply because due | |
2829 | * to priority we are exposing a smaller subset of | |
2830 | * memory to reclaim from. Consider this as a longer | |
2831 | * term TODO. | |
2832 | */ | |
ef8745c1 KH |
2833 | /* If excess == 0, no tree ops */ |
2834 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
2835 | spin_unlock(&mctz->lock); |
2836 | css_put(&mz->mem->css); | |
2837 | loop++; | |
2838 | /* | |
2839 | * Could not reclaim anything and there are no more | |
2840 | * mem cgroups to try or we seem to be looping without | |
2841 | * reclaiming anything. | |
2842 | */ | |
2843 | if (!nr_reclaimed && | |
2844 | (next_mz == NULL || | |
2845 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2846 | break; | |
2847 | } while (!nr_reclaimed); | |
2848 | if (next_mz) | |
2849 | css_put(&next_mz->mem->css); | |
2850 | return nr_reclaimed; | |
2851 | } | |
2852 | ||
cc847582 KH |
2853 | /* |
2854 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
2855 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
2856 | */ | |
f817ed48 | 2857 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 2858 | int node, int zid, enum lru_list lru) |
cc847582 | 2859 | { |
08e552c6 KH |
2860 | struct zone *zone; |
2861 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 2862 | struct page_cgroup *pc, *busy; |
08e552c6 | 2863 | unsigned long flags, loop; |
072c56c1 | 2864 | struct list_head *list; |
f817ed48 | 2865 | int ret = 0; |
072c56c1 | 2866 | |
08e552c6 KH |
2867 | zone = &NODE_DATA(node)->node_zones[zid]; |
2868 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 2869 | list = &mz->lists[lru]; |
cc847582 | 2870 | |
f817ed48 KH |
2871 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
2872 | /* give some margin against EBUSY etc...*/ | |
2873 | loop += 256; | |
2874 | busy = NULL; | |
2875 | while (loop--) { | |
2876 | ret = 0; | |
08e552c6 | 2877 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 2878 | if (list_empty(list)) { |
08e552c6 | 2879 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 2880 | break; |
f817ed48 KH |
2881 | } |
2882 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
2883 | if (busy == pc) { | |
2884 | list_move(&pc->lru, list); | |
648bcc77 | 2885 | busy = NULL; |
08e552c6 | 2886 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
2887 | continue; |
2888 | } | |
08e552c6 | 2889 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 2890 | |
2c26fdd7 | 2891 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 2892 | if (ret == -ENOMEM) |
52d4b9ac | 2893 | break; |
f817ed48 KH |
2894 | |
2895 | if (ret == -EBUSY || ret == -EINVAL) { | |
2896 | /* found lock contention or "pc" is obsolete. */ | |
2897 | busy = pc; | |
2898 | cond_resched(); | |
2899 | } else | |
2900 | busy = NULL; | |
cc847582 | 2901 | } |
08e552c6 | 2902 | |
f817ed48 KH |
2903 | if (!ret && !list_empty(list)) |
2904 | return -EBUSY; | |
2905 | return ret; | |
cc847582 KH |
2906 | } |
2907 | ||
2908 | /* | |
2909 | * make mem_cgroup's charge to be 0 if there is no task. | |
2910 | * This enables deleting this mem_cgroup. | |
2911 | */ | |
c1e862c1 | 2912 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 2913 | { |
f817ed48 KH |
2914 | int ret; |
2915 | int node, zid, shrink; | |
2916 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 2917 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 2918 | |
cc847582 | 2919 | css_get(&mem->css); |
f817ed48 KH |
2920 | |
2921 | shrink = 0; | |
c1e862c1 KH |
2922 | /* should free all ? */ |
2923 | if (free_all) | |
2924 | goto try_to_free; | |
f817ed48 | 2925 | move_account: |
fce66477 | 2926 | do { |
f817ed48 | 2927 | ret = -EBUSY; |
c1e862c1 KH |
2928 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
2929 | goto out; | |
2930 | ret = -EINTR; | |
2931 | if (signal_pending(current)) | |
cc847582 | 2932 | goto out; |
52d4b9ac KH |
2933 | /* This is for making all *used* pages to be on LRU. */ |
2934 | lru_add_drain_all(); | |
cdec2e42 | 2935 | drain_all_stock_sync(); |
f817ed48 | 2936 | ret = 0; |
299b4eaa | 2937 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 2938 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 2939 | enum lru_list l; |
f817ed48 KH |
2940 | for_each_lru(l) { |
2941 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 2942 | node, zid, l); |
f817ed48 KH |
2943 | if (ret) |
2944 | break; | |
2945 | } | |
1ecaab2b | 2946 | } |
f817ed48 KH |
2947 | if (ret) |
2948 | break; | |
2949 | } | |
3c11ecf4 | 2950 | memcg_oom_recover(mem); |
f817ed48 KH |
2951 | /* it seems parent cgroup doesn't have enough mem */ |
2952 | if (ret == -ENOMEM) | |
2953 | goto try_to_free; | |
52d4b9ac | 2954 | cond_resched(); |
fce66477 DN |
2955 | /* "ret" should also be checked to ensure all lists are empty. */ |
2956 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
2957 | out: |
2958 | css_put(&mem->css); | |
2959 | return ret; | |
f817ed48 KH |
2960 | |
2961 | try_to_free: | |
c1e862c1 KH |
2962 | /* returns EBUSY if there is a task or if we come here twice. */ |
2963 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
2964 | ret = -EBUSY; |
2965 | goto out; | |
2966 | } | |
c1e862c1 KH |
2967 | /* we call try-to-free pages for make this cgroup empty */ |
2968 | lru_add_drain_all(); | |
f817ed48 KH |
2969 | /* try to free all pages in this cgroup */ |
2970 | shrink = 1; | |
2971 | while (nr_retries && mem->res.usage > 0) { | |
2972 | int progress; | |
c1e862c1 KH |
2973 | |
2974 | if (signal_pending(current)) { | |
2975 | ret = -EINTR; | |
2976 | goto out; | |
2977 | } | |
a7885eb8 KM |
2978 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
2979 | false, get_swappiness(mem)); | |
c1e862c1 | 2980 | if (!progress) { |
f817ed48 | 2981 | nr_retries--; |
c1e862c1 | 2982 | /* maybe some writeback is necessary */ |
8aa7e847 | 2983 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2984 | } |
f817ed48 KH |
2985 | |
2986 | } | |
08e552c6 | 2987 | lru_add_drain(); |
f817ed48 | 2988 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 2989 | goto move_account; |
cc847582 KH |
2990 | } |
2991 | ||
c1e862c1 KH |
2992 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
2993 | { | |
2994 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
2995 | } | |
2996 | ||
2997 | ||
18f59ea7 BS |
2998 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
2999 | { | |
3000 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3001 | } | |
3002 | ||
3003 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3004 | u64 val) | |
3005 | { | |
3006 | int retval = 0; | |
3007 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3008 | struct cgroup *parent = cont->parent; | |
3009 | struct mem_cgroup *parent_mem = NULL; | |
3010 | ||
3011 | if (parent) | |
3012 | parent_mem = mem_cgroup_from_cont(parent); | |
3013 | ||
3014 | cgroup_lock(); | |
3015 | /* | |
af901ca1 | 3016 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3017 | * in the child subtrees. If it is unset, then the change can |
3018 | * occur, provided the current cgroup has no children. | |
3019 | * | |
3020 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3021 | * set if there are no children. | |
3022 | */ | |
3023 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
3024 | (val == 1 || val == 0)) { | |
3025 | if (list_empty(&cont->children)) | |
3026 | mem->use_hierarchy = val; | |
3027 | else | |
3028 | retval = -EBUSY; | |
3029 | } else | |
3030 | retval = -EINVAL; | |
3031 | cgroup_unlock(); | |
3032 | ||
3033 | return retval; | |
3034 | } | |
3035 | ||
0c3e73e8 BS |
3036 | struct mem_cgroup_idx_data { |
3037 | s64 val; | |
3038 | enum mem_cgroup_stat_index idx; | |
3039 | }; | |
3040 | ||
3041 | static int | |
3042 | mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) | |
3043 | { | |
3044 | struct mem_cgroup_idx_data *d = data; | |
c62b1a3b | 3045 | d->val += mem_cgroup_read_stat(mem, d->idx); |
0c3e73e8 BS |
3046 | return 0; |
3047 | } | |
3048 | ||
3049 | static void | |
3050 | mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, | |
