Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
8cdea7c0 BS |
13 | * This program is free software; you can redistribute it and/or modify |
14 | * it under the terms of the GNU General Public License as published by | |
15 | * the Free Software Foundation; either version 2 of the License, or | |
16 | * (at your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, | |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | * GNU General Public License for more details. | |
22 | */ | |
23 | ||
24 | #include <linux/res_counter.h> | |
25 | #include <linux/memcontrol.h> | |
26 | #include <linux/cgroup.h> | |
78fb7466 | 27 | #include <linux/mm.h> |
4ffef5fe | 28 | #include <linux/hugetlb.h> |
d13d1443 | 29 | #include <linux/pagemap.h> |
d52aa412 | 30 | #include <linux/smp.h> |
8a9f3ccd | 31 | #include <linux/page-flags.h> |
66e1707b | 32 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
33 | #include <linux/bit_spinlock.h> |
34 | #include <linux/rcupdate.h> | |
e222432b | 35 | #include <linux/limits.h> |
8c7c6e34 | 36 | #include <linux/mutex.h> |
f64c3f54 | 37 | #include <linux/rbtree.h> |
b6ac57d5 | 38 | #include <linux/slab.h> |
66e1707b | 39 | #include <linux/swap.h> |
02491447 | 40 | #include <linux/swapops.h> |
66e1707b | 41 | #include <linux/spinlock.h> |
2e72b634 KS |
42 | #include <linux/eventfd.h> |
43 | #include <linux/sort.h> | |
66e1707b | 44 | #include <linux/fs.h> |
d2ceb9b7 | 45 | #include <linux/seq_file.h> |
33327948 | 46 | #include <linux/vmalloc.h> |
b69408e8 | 47 | #include <linux/mm_inline.h> |
52d4b9ac | 48 | #include <linux/page_cgroup.h> |
cdec2e42 | 49 | #include <linux/cpu.h> |
08e552c6 | 50 | #include "internal.h" |
8cdea7c0 | 51 | |
8697d331 BS |
52 | #include <asm/uaccess.h> |
53 | ||
a181b0e8 | 54 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 55 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 56 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 57 | |
c077719b | 58 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 59 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b KH |
60 | int do_swap_account __read_mostly; |
61 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
62 | #else | |
63 | #define do_swap_account (0) | |
64 | #endif | |
65 | ||
d2265e6f KH |
66 | /* |
67 | * Per memcg event counter is incremented at every pagein/pageout. This counter | |
68 | * is used for trigger some periodic events. This is straightforward and better | |
69 | * than using jiffies etc. to handle periodic memcg event. | |
70 | * | |
71 | * These values will be used as !((event) & ((1 <<(thresh)) - 1)) | |
72 | */ | |
73 | #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */ | |
74 | #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */ | |
c077719b | 75 | |
d52aa412 KH |
76 | /* |
77 | * Statistics for memory cgroup. | |
78 | */ | |
79 | enum mem_cgroup_stat_index { | |
80 | /* | |
81 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
82 | */ | |
83 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 84 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 85 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
55e462b0 BR |
86 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
87 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
0c3e73e8 | 88 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
d2265e6f | 89 | MEM_CGROUP_EVENTS, /* incremented at every pagein/pageout */ |
d52aa412 KH |
90 | |
91 | MEM_CGROUP_STAT_NSTATS, | |
92 | }; | |
93 | ||
94 | struct mem_cgroup_stat_cpu { | |
95 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
d52aa412 KH |
96 | }; |
97 | ||
6d12e2d8 KH |
98 | /* |
99 | * per-zone information in memory controller. | |
100 | */ | |
6d12e2d8 | 101 | struct mem_cgroup_per_zone { |
072c56c1 KH |
102 | /* |
103 | * spin_lock to protect the per cgroup LRU | |
104 | */ | |
b69408e8 CL |
105 | struct list_head lists[NR_LRU_LISTS]; |
106 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
107 | |
108 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
109 | struct rb_node tree_node; /* RB tree node */ |
110 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
111 | /* the soft limit is exceeded*/ | |
112 | bool on_tree; | |
4e416953 BS |
113 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
114 | /* use container_of */ | |
6d12e2d8 KH |
115 | }; |
116 | /* Macro for accessing counter */ | |
117 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
118 | ||
119 | struct mem_cgroup_per_node { | |
120 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
121 | }; | |
122 | ||
123 | struct mem_cgroup_lru_info { | |
124 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
125 | }; | |
126 | ||
f64c3f54 BS |
127 | /* |
128 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
129 | * their hierarchy representation | |
130 | */ | |
131 | ||
132 | struct mem_cgroup_tree_per_zone { | |
133 | struct rb_root rb_root; | |
134 | spinlock_t lock; | |
135 | }; | |
136 | ||
137 | struct mem_cgroup_tree_per_node { | |
138 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
139 | }; | |
140 | ||
141 | struct mem_cgroup_tree { | |
142 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
143 | }; | |
144 | ||
145 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
146 | ||
2e72b634 KS |
147 | struct mem_cgroup_threshold { |
148 | struct eventfd_ctx *eventfd; | |
149 | u64 threshold; | |
150 | }; | |
151 | ||
9490ff27 | 152 | /* For threshold */ |
2e72b634 KS |
153 | struct mem_cgroup_threshold_ary { |
154 | /* An array index points to threshold just below usage. */ | |
5407a562 | 155 | int current_threshold; |
2e72b634 KS |
156 | /* Size of entries[] */ |
157 | unsigned int size; | |
158 | /* Array of thresholds */ | |
159 | struct mem_cgroup_threshold entries[0]; | |
160 | }; | |
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 */ |
87946a72 | 253 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
254 | NR_MOVE_TYPE, |
255 | }; | |
256 | ||
4ffef5fe DN |
257 | /* "mc" and its members are protected by cgroup_mutex */ |
258 | static struct move_charge_struct { | |
259 | struct mem_cgroup *from; | |
260 | struct mem_cgroup *to; | |
261 | unsigned long precharge; | |
854ffa8d | 262 | unsigned long moved_charge; |
483c30b5 | 263 | unsigned long moved_swap; |
8033b97c DN |
264 | struct task_struct *moving_task; /* a task moving charges */ |
265 | wait_queue_head_t waitq; /* a waitq for other context */ | |
266 | } mc = { | |
267 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), | |
268 | }; | |
4ffef5fe | 269 | |
90254a65 DN |
270 | static bool move_anon(void) |
271 | { | |
272 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
273 | &mc.to->move_charge_at_immigrate); | |
274 | } | |
275 | ||
87946a72 DN |
276 | static bool move_file(void) |
277 | { | |
278 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
279 | &mc.to->move_charge_at_immigrate); | |
280 | } | |
281 | ||
4e416953 BS |
282 | /* |
283 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
284 | * limit reclaim to prevent infinite loops, if they ever occur. | |
285 | */ | |
286 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
287 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
288 | ||
217bc319 KH |
289 | enum charge_type { |
290 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
291 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 292 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 293 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 294 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 295 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
296 | NR_CHARGE_TYPE, |
297 | }; | |
298 | ||
52d4b9ac KH |
299 | /* only for here (for easy reading.) */ |
300 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
301 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 302 | #define PCGF_LOCK (1UL << PCG_LOCK) |
4b3bde4c BS |
303 | /* Not used, but added here for completeness */ |
304 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
217bc319 | 305 | |
8c7c6e34 KH |
306 | /* for encoding cft->private value on file */ |
307 | #define _MEM (0) | |
308 | #define _MEMSWAP (1) | |
9490ff27 | 309 | #define _OOM_TYPE (2) |
8c7c6e34 KH |
310 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
311 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
312 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
313 | /* Used for OOM nofiier */ |
314 | #define OOM_CONTROL (0) | |
8c7c6e34 | 315 | |
75822b44 BS |
316 | /* |
317 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
318 | */ | |
319 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
320 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
321 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
322 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
323 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
324 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 325 | |
8c7c6e34 KH |
326 | static void mem_cgroup_get(struct mem_cgroup *mem); |
327 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 328 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 329 | static void drain_all_stock_async(void); |
8c7c6e34 | 330 | |
f64c3f54 BS |
331 | static struct mem_cgroup_per_zone * |
332 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
333 | { | |
334 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
335 | } | |
336 | ||
d324236b WF |
337 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
338 | { | |
339 | return &mem->css; | |
340 | } | |
341 | ||
f64c3f54 BS |
342 | static struct mem_cgroup_per_zone * |
343 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
344 | { | |
345 | struct mem_cgroup *mem = pc->mem_cgroup; | |
346 | int nid = page_cgroup_nid(pc); | |
347 | int zid = page_cgroup_zid(pc); | |
348 | ||
349 | if (!mem) | |
350 | return NULL; | |
351 | ||
352 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
353 | } | |
354 | ||
355 | static struct mem_cgroup_tree_per_zone * | |
356 | soft_limit_tree_node_zone(int nid, int zid) | |
357 | { | |
358 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
359 | } | |
360 | ||
361 | static struct mem_cgroup_tree_per_zone * | |
362 | soft_limit_tree_from_page(struct page *page) | |
363 | { | |
364 | int nid = page_to_nid(page); | |
365 | int zid = page_zonenum(page); | |
366 | ||
367 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
368 | } | |
369 | ||
370 | static void | |
4e416953 | 371 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 372 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
373 | struct mem_cgroup_tree_per_zone *mctz, |
374 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
375 | { |
376 | struct rb_node **p = &mctz->rb_root.rb_node; | |
377 | struct rb_node *parent = NULL; | |
378 | struct mem_cgroup_per_zone *mz_node; | |
379 | ||
380 | if (mz->on_tree) | |
381 | return; | |
382 | ||
ef8745c1 KH |
383 | mz->usage_in_excess = new_usage_in_excess; |
384 | if (!mz->usage_in_excess) | |
385 | return; | |
f64c3f54 BS |
386 | while (*p) { |
387 | parent = *p; | |
388 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
389 | tree_node); | |
390 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
391 | p = &(*p)->rb_left; | |
392 | /* | |
393 | * We can't avoid mem cgroups that are over their soft | |
394 | * limit by the same amount | |
395 | */ | |
396 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
397 | p = &(*p)->rb_right; | |
398 | } | |
399 | rb_link_node(&mz->tree_node, parent, p); | |
400 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
401 | mz->on_tree = true; | |
4e416953 BS |
402 | } |
403 | ||
404 | static void | |
405 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
406 | struct mem_cgroup_per_zone *mz, | |
407 | struct mem_cgroup_tree_per_zone *mctz) | |
408 | { | |
409 | if (!mz->on_tree) | |
410 | return; | |
411 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
412 | mz->on_tree = false; | |
413 | } | |
414 | ||
f64c3f54 BS |
415 | static void |
416 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
417 | struct mem_cgroup_per_zone *mz, | |
418 | struct mem_cgroup_tree_per_zone *mctz) | |
419 | { | |
420 | spin_lock(&mctz->lock); | |
4e416953 | 421 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
422 | spin_unlock(&mctz->lock); |
423 | } | |
424 | ||
f64c3f54 BS |
425 | |
426 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
427 | { | |
ef8745c1 | 428 | unsigned long long excess; |
f64c3f54 BS |
429 | struct mem_cgroup_per_zone *mz; |
430 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
431 | int nid = page_to_nid(page); |
432 | int zid = page_zonenum(page); | |
f64c3f54 BS |
433 | mctz = soft_limit_tree_from_page(page); |
434 | ||
435 | /* | |
4e649152 KH |
436 | * Necessary to update all ancestors when hierarchy is used. |
437 | * because their event counter is not touched. | |
f64c3f54 | 438 | */ |
4e649152 KH |
439 | for (; mem; mem = parent_mem_cgroup(mem)) { |
440 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 441 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
442 | /* |
443 | * We have to update the tree if mz is on RB-tree or | |
444 | * mem is over its softlimit. | |
445 | */ | |
ef8745c1 | 446 | if (excess || mz->on_tree) { |
4e649152 KH |
447 | spin_lock(&mctz->lock); |
448 | /* if on-tree, remove it */ | |
449 | if (mz->on_tree) | |
450 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
451 | /* | |
ef8745c1 KH |
452 | * Insert again. mz->usage_in_excess will be updated. |
453 | * If excess is 0, no tree ops. | |
4e649152 | 454 | */ |
ef8745c1 | 455 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
456 | spin_unlock(&mctz->lock); |
457 | } | |
f64c3f54 BS |
458 | } |
459 | } | |
460 | ||
461 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
462 | { | |
463 | int node, zone; | |
464 | struct mem_cgroup_per_zone *mz; | |
465 | struct mem_cgroup_tree_per_zone *mctz; | |
466 | ||
467 | for_each_node_state(node, N_POSSIBLE) { | |
468 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
469 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
470 | mctz = soft_limit_tree_node_zone(node, zone); | |
471 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
472 | } | |
473 | } | |
474 | } | |
475 | ||
4e416953 BS |
476 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
477 | { | |
478 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; | |
479 | } | |
480 | ||
481 | static struct mem_cgroup_per_zone * | |
482 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
483 | { | |
484 | struct rb_node *rightmost = NULL; | |
26251eaf | 485 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
486 | |
487 | retry: | |
26251eaf | 488 | mz = NULL; |
4e416953 BS |
489 | rightmost = rb_last(&mctz->rb_root); |
490 | if (!rightmost) | |
491 | goto done; /* Nothing to reclaim from */ | |
492 | ||
493 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
494 | /* | |
495 | * Remove the node now but someone else can add it back, | |
496 | * we will to add it back at the end of reclaim to its correct | |
497 | * position in the tree. | |
498 | */ | |
499 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
500 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
501 | !css_tryget(&mz->mem->css)) | |
502 | goto retry; | |
503 | done: | |
504 | return mz; | |
505 | } | |
506 | ||
507 | static struct mem_cgroup_per_zone * | |
508 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
509 | { | |
510 | struct mem_cgroup_per_zone *mz; | |
511 | ||
512 | spin_lock(&mctz->lock); | |
513 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
514 | spin_unlock(&mctz->lock); | |
515 | return mz; | |
516 | } | |
517 | ||
c62b1a3b KH |
518 | static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, |
519 | enum mem_cgroup_stat_index idx) | |
520 | { | |
521 | int cpu; | |
522 | s64 val = 0; | |
523 | ||
524 | for_each_possible_cpu(cpu) | |
525 | val += per_cpu(mem->stat->count[idx], cpu); | |
526 | return val; | |
527 | } | |
528 | ||
529 | static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) | |
530 | { | |
531 | s64 ret; | |
532 | ||
533 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); | |
534 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); | |
535 | return ret; | |
536 | } | |
537 | ||
0c3e73e8 BS |
538 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
539 | bool charge) | |
540 | { | |
541 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 542 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
543 | } |
544 | ||
c05555b5 KH |
545 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
546 | struct page_cgroup *pc, | |
547 | bool charge) | |
d52aa412 | 548 | { |
0c3e73e8 | 549 | int val = (charge) ? 1 : -1; |
d52aa412 | 550 | |
c62b1a3b KH |
551 | preempt_disable(); |
552 | ||
c05555b5 | 553 | if (PageCgroupCache(pc)) |
c62b1a3b | 554 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val); |
d52aa412 | 555 | else |
c62b1a3b | 556 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val); |
55e462b0 BR |
557 | |
558 | if (charge) | |
c62b1a3b | 559 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); |
55e462b0 | 560 | else |
