]>
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. */ | |
155 | atomic_t current_threshold; | |
156 | /* Size of entries[] */ | |
157 | unsigned int size; | |
158 | /* Array of thresholds */ | |
159 | struct mem_cgroup_threshold entries[0]; | |
160 | }; | |
9490ff27 KH |
161 | /* for OOM */ |
162 | struct mem_cgroup_eventfd_list { | |
163 | struct list_head list; | |
164 | struct eventfd_ctx *eventfd; | |
165 | }; | |
2e72b634 | 166 | |
2e72b634 | 167 | static void mem_cgroup_threshold(struct mem_cgroup *mem); |
9490ff27 | 168 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem); |
2e72b634 | 169 | |
8cdea7c0 BS |
170 | /* |
171 | * The memory controller data structure. The memory controller controls both | |
172 | * page cache and RSS per cgroup. We would eventually like to provide | |
173 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
174 | * to help the administrator determine what knobs to tune. | |
175 | * | |
176 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
177 | * we hit the water mark. May be even add a low water mark, such that |
178 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
179 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
180 | */ |
181 | struct mem_cgroup { | |
182 | struct cgroup_subsys_state css; | |
183 | /* | |
184 | * the counter to account for memory usage | |
185 | */ | |
186 | struct res_counter res; | |
8c7c6e34 KH |
187 | /* |
188 | * the counter to account for mem+swap usage. | |
189 | */ | |
190 | struct res_counter memsw; | |
78fb7466 PE |
191 | /* |
192 | * Per cgroup active and inactive list, similar to the | |
193 | * per zone LRU lists. | |
78fb7466 | 194 | */ |
6d12e2d8 | 195 | struct mem_cgroup_lru_info info; |
072c56c1 | 196 | |
2733c06a KM |
197 | /* |
198 | protect against reclaim related member. | |
199 | */ | |
200 | spinlock_t reclaim_param_lock; | |
201 | ||
6c48a1d0 | 202 | int prev_priority; /* for recording reclaim priority */ |
6d61ef40 BS |
203 | |
204 | /* | |
af901ca1 | 205 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 206 | * reclaimed from. |
6d61ef40 | 207 | */ |
04046e1a | 208 | int last_scanned_child; |
18f59ea7 BS |
209 | /* |
210 | * Should the accounting and control be hierarchical, per subtree? | |
211 | */ | |
212 | bool use_hierarchy; | |
867578cb | 213 | atomic_t oom_lock; |
8c7c6e34 | 214 | atomic_t refcnt; |
14797e23 | 215 | |
a7885eb8 | 216 | unsigned int swappiness; |
3c11ecf4 KH |
217 | /* OOM-Killer disable */ |
218 | int oom_kill_disable; | |
a7885eb8 | 219 | |
22a668d7 KH |
220 | /* set when res.limit == memsw.limit */ |
221 | bool memsw_is_minimum; | |
222 | ||
2e72b634 KS |
223 | /* protect arrays of thresholds */ |
224 | struct mutex thresholds_lock; | |
225 | ||
226 | /* thresholds for memory usage. RCU-protected */ | |
227 | struct mem_cgroup_threshold_ary *thresholds; | |
228 | ||
229 | /* thresholds for mem+swap usage. RCU-protected */ | |
230 | struct mem_cgroup_threshold_ary *memsw_thresholds; | |
231 | ||
9490ff27 KH |
232 | /* For oom notifier event fd */ |
233 | struct list_head oom_notify; | |
234 | ||
7dc74be0 DN |
235 | /* |
236 | * Should we move charges of a task when a task is moved into this | |
237 | * mem_cgroup ? And what type of charges should we move ? | |
238 | */ | |
239 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 240 | /* |
c62b1a3b | 241 | * percpu counter. |
d52aa412 | 242 | */ |
c62b1a3b | 243 | struct mem_cgroup_stat_cpu *stat; |
8cdea7c0 BS |
244 | }; |
245 | ||
7dc74be0 DN |
246 | /* Stuffs for move charges at task migration. */ |
247 | /* | |
248 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
249 | * left-shifted bitmap of these types. | |
250 | */ | |
251 | enum move_type { | |
4ffef5fe | 252 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
7dc74be0 DN |
253 | NR_MOVE_TYPE, |
254 | }; | |
255 | ||
4ffef5fe DN |
256 | /* "mc" and its members are protected by cgroup_mutex */ |
257 | static struct move_charge_struct { | |
258 | struct mem_cgroup *from; | |
259 | struct mem_cgroup *to; | |
260 | unsigned long precharge; | |
854ffa8d | 261 | unsigned long moved_charge; |
483c30b5 | 262 | unsigned long moved_swap; |
8033b97c DN |
263 | struct task_struct *moving_task; /* a task moving charges */ |
264 | wait_queue_head_t waitq; /* a waitq for other context */ | |
265 | } mc = { | |
266 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), | |
267 | }; | |
4ffef5fe | 268 | |
90254a65 DN |
269 | static bool move_anon(void) |
270 | { | |
271 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
272 | &mc.to->move_charge_at_immigrate); | |
273 | } | |
274 | ||
4e416953 BS |
275 | /* |
276 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
277 | * limit reclaim to prevent infinite loops, if they ever occur. | |
278 | */ | |
279 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
280 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
281 | ||
217bc319 KH |
282 | enum charge_type { |
283 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
284 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 285 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 286 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 287 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 288 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
289 | NR_CHARGE_TYPE, |
290 | }; | |
291 | ||
52d4b9ac KH |
292 | /* only for here (for easy reading.) */ |
293 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
294 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 295 | #define PCGF_LOCK (1UL << PCG_LOCK) |
4b3bde4c BS |
296 | /* Not used, but added here for completeness */ |
297 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
217bc319 | 298 | |
8c7c6e34 KH |
299 | /* for encoding cft->private value on file */ |
300 | #define _MEM (0) | |
301 | #define _MEMSWAP (1) | |
9490ff27 | 302 | #define _OOM_TYPE (2) |
8c7c6e34 KH |
303 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
304 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
305 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
306 | /* Used for OOM nofiier */ |
307 | #define OOM_CONTROL (0) | |
8c7c6e34 | 308 | |
75822b44 BS |
309 | /* |
310 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
311 | */ | |
312 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
313 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
314 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
315 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
316 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
317 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 318 | |
8c7c6e34 KH |
319 | static void mem_cgroup_get(struct mem_cgroup *mem); |
320 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 321 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 322 | static void drain_all_stock_async(void); |
8c7c6e34 | 323 | |
f64c3f54 BS |
324 | static struct mem_cgroup_per_zone * |
325 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
326 | { | |
327 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
328 | } | |
329 | ||
d324236b WF |
330 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
331 | { | |
332 | return &mem->css; | |
333 | } | |
334 | ||
f64c3f54 BS |
335 | static struct mem_cgroup_per_zone * |
336 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
337 | { | |
338 | struct mem_cgroup *mem = pc->mem_cgroup; | |
339 | int nid = page_cgroup_nid(pc); | |
340 | int zid = page_cgroup_zid(pc); | |
341 | ||
342 | if (!mem) | |
343 | return NULL; | |
344 | ||
345 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
346 | } | |
347 | ||
348 | static struct mem_cgroup_tree_per_zone * | |
349 | soft_limit_tree_node_zone(int nid, int zid) | |
350 | { | |
351 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
352 | } | |
353 | ||
354 | static struct mem_cgroup_tree_per_zone * | |
355 | soft_limit_tree_from_page(struct page *page) | |
356 | { | |
357 | int nid = page_to_nid(page); | |
358 | int zid = page_zonenum(page); | |
359 | ||
360 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
361 | } | |
362 | ||
363 | static void | |
4e416953 | 364 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 365 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
366 | struct mem_cgroup_tree_per_zone *mctz, |
367 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
368 | { |
369 | struct rb_node **p = &mctz->rb_root.rb_node; | |
370 | struct rb_node *parent = NULL; | |
371 | struct mem_cgroup_per_zone *mz_node; | |
372 | ||
373 | if (mz->on_tree) | |
374 | return; | |
375 | ||
ef8745c1 KH |
376 | mz->usage_in_excess = new_usage_in_excess; |
377 | if (!mz->usage_in_excess) | |
378 | return; | |
f64c3f54 BS |
379 | while (*p) { |
380 | parent = *p; | |
381 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
382 | tree_node); | |
383 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
384 | p = &(*p)->rb_left; | |
385 | /* | |
386 | * We can't avoid mem cgroups that are over their soft | |
387 | * limit by the same amount | |
388 | */ | |
389 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
390 | p = &(*p)->rb_right; | |
391 | } | |
392 | rb_link_node(&mz->tree_node, parent, p); | |
393 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
394 | mz->on_tree = true; | |
4e416953 BS |
395 | } |
396 | ||
397 | static void | |
398 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
399 | struct mem_cgroup_per_zone *mz, | |
400 | struct mem_cgroup_tree_per_zone *mctz) | |
401 | { | |
402 | if (!mz->on_tree) | |
403 | return; | |
404 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
405 | mz->on_tree = false; | |
406 | } | |
407 | ||
f64c3f54 BS |
408 | static void |
409 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
410 | struct mem_cgroup_per_zone *mz, | |
411 | struct mem_cgroup_tree_per_zone *mctz) | |
412 | { | |
413 | spin_lock(&mctz->lock); | |
4e416953 | 414 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
415 | spin_unlock(&mctz->lock); |
416 | } | |
417 | ||
f64c3f54 BS |
418 | |
419 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
420 | { | |
ef8745c1 | 421 | unsigned long long excess; |
f64c3f54 BS |
422 | struct mem_cgroup_per_zone *mz; |
423 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
424 | int nid = page_to_nid(page); |
425 | int zid = page_zonenum(page); | |
f64c3f54 BS |
426 | mctz = soft_limit_tree_from_page(page); |
427 | ||
428 | /* | |
4e649152 KH |
429 | * Necessary to update all ancestors when hierarchy is used. |
430 | * because their event counter is not touched. | |
f64c3f54 | 431 | */ |
4e649152 KH |
432 | for (; mem; mem = parent_mem_cgroup(mem)) { |
433 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 434 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
435 | /* |
436 | * We have to update the tree if mz is on RB-tree or | |
437 | * mem is over its softlimit. | |
438 | */ | |
ef8745c1 | 439 | if (excess || mz->on_tree) { |
4e649152 KH |
440 | spin_lock(&mctz->lock); |
441 | /* if on-tree, remove it */ | |
442 | if (mz->on_tree) | |
443 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
444 | /* | |
ef8745c1 KH |
445 | * Insert again. mz->usage_in_excess will be updated. |
446 | * If excess is 0, no tree ops. | |
4e649152 | 447 | */ |
ef8745c1 | 448 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
449 | spin_unlock(&mctz->lock); |
450 | } | |
f64c3f54 BS |
451 | } |
452 | } | |
453 | ||
454 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
455 | { | |
456 | int node, zone; | |
457 | struct mem_cgroup_per_zone *mz; | |
458 | struct mem_cgroup_tree_per_zone *mctz; | |
459 | ||
460 | for_each_node_state(node, N_POSSIBLE) { | |
461 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
462 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
463 | mctz = soft_limit_tree_node_zone(node, zone); | |
464 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
465 | } | |
466 | } | |
467 | } | |
468 | ||
4e416953 BS |
469 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
470 | { | |
471 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; | |
472 | } | |
473 | ||
474 | static struct mem_cgroup_per_zone * | |
475 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
476 | { | |
477 | struct rb_node *rightmost = NULL; | |
26251eaf | 478 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
479 | |
480 | retry: | |
26251eaf | 481 | mz = NULL; |
4e416953 BS |
482 | rightmost = rb_last(&mctz->rb_root); |
483 | if (!rightmost) | |
484 | goto done; /* Nothing to reclaim from */ | |
485 | ||
486 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
487 | /* | |
488 | * Remove the node now but someone else can add it back, | |
489 | * we will to add it back at the end of reclaim to its correct | |
490 | * position in the tree. | |
491 | */ | |
492 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
493 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
494 | !css_tryget(&mz->mem->css)) | |
495 | goto retry; | |
496 | done: | |
497 | return mz; | |
498 | } | |
499 | ||
500 | static struct mem_cgroup_per_zone * | |
501 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
502 | { | |
503 | struct mem_cgroup_per_zone *mz; | |
504 | ||
505 | spin_lock(&mctz->lock); | |
506 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
507 | spin_unlock(&mctz->lock); | |
508 | return mz; | |
509 | } | |
510 | ||
c62b1a3b KH |
511 | static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, |
512 | enum mem_cgroup_stat_index idx) | |
513 | { | |
514 | int cpu; | |
515 | s64 val = 0; | |
516 | ||
517 | for_each_possible_cpu(cpu) | |
518 | val += per_cpu(mem->stat->count[idx], cpu); | |
519 | return val; | |
520 | } | |
521 | ||
522 | static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) | |
523 | { | |
524 | s64 ret; | |
525 | ||
526 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); | |
527 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); | |
528 | return ret; | |
529 | } | |
530 | ||
0c3e73e8 BS |
531 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
532 | bool charge) | |
533 | { | |
534 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 535 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
536 | } |
537 | ||
c05555b5 KH |
538 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
539 | struct page_cgroup *pc, | |
540 | bool charge) | |
d52aa412 | 541 | { |
0c3e73e8 | 542 | int val = (charge) ? 1 : -1; |
d52aa412 | 543 | |
c62b1a3b KH |
544 | preempt_disable(); |
545 | ||
c05555b5 | 546 | if (PageCgroupCache(pc)) |
c62b1a3b | 547 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val); |
d52aa412 | 548 | else |
c62b1a3b | 549 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val); |
55e462b0 BR |
550 | |
551 | if (charge) | |
c62b1a3b | 552 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); |
55e462b0 | 553 | else |
c62b1a3b | 554 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); |
d2265e6f | 555 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]); |
2e72b634 | 556 | |
c62b1a3b | 557 | preempt_enable(); |
6d12e2d8 KH |
558 | } |
559 | ||
14067bb3 | 560 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 561 | enum lru_list idx) |
6d12e2d8 KH |
562 | { |
563 | int nid, zid; | |
564 | struct mem_cgroup_per_zone *mz; | |
565 | u64 total = 0; | |
566 | ||
567 | for_each_online_node(nid) | |
568 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
569 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
570 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
571 | } | |
572 | return total; | |
d52aa412 KH |
573 | } |
574 | ||
d2265e6f KH |
575 | static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift) |
576 | { | |
577 | s64 val; | |
578 | ||
579 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]); | |
580 | ||
581 | return !(val & ((1 << event_mask_shift) - 1)); | |
582 | } | |
583 | ||
584 | /* | |
585 | * Check events in order. | |
586 | * | |
587 | */ | |
588 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) | |
589 | { | |
590 | /* threshold event is triggered in finer grain than soft limit */ | |
591 | if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) { | |
592 | mem_cgroup_threshold(mem); | |
593 | if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH))) | |
594 | mem_cgroup_update_tree(mem, page); | |
595 | } | |
596 | } | |
597 | ||
d5b69e38 | 598 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
599 | { |
600 | return container_of(cgroup_subsys_state(cont, | |
601 | mem_cgroup_subsys_id), struct mem_cgroup, | |
602 | css); | |
603 | } | |
604 | ||
cf475ad2 | 605 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 606 | { |
31a78f23 BS |
607 | /* |
608 | * mm_update_next_owner() may clear mm->owner to NULL | |
609 | * if it races with swapoff, page migration, etc. | |
610 | * So this can be called with p == NULL. | |
611 | */ | |
612 | if (unlikely(!p)) | |
613 | return NULL; | |
614 | ||
78fb7466 PE |
615 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
616 | struct mem_cgroup, css); | |
617 | } | |
618 | ||
54595fe2 KH |
619 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
620 | { | |
621 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
622 | |
623 | if (!mm) | |
624 | return NULL; | |
54595fe2 KH |
625 | /* |
626 | * Because we have no locks, mm->owner's may be being moved to other | |
627 | * cgroup. We use css_tryget() here even if this looks | |
628 | * pessimistic (rather than adding locks here). | |
629 | */ | |
630 | rcu_read_lock(); | |
631 | do { | |
632 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
633 | if (unlikely(!mem)) | |
634 | break; | |
635 | } while (!css_tryget(&mem->css)); | |
636 | rcu_read_unlock(); | |
637 | return mem; | |
638 | } | |
639 | ||
14067bb3 KH |
640 | /* |
641 | * Call callback function against all cgroup under hierarchy tree. | |
642 | */ | |
643 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
644 | int (*func)(struct mem_cgroup *, void *)) | |
645 | { | |
646 | int found, ret, nextid; | |
647 | struct cgroup_subsys_state *css; | |
648 | struct mem_cgroup *mem; | |
649 | ||
650 | if (!root->use_hierarchy) | |
651 | return (*func)(root, data); | |
652 | ||
653 | nextid = 1; | |
654 | do { | |
655 | ret = 0; | |
656 | mem = NULL; | |
657 | ||
658 | rcu_read_lock(); | |
659 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
660 | &found); | |
661 | if (css && css_tryget(css)) | |
662 | mem = container_of(css, struct mem_cgroup, css); | |
663 | rcu_read_unlock(); | |
664 | ||
665 | if (mem) { | |
666 | ret = (*func)(mem, data); | |
667 | css_put(&mem->css); | |
668 | } | |
669 | nextid = found + 1; | |
670 | } while (!ret && css); | |
671 | ||
672 | return ret; | |
673 | } | |
674 | ||
4b3bde4c BS |
675 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
676 | { | |
677 | return (mem == root_mem_cgroup); | |
678 | } | |
679 | ||
08e552c6 KH |
680 | /* |
681 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
682 | * Operations are called by routine of global LRU independently from memcg. | |
683 | * What we have to take care of here is validness of pc->mem_cgroup. | |
684 | * | |
685 | * Changes to pc->mem_cgroup happens when | |
686 | * 1. charge | |
687 | * 2. moving account | |
688 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
689 | * It is added to LRU before charge. | |
690 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
691 | * When moving account, the page is not on LRU. It's isolated. | |
692 | */ | |
4f98a2fe | 693 | |
08e552c6 KH |
694 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
695 | { | |
696 | struct page_cgroup *pc; | |
08e552c6 | 697 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 698 | |
f8d66542 | 699 | if (mem_cgroup_disabled()) |
08e552c6 KH |
700 | return; |
701 | pc = lookup_page_cgroup(page); | |
702 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 703 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 704 | return; |
4b3bde4c | 705 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
706 | /* |
707 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