3051 | enum mem_cgroup_stat_index idx, s64 *val) | |
3052 | { | |
3053 | struct mem_cgroup_idx_data d; | |
3054 | d.idx = idx; | |
3055 | d.val = 0; | |
3056 | mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat); | |
3057 | *val = d.val; | |
3058 | } | |
3059 | ||
104f3928 KS |
3060 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
3061 | { | |
3062 | u64 idx_val, val; | |
3063 | ||
3064 | if (!mem_cgroup_is_root(mem)) { | |
3065 | if (!swap) | |
3066 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
3067 | else | |
3068 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
3069 | } | |
3070 | ||
3071 | mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val); | |
3072 | val = idx_val; | |
3073 | mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val); | |
3074 | val += idx_val; | |
3075 | ||
3076 | if (swap) { | |
3077 | mem_cgroup_get_recursive_idx_stat(mem, | |
3078 | MEM_CGROUP_STAT_SWAPOUT, &idx_val); | |
3079 | val += idx_val; | |
3080 | } | |
3081 | ||
3082 | return val << PAGE_SHIFT; | |
3083 | } | |
3084 | ||
2c3daa72 | 3085 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3086 | { |
8c7c6e34 | 3087 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 3088 | u64 val; |
8c7c6e34 KH |
3089 | int type, name; |
3090 | ||
3091 | type = MEMFILE_TYPE(cft->private); | |
3092 | name = MEMFILE_ATTR(cft->private); | |
3093 | switch (type) { | |
3094 | case _MEM: | |
104f3928 KS |
3095 | if (name == RES_USAGE) |
3096 | val = mem_cgroup_usage(mem, false); | |
3097 | else | |
0c3e73e8 | 3098 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
3099 | break; |
3100 | case _MEMSWAP: | |
104f3928 KS |
3101 | if (name == RES_USAGE) |
3102 | val = mem_cgroup_usage(mem, true); | |
3103 | else | |
0c3e73e8 | 3104 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
3105 | break; |
3106 | default: | |
3107 | BUG(); | |
3108 | break; | |
3109 | } | |
3110 | return val; | |
8cdea7c0 | 3111 | } |
628f4235 KH |
3112 | /* |
3113 | * The user of this function is... | |
3114 | * RES_LIMIT. | |
3115 | */ | |
856c13aa PM |
3116 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3117 | const char *buffer) | |
8cdea7c0 | 3118 | { |
628f4235 | 3119 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3120 | int type, name; |
628f4235 KH |
3121 | unsigned long long val; |
3122 | int ret; | |
3123 | ||
8c7c6e34 KH |
3124 | type = MEMFILE_TYPE(cft->private); |
3125 | name = MEMFILE_ATTR(cft->private); | |
3126 | switch (name) { | |
628f4235 | 3127 | case RES_LIMIT: |
4b3bde4c BS |
3128 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3129 | ret = -EINVAL; | |
3130 | break; | |
3131 | } | |
628f4235 KH |
3132 | /* This function does all necessary parse...reuse it */ |
3133 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3134 | if (ret) |
3135 | break; | |
3136 | if (type == _MEM) | |
628f4235 | 3137 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3138 | else |
3139 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3140 | break; |
296c81d8 BS |
3141 | case RES_SOFT_LIMIT: |
3142 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3143 | if (ret) | |
3144 | break; | |
3145 | /* | |
3146 | * For memsw, soft limits are hard to implement in terms | |
3147 | * of semantics, for now, we support soft limits for | |
3148 | * control without swap | |
3149 | */ | |
3150 | if (type == _MEM) | |
3151 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3152 | else | |
3153 | ret = -EINVAL; | |
3154 | break; | |
628f4235 KH |
3155 | default: |
3156 | ret = -EINVAL; /* should be BUG() ? */ | |
3157 | break; | |
3158 | } | |
3159 | return ret; | |
8cdea7c0 BS |
3160 | } |
3161 | ||
fee7b548 KH |
3162 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3163 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3164 | { | |
3165 | struct cgroup *cgroup; | |
3166 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3167 | ||
3168 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3169 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3170 | cgroup = memcg->css.cgroup; | |
3171 | if (!memcg->use_hierarchy) | |
3172 | goto out; | |
3173 | ||
3174 | while (cgroup->parent) { | |
3175 | cgroup = cgroup->parent; | |
3176 | memcg = mem_cgroup_from_cont(cgroup); | |
3177 | if (!memcg->use_hierarchy) | |
3178 | break; | |
3179 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3180 | min_limit = min(min_limit, tmp); | |
3181 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3182 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3183 | } | |
3184 | out: | |
3185 | *mem_limit = min_limit; | |
3186 | *memsw_limit = min_memsw_limit; | |
3187 | return; | |
3188 | } | |
3189 | ||
29f2a4da | 3190 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
3191 | { |
3192 | struct mem_cgroup *mem; | |
8c7c6e34 | 3193 | int type, name; |
c84872e1 PE |
3194 | |
3195 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
3196 | type = MEMFILE_TYPE(event); |
3197 | name = MEMFILE_ATTR(event); | |
3198 | switch (name) { | |
29f2a4da | 3199 | case RES_MAX_USAGE: |
8c7c6e34 KH |
3200 | if (type == _MEM) |
3201 | res_counter_reset_max(&mem->res); | |
3202 | else | |
3203 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
3204 | break; |
3205 | case RES_FAILCNT: | |
8c7c6e34 KH |
3206 | if (type == _MEM) |
3207 | res_counter_reset_failcnt(&mem->res); | |
3208 | else | |
3209 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
3210 | break; |
3211 | } | |
f64c3f54 | 3212 | |
85cc59db | 3213 | return 0; |
c84872e1 PE |
3214 | } |
3215 | ||
7dc74be0 DN |
3216 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3217 | struct cftype *cft) | |
3218 | { | |
3219 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3220 | } | |
3221 | ||
02491447 | 3222 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3223 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3224 | struct cftype *cft, u64 val) | |
3225 | { | |
3226 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3227 | ||
3228 | if (val >= (1 << NR_MOVE_TYPE)) | |
3229 | return -EINVAL; | |
3230 | /* | |
3231 | * We check this value several times in both in can_attach() and | |
3232 | * attach(), so we need cgroup lock to prevent this value from being | |
3233 | * inconsistent. | |
3234 | */ | |
3235 | cgroup_lock(); | |
3236 | mem->move_charge_at_immigrate = val; | |
3237 | cgroup_unlock(); | |
3238 | ||
3239 | return 0; | |
3240 | } | |
02491447 DN |
3241 | #else |
3242 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3243 | struct cftype *cft, u64 val) | |
3244 | { | |
3245 | return -ENOSYS; | |
3246 | } | |
3247 | #endif | |
7dc74be0 | 3248 | |
14067bb3 KH |
3249 | |
3250 | /* For read statistics */ | |
3251 | enum { | |
3252 | MCS_CACHE, | |
3253 | MCS_RSS, | |
d8046582 | 3254 | MCS_FILE_MAPPED, |
14067bb3 KH |
3255 | MCS_PGPGIN, |
3256 | MCS_PGPGOUT, | |
1dd3a273 | 3257 | MCS_SWAP, |
14067bb3 KH |
3258 | MCS_INACTIVE_ANON, |
3259 | MCS_ACTIVE_ANON, | |
3260 | MCS_INACTIVE_FILE, | |
3261 | MCS_ACTIVE_FILE, | |
3262 | MCS_UNEVICTABLE, | |
3263 | NR_MCS_STAT, | |
3264 | }; | |
3265 | ||
3266 | struct mcs_total_stat { | |
3267 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
3268 | }; |
3269 | ||
14067bb3 KH |
3270 | struct { |
3271 | char *local_name; | |
3272 | char *total_name; | |
3273 | } memcg_stat_strings[NR_MCS_STAT] = { | |
3274 | {"cache", "total_cache"}, | |
3275 | {"rss", "total_rss"}, | |
d69b042f | 3276 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
3277 | {"pgpgin", "total_pgpgin"}, |
3278 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3279 | {"swap", "total_swap"}, |
14067bb3 KH |
3280 | {"inactive_anon", "total_inactive_anon"}, |
3281 | {"active_anon", "total_active_anon"}, | |
3282 | {"inactive_file", "total_inactive_file"}, | |
3283 | {"active_file", "total_active_file"}, | |
3284 | {"unevictable", "total_unevictable"} | |
3285 | }; | |
3286 | ||
3287 | ||
3288 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
3289 | { | |
3290 | struct mcs_total_stat *s = data; | |