c62b1a3b | 561 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); |
d2265e6f | 562 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]); |
2e72b634 | 563 | |
c62b1a3b | 564 | preempt_enable(); |
6d12e2d8 KH |
565 | } |
566 | ||
14067bb3 | 567 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 568 | enum lru_list idx) |
6d12e2d8 KH |
569 | { |
570 | int nid, zid; | |
571 | struct mem_cgroup_per_zone *mz; | |
572 | u64 total = 0; | |
573 | ||
574 | for_each_online_node(nid) | |
575 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
576 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
577 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
578 | } | |
579 | return total; | |
d52aa412 KH |
580 | } |
581 | ||
d2265e6f KH |
582 | static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift) |
583 | { | |
584 | s64 val; | |
585 | ||
586 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]); | |
587 | ||
588 | return !(val & ((1 << event_mask_shift) - 1)); | |
589 | } | |
590 | ||
591 | /* | |
592 | * Check events in order. | |
593 | * | |
594 | */ | |
595 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) | |
596 | { | |
597 | /* threshold event is triggered in finer grain than soft limit */ | |
598 | if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) { | |
599 | mem_cgroup_threshold(mem); | |
600 | if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH))) | |
601 | mem_cgroup_update_tree(mem, page); | |
602 | } | |
603 | } | |
604 | ||
d5b69e38 | 605 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
606 | { |
607 | return container_of(cgroup_subsys_state(cont, | |
608 | mem_cgroup_subsys_id), struct mem_cgroup, | |
609 | css); | |
610 | } | |
611 | ||
cf475ad2 | 612 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 613 | { |
31a78f23 BS |
614 | /* |
615 | * mm_update_next_owner() may clear mm->owner to NULL | |
616 | * if it races with swapoff, page migration, etc. | |
617 | * So this can be called with p == NULL. | |
618 | */ | |
619 | if (unlikely(!p)) | |
620 | return NULL; | |
621 | ||
78fb7466 PE |
622 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
623 | struct mem_cgroup, css); | |
624 | } | |
625 | ||
54595fe2 KH |
626 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
627 | { | |
628 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
629 | |
630 | if (!mm) | |
631 | return NULL; | |
54595fe2 KH |
632 | /* |
633 | * Because we have no locks, mm->owner's may be being moved to other | |
634 | * cgroup. We use css_tryget() here even if this looks | |
635 | * pessimistic (rather than adding locks here). | |
636 | */ | |
637 | rcu_read_lock(); | |
638 | do { | |
639 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
640 | if (unlikely(!mem)) | |
641 | break; | |
642 | } while (!css_tryget(&mem->css)); | |
643 | rcu_read_unlock(); | |
644 | return mem; | |
645 | } | |
646 | ||
14067bb3 KH |
647 | /* |
648 | * Call callback function against all cgroup under hierarchy tree. | |
649 | */ | |
650 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
651 | int (*func)(struct mem_cgroup *, void *)) | |
652 | { | |
653 | int found, ret, nextid; | |
654 | struct cgroup_subsys_state *css; | |
655 | struct mem_cgroup *mem; | |
656 | ||
657 | if (!root->use_hierarchy) | |
658 | return (*func)(root, data); | |
659 | ||
660 | nextid = 1; | |
661 | do { | |
662 | ret = 0; | |
663 | mem = NULL; | |
664 | ||
665 | rcu_read_lock(); | |
666 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
667 | &found); | |
668 | if (css && css_tryget(css)) | |
669 | mem = container_of(css, struct mem_cgroup, css); | |
670 | rcu_read_unlock(); | |
671 | ||
672 | if (mem) { | |
673 | ret = (*func)(mem, data); | |
674 | css_put(&mem->css); | |
675 | } | |
676 | nextid = found + 1; | |
677 | } while (!ret && css); | |
678 | ||
679 | return ret; | |
680 | } | |
681 | ||
4b3bde4c BS |
682 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
683 | { | |
684 | return (mem == root_mem_cgroup); | |
685 | } | |
686 | ||
08e552c6 KH |
687 | /* |
688 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
689 | * Operations are called by routine of global LRU independently from memcg. | |
690 | * What we have to take care of here is validness of pc->mem_cgroup. | |
691 | * | |
692 | * Changes to pc->mem_cgroup happens when | |
693 | * 1. charge | |
694 | * 2. moving account | |
695 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
696 | * It is added to LRU before charge. | |
697 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
698 | * When moving account, the page is not on LRU. It's isolated. | |
699 | */ | |
4f98a2fe | 700 | |
08e552c6 KH |
701 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
702 | { | |
703 | struct page_cgroup *pc; | |
08e552c6 | 704 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 705 | |
f8d66542 | 706 | if (mem_cgroup_disabled()) |
08e552c6 KH |
707 | return; |
708 | pc = lookup_page_cgroup(page); | |
709 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 710 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 711 | return; |
4b3bde4c | 712 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
713 | /* |
714 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
715 | * removed from global LRU. | |
716 | */ | |
08e552c6 | 717 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 718 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
4b3bde4c BS |
719 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
720 | return; | |
721 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 KH |
722 | list_del_init(&pc->lru); |
723 | return; | |
6d12e2d8 KH |
724 | } |
725 | ||
08e552c6 | 726 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 727 | { |
08e552c6 KH |
728 | mem_cgroup_del_lru_list(page, page_lru(page)); |
729 | } | |
b69408e8 | 730 | |
08e552c6 KH |
731 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
732 | { | |
733 | struct mem_cgroup_per_zone *mz; | |
734 | struct page_cgroup *pc; | |
b69408e8 | 735 | |
f8d66542 | 736 | if (mem_cgroup_disabled()) |
08e552c6 | 737 | return; |
6d12e2d8 | 738 | |
08e552c6 | 739 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
740 | /* |
741 | * Used bit is set without atomic ops but after smp_wmb(). | |
742 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
743 | */ | |
08e552c6 | 744 | smp_rmb(); |
4b3bde4c BS |
745 | /* unused or root page is not rotated. */ |
746 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 KH |
747 | return; |
748 | mz = page_cgroup_zoneinfo(pc); | |
749 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
750 | } |
751 | ||
08e552c6 | 752 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 753 | { |
08e552c6 KH |
754 | struct page_cgroup *pc; |
755 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 756 | |
f8d66542 | 757 | if (mem_cgroup_disabled()) |
08e552c6 KH |
758 | return; |
759 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 760 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
bd112db8 DN |
761 | /* |
762 | * Used bit is set without atomic ops but after smp_wmb(). | |
763 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
764 | */ | |
08e552c6 KH |
765 | smp_rmb(); |
766 | if (!PageCgroupUsed(pc)) | |
894bc310 | 767 | return; |
b69408e8 | 768 | |
08e552c6 | 769 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 770 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
4b3bde4c BS |
771 | SetPageCgroupAcctLRU(pc); |
772 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
773 | return; | |
08e552c6 KH |
774 | list_add(&pc->lru, &mz->lists[lru]); |
775 | } | |
544122e5 | 776 | |
08e552c6 | 777 | /* |
544122e5 KH |
778 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
779 | * lru because the page may.be reused after it's fully uncharged (because of | |
780 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
781 | * it again. This function is only used to charge SwapCache. It's done under | |
782 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 783 | */ |
544122e5 | 784 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 785 | { |
544122e5 KH |
786 | unsigned long flags; |
787 | struct zone *zone = page_zone(page); | |
788 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
789 | ||
790 | spin_lock_irqsave(&zone->lru_lock, flags); | |
791 | /* | |
792 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
793 | * is guarded by lock_page() because the page is SwapCache. | |
794 | */ | |
795 | if (!PageCgroupUsed(pc)) | |
796 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
797 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
798 | } |
799 | ||
544122e5 KH |
800 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
801 | { | |
802 | unsigned long flags; | |
803 | struct zone *zone = page_zone(page); | |
804 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
805 | ||
806 | spin_lock_irqsave(&zone->lru_lock, flags); | |
807 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 808 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
809 | mem_cgroup_add_lru_list(page, page_lru(page)); |
810 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
811 | } | |
812 | ||
813 | ||
08e552c6 KH |
814 | void mem_cgroup_move_lists(struct page *page, |
815 | enum lru_list from, enum lru_list to) | |
816 | { | |
f8d66542 | 817 | if (mem_cgroup_disabled()) |
08e552c6 KH |
818 | return; |
819 | mem_cgroup_del_lru_list(page, from); | |
820 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
821 | } |
822 | ||
4c4a2214 DR |
823 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
824 | { | |
825 | int ret; | |
0b7f569e | 826 | struct mem_cgroup *curr = NULL; |
4c4a2214 DR |
827 | |
828 | task_lock(task); | |
0b7f569e KH |
829 | rcu_read_lock(); |
830 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
831 | rcu_read_unlock(); | |
4c4a2214 | 832 | task_unlock(task); |
0b7f569e KH |
833 | if (!curr) |
834 | return 0; | |
d31f56db DN |
835 | /* |
836 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
837 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
838 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
839 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
840 | */ | |
841 | if (mem->use_hierarchy) | |
0b7f569e KH |
842 | ret = css_is_ancestor(&curr->css, &mem->css); |
843 | else | |
844 | ret = (curr == mem); | |
845 | css_put(&curr->css); | |
4c4a2214 DR |
846 | return ret; |
847 | } | |
848 | ||
6c48a1d0 KH |
849 | /* |
850 | * prev_priority control...this will be used in memory reclaim path. | |
851 | */ | |
852 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
853 | { | |
2733c06a KM |
854 | int prev_priority; |
855 | ||
856 | spin_lock(&mem->reclaim_param_lock); | |
857 | prev_priority = mem->prev_priority; | |
858 | spin_unlock(&mem->reclaim_param_lock); | |
859 | ||
860 | return prev_priority; | |
6c48a1d0 KH |
861 | } |
862 | ||
863 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
864 | { | |
2733c06a | 865 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
866 | if (priority < mem->prev_priority) |
867 | mem->prev_priority = priority; | |
2733c06a | 868 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
869 | } |
870 | ||
871 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
872 | { | |
2733c06a | 873 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 874 | mem->prev_priority = priority; |
2733c06a | 875 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
876 | } |
877 | ||
c772be93 | 878 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
879 | { |
880 | unsigned long active; | |
881 | unsigned long inactive; | |
c772be93 KM |
882 | unsigned long gb; |
883 | unsigned long inactive_ratio; | |
14797e23 | 884 | |
14067bb3 KH |
885 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
886 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 887 | |
c772be93 KM |
888 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
889 | if (gb) | |
890 | inactive_ratio = int_sqrt(10 * gb); | |
891 | else | |
892 | inactive_ratio = 1; | |
893 | ||
894 | if (present_pages) { | |
895 | present_pages[0] = inactive; | |
896 | present_pages[1] = active; | |
897 | } | |
898 | ||
899 | return inactive_ratio; | |
900 | } | |
901 | ||
902 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
903 | { | |
904 | unsigned long active; | |
905 | unsigned long inactive; | |
906 | unsigned long present_pages[2]; | |
907 | unsigned long inactive_ratio; | |
908 | ||
909 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
910 | ||
911 | inactive = present_pages[0]; | |
912 | active = present_pages[1]; | |
913 | ||
914 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
915 | return 1; |
916 | ||
917 | return 0; | |
918 | } | |
919 | ||
56e49d21 RR |
920 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
921 | { | |
922 | unsigned long active; | |
923 | unsigned long inactive; | |
924 | ||
925 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
926 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
927 | ||
928 | return (active > inactive); | |
929 | } | |
930 | ||
a3d8e054 KM |
931 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
932 | struct zone *zone, | |
933 | enum lru_list lru) | |
934 | { | |
935 | int nid = zone->zone_pgdat->node_id; | |
936 | int zid = zone_idx(zone); | |
937 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
938 | ||
939 | return MEM_CGROUP_ZSTAT(mz, lru); | |
940 | } | |
941 | ||
3e2f41f1 KM |
942 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
943 | struct zone *zone) | |
944 | { | |
945 | int nid = zone->zone_pgdat->node_id; | |
946 | int zid = zone_idx(zone); | |
947 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
948 | ||
949 | return &mz->reclaim_stat; | |
950 | } | |
951 | ||
952 | struct zone_reclaim_stat * | |
953 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
954 | { | |
955 | struct page_cgroup *pc; | |
956 | struct mem_cgroup_per_zone *mz; | |
957 | ||
958 | if (mem_cgroup_disabled()) | |
959 | return NULL; | |
960 | ||
961 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
962 | /* |
963 | * Used bit is set without atomic ops but after smp_wmb(). | |
964 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
965 | */ | |
966 | smp_rmb(); | |
967 | if (!PageCgroupUsed(pc)) | |
968 | return NULL; | |
969 | ||
3e2f41f1 KM |
970 | mz = page_cgroup_zoneinfo(pc); |
971 | if (!mz) | |
972 | return NULL; | |
973 | ||
974 | return &mz->reclaim_stat; | |
975 | } | |
976 | ||
66e1707b BS |
977 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
978 | struct list_head *dst, | |
979 | unsigned long *scanned, int order, | |
980 | int mode, struct zone *z, | |
981 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 982 | int active, int file) |
66e1707b BS |
983 | { |
984 | unsigned long nr_taken = 0; | |
985 | struct page *page; | |
986 | unsigned long scan; | |
987 | LIST_HEAD(pc_list); | |
988 | struct list_head *src; | |
ff7283fa | 989 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
990 | int nid = z->zone_pgdat->node_id; |
991 | int zid = zone_idx(z); | |
992 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 993 | int lru = LRU_FILE * file + active; |
2ffebca6 | 994 | int ret; |
66e1707b | 995 | |
cf475ad2 | 996 | BUG_ON(!mem_cont); |
1ecaab2b | 997 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 998 | src = &mz->lists[lru]; |
66e1707b | 999 | |
ff7283fa KH |
1000 | scan = 0; |
1001 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 1002 | if (scan >= nr_to_scan) |
ff7283fa | 1003 | break; |
08e552c6 KH |
1004 | |
1005 | page = pc->page; | |
52d4b9ac KH |
1006 | if (unlikely(!PageCgroupUsed(pc))) |
1007 | continue; | |
436c6541 | 1008 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1009 | continue; |
ff7283fa | 1010 | |
436c6541 | 1011 | scan++; |
2ffebca6 KH |
1012 | ret = __isolate_lru_page(page, mode, file); |
1013 | switch (ret) { | |
1014 | case 0: | |
66e1707b | 1015 | list_move(&page->lru, dst); |
2ffebca6 | 1016 | mem_cgroup_del_lru(page); |
66e1707b | 1017 | nr_taken++; |
2ffebca6 KH |
1018 | break; |
1019 | case -EBUSY: | |
1020 | /* we don't affect global LRU but rotate in our LRU */ | |
1021 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1022 | break; | |
1023 | default: | |
1024 | break; | |
66e1707b BS |
1025 | } |
1026 | } | |
1027 | ||
66e1707b BS |
1028 | *scanned = scan; |
1029 | return nr_taken; | |
1030 | } | |
1031 | ||
6d61ef40 BS |
1032 | #define mem_cgroup_from_res_counter(counter, member) \ |
1033 | container_of(counter, struct mem_cgroup, member) | |
1034 | ||
b85a96c0 DN |
1035 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
1036 | { | |
1037 | if (do_swap_account) { | |
1038 | if (res_counter_check_under_limit(&mem->res) && | |
1039 | res_counter_check_under_limit(&mem->memsw)) | |
1040 | return true; | |
1041 | } else | |
1042 | if (res_counter_check_under_limit(&mem->res)) | |
1043 | return true; | |
1044 | return false; | |
1045 | } | |
1046 | ||
a7885eb8 KM |
1047 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1048 | { | |
1049 | struct cgroup *cgrp = memcg->css.cgroup; | |
1050 | unsigned int swappiness; | |
1051 | ||