708 | * removed from global LRU. | |
709 | */ | |
08e552c6 | 710 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 711 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
4b3bde4c BS |
712 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
713 | return; | |
714 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 KH |
715 | list_del_init(&pc->lru); |
716 | return; | |
6d12e2d8 KH |
717 | } |
718 | ||
08e552c6 | 719 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 720 | { |
08e552c6 KH |
721 | mem_cgroup_del_lru_list(page, page_lru(page)); |
722 | } | |
b69408e8 | 723 | |
08e552c6 KH |
724 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
725 | { | |
726 | struct mem_cgroup_per_zone *mz; | |
727 | struct page_cgroup *pc; | |
b69408e8 | 728 | |
f8d66542 | 729 | if (mem_cgroup_disabled()) |
08e552c6 | 730 | return; |
6d12e2d8 | 731 | |
08e552c6 | 732 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
733 | /* |
734 | * Used bit is set without atomic ops but after smp_wmb(). | |
735 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
736 | */ | |
08e552c6 | 737 | smp_rmb(); |
4b3bde4c BS |
738 | /* unused or root page is not rotated. */ |
739 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 KH |
740 | return; |
741 | mz = page_cgroup_zoneinfo(pc); | |
742 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
743 | } |
744 | ||
08e552c6 | 745 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 746 | { |
08e552c6 KH |
747 | struct page_cgroup *pc; |
748 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 749 | |
f8d66542 | 750 | if (mem_cgroup_disabled()) |
08e552c6 KH |
751 | return; |
752 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 753 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
bd112db8 DN |
754 | /* |
755 | * Used bit is set without atomic ops but after smp_wmb(). | |
756 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
757 | */ | |
08e552c6 KH |
758 | smp_rmb(); |
759 | if (!PageCgroupUsed(pc)) | |
894bc310 | 760 | return; |
b69408e8 | 761 | |
08e552c6 | 762 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 763 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
4b3bde4c BS |
764 | SetPageCgroupAcctLRU(pc); |
765 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
766 | return; | |
08e552c6 KH |
767 | list_add(&pc->lru, &mz->lists[lru]); |
768 | } | |
544122e5 | 769 | |
08e552c6 | 770 | /* |
544122e5 KH |
771 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
772 | * lru because the page may.be reused after it's fully uncharged (because of | |
773 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
774 | * it again. This function is only used to charge SwapCache. It's done under | |
775 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 776 | */ |
544122e5 | 777 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 778 | { |
544122e5 KH |
779 | unsigned long flags; |
780 | struct zone *zone = page_zone(page); | |
781 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
782 | ||
783 | spin_lock_irqsave(&zone->lru_lock, flags); | |
784 | /* | |
785 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
786 | * is guarded by lock_page() because the page is SwapCache. | |
787 | */ | |
788 | if (!PageCgroupUsed(pc)) | |
789 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
790 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
791 | } |
792 | ||
544122e5 KH |
793 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
794 | { | |
795 | unsigned long flags; | |
796 | struct zone *zone = page_zone(page); | |
797 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
798 | ||
799 | spin_lock_irqsave(&zone->lru_lock, flags); | |
800 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 801 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
802 | mem_cgroup_add_lru_list(page, page_lru(page)); |
803 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
804 | } | |
805 | ||
806 | ||
08e552c6 KH |
807 | void mem_cgroup_move_lists(struct page *page, |
808 | enum lru_list from, enum lru_list to) | |
809 | { | |
f8d66542 | 810 | if (mem_cgroup_disabled()) |
08e552c6 KH |
811 | return; |
812 | mem_cgroup_del_lru_list(page, from); | |
813 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
814 | } |
815 | ||
4c4a2214 DR |
816 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
817 | { | |
818 | int ret; | |
0b7f569e | 819 | struct mem_cgroup *curr = NULL; |
4c4a2214 DR |
820 | |
821 | task_lock(task); | |
0b7f569e KH |
822 | rcu_read_lock(); |
823 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
824 | rcu_read_unlock(); | |
4c4a2214 | 825 | task_unlock(task); |
0b7f569e KH |
826 | if (!curr) |
827 | return 0; | |
d31f56db DN |
828 | /* |
829 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
830 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
831 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
832 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
833 | */ | |
834 | if (mem->use_hierarchy) | |
0b7f569e KH |
835 | ret = css_is_ancestor(&curr->css, &mem->css); |
836 | else | |
837 | ret = (curr == mem); | |
838 | css_put(&curr->css); | |
4c4a2214 DR |
839 | return ret; |
840 | } | |
841 | ||
6c48a1d0 KH |
842 | /* |
843 | * prev_priority control...this will be used in memory reclaim path. | |
844 | */ | |
845 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
846 | { | |
2733c06a KM |
847 | int prev_priority; |
848 | ||
849 | spin_lock(&mem->reclaim_param_lock); | |
850 | prev_priority = mem->prev_priority; | |
851 | spin_unlock(&mem->reclaim_param_lock); | |
852 | ||
853 | return prev_priority; | |
6c48a1d0 KH |
854 | } |
855 | ||
856 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
857 | { | |
2733c06a | 858 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
859 | if (priority < mem->prev_priority) |
860 | mem->prev_priority = priority; | |
2733c06a | 861 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
862 | } |
863 | ||
864 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
865 | { | |
2733c06a | 866 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 867 | mem->prev_priority = priority; |
2733c06a | 868 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
869 | } |
870 | ||
c772be93 | 871 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
872 | { |
873 | unsigned long active; | |
874 | unsigned long inactive; | |
c772be93 KM |
875 | unsigned long gb; |
876 | unsigned long inactive_ratio; | |
14797e23 | 877 | |
14067bb3 KH |
878 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
879 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 880 | |
c772be93 KM |
881 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
882 | if (gb) | |
883 | inactive_ratio = int_sqrt(10 * gb); | |
884 | else | |
885 | inactive_ratio = 1; | |
886 | ||
887 | if (present_pages) { | |
888 | present_pages[0] = inactive; | |
889 | present_pages[1] = active; | |
890 | } | |
891 | ||
892 | return inactive_ratio; | |
893 | } | |
894 | ||
895 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
896 | { | |
897 | unsigned long active; | |
898 | unsigned long inactive; | |
899 | unsigned long present_pages[2]; | |
900 | unsigned long inactive_ratio; | |
901 | ||
902 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
903 | ||
904 | inactive = present_pages[0]; | |
905 | active = present_pages[1]; | |
906 | ||
907 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
908 | return 1; |
909 | ||
910 | return 0; | |
911 | } | |
912 | ||
56e49d21 RR |
913 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
914 | { | |
915 | unsigned long active; | |
916 | unsigned long inactive; | |
917 | ||
918 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
919 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
920 | ||
921 | return (active > inactive); | |
922 | } | |
923 | ||
a3d8e054 KM |
924 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
925 | struct zone *zone, | |
926 | enum lru_list lru) | |
927 | { | |
928 | int nid = zone->zone_pgdat->node_id; | |
929 | int zid = zone_idx(zone); | |
930 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
931 | ||
932 | return MEM_CGROUP_ZSTAT(mz, lru); | |
933 | } | |
934 | ||
3e2f41f1 KM |
935 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
936 | struct zone *zone) | |
937 | { | |
938 | int nid = zone->zone_pgdat->node_id; | |
939 | int zid = zone_idx(zone); | |
940 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
941 | ||
942 | return &mz->reclaim_stat; | |
943 | } | |
944 | ||
945 | struct zone_reclaim_stat * | |
946 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
947 | { | |
948 | struct page_cgroup *pc; | |
949 | struct mem_cgroup_per_zone *mz; | |
950 | ||
951 | if (mem_cgroup_disabled()) | |
952 | return NULL; | |
953 | ||
954 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
955 | /* |
956 | * Used bit is set without atomic ops but after smp_wmb(). | |
957 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
958 | */ | |
959 | smp_rmb(); | |
960 | if (!PageCgroupUsed(pc)) | |
961 | return NULL; | |
962 | ||
3e2f41f1 KM |
963 | mz = page_cgroup_zoneinfo(pc); |
964 | if (!mz) | |
965 | return NULL; | |
966 | ||
967 | return &mz->reclaim_stat; | |
968 | } | |
969 | ||
66e1707b BS |
970 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
971 | struct list_head *dst, | |
972 | unsigned long *scanned, int order, | |
973 | int mode, struct zone *z, | |
974 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 975 | int active, int file) |
66e1707b BS |
976 | { |
977 | unsigned long nr_taken = 0; | |
978 | struct page *page; | |
979 | unsigned long scan; | |
980 | LIST_HEAD(pc_list); | |
981 | struct list_head *src; | |
ff7283fa | 982 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
983 | int nid = z->zone_pgdat->node_id; |
984 | int zid = zone_idx(z); | |
985 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 986 | int lru = LRU_FILE * file + active; |
2ffebca6 | 987 | int ret; |
66e1707b | 988 | |
cf475ad2 | 989 | BUG_ON(!mem_cont); |
1ecaab2b | 990 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 991 | src = &mz->lists[lru]; |
66e1707b | 992 | |
ff7283fa KH |
993 | scan = 0; |
994 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 995 | if (scan >= nr_to_scan) |
ff7283fa | 996 | break; |
08e552c6 KH |
997 | |
998 | page = pc->page; | |
52d4b9ac KH |
999 | if (unlikely(!PageCgroupUsed(pc))) |
1000 | continue; | |
436c6541 | 1001 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1002 | continue; |
ff7283fa | 1003 | |
436c6541 | 1004 | scan++; |
2ffebca6 KH |
1005 | ret = __isolate_lru_page(page, mode, file); |
1006 | switch (ret) { | |
1007 | case 0: | |
66e1707b | 1008 | list_move(&page->lru, dst); |
2ffebca6 | 1009 | mem_cgroup_del_lru(page); |
66e1707b | 1010 | nr_taken++; |
2ffebca6 KH |
1011 | break; |
1012 | case -EBUSY: | |
1013 | /* we don't affect global LRU but rotate in our LRU */ | |
1014 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1015 | break; | |
1016 | default: | |
1017 | break; | |
66e1707b BS |
1018 | } |
1019 | } | |
1020 | ||
66e1707b BS |
1021 | *scanned = scan; |
1022 | return nr_taken; | |
1023 | } | |
1024 | ||
6d61ef40 BS |
1025 | #define mem_cgroup_from_res_counter(counter, member) \ |
1026 | container_of(counter, struct mem_cgroup, member) | |
1027 | ||
b85a96c0 DN |
1028 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
1029 | { | |
1030 | if (do_swap_account) { | |
1031 | if (res_counter_check_under_limit(&mem->res) && | |
1032 | res_counter_check_under_limit(&mem->memsw)) | |
1033 | return true; | |
1034 | } else | |
1035 | if (res_counter_check_under_limit(&mem->res)) | |
1036 | return true; | |
1037 | return false; | |
1038 | } | |
1039 | ||
a7885eb8 KM |
1040 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1041 | { | |
1042 | struct cgroup *cgrp = memcg->css.cgroup; | |
1043 | unsigned int swappiness; | |
1044 | ||
1045 | /* root ? */ | |
1046 | if (cgrp->parent == NULL) | |
1047 | return vm_swappiness; | |
1048 | ||
1049 | spin_lock(&memcg->reclaim_param_lock); | |
1050 | swappiness = memcg->swappiness; | |
1051 | spin_unlock(&memcg->reclaim_param_lock); | |
1052 | ||
1053 | return swappiness; | |
1054 | } | |
1055 | ||
81d39c20 KH |
1056 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
1057 | { | |
1058 | int *val = data; | |
1059 | (*val)++; | |
1060 | return 0; | |
1061 | } | |
e222432b BS |
1062 | |
1063 | /** | |
6a6135b6 | 1064 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1065 | * @memcg: The memory cgroup that went over limit |
1066 | * @p: Task that is going to be killed | |
1067 | * | |
1068 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1069 | * enabled | |
1070 | */ | |
1071 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1072 | { | |
1073 | struct cgroup *task_cgrp; | |
1074 | struct cgroup *mem_cgrp; | |
1075 | /* | |
1076 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1077 | * on the assumption that OOM is serialized for memory controller. | |
1078 | * If this assumption is broken, revisit this code. | |
1079 | */ | |
1080 | static char memcg_name[PATH_MAX]; | |
1081 | int ret; | |
1082 | ||
d31f56db | 1083 | if (!memcg || !p) |
e222432b BS |
1084 | return; |
1085 | ||
1086 | ||
1087 | rcu_read_lock(); | |
1088 | ||
1089 | mem_cgrp = memcg->css.cgroup; | |
1090 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1091 | ||
1092 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1093 | if (ret < 0) { | |
1094 | /* | |
1095 | * Unfortunately, we are unable to convert to a useful name | |
1096 | * But we'll still print out the usage information | |
1097 | */ | |
1098 | rcu_read_unlock(); | |
1099 | goto done; | |
1100 | } | |
1101 | rcu_read_unlock(); | |
1102 | ||
1103 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1104 | ||
1105 | rcu_read_lock(); | |
1106 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1107 | if (ret < 0) { | |
1108 | rcu_read_unlock(); | |
1109 | goto done; | |
1110 | } | |
1111 | rcu_read_unlock(); | |
1112 | ||
1113 | /* | |
1114 | * Continues from above, so we don't need an KERN_ level | |
1115 | */ | |
1116 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1117 | done: | |
1118 | ||
1119 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1120 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1121 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1122 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1123 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1124 | "failcnt %llu\n", | |
1125 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1126 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1127 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1128 | } | |
1129 | ||
81d39c20 KH |
1130 | /* |
1131 | * This function returns the number of memcg under hierarchy tree. Returns | |
1132 | * 1(self count) if no children. | |
1133 | */ | |
1134 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1135 | { | |
1136 | int num = 0; | |
1137 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
1138 | return num; | |
1139 | } | |
1140 | ||
6d61ef40 | 1141 | /* |
04046e1a KH |
1142 | * Visit the first child (need not be the first child as per the ordering |
1143 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1144 | * that to reclaim free pages from. | |
1145 | */ | |
1146 | static struct mem_cgroup * | |
1147 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1148 | { | |
1149 | struct mem_cgroup *ret = NULL; | |
1150 | struct cgroup_subsys_state *css; | |
1151 | int nextid, found; | |
1152 | ||
1153 | if (!root_mem->use_hierarchy) { | |
1154 | css_get(&root_mem->css); | |
1155 | ret = root_mem; | |
1156 | } | |
1157 | ||
1158 | while (!ret) { | |
1159 | rcu_read_lock(); | |
1160 | nextid = root_mem->last_scanned_child + 1; | |
1161 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1162 | &found); | |
1163 | if (css && css_tryget(css)) | |
1164 | ret = container_of(css, struct mem_cgroup, css); | |
1165 | ||
1166 | rcu_read_unlock(); | |
1167 | /* Updates scanning parameter */ | |
1168 | spin_lock(&root_mem->reclaim_param_lock); | |
1169 | if (!css) { | |
1170 | /* this means start scan from ID:1 */ | |
1171 | root_mem->last_scanned_child = 0; | |
1172 | } else | |
1173 | root_mem->last_scanned_child = found; | |
1174 | spin_unlock(&root_mem->reclaim_param_lock); | |
1175 | } | |
1176 | ||
1177 | return ret; | |
1178 | } | |
1179 | ||
1180 | /* | |
1181 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1182 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1183 | * based on its position in the children list. | |
6d61ef40 BS |
1184 | * |
1185 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1186 | * |
1187 | * We give up and return to the caller when we visit root_mem twice. | |
1188 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1189 | * |
1190 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1191 | */ |
1192 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1193 | struct zone *zone, |
75822b44 BS |
1194 | gfp_t gfp_mask, |
1195 | unsigned long reclaim_options) | |
6d61ef40 | 1196 | { |
04046e1a KH |
1197 | struct mem_cgroup *victim; |
1198 | int ret, total = 0; | |
1199 | int loop = 0; | |
75822b44 BS |
1200 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1201 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 BS |
1202 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
1203 | unsigned long excess = mem_cgroup_get_excess(root_mem); | |
04046e1a | 1204 | |
22a668d7 KH |
1205 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1206 | if (root_mem->memsw_is_minimum) | |
1207 | noswap = true; | |
1208 | ||
4e416953 | 1209 | while (1) { |
04046e1a | 1210 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1211 | if (victim == root_mem) { |
04046e1a | 1212 | loop++; |
cdec2e42 KH |
1213 | if (loop >= 1) |
1214 | drain_all_stock_async(); | |
4e416953 BS |
1215 | if (loop >= 2) { |
1216 | /* | |
1217 | * If we have not been able to reclaim | |
1218 | * anything, it might because there are | |
1219 | * no reclaimable pages under this hierarchy | |
1220 | */ | |
1221 | if (!check_soft || !total) { | |
1222 | css_put(&victim->css); | |
1223 | break; | |
1224 | } | |
1225 | /* | |
1226 | * We want to do more targetted reclaim. | |
1227 | * excess >> 2 is not to excessive so as to | |
1228 | * reclaim too much, nor too less that we keep | |
1229 | * coming back to reclaim from this cgroup | |
1230 | */ | |
1231 | if (total >= (excess >> 2) || | |
1232 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1233 | css_put(&victim->css); | |
1234 | break; | |
1235 | } | |
1236 | } | |
1237 | } | |
c62b1a3b | 1238 | if (!mem_cgroup_local_usage(victim)) { |
04046e1a KH |
1239 | /* this cgroup's local usage == 0 */ |
1240 | css_put(&victim->css); | |
6d61ef40 BS |
1241 | continue; |
1242 | } | |
04046e1a | 1243 | /* we use swappiness of local cgroup */ |
4e416953 BS |
1244 | if (check_soft) |
1245 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
1246 | noswap, get_swappiness(victim), zone, | |
1247 | zone->zone_pgdat->node_id); | |
1248 | else | |
1249 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1250 | noswap, get_swappiness(victim)); | |
04046e1a | 1251 | css_put(&victim->css); |
81d39c20 KH |
1252 | /* |
1253 | * At shrinking usage, we can't check we should stop here or | |
1254 | * reclaim more. It's depends on callers. last_scanned_child | |
1255 | * will work enough for keeping fairness under tree. | |
1256 | */ | |
1257 | if (shrink) | |
1258 | return ret; | |
04046e1a | 1259 | total += ret; |
4e416953 BS |
1260 | if (check_soft) { |
1261 | if (res_counter_check_under_soft_limit(&root_mem->res)) | |
1262 | return total; | |
1263 | } else if (mem_cgroup_check_under_limit(root_mem)) | |
04046e1a | 1264 | return 1 + total; |
6d61ef40 | 1265 | } |
04046e1a | 1266 | return total; |
6d61ef40 BS |
1267 | } |
1268 | ||
867578cb | 1269 | static int mem_cgroup_oom_lock_cb(struct mem_cgroup *mem, void *data) |
a636b327 | 1270 | { |
867578cb KH |
1271 | int *val = (int *)data; |
1272 | int x; | |