3291 | s64 val; | |
3292 | ||
3293 | /* per cpu stat */ | |
c62b1a3b | 3294 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 3295 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 3296 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 3297 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 3298 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 3299 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c62b1a3b | 3300 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); |
14067bb3 | 3301 | s->stat[MCS_PGPGIN] += val; |
c62b1a3b | 3302 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
14067bb3 | 3303 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 3304 | if (do_swap_account) { |
c62b1a3b | 3305 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
3306 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
3307 | } | |
14067bb3 KH |
3308 | |
3309 | /* per zone stat */ | |
3310 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
3311 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
3312 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
3313 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
3314 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
3315 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
3316 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
3317 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
3318 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
3319 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
3320 | return 0; | |
3321 | } | |
3322 | ||
3323 | static void | |
3324 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
3325 | { | |
3326 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
3327 | } | |
3328 | ||
c64745cf PM |
3329 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
3330 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 3331 | { |
d2ceb9b7 | 3332 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 3333 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
3334 | int i; |
3335 | ||
14067bb3 KH |
3336 | memset(&mystat, 0, sizeof(mystat)); |
3337 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 3338 | |
1dd3a273 DN |
3339 | for (i = 0; i < NR_MCS_STAT; i++) { |
3340 | if (i == MCS_SWAP && !do_swap_account) | |
3341 | continue; | |
14067bb3 | 3342 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 3343 | } |
7b854121 | 3344 | |
14067bb3 | 3345 | /* Hierarchical information */ |
fee7b548 KH |
3346 | { |
3347 | unsigned long long limit, memsw_limit; | |
3348 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
3349 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
3350 | if (do_swap_account) | |
3351 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
3352 | } | |
7f016ee8 | 3353 | |
14067bb3 KH |
3354 | memset(&mystat, 0, sizeof(mystat)); |
3355 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
3356 | for (i = 0; i < NR_MCS_STAT; i++) { |
3357 | if (i == MCS_SWAP && !do_swap_account) | |
3358 | continue; | |
14067bb3 | 3359 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 3360 | } |
14067bb3 | 3361 | |
7f016ee8 | 3362 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 3363 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
3364 | |
3365 | { | |
3366 | int nid, zid; | |
3367 | struct mem_cgroup_per_zone *mz; | |
3368 | unsigned long recent_rotated[2] = {0, 0}; | |
3369 | unsigned long recent_scanned[2] = {0, 0}; | |
3370 | ||
3371 | for_each_online_node(nid) | |
3372 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3373 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3374 | ||
3375 | recent_rotated[0] += | |
3376 | mz->reclaim_stat.recent_rotated[0]; | |
3377 | recent_rotated[1] += | |
3378 | mz->reclaim_stat.recent_rotated[1]; | |
3379 | recent_scanned[0] += | |
3380 | mz->reclaim_stat.recent_scanned[0]; | |
3381 | recent_scanned[1] += | |
3382 | mz->reclaim_stat.recent_scanned[1]; | |
3383 | } | |
3384 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3385 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3386 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3387 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3388 | } | |
3389 | #endif | |
3390 | ||
d2ceb9b7 KH |
3391 | return 0; |
3392 | } | |
3393 | ||
a7885eb8 KM |
3394 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
3395 | { | |
3396 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3397 | ||
3398 | return get_swappiness(memcg); | |
3399 | } | |
3400 | ||
3401 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3402 | u64 val) | |
3403 | { | |
3404 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3405 | struct mem_cgroup *parent; | |
068b38c1 | 3406 | |
a7885eb8 KM |
3407 | if (val > 100) |
3408 | return -EINVAL; | |
3409 | ||
3410 | if (cgrp->parent == NULL) | |
3411 | return -EINVAL; | |
3412 | ||
3413 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
3414 | |
3415 | cgroup_lock(); | |
3416 | ||
a7885eb8 KM |
3417 | /* If under hierarchy, only empty-root can set this value */ |
3418 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
3419 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3420 | cgroup_unlock(); | |
a7885eb8 | 3421 | return -EINVAL; |
068b38c1 | 3422 | } |
a7885eb8 KM |
3423 | |
3424 | spin_lock(&memcg->reclaim_param_lock); | |
3425 | memcg->swappiness = val; | |
3426 | spin_unlock(&memcg->reclaim_param_lock); | |
3427 | ||
068b38c1 LZ |
3428 | cgroup_unlock(); |
3429 | ||
a7885eb8 KM |
3430 | return 0; |
3431 | } | |
3432 | ||
2e72b634 KS |
3433 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3434 | { | |
3435 | struct mem_cgroup_threshold_ary *t; | |
3436 | u64 usage; | |
3437 | int i; | |
3438 | ||
3439 | rcu_read_lock(); | |
3440 | if (!swap) | |
2c488db2 | 3441 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3442 | else |
2c488db2 | 3443 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3444 | |
3445 | if (!t) | |
3446 | goto unlock; | |
3447 | ||
3448 | usage = mem_cgroup_usage(memcg, swap); | |
3449 | ||
3450 | /* | |
3451 | * current_threshold points to threshold just below usage. | |
3452 | * If it's not true, a threshold was crossed after last | |
3453 | * call of __mem_cgroup_threshold(). | |
3454 | */ | |
5407a562 | 3455 | i = t->current_threshold; |
2e72b634 KS |
3456 | |
3457 | /* | |
3458 | * Iterate backward over array of thresholds starting from | |
3459 | * current_threshold and check if a threshold is crossed. | |
3460 | * If none of thresholds below usage is crossed, we read | |
3461 | * only one element of the array here. | |
3462 | */ | |
3463 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3464 | eventfd_signal(t->entries[i].eventfd, 1); | |
3465 | ||
3466 | /* i = current_threshold + 1 */ | |
3467 | i++; | |
3468 | ||
3469 | /* | |
3470 | * Iterate forward over array of thresholds starting from | |
3471 | * current_threshold+1 and check if a threshold is crossed. | |
3472 | * If none of thresholds above usage is crossed, we read | |
3473 | * only one element of the array here. | |
3474 | */ | |
3475 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3476 | eventfd_signal(t->entries[i].eventfd, 1); | |
3477 | ||
3478 | /* Update current_threshold */ | |
5407a562 | 3479 | t->current_threshold = i - 1; |
2e72b634 KS |
3480 | unlock: |
3481 | rcu_read_unlock(); | |
3482 | } | |
3483 | ||
3484 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3485 | { | |
3486 | __mem_cgroup_threshold(memcg, false); | |
3487 | if (do_swap_account) | |
3488 | __mem_cgroup_threshold(memcg, true); | |
3489 | } | |
3490 | ||
3491 | static int compare_thresholds(const void *a, const void *b) | |
3492 | { | |
3493 | const struct mem_cgroup_threshold *_a = a; | |
3494 | const struct mem_cgroup_threshold *_b = b; | |
3495 | ||
3496 | return _a->threshold - _b->threshold; | |
3497 | } | |
3498 | ||
9490ff27 KH |
3499 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem, void *data) |
3500 | { | |
3501 | struct mem_cgroup_eventfd_list *ev; | |
3502 | ||
3503 | list_for_each_entry(ev, &mem->oom_notify, list) | |