1052 | /* root ? */ | |
1053 | if (cgrp->parent == NULL) | |
1054 | return vm_swappiness; | |
1055 | ||
1056 | spin_lock(&memcg->reclaim_param_lock); | |
1057 | swappiness = memcg->swappiness; | |
1058 | spin_unlock(&memcg->reclaim_param_lock); | |
1059 | ||
1060 | return swappiness; | |
1061 | } | |
1062 | ||
81d39c20 KH |
1063 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
1064 | { | |
1065 | int *val = data; | |
1066 | (*val)++; | |
1067 | return 0; | |
1068 | } | |
e222432b BS |
1069 | |
1070 | /** | |
6a6135b6 | 1071 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1072 | * @memcg: The memory cgroup that went over limit |
1073 | * @p: Task that is going to be killed | |
1074 | * | |
1075 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1076 | * enabled | |
1077 | */ | |
1078 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1079 | { | |
1080 | struct cgroup *task_cgrp; | |
1081 | struct cgroup *mem_cgrp; | |
1082 | /* | |
1083 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1084 | * on the assumption that OOM is serialized for memory controller. | |
1085 | * If this assumption is broken, revisit this code. | |
1086 | */ | |
1087 | static char memcg_name[PATH_MAX]; | |
1088 | int ret; | |
1089 | ||
d31f56db | 1090 | if (!memcg || !p) |
e222432b BS |
1091 | return; |
1092 | ||
1093 | ||
1094 | rcu_read_lock(); | |
1095 | ||
1096 | mem_cgrp = memcg->css.cgroup; | |
1097 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1098 | ||
1099 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1100 | if (ret < 0) { | |
1101 | /* | |
1102 | * Unfortunately, we are unable to convert to a useful name | |
1103 | * But we'll still print out the usage information | |
1104 | */ | |
1105 | rcu_read_unlock(); | |
1106 | goto done; | |
1107 | } | |
1108 | rcu_read_unlock(); | |
1109 | ||
1110 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1111 | ||
1112 | rcu_read_lock(); | |
1113 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1114 | if (ret < 0) { | |
1115 | rcu_read_unlock(); | |
1116 | goto done; | |
1117 | } | |
1118 | rcu_read_unlock(); | |
1119 | ||
1120 | /* | |
1121 | * Continues from above, so we don't need an KERN_ level | |
1122 | */ | |
1123 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1124 | done: | |
1125 | ||
1126 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1127 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1128 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1129 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1130 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1131 | "failcnt %llu\n", | |
1132 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1133 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1134 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1135 | } | |
1136 | ||
81d39c20 KH |
1137 | /* |
1138 | * This function returns the number of memcg under hierarchy tree. Returns | |
1139 | * 1(self count) if no children. | |
1140 | */ | |
1141 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1142 | { | |
1143 | int num = 0; | |
1144 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
1145 | return num; | |
1146 | } | |
1147 | ||
6d61ef40 | 1148 | /* |
04046e1a KH |
1149 | * Visit the first child (need not be the first child as per the ordering |
1150 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1151 | * that to reclaim free pages from. | |
1152 | */ | |
1153 | static struct mem_cgroup * | |
1154 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1155 | { | |
1156 | struct mem_cgroup *ret = NULL; | |
1157 | struct cgroup_subsys_state *css; | |
1158 | int nextid, found; | |
1159 | ||
1160 | if (!root_mem->use_hierarchy) { | |
1161 | css_get(&root_mem->css); | |
1162 | ret = root_mem; | |
1163 | } | |
1164 | ||
1165 | while (!ret) { | |
1166 | rcu_read_lock(); | |
1167 | nextid = root_mem->last_scanned_child + 1; | |
1168 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1169 | &found); | |
1170 | if (css && css_tryget(css)) | |
1171 | ret = container_of(css, struct mem_cgroup, css); | |
1172 | ||
1173 | rcu_read_unlock(); | |
1174 | /* Updates scanning parameter */ | |
1175 | spin_lock(&root_mem->reclaim_param_lock); | |
1176 | if (!css) { | |
1177 | /* this means start scan from ID:1 */ | |
1178 | root_mem->last_scanned_child = 0; | |
1179 | } else | |
1180 | root_mem->last_scanned_child = found; | |
1181 | spin_unlock(&root_mem->reclaim_param_lock); | |
1182 | } | |
1183 | ||
1184 | return ret; | |
1185 | } | |
1186 | ||
1187 | /* | |
1188 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1189 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1190 | * based on its position in the children list. | |
6d61ef40 BS |
1191 | * |
1192 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1193 | * |
1194 | * We give up and return to the caller when we visit root_mem twice. | |
1195 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1196 | * |
1197 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1198 | */ |
1199 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1200 | struct zone *zone, |
75822b44 BS |
1201 | gfp_t gfp_mask, |
1202 | unsigned long reclaim_options) | |
6d61ef40 | 1203 | { |
04046e1a KH |
1204 | struct mem_cgroup *victim; |
1205 | int ret, total = 0; | |
1206 | int loop = 0; | |
75822b44 BS |
1207 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1208 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 BS |
1209 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
1210 | unsigned long excess = mem_cgroup_get_excess(root_mem); | |
04046e1a | 1211 | |
22a668d7 KH |
1212 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1213 | if (root_mem->memsw_is_minimum) | |
1214 | noswap = true; | |
1215 | ||
4e416953 | 1216 | while (1) { |
04046e1a | 1217 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1218 | if (victim == root_mem) { |
04046e1a | 1219 | loop++; |
cdec2e42 KH |
1220 | if (loop >= 1) |
1221 | drain_all_stock_async(); | |
4e416953 BS |
1222 | if (loop >= 2) { |
1223 | /* | |
1224 | * If we have not been able to reclaim | |
1225 | * anything, it might because there are | |
1226 | * no reclaimable pages under this hierarchy | |
1227 | */ | |
1228 | if (!check_soft || !total) { | |
1229 | css_put(&victim->css); | |
1230 | break; | |
1231 | } | |
1232 | /* | |
1233 | * We want to do more targetted reclaim. | |
1234 | * excess >> 2 is not to excessive so as to | |
1235 | * reclaim too much, nor too less that we keep | |
1236 | * coming back to reclaim from this cgroup | |
1237 | */ | |
1238 | if (total >= (excess >> 2) || | |
1239 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1240 | css_put(&victim->css); | |
1241 | break; | |
1242 | } | |
1243 | } | |
1244 | } | |
c62b1a3b | 1245 | if (!mem_cgroup_local_usage(victim)) { |
04046e1a KH |
1246 | /* this cgroup's local usage == 0 */ |
1247 | css_put(&victim->css); | |
6d61ef40 BS |
1248 | continue; |
1249 | } | |
04046e1a | 1250 | /* we use swappiness of local cgroup */ |
4e416953 BS |
1251 | if (check_soft) |
1252 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
1253 | noswap, get_swappiness(victim), zone, | |
1254 | zone->zone_pgdat->node_id); | |
1255 | else | |
1256 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1257 | noswap, get_swappiness(victim)); | |
04046e1a | 1258 | css_put(&victim->css); |
81d39c20 KH |
1259 | /* |
1260 | * At shrinking usage, we can't check we should stop here or | |
1261 | * reclaim more. It's depends on callers. last_scanned_child | |
1262 | * will work enough for keeping fairness under tree. | |
1263 | */ | |
1264 | if (shrink) | |
1265 | return ret; | |
04046e1a | 1266 | total += ret; |
4e416953 BS |
1267 | if (check_soft) { |
1268 | if (res_counter_check_under_soft_limit(&root_mem->res)) | |
1269 | return total; | |
1270 | } else if (mem_cgroup_check_under_limit(root_mem)) | |
04046e1a | 1271 | return 1 + total; |
6d61ef40 | 1272 | } |
04046e1a | 1273 | return total; |
6d61ef40 BS |
1274 | } |
1275 | ||
867578cb | 1276 | static int mem_cgroup_oom_lock_cb(struct mem_cgroup *mem, void *data) |
a636b327 | 1277 | { |
867578cb KH |
1278 | int *val = (int *)data; |
1279 | int x; | |
1280 | /* | |
1281 | * Logically, we can stop scanning immediately when we find | |
1282 | * a memcg is already locked. But condidering unlock ops and | |
1283 | * creation/removal of memcg, scan-all is simple operation. | |
1284 | */ | |
1285 | x = atomic_inc_return(&mem->oom_lock); | |
1286 | *val = max(x, *val); | |
1287 | return 0; | |
1288 | } | |
1289 | /* | |
1290 | * Check OOM-Killer is already running under our hierarchy. | |
1291 | * If someone is running, return false. | |
1292 | */ | |
1293 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) | |
1294 | { | |
1295 | int lock_count = 0; | |
a636b327 | 1296 | |
867578cb KH |
1297 | mem_cgroup_walk_tree(mem, &lock_count, mem_cgroup_oom_lock_cb); |
1298 | ||
1299 | if (lock_count == 1) | |
1300 | return true; | |
1301 | return false; | |
a636b327 | 1302 | } |
0b7f569e | 1303 | |
867578cb | 1304 | static int mem_cgroup_oom_unlock_cb(struct mem_cgroup *mem, void *data) |
0b7f569e | 1305 | { |
867578cb KH |
1306 | /* |
1307 | * When a new child is created while the hierarchy is under oom, | |
1308 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1309 | * atomic_add_unless() here. | |
1310 | */ | |
1311 | atomic_add_unless(&mem->oom_lock, -1, 0); | |
0b7f569e KH |
1312 | return 0; |
1313 | } | |
1314 | ||
867578cb KH |
1315 | static void mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
1316 | { | |
1317 | mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_unlock_cb); | |
1318 | } | |
1319 | ||
1320 | static DEFINE_MUTEX(memcg_oom_mutex); | |
1321 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); | |
1322 | ||
dc98df5a KH |
1323 | struct oom_wait_info { |
1324 | struct mem_cgroup *mem; | |
1325 | wait_queue_t wait; | |
1326 | }; | |
1327 | ||
1328 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1329 | unsigned mode, int sync, void *arg) | |
1330 | { | |
1331 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; | |
1332 | struct oom_wait_info *oom_wait_info; | |
1333 | ||
1334 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
1335 | ||
1336 | if (oom_wait_info->mem == wake_mem) | |
1337 | goto wakeup; | |
1338 | /* if no hierarchy, no match */ | |
1339 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) | |
1340 | return 0; | |
1341 | /* | |
1342 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1343 | * Then we can use css_is_ancestor without taking care of RCU. | |
1344 | */ | |
1345 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && | |
1346 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) | |
1347 | return 0; | |
1348 | ||
1349 | wakeup: | |
1350 | return autoremove_wake_function(wait, mode, sync, arg); | |
1351 | } | |
1352 | ||
1353 | static void memcg_wakeup_oom(struct mem_cgroup *mem) | |
1354 | { | |
1355 | /* for filtering, pass "mem" as argument. */ | |
1356 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); | |
1357 | } | |
1358 | ||
3c11ecf4 KH |
1359 | static void memcg_oom_recover(struct mem_cgroup *mem) |
1360 | { | |
1361 | if (mem->oom_kill_disable && atomic_read(&mem->oom_lock)) | |
1362 | memcg_wakeup_oom(mem); | |
1363 | } | |
1364 | ||
867578cb KH |
1365 | /* |
1366 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1367 | */ | |
1368 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) | |
0b7f569e | 1369 | { |
dc98df5a | 1370 | struct oom_wait_info owait; |
3c11ecf4 | 1371 | bool locked, need_to_kill; |
867578cb | 1372 | |
dc98df5a KH |
1373 | owait.mem = mem; |
1374 | owait.wait.flags = 0; | |
1375 | owait.wait.func = memcg_oom_wake_function; | |
1376 | owait.wait.private = current; | |
1377 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1378 | need_to_kill = true; |
867578cb KH |
1379 | /* At first, try to OOM lock hierarchy under mem.*/ |
1380 | mutex_lock(&memcg_oom_mutex); | |
1381 | locked = mem_cgroup_oom_lock(mem); | |
1382 | /* | |
1383 | * Even if signal_pending(), we can't quit charge() loop without | |
1384 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1385 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1386 | */ | |
3c11ecf4 KH |
1387 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
1388 | if (!locked || mem->oom_kill_disable) | |
1389 | need_to_kill = false; | |
1390 | if (locked) | |
9490ff27 | 1391 | mem_cgroup_oom_notify(mem); |
867578cb KH |
1392 | mutex_unlock(&memcg_oom_mutex); |
1393 | ||
3c11ecf4 KH |
1394 | if (need_to_kill) { |
1395 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
867578cb | 1396 | mem_cgroup_out_of_memory(mem, mask); |
3c11ecf4 | 1397 | } else { |
867578cb | 1398 | schedule(); |
dc98df5a | 1399 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb KH |
1400 | } |
1401 | mutex_lock(&memcg_oom_mutex); | |
1402 | mem_cgroup_oom_unlock(mem); | |
dc98df5a | 1403 | memcg_wakeup_oom(mem); |
867578cb KH |
1404 | mutex_unlock(&memcg_oom_mutex); |
1405 | ||
1406 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) | |
1407 | return false; | |
1408 | /* Give chance to dying process */ | |
1409 | schedule_timeout(1); | |
1410 | return true; | |
0b7f569e KH |
1411 | } |
1412 | ||
d69b042f BS |
1413 | /* |
1414 | * Currently used to update mapped file statistics, but the routine can be | |
1415 | * generalized to update other statistics as well. | |
1416 | */ | |
d8046582 | 1417 | void mem_cgroup_update_file_mapped(struct page *page, int val) |
d69b042f BS |
1418 | { |
1419 | struct mem_cgroup *mem; | |
d69b042f BS |
1420 | struct page_cgroup *pc; |
1421 | ||
d69b042f BS |
1422 | pc = lookup_page_cgroup(page); |
1423 | if (unlikely(!pc)) | |
1424 | return; | |
1425 | ||
1426 | lock_page_cgroup(pc); | |
1427 | mem = pc->mem_cgroup; | |
8725d541 | 1428 | if (!mem || !PageCgroupUsed(pc)) |
d69b042f BS |
1429 | goto done; |
1430 | ||
1431 | /* | |
c62b1a3b | 1432 | * Preemption is already disabled. We can use __this_cpu_xxx |
d69b042f | 1433 | */ |
8725d541 KH |
1434 | if (val > 0) { |
1435 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1436 | SetPageCgroupFileMapped(pc); | |
1437 | } else { | |
1438 | __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1439 | ClearPageCgroupFileMapped(pc); | |
1440 | } | |
d69b042f | 1441 | |
d69b042f BS |
1442 | done: |
1443 | unlock_page_cgroup(pc); | |
1444 | } | |
0b7f569e | 1445 | |
cdec2e42 KH |
1446 | /* |
1447 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1448 | * TODO: maybe necessary to use big numbers in big irons. | |
1449 | */ | |
1450 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1451 | struct memcg_stock_pcp { | |
1452 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1453 | int charge; | |
1454 | struct work_struct work; | |
1455 | }; | |
1456 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1457 | static atomic_t memcg_drain_count; | |
1458 | ||
1459 | /* | |
1460 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1461 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1462 | * cgroup which is not current target, returns false. This stock will be | |
1463 | * refilled. | |
1464 | */ | |
1465 | static bool consume_stock(struct mem_cgroup *mem) | |
1466 | { | |
1467 | struct memcg_stock_pcp *stock; | |
1468 | bool ret = true; | |
1469 | ||
1470 | stock = &get_cpu_var(memcg_stock); | |
1471 | if (mem == stock->cached && stock->charge) | |
1472 | stock->charge -= PAGE_SIZE; | |
1473 | else /* need to call res_counter_charge */ | |
1474 | ret = false; | |
1475 | put_cpu_var(memcg_stock); | |
1476 | return ret; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1481 | */ | |
1482 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1483 | { | |
1484 | struct mem_cgroup *old = stock->cached; | |
1485 | ||
1486 | if (stock->charge) { | |
1487 | res_counter_uncharge(&old->res, stock->charge); | |
1488 | if (do_swap_account) | |
1489 | res_counter_uncharge(&old->memsw, stock->charge); | |
1490 | } | |
1491 | stock->cached = NULL; | |
1492 | stock->charge = 0; | |
1493 | } | |
1494 | ||
1495 | /* | |
1496 | * This must be called under preempt disabled or must be called by | |
1497 | * a thread which is pinned to local cpu. | |
1498 | */ | |
1499 | static void drain_local_stock(struct work_struct *dummy) | |
1500 | { | |
1501 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1502 | drain_stock(stock); | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 1507 | * This will be consumed by consume_stock() function, later. |
cdec2e42 KH |
1508 | */ |
1509 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1510 | { | |
1511 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1512 | ||
1513 | if (stock->cached != mem) { /* reset if necessary */ | |
1514 | drain_stock(stock); | |
1515 | stock->cached = mem; | |
1516 | } | |
1517 | stock->charge += val; | |
1518 | put_cpu_var(memcg_stock); | |
1519 | } | |
1520 | ||
1521 | /* | |
1522 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1523 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1524 | * expects some charges will be back to res_counter later but cannot wait for | |