1273 | /* | |
1274 | * Logically, we can stop scanning immediately when we find | |
1275 | * a memcg is already locked. But condidering unlock ops and | |
1276 | * creation/removal of memcg, scan-all is simple operation. | |
1277 | */ | |
1278 | x = atomic_inc_return(&mem->oom_lock); | |
1279 | *val = max(x, *val); | |
1280 | return 0; | |
1281 | } | |
1282 | /* | |
1283 | * Check OOM-Killer is already running under our hierarchy. | |
1284 | * If someone is running, return false. | |
1285 | */ | |
1286 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) | |
1287 | { | |
1288 | int lock_count = 0; | |
a636b327 | 1289 | |
867578cb KH |
1290 | mem_cgroup_walk_tree(mem, &lock_count, mem_cgroup_oom_lock_cb); |
1291 | ||
1292 | if (lock_count == 1) | |
1293 | return true; | |
1294 | return false; | |
a636b327 | 1295 | } |
0b7f569e | 1296 | |
867578cb | 1297 | static int mem_cgroup_oom_unlock_cb(struct mem_cgroup *mem, void *data) |
0b7f569e | 1298 | { |
867578cb KH |
1299 | /* |
1300 | * When a new child is created while the hierarchy is under oom, | |
1301 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1302 | * atomic_add_unless() here. | |
1303 | */ | |
1304 | atomic_add_unless(&mem->oom_lock, -1, 0); | |
0b7f569e KH |
1305 | return 0; |
1306 | } | |
1307 | ||
867578cb KH |
1308 | static void mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
1309 | { | |
1310 | mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_unlock_cb); | |
1311 | } | |
1312 | ||
1313 | static DEFINE_MUTEX(memcg_oom_mutex); | |
1314 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); | |
1315 | ||
dc98df5a KH |
1316 | struct oom_wait_info { |
1317 | struct mem_cgroup *mem; | |
1318 | wait_queue_t wait; | |
1319 | }; | |
1320 | ||
1321 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1322 | unsigned mode, int sync, void *arg) | |
1323 | { | |
1324 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; | |
1325 | struct oom_wait_info *oom_wait_info; | |
1326 | ||
1327 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
1328 | ||
1329 | if (oom_wait_info->mem == wake_mem) | |
1330 | goto wakeup; | |
1331 | /* if no hierarchy, no match */ | |
1332 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) | |
1333 | return 0; | |
1334 | /* | |
1335 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1336 | * Then we can use css_is_ancestor without taking care of RCU. | |
1337 | */ | |
1338 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && | |
1339 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) | |
1340 | return 0; | |
1341 | ||
1342 | wakeup: | |
1343 | return autoremove_wake_function(wait, mode, sync, arg); | |
1344 | } | |
1345 | ||
1346 | static void memcg_wakeup_oom(struct mem_cgroup *mem) | |
1347 | { | |
1348 | /* for filtering, pass "mem" as argument. */ | |
1349 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); | |
1350 | } | |
1351 | ||
3c11ecf4 KH |
1352 | static void memcg_oom_recover(struct mem_cgroup *mem) |
1353 | { | |
1354 | if (mem->oom_kill_disable && atomic_read(&mem->oom_lock)) | |
1355 | memcg_wakeup_oom(mem); | |
1356 | } | |
1357 | ||
867578cb KH |
1358 | /* |
1359 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1360 | */ | |
1361 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) | |
0b7f569e | 1362 | { |
dc98df5a | 1363 | struct oom_wait_info owait; |
3c11ecf4 | 1364 | bool locked, need_to_kill; |
867578cb | 1365 | |
dc98df5a KH |
1366 | owait.mem = mem; |
1367 | owait.wait.flags = 0; | |
1368 | owait.wait.func = memcg_oom_wake_function; | |
1369 | owait.wait.private = current; | |
1370 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1371 | need_to_kill = true; |
867578cb KH |
1372 | /* At first, try to OOM lock hierarchy under mem.*/ |
1373 | mutex_lock(&memcg_oom_mutex); | |
1374 | locked = mem_cgroup_oom_lock(mem); | |
1375 | /* | |
1376 | * Even if signal_pending(), we can't quit charge() loop without | |
1377 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1378 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1379 | */ | |
3c11ecf4 KH |
1380 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
1381 | if (!locked || mem->oom_kill_disable) | |
1382 | need_to_kill = false; | |
1383 | if (locked) | |
9490ff27 | 1384 | mem_cgroup_oom_notify(mem); |
867578cb KH |
1385 | mutex_unlock(&memcg_oom_mutex); |
1386 | ||
3c11ecf4 KH |
1387 | if (need_to_kill) { |
1388 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
867578cb | 1389 | mem_cgroup_out_of_memory(mem, mask); |
3c11ecf4 | 1390 | } else { |
867578cb | 1391 | schedule(); |
dc98df5a | 1392 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb KH |
1393 | } |
1394 | mutex_lock(&memcg_oom_mutex); | |
1395 | mem_cgroup_oom_unlock(mem); | |
dc98df5a | 1396 | memcg_wakeup_oom(mem); |
867578cb KH |
1397 | mutex_unlock(&memcg_oom_mutex); |
1398 | ||
1399 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) | |
1400 | return false; | |
1401 | /* Give chance to dying process */ | |
1402 | schedule_timeout(1); | |
1403 | return true; | |
0b7f569e KH |
1404 | } |
1405 | ||
d69b042f BS |
1406 | /* |
1407 | * Currently used to update mapped file statistics, but the routine can be | |
1408 | * generalized to update other statistics as well. | |
1409 | */ | |
d8046582 | 1410 | void mem_cgroup_update_file_mapped(struct page *page, int val) |
d69b042f BS |
1411 | { |
1412 | struct mem_cgroup *mem; | |
d69b042f BS |
1413 | struct page_cgroup *pc; |
1414 | ||
d69b042f BS |
1415 | pc = lookup_page_cgroup(page); |
1416 | if (unlikely(!pc)) | |
1417 | return; | |
1418 | ||
1419 | lock_page_cgroup(pc); | |
1420 | mem = pc->mem_cgroup; | |
8725d541 | 1421 | if (!mem || !PageCgroupUsed(pc)) |
d69b042f BS |
1422 | goto done; |
1423 | ||
1424 | /* | |
c62b1a3b | 1425 | * Preemption is already disabled. We can use __this_cpu_xxx |
d69b042f | 1426 | */ |
8725d541 KH |
1427 | if (val > 0) { |
1428 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1429 | SetPageCgroupFileMapped(pc); | |
1430 | } else { | |
1431 | __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1432 | ClearPageCgroupFileMapped(pc); | |
1433 | } | |
d69b042f | 1434 | |
d69b042f BS |
1435 | done: |
1436 | unlock_page_cgroup(pc); | |
1437 | } | |
0b7f569e | 1438 | |
cdec2e42 KH |
1439 | /* |
1440 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1441 | * TODO: maybe necessary to use big numbers in big irons. | |
1442 | */ | |
1443 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1444 | struct memcg_stock_pcp { | |
1445 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1446 | int charge; | |
1447 | struct work_struct work; | |
1448 | }; | |
1449 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1450 | static atomic_t memcg_drain_count; | |
1451 | ||
1452 | /* | |
1453 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1454 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1455 | * cgroup which is not current target, returns false. This stock will be | |
1456 | * refilled. | |
1457 | */ | |
1458 | static bool consume_stock(struct mem_cgroup *mem) | |
1459 | { | |
1460 | struct memcg_stock_pcp *stock; | |
1461 | bool ret = true; | |
1462 | ||
1463 | stock = &get_cpu_var(memcg_stock); | |
1464 | if (mem == stock->cached && stock->charge) | |
1465 | stock->charge -= PAGE_SIZE; | |
1466 | else /* need to call res_counter_charge */ | |
1467 | ret = false; | |
1468 | put_cpu_var(memcg_stock); | |
1469 | return ret; | |
1470 | } | |
1471 | ||
1472 | /* | |
1473 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1474 | */ | |
1475 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1476 | { | |
1477 | struct mem_cgroup *old = stock->cached; | |
1478 | ||
1479 | if (stock->charge) { | |
1480 | res_counter_uncharge(&old->res, stock->charge); | |
1481 | if (do_swap_account) | |
1482 | res_counter_uncharge(&old->memsw, stock->charge); | |
1483 | } | |
1484 | stock->cached = NULL; | |
1485 | stock->charge = 0; | |
1486 | } | |
1487 | ||
1488 | /* | |
1489 | * This must be called under preempt disabled or must be called by | |
1490 | * a thread which is pinned to local cpu. | |
1491 | */ | |
1492 | static void drain_local_stock(struct work_struct *dummy) | |
1493 | { | |
1494 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1495 | drain_stock(stock); | |
1496 | } | |
1497 | ||
1498 | /* | |
1499 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 1500 | * This will be consumed by consume_stock() function, later. |
cdec2e42 KH |
1501 | */ |
1502 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1503 | { | |
1504 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1505 | ||
1506 | if (stock->cached != mem) { /* reset if necessary */ | |
1507 | drain_stock(stock); | |
1508 | stock->cached = mem; | |
1509 | } | |
1510 | stock->charge += val; | |
1511 | put_cpu_var(memcg_stock); | |
1512 | } | |
1513 | ||
1514 | /* | |
1515 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1516 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1517 | * expects some charges will be back to res_counter later but cannot wait for | |
1518 | * it. | |
1519 | */ | |
1520 | static void drain_all_stock_async(void) | |
1521 | { | |
1522 | int cpu; | |
1523 | /* This function is for scheduling "drain" in asynchronous way. | |
1524 | * The result of "drain" is not directly handled by callers. Then, | |
1525 | * if someone is calling drain, we don't have to call drain more. | |
1526 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1527 | * there is a race. We just do loose check here. | |
1528 | */ | |
1529 | if (atomic_read(&memcg_drain_count)) | |
1530 | return; | |
1531 | /* Notify other cpus that system-wide "drain" is running */ | |
1532 | atomic_inc(&memcg_drain_count); | |
1533 | get_online_cpus(); | |
1534 | for_each_online_cpu(cpu) { | |
1535 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1536 | schedule_work_on(cpu, &stock->work); | |
1537 | } | |
1538 | put_online_cpus(); | |
1539 | atomic_dec(&memcg_drain_count); | |
1540 | /* We don't wait for flush_work */ | |
1541 | } | |
1542 | ||
1543 | /* This is a synchronous drain interface. */ | |
1544 | static void drain_all_stock_sync(void) | |
1545 | { | |
1546 | /* called when force_empty is called */ | |
1547 | atomic_inc(&memcg_drain_count); | |
1548 | schedule_on_each_cpu(drain_local_stock); | |
1549 | atomic_dec(&memcg_drain_count); | |
1550 | } | |
1551 | ||
1552 | static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb, | |
1553 | unsigned long action, | |
1554 | void *hcpu) | |
1555 | { | |
1556 | int cpu = (unsigned long)hcpu; | |
1557 | struct memcg_stock_pcp *stock; | |
1558 | ||
1559 | if (action != CPU_DEAD) | |
1560 | return NOTIFY_OK; | |
1561 | stock = &per_cpu(memcg_stock, cpu); | |
1562 | drain_stock(stock); | |
1563 | return NOTIFY_OK; | |
1564 | } | |
1565 | ||
f817ed48 KH |
1566 | /* |
1567 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1568 | * oom-killer can be invoked. | |
8a9f3ccd | 1569 | */ |
f817ed48 | 1570 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
430e4863 | 1571 | gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom) |
8a9f3ccd | 1572 | { |
4e649152 | 1573 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 1574 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
4e649152 | 1575 | struct res_counter *fail_res; |
cdec2e42 | 1576 | int csize = CHARGE_SIZE; |
a636b327 | 1577 | |
867578cb KH |
1578 | /* |
1579 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
1580 | * in system level. So, allow to go ahead dying process in addition to | |
1581 | * MEMDIE process. | |
1582 | */ | |
1583 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
1584 | || fatal_signal_pending(current))) | |
1585 | goto bypass; | |
a636b327 | 1586 | |
8a9f3ccd | 1587 | /* |
3be91277 HD |
1588 | * We always charge the cgroup the mm_struct belongs to. |
1589 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
1590 | * thread group leader migrates. It's possible that mm is not |
1591 | * set, if so charge the init_mm (happens for pagecache usage). | |
1592 | */ | |
54595fe2 KH |
1593 | mem = *memcg; |
1594 | if (likely(!mem)) { | |
1595 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 1596 | *memcg = mem; |
e8589cc1 | 1597 | } else { |
7a81b88c | 1598 | css_get(&mem->css); |
e8589cc1 | 1599 | } |
54595fe2 KH |
1600 | if (unlikely(!mem)) |
1601 | return 0; | |
1602 | ||
46f7e602 | 1603 | VM_BUG_ON(css_is_removed(&mem->css)); |
cdec2e42 KH |
1604 | if (mem_cgroup_is_root(mem)) |
1605 | goto done; | |
8a9f3ccd | 1606 | |
8c7c6e34 | 1607 | while (1) { |
0c3e73e8 | 1608 | int ret = 0; |
75822b44 | 1609 | unsigned long flags = 0; |
7a81b88c | 1610 | |
cdec2e42 | 1611 | if (consume_stock(mem)) |
430e4863 | 1612 | goto done; |
cdec2e42 KH |
1613 | |
1614 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
8c7c6e34 KH |
1615 | if (likely(!ret)) { |
1616 | if (!do_swap_account) | |
1617 | break; | |
cdec2e42 | 1618 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); |
8c7c6e34 KH |
1619 | if (likely(!ret)) |
1620 | break; | |
1621 | /* mem+swap counter fails */ | |
cdec2e42 | 1622 | res_counter_uncharge(&mem->res, csize); |
75822b44 | 1623 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
6d61ef40 BS |
1624 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1625 | memsw); | |
1626 | } else | |
1627 | /* mem counter fails */ | |
1628 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1629 | res); | |
1630 | ||
cdec2e42 KH |
1631 | /* reduce request size and retry */ |
1632 | if (csize > PAGE_SIZE) { | |
1633 | csize = PAGE_SIZE; | |
1634 | continue; | |
1635 | } | |
3be91277 | 1636 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 1637 | goto nomem; |
e1a1cd59 | 1638 | |
4e416953 BS |
1639 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
1640 | gfp_mask, flags); | |
4d1c6273 DN |
1641 | if (ret) |
1642 | continue; | |
66e1707b BS |
1643 | |
1644 | /* | |
8869b8f6 HD |
1645 | * try_to_free_mem_cgroup_pages() might not give us a full |
1646 | * picture of reclaim. Some pages are reclaimed and might be | |
1647 | * moved to swap cache or just unmapped from the cgroup. | |
1648 | * Check the limit again to see if the reclaim reduced the | |
1649 | * current usage of the cgroup before giving up | |
8c7c6e34 | 1650 | * |
8869b8f6 | 1651 | */ |
b85a96c0 DN |
1652 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
1653 | continue; | |
3be91277 | 1654 | |
8033b97c DN |
1655 | /* try to avoid oom while someone is moving charge */ |
1656 | if (mc.moving_task && current != mc.moving_task) { | |
1657 | struct mem_cgroup *from, *to; | |
1658 | bool do_continue = false; | |
1659 | /* | |
1660 | * There is a small race that "from" or "to" can be | |
1661 | * freed by rmdir, so we use css_tryget(). | |
1662 | */ | |
8033b97c DN |
1663 | from = mc.from; |
1664 | to = mc.to; | |
1665 | if (from && css_tryget(&from->css)) { | |
1666 | if (mem_over_limit->use_hierarchy) | |
1667 | do_continue = css_is_ancestor( | |
1668 | &from->css, | |
1669 | &mem_over_limit->css); | |
1670 | else | |
1671 | do_continue = (from == mem_over_limit); | |
1672 | css_put(&from->css); | |
1673 | } | |
1674 | if (!do_continue && to && css_tryget(&to->css)) { | |
1675 | if (mem_over_limit->use_hierarchy) | |
1676 | do_continue = css_is_ancestor( | |
1677 | &to->css, | |
1678 | &mem_over_limit->css); | |
1679 | else | |
1680 | do_continue = (to == mem_over_limit); | |
1681 | css_put(&to->css); | |
1682 | } | |
8033b97c DN |
1683 | if (do_continue) { |
1684 | DEFINE_WAIT(wait); | |
1685 | prepare_to_wait(&mc.waitq, &wait, | |
1686 | TASK_INTERRUPTIBLE); | |
1687 | /* moving charge context might have finished. */ | |
1688 | if (mc.moving_task) | |
1689 | schedule(); | |
1690 | finish_wait(&mc.waitq, &wait); | |
1691 | continue; | |
1692 | } | |
1693 | } | |
1694 | ||
3be91277 | 1695 | if (!nr_retries--) { |
867578cb KH |
1696 | if (!oom) |
1697 | goto nomem; | |
1698 | if (mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) { | |
1699 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
1700 | continue; | |
a636b327 | 1701 | } |
867578cb KH |
1702 | /* When we reach here, current task is dying .*/ |
1703 | css_put(&mem->css); | |
1704 | goto bypass; | |
66e1707b | 1705 | } |
8a9f3ccd | 1706 | } |
cdec2e42 KH |
1707 | if (csize > PAGE_SIZE) |
1708 | refill_stock(mem, csize - PAGE_SIZE); | |
0c3e73e8 | 1709 | done: |
7a81b88c KH |
1710 | return 0; |
1711 | nomem: | |
1712 | css_put(&mem->css); | |
1713 | return -ENOMEM; | |
867578cb KH |
1714 | bypass: |
1715 | *memcg = NULL; | |
1716 | return 0; | |
7a81b88c | 1717 | } |
8a9f3ccd | 1718 | |
a3032a2c DN |
1719 | /* |
1720 | * Somemtimes we have to undo a charge we got by try_charge(). | |
1721 | * This function is for that and do uncharge, put css's refcnt. | |
1722 | * gotten by try_charge(). | |
1723 | */ | |
854ffa8d DN |
1724 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
1725 | unsigned long count) | |
a3032a2c DN |
1726 | { |
1727 | if (!mem_cgroup_is_root(mem)) { | |
854ffa8d | 1728 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
a3032a2c | 1729 | if (do_swap_account) |
854ffa8d DN |
1730 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
1731 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
1732 | WARN_ON_ONCE(count > INT_MAX); | |
1733 | __css_put(&mem->css, (int)count); | |
a3032a2c | 1734 | } |
854ffa8d DN |
1735 | /* we don't need css_put for root */ |
1736 | } | |
1737 | ||
1738 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem) | |
1739 | { | |
1740 | __mem_cgroup_cancel_charge(mem, 1); | |
a3032a2c DN |
1741 | } |
1742 | ||
a3b2d692 KH |
1743 | /* |
1744 | * A helper function to get mem_cgroup from ID. must be called under | |
1745 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
1746 | * it's concern. (dropping refcnt from swap can be called against removed | |
1747 | * memcg.) | |
1748 | */ | |
1749 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
1750 | { | |
1751 | struct cgroup_subsys_state *css; | |
1752 | ||
1753 | /* ID 0 is unused ID */ | |
1754 | if (!id) | |
1755 | return NULL; | |
1756 | css = css_lookup(&mem_cgroup_subsys, id); | |
1757 | if (!css) | |
1758 | return NULL; | |
1759 | return container_of(css, struct mem_cgroup, css); | |
1760 | } | |
1761 | ||
e42d9d5d | 1762 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 1763 | { |
e42d9d5d | 1764 | struct mem_cgroup *mem = NULL; |
3c776e64 | 1765 | struct page_cgroup *pc; |
a3b2d692 | 1766 | unsigned short id; |
b5a84319 KH |
1767 | swp_entry_t ent; |
1768 | ||
3c776e64 DN |
1769 | VM_BUG_ON(!PageLocked(page)); |
1770 | ||
3c776e64 | 1771 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 1772 | lock_page_cgroup(pc); |
a3b2d692 | 1773 | if (PageCgroupUsed(pc)) { |
3c776e64 | 1774 | mem = pc->mem_cgroup; |
a3b2d692 KH |
1775 | if (mem && !css_tryget(&mem->css)) |
1776 | mem = NULL; | |
e42d9d5d | 1777 | } else if (PageSwapCache(page)) { |
3c776e64 | 1778 | ent.val = page_private(page); |
a3b2d692 KH |
1779 | id = lookup_swap_cgroup(ent); |
1780 | rcu_read_lock(); | |
1781 | mem = mem_cgroup_lookup(id); | |
1782 | if (mem && !css_tryget(&mem->css)) | |
1783 | mem = NULL; | |
1784 | rcu_read_unlock(); | |
3c776e64 | 1785 | } |
c0bd3f63 | 1786 | unlock_page_cgroup(pc); |
b5a84319 KH |
1787 | return mem; |
1788 | } | |
1789 | ||
7a81b88c | 1790 | /* |
a5e924f5 | 1791 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
1792 | * USED state. If already USED, uncharge and return. |
1793 | */ | |
1794 | ||
1795 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
1796 | struct page_cgroup *pc, | |
1797 | enum charge_type ctype) | |
1798 | { | |
7a81b88c KH |
1799 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
1800 | if (!mem) | |
1801 | return; | |
52d4b9ac KH |
1802 | |
1803 | lock_page_cgroup(pc); | |
1804 | if (unlikely(PageCgroupUsed(pc))) { | |
1805 | unlock_page_cgroup(pc); | |
a3032a2c | 1806 | mem_cgroup_cancel_charge(mem); |
7a81b88c | 1807 | return; |
52d4b9ac | 1808 | } |
4b3bde4c | 1809 | |
8a9f3ccd | 1810 | pc->mem_cgroup = mem; |