3504 | eventfd_signal(ev->eventfd, 1); | |
3505 | return 0; | |
3506 | } | |
3507 | ||
3508 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) | |
3509 | { | |
3510 | mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_notify_cb); | |
3511 | } | |
3512 | ||
3513 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
3514 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
3515 | { |
3516 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
3517 | struct mem_cgroup_thresholds *thresholds; |
3518 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
3519 | int type = MEMFILE_TYPE(cft->private); |
3520 | u64 threshold, usage; | |
2c488db2 | 3521 | int i, size, ret; |
2e72b634 KS |
3522 | |
3523 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
3524 | if (ret) | |
3525 | return ret; | |
3526 | ||
3527 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3528 | |
2e72b634 | 3529 | if (type == _MEM) |
2c488db2 | 3530 | thresholds = &memcg->thresholds; |
2e72b634 | 3531 | else if (type == _MEMSWAP) |
2c488db2 | 3532 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
3533 | else |
3534 | BUG(); | |
3535 | ||
3536 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3537 | ||
3538 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 3539 | if (thresholds->primary) |
2e72b634 KS |
3540 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3541 | ||
2c488db2 | 3542 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3543 | |
3544 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3545 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3546 | GFP_KERNEL); |
2c488db2 | 3547 | if (!new) { |
2e72b634 KS |
3548 | ret = -ENOMEM; |
3549 | goto unlock; | |
3550 | } | |
2c488db2 | 3551 | new->size = size; |
2e72b634 KS |
3552 | |
3553 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3554 | if (thresholds->primary) { |
3555 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3556 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3557 | } |
3558 | ||
2e72b634 | 3559 | /* Add new threshold */ |
2c488db2 KS |
3560 | new->entries[size - 1].eventfd = eventfd; |
3561 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3562 | |
3563 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3564 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3565 | compare_thresholds, NULL); |
3566 | ||
3567 | /* Find current threshold */ | |
2c488db2 | 3568 | new->current_threshold = -1; |
2e72b634 | 3569 | for (i = 0; i < size; i++) { |
2c488db2 | 3570 | if (new->entries[i].threshold < usage) { |
2e72b634 | 3571 | /* |
2c488db2 KS |
3572 | * new->current_threshold will not be used until |
3573 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3574 | * it here. |
3575 | */ | |
2c488db2 | 3576 | ++new->current_threshold; |
2e72b634 KS |
3577 | } |
3578 | } | |
3579 | ||
2c488db2 KS |
3580 | /* Free old spare buffer and save old primary buffer as spare */ |
3581 | kfree(thresholds->spare); | |
3582 | thresholds->spare = thresholds->primary; | |
3583 | ||
3584 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3585 | |
907860ed | 3586 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3587 | synchronize_rcu(); |
3588 | ||
2e72b634 KS |
3589 | unlock: |
3590 | mutex_unlock(&memcg->thresholds_lock); | |
3591 | ||
3592 | return ret; | |
3593 | } | |
3594 | ||
907860ed | 3595 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 3596 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
3597 | { |
3598 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
3599 | struct mem_cgroup_thresholds *thresholds; |
3600 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
3601 | int type = MEMFILE_TYPE(cft->private); |
3602 | u64 usage; | |
2c488db2 | 3603 | int i, j, size; |
2e72b634 KS |
3604 | |
3605 | mutex_lock(&memcg->thresholds_lock); | |
3606 | if (type == _MEM) | |
2c488db2 | 3607 | thresholds = &memcg->thresholds; |
2e72b634 | 3608 | else if (type == _MEMSWAP) |
2c488db2 | 3609 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
3610 | else |
3611 | BUG(); | |
3612 | ||
3613 | /* | |
3614 | * Something went wrong if we trying to unregister a threshold | |
3615 | * if we don't have thresholds | |
3616 | */ | |
3617 | BUG_ON(!thresholds); | |
3618 | ||
3619 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3620 | ||
3621 | /* Check if a threshold crossed before removing */ | |
3622 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3623 | ||
3624 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3625 | size = 0; |
3626 | for (i = 0; i < thresholds->primary->size; i++) { | |
3627 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3628 | size++; |
3629 | } | |
3630 | ||
2c488db2 | 3631 | new = thresholds->spare; |
907860ed | 3632 | |
2e72b634 KS |
3633 | /* Set thresholds array to NULL if we don't have thresholds */ |
3634 | if (!size) { | |
2c488db2 KS |
3635 | kfree(new); |
3636 | new = NULL; | |
907860ed | 3637 | goto swap_buffers; |
2e72b634 KS |
3638 | } |
3639 | ||
2c488db2 | 3640 | new->size = size; |
2e72b634 KS |
3641 | |
3642 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
3643 | new->current_threshold = -1; |
3644 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
3645 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
3646 | continue; |
3647 | ||
2c488db2 KS |
3648 | new->entries[j] = thresholds->primary->entries[i]; |
3649 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 3650 | /* |
2c488db2 | 3651 | * new->current_threshold will not be used |
2e72b634 KS |
3652 | * until rcu_assign_pointer(), so it's safe to increment |
3653 | * it here. | |
3654 | */ | |
2c488db2 | 3655 | ++new->current_threshold; |
2e72b634 KS |
3656 | } |
3657 | j++; | |
3658 | } | |
3659 | ||
907860ed | 3660 | swap_buffers: |
2c488db2 KS |
3661 | /* Swap primary and spare array */ |
3662 | thresholds->spare = thresholds->primary; | |
3663 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3664 | |
907860ed | 3665 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3666 | synchronize_rcu(); |
3667 | ||
2e72b634 | 3668 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 3669 | } |
c1e862c1 | 3670 | |
9490ff27 KH |
3671 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
3672 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
3673 | { | |
3674 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3675 | struct mem_cgroup_eventfd_list *event; | |
3676 | int type = MEMFILE_TYPE(cft->private); | |
3677 | ||
3678 | BUG_ON(type != _OOM_TYPE); | |
3679 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
3680 | if (!event) | |
3681 | return -ENOMEM; | |
3682 | ||
3683 | mutex_lock(&memcg_oom_mutex); | |
3684 | ||
3685 | event->eventfd = eventfd; | |
3686 | list_add(&event->list, &memcg->oom_notify); | |
3687 | ||
3688 | /* already in OOM ? */ | |
3689 | if (atomic_read(&memcg->oom_lock)) | |
3690 | eventfd_signal(eventfd, 1); | |
3691 | mutex_unlock(&memcg_oom_mutex); | |
3692 | ||
3693 | return 0; | |
3694 | } | |
3695 | ||
907860ed | 3696 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
3697 | struct cftype *cft, struct eventfd_ctx *eventfd) |
3698 | { | |
3699 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3700 | struct mem_cgroup_eventfd_list *ev, *tmp; | |
3701 | int type = MEMFILE_TYPE(cft->private); | |
3702 | ||
3703 | BUG_ON(type != _OOM_TYPE); | |
3704 | ||
3705 | mutex_lock(&memcg_oom_mutex); | |
3706 | ||
3707 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { | |
3708 | if (ev->eventfd == eventfd) { | |
3709 | list_del(&ev->list); | |
3710 | kfree(ev); | |
3711 | } | |
3712 | } | |
3713 | ||
3714 | mutex_unlock(&memcg_oom_mutex); | |
9490ff27 KH |
3715 | } |
3716 | ||
3c11ecf4 KH |
3717 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
3718 | struct cftype *cft, struct cgroup_map_cb *cb) | |
3719 | { | |
3720 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3721 | ||
3722 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); | |
3723 | ||
3724 | if (atomic_read(&mem->oom_lock)) | |