1525 | * it. | |
1526 | */ | |
1527 | static void drain_all_stock_async(void) | |
1528 | { | |
1529 | int cpu; | |
1530 | /* This function is for scheduling "drain" in asynchronous way. | |
1531 | * The result of "drain" is not directly handled by callers. Then, | |
1532 | * if someone is calling drain, we don't have to call drain more. | |
1533 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1534 | * there is a race. We just do loose check here. | |
1535 | */ | |
1536 | if (atomic_read(&memcg_drain_count)) | |
1537 | return; | |
1538 | /* Notify other cpus that system-wide "drain" is running */ | |
1539 | atomic_inc(&memcg_drain_count); | |
1540 | get_online_cpus(); | |
1541 | for_each_online_cpu(cpu) { | |
1542 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1543 | schedule_work_on(cpu, &stock->work); | |
1544 | } | |
1545 | put_online_cpus(); | |
1546 | atomic_dec(&memcg_drain_count); | |
1547 | /* We don't wait for flush_work */ | |
1548 | } | |
1549 | ||
1550 | /* This is a synchronous drain interface. */ | |
1551 | static void drain_all_stock_sync(void) | |
1552 | { | |
1553 | /* called when force_empty is called */ | |
1554 | atomic_inc(&memcg_drain_count); | |
1555 | schedule_on_each_cpu(drain_local_stock); | |
1556 | atomic_dec(&memcg_drain_count); | |
1557 | } | |
1558 | ||
1559 | static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb, | |
1560 | unsigned long action, | |
1561 | void *hcpu) | |
1562 | { | |
1563 | int cpu = (unsigned long)hcpu; | |
1564 | struct memcg_stock_pcp *stock; | |
1565 | ||
1566 | if (action != CPU_DEAD) | |
1567 | return NOTIFY_OK; | |
1568 | stock = &per_cpu(memcg_stock, cpu); | |
1569 | drain_stock(stock); | |
1570 | return NOTIFY_OK; | |
1571 | } | |
1572 | ||
f817ed48 KH |
1573 | /* |
1574 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1575 | * oom-killer can be invoked. | |
8a9f3ccd | 1576 | */ |
f817ed48 | 1577 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
430e4863 | 1578 | gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom) |
8a9f3ccd | 1579 | { |
4e649152 | 1580 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 1581 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
4e649152 | 1582 | struct res_counter *fail_res; |
cdec2e42 | 1583 | int csize = CHARGE_SIZE; |
a636b327 | 1584 | |
867578cb KH |
1585 | /* |
1586 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
1587 | * in system level. So, allow to go ahead dying process in addition to | |
1588 | * MEMDIE process. | |
1589 | */ | |
1590 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
1591 | || fatal_signal_pending(current))) | |
1592 | goto bypass; | |
a636b327 | 1593 | |
8a9f3ccd | 1594 | /* |
3be91277 HD |
1595 | * We always charge the cgroup the mm_struct belongs to. |
1596 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
1597 | * thread group leader migrates. It's possible that mm is not |
1598 | * set, if so charge the init_mm (happens for pagecache usage). | |
1599 | */ | |
54595fe2 KH |
1600 | mem = *memcg; |
1601 | if (likely(!mem)) { | |
1602 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 1603 | *memcg = mem; |
e8589cc1 | 1604 | } else { |
7a81b88c | 1605 | css_get(&mem->css); |
e8589cc1 | 1606 | } |
54595fe2 KH |
1607 | if (unlikely(!mem)) |
1608 | return 0; | |
1609 | ||
46f7e602 | 1610 | VM_BUG_ON(css_is_removed(&mem->css)); |
cdec2e42 KH |
1611 | if (mem_cgroup_is_root(mem)) |
1612 | goto done; | |
8a9f3ccd | 1613 | |
8c7c6e34 | 1614 | while (1) { |
0c3e73e8 | 1615 | int ret = 0; |
75822b44 | 1616 | unsigned long flags = 0; |
7a81b88c | 1617 | |
cdec2e42 | 1618 | if (consume_stock(mem)) |
430e4863 | 1619 | goto done; |
cdec2e42 KH |
1620 | |
1621 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
8c7c6e34 KH |
1622 | if (likely(!ret)) { |
1623 | if (!do_swap_account) | |
1624 | break; | |
cdec2e42 | 1625 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); |
8c7c6e34 KH |
1626 | if (likely(!ret)) |
1627 | break; | |
1628 | /* mem+swap counter fails */ | |
cdec2e42 | 1629 | res_counter_uncharge(&mem->res, csize); |
75822b44 | 1630 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
6d61ef40 BS |
1631 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1632 | memsw); | |
1633 | } else | |
1634 | /* mem counter fails */ | |
1635 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1636 | res); | |
1637 | ||
cdec2e42 KH |
1638 | /* reduce request size and retry */ |
1639 | if (csize > PAGE_SIZE) { | |
1640 | csize = PAGE_SIZE; | |
1641 | continue; | |
1642 | } | |
3be91277 | 1643 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 1644 | goto nomem; |
e1a1cd59 | 1645 | |
4e416953 BS |
1646 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
1647 | gfp_mask, flags); | |
4d1c6273 DN |
1648 | if (ret) |
1649 | continue; | |
66e1707b BS |
1650 | |
1651 | /* | |
8869b8f6 HD |
1652 | * try_to_free_mem_cgroup_pages() might not give us a full |
1653 | * picture of reclaim. Some pages are reclaimed and might be | |
1654 | * moved to swap cache or just unmapped from the cgroup. | |
1655 | * Check the limit again to see if the reclaim reduced the | |
1656 | * current usage of the cgroup before giving up | |
8c7c6e34 | 1657 | * |
8869b8f6 | 1658 | */ |
b85a96c0 DN |
1659 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
1660 | continue; | |
3be91277 | 1661 | |
8033b97c DN |
1662 | /* try to avoid oom while someone is moving charge */ |
1663 | if (mc.moving_task && current != mc.moving_task) { | |
1664 | struct mem_cgroup *from, *to; | |
1665 | bool do_continue = false; | |
1666 | /* | |
1667 | * There is a small race that "from" or "to" can be | |
1668 | * freed by rmdir, so we use css_tryget(). | |
1669 | */ | |
8033b97c DN |
1670 | from = mc.from; |
1671 | to = mc.to; | |
1672 | if (from && css_tryget(&from->css)) { | |
1673 | if (mem_over_limit->use_hierarchy) | |
1674 | do_continue = css_is_ancestor( | |
1675 | &from->css, | |
1676 | &mem_over_limit->css); | |
1677 | else | |
1678 | do_continue = (from == mem_over_limit); | |
1679 | css_put(&from->css); | |
1680 | } | |
1681 | if (!do_continue && to && css_tryget(&to->css)) { | |
1682 | if (mem_over_limit->use_hierarchy) | |
1683 | do_continue = css_is_ancestor( | |
1684 | &to->css, | |
1685 | &mem_over_limit->css); | |
1686 | else | |
1687 | do_continue = (to == mem_over_limit); | |
1688 | css_put(&to->css); | |
1689 | } | |
8033b97c DN |
1690 | if (do_continue) { |
1691 | DEFINE_WAIT(wait); | |
1692 | prepare_to_wait(&mc.waitq, &wait, | |
1693 | TASK_INTERRUPTIBLE); | |
1694 | /* moving charge context might have finished. */ | |
1695 | if (mc.moving_task) | |
1696 | schedule(); | |
1697 | finish_wait(&mc.waitq, &wait); | |
1698 | continue; | |
1699 | } | |
1700 | } | |
1701 | ||
3be91277 | 1702 | if (!nr_retries--) { |
867578cb KH |
1703 | if (!oom) |
1704 | goto nomem; | |
1705 | if (mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) { | |
1706 | nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
1707 | continue; | |
a636b327 | 1708 | } |
867578cb KH |
1709 | /* When we reach here, current task is dying .*/ |
1710 | css_put(&mem->css); | |
1711 | goto bypass; | |
66e1707b | 1712 | } |
8a9f3ccd | 1713 | } |
cdec2e42 KH |
1714 | if (csize > PAGE_SIZE) |
1715 | refill_stock(mem, csize - PAGE_SIZE); | |
0c3e73e8 | 1716 | done: |
7a81b88c KH |
1717 | return 0; |
1718 | nomem: | |
1719 | css_put(&mem->css); | |
1720 | return -ENOMEM; | |
867578cb KH |
1721 | bypass: |
1722 | *memcg = NULL; | |
1723 | return 0; | |
7a81b88c | 1724 | } |
8a9f3ccd | 1725 | |
a3032a2c DN |
1726 | /* |
1727 | * Somemtimes we have to undo a charge we got by try_charge(). | |
1728 | * This function is for that and do uncharge, put css's refcnt. | |
1729 | * gotten by try_charge(). | |
1730 | */ | |
854ffa8d DN |
1731 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
1732 | unsigned long count) | |
a3032a2c DN |
1733 | { |
1734 | if (!mem_cgroup_is_root(mem)) { | |
854ffa8d | 1735 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
a3032a2c | 1736 | if (do_swap_account) |
854ffa8d DN |
1737 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
1738 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
1739 | WARN_ON_ONCE(count > INT_MAX); | |
1740 | __css_put(&mem->css, (int)count); | |
a3032a2c | 1741 | } |
854ffa8d DN |
1742 | /* we don't need css_put for root */ |
1743 | } | |
1744 | ||
1745 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem) | |
1746 | { | |
1747 | __mem_cgroup_cancel_charge(mem, 1); | |
a3032a2c DN |
1748 | } |
1749 | ||
a3b2d692 KH |
1750 | /* |
1751 | * A helper function to get mem_cgroup from ID. must be called under | |
1752 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
1753 | * it's concern. (dropping refcnt from swap can be called against removed | |
1754 | * memcg.) | |
1755 | */ | |
1756 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
1757 | { | |
1758 | struct cgroup_subsys_state *css; | |
1759 | ||
1760 | /* ID 0 is unused ID */ | |
1761 | if (!id) | |
1762 | return NULL; | |
1763 | css = css_lookup(&mem_cgroup_subsys, id); | |
1764 | if (!css) | |
1765 | return NULL; | |
1766 | return container_of(css, struct mem_cgroup, css); | |
1767 | } | |
1768 | ||
e42d9d5d | 1769 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 1770 | { |
e42d9d5d | 1771 | struct mem_cgroup *mem = NULL; |
3c776e64 | 1772 | struct page_cgroup *pc; |
a3b2d692 | 1773 | unsigned short id; |
b5a84319 KH |
1774 | swp_entry_t ent; |
1775 | ||
3c776e64 DN |
1776 | VM_BUG_ON(!PageLocked(page)); |
1777 | ||
3c776e64 | 1778 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 1779 | lock_page_cgroup(pc); |
a3b2d692 | 1780 | if (PageCgroupUsed(pc)) { |
3c776e64 | 1781 | mem = pc->mem_cgroup; |
a3b2d692 KH |
1782 | if (mem && !css_tryget(&mem->css)) |
1783 | mem = NULL; | |
e42d9d5d | 1784 | } else if (PageSwapCache(page)) { |
3c776e64 | 1785 | ent.val = page_private(page); |
a3b2d692 KH |
1786 | id = lookup_swap_cgroup(ent); |
1787 | rcu_read_lock(); | |
1788 | mem = mem_cgroup_lookup(id); | |
1789 | if (mem && !css_tryget(&mem->css)) | |
1790 | mem = NULL; | |
1791 | rcu_read_unlock(); | |
3c776e64 | 1792 | } |
c0bd3f63 | 1793 | unlock_page_cgroup(pc); |
b5a84319 KH |
1794 | return mem; |
1795 | } | |
1796 | ||
7a81b88c | 1797 | /* |
a5e924f5 | 1798 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
1799 | * USED state. If already USED, uncharge and return. |
1800 | */ | |
1801 | ||
1802 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
1803 | struct page_cgroup *pc, | |
1804 | enum charge_type ctype) | |
1805 | { | |
7a81b88c KH |
1806 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
1807 | if (!mem) | |
1808 | return; | |
52d4b9ac KH |
1809 | |
1810 | lock_page_cgroup(pc); | |
1811 | if (unlikely(PageCgroupUsed(pc))) { | |
1812 | unlock_page_cgroup(pc); | |
a3032a2c | 1813 | mem_cgroup_cancel_charge(mem); |
7a81b88c | 1814 | return; |
52d4b9ac | 1815 | } |
4b3bde4c | 1816 | |
8a9f3ccd | 1817 | pc->mem_cgroup = mem; |
261fb61a KH |
1818 | /* |
1819 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
1820 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
1821 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
1822 | * before USED bit, we need memory barrier here. | |
1823 | * See mem_cgroup_add_lru_list(), etc. | |
1824 | */ | |
08e552c6 | 1825 | smp_wmb(); |
4b3bde4c BS |
1826 | switch (ctype) { |
1827 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
1828 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
1829 | SetPageCgroupCache(pc); | |
1830 | SetPageCgroupUsed(pc); | |
1831 | break; | |
1832 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1833 | ClearPageCgroupCache(pc); | |
1834 | SetPageCgroupUsed(pc); | |
1835 | break; | |
1836 | default: | |
1837 | break; | |
1838 | } | |
3be91277 | 1839 | |
08e552c6 | 1840 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 1841 | |
52d4b9ac | 1842 | unlock_page_cgroup(pc); |
430e4863 KH |
1843 | /* |
1844 | * "charge_statistics" updated event counter. Then, check it. | |
1845 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
1846 | * if they exceeds softlimit. | |
1847 | */ | |
d2265e6f | 1848 | memcg_check_events(mem, pc->page); |
7a81b88c | 1849 | } |
66e1707b | 1850 | |
f817ed48 | 1851 | /** |
57f9fd7d | 1852 | * __mem_cgroup_move_account - move account of the page |
f817ed48 KH |
1853 | * @pc: page_cgroup of the page. |
1854 | * @from: mem_cgroup which the page is moved from. | |
1855 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 1856 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
1857 | * |
1858 | * The caller must confirm following. | |
08e552c6 | 1859 | * - page is not on LRU (isolate_page() is useful.) |
57f9fd7d | 1860 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
f817ed48 | 1861 | * |
854ffa8d DN |
1862 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
1863 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is | |
1864 | * true, this function does "uncharge" from old cgroup, but it doesn't if | |
1865 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 KH |
1866 | */ |
1867 | ||
57f9fd7d | 1868 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
854ffa8d | 1869 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
f817ed48 | 1870 | { |
f817ed48 | 1871 | VM_BUG_ON(from == to); |
08e552c6 | 1872 | VM_BUG_ON(PageLRU(pc->page)); |
57f9fd7d DN |
1873 | VM_BUG_ON(!PageCgroupLocked(pc)); |
1874 | VM_BUG_ON(!PageCgroupUsed(pc)); | |
1875 | VM_BUG_ON(pc->mem_cgroup != from); | |
f817ed48 | 1876 | |
8725d541 | 1877 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
1878 | /* Update mapped_file data for mem_cgroup */ |
1879 | preempt_disable(); | |
1880 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1881 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1882 | preempt_enable(); | |
d69b042f | 1883 | } |
854ffa8d DN |
1884 | mem_cgroup_charge_statistics(from, pc, false); |
1885 | if (uncharge) | |
1886 | /* This is not "cancel", but cancel_charge does all we need. */ | |
1887 | mem_cgroup_cancel_charge(from); | |
d69b042f | 1888 | |
854ffa8d | 1889 | /* caller should have done css_get */ |
08e552c6 KH |
1890 | pc->mem_cgroup = to; |
1891 | mem_cgroup_charge_statistics(to, pc, true); | |
88703267 KH |
1892 | /* |
1893 | * We charges against "to" which may not have any tasks. Then, "to" | |
1894 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe DN |
1895 | * this function is just force_empty() and move charge, so it's |
1896 | * garanteed that "to" is never removed. So, we don't check rmdir | |
1897 | * status here. | |
88703267 | 1898 | */ |
57f9fd7d DN |
1899 | } |
1900 | ||
1901 | /* | |
1902 | * check whether the @pc is valid for moving account and call | |
1903 | * __mem_cgroup_move_account() | |
1904 | */ | |
1905 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
854ffa8d | 1906 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
57f9fd7d DN |
1907 | { |
1908 | int ret = -EINVAL; | |
1909 | lock_page_cgroup(pc); | |
1910 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { | |
854ffa8d | 1911 | __mem_cgroup_move_account(pc, from, to, uncharge); |
57f9fd7d DN |
1912 | ret = 0; |
1913 | } | |
1914 | unlock_page_cgroup(pc); | |
d2265e6f KH |
1915 | /* |
1916 | * check events | |
1917 | */ | |
1918 | memcg_check_events(to, pc->page); | |
1919 | memcg_check_events(from, pc->page); | |
f817ed48 KH |
1920 | return ret; |
1921 | } | |
1922 | ||
1923 | /* | |
1924 | * move charges to its parent. | |
1925 | */ | |
1926 | ||
1927 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1928 | struct mem_cgroup *child, | |
1929 | gfp_t gfp_mask) | |
1930 | { | |
08e552c6 | 1931 | struct page *page = pc->page; |
f817ed48 KH |
1932 | struct cgroup *cg = child->css.cgroup; |
1933 | struct cgroup *pcg = cg->parent; | |
1934 | struct mem_cgroup *parent; | |
f817ed48 KH |
1935 | int ret; |
1936 | ||
1937 | /* Is ROOT ? */ | |
1938 | if (!pcg) | |
1939 | return -EINVAL; | |
1940 | ||
57f9fd7d DN |
1941 | ret = -EBUSY; |
1942 | if (!get_page_unless_zero(page)) | |
1943 | goto out; | |
1944 | if (isolate_lru_page(page)) | |
1945 | goto put; | |
08e552c6 | 1946 | |
f817ed48 | 1947 | parent = mem_cgroup_from_cont(pcg); |
430e4863 | 1948 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 1949 | if (ret || !parent) |
57f9fd7d | 1950 | goto put_back; |
f817ed48 | 1951 | |
854ffa8d DN |
1952 | ret = mem_cgroup_move_account(pc, child, parent, true); |
1953 | if (ret) | |
1954 | mem_cgroup_cancel_charge(parent); | |
57f9fd7d | 1955 | put_back: |
08e552c6 | 1956 | putback_lru_page(page); |
57f9fd7d | 1957 | put: |
40d58138 | 1958 | put_page(page); |
57f9fd7d | 1959 | out: |
f817ed48 KH |
1960 | return ret; |
1961 | } | |
1962 | ||
7a81b88c KH |
1963 | /* |
1964 | * Charge the memory controller for page usage. | |
1965 | * Return | |
1966 | * 0 if the charge was successful | |
1967 | * < 0 if the cgroup is over its limit | |
1968 | */ | |
1969 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1970 | gfp_t gfp_mask, enum charge_type ctype, | |
1971 | struct mem_cgroup *memcg) | |