261fb61a KH |
1811 | /* |
1812 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
1813 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
1814 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
1815 | * before USED bit, we need memory barrier here. | |
1816 | * See mem_cgroup_add_lru_list(), etc. | |
1817 | */ | |
08e552c6 | 1818 | smp_wmb(); |
4b3bde4c BS |
1819 | switch (ctype) { |
1820 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
1821 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
1822 | SetPageCgroupCache(pc); | |
1823 | SetPageCgroupUsed(pc); | |
1824 | break; | |
1825 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1826 | ClearPageCgroupCache(pc); | |
1827 | SetPageCgroupUsed(pc); | |
1828 | break; | |
1829 | default: | |
1830 | break; | |
1831 | } | |
3be91277 | 1832 | |
08e552c6 | 1833 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 1834 | |
52d4b9ac | 1835 | unlock_page_cgroup(pc); |
430e4863 KH |
1836 | /* |
1837 | * "charge_statistics" updated event counter. Then, check it. | |
1838 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
1839 | * if they exceeds softlimit. | |
1840 | */ | |
d2265e6f | 1841 | memcg_check_events(mem, pc->page); |
7a81b88c | 1842 | } |
66e1707b | 1843 | |
f817ed48 | 1844 | /** |
57f9fd7d | 1845 | * __mem_cgroup_move_account - move account of the page |
f817ed48 KH |
1846 | * @pc: page_cgroup of the page. |
1847 | * @from: mem_cgroup which the page is moved from. | |
1848 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 1849 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
1850 | * |
1851 | * The caller must confirm following. | |
08e552c6 | 1852 | * - page is not on LRU (isolate_page() is useful.) |
57f9fd7d | 1853 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
f817ed48 | 1854 | * |
854ffa8d DN |
1855 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
1856 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is | |
1857 | * true, this function does "uncharge" from old cgroup, but it doesn't if | |
1858 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 KH |
1859 | */ |
1860 | ||
57f9fd7d | 1861 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
854ffa8d | 1862 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
f817ed48 | 1863 | { |
f817ed48 | 1864 | VM_BUG_ON(from == to); |
08e552c6 | 1865 | VM_BUG_ON(PageLRU(pc->page)); |
57f9fd7d DN |
1866 | VM_BUG_ON(!PageCgroupLocked(pc)); |
1867 | VM_BUG_ON(!PageCgroupUsed(pc)); | |
1868 | VM_BUG_ON(pc->mem_cgroup != from); | |
f817ed48 | 1869 | |
8725d541 | 1870 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
1871 | /* Update mapped_file data for mem_cgroup */ |
1872 | preempt_disable(); | |
1873 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1874 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1875 | preempt_enable(); | |
d69b042f | 1876 | } |
854ffa8d DN |
1877 | mem_cgroup_charge_statistics(from, pc, false); |
1878 | if (uncharge) | |
1879 | /* This is not "cancel", but cancel_charge does all we need. */ | |
1880 | mem_cgroup_cancel_charge(from); | |
d69b042f | 1881 | |
854ffa8d | 1882 | /* caller should have done css_get */ |
08e552c6 KH |
1883 | pc->mem_cgroup = to; |
1884 | mem_cgroup_charge_statistics(to, pc, true); | |
88703267 KH |
1885 | /* |
1886 | * We charges against "to" which may not have any tasks. Then, "to" | |
1887 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe DN |
1888 | * this function is just force_empty() and move charge, so it's |
1889 | * garanteed that "to" is never removed. So, we don't check rmdir | |
1890 | * status here. | |
88703267 | 1891 | */ |
57f9fd7d DN |
1892 | } |
1893 | ||
1894 | /* | |
1895 | * check whether the @pc is valid for moving account and call | |
1896 | * __mem_cgroup_move_account() | |
1897 | */ | |
1898 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
854ffa8d | 1899 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
57f9fd7d DN |
1900 | { |
1901 | int ret = -EINVAL; | |
1902 | lock_page_cgroup(pc); | |
1903 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { | |
854ffa8d | 1904 | __mem_cgroup_move_account(pc, from, to, uncharge); |
57f9fd7d DN |
1905 | ret = 0; |
1906 | } | |
1907 | unlock_page_cgroup(pc); | |
d2265e6f KH |
1908 | /* |
1909 | * check events | |
1910 | */ | |
1911 | memcg_check_events(to, pc->page); | |
1912 | memcg_check_events(from, pc->page); | |
f817ed48 KH |
1913 | return ret; |
1914 | } | |
1915 | ||
1916 | /* | |
1917 | * move charges to its parent. | |
1918 | */ | |
1919 | ||
1920 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1921 | struct mem_cgroup *child, | |
1922 | gfp_t gfp_mask) | |
1923 | { | |
08e552c6 | 1924 | struct page *page = pc->page; |
f817ed48 KH |
1925 | struct cgroup *cg = child->css.cgroup; |
1926 | struct cgroup *pcg = cg->parent; | |
1927 | struct mem_cgroup *parent; | |
f817ed48 KH |
1928 | int ret; |
1929 | ||
1930 | /* Is ROOT ? */ | |
1931 | if (!pcg) | |
1932 | return -EINVAL; | |
1933 | ||
57f9fd7d DN |
1934 | ret = -EBUSY; |
1935 | if (!get_page_unless_zero(page)) | |
1936 | goto out; | |
1937 | if (isolate_lru_page(page)) | |
1938 | goto put; | |
08e552c6 | 1939 | |
f817ed48 | 1940 | parent = mem_cgroup_from_cont(pcg); |
430e4863 | 1941 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 1942 | if (ret || !parent) |
57f9fd7d | 1943 | goto put_back; |
f817ed48 | 1944 | |
854ffa8d DN |
1945 | ret = mem_cgroup_move_account(pc, child, parent, true); |
1946 | if (ret) | |
1947 | mem_cgroup_cancel_charge(parent); | |
57f9fd7d | 1948 | put_back: |
08e552c6 | 1949 | putback_lru_page(page); |
57f9fd7d | 1950 | put: |
40d58138 | 1951 | put_page(page); |
57f9fd7d | 1952 | out: |
f817ed48 KH |
1953 | return ret; |
1954 | } | |
1955 | ||
7a81b88c KH |
1956 | /* |
1957 | * Charge the memory controller for page usage. | |
1958 | * Return | |
1959 | * 0 if the charge was successful | |
1960 | * < 0 if the cgroup is over its limit | |
1961 | */ | |
1962 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1963 | gfp_t gfp_mask, enum charge_type ctype, | |
1964 | struct mem_cgroup *memcg) | |
1965 | { | |
1966 | struct mem_cgroup *mem; | |
1967 | struct page_cgroup *pc; | |
1968 | int ret; | |
1969 | ||
1970 | pc = lookup_page_cgroup(page); | |
1971 | /* can happen at boot */ | |
1972 | if (unlikely(!pc)) | |
1973 | return 0; | |
1974 | prefetchw(pc); | |
1975 | ||
1976 | mem = memcg; | |
430e4863 | 1977 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 1978 | if (ret || !mem) |
7a81b88c KH |
1979 | return ret; |
1980 | ||
1981 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1982 | return 0; |
8a9f3ccd BS |
1983 | } |
1984 | ||
7a81b88c KH |
1985 | int mem_cgroup_newpage_charge(struct page *page, |
1986 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1987 | { |
f8d66542 | 1988 | if (mem_cgroup_disabled()) |
cede86ac | 1989 | return 0; |
52d4b9ac KH |
1990 | if (PageCompound(page)) |
1991 | return 0; | |
69029cd5 KH |
1992 | /* |
1993 | * If already mapped, we don't have to account. | |
1994 | * If page cache, page->mapping has address_space. | |
1995 | * But page->mapping may have out-of-use anon_vma pointer, | |
1996 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1997 | * is NULL. | |
1998 | */ | |
1999 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2000 | return 0; | |
2001 | if (unlikely(!mm)) | |
2002 | mm = &init_mm; | |
217bc319 | 2003 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 2004 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
2005 | } |
2006 | ||
83aae4c7 DN |
2007 | static void |
2008 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2009 | enum charge_type ctype); | |
2010 | ||
e1a1cd59 BS |
2011 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2012 | gfp_t gfp_mask) | |
8697d331 | 2013 | { |
b5a84319 KH |
2014 | struct mem_cgroup *mem = NULL; |
2015 | int ret; | |
2016 | ||
f8d66542 | 2017 | if (mem_cgroup_disabled()) |
cede86ac | 2018 | return 0; |
52d4b9ac KH |
2019 | if (PageCompound(page)) |
2020 | return 0; | |
accf163e KH |
2021 | /* |
2022 | * Corner case handling. This is called from add_to_page_cache() | |
2023 | * in usual. But some FS (shmem) precharges this page before calling it | |
2024 | * and call add_to_page_cache() with GFP_NOWAIT. | |
2025 | * | |
2026 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
2027 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
2028 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
2029 | * And when the page is SwapCache, it should take swap information |
2030 | * into account. This is under lock_page() now. | |
accf163e KH |
2031 | */ |
2032 | if (!(gfp_mask & __GFP_WAIT)) { | |
2033 | struct page_cgroup *pc; | |
2034 | ||
52d4b9ac KH |
2035 | |
2036 | pc = lookup_page_cgroup(page); | |
2037 | if (!pc) | |
2038 | return 0; | |
2039 | lock_page_cgroup(pc); | |
2040 | if (PageCgroupUsed(pc)) { | |
2041 | unlock_page_cgroup(pc); | |
accf163e KH |
2042 | return 0; |
2043 | } | |
52d4b9ac | 2044 | unlock_page_cgroup(pc); |
accf163e KH |
2045 | } |
2046 | ||
b5a84319 | 2047 | if (unlikely(!mm && !mem)) |
8697d331 | 2048 | mm = &init_mm; |
accf163e | 2049 | |
c05555b5 KH |
2050 | if (page_is_file_cache(page)) |
2051 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 2052 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 | 2053 | |
83aae4c7 DN |
2054 | /* shmem */ |
2055 | if (PageSwapCache(page)) { | |
2056 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
2057 | if (!ret) | |
2058 | __mem_cgroup_commit_charge_swapin(page, mem, | |
2059 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
2060 | } else | |
2061 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
2062 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
b5a84319 | 2063 | |
b5a84319 | 2064 | return ret; |
e8589cc1 KH |
2065 | } |
2066 | ||
54595fe2 KH |
2067 | /* |
2068 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2069 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2070 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2071 | * "commit()" or removed by "cancel()" |
2072 | */ | |
8c7c6e34 KH |
2073 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2074 | struct page *page, | |
2075 | gfp_t mask, struct mem_cgroup **ptr) | |
2076 | { | |
2077 | struct mem_cgroup *mem; | |
54595fe2 | 2078 | int ret; |
8c7c6e34 | 2079 | |
f8d66542 | 2080 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2081 | return 0; |
2082 | ||
2083 | if (!do_swap_account) | |
2084 | goto charge_cur_mm; | |
8c7c6e34 KH |
2085 | /* |
2086 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2087 | * the pte, and even removed page from swap cache: in those cases |
2088 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2089 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2090 | */ |
2091 | if (!PageSwapCache(page)) | |
407f9c8b | 2092 | goto charge_cur_mm; |
e42d9d5d | 2093 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
2094 | if (!mem) |
2095 | goto charge_cur_mm; | |
8c7c6e34 | 2096 | *ptr = mem; |
430e4863 | 2097 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); |
54595fe2 KH |
2098 | /* drop extra refcnt from tryget */ |
2099 | css_put(&mem->css); | |
2100 | return ret; | |
8c7c6e34 KH |
2101 | charge_cur_mm: |
2102 | if (unlikely(!mm)) | |
2103 | mm = &init_mm; | |
430e4863 | 2104 | return __mem_cgroup_try_charge(mm, mask, ptr, true); |
8c7c6e34 KH |
2105 | } |
2106 | ||
83aae4c7 DN |
2107 | static void |
2108 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2109 | enum charge_type ctype) | |
7a81b88c KH |
2110 | { |
2111 | struct page_cgroup *pc; | |
2112 | ||
f8d66542 | 2113 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2114 | return; |
2115 | if (!ptr) | |
2116 | return; | |
88703267 | 2117 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 2118 | pc = lookup_page_cgroup(page); |
544122e5 | 2119 | mem_cgroup_lru_del_before_commit_swapcache(page); |
83aae4c7 | 2120 | __mem_cgroup_commit_charge(ptr, pc, ctype); |
544122e5 | 2121 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
2122 | /* |
2123 | * Now swap is on-memory. This means this page may be | |
2124 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2125 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2126 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2127 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2128 | */ |
03f3c433 | 2129 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2130 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2131 | unsigned short id; |
8c7c6e34 | 2132 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2133 | |
2134 | id = swap_cgroup_record(ent, 0); | |
2135 | rcu_read_lock(); | |
2136 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2137 | if (memcg) { |
a3b2d692 KH |
2138 | /* |
2139 | * This recorded memcg can be obsolete one. So, avoid | |
2140 | * calling css_tryget | |
2141 | */ | |
0c3e73e8 | 2142 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2143 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2144 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2145 | mem_cgroup_put(memcg); |
2146 | } | |
a3b2d692 | 2147 | rcu_read_unlock(); |
8c7c6e34 | 2148 | } |
88703267 KH |
2149 | /* |
2150 | * At swapin, we may charge account against cgroup which has no tasks. | |
2151 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2152 | * In that case, we need to call pre_destroy() again. check it here. | |
2153 | */ | |
2154 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2155 | } |
2156 | ||
83aae4c7 DN |
2157 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2158 | { | |
2159 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2160 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2161 | } | |
2162 | ||
7a81b88c KH |
2163 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2164 | { | |
f8d66542 | 2165 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2166 | return; |
2167 | if (!mem) | |
2168 | return; | |
a3032a2c | 2169 | mem_cgroup_cancel_charge(mem); |
7a81b88c KH |
2170 | } |
2171 | ||
569b846d KH |
2172 | static void |
2173 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype) | |
2174 | { | |
2175 | struct memcg_batch_info *batch = NULL; | |
2176 | bool uncharge_memsw = true; | |
2177 | /* If swapout, usage of swap doesn't decrease */ | |
2178 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2179 | uncharge_memsw = false; | |
569b846d KH |
2180 | |
2181 | batch = ¤t->memcg_batch; | |
2182 | /* | |
2183 | * In usual, we do css_get() when we remember memcg pointer. | |
2184 | * But in this case, we keep res->usage until end of a series of | |
2185 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2186 | */ | |
2187 | if (!batch->memcg) | |
2188 | batch->memcg = mem; | |
3c11ecf4 KH |
2189 | /* |
2190 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
2191 | * In those cases, all pages freed continously can be expected to be in | |
2192 | * the same cgroup and we have chance to coalesce uncharges. | |
2193 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2194 | * because we want to do uncharge as soon as possible. | |
2195 | */ | |
2196 | ||
2197 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2198 | goto direct_uncharge; | |
2199 | ||
569b846d KH |
2200 | /* |
2201 | * In typical case, batch->memcg == mem. This means we can | |
2202 | * merge a series of uncharges to an uncharge of res_counter. | |
2203 | * If not, we uncharge res_counter ony by one. | |
2204 | */ | |
2205 | if (batch->memcg != mem) | |
2206 | goto direct_uncharge; | |
2207 | /* remember freed charge and uncharge it later */ | |
2208 | batch->bytes += PAGE_SIZE; | |
2209 | if (uncharge_memsw) | |
2210 | batch->memsw_bytes += PAGE_SIZE; | |
2211 | return; | |
2212 | direct_uncharge: | |
2213 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
2214 | if (uncharge_memsw) | |
2215 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
3c11ecf4 KH |
2216 | if (unlikely(batch->memcg != mem)) |
2217 | memcg_oom_recover(mem); | |
569b846d KH |
2218 | return; |
2219 | } | |
7a81b88c | 2220 | |
8a9f3ccd | 2221 | /* |
69029cd5 | 2222 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2223 | */ |
8c7c6e34 | 2224 | static struct mem_cgroup * |
69029cd5 | 2225 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2226 | { |
8289546e | 2227 | struct page_cgroup *pc; |
8c7c6e34 | 2228 | struct mem_cgroup *mem = NULL; |
072c56c1 | 2229 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 2230 | |
f8d66542 | 2231 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2232 | return NULL; |
4077960e | 2233 | |
d13d1443 | 2234 | if (PageSwapCache(page)) |
8c7c6e34 | 2235 | return NULL; |
d13d1443 | 2236 | |
8697d331 | 2237 | /* |
3c541e14 | 2238 | * Check if our page_cgroup is valid |
8697d331 | 2239 | */ |
52d4b9ac KH |
2240 | pc = lookup_page_cgroup(page); |
2241 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2242 | return NULL; |
b9c565d5 | 2243 | |
52d4b9ac | 2244 | lock_page_cgroup(pc); |
d13d1443 | 2245 | |
8c7c6e34 KH |
2246 | mem = pc->mem_cgroup; |
2247 | ||
d13d1443 KH |
2248 | if (!PageCgroupUsed(pc)) |
2249 | goto unlock_out; | |
2250 | ||
2251 | switch (ctype) { | |
2252 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2253 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
d13d1443 KH |
2254 | if (page_mapped(page)) |
2255 | goto unlock_out; | |
2256 | break; | |
2257 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2258 | if (!PageAnon(page)) { /* Shared memory */ | |
2259 | if (page->mapping && !page_is_file_cache(page)) | |
2260 | goto unlock_out; | |
2261 | } else if (page_mapped(page)) /* Anon */ | |
2262 | goto unlock_out; | |
2263 | break; | |
2264 | default: | |
2265 | break; | |
52d4b9ac | 2266 | } |
d13d1443 | 2267 | |
569b846d KH |
2268 | if (!mem_cgroup_is_root(mem)) |
2269 | __do_uncharge(mem, ctype); | |
0c3e73e8 BS |
2270 | if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
2271 | mem_cgroup_swap_statistics(mem, true); | |
08e552c6 | 2272 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 2273 | |
52d4b9ac | 2274 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2275 | /* |
2276 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2277 | * freed from LRU. This is safe because uncharged page is expected not | |
2278 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2279 | * special functions. | |
2280 | */ | |
b9c565d5 | 2281 | |
69029cd5 | 2282 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 2283 | unlock_page_cgroup(pc); |
fb59e9f1 | 2284 | |
d2265e6f | 2285 | memcg_check_events(mem, page); |
a7fe942e KH |
2286 | /* at swapout, this memcg will be accessed to record to swap */ |
2287 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2288 | css_put(&mem->css); | |
6d12e2d8 | 2289 | |
8c7c6e34 | 2290 | return mem; |
d13d1443 KH |
2291 | |
2292 | unlock_out: | |
2293 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2294 | return NULL; |
3c541e14 BS |
2295 | } |
2296 | ||
69029cd5 KH |
2297 | void mem_cgroup_uncharge_page(struct page *page) |
2298 | { | |
52d4b9ac KH |
2299 | /* early check. */ |
2300 | if (page_mapped(page)) | |
2301 | return; | |
2302 | if (page->mapping && !PageAnon(page)) | |
2303 | return; | |
69029cd5 KH |
2304 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2305 | } | |
2306 | ||
2307 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2308 | { | |
2309 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2310 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2311 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2312 | } | |
2313 | ||
569b846d KH |
2314 | /* |
2315 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2316 | * In that cases, pages are freed continuously and we can expect pages | |
2317 | * are in the same memcg. All these calls itself limits the number of | |
2318 | * pages freed at once, then uncharge_start/end() is called properly. | |
2319 | * This may be called prural(2) times in a context, | |
2320 | */ | |
2321 | ||
2322 | void mem_cgroup_uncharge_start(void) | |
2323 | { | |
2324 | current->memcg_batch.do_batch++; | |