3725 | cb->fill(cb, "under_oom", 1); | |
3726 | else | |
3727 | cb->fill(cb, "under_oom", 0); | |
3728 | return 0; | |
3729 | } | |
3730 | ||
3731 | /* | |
3732 | */ | |
3733 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, | |
3734 | struct cftype *cft, u64 val) | |
3735 | { | |
3736 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3737 | struct mem_cgroup *parent; | |
3738 | ||
3739 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
3740 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
3741 | return -EINVAL; | |
3742 | ||
3743 | parent = mem_cgroup_from_cont(cgrp->parent); | |
3744 | ||
3745 | cgroup_lock(); | |
3746 | /* oom-kill-disable is a flag for subhierarchy. */ | |
3747 | if ((parent->use_hierarchy) || | |
3748 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { | |
3749 | cgroup_unlock(); | |
3750 | return -EINVAL; | |
3751 | } | |
3752 | mem->oom_kill_disable = val; | |
4d845ebf KH |
3753 | if (!val) |
3754 | memcg_oom_recover(mem); | |
3c11ecf4 KH |
3755 | cgroup_unlock(); |
3756 | return 0; | |
3757 | } | |
3758 | ||
8cdea7c0 BS |
3759 | static struct cftype mem_cgroup_files[] = { |
3760 | { | |
0eea1030 | 3761 | .name = "usage_in_bytes", |
8c7c6e34 | 3762 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 3763 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
3764 | .register_event = mem_cgroup_usage_register_event, |
3765 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 3766 | }, |
c84872e1 PE |
3767 | { |
3768 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3769 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 3770 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
3771 | .read_u64 = mem_cgroup_read, |
3772 | }, | |
8cdea7c0 | 3773 | { |
0eea1030 | 3774 | .name = "limit_in_bytes", |
8c7c6e34 | 3775 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 3776 | .write_string = mem_cgroup_write, |
2c3daa72 | 3777 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3778 | }, |
296c81d8 BS |
3779 | { |
3780 | .name = "soft_limit_in_bytes", | |
3781 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
3782 | .write_string = mem_cgroup_write, | |
3783 | .read_u64 = mem_cgroup_read, | |
3784 | }, | |
8cdea7c0 BS |
3785 | { |
3786 | .name = "failcnt", | |
8c7c6e34 | 3787 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 3788 | .trigger = mem_cgroup_reset, |
2c3daa72 | 3789 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3790 | }, |
d2ceb9b7 KH |
3791 | { |
3792 | .name = "stat", | |
c64745cf | 3793 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 3794 | }, |
c1e862c1 KH |
3795 | { |
3796 | .name = "force_empty", | |
3797 | .trigger = mem_cgroup_force_empty_write, | |
3798 | }, | |
18f59ea7 BS |
3799 | { |
3800 | .name = "use_hierarchy", | |
3801 | .write_u64 = mem_cgroup_hierarchy_write, | |
3802 | .read_u64 = mem_cgroup_hierarchy_read, | |
3803 | }, | |
a7885eb8 KM |
3804 | { |
3805 | .name = "swappiness", | |
3806 | .read_u64 = mem_cgroup_swappiness_read, | |
3807 | .write_u64 = mem_cgroup_swappiness_write, | |
3808 | }, | |
7dc74be0 DN |
3809 | { |
3810 | .name = "move_charge_at_immigrate", | |
3811 | .read_u64 = mem_cgroup_move_charge_read, | |
3812 | .write_u64 = mem_cgroup_move_charge_write, | |
3813 | }, | |
9490ff27 KH |
3814 | { |
3815 | .name = "oom_control", | |
3c11ecf4 KH |
3816 | .read_map = mem_cgroup_oom_control_read, |
3817 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
3818 | .register_event = mem_cgroup_oom_register_event, |
3819 | .unregister_event = mem_cgroup_oom_unregister_event, | |
3820 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
3821 | }, | |
8cdea7c0 BS |
3822 | }; |
3823 | ||
8c7c6e34 KH |
3824 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3825 | static struct cftype memsw_cgroup_files[] = { | |
3826 | { | |
3827 | .name = "memsw.usage_in_bytes", | |
3828 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
3829 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
3830 | .register_event = mem_cgroup_usage_register_event, |
3831 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
3832 | }, |
3833 | { | |
3834 | .name = "memsw.max_usage_in_bytes", | |
3835 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
3836 | .trigger = mem_cgroup_reset, | |
3837 | .read_u64 = mem_cgroup_read, | |
3838 | }, | |
3839 | { | |
3840 | .name = "memsw.limit_in_bytes", | |
3841 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
3842 | .write_string = mem_cgroup_write, | |
3843 | .read_u64 = mem_cgroup_read, | |
3844 | }, | |
3845 | { | |
3846 | .name = "memsw.failcnt", | |
3847 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
3848 | .trigger = mem_cgroup_reset, | |
3849 | .read_u64 = mem_cgroup_read, | |
3850 | }, | |
3851 | }; | |
3852 | ||
3853 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3854 | { | |
3855 | if (!do_swap_account) | |
3856 | return 0; | |
3857 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
3858 | ARRAY_SIZE(memsw_cgroup_files)); | |
3859 | }; | |
3860 | #else | |
3861 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3862 | { | |
3863 | return 0; | |
3864 | } | |
3865 | #endif | |
3866 | ||
6d12e2d8 KH |
3867 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3868 | { | |
3869 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 3870 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 3871 | enum lru_list l; |
41e3355d | 3872 | int zone, tmp = node; |
1ecaab2b KH |
3873 | /* |
3874 | * This routine is called against possible nodes. | |
3875 | * But it's BUG to call kmalloc() against offline node. | |
3876 | * | |
3877 | * TODO: this routine can waste much memory for nodes which will | |
3878 | * never be onlined. It's better to use memory hotplug callback | |
3879 | * function. | |
3880 | */ | |
41e3355d KH |
3881 | if (!node_state(node, N_NORMAL_MEMORY)) |
3882 | tmp = -1; | |
3883 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
3884 | if (!pn) |
3885 | return 1; | |
1ecaab2b | 3886 | |
6d12e2d8 KH |
3887 | mem->info.nodeinfo[node] = pn; |
3888 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
3889 | |
3890 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3891 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
3892 | for_each_lru(l) |
3893 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 3894 | mz->usage_in_excess = 0; |
4e416953 BS |
3895 | mz->on_tree = false; |
3896 | mz->mem = mem; | |
1ecaab2b | 3897 | } |
6d12e2d8 KH |
3898 | return 0; |
3899 | } | |
3900 | ||
1ecaab2b KH |
3901 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3902 | { | |
3903 | kfree(mem->info.nodeinfo[node]); | |
3904 | } | |
3905 | ||
33327948 KH |
3906 | static struct mem_cgroup *mem_cgroup_alloc(void) |
3907 | { | |
3908 | struct mem_cgroup *mem; | |
c62b1a3b | 3909 | int size = sizeof(struct mem_cgroup); |
33327948 | 3910 | |
c62b1a3b | 3911 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb JB |
3912 | if (size < PAGE_SIZE) |
3913 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 3914 | else |
c8dad2bb | 3915 | mem = vmalloc(size); |
33327948 | 3916 | |
e7bbcdf3 DC |
3917 | if (!mem) |
3918 | return NULL; | |
3919 | ||
3920 | memset(mem, 0, size); | |
c62b1a3b KH |
3921 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
3922 | if (!mem->stat) { | |
3923 | if (size < PAGE_SIZE) | |
3924 | kfree(mem); | |
3925 | else | |
3926 | vfree(mem); | |
3927 | mem = NULL; | |
3928 | } | |
33327948 KH |
3929 | return mem; |
3930 | } | |
3931 | ||
8c7c6e34 KH |
3932 | /* |
3933 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
3934 | * (scanning all at force_empty is too costly...) | |
3935 | * | |
3936 | * Instead of clearing all references at force_empty, we remember | |
3937 | * the number of reference from swap_cgroup and free mem_cgroup when | |
3938 | * it goes down to 0. | |
3939 | * | |
8c7c6e34 KH |
3940 | * Removal of cgroup itself succeeds regardless of refs from swap. |
3941 | */ | |
3942 | ||
a7ba0eef | 3943 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 3944 | { |
08e552c6 KH |