1972 | { | |
1973 | struct mem_cgroup *mem; | |
1974 | struct page_cgroup *pc; | |
1975 | int ret; | |
1976 | ||
1977 | pc = lookup_page_cgroup(page); | |
1978 | /* can happen at boot */ | |
1979 | if (unlikely(!pc)) | |
1980 | return 0; | |
1981 | prefetchw(pc); | |
1982 | ||
1983 | mem = memcg; | |
430e4863 | 1984 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 1985 | if (ret || !mem) |
7a81b88c KH |
1986 | return ret; |
1987 | ||
1988 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1989 | return 0; |
8a9f3ccd BS |
1990 | } |
1991 | ||
7a81b88c KH |
1992 | int mem_cgroup_newpage_charge(struct page *page, |
1993 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1994 | { |
f8d66542 | 1995 | if (mem_cgroup_disabled()) |
cede86ac | 1996 | return 0; |
52d4b9ac KH |
1997 | if (PageCompound(page)) |
1998 | return 0; | |
69029cd5 KH |
1999 | /* |
2000 | * If already mapped, we don't have to account. | |
2001 | * If page cache, page->mapping has address_space. | |
2002 | * But page->mapping may have out-of-use anon_vma pointer, | |
2003 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
2004 | * is NULL. | |
2005 | */ | |
2006 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2007 | return 0; | |
2008 | if (unlikely(!mm)) | |
2009 | mm = &init_mm; | |
217bc319 | 2010 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 2011 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
2012 | } |
2013 | ||
83aae4c7 DN |
2014 | static void |
2015 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2016 | enum charge_type ctype); | |
2017 | ||
e1a1cd59 BS |
2018 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2019 | gfp_t gfp_mask) | |
8697d331 | 2020 | { |
b5a84319 KH |
2021 | struct mem_cgroup *mem = NULL; |
2022 | int ret; | |
2023 | ||
f8d66542 | 2024 | if (mem_cgroup_disabled()) |
cede86ac | 2025 | return 0; |
52d4b9ac KH |
2026 | if (PageCompound(page)) |
2027 | return 0; | |
accf163e KH |
2028 | /* |
2029 | * Corner case handling. This is called from add_to_page_cache() | |
2030 | * in usual. But some FS (shmem) precharges this page before calling it | |
2031 | * and call add_to_page_cache() with GFP_NOWAIT. | |
2032 | * | |
2033 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
2034 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
2035 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
2036 | * And when the page is SwapCache, it should take swap information |
2037 | * into account. This is under lock_page() now. | |
accf163e KH |
2038 | */ |
2039 | if (!(gfp_mask & __GFP_WAIT)) { | |
2040 | struct page_cgroup *pc; | |
2041 | ||
52d4b9ac KH |
2042 | |
2043 | pc = lookup_page_cgroup(page); | |
2044 | if (!pc) | |
2045 | return 0; | |
2046 | lock_page_cgroup(pc); | |
2047 | if (PageCgroupUsed(pc)) { | |
2048 | unlock_page_cgroup(pc); | |
accf163e KH |
2049 | return 0; |
2050 | } | |
52d4b9ac | 2051 | unlock_page_cgroup(pc); |
accf163e KH |
2052 | } |
2053 | ||
b5a84319 | 2054 | if (unlikely(!mm && !mem)) |
8697d331 | 2055 | mm = &init_mm; |
accf163e | 2056 | |
c05555b5 KH |
2057 | if (page_is_file_cache(page)) |
2058 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 2059 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 | 2060 | |
83aae4c7 DN |
2061 | /* shmem */ |
2062 | if (PageSwapCache(page)) { | |
2063 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
2064 | if (!ret) | |
2065 | __mem_cgroup_commit_charge_swapin(page, mem, | |
2066 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
2067 | } else | |
2068 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
2069 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
b5a84319 | 2070 | |
b5a84319 | 2071 | return ret; |
e8589cc1 KH |
2072 | } |
2073 | ||
54595fe2 KH |
2074 | /* |
2075 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2076 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2077 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2078 | * "commit()" or removed by "cancel()" |
2079 | */ | |
8c7c6e34 KH |
2080 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2081 | struct page *page, | |
2082 | gfp_t mask, struct mem_cgroup **ptr) | |
2083 | { | |
2084 | struct mem_cgroup *mem; | |
54595fe2 | 2085 | int ret; |
8c7c6e34 | 2086 | |
f8d66542 | 2087 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2088 | return 0; |
2089 | ||
2090 | if (!do_swap_account) | |
2091 | goto charge_cur_mm; | |
8c7c6e34 KH |
2092 | /* |
2093 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2094 | * the pte, and even removed page from swap cache: in those cases |
2095 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2096 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2097 | */ |
2098 | if (!PageSwapCache(page)) | |
407f9c8b | 2099 | goto charge_cur_mm; |
e42d9d5d | 2100 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
2101 | if (!mem) |
2102 | goto charge_cur_mm; | |
8c7c6e34 | 2103 | *ptr = mem; |
430e4863 | 2104 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); |
54595fe2 KH |
2105 | /* drop extra refcnt from tryget */ |
2106 | css_put(&mem->css); | |
2107 | return ret; | |
8c7c6e34 KH |
2108 | charge_cur_mm: |
2109 | if (unlikely(!mm)) | |
2110 | mm = &init_mm; | |
430e4863 | 2111 | return __mem_cgroup_try_charge(mm, mask, ptr, true); |
8c7c6e34 KH |
2112 | } |
2113 | ||
83aae4c7 DN |
2114 | static void |
2115 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2116 | enum charge_type ctype) | |
7a81b88c KH |
2117 | { |
2118 | struct page_cgroup *pc; | |
2119 | ||
f8d66542 | 2120 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2121 | return; |
2122 | if (!ptr) | |
2123 | return; | |
88703267 | 2124 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 2125 | pc = lookup_page_cgroup(page); |
544122e5 | 2126 | mem_cgroup_lru_del_before_commit_swapcache(page); |
83aae4c7 | 2127 | __mem_cgroup_commit_charge(ptr, pc, ctype); |
544122e5 | 2128 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
2129 | /* |
2130 | * Now swap is on-memory. This means this page may be | |
2131 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2132 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2133 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2134 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2135 | */ |
03f3c433 | 2136 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2137 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2138 | unsigned short id; |
8c7c6e34 | 2139 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2140 | |
2141 | id = swap_cgroup_record(ent, 0); | |
2142 | rcu_read_lock(); | |
2143 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2144 | if (memcg) { |
a3b2d692 KH |
2145 | /* |
2146 | * This recorded memcg can be obsolete one. So, avoid | |
2147 | * calling css_tryget | |
2148 | */ | |
0c3e73e8 | 2149 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2150 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2151 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2152 | mem_cgroup_put(memcg); |
2153 | } | |
a3b2d692 | 2154 | rcu_read_unlock(); |
8c7c6e34 | 2155 | } |
88703267 KH |
2156 | /* |
2157 | * At swapin, we may charge account against cgroup which has no tasks. | |
2158 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2159 | * In that case, we need to call pre_destroy() again. check it here. | |
2160 | */ | |
2161 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2162 | } |
2163 | ||
83aae4c7 DN |
2164 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2165 | { | |
2166 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2167 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2168 | } | |
2169 | ||
7a81b88c KH |
2170 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2171 | { | |
f8d66542 | 2172 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2173 | return; |
2174 | if (!mem) | |
2175 | return; | |
a3032a2c | 2176 | mem_cgroup_cancel_charge(mem); |
7a81b88c KH |
2177 | } |
2178 | ||
569b846d KH |
2179 | static void |
2180 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype) | |
2181 | { | |
2182 | struct memcg_batch_info *batch = NULL; | |
2183 | bool uncharge_memsw = true; | |
2184 | /* If swapout, usage of swap doesn't decrease */ | |
2185 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2186 | uncharge_memsw = false; | |
569b846d KH |
2187 | |
2188 | batch = ¤t->memcg_batch; | |
2189 | /* | |
2190 | * In usual, we do css_get() when we remember memcg pointer. | |
2191 | * But in this case, we keep res->usage until end of a series of | |
2192 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2193 | */ | |
2194 | if (!batch->memcg) | |
2195 | batch->memcg = mem; | |
3c11ecf4 KH |
2196 | /* |
2197 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
2198 | * In those cases, all pages freed continously can be expected to be in | |
2199 | * the same cgroup and we have chance to coalesce uncharges. | |
2200 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2201 | * because we want to do uncharge as soon as possible. | |
2202 | */ | |
2203 | ||
2204 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2205 | goto direct_uncharge; | |
2206 | ||
569b846d KH |
2207 | /* |
2208 | * In typical case, batch->memcg == mem. This means we can | |
2209 | * merge a series of uncharges to an uncharge of res_counter. | |
2210 | * If not, we uncharge res_counter ony by one. | |
2211 | */ | |
2212 | if (batch->memcg != mem) | |
2213 | goto direct_uncharge; | |
2214 | /* remember freed charge and uncharge it later */ | |
2215 | batch->bytes += PAGE_SIZE; | |
2216 | if (uncharge_memsw) | |
2217 | batch->memsw_bytes += PAGE_SIZE; | |
2218 | return; | |
2219 | direct_uncharge: | |
2220 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
2221 | if (uncharge_memsw) | |
2222 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
3c11ecf4 KH |
2223 | if (unlikely(batch->memcg != mem)) |
2224 | memcg_oom_recover(mem); | |
569b846d KH |
2225 | return; |
2226 | } | |
7a81b88c | 2227 | |
8a9f3ccd | 2228 | /* |
69029cd5 | 2229 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2230 | */ |
8c7c6e34 | 2231 | static struct mem_cgroup * |
69029cd5 | 2232 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2233 | { |
8289546e | 2234 | struct page_cgroup *pc; |
8c7c6e34 | 2235 | struct mem_cgroup *mem = NULL; |
072c56c1 | 2236 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 2237 | |
f8d66542 | 2238 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2239 | return NULL; |
4077960e | 2240 | |
d13d1443 | 2241 | if (PageSwapCache(page)) |
8c7c6e34 | 2242 | return NULL; |
d13d1443 | 2243 | |
8697d331 | 2244 | /* |
3c541e14 | 2245 | * Check if our page_cgroup is valid |
8697d331 | 2246 | */ |
52d4b9ac KH |
2247 | pc = lookup_page_cgroup(page); |
2248 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2249 | return NULL; |
b9c565d5 | 2250 | |
52d4b9ac | 2251 | lock_page_cgroup(pc); |
d13d1443 | 2252 | |
8c7c6e34 KH |
2253 | mem = pc->mem_cgroup; |
2254 | ||
d13d1443 KH |
2255 | if (!PageCgroupUsed(pc)) |
2256 | goto unlock_out; | |
2257 | ||
2258 | switch (ctype) { | |
2259 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2260 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2261 | /* See mem_cgroup_prepare_migration() */ |
2262 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2263 | goto unlock_out; |
2264 | break; | |
2265 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2266 | if (!PageAnon(page)) { /* Shared memory */ | |
2267 | if (page->mapping && !page_is_file_cache(page)) | |
2268 | goto unlock_out; | |
2269 | } else if (page_mapped(page)) /* Anon */ | |
2270 | goto unlock_out; | |
2271 | break; | |
2272 | default: | |
2273 | break; | |
52d4b9ac | 2274 | } |
d13d1443 | 2275 | |
569b846d KH |
2276 | if (!mem_cgroup_is_root(mem)) |
2277 | __do_uncharge(mem, ctype); | |
0c3e73e8 BS |
2278 | if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
2279 | mem_cgroup_swap_statistics(mem, true); | |
08e552c6 | 2280 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 2281 | |
52d4b9ac | 2282 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2283 | /* |
2284 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2285 | * freed from LRU. This is safe because uncharged page is expected not | |
2286 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2287 | * special functions. | |
2288 | */ | |
b9c565d5 | 2289 | |
69029cd5 | 2290 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 2291 | unlock_page_cgroup(pc); |
fb59e9f1 | 2292 | |
d2265e6f | 2293 | memcg_check_events(mem, page); |
a7fe942e KH |
2294 | /* at swapout, this memcg will be accessed to record to swap */ |
2295 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2296 | css_put(&mem->css); | |
6d12e2d8 | 2297 | |
8c7c6e34 | 2298 | return mem; |
d13d1443 KH |
2299 | |
2300 | unlock_out: | |
2301 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2302 | return NULL; |
3c541e14 BS |
2303 | } |
2304 | ||
69029cd5 KH |
2305 | void mem_cgroup_uncharge_page(struct page *page) |
2306 | { | |
52d4b9ac KH |
2307 | /* early check. */ |
2308 | if (page_mapped(page)) | |
2309 | return; | |
2310 | if (page->mapping && !PageAnon(page)) | |
2311 | return; | |
69029cd5 KH |
2312 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2313 | } | |
2314 | ||
2315 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2316 | { | |
2317 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2318 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2319 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2320 | } | |
2321 | ||
569b846d KH |
2322 | /* |
2323 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2324 | * In that cases, pages are freed continuously and we can expect pages | |
2325 | * are in the same memcg. All these calls itself limits the number of | |
2326 | * pages freed at once, then uncharge_start/end() is called properly. | |
2327 | * This may be called prural(2) times in a context, | |
2328 | */ | |
2329 | ||
2330 | void mem_cgroup_uncharge_start(void) | |
2331 | { | |
2332 | current->memcg_batch.do_batch++; | |
2333 | /* We can do nest. */ | |
2334 | if (current->memcg_batch.do_batch == 1) { | |
2335 | current->memcg_batch.memcg = NULL; | |
2336 | current->memcg_batch.bytes = 0; | |
2337 | current->memcg_batch.memsw_bytes = 0; | |
2338 | } | |
2339 | } | |
2340 | ||
2341 | void mem_cgroup_uncharge_end(void) | |
2342 | { | |
2343 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2344 | ||
2345 | if (!batch->do_batch) | |
2346 | return; | |
2347 | ||
2348 | batch->do_batch--; | |
2349 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2350 | return; | |
2351 | ||
2352 | if (!batch->memcg) | |
2353 | return; | |
2354 | /* | |
2355 | * This "batch->memcg" is valid without any css_get/put etc... | |
2356 | * bacause we hide charges behind us. | |
2357 | */ | |
2358 | if (batch->bytes) | |
2359 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2360 | if (batch->memsw_bytes) | |
2361 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
3c11ecf4 | 2362 | memcg_oom_recover(batch->memcg); |
569b846d KH |
2363 | /* forget this pointer (for sanity check) */ |
2364 | batch->memcg = NULL; | |
2365 | } | |
2366 | ||
e767e056 | 2367 | #ifdef CONFIG_SWAP |
8c7c6e34 | 2368 | /* |
e767e056 | 2369 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
2370 | * memcg information is recorded to swap_cgroup of "ent" |
2371 | */ | |
8a9478ca KH |
2372 | void |
2373 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
2374 | { |
2375 | struct mem_cgroup *memcg; | |
8a9478ca KH |
2376 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
2377 | ||
2378 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2379 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2380 | ||
2381 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 2382 | |
8c7c6e34 | 2383 | /* record memcg information */ |
8a9478ca | 2384 | if (do_swap_account && swapout && memcg) { |
a3b2d692 | 2385 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 KH |
2386 | mem_cgroup_get(memcg); |
2387 | } | |
8a9478ca | 2388 | if (swapout && memcg) |
a7fe942e | 2389 | css_put(&memcg->css); |
8c7c6e34 | 2390 | } |
e767e056 | 2391 | #endif |
8c7c6e34 KH |
2392 | |
2393 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2394 | /* | |
2395 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2396 | * uncharge "memsw" account. | |
2397 | */ | |
2398 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 2399 | { |
8c7c6e34 | 2400 | struct mem_cgroup *memcg; |
a3b2d692 | 2401 | unsigned short id; |
8c7c6e34 KH |
2402 | |
2403 | if (!do_swap_account) | |
2404 | return; | |
2405 | ||
a3b2d692 KH |
2406 | id = swap_cgroup_record(ent, 0); |
2407 | rcu_read_lock(); | |
2408 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2409 | if (memcg) { |
a3b2d692 KH |
2410 | /* |
2411 | * We uncharge this because swap is freed. | |
2412 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2413 | */ | |
0c3e73e8 | 2414 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2415 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2416 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2417 | mem_cgroup_put(memcg); |