2325 | /* We can do nest. */ | |
2326 | if (current->memcg_batch.do_batch == 1) { | |
2327 | current->memcg_batch.memcg = NULL; | |
2328 | current->memcg_batch.bytes = 0; | |
2329 | current->memcg_batch.memsw_bytes = 0; | |
2330 | } | |
2331 | } | |
2332 | ||
2333 | void mem_cgroup_uncharge_end(void) | |
2334 | { | |
2335 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2336 | ||
2337 | if (!batch->do_batch) | |
2338 | return; | |
2339 | ||
2340 | batch->do_batch--; | |
2341 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2342 | return; | |
2343 | ||
2344 | if (!batch->memcg) | |
2345 | return; | |
2346 | /* | |
2347 | * This "batch->memcg" is valid without any css_get/put etc... | |
2348 | * bacause we hide charges behind us. | |
2349 | */ | |
2350 | if (batch->bytes) | |
2351 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2352 | if (batch->memsw_bytes) | |
2353 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
3c11ecf4 | 2354 | memcg_oom_recover(batch->memcg); |
569b846d KH |
2355 | /* forget this pointer (for sanity check) */ |
2356 | batch->memcg = NULL; | |
2357 | } | |
2358 | ||
e767e056 | 2359 | #ifdef CONFIG_SWAP |
8c7c6e34 | 2360 | /* |
e767e056 | 2361 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
2362 | * memcg information is recorded to swap_cgroup of "ent" |
2363 | */ | |
8a9478ca KH |
2364 | void |
2365 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
2366 | { |
2367 | struct mem_cgroup *memcg; | |
8a9478ca KH |
2368 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
2369 | ||
2370 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2371 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2372 | ||
2373 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 2374 | |
8c7c6e34 | 2375 | /* record memcg information */ |
8a9478ca | 2376 | if (do_swap_account && swapout && memcg) { |
a3b2d692 | 2377 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 KH |
2378 | mem_cgroup_get(memcg); |
2379 | } | |
8a9478ca | 2380 | if (swapout && memcg) |
a7fe942e | 2381 | css_put(&memcg->css); |
8c7c6e34 | 2382 | } |
e767e056 | 2383 | #endif |
8c7c6e34 KH |
2384 | |
2385 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2386 | /* | |
2387 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2388 | * uncharge "memsw" account. | |
2389 | */ | |
2390 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 2391 | { |
8c7c6e34 | 2392 | struct mem_cgroup *memcg; |
a3b2d692 | 2393 | unsigned short id; |
8c7c6e34 KH |
2394 | |
2395 | if (!do_swap_account) | |
2396 | return; | |
2397 | ||
a3b2d692 KH |
2398 | id = swap_cgroup_record(ent, 0); |
2399 | rcu_read_lock(); | |
2400 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2401 | if (memcg) { |
a3b2d692 KH |
2402 | /* |
2403 | * We uncharge this because swap is freed. | |
2404 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2405 | */ | |
0c3e73e8 | 2406 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2407 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2408 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2409 | mem_cgroup_put(memcg); |
2410 | } | |
a3b2d692 | 2411 | rcu_read_unlock(); |
d13d1443 | 2412 | } |
02491447 DN |
2413 | |
2414 | /** | |
2415 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2416 | * @entry: swap entry to be moved | |
2417 | * @from: mem_cgroup which the entry is moved from | |
2418 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 2419 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
2420 | * |
2421 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2422 | * as the mem_cgroup's id of @from. | |
2423 | * | |
2424 | * Returns 0 on success, -EINVAL on failure. | |
2425 | * | |
2426 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
2427 | * both res and memsw, and called css_get(). | |
2428 | */ | |
2429 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2430 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2431 | { |
2432 | unsigned short old_id, new_id; | |
2433 | ||
2434 | old_id = css_id(&from->css); | |
2435 | new_id = css_id(&to->css); | |
2436 | ||
2437 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2438 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2439 | mem_cgroup_swap_statistics(to, true); |
02491447 | 2440 | /* |
483c30b5 DN |
2441 | * This function is only called from task migration context now. |
2442 | * It postpones res_counter and refcount handling till the end | |
2443 | * of task migration(mem_cgroup_clear_mc()) for performance | |
2444 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
2445 | * because if the process that has been moved to @to does | |
2446 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 2447 | */ |
02491447 | 2448 | mem_cgroup_get(to); |
483c30b5 DN |
2449 | if (need_fixup) { |
2450 | if (!mem_cgroup_is_root(from)) | |
2451 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
2452 | mem_cgroup_put(from); | |
2453 | /* | |
2454 | * we charged both to->res and to->memsw, so we should | |
2455 | * uncharge to->res. | |
2456 | */ | |
2457 | if (!mem_cgroup_is_root(to)) | |
2458 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
2459 | css_put(&to->css); | |
2460 | } | |
02491447 DN |
2461 | return 0; |
2462 | } | |
2463 | return -EINVAL; | |
2464 | } | |
2465 | #else | |
2466 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2467 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2468 | { |
2469 | return -EINVAL; | |
2470 | } | |
8c7c6e34 | 2471 | #endif |
d13d1443 | 2472 | |
ae41be37 | 2473 | /* |
01b1ae63 KH |
2474 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
2475 | * page belongs to. | |
ae41be37 | 2476 | */ |
01b1ae63 | 2477 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
2478 | { |
2479 | struct page_cgroup *pc; | |
e8589cc1 | 2480 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 2481 | int ret = 0; |
8869b8f6 | 2482 | |
f8d66542 | 2483 | if (mem_cgroup_disabled()) |
4077960e BS |
2484 | return 0; |
2485 | ||
52d4b9ac KH |
2486 | pc = lookup_page_cgroup(page); |
2487 | lock_page_cgroup(pc); | |
2488 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
2489 | mem = pc->mem_cgroup; |
2490 | css_get(&mem->css); | |
e8589cc1 | 2491 | } |
52d4b9ac | 2492 | unlock_page_cgroup(pc); |
01b1ae63 | 2493 | |
93d5c9be | 2494 | *ptr = mem; |
e8589cc1 | 2495 | if (mem) { |
93d5c9be | 2496 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false); |
e8589cc1 KH |
2497 | css_put(&mem->css); |
2498 | } | |
2499 | return ret; | |
ae41be37 | 2500 | } |
8869b8f6 | 2501 | |
69029cd5 | 2502 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
2503 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
2504 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 2505 | { |
01b1ae63 KH |
2506 | struct page *target, *unused; |
2507 | struct page_cgroup *pc; | |
2508 | enum charge_type ctype; | |
2509 | ||
2510 | if (!mem) | |
2511 | return; | |
88703267 | 2512 | cgroup_exclude_rmdir(&mem->css); |
01b1ae63 KH |
2513 | /* at migration success, oldpage->mapping is NULL. */ |
2514 | if (oldpage->mapping) { | |
2515 | target = oldpage; | |
2516 | unused = NULL; | |
2517 | } else { | |
2518 | target = newpage; | |
2519 | unused = oldpage; | |
2520 | } | |
2521 | ||
2522 | if (PageAnon(target)) | |
2523 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2524 | else if (page_is_file_cache(target)) | |
2525 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2526 | else | |
2527 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
2528 | ||
2529 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 2530 | if (unused) |
01b1ae63 KH |
2531 | __mem_cgroup_uncharge_common(unused, ctype); |
2532 | ||
2533 | pc = lookup_page_cgroup(target); | |
69029cd5 | 2534 | /* |
01b1ae63 KH |
2535 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
2536 | * So, double-counting is effectively avoided. | |
2537 | */ | |
2538 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
2539 | ||
2540 | /* | |
2541 | * Both of oldpage and newpage are still under lock_page(). | |
2542 | * Then, we don't have to care about race in radix-tree. | |
2543 | * But we have to be careful that this page is unmapped or not. | |
2544 | * | |
2545 | * There is a case for !page_mapped(). At the start of | |
2546 | * migration, oldpage was mapped. But now, it's zapped. | |
2547 | * But we know *target* page is not freed/reused under us. | |
2548 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 2549 | */ |
01b1ae63 KH |
2550 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
2551 | mem_cgroup_uncharge_page(target); | |
88703267 KH |
2552 | /* |
2553 | * At migration, we may charge account against cgroup which has no tasks | |
2554 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2555 | * In that case, we need to call pre_destroy() again. check it here. | |
2556 | */ | |
2557 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 2558 | } |
78fb7466 | 2559 | |
c9b0ed51 | 2560 | /* |
ae3abae6 DN |
2561 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
2562 | * Calling hierarchical_reclaim is not enough because we should update | |
2563 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
2564 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
2565 | * not from the memcg which this page would be charged to. | |
2566 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 2567 | */ |
ae3abae6 | 2568 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
2569 | struct mm_struct *mm, |
2570 | gfp_t gfp_mask) | |
c9b0ed51 | 2571 | { |
b5a84319 | 2572 | struct mem_cgroup *mem = NULL; |
ae3abae6 | 2573 | int ret; |
c9b0ed51 | 2574 | |
f8d66542 | 2575 | if (mem_cgroup_disabled()) |
cede86ac | 2576 | return 0; |
c9b0ed51 | 2577 | |
ae3abae6 DN |
2578 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2579 | if (!ret) | |
2580 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 2581 | |
ae3abae6 | 2582 | return ret; |
c9b0ed51 KH |
2583 | } |
2584 | ||
8c7c6e34 KH |
2585 | static DEFINE_MUTEX(set_limit_mutex); |
2586 | ||
d38d2a75 | 2587 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 2588 | unsigned long long val) |
628f4235 | 2589 | { |
81d39c20 | 2590 | int retry_count; |
3c11ecf4 | 2591 | u64 memswlimit, memlimit; |
628f4235 | 2592 | int ret = 0; |
81d39c20 KH |
2593 | int children = mem_cgroup_count_children(memcg); |
2594 | u64 curusage, oldusage; | |
3c11ecf4 | 2595 | int enlarge; |
81d39c20 KH |
2596 | |
2597 | /* | |
2598 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2599 | * is depends on callers. We set our retry-count to be function | |
2600 | * of # of children which we should visit in this loop. | |
2601 | */ | |
2602 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
2603 | ||
2604 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 2605 | |
3c11ecf4 | 2606 | enlarge = 0; |
8c7c6e34 | 2607 | while (retry_count) { |
628f4235 KH |
2608 | if (signal_pending(current)) { |
2609 | ret = -EINTR; | |
2610 | break; | |
2611 | } | |
8c7c6e34 KH |
2612 | /* |
2613 | * Rather than hide all in some function, I do this in | |
2614 | * open coded manner. You see what this really does. | |
2615 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2616 | */ | |
2617 | mutex_lock(&set_limit_mutex); | |
2618 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2619 | if (memswlimit < val) { | |
2620 | ret = -EINVAL; | |
2621 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
2622 | break; |
2623 | } | |
3c11ecf4 KH |
2624 | |
2625 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2626 | if (memlimit < val) | |
2627 | enlarge = 1; | |
2628 | ||
8c7c6e34 | 2629 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
2630 | if (!ret) { |
2631 | if (memswlimit == val) | |
2632 | memcg->memsw_is_minimum = true; | |
2633 | else | |
2634 | memcg->memsw_is_minimum = false; | |
2635 | } | |
8c7c6e34 KH |
2636 | mutex_unlock(&set_limit_mutex); |
2637 | ||
2638 | if (!ret) | |
2639 | break; | |
2640 | ||
aa20d489 | 2641 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
4e416953 | 2642 | MEM_CGROUP_RECLAIM_SHRINK); |
81d39c20 KH |
2643 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
2644 | /* Usage is reduced ? */ | |
2645 | if (curusage >= oldusage) | |
2646 | retry_count--; | |
2647 | else | |
2648 | oldusage = curusage; | |
8c7c6e34 | 2649 | } |
3c11ecf4 KH |
2650 | if (!ret && enlarge) |
2651 | memcg_oom_recover(memcg); | |
14797e23 | 2652 | |
8c7c6e34 KH |
2653 | return ret; |
2654 | } | |
2655 | ||
338c8431 LZ |
2656 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
2657 | unsigned long long val) | |
8c7c6e34 | 2658 | { |
81d39c20 | 2659 | int retry_count; |
3c11ecf4 | 2660 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
2661 | int children = mem_cgroup_count_children(memcg); |
2662 | int ret = -EBUSY; | |
3c11ecf4 | 2663 | int enlarge = 0; |
8c7c6e34 | 2664 | |
81d39c20 KH |
2665 | /* see mem_cgroup_resize_res_limit */ |
2666 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
2667 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
2668 | while (retry_count) { |
2669 | if (signal_pending(current)) { | |
2670 | ret = -EINTR; | |
2671 | break; | |
2672 | } | |
2673 | /* | |
2674 | * Rather than hide all in some function, I do this in | |
2675 | * open coded manner. You see what this really does. | |
2676 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2677 | */ | |
2678 | mutex_lock(&set_limit_mutex); | |
2679 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2680 | if (memlimit > val) { | |
2681 | ret = -EINVAL; | |
2682 | mutex_unlock(&set_limit_mutex); | |
2683 | break; | |
2684 | } | |
3c11ecf4 KH |
2685 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
2686 | if (memswlimit < val) | |
2687 | enlarge = 1; | |
8c7c6e34 | 2688 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
2689 | if (!ret) { |
2690 | if (memlimit == val) | |
2691 | memcg->memsw_is_minimum = true; | |
2692 | else | |
2693 | memcg->memsw_is_minimum = false; | |
2694 | } | |
8c7c6e34 KH |
2695 | mutex_unlock(&set_limit_mutex); |
2696 | ||
2697 | if (!ret) | |
2698 | break; | |
2699 | ||
4e416953 | 2700 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 BS |
2701 | MEM_CGROUP_RECLAIM_NOSWAP | |
2702 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 2703 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 2704 | /* Usage is reduced ? */ |
8c7c6e34 | 2705 | if (curusage >= oldusage) |
628f4235 | 2706 | retry_count--; |
81d39c20 KH |
2707 | else |
2708 | oldusage = curusage; | |
628f4235 | 2709 | } |
3c11ecf4 KH |
2710 | if (!ret && enlarge) |
2711 | memcg_oom_recover(memcg); | |
628f4235 KH |
2712 | return ret; |
2713 | } | |
2714 | ||
4e416953 BS |
2715 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2716 | gfp_t gfp_mask, int nid, | |
2717 | int zid) | |
2718 | { | |
2719 | unsigned long nr_reclaimed = 0; | |
2720 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2721 | unsigned long reclaimed; | |
2722 | int loop = 0; | |
2723 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 2724 | unsigned long long excess; |
4e416953 BS |
2725 | |
2726 | if (order > 0) | |
2727 | return 0; | |
2728 | ||
2729 | mctz = soft_limit_tree_node_zone(nid, zid); | |
2730 | /* | |
2731 | * This loop can run a while, specially if mem_cgroup's continuously | |
2732 | * keep exceeding their soft limit and putting the system under | |
2733 | * pressure | |
2734 | */ | |
2735 | do { | |
2736 | if (next_mz) | |
2737 | mz = next_mz; | |
2738 | else | |
2739 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2740 | if (!mz) | |
2741 | break; | |
2742 | ||
2743 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
2744 | gfp_mask, | |
2745 | MEM_CGROUP_RECLAIM_SOFT); | |
2746 | nr_reclaimed += reclaimed; | |
2747 | spin_lock(&mctz->lock); | |
2748 | ||
2749 | /* | |
2750 | * If we failed to reclaim anything from this memory cgroup | |
2751 | * it is time to move on to the next cgroup | |
2752 | */ | |
2753 | next_mz = NULL; | |
2754 | if (!reclaimed) { | |
2755 | do { | |
2756 | /* | |
2757 | * Loop until we find yet another one. | |
2758 | * | |
2759 | * By the time we get the soft_limit lock | |
2760 | * again, someone might have aded the | |
2761 | * group back on the RB tree. Iterate to | |
2762 | * make sure we get a different mem. | |
2763 | * mem_cgroup_largest_soft_limit_node returns | |
2764 | * NULL if no other cgroup is present on | |
2765 | * the tree | |
2766 | */ | |
2767 | next_mz = | |
2768 | __mem_cgroup_largest_soft_limit_node(mctz); | |
2769 | if (next_mz == mz) { | |
2770 | css_put(&next_mz->mem->css); | |
2771 | next_mz = NULL; | |
2772 | } else /* next_mz == NULL or other memcg */ | |
2773 | break; | |
2774 | } while (1); | |
2775 | } | |
4e416953 | 2776 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 2777 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
2778 | /* |
2779 | * One school of thought says that we should not add | |
2780 | * back the node to the tree if reclaim returns 0. | |
2781 | * But our reclaim could return 0, simply because due | |
2782 | * to priority we are exposing a smaller subset of | |
2783 | * memory to reclaim from. Consider this as a longer | |
2784 | * term TODO. | |
2785 | */ | |
ef8745c1 KH |
2786 | /* If excess == 0, no tree ops */ |
2787 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
2788 | spin_unlock(&mctz->lock); |
2789 | css_put(&mz->mem->css); | |
2790 | loop++; | |
2791 | /* | |
2792 | * Could not reclaim anything and there are no more | |
2793 | * mem cgroups to try or we seem to be looping without | |
2794 | * reclaiming anything. | |
2795 | */ | |
2796 | if (!nr_reclaimed && | |
2797 | (next_mz == NULL || | |
2798 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2799 | break; | |
2800 | } while (!nr_reclaimed); | |
2801 | if (next_mz) | |
2802 | css_put(&next_mz->mem->css); | |
2803 | return nr_reclaimed; | |
2804 | } | |
2805 | ||
cc847582 KH |
2806 | /* |
2807 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
2808 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
2809 | */ | |
f817ed48 | 2810 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 2811 | int node, int zid, enum lru_list lru) |
cc847582 | 2812 | { |
08e552c6 KH |
2813 | struct zone *zone; |
2814 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 2815 | struct page_cgroup *pc, *busy; |
08e552c6 | 2816 | unsigned long flags, loop; |
072c56c1 | 2817 | struct list_head *list; |
f817ed48 | 2818 | int ret = 0; |
072c56c1 | 2819 | |
08e552c6 KH |
2820 | zone = &NODE_DATA(node)->node_zones[zid]; |
2821 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 2822 | list = &mz->lists[lru]; |
cc847582 | 2823 | |
f817ed48 KH |
2824 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
2825 | /* give some margin against EBUSY etc...*/ | |
2826 | loop += 256; | |
2827 | busy = NULL; | |
2828 | while (loop--) { | |
2829 | ret = 0; | |
08e552c6 | 2830 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 2831 | if (list_empty(list)) { |
08e552c6 | 2832 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 2833 | break; |
f817ed48 KH |
2834 | } |
2835 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
2836 | if (busy == pc) { | |
2837 | list_move(&pc->lru, list); | |
648bcc77 | 2838 | busy = NULL; |
08e552c6 | 2839 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
2840 | continue; |
2841 | } | |
08e552c6 | 2842 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 2843 | |
2c26fdd7 | 2844 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 2845 | if (ret == -ENOMEM) |
52d4b9ac | 2846 | break; |
f817ed48 KH |
2847 | |
2848 | if (ret == -EBUSY || ret == -EINVAL) { | |
2849 | /* found lock contention or "pc" is obsolete. */ | |
2850 | busy = pc; | |
2851 | cond_resched(); | |
2852 | } else | |
2853 | busy = NULL; | |
cc847582 | 2854 | } |
08e552c6 | 2855 | |
f817ed48 KH |
2856 | if (!ret && !list_empty(list)) |
2857 | return -EBUSY; | |
2858 | return ret; | |
cc847582 KH |
2859 | } |
2860 | ||
2861 | /* | |
2862 | * make mem_cgroup's charge to be 0 if there is no task. | |
2863 | * This enables deleting this mem_cgroup. | |