3945 | int node; |
3946 | ||
f64c3f54 | 3947 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
3948 | free_css_id(&mem_cgroup_subsys, &mem->css); |
3949 | ||
08e552c6 KH |
3950 | for_each_node_state(node, N_POSSIBLE) |
3951 | free_mem_cgroup_per_zone_info(mem, node); | |
3952 | ||
c62b1a3b KH |
3953 | free_percpu(mem->stat); |
3954 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
3955 | kfree(mem); |
3956 | else | |
3957 | vfree(mem); | |
3958 | } | |
3959 | ||
8c7c6e34 KH |
3960 | static void mem_cgroup_get(struct mem_cgroup *mem) |
3961 | { | |
3962 | atomic_inc(&mem->refcnt); | |
3963 | } | |
3964 | ||
483c30b5 | 3965 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 3966 | { |
483c30b5 | 3967 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 3968 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 3969 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
3970 | if (parent) |
3971 | mem_cgroup_put(parent); | |
3972 | } | |
8c7c6e34 KH |
3973 | } |
3974 | ||
483c30b5 DN |
3975 | static void mem_cgroup_put(struct mem_cgroup *mem) |
3976 | { | |
3977 | __mem_cgroup_put(mem, 1); | |
3978 | } | |
3979 | ||
7bcc1bb1 DN |
3980 | /* |
3981 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
3982 | */ | |
3983 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
3984 | { | |
3985 | if (!mem->res.parent) | |
3986 | return NULL; | |
3987 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
3988 | } | |
33327948 | 3989 | |
c077719b KH |
3990 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3991 | static void __init enable_swap_cgroup(void) | |
3992 | { | |
f8d66542 | 3993 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
3994 | do_swap_account = 1; |
3995 | } | |
3996 | #else | |
3997 | static void __init enable_swap_cgroup(void) | |
3998 | { | |
3999 | } | |
4000 | #endif | |
4001 | ||
f64c3f54 BS |
4002 | static int mem_cgroup_soft_limit_tree_init(void) |
4003 | { | |
4004 | struct mem_cgroup_tree_per_node *rtpn; | |
4005 | struct mem_cgroup_tree_per_zone *rtpz; | |
4006 | int tmp, node, zone; | |
4007 | ||
4008 | for_each_node_state(node, N_POSSIBLE) { | |
4009 | tmp = node; | |
4010 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4011 | tmp = -1; | |
4012 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4013 | if (!rtpn) | |
4014 | return 1; | |
4015 | ||
4016 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4017 | ||
4018 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4019 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4020 | rtpz->rb_root = RB_ROOT; | |
4021 | spin_lock_init(&rtpz->lock); | |
4022 | } | |
4023 | } | |
4024 | return 0; | |
4025 | } | |
4026 | ||
0eb253e2 | 4027 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4028 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4029 | { | |
28dbc4b6 | 4030 | struct mem_cgroup *mem, *parent; |
04046e1a | 4031 | long error = -ENOMEM; |
6d12e2d8 | 4032 | int node; |
8cdea7c0 | 4033 | |
c8dad2bb JB |
4034 | mem = mem_cgroup_alloc(); |
4035 | if (!mem) | |
04046e1a | 4036 | return ERR_PTR(error); |
78fb7466 | 4037 | |
6d12e2d8 KH |
4038 | for_each_node_state(node, N_POSSIBLE) |
4039 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
4040 | goto free_out; | |
f64c3f54 | 4041 | |
c077719b | 4042 | /* root ? */ |
28dbc4b6 | 4043 | if (cont->parent == NULL) { |
cdec2e42 | 4044 | int cpu; |
c077719b | 4045 | enable_swap_cgroup(); |
28dbc4b6 | 4046 | parent = NULL; |
4b3bde4c | 4047 | root_mem_cgroup = mem; |
f64c3f54 BS |
4048 | if (mem_cgroup_soft_limit_tree_init()) |
4049 | goto free_out; | |
cdec2e42 KH |
4050 | for_each_possible_cpu(cpu) { |
4051 | struct memcg_stock_pcp *stock = | |
4052 | &per_cpu(memcg_stock, cpu); | |
4053 | INIT_WORK(&stock->work, drain_local_stock); | |
4054 | } | |
4055 | hotcpu_notifier(memcg_stock_cpu_callback, 0); | |
18f59ea7 | 4056 | } else { |
28dbc4b6 | 4057 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 | 4058 | mem->use_hierarchy = parent->use_hierarchy; |
3c11ecf4 | 4059 | mem->oom_kill_disable = parent->oom_kill_disable; |
18f59ea7 | 4060 | } |
28dbc4b6 | 4061 | |
18f59ea7 BS |
4062 | if (parent && parent->use_hierarchy) { |
4063 | res_counter_init(&mem->res, &parent->res); | |
4064 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4065 | /* |
4066 | * We increment refcnt of the parent to ensure that we can | |
4067 | * safely access it on res_counter_charge/uncharge. | |
4068 | * This refcnt will be decremented when freeing this | |
4069 | * mem_cgroup(see mem_cgroup_put). | |
4070 | */ | |
4071 | mem_cgroup_get(parent); | |
18f59ea7 BS |
4072 | } else { |
4073 | res_counter_init(&mem->res, NULL); | |
4074 | res_counter_init(&mem->memsw, NULL); | |
4075 | } | |
04046e1a | 4076 | mem->last_scanned_child = 0; |
2733c06a | 4077 | spin_lock_init(&mem->reclaim_param_lock); |
9490ff27 | 4078 | INIT_LIST_HEAD(&mem->oom_notify); |
6d61ef40 | 4079 | |
a7885eb8 KM |
4080 | if (parent) |
4081 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 4082 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 4083 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 4084 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 4085 | return &mem->css; |
6d12e2d8 | 4086 | free_out: |
a7ba0eef | 4087 | __mem_cgroup_free(mem); |
4b3bde4c | 4088 | root_mem_cgroup = NULL; |
04046e1a | 4089 | return ERR_PTR(error); |
8cdea7c0 BS |
4090 | } |
4091 | ||
ec64f515 | 4092 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4093 | struct cgroup *cont) |
4094 | { | |
4095 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
4096 | |
4097 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
4098 | } |
4099 | ||
8cdea7c0 BS |
4100 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4101 | struct cgroup *cont) | |
4102 | { | |
c268e994 | 4103 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 4104 | |
c268e994 | 4105 | mem_cgroup_put(mem); |
8cdea7c0 BS |
4106 | } |
4107 | ||
4108 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4109 | struct cgroup *cont) | |
4110 | { | |
8c7c6e34 KH |
4111 | int ret; |
4112 | ||
4113 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4114 | ARRAY_SIZE(mem_cgroup_files)); | |
4115 | ||
4116 | if (!ret) | |
4117 | ret = register_memsw_files(cont, ss); | |
4118 | return ret; | |
8cdea7c0 BS |
4119 | } |
4120 | ||
02491447 | 4121 | #ifdef CONFIG_MMU |
7dc74be0 | 4122 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4123 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4124 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4125 | { |
854ffa8d DN |
4126 | int ret = 0; |
4127 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
4128 | struct mem_cgroup *mem = mc.to; |
4129 | ||
854ffa8d DN |
4130 | if (mem_cgroup_is_root(mem)) { |
4131 | mc.precharge += count; | |
4132 | /* we don't need css_get for root */ | |
4133 | return ret; | |
4134 | } | |
4135 | /* try to charge at once */ | |
4136 | if (count > 1) { | |
4137 | struct res_counter *dummy; | |
4138 | /* | |
4139 | * "mem" cannot be under rmdir() because we've already checked | |
4140 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
4141 | * are still under the same cgroup_mutex. So we can postpone | |
4142 | * css_get(). | |
4143 | */ | |
4144 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
4145 | goto one_by_one; | |
4146 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
4147 | PAGE_SIZE * count, &dummy)) { | |
4148 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
4149 | goto one_by_one; | |
4150 | } | |
4151 | mc.precharge += count; | |
4152 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
4153 | WARN_ON_ONCE(count > INT_MAX); | |
4154 | __css_get(&mem->css, (int)count); | |
4155 | return ret; | |
4156 | } | |
4157 | one_by_one: | |
4158 | /* fall back to one by one charge */ | |
4159 | while (count--) { | |
4160 | if (signal_pending(current)) { | |
4161 | ret = -EINTR; | |
4162 | break; | |
4163 | } | |
4164 | if (!batch_count--) { | |
4165 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4166 | cond_resched(); | |