2418 | } | |
a3b2d692 | 2419 | rcu_read_unlock(); |
d13d1443 | 2420 | } |
02491447 DN |
2421 | |
2422 | /** | |
2423 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2424 | * @entry: swap entry to be moved | |
2425 | * @from: mem_cgroup which the entry is moved from | |
2426 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 2427 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
2428 | * |
2429 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2430 | * as the mem_cgroup's id of @from. | |
2431 | * | |
2432 | * Returns 0 on success, -EINVAL on failure. | |
2433 | * | |
2434 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
2435 | * both res and memsw, and called css_get(). | |
2436 | */ | |
2437 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2438 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2439 | { |
2440 | unsigned short old_id, new_id; | |
2441 | ||
2442 | old_id = css_id(&from->css); | |
2443 | new_id = css_id(&to->css); | |
2444 | ||
2445 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2446 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2447 | mem_cgroup_swap_statistics(to, true); |
02491447 | 2448 | /* |
483c30b5 DN |
2449 | * This function is only called from task migration context now. |
2450 | * It postpones res_counter and refcount handling till the end | |
2451 | * of task migration(mem_cgroup_clear_mc()) for performance | |
2452 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
2453 | * because if the process that has been moved to @to does | |
2454 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 2455 | */ |
02491447 | 2456 | mem_cgroup_get(to); |
483c30b5 DN |
2457 | if (need_fixup) { |
2458 | if (!mem_cgroup_is_root(from)) | |
2459 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
2460 | mem_cgroup_put(from); | |
2461 | /* | |
2462 | * we charged both to->res and to->memsw, so we should | |
2463 | * uncharge to->res. | |
2464 | */ | |
2465 | if (!mem_cgroup_is_root(to)) | |
2466 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
2467 | css_put(&to->css); | |
2468 | } | |
02491447 DN |
2469 | return 0; |
2470 | } | |
2471 | return -EINVAL; | |
2472 | } | |
2473 | #else | |
2474 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2475 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2476 | { |
2477 | return -EINVAL; | |
2478 | } | |
8c7c6e34 | 2479 | #endif |
d13d1443 | 2480 | |
ae41be37 | 2481 | /* |
01b1ae63 KH |
2482 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
2483 | * page belongs to. | |
ae41be37 | 2484 | */ |
ac39cf8c | 2485 | int mem_cgroup_prepare_migration(struct page *page, |
2486 | struct page *newpage, struct mem_cgroup **ptr) | |
ae41be37 KH |
2487 | { |
2488 | struct page_cgroup *pc; | |
e8589cc1 | 2489 | struct mem_cgroup *mem = NULL; |
ac39cf8c | 2490 | enum charge_type ctype; |
e8589cc1 | 2491 | int ret = 0; |
8869b8f6 | 2492 | |
f8d66542 | 2493 | if (mem_cgroup_disabled()) |
4077960e BS |
2494 | return 0; |
2495 | ||
52d4b9ac KH |
2496 | pc = lookup_page_cgroup(page); |
2497 | lock_page_cgroup(pc); | |
2498 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
2499 | mem = pc->mem_cgroup; |
2500 | css_get(&mem->css); | |
ac39cf8c | 2501 | /* |
2502 | * At migrating an anonymous page, its mapcount goes down | |
2503 | * to 0 and uncharge() will be called. But, even if it's fully | |
2504 | * unmapped, migration may fail and this page has to be | |
2505 | * charged again. We set MIGRATION flag here and delay uncharge | |
2506 | * until end_migration() is called | |
2507 | * | |
2508 | * Corner Case Thinking | |
2509 | * A) | |
2510 | * When the old page was mapped as Anon and it's unmap-and-freed | |
2511 | * while migration was ongoing. | |
2512 | * If unmap finds the old page, uncharge() of it will be delayed | |
2513 | * until end_migration(). If unmap finds a new page, it's | |
2514 | * uncharged when it make mapcount to be 1->0. If unmap code | |
2515 | * finds swap_migration_entry, the new page will not be mapped | |
2516 | * and end_migration() will find it(mapcount==0). | |
2517 | * | |
2518 | * B) | |
2519 | * When the old page was mapped but migraion fails, the kernel | |
2520 | * remaps it. A charge for it is kept by MIGRATION flag even | |
2521 | * if mapcount goes down to 0. We can do remap successfully | |
2522 | * without charging it again. | |
2523 | * | |
2524 | * C) | |
2525 | * The "old" page is under lock_page() until the end of | |
2526 | * migration, so, the old page itself will not be swapped-out. | |
2527 | * If the new page is swapped out before end_migraton, our | |
2528 | * hook to usual swap-out path will catch the event. | |
2529 | */ | |
2530 | if (PageAnon(page)) | |
2531 | SetPageCgroupMigration(pc); | |
e8589cc1 | 2532 | } |
52d4b9ac | 2533 | unlock_page_cgroup(pc); |
ac39cf8c | 2534 | /* |
2535 | * If the page is not charged at this point, | |
2536 | * we return here. | |
2537 | */ | |
2538 | if (!mem) | |
2539 | return 0; | |
01b1ae63 | 2540 | |
93d5c9be | 2541 | *ptr = mem; |
ac39cf8c | 2542 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false); |
2543 | css_put(&mem->css);/* drop extra refcnt */ | |
2544 | if (ret || *ptr == NULL) { | |
2545 | if (PageAnon(page)) { | |
2546 | lock_page_cgroup(pc); | |
2547 | ClearPageCgroupMigration(pc); | |
2548 | unlock_page_cgroup(pc); | |
2549 | /* | |
2550 | * The old page may be fully unmapped while we kept it. | |
2551 | */ | |
2552 | mem_cgroup_uncharge_page(page); | |
2553 | } | |
2554 | return -ENOMEM; | |
e8589cc1 | 2555 | } |
ac39cf8c | 2556 | /* |
2557 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
2558 | * is called before end_migration, we can catch all events on this new | |
2559 | * page. In the case new page is migrated but not remapped, new page's | |
2560 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
2561 | */ | |
2562 | pc = lookup_page_cgroup(newpage); | |
2563 | if (PageAnon(page)) | |
2564 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2565 | else if (page_is_file_cache(page)) | |
2566 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2567 | else | |
2568 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
2569 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
e8589cc1 | 2570 | return ret; |
ae41be37 | 2571 | } |
8869b8f6 | 2572 | |
69029cd5 | 2573 | /* remove redundant charge if migration failed*/ |
01b1ae63 | 2574 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
ac39cf8c | 2575 | struct page *oldpage, struct page *newpage) |
ae41be37 | 2576 | { |
ac39cf8c | 2577 | struct page *used, *unused; |
01b1ae63 | 2578 | struct page_cgroup *pc; |
01b1ae63 KH |
2579 | |
2580 | if (!mem) | |
2581 | return; | |
ac39cf8c | 2582 | /* blocks rmdir() */ |
88703267 | 2583 | cgroup_exclude_rmdir(&mem->css); |
01b1ae63 KH |
2584 | /* at migration success, oldpage->mapping is NULL. */ |
2585 | if (oldpage->mapping) { | |
ac39cf8c | 2586 | used = oldpage; |
2587 | unused = newpage; | |
01b1ae63 | 2588 | } else { |
ac39cf8c | 2589 | used = newpage; |
01b1ae63 KH |
2590 | unused = oldpage; |
2591 | } | |
69029cd5 | 2592 | /* |
ac39cf8c | 2593 | * We disallowed uncharge of pages under migration because mapcount |
2594 | * of the page goes down to zero, temporarly. | |
2595 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 2596 | */ |
ac39cf8c | 2597 | pc = lookup_page_cgroup(oldpage); |
2598 | lock_page_cgroup(pc); | |
2599 | ClearPageCgroupMigration(pc); | |
2600 | unlock_page_cgroup(pc); | |
01b1ae63 | 2601 | |
ac39cf8c | 2602 | if (unused != oldpage) |
2603 | pc = lookup_page_cgroup(unused); | |
2604 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); | |
2605 | ||
2606 | pc = lookup_page_cgroup(used); | |
01b1ae63 | 2607 | /* |
ac39cf8c | 2608 | * If a page is a file cache, radix-tree replacement is very atomic |
2609 | * and we can skip this check. When it was an Anon page, its mapcount | |
2610 | * goes down to 0. But because we added MIGRATION flage, it's not | |
2611 | * uncharged yet. There are several case but page->mapcount check | |
2612 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
2613 | * check. (see prepare_charge() also) | |
69029cd5 | 2614 | */ |
ac39cf8c | 2615 | if (PageAnon(used)) |
2616 | mem_cgroup_uncharge_page(used); | |
88703267 | 2617 | /* |
ac39cf8c | 2618 | * At migration, we may charge account against cgroup which has no |
2619 | * tasks. | |
88703267 KH |
2620 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
2621 | * In that case, we need to call pre_destroy() again. check it here. | |
2622 | */ | |
2623 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 2624 | } |
78fb7466 | 2625 | |
c9b0ed51 | 2626 | /* |
ae3abae6 DN |
2627 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
2628 | * Calling hierarchical_reclaim is not enough because we should update | |
2629 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
2630 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
2631 | * not from the memcg which this page would be charged to. | |
2632 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 2633 | */ |
ae3abae6 | 2634 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
2635 | struct mm_struct *mm, |
2636 | gfp_t gfp_mask) | |
c9b0ed51 | 2637 | { |
b5a84319 | 2638 | struct mem_cgroup *mem = NULL; |
ae3abae6 | 2639 | int ret; |
c9b0ed51 | 2640 | |
f8d66542 | 2641 | if (mem_cgroup_disabled()) |
cede86ac | 2642 | return 0; |
c9b0ed51 | 2643 | |
ae3abae6 DN |
2644 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2645 | if (!ret) | |
2646 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 2647 | |
ae3abae6 | 2648 | return ret; |
c9b0ed51 KH |
2649 | } |
2650 | ||
8c7c6e34 KH |
2651 | static DEFINE_MUTEX(set_limit_mutex); |
2652 | ||
d38d2a75 | 2653 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 2654 | unsigned long long val) |
628f4235 | 2655 | { |
81d39c20 | 2656 | int retry_count; |
3c11ecf4 | 2657 | u64 memswlimit, memlimit; |
628f4235 | 2658 | int ret = 0; |
81d39c20 KH |
2659 | int children = mem_cgroup_count_children(memcg); |
2660 | u64 curusage, oldusage; | |
3c11ecf4 | 2661 | int enlarge; |
81d39c20 KH |
2662 | |
2663 | /* | |
2664 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2665 | * is depends on callers. We set our retry-count to be function | |
2666 | * of # of children which we should visit in this loop. | |
2667 | */ | |
2668 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
2669 | ||
2670 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 2671 | |
3c11ecf4 | 2672 | enlarge = 0; |
8c7c6e34 | 2673 | while (retry_count) { |
628f4235 KH |
2674 | if (signal_pending(current)) { |
2675 | ret = -EINTR; | |
2676 | break; | |
2677 | } | |
8c7c6e34 KH |
2678 | /* |
2679 | * Rather than hide all in some function, I do this in | |
2680 | * open coded manner. You see what this really does. | |
2681 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2682 | */ | |
2683 | mutex_lock(&set_limit_mutex); | |
2684 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2685 | if (memswlimit < val) { | |
2686 | ret = -EINVAL; | |
2687 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
2688 | break; |
2689 | } | |
3c11ecf4 KH |
2690 | |
2691 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2692 | if (memlimit < val) | |
2693 | enlarge = 1; | |
2694 | ||
8c7c6e34 | 2695 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
2696 | if (!ret) { |
2697 | if (memswlimit == val) | |
2698 | memcg->memsw_is_minimum = true; | |
2699 | else | |
2700 | memcg->memsw_is_minimum = false; | |
2701 | } | |
8c7c6e34 KH |
2702 | mutex_unlock(&set_limit_mutex); |
2703 | ||
2704 | if (!ret) | |
2705 | break; | |
2706 | ||
aa20d489 | 2707 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
4e416953 | 2708 | MEM_CGROUP_RECLAIM_SHRINK); |
81d39c20 KH |
2709 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
2710 | /* Usage is reduced ? */ | |
2711 | if (curusage >= oldusage) | |
2712 | retry_count--; | |
2713 | else | |
2714 | oldusage = curusage; | |
8c7c6e34 | 2715 | } |
3c11ecf4 KH |
2716 | if (!ret && enlarge) |
2717 | memcg_oom_recover(memcg); | |
14797e23 | 2718 | |
8c7c6e34 KH |
2719 | return ret; |
2720 | } | |
2721 | ||
338c8431 LZ |
2722 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
2723 | unsigned long long val) | |
8c7c6e34 | 2724 | { |
81d39c20 | 2725 | int retry_count; |
3c11ecf4 | 2726 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
2727 | int children = mem_cgroup_count_children(memcg); |
2728 | int ret = -EBUSY; | |
3c11ecf4 | 2729 | int enlarge = 0; |
8c7c6e34 | 2730 | |
81d39c20 KH |
2731 | /* see mem_cgroup_resize_res_limit */ |
2732 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
2733 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
2734 | while (retry_count) { |
2735 | if (signal_pending(current)) { | |
2736 | ret = -EINTR; | |
2737 | break; | |
2738 | } | |
2739 | /* | |
2740 | * Rather than hide all in some function, I do this in | |
2741 | * open coded manner. You see what this really does. | |
2742 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2743 | */ | |
2744 | mutex_lock(&set_limit_mutex); | |
2745 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2746 | if (memlimit > val) { | |
2747 | ret = -EINVAL; | |
2748 | mutex_unlock(&set_limit_mutex); | |
2749 | break; | |
2750 | } | |
3c11ecf4 KH |
2751 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
2752 | if (memswlimit < val) | |
2753 | enlarge = 1; | |
8c7c6e34 | 2754 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
2755 | if (!ret) { |
2756 | if (memlimit == val) | |
2757 | memcg->memsw_is_minimum = true; | |
2758 | else | |
2759 | memcg->memsw_is_minimum = false; | |
2760 | } | |
8c7c6e34 KH |
2761 | mutex_unlock(&set_limit_mutex); |
2762 | ||
2763 | if (!ret) | |
2764 | break; | |
2765 | ||
4e416953 | 2766 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 BS |
2767 | MEM_CGROUP_RECLAIM_NOSWAP | |
2768 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 2769 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 2770 | /* Usage is reduced ? */ |
8c7c6e34 | 2771 | if (curusage >= oldusage) |
628f4235 | 2772 | retry_count--; |
81d39c20 KH |
2773 | else |
2774 | oldusage = curusage; | |
628f4235 | 2775 | } |
3c11ecf4 KH |
2776 | if (!ret && enlarge) |
2777 | memcg_oom_recover(memcg); | |
628f4235 KH |
2778 | return ret; |
2779 | } | |
2780 | ||
4e416953 BS |
2781 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2782 | gfp_t gfp_mask, int nid, | |
2783 | int zid) | |
2784 | { | |
2785 | unsigned long nr_reclaimed = 0; | |
2786 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2787 | unsigned long reclaimed; | |
2788 | int loop = 0; | |
2789 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 2790 | unsigned long long excess; |
4e416953 BS |
2791 | |
2792 | if (order > 0) | |
2793 | return 0; | |
2794 | ||
2795 | mctz = soft_limit_tree_node_zone(nid, zid); | |
2796 | /* | |
2797 | * This loop can run a while, specially if mem_cgroup's continuously | |
2798 | * keep exceeding their soft limit and putting the system under | |
2799 | * pressure | |
2800 | */ | |
2801 | do { | |
2802 | if (next_mz) | |
2803 | mz = next_mz; | |
2804 | else | |
2805 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2806 | if (!mz) | |
2807 | break; | |
2808 | ||
2809 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
2810 | gfp_mask, | |
2811 | MEM_CGROUP_RECLAIM_SOFT); | |
2812 | nr_reclaimed += reclaimed; | |
2813 | spin_lock(&mctz->lock); | |
2814 | ||
2815 | /* | |
2816 | * If we failed to reclaim anything from this memory cgroup | |
2817 | * it is time to move on to the next cgroup | |
2818 | */ | |
2819 | next_mz = NULL; | |
2820 | if (!reclaimed) { | |
2821 | do { | |
2822 | /* | |
2823 | * Loop until we find yet another one. | |
2824 | * | |
2825 | * By the time we get the soft_limit lock | |
2826 | * again, someone might have aded the | |
2827 | * group back on the RB tree. Iterate to | |
2828 | * make sure we get a different mem. | |
2829 | * mem_cgroup_largest_soft_limit_node returns | |
2830 | * NULL if no other cgroup is present on | |
2831 | * the tree | |
2832 | */ | |
2833 | next_mz = | |
2834 | __mem_cgroup_largest_soft_limit_node(mctz); | |
2835 | if (next_mz == mz) { | |
2836 | css_put(&next_mz->mem->css); | |
2837 | next_mz = NULL; | |
2838 | } else /* next_mz == NULL or other memcg */ | |
2839 | break; | |
2840 | } while (1); | |
2841 | } | |
4e416953 | 2842 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 2843 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
2844 | /* |
2845 | * One school of thought says that we should not add | |
2846 | * back the node to the tree if reclaim returns 0. | |
2847 | * But our reclaim could return 0, simply because due | |
2848 | * to priority we are exposing a smaller subset of | |
2849 | * memory to reclaim from. Consider this as a longer | |
2850 | * term TODO. | |
2851 | */ | |
ef8745c1 KH |
2852 | /* If excess == 0, no tree ops */ |
2853 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
2854 | spin_unlock(&mctz->lock); |
2855 | css_put(&mz->mem->css); | |
2856 | loop++; | |
2857 | /* | |