2864 | */ | |
c1e862c1 | 2865 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 2866 | { |
f817ed48 KH |
2867 | int ret; |
2868 | int node, zid, shrink; | |
2869 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 2870 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 2871 | |
cc847582 | 2872 | css_get(&mem->css); |
f817ed48 KH |
2873 | |
2874 | shrink = 0; | |
c1e862c1 KH |
2875 | /* should free all ? */ |
2876 | if (free_all) | |
2877 | goto try_to_free; | |
f817ed48 | 2878 | move_account: |
fce66477 | 2879 | do { |
f817ed48 | 2880 | ret = -EBUSY; |
c1e862c1 KH |
2881 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
2882 | goto out; | |
2883 | ret = -EINTR; | |
2884 | if (signal_pending(current)) | |
cc847582 | 2885 | goto out; |
52d4b9ac KH |
2886 | /* This is for making all *used* pages to be on LRU. */ |
2887 | lru_add_drain_all(); | |
cdec2e42 | 2888 | drain_all_stock_sync(); |
f817ed48 | 2889 | ret = 0; |
299b4eaa | 2890 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 2891 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 2892 | enum lru_list l; |
f817ed48 KH |
2893 | for_each_lru(l) { |
2894 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 2895 | node, zid, l); |
f817ed48 KH |
2896 | if (ret) |
2897 | break; | |
2898 | } | |
1ecaab2b | 2899 | } |
f817ed48 KH |
2900 | if (ret) |
2901 | break; | |
2902 | } | |
3c11ecf4 | 2903 | memcg_oom_recover(mem); |
f817ed48 KH |
2904 | /* it seems parent cgroup doesn't have enough mem */ |
2905 | if (ret == -ENOMEM) | |
2906 | goto try_to_free; | |
52d4b9ac | 2907 | cond_resched(); |
fce66477 DN |
2908 | /* "ret" should also be checked to ensure all lists are empty. */ |
2909 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
2910 | out: |
2911 | css_put(&mem->css); | |
2912 | return ret; | |
f817ed48 KH |
2913 | |
2914 | try_to_free: | |
c1e862c1 KH |
2915 | /* returns EBUSY if there is a task or if we come here twice. */ |
2916 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
2917 | ret = -EBUSY; |
2918 | goto out; | |
2919 | } | |
c1e862c1 KH |
2920 | /* we call try-to-free pages for make this cgroup empty */ |
2921 | lru_add_drain_all(); | |
f817ed48 KH |
2922 | /* try to free all pages in this cgroup */ |
2923 | shrink = 1; | |
2924 | while (nr_retries && mem->res.usage > 0) { | |
2925 | int progress; | |
c1e862c1 KH |
2926 | |
2927 | if (signal_pending(current)) { | |
2928 | ret = -EINTR; | |
2929 | goto out; | |
2930 | } | |
a7885eb8 KM |
2931 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
2932 | false, get_swappiness(mem)); | |
c1e862c1 | 2933 | if (!progress) { |
f817ed48 | 2934 | nr_retries--; |
c1e862c1 | 2935 | /* maybe some writeback is necessary */ |
8aa7e847 | 2936 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2937 | } |
f817ed48 KH |
2938 | |
2939 | } | |
08e552c6 | 2940 | lru_add_drain(); |
f817ed48 | 2941 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 2942 | goto move_account; |
cc847582 KH |
2943 | } |
2944 | ||
c1e862c1 KH |
2945 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
2946 | { | |
2947 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
2948 | } | |
2949 | ||
2950 | ||
18f59ea7 BS |
2951 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
2952 | { | |
2953 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
2954 | } | |
2955 | ||
2956 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
2957 | u64 val) | |
2958 | { | |
2959 | int retval = 0; | |
2960 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
2961 | struct cgroup *parent = cont->parent; | |
2962 | struct mem_cgroup *parent_mem = NULL; | |
2963 | ||
2964 | if (parent) | |
2965 | parent_mem = mem_cgroup_from_cont(parent); | |
2966 | ||
2967 | cgroup_lock(); | |
2968 | /* | |
af901ca1 | 2969 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2970 | * in the child subtrees. If it is unset, then the change can |
2971 | * occur, provided the current cgroup has no children. | |
2972 | * | |
2973 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2974 | * set if there are no children. | |
2975 | */ | |
2976 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
2977 | (val == 1 || val == 0)) { | |
2978 | if (list_empty(&cont->children)) | |
2979 | mem->use_hierarchy = val; | |
2980 | else | |
2981 | retval = -EBUSY; | |
2982 | } else | |
2983 | retval = -EINVAL; | |
2984 | cgroup_unlock(); | |
2985 | ||
2986 | return retval; | |
2987 | } | |
2988 | ||
0c3e73e8 BS |
2989 | struct mem_cgroup_idx_data { |
2990 | s64 val; | |
2991 | enum mem_cgroup_stat_index idx; | |
2992 | }; | |
2993 | ||
2994 | static int | |
2995 | mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) | |
2996 | { | |
2997 | struct mem_cgroup_idx_data *d = data; | |
c62b1a3b | 2998 | d->val += mem_cgroup_read_stat(mem, d->idx); |
0c3e73e8 BS |
2999 | return 0; |
3000 | } | |
3001 | ||
3002 | static void | |
3003 | mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, | |
3004 | enum mem_cgroup_stat_index idx, s64 *val) | |
3005 | { | |
3006 | struct mem_cgroup_idx_data d; | |
3007 | d.idx = idx; | |
3008 | d.val = 0; | |
3009 | mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat); | |
3010 | *val = d.val; | |
3011 | } | |
3012 | ||
104f3928 KS |
3013 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
3014 | { | |
3015 | u64 idx_val, val; | |
3016 | ||
3017 | if (!mem_cgroup_is_root(mem)) { | |
3018 | if (!swap) | |
3019 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
3020 | else | |
3021 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
3022 | } | |
3023 | ||
3024 | mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val); | |
3025 | val = idx_val; | |
3026 | mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val); | |
3027 | val += idx_val; | |
3028 | ||
3029 | if (swap) { | |
3030 | mem_cgroup_get_recursive_idx_stat(mem, | |
3031 | MEM_CGROUP_STAT_SWAPOUT, &idx_val); | |
3032 | val += idx_val; | |
3033 | } | |
3034 | ||
3035 | return val << PAGE_SHIFT; | |
3036 | } | |
3037 | ||
2c3daa72 | 3038 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3039 | { |
8c7c6e34 | 3040 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 3041 | u64 val; |
8c7c6e34 KH |
3042 | int type, name; |
3043 | ||
3044 | type = MEMFILE_TYPE(cft->private); | |
3045 | name = MEMFILE_ATTR(cft->private); | |
3046 | switch (type) { | |
3047 | case _MEM: | |
104f3928 KS |
3048 | if (name == RES_USAGE) |
3049 | val = mem_cgroup_usage(mem, false); | |
3050 | else | |
0c3e73e8 | 3051 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
3052 | break; |
3053 | case _MEMSWAP: | |
104f3928 KS |
3054 | if (name == RES_USAGE) |
3055 | val = mem_cgroup_usage(mem, true); | |
3056 | else | |
0c3e73e8 | 3057 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
3058 | break; |
3059 | default: | |
3060 | BUG(); | |
3061 | break; | |
3062 | } | |
3063 | return val; | |
8cdea7c0 | 3064 | } |
628f4235 KH |
3065 | /* |
3066 | * The user of this function is... | |
3067 | * RES_LIMIT. | |
3068 | */ | |
856c13aa PM |
3069 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3070 | const char *buffer) | |
8cdea7c0 | 3071 | { |
628f4235 | 3072 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3073 | int type, name; |
628f4235 KH |
3074 | unsigned long long val; |
3075 | int ret; | |
3076 | ||
8c7c6e34 KH |
3077 | type = MEMFILE_TYPE(cft->private); |
3078 | name = MEMFILE_ATTR(cft->private); | |
3079 | switch (name) { | |
628f4235 | 3080 | case RES_LIMIT: |
4b3bde4c BS |
3081 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3082 | ret = -EINVAL; | |
3083 | break; | |
3084 | } | |
628f4235 KH |
3085 | /* This function does all necessary parse...reuse it */ |
3086 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3087 | if (ret) |
3088 | break; | |
3089 | if (type == _MEM) | |
628f4235 | 3090 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3091 | else |
3092 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3093 | break; |
296c81d8 BS |
3094 | case RES_SOFT_LIMIT: |
3095 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3096 | if (ret) | |
3097 | break; | |
3098 | /* | |
3099 | * For memsw, soft limits are hard to implement in terms | |
3100 | * of semantics, for now, we support soft limits for | |
3101 | * control without swap | |
3102 | */ | |
3103 | if (type == _MEM) | |
3104 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3105 | else | |
3106 | ret = -EINVAL; | |
3107 | break; | |
628f4235 KH |
3108 | default: |
3109 | ret = -EINVAL; /* should be BUG() ? */ | |
3110 | break; | |
3111 | } | |
3112 | return ret; | |
8cdea7c0 BS |
3113 | } |
3114 | ||
fee7b548 KH |
3115 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3116 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3117 | { | |
3118 | struct cgroup *cgroup; | |
3119 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3120 | ||
3121 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3122 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3123 | cgroup = memcg->css.cgroup; | |
3124 | if (!memcg->use_hierarchy) | |
3125 | goto out; | |
3126 | ||
3127 | while (cgroup->parent) { | |
3128 | cgroup = cgroup->parent; | |
3129 | memcg = mem_cgroup_from_cont(cgroup); | |
3130 | if (!memcg->use_hierarchy) | |
3131 | break; | |
3132 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3133 | min_limit = min(min_limit, tmp); | |
3134 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3135 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3136 | } | |
3137 | out: | |
3138 | *mem_limit = min_limit; | |
3139 | *memsw_limit = min_memsw_limit; | |
3140 | return; | |
3141 | } | |
3142 | ||
29f2a4da | 3143 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
3144 | { |
3145 | struct mem_cgroup *mem; | |
8c7c6e34 | 3146 | int type, name; |
c84872e1 PE |
3147 | |
3148 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
3149 | type = MEMFILE_TYPE(event); |
3150 | name = MEMFILE_ATTR(event); | |
3151 | switch (name) { | |
29f2a4da | 3152 | case RES_MAX_USAGE: |
8c7c6e34 KH |
3153 | if (type == _MEM) |
3154 | res_counter_reset_max(&mem->res); | |
3155 | else | |
3156 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
3157 | break; |
3158 | case RES_FAILCNT: | |
8c7c6e34 KH |
3159 | if (type == _MEM) |
3160 | res_counter_reset_failcnt(&mem->res); | |
3161 | else | |
3162 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
3163 | break; |
3164 | } | |
f64c3f54 | 3165 | |
85cc59db | 3166 | return 0; |
c84872e1 PE |
3167 | } |
3168 | ||
7dc74be0 DN |
3169 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3170 | struct cftype *cft) | |
3171 | { | |
3172 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3173 | } | |
3174 | ||
02491447 | 3175 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3176 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3177 | struct cftype *cft, u64 val) | |
3178 | { | |
3179 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3180 | ||
3181 | if (val >= (1 << NR_MOVE_TYPE)) | |
3182 | return -EINVAL; | |
3183 | /* | |
3184 | * We check this value several times in both in can_attach() and | |
3185 | * attach(), so we need cgroup lock to prevent this value from being | |
3186 | * inconsistent. | |
3187 | */ | |
3188 | cgroup_lock(); | |
3189 | mem->move_charge_at_immigrate = val; | |
3190 | cgroup_unlock(); | |
3191 | ||
3192 | return 0; | |
3193 | } | |
02491447 DN |
3194 | #else |
3195 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3196 | struct cftype *cft, u64 val) | |
3197 | { | |
3198 | return -ENOSYS; | |
3199 | } | |
3200 | #endif | |
7dc74be0 | 3201 | |
14067bb3 KH |
3202 | |
3203 | /* For read statistics */ | |
3204 | enum { | |
3205 | MCS_CACHE, | |
3206 | MCS_RSS, | |
d8046582 | 3207 | MCS_FILE_MAPPED, |
14067bb3 KH |
3208 | MCS_PGPGIN, |
3209 | MCS_PGPGOUT, | |
1dd3a273 | 3210 | MCS_SWAP, |
14067bb3 KH |
3211 | MCS_INACTIVE_ANON, |
3212 | MCS_ACTIVE_ANON, | |
3213 | MCS_INACTIVE_FILE, | |
3214 | MCS_ACTIVE_FILE, | |
3215 | MCS_UNEVICTABLE, | |
3216 | NR_MCS_STAT, | |
3217 | }; | |
3218 | ||
3219 | struct mcs_total_stat { | |
3220 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
3221 | }; |
3222 | ||
14067bb3 KH |
3223 | struct { |
3224 | char *local_name; | |
3225 | char *total_name; | |
3226 | } memcg_stat_strings[NR_MCS_STAT] = { | |
3227 | {"cache", "total_cache"}, | |
3228 | {"rss", "total_rss"}, | |
d69b042f | 3229 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
3230 | {"pgpgin", "total_pgpgin"}, |
3231 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3232 | {"swap", "total_swap"}, |
14067bb3 KH |
3233 | {"inactive_anon", "total_inactive_anon"}, |
3234 | {"active_anon", "total_active_anon"}, | |
3235 | {"inactive_file", "total_inactive_file"}, | |
3236 | {"active_file", "total_active_file"}, | |
3237 | {"unevictable", "total_unevictable"} | |
3238 | }; | |
3239 | ||
3240 | ||
3241 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
3242 | { | |
3243 | struct mcs_total_stat *s = data; | |
3244 | s64 val; | |
3245 | ||
3246 | /* per cpu stat */ | |
c62b1a3b | 3247 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 3248 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 3249 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 3250 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 3251 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 3252 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c62b1a3b | 3253 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); |
14067bb3 | 3254 | s->stat[MCS_PGPGIN] += val; |
c62b1a3b | 3255 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
14067bb3 | 3256 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 3257 | if (do_swap_account) { |
c62b1a3b | 3258 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
3259 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
3260 | } | |
14067bb3 KH |
3261 | |
3262 | /* per zone stat */ | |
3263 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
3264 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
3265 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
3266 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
3267 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
3268 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
3269 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
3270 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
3271 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
3272 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
3273 | return 0; | |
3274 | } | |
3275 | ||
3276 | static void | |
3277 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
3278 | { | |
3279 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
3280 | } | |
3281 | ||
c64745cf PM |
3282 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
3283 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 3284 | { |
d2ceb9b7 | 3285 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 3286 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
3287 | int i; |
3288 | ||
14067bb3 KH |
3289 | memset(&mystat, 0, sizeof(mystat)); |
3290 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 3291 | |
1dd3a273 DN |
3292 | for (i = 0; i < NR_MCS_STAT; i++) { |
3293 | if (i == MCS_SWAP && !do_swap_account) | |
3294 | continue; | |
14067bb3 | 3295 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 3296 | } |
7b854121 | 3297 | |
14067bb3 | 3298 | /* Hierarchical information */ |
fee7b548 KH |
3299 | { |
3300 | unsigned long long limit, memsw_limit; | |
3301 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
3302 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
3303 | if (do_swap_account) | |
3304 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
3305 | } | |
7f016ee8 | 3306 | |
14067bb3 KH |
3307 | memset(&mystat, 0, sizeof(mystat)); |
3308 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
3309 | for (i = 0; i < NR_MCS_STAT; i++) { |
3310 | if (i == MCS_SWAP && !do_swap_account) | |
3311 | continue; | |
14067bb3 | 3312 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 3313 | } |
14067bb3 | 3314 | |
7f016ee8 | 3315 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 3316 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
3317 | |
3318 | { | |
3319 | int nid, zid; | |
3320 | struct mem_cgroup_per_zone *mz; | |
3321 | unsigned long recent_rotated[2] = {0, 0}; | |
3322 | unsigned long recent_scanned[2] = {0, 0}; | |
3323 | ||
3324 | for_each_online_node(nid) | |
3325 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3326 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3327 | ||
3328 | recent_rotated[0] += | |
3329 | mz->reclaim_stat.recent_rotated[0]; | |
3330 | recent_rotated[1] += | |
3331 | mz->reclaim_stat.recent_rotated[1]; | |
3332 | recent_scanned[0] += | |
3333 | mz->reclaim_stat.recent_scanned[0]; | |
3334 | recent_scanned[1] += | |
3335 | mz->reclaim_stat.recent_scanned[1]; | |
3336 | } | |
3337 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3338 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3339 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3340 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3341 | } | |
3342 | #endif | |
3343 | ||
d2ceb9b7 KH |
3344 | return 0; |
3345 | } | |
3346 | ||
a7885eb8 KM |
3347 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
3348 | { | |
3349 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3350 | ||
3351 | return get_swappiness(memcg); | |
3352 | } | |
3353 | ||
3354 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3355 | u64 val) | |
3356 | { | |
3357 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3358 | struct mem_cgroup *parent; | |
068b38c1 | 3359 | |
a7885eb8 KM |
3360 | if (val > 100) |
3361 | return -EINVAL; | |
3362 | ||
3363 | if (cgrp->parent == NULL) | |
3364 | return -EINVAL; | |
3365 | ||
3366 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
3367 | |
3368 | cgroup_lock(); | |
3369 | ||
a7885eb8 KM |
3370 | /* If under hierarchy, only empty-root can set this value */ |
3371 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
3372 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3373 | cgroup_unlock(); | |
a7885eb8 | 3374 | return -EINVAL; |
068b38c1 | 3375 | } |
a7885eb8 KM |
3376 | |
3377 | spin_lock(&memcg->reclaim_param_lock); | |
3378 | memcg->swappiness = val; | |
3379 | spin_unlock(&memcg->reclaim_param_lock); | |
3380 | ||
068b38c1 LZ |
3381 | cgroup_unlock(); |
3382 | ||
a7885eb8 KM |
3383 | return 0; |
3384 | } | |
3385 | ||
2e72b634 KS |
3386 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3387 | { | |
3388 | struct mem_cgroup_threshold_ary *t; | |
3389 | u64 usage; | |
3390 | int i; | |
3391 | ||
3392 | rcu_read_lock(); | |
3393 | if (!swap) | |
3394 | t = rcu_dereference(memcg->thresholds); | |
3395 | else | |
3396 | t = rcu_dereference(memcg->memsw_thresholds); | |
3397 | ||
3398 | if (!t) | |
3399 | goto unlock; | |
3400 | ||
3401 | usage = mem_cgroup_usage(memcg, swap); | |
3402 | ||
3403 | /* | |
3404 | * current_threshold points to threshold just below usage. | |
3405 | * If it's not true, a threshold was crossed after last | |
3406 | * call of __mem_cgroup_threshold(). | |
3407 | */ | |
3408 | i = atomic_read(&t->current_threshold); | |
3409 | ||
3410 | /* | |
3411 | * Iterate backward over array of thresholds starting from | |
3412 | * current_threshold and check if a threshold is crossed. | |
3413 | * If none of thresholds below usage is crossed, we read | |
3414 | * only one element of the array here. | |
3415 | */ | |
3416 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3417 | eventfd_signal(t->entries[i].eventfd, 1); | |
3418 | ||
3419 | /* i = current_threshold + 1 */ | |
3420 | i++; | |
3421 | ||
3422 | /* | |
3423 | * Iterate forward over array of thresholds starting from | |
3424 | * current_threshold+1 and check if a threshold is crossed. | |
3425 | * If none of thresholds above usage is crossed, we read | |
3426 | * only one element of the array here. | |
3427 | */ | |
3428 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3429 | eventfd_signal(t->entries[i].eventfd, 1); | |