4167 | } | |
430e4863 | 4168 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
854ffa8d DN |
4169 | if (ret || !mem) |
4170 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
4171 | return -ENOMEM; | |
4172 | mc.precharge++; | |
4173 | } | |
4ffef5fe DN |
4174 | return ret; |
4175 | } | |
4176 | ||
4177 | /** | |
4178 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
4179 | * @vma: the vma the pte to be checked belongs | |
4180 | * @addr: the address corresponding to the pte to be checked | |
4181 | * @ptent: the pte to be checked | |
02491447 | 4182 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4183 | * |
4184 | * Returns | |
4185 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4186 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4187 | * move charge. if @target is not NULL, the page is stored in target->page | |
4188 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4189 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4190 | * target for charge migration. if @target is not NULL, the entry is stored | |
4191 | * in target->ent. | |
4ffef5fe DN |
4192 | * |
4193 | * Called with pte lock held. | |
4194 | */ | |
4ffef5fe DN |
4195 | union mc_target { |
4196 | struct page *page; | |
02491447 | 4197 | swp_entry_t ent; |
4ffef5fe DN |
4198 | }; |
4199 | ||
4ffef5fe DN |
4200 | enum mc_target_type { |
4201 | MC_TARGET_NONE, /* not used */ | |
4202 | MC_TARGET_PAGE, | |
02491447 | 4203 | MC_TARGET_SWAP, |
4ffef5fe DN |
4204 | }; |
4205 | ||
90254a65 DN |
4206 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4207 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4208 | { |
90254a65 | 4209 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4210 | |
90254a65 DN |
4211 | if (!page || !page_mapped(page)) |
4212 | return NULL; | |
4213 | if (PageAnon(page)) { | |
4214 | /* we don't move shared anon */ | |
4215 | if (!move_anon() || page_mapcount(page) > 2) | |
4216 | return NULL; | |
87946a72 DN |
4217 | } else if (!move_file()) |
4218 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
4219 | return NULL; |
4220 | if (!get_page_unless_zero(page)) | |
4221 | return NULL; | |
4222 | ||
4223 | return page; | |
4224 | } | |
4225 | ||
4226 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4227 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4228 | { | |
4229 | int usage_count; | |
4230 | struct page *page = NULL; | |
4231 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4232 | ||
4233 | if (!move_anon() || non_swap_entry(ent)) | |
4234 | return NULL; | |
4235 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
4236 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
4237 | if (page) |
4238 | put_page(page); | |
90254a65 | 4239 | return NULL; |
02491447 | 4240 | } |
90254a65 DN |
4241 | if (do_swap_account) |
4242 | entry->val = ent.val; | |
4243 | ||
4244 | return page; | |
4245 | } | |
4246 | ||
87946a72 DN |
4247 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4248 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4249 | { | |
4250 | struct page *page = NULL; | |
4251 | struct inode *inode; | |
4252 | struct address_space *mapping; | |
4253 | pgoff_t pgoff; | |
4254 | ||
4255 | if (!vma->vm_file) /* anonymous vma */ | |
4256 | return NULL; | |
4257 | if (!move_file()) | |
4258 | return NULL; | |
4259 | ||
4260 | inode = vma->vm_file->f_path.dentry->d_inode; | |
4261 | mapping = vma->vm_file->f_mapping; | |
4262 | if (pte_none(ptent)) | |
4263 | pgoff = linear_page_index(vma, addr); | |
4264 | else /* pte_file(ptent) is true */ | |
4265 | pgoff = pte_to_pgoff(ptent); | |
4266 | ||
4267 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
4268 | if (!mapping_cap_swap_backed(mapping)) { /* normal file */ | |
4269 | page = find_get_page(mapping, pgoff); | |
4270 | } else { /* shmem/tmpfs file. we should take account of swap too. */ | |
4271 | swp_entry_t ent; | |
4272 | mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent); | |
4273 | if (do_swap_account) | |
4274 | entry->val = ent.val; | |
4275 | } | |
4276 | ||
4277 | return page; | |
4278 | } | |
4279 | ||
90254a65 DN |
4280 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
4281 | unsigned long addr, pte_t ptent, union mc_target *target) | |
4282 | { | |
4283 | struct page *page = NULL; | |
4284 | struct page_cgroup *pc; | |
4285 | int ret = 0; | |
4286 | swp_entry_t ent = { .val = 0 }; | |
4287 | ||
4288 | if (pte_present(ptent)) | |
4289 | page = mc_handle_present_pte(vma, addr, ptent); | |
4290 | else if (is_swap_pte(ptent)) | |
4291 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
4292 | else if (pte_none(ptent) || pte_file(ptent)) |
4293 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
4294 | |
4295 | if (!page && !ent.val) | |
4296 | return 0; | |
02491447 DN |
4297 | if (page) { |
4298 | pc = lookup_page_cgroup(page); | |
4299 | /* | |
4300 | * Do only loose check w/o page_cgroup lock. | |
4301 | * mem_cgroup_move_account() checks the pc is valid or not under | |
4302 | * the lock. | |
4303 | */ | |
4304 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
4305 | ret = MC_TARGET_PAGE; | |
4306 | if (target) | |
4307 | target->page = page; | |
4308 | } | |
4309 | if (!ret || !target) | |
4310 | put_page(page); | |
4311 | } | |
90254a65 DN |
4312 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4313 | if (ent.val && !ret && | |
7f0f1546 KH |
4314 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
4315 | ret = MC_TARGET_SWAP; | |
4316 | if (target) | |
4317 | target->ent = ent; | |
4ffef5fe | 4318 | } |
4ffef5fe DN |
4319 | return ret; |
4320 | } | |
4321 | ||
4322 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
4323 | unsigned long addr, unsigned long end, | |
4324 | struct mm_walk *walk) | |
4325 | { | |
4326 | struct vm_area_struct *vma = walk->private; | |
4327 | pte_t *pte; | |
4328 | spinlock_t *ptl; | |
4329 | ||
4330 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4331 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
4332 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
4333 | mc.precharge++; /* increment precharge temporarily */ | |
4334 | pte_unmap_unlock(pte - 1, ptl); | |
4335 | cond_resched(); | |
4336 | ||
7dc74be0 DN |
4337 | return 0; |
4338 | } | |
4339 | ||
4ffef5fe DN |
4340 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4341 | { | |
4342 | unsigned long precharge; | |
4343 | struct vm_area_struct *vma; | |
4344 | ||
4345 | down_read(&mm->mmap_sem); | |
4346 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
4347 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
4348 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4349 | .mm = mm, | |
4350 | .private = vma, | |
4351 | }; | |
4352 | if (is_vm_hugetlb_page(vma)) | |
4353 | continue; | |
4ffef5fe DN |
4354 | walk_page_range(vma->vm_start, vma->vm_end, |
4355 | &mem_cgroup_count_precharge_walk); | |
4356 | } | |
4357 | up_read(&mm->mmap_sem); | |
4358 | ||
4359 | precharge = mc.precharge; | |
4360 | mc.precharge = 0; | |
4361 | ||
4362 | return precharge; | |
4363 | } | |
4364 | ||
4ffef5fe DN |
4365 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4366 | { | |
854ffa8d | 4367 | return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm)); |
4ffef5fe DN |
4368 | } |
4369 | ||
4370 | static void mem_cgroup_clear_mc(void) | |
4371 | { | |
4372 | /* we must uncharge all the leftover precharges from mc.to */ | |
854ffa8d DN |
4373 | if (mc.precharge) { |
4374 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
4375 | mc.precharge = 0; | |
3c11ecf4 | 4376 | memcg_oom_recover(mc.to); |
854ffa8d DN |
4377 | } |
4378 | /* | |
4379 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4380 | * we must uncharge here. | |
4381 | */ | |
4382 | if (mc.moved_charge) { | |
4383 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
4384 | mc.moved_charge = 0; | |
3c11ecf4 | 4385 | memcg_oom_recover(mc.from); |
4ffef5fe | 4386 | } |
483c30b5 DN |
4387 | /* we must fixup refcnts and charges */ |