2858 | * Could not reclaim anything and there are no more | |
2859 | * mem cgroups to try or we seem to be looping without | |
2860 | * reclaiming anything. | |
2861 | */ | |
2862 | if (!nr_reclaimed && | |
2863 | (next_mz == NULL || | |
2864 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2865 | break; | |
2866 | } while (!nr_reclaimed); | |
2867 | if (next_mz) | |
2868 | css_put(&next_mz->mem->css); | |
2869 | return nr_reclaimed; | |
2870 | } | |
2871 | ||
cc847582 KH |
2872 | /* |
2873 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
2874 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
2875 | */ | |
f817ed48 | 2876 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 2877 | int node, int zid, enum lru_list lru) |
cc847582 | 2878 | { |
08e552c6 KH |
2879 | struct zone *zone; |
2880 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 2881 | struct page_cgroup *pc, *busy; |
08e552c6 | 2882 | unsigned long flags, loop; |
072c56c1 | 2883 | struct list_head *list; |
f817ed48 | 2884 | int ret = 0; |
072c56c1 | 2885 | |
08e552c6 KH |
2886 | zone = &NODE_DATA(node)->node_zones[zid]; |
2887 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 2888 | list = &mz->lists[lru]; |
cc847582 | 2889 | |
f817ed48 KH |
2890 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
2891 | /* give some margin against EBUSY etc...*/ | |
2892 | loop += 256; | |
2893 | busy = NULL; | |
2894 | while (loop--) { | |
2895 | ret = 0; | |
08e552c6 | 2896 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 2897 | if (list_empty(list)) { |
08e552c6 | 2898 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 2899 | break; |
f817ed48 KH |
2900 | } |
2901 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
2902 | if (busy == pc) { | |
2903 | list_move(&pc->lru, list); | |
648bcc77 | 2904 | busy = NULL; |
08e552c6 | 2905 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
2906 | continue; |
2907 | } | |
08e552c6 | 2908 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 2909 | |
2c26fdd7 | 2910 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 2911 | if (ret == -ENOMEM) |
52d4b9ac | 2912 | break; |
f817ed48 KH |
2913 | |
2914 | if (ret == -EBUSY || ret == -EINVAL) { | |
2915 | /* found lock contention or "pc" is obsolete. */ | |
2916 | busy = pc; | |
2917 | cond_resched(); | |
2918 | } else | |
2919 | busy = NULL; | |
cc847582 | 2920 | } |
08e552c6 | 2921 | |
f817ed48 KH |
2922 | if (!ret && !list_empty(list)) |
2923 | return -EBUSY; | |
2924 | return ret; | |
cc847582 KH |
2925 | } |
2926 | ||
2927 | /* | |
2928 | * make mem_cgroup's charge to be 0 if there is no task. | |
2929 | * This enables deleting this mem_cgroup. | |
2930 | */ | |
c1e862c1 | 2931 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 2932 | { |
f817ed48 KH |
2933 | int ret; |
2934 | int node, zid, shrink; | |
2935 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 2936 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 2937 | |
cc847582 | 2938 | css_get(&mem->css); |
f817ed48 KH |
2939 | |
2940 | shrink = 0; | |
c1e862c1 KH |
2941 | /* should free all ? */ |
2942 | if (free_all) | |
2943 | goto try_to_free; | |
f817ed48 | 2944 | move_account: |
fce66477 | 2945 | do { |
f817ed48 | 2946 | ret = -EBUSY; |
c1e862c1 KH |
2947 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
2948 | goto out; | |
2949 | ret = -EINTR; | |
2950 | if (signal_pending(current)) | |
cc847582 | 2951 | goto out; |
52d4b9ac KH |
2952 | /* This is for making all *used* pages to be on LRU. */ |
2953 | lru_add_drain_all(); | |
cdec2e42 | 2954 | drain_all_stock_sync(); |
f817ed48 | 2955 | ret = 0; |
299b4eaa | 2956 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 2957 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 2958 | enum lru_list l; |
f817ed48 KH |
2959 | for_each_lru(l) { |
2960 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 2961 | node, zid, l); |
f817ed48 KH |
2962 | if (ret) |
2963 | break; | |
2964 | } | |
1ecaab2b | 2965 | } |
f817ed48 KH |
2966 | if (ret) |
2967 | break; | |
2968 | } | |
3c11ecf4 | 2969 | memcg_oom_recover(mem); |
f817ed48 KH |
2970 | /* it seems parent cgroup doesn't have enough mem */ |
2971 | if (ret == -ENOMEM) | |
2972 | goto try_to_free; | |
52d4b9ac | 2973 | cond_resched(); |
fce66477 DN |
2974 | /* "ret" should also be checked to ensure all lists are empty. */ |
2975 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
2976 | out: |
2977 | css_put(&mem->css); | |
2978 | return ret; | |
f817ed48 KH |
2979 | |
2980 | try_to_free: | |
c1e862c1 KH |
2981 | /* returns EBUSY if there is a task or if we come here twice. */ |
2982 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
2983 | ret = -EBUSY; |
2984 | goto out; | |
2985 | } | |
c1e862c1 KH |
2986 | /* we call try-to-free pages for make this cgroup empty */ |
2987 | lru_add_drain_all(); | |
f817ed48 KH |
2988 | /* try to free all pages in this cgroup */ |
2989 | shrink = 1; | |
2990 | while (nr_retries && mem->res.usage > 0) { | |
2991 | int progress; | |
c1e862c1 KH |
2992 | |
2993 | if (signal_pending(current)) { | |
2994 | ret = -EINTR; | |
2995 | goto out; | |
2996 | } | |
a7885eb8 KM |
2997 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
2998 | false, get_swappiness(mem)); | |
c1e862c1 | 2999 | if (!progress) { |
f817ed48 | 3000 | nr_retries--; |
c1e862c1 | 3001 | /* maybe some writeback is necessary */ |
8aa7e847 | 3002 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3003 | } |
f817ed48 KH |
3004 | |
3005 | } | |
08e552c6 | 3006 | lru_add_drain(); |
f817ed48 | 3007 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3008 | goto move_account; |
cc847582 KH |
3009 | } |
3010 | ||
c1e862c1 KH |
3011 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3012 | { | |
3013 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3014 | } | |
3015 | ||
3016 | ||
18f59ea7 BS |
3017 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3018 | { | |
3019 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3020 | } | |
3021 | ||
3022 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3023 | u64 val) | |
3024 | { | |
3025 | int retval = 0; | |
3026 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3027 | struct cgroup *parent = cont->parent; | |
3028 | struct mem_cgroup *parent_mem = NULL; | |
3029 | ||
3030 | if (parent) | |
3031 | parent_mem = mem_cgroup_from_cont(parent); | |
3032 | ||
3033 | cgroup_lock(); | |
3034 | /* | |
af901ca1 | 3035 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3036 | * in the child subtrees. If it is unset, then the change can |
3037 | * occur, provided the current cgroup has no children. | |
3038 | * | |
3039 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3040 | * set if there are no children. | |
3041 | */ | |
3042 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
3043 | (val == 1 || val == 0)) { | |
3044 | if (list_empty(&cont->children)) | |
3045 | mem->use_hierarchy = val; | |
3046 | else | |
3047 | retval = -EBUSY; | |
3048 | } else | |
3049 | retval = -EINVAL; | |
3050 | cgroup_unlock(); | |
3051 | ||
3052 | return retval; | |
3053 | } | |
3054 | ||
0c3e73e8 BS |
3055 | struct mem_cgroup_idx_data { |
3056 | s64 val; | |
3057 | enum mem_cgroup_stat_index idx; | |
3058 | }; | |
3059 | ||
3060 | static int | |
3061 | mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) | |
3062 | { | |
3063 | struct mem_cgroup_idx_data *d = data; | |
c62b1a3b | 3064 | d->val += mem_cgroup_read_stat(mem, d->idx); |
0c3e73e8 BS |
3065 | return 0; |
3066 | } | |
3067 | ||
3068 | static void | |
3069 | mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, | |
3070 | enum mem_cgroup_stat_index idx, s64 *val) | |
3071 | { | |
3072 | struct mem_cgroup_idx_data d; | |
3073 | d.idx = idx; | |
3074 | d.val = 0; | |
3075 | mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat); | |
3076 | *val = d.val; | |
3077 | } | |
3078 | ||
104f3928 KS |
3079 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
3080 | { | |
3081 | u64 idx_val, val; | |
3082 | ||
3083 | if (!mem_cgroup_is_root(mem)) { | |
3084 | if (!swap) | |
3085 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
3086 | else | |
3087 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
3088 | } | |
3089 | ||
3090 | mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val); | |
3091 | val = idx_val; | |
3092 | mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val); | |
3093 | val += idx_val; | |
3094 | ||
3095 | if (swap) { | |
3096 | mem_cgroup_get_recursive_idx_stat(mem, | |
3097 | MEM_CGROUP_STAT_SWAPOUT, &idx_val); | |
3098 | val += idx_val; | |
3099 | } | |
3100 | ||
3101 | return val << PAGE_SHIFT; | |
3102 | } | |
3103 | ||
2c3daa72 | 3104 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3105 | { |
8c7c6e34 | 3106 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 3107 | u64 val; |
8c7c6e34 KH |
3108 | int type, name; |
3109 | ||
3110 | type = MEMFILE_TYPE(cft->private); | |
3111 | name = MEMFILE_ATTR(cft->private); | |
3112 | switch (type) { | |
3113 | case _MEM: | |
104f3928 KS |
3114 | if (name == RES_USAGE) |
3115 | val = mem_cgroup_usage(mem, false); | |
3116 | else | |
0c3e73e8 | 3117 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
3118 | break; |
3119 | case _MEMSWAP: | |
104f3928 KS |
3120 | if (name == RES_USAGE) |
3121 | val = mem_cgroup_usage(mem, true); | |
3122 | else | |
0c3e73e8 | 3123 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
3124 | break; |
3125 | default: | |
3126 | BUG(); | |
3127 | break; | |
3128 | } | |
3129 | return val; | |
8cdea7c0 | 3130 | } |
628f4235 KH |
3131 | /* |
3132 | * The user of this function is... | |
3133 | * RES_LIMIT. | |
3134 | */ | |
856c13aa PM |
3135 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3136 | const char *buffer) | |
8cdea7c0 | 3137 | { |
628f4235 | 3138 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3139 | int type, name; |
628f4235 KH |
3140 | unsigned long long val; |
3141 | int ret; | |
3142 | ||
8c7c6e34 KH |
3143 | type = MEMFILE_TYPE(cft->private); |
3144 | name = MEMFILE_ATTR(cft->private); | |
3145 | switch (name) { | |
628f4235 | 3146 | case RES_LIMIT: |
4b3bde4c BS |
3147 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3148 | ret = -EINVAL; | |
3149 | break; | |
3150 | } | |
628f4235 KH |
3151 | /* This function does all necessary parse...reuse it */ |
3152 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3153 | if (ret) |
3154 | break; | |
3155 | if (type == _MEM) | |
628f4235 | 3156 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3157 | else |
3158 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3159 | break; |
296c81d8 BS |
3160 | case RES_SOFT_LIMIT: |
3161 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3162 | if (ret) | |
3163 | break; | |
3164 | /* | |
3165 | * For memsw, soft limits are hard to implement in terms | |
3166 | * of semantics, for now, we support soft limits for | |
3167 | * control without swap | |
3168 | */ | |
3169 | if (type == _MEM) | |
3170 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3171 | else | |
3172 | ret = -EINVAL; | |
3173 | break; | |
628f4235 KH |
3174 | default: |
3175 | ret = -EINVAL; /* should be BUG() ? */ | |
3176 | break; | |
3177 | } | |
3178 | return ret; | |
8cdea7c0 BS |
3179 | } |
3180 | ||
fee7b548 KH |
3181 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3182 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3183 | { | |
3184 | struct cgroup *cgroup; | |
3185 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3186 | ||
3187 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3188 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3189 | cgroup = memcg->css.cgroup; | |
3190 | if (!memcg->use_hierarchy) | |
3191 | goto out; | |
3192 | ||
3193 | while (cgroup->parent) { | |
3194 | cgroup = cgroup->parent; | |
3195 | memcg = mem_cgroup_from_cont(cgroup); | |
3196 | if (!memcg->use_hierarchy) | |
3197 | break; | |
3198 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3199 | min_limit = min(min_limit, tmp); | |
3200 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3201 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3202 | } | |
3203 | out: | |
3204 | *mem_limit = min_limit; | |
3205 | *memsw_limit = min_memsw_limit; | |
3206 | return; | |
3207 | } | |
3208 | ||
29f2a4da | 3209 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
3210 | { |
3211 | struct mem_cgroup *mem; | |
8c7c6e34 | 3212 | int type, name; |
c84872e1 PE |
3213 | |
3214 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
3215 | type = MEMFILE_TYPE(event); |
3216 | name = MEMFILE_ATTR(event); | |
3217 | switch (name) { | |
29f2a4da | 3218 | case RES_MAX_USAGE: |
8c7c6e34 KH |
3219 | if (type == _MEM) |
3220 | res_counter_reset_max(&mem->res); | |
3221 | else | |
3222 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
3223 | break; |
3224 | case RES_FAILCNT: | |
8c7c6e34 KH |
3225 | if (type == _MEM) |
3226 | res_counter_reset_failcnt(&mem->res); | |
3227 | else | |
3228 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
3229 | break; |
3230 | } | |
f64c3f54 | 3231 | |
85cc59db | 3232 | return 0; |
c84872e1 PE |
3233 | } |
3234 | ||
7dc74be0 DN |
3235 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3236 | struct cftype *cft) | |
3237 | { | |
3238 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3239 | } | |
3240 | ||
02491447 | 3241 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3242 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3243 | struct cftype *cft, u64 val) | |
3244 | { | |
3245 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3246 | ||
3247 | if (val >= (1 << NR_MOVE_TYPE)) | |
3248 | return -EINVAL; | |
3249 | /* | |
3250 | * We check this value several times in both in can_attach() and | |
3251 | * attach(), so we need cgroup lock to prevent this value from being | |
3252 | * inconsistent. | |
3253 | */ | |
3254 | cgroup_lock(); | |
3255 | mem->move_charge_at_immigrate = val; | |
3256 | cgroup_unlock(); | |
3257 | ||
3258 | return 0; | |
3259 | } | |
02491447 DN |
3260 | #else |
3261 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3262 | struct cftype *cft, u64 val) | |
3263 | { | |
3264 | return -ENOSYS; | |
3265 | } | |
3266 | #endif | |
7dc74be0 | 3267 | |
14067bb3 KH |
3268 | |
3269 | /* For read statistics */ | |
3270 | enum { | |
3271 | MCS_CACHE, | |
3272 | MCS_RSS, | |
d8046582 | 3273 | MCS_FILE_MAPPED, |
14067bb3 KH |
3274 | MCS_PGPGIN, |
3275 | MCS_PGPGOUT, | |
1dd3a273 | 3276 | MCS_SWAP, |
14067bb3 KH |
3277 | MCS_INACTIVE_ANON, |
3278 | MCS_ACTIVE_ANON, | |
3279 | MCS_INACTIVE_FILE, | |
3280 | MCS_ACTIVE_FILE, | |
3281 | MCS_UNEVICTABLE, | |
3282 | NR_MCS_STAT, | |
3283 | }; | |
3284 | ||
3285 | struct mcs_total_stat { | |
3286 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
3287 | }; |
3288 | ||
14067bb3 KH |
3289 | struct { |
3290 | char *local_name; | |
3291 | char *total_name; | |
3292 | } memcg_stat_strings[NR_MCS_STAT] = { | |
3293 | {"cache", "total_cache"}, | |
3294 | {"rss", "total_rss"}, | |
d69b042f | 3295 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
3296 | {"pgpgin", "total_pgpgin"}, |
3297 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3298 | {"swap", "total_swap"}, |
14067bb3 KH |
3299 | {"inactive_anon", "total_inactive_anon"}, |
3300 | {"active_anon", "total_active_anon"}, | |
3301 | {"inactive_file", "total_inactive_file"}, | |
3302 | {"active_file", "total_active_file"}, | |
3303 | {"unevictable", "total_unevictable"} | |
3304 | }; | |
3305 | ||
3306 | ||
3307 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
3308 | { | |
3309 | struct mcs_total_stat *s = data; | |
3310 | s64 val; | |
3311 | ||
3312 | /* per cpu stat */ | |
c62b1a3b | 3313 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 3314 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 3315 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 3316 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 3317 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 3318 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c62b1a3b | 3319 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); |
14067bb3 | 3320 | s->stat[MCS_PGPGIN] += val; |
c62b1a3b | 3321 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
14067bb3 | 3322 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 3323 | if (do_swap_account) { |
c62b1a3b | 3324 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
3325 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
3326 | } | |
14067bb3 KH |
3327 | |
3328 | /* per zone stat */ | |
3329 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
3330 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
3331 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
3332 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
3333 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
3334 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
3335 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
3336 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
3337 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
3338 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