3430 | ||
3431 | /* Update current_threshold */ | |
3432 | atomic_set(&t->current_threshold, i - 1); | |
3433 | unlock: | |
3434 | rcu_read_unlock(); | |
3435 | } | |
3436 | ||
3437 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3438 | { | |
3439 | __mem_cgroup_threshold(memcg, false); | |
3440 | if (do_swap_account) | |
3441 | __mem_cgroup_threshold(memcg, true); | |
3442 | } | |
3443 | ||
3444 | static int compare_thresholds(const void *a, const void *b) | |
3445 | { | |
3446 | const struct mem_cgroup_threshold *_a = a; | |
3447 | const struct mem_cgroup_threshold *_b = b; | |
3448 | ||
3449 | return _a->threshold - _b->threshold; | |
3450 | } | |
3451 | ||
9490ff27 KH |
3452 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem, void *data) |
3453 | { | |
3454 | struct mem_cgroup_eventfd_list *ev; | |
3455 | ||
3456 | list_for_each_entry(ev, &mem->oom_notify, list) | |
3457 | eventfd_signal(ev->eventfd, 1); | |
3458 | return 0; | |
3459 | } | |
3460 | ||
3461 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) | |
3462 | { | |
3463 | mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_notify_cb); | |
3464 | } | |
3465 | ||
3466 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
3467 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
3468 | { |
3469 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3470 | struct mem_cgroup_threshold_ary *thresholds, *thresholds_new; | |
3471 | int type = MEMFILE_TYPE(cft->private); | |
3472 | u64 threshold, usage; | |
3473 | int size; | |
3474 | int i, ret; | |
3475 | ||
3476 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
3477 | if (ret) | |
3478 | return ret; | |
3479 | ||
3480 | mutex_lock(&memcg->thresholds_lock); | |
3481 | if (type == _MEM) | |
3482 | thresholds = memcg->thresholds; | |
3483 | else if (type == _MEMSWAP) | |
3484 | thresholds = memcg->memsw_thresholds; | |
3485 | else | |
3486 | BUG(); | |
3487 | ||
3488 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3489 | ||
3490 | /* Check if a threshold crossed before adding a new one */ | |
3491 | if (thresholds) | |
3492 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3493 | ||
3494 | if (thresholds) | |
3495 | size = thresholds->size + 1; | |
3496 | else | |
3497 | size = 1; | |
3498 | ||
3499 | /* Allocate memory for new array of thresholds */ | |
3500 | thresholds_new = kmalloc(sizeof(*thresholds_new) + | |
3501 | size * sizeof(struct mem_cgroup_threshold), | |
3502 | GFP_KERNEL); | |
3503 | if (!thresholds_new) { | |
3504 | ret = -ENOMEM; | |
3505 | goto unlock; | |
3506 | } | |
3507 | thresholds_new->size = size; | |
3508 | ||
3509 | /* Copy thresholds (if any) to new array */ | |
3510 | if (thresholds) | |
3511 | memcpy(thresholds_new->entries, thresholds->entries, | |
3512 | thresholds->size * | |
3513 | sizeof(struct mem_cgroup_threshold)); | |
3514 | /* Add new threshold */ | |
3515 | thresholds_new->entries[size - 1].eventfd = eventfd; | |
3516 | thresholds_new->entries[size - 1].threshold = threshold; | |
3517 | ||
3518 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
3519 | sort(thresholds_new->entries, size, | |
3520 | sizeof(struct mem_cgroup_threshold), | |
3521 | compare_thresholds, NULL); | |
3522 | ||
3523 | /* Find current threshold */ | |
3524 | atomic_set(&thresholds_new->current_threshold, -1); | |
3525 | for (i = 0; i < size; i++) { | |
3526 | if (thresholds_new->entries[i].threshold < usage) { | |
3527 | /* | |
3528 | * thresholds_new->current_threshold will not be used | |
3529 | * until rcu_assign_pointer(), so it's safe to increment | |
3530 | * it here. | |
3531 | */ | |
3532 | atomic_inc(&thresholds_new->current_threshold); | |
3533 | } | |
3534 | } | |
3535 | ||
2e72b634 KS |
3536 | if (type == _MEM) |
3537 | rcu_assign_pointer(memcg->thresholds, thresholds_new); | |
3538 | else | |
3539 | rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new); | |
3540 | ||
3541 | /* To be sure that nobody uses thresholds before freeing it */ | |
3542 | synchronize_rcu(); | |
3543 | ||
3544 | kfree(thresholds); | |
3545 | unlock: | |
3546 | mutex_unlock(&memcg->thresholds_lock); | |
3547 | ||
3548 | return ret; | |
3549 | } | |
3550 | ||
9490ff27 KH |
3551 | static int mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
3552 | struct cftype *cft, struct eventfd_ctx *eventfd) | |
2e72b634 KS |
3553 | { |
3554 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3555 | struct mem_cgroup_threshold_ary *thresholds, *thresholds_new; | |
3556 | int type = MEMFILE_TYPE(cft->private); | |
3557 | u64 usage; | |
3558 | int size = 0; | |
3559 | int i, j, ret; | |
3560 | ||
3561 | mutex_lock(&memcg->thresholds_lock); | |
3562 | if (type == _MEM) | |
3563 | thresholds = memcg->thresholds; | |
3564 | else if (type == _MEMSWAP) | |
3565 | thresholds = memcg->memsw_thresholds; | |
3566 | else | |
3567 | BUG(); | |
3568 | ||
3569 | /* | |
3570 | * Something went wrong if we trying to unregister a threshold | |
3571 | * if we don't have thresholds | |
3572 | */ | |
3573 | BUG_ON(!thresholds); | |
3574 | ||
3575 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3576 | ||
3577 | /* Check if a threshold crossed before removing */ | |
3578 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3579 | ||
3580 | /* Calculate new number of threshold */ | |
3581 | for (i = 0; i < thresholds->size; i++) { | |
3582 | if (thresholds->entries[i].eventfd != eventfd) | |
3583 | size++; | |
3584 | } | |
3585 | ||
3586 | /* Set thresholds array to NULL if we don't have thresholds */ | |
3587 | if (!size) { | |
3588 | thresholds_new = NULL; | |
3589 | goto assign; | |
3590 | } | |
3591 | ||
3592 | /* Allocate memory for new array of thresholds */ | |
3593 | thresholds_new = kmalloc(sizeof(*thresholds_new) + | |
3594 | size * sizeof(struct mem_cgroup_threshold), | |
3595 | GFP_KERNEL); | |
3596 | if (!thresholds_new) { | |
3597 | ret = -ENOMEM; | |
3598 | goto unlock; | |
3599 | } | |
3600 | thresholds_new->size = size; | |
3601 | ||
3602 | /* Copy thresholds and find current threshold */ | |
3603 | atomic_set(&thresholds_new->current_threshold, -1); | |
3604 | for (i = 0, j = 0; i < thresholds->size; i++) { | |
3605 | if (thresholds->entries[i].eventfd == eventfd) | |
3606 | continue; | |
3607 | ||
3608 | thresholds_new->entries[j] = thresholds->entries[i]; | |
3609 | if (thresholds_new->entries[j].threshold < usage) { | |
3610 | /* | |
3611 | * thresholds_new->current_threshold will not be used | |
3612 | * until rcu_assign_pointer(), so it's safe to increment | |
3613 | * it here. | |
3614 | */ | |
3615 | atomic_inc(&thresholds_new->current_threshold); | |
3616 | } | |
3617 | j++; | |
3618 | } | |
3619 | ||
3620 | assign: | |
3621 | if (type == _MEM) | |
3622 | rcu_assign_pointer(memcg->thresholds, thresholds_new); | |
3623 | else | |
3624 | rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new); | |
3625 | ||
3626 | /* To be sure that nobody uses thresholds before freeing it */ | |
3627 | synchronize_rcu(); | |
3628 | ||
2e72b634 KS |
3629 | kfree(thresholds); |
3630 | unlock: | |
3631 | mutex_unlock(&memcg->thresholds_lock); | |
3632 | ||
3633 | return ret; | |
3634 | } | |
c1e862c1 | 3635 | |
9490ff27 KH |
3636 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
3637 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
3638 | { | |
3639 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3640 | struct mem_cgroup_eventfd_list *event; | |
3641 | int type = MEMFILE_TYPE(cft->private); | |
3642 | ||
3643 | BUG_ON(type != _OOM_TYPE); | |
3644 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
3645 | if (!event) | |
3646 | return -ENOMEM; | |
3647 | ||
3648 | mutex_lock(&memcg_oom_mutex); | |
3649 | ||
3650 | event->eventfd = eventfd; | |
3651 | list_add(&event->list, &memcg->oom_notify); | |
3652 | ||
3653 | /* already in OOM ? */ | |
3654 | if (atomic_read(&memcg->oom_lock)) | |
3655 | eventfd_signal(eventfd, 1); | |
3656 | mutex_unlock(&memcg_oom_mutex); | |
3657 | ||
3658 | return 0; | |
3659 | } | |
3660 | ||
3661 | static int mem_cgroup_oom_unregister_event(struct cgroup *cgrp, | |
3662 | struct cftype *cft, struct eventfd_ctx *eventfd) | |
3663 | { | |
3664 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3665 | struct mem_cgroup_eventfd_list *ev, *tmp; | |
3666 | int type = MEMFILE_TYPE(cft->private); | |
3667 | ||
3668 | BUG_ON(type != _OOM_TYPE); | |
3669 | ||
3670 | mutex_lock(&memcg_oom_mutex); | |
3671 | ||
3672 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { | |
3673 | if (ev->eventfd == eventfd) { | |
3674 | list_del(&ev->list); | |
3675 | kfree(ev); | |
3676 | } | |
3677 | } | |
3678 | ||
3679 | mutex_unlock(&memcg_oom_mutex); | |
3680 | ||
3681 | return 0; | |
3682 | } | |
3683 | ||
3c11ecf4 KH |
3684 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
3685 | struct cftype *cft, struct cgroup_map_cb *cb) | |
3686 | { | |
3687 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3688 | ||
3689 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); | |
3690 | ||
3691 | if (atomic_read(&mem->oom_lock)) | |
3692 | cb->fill(cb, "under_oom", 1); | |
3693 | else | |
3694 | cb->fill(cb, "under_oom", 0); | |
3695 | return 0; | |
3696 | } | |
3697 | ||
3698 | /* | |
3699 | */ | |
3700 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, | |
3701 | struct cftype *cft, u64 val) | |
3702 | { | |
3703 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3704 | struct mem_cgroup *parent; | |
3705 | ||
3706 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
3707 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
3708 | return -EINVAL; | |
3709 | ||
3710 | parent = mem_cgroup_from_cont(cgrp->parent); | |
3711 | ||
3712 | cgroup_lock(); | |
3713 | /* oom-kill-disable is a flag for subhierarchy. */ | |
3714 | if ((parent->use_hierarchy) || | |
3715 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { | |
3716 | cgroup_unlock(); | |
3717 | return -EINVAL; | |
3718 | } | |
3719 | mem->oom_kill_disable = val; | |
3720 | cgroup_unlock(); | |
3721 | return 0; | |
3722 | } | |
3723 | ||
8cdea7c0 BS |
3724 | static struct cftype mem_cgroup_files[] = { |
3725 | { | |
0eea1030 | 3726 | .name = "usage_in_bytes", |
8c7c6e34 | 3727 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 3728 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
3729 | .register_event = mem_cgroup_usage_register_event, |
3730 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 3731 | }, |
c84872e1 PE |
3732 | { |
3733 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3734 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 3735 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
3736 | .read_u64 = mem_cgroup_read, |
3737 | }, | |
8cdea7c0 | 3738 | { |
0eea1030 | 3739 | .name = "limit_in_bytes", |
8c7c6e34 | 3740 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 3741 | .write_string = mem_cgroup_write, |
2c3daa72 | 3742 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3743 | }, |
296c81d8 BS |
3744 | { |
3745 | .name = "soft_limit_in_bytes", | |
3746 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
3747 | .write_string = mem_cgroup_write, | |
3748 | .read_u64 = mem_cgroup_read, | |
3749 | }, | |
8cdea7c0 BS |
3750 | { |
3751 | .name = "failcnt", | |
8c7c6e34 | 3752 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 3753 | .trigger = mem_cgroup_reset, |
2c3daa72 | 3754 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3755 | }, |
d2ceb9b7 KH |
3756 | { |
3757 | .name = "stat", | |
c64745cf | 3758 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 3759 | }, |
c1e862c1 KH |
3760 | { |
3761 | .name = "force_empty", | |
3762 | .trigger = mem_cgroup_force_empty_write, | |
3763 | }, | |
18f59ea7 BS |
3764 | { |
3765 | .name = "use_hierarchy", | |
3766 | .write_u64 = mem_cgroup_hierarchy_write, | |
3767 | .read_u64 = mem_cgroup_hierarchy_read, | |
3768 | }, | |
a7885eb8 KM |
3769 | { |
3770 | .name = "swappiness", | |
3771 | .read_u64 = mem_cgroup_swappiness_read, | |
3772 | .write_u64 = mem_cgroup_swappiness_write, | |
3773 | }, | |
7dc74be0 DN |
3774 | { |
3775 | .name = "move_charge_at_immigrate", | |
3776 | .read_u64 = mem_cgroup_move_charge_read, | |
3777 | .write_u64 = mem_cgroup_move_charge_write, | |
3778 | }, | |
9490ff27 KH |
3779 | { |
3780 | .name = "oom_control", | |
3c11ecf4 KH |
3781 | .read_map = mem_cgroup_oom_control_read, |
3782 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
3783 | .register_event = mem_cgroup_oom_register_event, |
3784 | .unregister_event = mem_cgroup_oom_unregister_event, | |
3785 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
3786 | }, | |
8cdea7c0 BS |
3787 | }; |
3788 | ||
8c7c6e34 KH |
3789 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3790 | static struct cftype memsw_cgroup_files[] = { | |
3791 | { | |
3792 | .name = "memsw.usage_in_bytes", | |
3793 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
3794 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
3795 | .register_event = mem_cgroup_usage_register_event, |
3796 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
3797 | }, |
3798 | { | |
3799 | .name = "memsw.max_usage_in_bytes", | |
3800 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
3801 | .trigger = mem_cgroup_reset, | |
3802 | .read_u64 = mem_cgroup_read, | |
3803 | }, | |
3804 | { | |
3805 | .name = "memsw.limit_in_bytes", | |
3806 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
3807 | .write_string = mem_cgroup_write, | |
3808 | .read_u64 = mem_cgroup_read, | |
3809 | }, | |
3810 | { | |
3811 | .name = "memsw.failcnt", | |
3812 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
3813 | .trigger = mem_cgroup_reset, | |
3814 | .read_u64 = mem_cgroup_read, | |
3815 | }, | |
3816 | }; | |
3817 | ||
3818 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3819 | { | |
3820 | if (!do_swap_account) | |
3821 | return 0; | |
3822 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
3823 | ARRAY_SIZE(memsw_cgroup_files)); | |
3824 | }; | |
3825 | #else | |
3826 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3827 | { | |
3828 | return 0; | |
3829 | } | |
3830 | #endif | |
3831 | ||
6d12e2d8 KH |
3832 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3833 | { | |
3834 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 3835 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 3836 | enum lru_list l; |
41e3355d | 3837 | int zone, tmp = node; |
1ecaab2b KH |
3838 | /* |
3839 | * This routine is called against possible nodes. | |
3840 | * But it's BUG to call kmalloc() against offline node. | |
3841 | * | |
3842 | * TODO: this routine can waste much memory for nodes which will | |
3843 | * never be onlined. It's better to use memory hotplug callback | |
3844 | * function. | |
3845 | */ | |
41e3355d KH |
3846 | if (!node_state(node, N_NORMAL_MEMORY)) |
3847 | tmp = -1; | |
3848 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
3849 | if (!pn) |
3850 | return 1; | |
1ecaab2b | 3851 | |
6d12e2d8 KH |
3852 | mem->info.nodeinfo[node] = pn; |
3853 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
3854 | |
3855 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3856 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
3857 | for_each_lru(l) |
3858 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 3859 | mz->usage_in_excess = 0; |
4e416953 BS |
3860 | mz->on_tree = false; |
3861 | mz->mem = mem; | |
1ecaab2b | 3862 | } |
6d12e2d8 KH |
3863 | return 0; |
3864 | } | |
3865 | ||
1ecaab2b KH |
3866 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3867 | { | |
3868 | kfree(mem->info.nodeinfo[node]); | |
3869 | } | |
3870 | ||
33327948 KH |
3871 | static struct mem_cgroup *mem_cgroup_alloc(void) |
3872 | { | |
3873 | struct mem_cgroup *mem; | |
c62b1a3b | 3874 | int size = sizeof(struct mem_cgroup); |
33327948 | 3875 | |
c62b1a3b | 3876 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb JB |
3877 | if (size < PAGE_SIZE) |
3878 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 3879 | else |
c8dad2bb | 3880 | mem = vmalloc(size); |
33327948 | 3881 | |
e7bbcdf3 DC |
3882 | if (!mem) |
3883 | return NULL; | |
3884 | ||
3885 | memset(mem, 0, size); | |
c62b1a3b KH |
3886 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
3887 | if (!mem->stat) { | |
3888 | if (size < PAGE_SIZE) | |
3889 | kfree(mem); | |
3890 | else | |
3891 | vfree(mem); | |
3892 | mem = NULL; | |
3893 | } | |
33327948 KH |
3894 | return mem; |
3895 | } | |
3896 | ||
8c7c6e34 KH |
3897 | /* |
3898 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
3899 | * (scanning all at force_empty is too costly...) | |
3900 | * | |
3901 | * Instead of clearing all references at force_empty, we remember | |
3902 | * the number of reference from swap_cgroup and free mem_cgroup when | |
3903 | * it goes down to 0. | |
3904 | * | |
8c7c6e34 KH |
3905 | * Removal of cgroup itself succeeds regardless of refs from swap. |
3906 | */ | |
3907 | ||
a7ba0eef | 3908 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 3909 | { |
08e552c6 KH |
3910 | int node; |
3911 | ||
f64c3f54 | 3912 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
3913 | free_css_id(&mem_cgroup_subsys, &mem->css); |
3914 | ||
08e552c6 KH |
3915 | for_each_node_state(node, N_POSSIBLE) |
3916 | free_mem_cgroup_per_zone_info(mem, node); | |
3917 | ||
c62b1a3b KH |
3918 | free_percpu(mem->stat); |
3919 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
3920 | kfree(mem); |
3921 | else | |
3922 | vfree(mem); | |
3923 | } | |
3924 | ||
8c7c6e34 KH |
3925 | static void mem_cgroup_get(struct mem_cgroup *mem) |
3926 | { | |
3927 | atomic_inc(&mem->refcnt); | |
3928 | } | |
3929 | ||
483c30b5 | 3930 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 3931 | { |
483c30b5 | 3932 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 3933 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 3934 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
3935 | if (parent) |
3936 | mem_cgroup_put(parent); | |
3937 | } | |
8c7c6e34 KH |
3938 | } |
3939 | ||
483c30b5 DN |
3940 | static void mem_cgroup_put(struct mem_cgroup *mem) |
3941 | { | |
3942 | __mem_cgroup_put(mem, 1); | |
3943 | } | |
3944 | ||
7bcc1bb1 DN |
3945 | /* |
3946 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
3947 | */ | |
3948 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
3949 | { | |
3950 | if (!mem->res.parent) | |
3951 | return NULL; | |
3952 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
3953 | } | |
33327948 | 3954 | |
c077719b KH |
3955 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3956 | static void __init enable_swap_cgroup(void) | |
3957 | { | |
f8d66542 | 3958 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
3959 | do_swap_account = 1; |
3960 | } | |
3961 | #else | |
3962 | static void __init enable_swap_cgroup(void) | |
3963 | { | |
3964 | } | |
3965 | #endif | |
3966 | ||
f64c3f54 BS |
3967 | static int mem_cgroup_soft_limit_tree_init(void) |
3968 | { | |
3969 | struct mem_cgroup_tree_per_node *rtpn; | |
3970 | struct mem_cgroup_tree_per_zone *rtpz; | |
3971 | int tmp, node, zone; | |
3972 | ||
3973 | for_each_node_state(node, N_POSSIBLE) { | |
3974 | tmp = node; | |
3975 | if (!node_state(node, N_NORMAL_MEMORY)) | |
3976 | tmp = -1; | |
3977 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
3978 | if (!rtpn) | |
3979 | return 1; | |
3980 | ||
3981 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
3982 | ||
3983 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3984 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
3985 | rtpz->rb_root = RB_ROOT; | |
3986 | spin_lock_init(&rtpz->lock); | |
3987 | } | |
3988 | } | |
3989 | return 0; | |
3990 | } | |
3991 | ||
0eb253e2 | 3992 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
3993 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
3994 | { | |
28dbc4b6 | 3995 | struct mem_cgroup *mem, *parent; |
04046e1a | 3996 | long error = -ENOMEM; |
6d12e2d8 | 3997 | int node; |
8cdea7c0 | 3998 | |
c8dad2bb JB |
3999 | mem = mem_cgroup_alloc(); |
4000 | if (!mem) | |
04046e1a | 4001 | return ERR_PTR(error); |