4388 | if (mc.moved_swap) { | |
4389 | WARN_ON_ONCE(mc.moved_swap > INT_MAX); | |
4390 | /* uncharge swap account from the old cgroup */ | |
4391 | if (!mem_cgroup_is_root(mc.from)) | |
4392 | res_counter_uncharge(&mc.from->memsw, | |
4393 | PAGE_SIZE * mc.moved_swap); | |
4394 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
4395 | ||
4396 | if (!mem_cgroup_is_root(mc.to)) { | |
4397 | /* | |
4398 | * we charged both to->res and to->memsw, so we should | |
4399 | * uncharge to->res. | |
4400 | */ | |
4401 | res_counter_uncharge(&mc.to->res, | |
4402 | PAGE_SIZE * mc.moved_swap); | |
4403 | VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags)); | |
4404 | __css_put(&mc.to->css, mc.moved_swap); | |
4405 | } | |
4406 | /* we've already done mem_cgroup_get(mc.to) */ | |
4407 | ||
4408 | mc.moved_swap = 0; | |
4409 | } | |
4ffef5fe DN |
4410 | mc.from = NULL; |
4411 | mc.to = NULL; | |
8033b97c DN |
4412 | mc.moving_task = NULL; |
4413 | wake_up_all(&mc.waitq); | |
4ffef5fe DN |
4414 | } |
4415 | ||
7dc74be0 DN |
4416 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
4417 | struct cgroup *cgroup, | |
4418 | struct task_struct *p, | |
4419 | bool threadgroup) | |
4420 | { | |
4421 | int ret = 0; | |
4422 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
4423 | ||
4424 | if (mem->move_charge_at_immigrate) { | |
4425 | struct mm_struct *mm; | |
4426 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
4427 | ||
4428 | VM_BUG_ON(from == mem); | |
4429 | ||
4430 | mm = get_task_mm(p); | |
4431 | if (!mm) | |
4432 | return 0; | |
7dc74be0 | 4433 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
4434 | if (mm->owner == p) { |
4435 | VM_BUG_ON(mc.from); | |
4436 | VM_BUG_ON(mc.to); | |
4437 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 4438 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 4439 | VM_BUG_ON(mc.moved_swap); |
8033b97c | 4440 | VM_BUG_ON(mc.moving_task); |
4ffef5fe DN |
4441 | mc.from = from; |
4442 | mc.to = mem; | |
4443 | mc.precharge = 0; | |
854ffa8d | 4444 | mc.moved_charge = 0; |
483c30b5 | 4445 | mc.moved_swap = 0; |
8033b97c | 4446 | mc.moving_task = current; |
4ffef5fe DN |
4447 | |
4448 | ret = mem_cgroup_precharge_mc(mm); | |
4449 | if (ret) | |
4450 | mem_cgroup_clear_mc(); | |
4451 | } | |
7dc74be0 DN |
4452 | mmput(mm); |
4453 | } | |
4454 | return ret; | |
4455 | } | |
4456 | ||
4457 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4458 | struct cgroup *cgroup, | |
4459 | struct task_struct *p, | |
4460 | bool threadgroup) | |
4461 | { | |
4ffef5fe | 4462 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
4463 | } |
4464 | ||
4ffef5fe DN |
4465 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4466 | unsigned long addr, unsigned long end, | |
4467 | struct mm_walk *walk) | |
7dc74be0 | 4468 | { |
4ffef5fe DN |
4469 | int ret = 0; |
4470 | struct vm_area_struct *vma = walk->private; | |
4471 | pte_t *pte; | |
4472 | spinlock_t *ptl; | |
4473 | ||
4474 | retry: | |
4475 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4476 | for (; addr != end; addr += PAGE_SIZE) { | |
4477 | pte_t ptent = *(pte++); | |
4478 | union mc_target target; | |
4479 | int type; | |
4480 | struct page *page; | |
4481 | struct page_cgroup *pc; | |
02491447 | 4482 | swp_entry_t ent; |
4ffef5fe DN |
4483 | |
4484 | if (!mc.precharge) | |
4485 | break; | |
4486 | ||
4487 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
4488 | switch (type) { | |
4489 | case MC_TARGET_PAGE: | |
4490 | page = target.page; | |
4491 | if (isolate_lru_page(page)) | |
4492 | goto put; | |
4493 | pc = lookup_page_cgroup(page); | |
854ffa8d DN |
4494 | if (!mem_cgroup_move_account(pc, |
4495 | mc.from, mc.to, false)) { | |
4ffef5fe | 4496 | mc.precharge--; |
854ffa8d DN |
4497 | /* we uncharge from mc.from later. */ |
4498 | mc.moved_charge++; | |
4ffef5fe DN |
4499 | } |
4500 | putback_lru_page(page); | |
4501 | put: /* is_target_pte_for_mc() gets the page */ | |
4502 | put_page(page); | |
4503 | break; | |
02491447 DN |
4504 | case MC_TARGET_SWAP: |
4505 | ent = target.ent; | |
483c30b5 DN |
4506 | if (!mem_cgroup_move_swap_account(ent, |
4507 | mc.from, mc.to, false)) { | |
02491447 | 4508 | mc.precharge--; |
483c30b5 DN |
4509 | /* we fixup refcnts and charges later. */ |
4510 | mc.moved_swap++; | |
4511 | } | |
02491447 | 4512 | break; |
4ffef5fe DN |
4513 | default: |
4514 | break; | |
4515 | } | |
4516 | } | |
4517 | pte_unmap_unlock(pte - 1, ptl); | |
4518 | cond_resched(); | |
4519 | ||
4520 | if (addr != end) { | |
4521 | /* | |
4522 | * We have consumed all precharges we got in can_attach(). | |
4523 | * We try charge one by one, but don't do any additional | |
4524 | * charges to mc.to if we have failed in charge once in attach() | |
4525 | * phase. | |
4526 | */ | |
854ffa8d | 4527 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4528 | if (!ret) |
4529 | goto retry; | |
4530 | } | |
4531 | ||
4532 | return ret; | |
4533 | } | |
4534 | ||
4535 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4536 | { | |
4537 | struct vm_area_struct *vma; | |
4538 | ||
4539 | lru_add_drain_all(); | |
4540 | down_read(&mm->mmap_sem); | |
4541 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
4542 | int ret; | |
4543 | struct mm_walk mem_cgroup_move_charge_walk = { | |
4544 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4545 | .mm = mm, | |
4546 | .private = vma, | |
4547 | }; | |
4548 | if (is_vm_hugetlb_page(vma)) | |
4549 | continue; | |
4ffef5fe DN |
4550 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
4551 | &mem_cgroup_move_charge_walk); | |
4552 | if (ret) | |
4553 | /* | |
4554 | * means we have consumed all precharges and failed in | |
4555 | * doing additional charge. Just abandon here. | |
4556 | */ | |
4557 | break; | |
4558 | } | |
4559 | up_read(&mm->mmap_sem); | |
7dc74be0 DN |
4560 | } |
4561 | ||
67e465a7 BS |
4562 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
4563 | struct cgroup *cont, | |
4564 | struct cgroup *old_cont, | |
be367d09 BB |
4565 | struct task_struct *p, |
4566 | bool threadgroup) | |
67e465a7 | 4567 | { |
4ffef5fe DN |
4568 | struct mm_struct *mm; |
4569 | ||
4570 | if (!mc.to) | |
4571 | /* no need to move charge */ | |
4572 | return; | |
4573 | ||
4574 | mm = get_task_mm(p); | |
4575 | if (mm) { | |
4576 | mem_cgroup_move_charge(mm); | |
4577 | mmput(mm); | |
4578 | } | |
4579 | mem_cgroup_clear_mc(); | |
67e465a7 | 4580 | } |
5cfb80a7 DN |
4581 | #else /* !CONFIG_MMU */ |
4582 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
4583 | struct cgroup *cgroup, | |
4584 | struct task_struct *p, | |
4585 | bool threadgroup) | |
4586 | { | |
4587 | return 0; | |
4588 | } | |
4589 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4590 | struct cgroup *cgroup, | |
4591 | struct task_struct *p, | |
4592 | bool threadgroup) | |
4593 | { | |
4594 | } | |
4595 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
4596 | struct cgroup *cont, | |
4597 | struct cgroup *old_cont, | |
4598 | struct task_struct *p, | |
4599 | bool threadgroup) | |
4600 | { | |
4601 | } | |
4602 | #endif | |
67e465a7 | 4603 | |
8cdea7c0 BS |
4604 | struct cgroup_subsys mem_cgroup_subsys = { |
4605 | .name = "memory", | |
4606 | .subsys_id = mem_cgroup_subsys_id, | |
4607 | .create = mem_cgroup_create, | |
df878fb0 | 4608 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
4609 | .destroy = mem_cgroup_destroy, |
4610 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
4611 | .can_attach = mem_cgroup_can_attach, |
4612 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 4613 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 4614 | .early_init = 0, |
04046e1a | 4615 | .use_id = 1, |
8cdea7c0 | 4616 | }; |
c077719b KH |
4617 | |
4618 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
4619 | ||
4620 | static int __init disable_swap_account(char *s) | |
4621 | { | |
4622 | really_do_swap_account = 0; | |
4623 | return 1; | |
4624 | } | |
4625 | __setup("noswapaccount", disable_swap_account); | |
4626 | #endif |