78fb7466 | 4002 | |
6d12e2d8 KH |
4003 | for_each_node_state(node, N_POSSIBLE) |
4004 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
4005 | goto free_out; | |
f64c3f54 | 4006 | |
c077719b | 4007 | /* root ? */ |
28dbc4b6 | 4008 | if (cont->parent == NULL) { |
cdec2e42 | 4009 | int cpu; |
c077719b | 4010 | enable_swap_cgroup(); |
28dbc4b6 | 4011 | parent = NULL; |
4b3bde4c | 4012 | root_mem_cgroup = mem; |
f64c3f54 BS |
4013 | if (mem_cgroup_soft_limit_tree_init()) |
4014 | goto free_out; | |
cdec2e42 KH |
4015 | for_each_possible_cpu(cpu) { |
4016 | struct memcg_stock_pcp *stock = | |
4017 | &per_cpu(memcg_stock, cpu); | |
4018 | INIT_WORK(&stock->work, drain_local_stock); | |
4019 | } | |
4020 | hotcpu_notifier(memcg_stock_cpu_callback, 0); | |
18f59ea7 | 4021 | } else { |
28dbc4b6 | 4022 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 | 4023 | mem->use_hierarchy = parent->use_hierarchy; |
3c11ecf4 | 4024 | mem->oom_kill_disable = parent->oom_kill_disable; |
18f59ea7 | 4025 | } |
28dbc4b6 | 4026 | |
18f59ea7 BS |
4027 | if (parent && parent->use_hierarchy) { |
4028 | res_counter_init(&mem->res, &parent->res); | |
4029 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4030 | /* |
4031 | * We increment refcnt of the parent to ensure that we can | |
4032 | * safely access it on res_counter_charge/uncharge. | |
4033 | * This refcnt will be decremented when freeing this | |
4034 | * mem_cgroup(see mem_cgroup_put). | |
4035 | */ | |
4036 | mem_cgroup_get(parent); | |
18f59ea7 BS |
4037 | } else { |
4038 | res_counter_init(&mem->res, NULL); | |
4039 | res_counter_init(&mem->memsw, NULL); | |
4040 | } | |
04046e1a | 4041 | mem->last_scanned_child = 0; |
2733c06a | 4042 | spin_lock_init(&mem->reclaim_param_lock); |
9490ff27 | 4043 | INIT_LIST_HEAD(&mem->oom_notify); |
6d61ef40 | 4044 | |
a7885eb8 KM |
4045 | if (parent) |
4046 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 4047 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 4048 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 4049 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 4050 | return &mem->css; |
6d12e2d8 | 4051 | free_out: |
a7ba0eef | 4052 | __mem_cgroup_free(mem); |
4b3bde4c | 4053 | root_mem_cgroup = NULL; |
04046e1a | 4054 | return ERR_PTR(error); |
8cdea7c0 BS |
4055 | } |
4056 | ||
ec64f515 | 4057 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4058 | struct cgroup *cont) |
4059 | { | |
4060 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
4061 | |
4062 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
4063 | } |
4064 | ||
8cdea7c0 BS |
4065 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4066 | struct cgroup *cont) | |
4067 | { | |
c268e994 | 4068 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 4069 | |
c268e994 | 4070 | mem_cgroup_put(mem); |
8cdea7c0 BS |
4071 | } |
4072 | ||
4073 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4074 | struct cgroup *cont) | |
4075 | { | |
8c7c6e34 KH |
4076 | int ret; |
4077 | ||
4078 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4079 | ARRAY_SIZE(mem_cgroup_files)); | |
4080 | ||
4081 | if (!ret) | |
4082 | ret = register_memsw_files(cont, ss); | |
4083 | return ret; | |
8cdea7c0 BS |
4084 | } |
4085 | ||
02491447 | 4086 | #ifdef CONFIG_MMU |
7dc74be0 | 4087 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4088 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4089 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4090 | { |
854ffa8d DN |
4091 | int ret = 0; |
4092 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
4093 | struct mem_cgroup *mem = mc.to; |
4094 | ||
854ffa8d DN |
4095 | if (mem_cgroup_is_root(mem)) { |
4096 | mc.precharge += count; | |
4097 | /* we don't need css_get for root */ | |
4098 | return ret; | |
4099 | } | |
4100 | /* try to charge at once */ | |
4101 | if (count > 1) { | |
4102 | struct res_counter *dummy; | |
4103 | /* | |
4104 | * "mem" cannot be under rmdir() because we've already checked | |
4105 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
4106 | * are still under the same cgroup_mutex. So we can postpone | |
4107 | * css_get(). | |
4108 | */ | |
4109 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
4110 | goto one_by_one; | |
4111 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
4112 | PAGE_SIZE * count, &dummy)) { | |
4113 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
4114 | goto one_by_one; | |
4115 | } | |
4116 | mc.precharge += count; | |
4117 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
4118 | WARN_ON_ONCE(count > INT_MAX); | |
4119 | __css_get(&mem->css, (int)count); | |
4120 | return ret; | |
4121 | } | |
4122 | one_by_one: | |
4123 | /* fall back to one by one charge */ | |
4124 | while (count--) { | |
4125 | if (signal_pending(current)) { | |
4126 | ret = -EINTR; | |
4127 | break; | |
4128 | } | |
4129 | if (!batch_count--) { | |
4130 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4131 | cond_resched(); | |
4132 | } | |
430e4863 | 4133 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
854ffa8d DN |
4134 | if (ret || !mem) |
4135 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
4136 | return -ENOMEM; | |
4137 | mc.precharge++; | |
4138 | } | |
4ffef5fe DN |
4139 | return ret; |
4140 | } | |
4141 | ||
4142 | /** | |
4143 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
4144 | * @vma: the vma the pte to be checked belongs | |
4145 | * @addr: the address corresponding to the pte to be checked | |
4146 | * @ptent: the pte to be checked | |
02491447 | 4147 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4148 | * |
4149 | * Returns | |
4150 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4151 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4152 | * move charge. if @target is not NULL, the page is stored in target->page | |
4153 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4154 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4155 | * target for charge migration. if @target is not NULL, the entry is stored | |
4156 | * in target->ent. | |
4ffef5fe DN |
4157 | * |
4158 | * Called with pte lock held. | |
4159 | */ | |
4ffef5fe DN |
4160 | union mc_target { |
4161 | struct page *page; | |
02491447 | 4162 | swp_entry_t ent; |
4ffef5fe DN |
4163 | }; |
4164 | ||
4ffef5fe DN |
4165 | enum mc_target_type { |
4166 | MC_TARGET_NONE, /* not used */ | |
4167 | MC_TARGET_PAGE, | |
02491447 | 4168 | MC_TARGET_SWAP, |
4ffef5fe DN |
4169 | }; |
4170 | ||
90254a65 DN |
4171 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4172 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4173 | { |
90254a65 | 4174 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4175 | |
90254a65 DN |
4176 | if (!page || !page_mapped(page)) |
4177 | return NULL; | |
4178 | if (PageAnon(page)) { | |
4179 | /* we don't move shared anon */ | |
4180 | if (!move_anon() || page_mapcount(page) > 2) | |
4181 | return NULL; | |
4182 | } else | |
02491447 DN |
4183 | /* |
4184 | * TODO: We don't move charges of file(including shmem/tmpfs) | |
4185 | * pages for now. | |
4186 | */ | |
90254a65 DN |
4187 | return NULL; |
4188 | if (!get_page_unless_zero(page)) | |
4189 | return NULL; | |
4190 | ||
4191 | return page; | |
4192 | } | |
4193 | ||
4194 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4195 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4196 | { | |
4197 | int usage_count; | |
4198 | struct page *page = NULL; | |
4199 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4200 | ||
4201 | if (!move_anon() || non_swap_entry(ent)) | |
4202 | return NULL; | |
4203 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
4204 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
4205 | if (page) |
4206 | put_page(page); | |
90254a65 | 4207 | return NULL; |
02491447 | 4208 | } |
90254a65 DN |
4209 | if (do_swap_account) |
4210 | entry->val = ent.val; | |
4211 | ||
4212 | return page; | |
4213 | } | |
4214 | ||
4215 | static int is_target_pte_for_mc(struct vm_area_struct *vma, | |
4216 | unsigned long addr, pte_t ptent, union mc_target *target) | |
4217 | { | |
4218 | struct page *page = NULL; | |
4219 | struct page_cgroup *pc; | |
4220 | int ret = 0; | |
4221 | swp_entry_t ent = { .val = 0 }; | |
4222 | ||
4223 | if (pte_present(ptent)) | |
4224 | page = mc_handle_present_pte(vma, addr, ptent); | |
4225 | else if (is_swap_pte(ptent)) | |
4226 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
4227 | /* TODO: handle swap of shmes/tmpfs */ | |
4228 | ||
4229 | if (!page && !ent.val) | |
4230 | return 0; | |
02491447 DN |
4231 | if (page) { |
4232 | pc = lookup_page_cgroup(page); | |
4233 | /* | |
4234 | * Do only loose check w/o page_cgroup lock. | |
4235 | * mem_cgroup_move_account() checks the pc is valid or not under | |
4236 | * the lock. | |
4237 | */ | |
4238 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
4239 | ret = MC_TARGET_PAGE; | |
4240 | if (target) | |
4241 | target->page = page; | |
4242 | } | |
4243 | if (!ret || !target) | |
4244 | put_page(page); | |
4245 | } | |
90254a65 DN |
4246 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4247 | if (ent.val && !ret && | |
7f0f1546 KH |
4248 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
4249 | ret = MC_TARGET_SWAP; | |
4250 | if (target) | |
4251 | target->ent = ent; | |
4ffef5fe | 4252 | } |
4ffef5fe DN |
4253 | return ret; |
4254 | } | |
4255 | ||
4256 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
4257 | unsigned long addr, unsigned long end, | |
4258 | struct mm_walk *walk) | |
4259 | { | |
4260 | struct vm_area_struct *vma = walk->private; | |
4261 | pte_t *pte; | |
4262 | spinlock_t *ptl; | |
4263 | ||
4264 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4265 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
4266 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
4267 | mc.precharge++; /* increment precharge temporarily */ | |
4268 | pte_unmap_unlock(pte - 1, ptl); | |
4269 | cond_resched(); | |
4270 | ||
7dc74be0 DN |
4271 | return 0; |
4272 | } | |
4273 | ||
4ffef5fe DN |
4274 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4275 | { | |
4276 | unsigned long precharge; | |
4277 | struct vm_area_struct *vma; | |
4278 | ||
4279 | down_read(&mm->mmap_sem); | |
4280 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
4281 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
4282 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4283 | .mm = mm, | |
4284 | .private = vma, | |
4285 | }; | |
4286 | if (is_vm_hugetlb_page(vma)) | |
4287 | continue; | |
4288 | /* TODO: We don't move charges of shmem/tmpfs pages for now. */ | |
4289 | if (vma->vm_flags & VM_SHARED) | |
4290 | continue; | |
4291 | walk_page_range(vma->vm_start, vma->vm_end, | |
4292 | &mem_cgroup_count_precharge_walk); | |
4293 | } | |
4294 | up_read(&mm->mmap_sem); | |
4295 | ||
4296 | precharge = mc.precharge; | |
4297 | mc.precharge = 0; | |
4298 | ||
4299 | return precharge; | |
4300 | } | |
4301 | ||
4ffef5fe DN |
4302 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4303 | { | |
854ffa8d | 4304 | return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm)); |
4ffef5fe DN |
4305 | } |
4306 | ||
4307 | static void mem_cgroup_clear_mc(void) | |
4308 | { | |
4309 | /* we must uncharge all the leftover precharges from mc.to */ | |
854ffa8d DN |
4310 | if (mc.precharge) { |
4311 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
4312 | mc.precharge = 0; | |
3c11ecf4 | 4313 | memcg_oom_recover(mc.to); |
854ffa8d DN |
4314 | } |
4315 | /* | |
4316 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4317 | * we must uncharge here. | |
4318 | */ | |
4319 | if (mc.moved_charge) { | |
4320 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
4321 | mc.moved_charge = 0; | |
3c11ecf4 | 4322 | memcg_oom_recover(mc.from); |
4ffef5fe | 4323 | } |
483c30b5 DN |
4324 | /* we must fixup refcnts and charges */ |
4325 | if (mc.moved_swap) { | |
4326 | WARN_ON_ONCE(mc.moved_swap > INT_MAX); | |
4327 | /* uncharge swap account from the old cgroup */ | |
4328 | if (!mem_cgroup_is_root(mc.from)) | |
4329 | res_counter_uncharge(&mc.from->memsw, | |
4330 | PAGE_SIZE * mc.moved_swap); | |
4331 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
4332 | ||
4333 | if (!mem_cgroup_is_root(mc.to)) { | |
4334 | /* | |
4335 | * we charged both to->res and to->memsw, so we should | |
4336 | * uncharge to->res. | |
4337 | */ | |
4338 | res_counter_uncharge(&mc.to->res, | |
4339 | PAGE_SIZE * mc.moved_swap); | |
4340 | VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags)); | |
4341 | __css_put(&mc.to->css, mc.moved_swap); | |
4342 | } | |
4343 | /* we've already done mem_cgroup_get(mc.to) */ | |
4344 | ||
4345 | mc.moved_swap = 0; | |
4346 | } | |
4ffef5fe DN |
4347 | mc.from = NULL; |
4348 | mc.to = NULL; | |
8033b97c DN |
4349 | mc.moving_task = NULL; |
4350 | wake_up_all(&mc.waitq); | |
4ffef5fe DN |
4351 | } |
4352 | ||
7dc74be0 DN |
4353 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
4354 | struct cgroup *cgroup, | |
4355 | struct task_struct *p, | |
4356 | bool threadgroup) | |
4357 | { | |
4358 | int ret = 0; | |
4359 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
4360 | ||
4361 | if (mem->move_charge_at_immigrate) { | |
4362 | struct mm_struct *mm; | |
4363 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
4364 | ||
4365 | VM_BUG_ON(from == mem); | |
4366 | ||
4367 | mm = get_task_mm(p); | |
4368 | if (!mm) | |
4369 | return 0; | |
7dc74be0 | 4370 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
4371 | if (mm->owner == p) { |
4372 | VM_BUG_ON(mc.from); | |
4373 | VM_BUG_ON(mc.to); | |
4374 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 4375 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 4376 | VM_BUG_ON(mc.moved_swap); |
8033b97c | 4377 | VM_BUG_ON(mc.moving_task); |
4ffef5fe DN |
4378 | mc.from = from; |
4379 | mc.to = mem; | |
4380 | mc.precharge = 0; | |
854ffa8d | 4381 | mc.moved_charge = 0; |
483c30b5 | 4382 | mc.moved_swap = 0; |
8033b97c | 4383 | mc.moving_task = current; |
4ffef5fe DN |
4384 | |
4385 | ret = mem_cgroup_precharge_mc(mm); | |
4386 | if (ret) | |
4387 | mem_cgroup_clear_mc(); | |
4388 | } | |
7dc74be0 DN |
4389 | mmput(mm); |
4390 | } | |
4391 | return ret; | |
4392 | } | |
4393 | ||
4394 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4395 | struct cgroup *cgroup, | |
4396 | struct task_struct *p, | |
4397 | bool threadgroup) | |
4398 | { | |
4ffef5fe | 4399 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
4400 | } |
4401 | ||
4ffef5fe DN |
4402 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4403 | unsigned long addr, unsigned long end, | |
4404 | struct mm_walk *walk) | |
7dc74be0 | 4405 | { |
4ffef5fe DN |
4406 | int ret = 0; |
4407 | struct vm_area_struct *vma = walk->private; | |
4408 | pte_t *pte; | |
4409 | spinlock_t *ptl; | |
4410 | ||
4411 | retry: | |
4412 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4413 | for (; addr != end; addr += PAGE_SIZE) { | |
4414 | pte_t ptent = *(pte++); | |
4415 | union mc_target target; | |
4416 | int type; | |
4417 | struct page *page; | |
4418 | struct page_cgroup *pc; | |
02491447 | 4419 | swp_entry_t ent; |
4ffef5fe DN |
4420 | |
4421 | if (!mc.precharge) | |
4422 | break; | |
4423 | ||
4424 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
4425 | switch (type) { | |
4426 | case MC_TARGET_PAGE: | |
4427 | page = target.page; | |
4428 | if (isolate_lru_page(page)) | |
4429 | goto put; | |
4430 | pc = lookup_page_cgroup(page); | |
854ffa8d DN |
4431 | if (!mem_cgroup_move_account(pc, |
4432 | mc.from, mc.to, false)) { | |
4ffef5fe | 4433 | mc.precharge--; |
854ffa8d DN |
4434 | /* we uncharge from mc.from later. */ |
4435 | mc.moved_charge++; | |
4ffef5fe DN |
4436 | } |
4437 | putback_lru_page(page); | |
4438 | put: /* is_target_pte_for_mc() gets the page */ | |
4439 | put_page(page); | |
4440 | break; | |
02491447 DN |
4441 | case MC_TARGET_SWAP: |
4442 | ent = target.ent; | |
483c30b5 DN |
4443 | if (!mem_cgroup_move_swap_account(ent, |
4444 | mc.from, mc.to, false)) { | |
02491447 | 4445 | mc.precharge--; |
483c30b5 DN |
4446 | /* we fixup refcnts and charges later. */ |
4447 | mc.moved_swap++; | |
4448 | } | |
02491447 | 4449 | break; |
4ffef5fe DN |
4450 | default: |
4451 | break; | |
4452 | } | |
4453 | } | |
4454 | pte_unmap_unlock(pte - 1, ptl); | |
4455 | cond_resched(); | |
4456 | ||
4457 | if (addr != end) { | |
4458 | /* | |
4459 | * We have consumed all precharges we got in can_attach(). | |
4460 | * We try charge one by one, but don't do any additional | |
4461 | * charges to mc.to if we have failed in charge once in attach() | |
4462 | * phase. | |
4463 | */ | |
854ffa8d | 4464 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4465 | if (!ret) |
4466 | goto retry; | |
4467 | } | |
4468 | ||
4469 | return ret; | |
4470 | } | |
4471 | ||
4472 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4473 | { | |
4474 | struct vm_area_struct *vma; | |
4475 | ||
4476 | lru_add_drain_all(); | |
4477 | down_read(&mm->mmap_sem); | |
4478 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
4479 | int ret; | |
4480 | struct mm_walk mem_cgroup_move_charge_walk = { | |
4481 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4482 | .mm = mm, | |
4483 | .private = vma, | |
4484 | }; | |
4485 | if (is_vm_hugetlb_page(vma)) | |
4486 | continue; | |
4487 | /* TODO: We don't move charges of shmem/tmpfs pages for now. */ | |
4488 | if (vma->vm_flags & VM_SHARED) | |
4489 | continue; | |
4490 | ret = walk_page_range(vma->vm_start, vma->vm_end, | |
4491 | &mem_cgroup_move_charge_walk); | |
4492 | if (ret) | |
4493 | /* | |
4494 | * means we have consumed all precharges and failed in | |
4495 | * doing additional charge. Just abandon here. | |
4496 | */ | |
4497 | break; | |
4498 | } | |
4499 | up_read(&mm->mmap_sem); | |
7dc74be0 DN |
4500 | } |
4501 | ||
67e465a7 BS |
4502 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
4503 | struct cgroup *cont, | |
4504 | struct cgroup *old_cont, | |
be367d09 BB |
4505 | struct task_struct *p, |
4506 | bool threadgroup) | |
67e465a7 | 4507 | { |
4ffef5fe DN |
4508 | struct mm_struct *mm; |
4509 | ||
4510 | if (!mc.to) | |
4511 | /* no need to move charge */ | |
4512 | return; | |
4513 | ||
4514 | mm = get_task_mm(p); | |
4515 | if (mm) { | |
4516 | mem_cgroup_move_charge(mm); | |
4517 | mmput(mm); | |
4518 | } | |
4519 | mem_cgroup_clear_mc(); | |
67e465a7 | 4520 | } |
5cfb80a7 DN |
4521 | #else /* !CONFIG_MMU */ |
4522 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
4523 | struct cgroup *cgroup, | |
4524 | struct task_struct *p, | |
4525 | bool threadgroup) | |
4526 | { | |
4527 | return 0; | |
4528 | } | |
4529 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4530 | struct cgroup *cgroup, | |
4531 | struct task_struct *p, | |
4532 | bool threadgroup) | |
4533 | { | |
4534 | } | |
4535 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
4536 | struct cgroup *cont, | |
4537 | struct cgroup *old_cont, | |
4538 | struct task_struct *p, | |
4539 | bool threadgroup) | |
4540 | { | |
4541 | } | |
4542 | #endif | |
67e465a7 | 4543 | |
8cdea7c0 BS |
4544 | struct cgroup_subsys mem_cgroup_subsys = { |
4545 | .name = "memory", | |
4546 | .subsys_id = mem_cgroup_subsys_id, | |
4547 | .create = mem_cgroup_create, | |
df878fb0 | 4548 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
4549 | .destroy = mem_cgroup_destroy, |
4550 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
4551 | .can_attach = mem_cgroup_can_attach, |
4552 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 4553 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 4554 | .early_init = 0, |
04046e1a | 4555 | .use_id = 1, |
8cdea7c0 | 4556 | }; |
c077719b KH |
4557 | |
4558 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
4559 | ||
4560 | static int __init disable_swap_account(char *s) | |
4561 | { | |
4562 | really_do_swap_account = 0; | |
4563 | return 1; | |
4564 | } | |
4565 | __setup("noswapaccount", disable_swap_account); | |
4566 | #endif |