3339 | return 0; | |
3340 | } | |
3341 | ||
3342 | static void | |
3343 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
3344 | { | |
3345 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
3346 | } | |
3347 | ||
c64745cf PM |
3348 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
3349 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 3350 | { |
d2ceb9b7 | 3351 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 3352 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
3353 | int i; |
3354 | ||
14067bb3 KH |
3355 | memset(&mystat, 0, sizeof(mystat)); |
3356 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 3357 | |
1dd3a273 DN |
3358 | for (i = 0; i < NR_MCS_STAT; i++) { |
3359 | if (i == MCS_SWAP && !do_swap_account) | |
3360 | continue; | |
14067bb3 | 3361 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 3362 | } |
7b854121 | 3363 | |
14067bb3 | 3364 | /* Hierarchical information */ |
fee7b548 KH |
3365 | { |
3366 | unsigned long long limit, memsw_limit; | |
3367 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
3368 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
3369 | if (do_swap_account) | |
3370 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
3371 | } | |
7f016ee8 | 3372 | |
14067bb3 KH |
3373 | memset(&mystat, 0, sizeof(mystat)); |
3374 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
3375 | for (i = 0; i < NR_MCS_STAT; i++) { |
3376 | if (i == MCS_SWAP && !do_swap_account) | |
3377 | continue; | |
14067bb3 | 3378 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 3379 | } |
14067bb3 | 3380 | |
7f016ee8 | 3381 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 3382 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
3383 | |
3384 | { | |
3385 | int nid, zid; | |
3386 | struct mem_cgroup_per_zone *mz; | |
3387 | unsigned long recent_rotated[2] = {0, 0}; | |
3388 | unsigned long recent_scanned[2] = {0, 0}; | |
3389 | ||
3390 | for_each_online_node(nid) | |
3391 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3392 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3393 | ||
3394 | recent_rotated[0] += | |
3395 | mz->reclaim_stat.recent_rotated[0]; | |
3396 | recent_rotated[1] += | |
3397 | mz->reclaim_stat.recent_rotated[1]; | |
3398 | recent_scanned[0] += | |
3399 | mz->reclaim_stat.recent_scanned[0]; | |
3400 | recent_scanned[1] += | |
3401 | mz->reclaim_stat.recent_scanned[1]; | |
3402 | } | |
3403 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3404 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3405 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3406 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3407 | } | |
3408 | #endif | |
3409 | ||
d2ceb9b7 KH |
3410 | return 0; |
3411 | } | |
3412 | ||
a7885eb8 KM |
3413 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
3414 | { | |
3415 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3416 | ||
3417 | return get_swappiness(memcg); | |
3418 | } | |
3419 | ||
3420 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3421 | u64 val) | |
3422 | { | |
3423 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3424 | struct mem_cgroup *parent; | |
068b38c1 | 3425 | |
a7885eb8 KM |
3426 | if (val > 100) |
3427 | return -EINVAL; | |
3428 | ||
3429 | if (cgrp->parent == NULL) | |
3430 | return -EINVAL; | |
3431 | ||
3432 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
3433 | |
3434 | cgroup_lock(); | |
3435 | ||
a7885eb8 KM |
3436 | /* If under hierarchy, only empty-root can set this value */ |
3437 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
3438 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3439 | cgroup_unlock(); | |
a7885eb8 | 3440 | return -EINVAL; |
068b38c1 | 3441 | } |
a7885eb8 KM |
3442 | |
3443 | spin_lock(&memcg->reclaim_param_lock); | |
3444 | memcg->swappiness = val; | |
3445 | spin_unlock(&memcg->reclaim_param_lock); | |
3446 | ||
068b38c1 LZ |
3447 | cgroup_unlock(); |
3448 | ||
a7885eb8 KM |
3449 | return 0; |
3450 | } | |
3451 | ||
2e72b634 KS |
3452 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3453 | { | |
3454 | struct mem_cgroup_threshold_ary *t; | |
3455 | u64 usage; | |
3456 | int i; | |
3457 | ||
3458 | rcu_read_lock(); | |
3459 | if (!swap) | |
3460 | t = rcu_dereference(memcg->thresholds); | |
3461 | else | |
3462 | t = rcu_dereference(memcg->memsw_thresholds); | |
3463 | ||
3464 | if (!t) | |
3465 | goto unlock; | |
3466 | ||
3467 | usage = mem_cgroup_usage(memcg, swap); | |
3468 | ||
3469 | /* | |
3470 | * current_threshold points to threshold just below usage. | |
3471 | * If it's not true, a threshold was crossed after last | |
3472 | * call of __mem_cgroup_threshold(). | |
3473 | */ | |
5407a562 | 3474 | i = t->current_threshold; |
2e72b634 KS |
3475 | |
3476 | /* | |
3477 | * Iterate backward over array of thresholds starting from | |
3478 | * current_threshold and check if a threshold is crossed. | |
3479 | * If none of thresholds below usage is crossed, we read | |
3480 | * only one element of the array here. | |
3481 | */ | |
3482 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3483 | eventfd_signal(t->entries[i].eventfd, 1); | |
3484 | ||
3485 | /* i = current_threshold + 1 */ | |
3486 | i++; | |
3487 | ||
3488 | /* | |
3489 | * Iterate forward over array of thresholds starting from | |
3490 | * current_threshold+1 and check if a threshold is crossed. | |
3491 | * If none of thresholds above usage is crossed, we read | |
3492 | * only one element of the array here. | |
3493 | */ | |
3494 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3495 | eventfd_signal(t->entries[i].eventfd, 1); | |
3496 | ||
3497 | /* Update current_threshold */ | |
5407a562 | 3498 | t->current_threshold = i - 1; |
2e72b634 KS |
3499 | unlock: |
3500 | rcu_read_unlock(); | |
3501 | } | |
3502 | ||
3503 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3504 | { | |
3505 | __mem_cgroup_threshold(memcg, false); | |
3506 | if (do_swap_account) | |
3507 | __mem_cgroup_threshold(memcg, true); | |
3508 | } | |
3509 | ||
3510 | static int compare_thresholds(const void *a, const void *b) | |
3511 | { | |
3512 | const struct mem_cgroup_threshold *_a = a; | |
3513 | const struct mem_cgroup_threshold *_b = b; | |
3514 | ||
3515 | return _a->threshold - _b->threshold; | |
3516 | } | |
3517 | ||
9490ff27 KH |
3518 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem, void *data) |
3519 | { | |
3520 | struct mem_cgroup_eventfd_list *ev; | |
3521 | ||
3522 | list_for_each_entry(ev, &mem->oom_notify, list) | |
3523 | eventfd_signal(ev->eventfd, 1); | |
3524 | return 0; | |
3525 | } | |
3526 | ||
3527 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) | |
3528 | { | |
3529 | mem_cgroup_walk_tree(mem, NULL, mem_cgroup_oom_notify_cb); | |
3530 | } | |
3531 | ||
3532 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
3533 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
3534 | { |
3535 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3536 | struct mem_cgroup_threshold_ary *thresholds, *thresholds_new; | |
3537 | int type = MEMFILE_TYPE(cft->private); | |
3538 | u64 threshold, usage; | |
3539 | int size; | |
3540 | int i, ret; | |
3541 | ||
3542 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
3543 | if (ret) | |
3544 | return ret; | |
3545 | ||
3546 | mutex_lock(&memcg->thresholds_lock); | |
3547 | if (type == _MEM) | |
3548 | thresholds = memcg->thresholds; | |
3549 | else if (type == _MEMSWAP) | |
3550 | thresholds = memcg->memsw_thresholds; | |
3551 | else | |
3552 | BUG(); | |
3553 | ||
3554 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3555 | ||
3556 | /* Check if a threshold crossed before adding a new one */ | |
3557 | if (thresholds) | |
3558 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3559 | ||
3560 | if (thresholds) | |
3561 | size = thresholds->size + 1; | |
3562 | else | |
3563 | size = 1; | |
3564 | ||
3565 | /* Allocate memory for new array of thresholds */ | |
3566 | thresholds_new = kmalloc(sizeof(*thresholds_new) + | |
3567 | size * sizeof(struct mem_cgroup_threshold), | |
3568 | GFP_KERNEL); | |
3569 | if (!thresholds_new) { | |
3570 | ret = -ENOMEM; | |
3571 | goto unlock; | |
3572 | } | |
3573 | thresholds_new->size = size; | |
3574 | ||
3575 | /* Copy thresholds (if any) to new array */ | |
3576 | if (thresholds) | |
3577 | memcpy(thresholds_new->entries, thresholds->entries, | |
3578 | thresholds->size * | |
3579 | sizeof(struct mem_cgroup_threshold)); | |
3580 | /* Add new threshold */ | |
3581 | thresholds_new->entries[size - 1].eventfd = eventfd; | |
3582 | thresholds_new->entries[size - 1].threshold = threshold; | |
3583 | ||
3584 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
3585 | sort(thresholds_new->entries, size, | |
3586 | sizeof(struct mem_cgroup_threshold), | |
3587 | compare_thresholds, NULL); | |
3588 | ||
3589 | /* Find current threshold */ | |
5407a562 | 3590 | thresholds_new->current_threshold = -1; |
2e72b634 KS |
3591 | for (i = 0; i < size; i++) { |
3592 | if (thresholds_new->entries[i].threshold < usage) { | |
3593 | /* | |
3594 | * thresholds_new->current_threshold will not be used | |
3595 | * until rcu_assign_pointer(), so it's safe to increment | |
3596 | * it here. | |
3597 | */ | |
5407a562 | 3598 | ++thresholds_new->current_threshold; |
2e72b634 KS |
3599 | } |
3600 | } | |
3601 | ||
2e72b634 KS |
3602 | if (type == _MEM) |
3603 | rcu_assign_pointer(memcg->thresholds, thresholds_new); | |
3604 | else | |
3605 | rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new); | |
3606 | ||
3607 | /* To be sure that nobody uses thresholds before freeing it */ | |
3608 | synchronize_rcu(); | |
3609 | ||
3610 | kfree(thresholds); | |
3611 | unlock: | |
3612 | mutex_unlock(&memcg->thresholds_lock); | |
3613 | ||
3614 | return ret; | |
3615 | } | |
3616 | ||
9490ff27 KH |
3617 | static int mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
3618 | struct cftype *cft, struct eventfd_ctx *eventfd) | |
2e72b634 KS |
3619 | { |
3620 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3621 | struct mem_cgroup_threshold_ary *thresholds, *thresholds_new; | |
3622 | int type = MEMFILE_TYPE(cft->private); | |
3623 | u64 usage; | |
3624 | int size = 0; | |
315c1998 | 3625 | int i, j, ret = 0; |
2e72b634 KS |
3626 | |
3627 | mutex_lock(&memcg->thresholds_lock); | |
3628 | if (type == _MEM) | |
3629 | thresholds = memcg->thresholds; | |
3630 | else if (type == _MEMSWAP) | |
3631 | thresholds = memcg->memsw_thresholds; | |
3632 | else | |
3633 | BUG(); | |
3634 | ||
3635 | /* | |
3636 | * Something went wrong if we trying to unregister a threshold | |
3637 | * if we don't have thresholds | |
3638 | */ | |
3639 | BUG_ON(!thresholds); | |
3640 | ||
3641 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3642 | ||
3643 | /* Check if a threshold crossed before removing */ | |
3644 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3645 | ||
3646 | /* Calculate new number of threshold */ | |
3647 | for (i = 0; i < thresholds->size; i++) { | |
3648 | if (thresholds->entries[i].eventfd != eventfd) | |
3649 | size++; | |
3650 | } | |
3651 | ||
3652 | /* Set thresholds array to NULL if we don't have thresholds */ | |
3653 | if (!size) { | |
3654 | thresholds_new = NULL; | |
3655 | goto assign; | |
3656 | } | |
3657 | ||
3658 | /* Allocate memory for new array of thresholds */ | |
3659 | thresholds_new = kmalloc(sizeof(*thresholds_new) + | |
3660 | size * sizeof(struct mem_cgroup_threshold), | |
3661 | GFP_KERNEL); | |
3662 | if (!thresholds_new) { | |
3663 | ret = -ENOMEM; | |
3664 | goto unlock; | |
3665 | } | |
3666 | thresholds_new->size = size; | |
3667 | ||
3668 | /* Copy thresholds and find current threshold */ | |
5407a562 | 3669 | thresholds_new->current_threshold = -1; |
2e72b634 KS |
3670 | for (i = 0, j = 0; i < thresholds->size; i++) { |
3671 | if (thresholds->entries[i].eventfd == eventfd) | |
3672 | continue; | |
3673 | ||
3674 | thresholds_new->entries[j] = thresholds->entries[i]; | |
3675 | if (thresholds_new->entries[j].threshold < usage) { | |
3676 | /* | |
3677 | * thresholds_new->current_threshold will not be used | |
3678 | * until rcu_assign_pointer(), so it's safe to increment | |
3679 | * it here. | |
3680 | */ | |
5407a562 | 3681 | ++thresholds_new->current_threshold; |
2e72b634 KS |
3682 | } |
3683 | j++; | |
3684 | } | |
3685 | ||
3686 | assign: | |
3687 | if (type == _MEM) | |
3688 | rcu_assign_pointer(memcg->thresholds, thresholds_new); | |
3689 | else | |
3690 | rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new); | |
3691 | ||
3692 | /* To be sure that nobody uses thresholds before freeing it */ | |
3693 | synchronize_rcu(); | |
3694 | ||
2e72b634 KS |
3695 | kfree(thresholds); |
3696 | unlock: | |
3697 | mutex_unlock(&memcg->thresholds_lock); | |
3698 | ||
3699 | return ret; | |
3700 | } | |
c1e862c1 | 3701 | |
9490ff27 KH |
3702 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
3703 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
3704 | { | |
3705 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3706 | struct mem_cgroup_eventfd_list *event; | |
3707 | int type = MEMFILE_TYPE(cft->private); | |
3708 | ||
3709 | BUG_ON(type != _OOM_TYPE); | |
3710 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
3711 | if (!event) | |
3712 | return -ENOMEM; | |
3713 | ||
3714 | mutex_lock(&memcg_oom_mutex); | |
3715 | ||
3716 | event->eventfd = eventfd; | |
3717 | list_add(&event->list, &memcg->oom_notify); | |
3718 | ||
3719 | /* already in OOM ? */ | |
3720 | if (atomic_read(&memcg->oom_lock)) | |
3721 | eventfd_signal(eventfd, 1); | |
3722 | mutex_unlock(&memcg_oom_mutex); | |
3723 | ||
3724 | return 0; | |
3725 | } | |
3726 | ||
3727 | static int mem_cgroup_oom_unregister_event(struct cgroup *cgrp, | |
3728 | struct cftype *cft, struct eventfd_ctx *eventfd) | |
3729 | { | |
3730 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3731 | struct mem_cgroup_eventfd_list *ev, *tmp; | |
3732 | int type = MEMFILE_TYPE(cft->private); | |
3733 | ||
3734 | BUG_ON(type != _OOM_TYPE); | |
3735 | ||
3736 | mutex_lock(&memcg_oom_mutex); | |
3737 | ||
3738 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { | |
3739 | if (ev->eventfd == eventfd) { | |
3740 | list_del(&ev->list); | |
3741 | kfree(ev); | |
3742 | } | |
3743 | } | |
3744 | ||
3745 | mutex_unlock(&memcg_oom_mutex); | |
3746 | ||
3747 | return 0; | |
3748 | } | |
3749 | ||
3c11ecf4 KH |
3750 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
3751 | struct cftype *cft, struct cgroup_map_cb *cb) | |
3752 | { | |
3753 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3754 | ||
3755 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); | |
3756 | ||
3757 | if (atomic_read(&mem->oom_lock)) | |
3758 | cb->fill(cb, "under_oom", 1); | |
3759 | else | |
3760 | cb->fill(cb, "under_oom", 0); | |
3761 | return 0; | |
3762 | } | |
3763 | ||
3764 | /* | |
3765 | */ | |
3766 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, | |
3767 | struct cftype *cft, u64 val) | |
3768 | { | |
3769 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3770 | struct mem_cgroup *parent; | |
3771 | ||
3772 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
3773 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
3774 | return -EINVAL; | |
3775 | ||
3776 | parent = mem_cgroup_from_cont(cgrp->parent); | |
3777 | ||
3778 | cgroup_lock(); | |
3779 | /* oom-kill-disable is a flag for subhierarchy. */ | |
3780 | if ((parent->use_hierarchy) || | |
3781 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { | |
3782 | cgroup_unlock(); | |
3783 | return -EINVAL; | |
3784 | } | |
3785 | mem->oom_kill_disable = val; | |
3786 | cgroup_unlock(); | |
3787 | return 0; | |
3788 | } | |
3789 | ||
8cdea7c0 BS |
3790 | static struct cftype mem_cgroup_files[] = { |
3791 | { | |
0eea1030 | 3792 | .name = "usage_in_bytes", |
8c7c6e34 | 3793 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 3794 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
3795 | .register_event = mem_cgroup_usage_register_event, |
3796 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 3797 | }, |
c84872e1 PE |
3798 | { |
3799 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3800 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 3801 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
3802 | .read_u64 = mem_cgroup_read, |
3803 | }, | |
8cdea7c0 | 3804 | { |
0eea1030 | 3805 | .name = "limit_in_bytes", |
8c7c6e34 | 3806 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 3807 | .write_string = mem_cgroup_write, |
2c3daa72 | 3808 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3809 | }, |
296c81d8 BS |
3810 | { |
3811 | .name = "soft_limit_in_bytes", | |
3812 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
3813 | .write_string = mem_cgroup_write, | |
3814 | .read_u64 = mem_cgroup_read, | |
3815 | }, | |
8cdea7c0 BS |
3816 | { |
3817 | .name = "failcnt", | |
8c7c6e34 | 3818 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 3819 | .trigger = mem_cgroup_reset, |
2c3daa72 | 3820 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3821 | }, |
d2ceb9b7 KH |
3822 | { |
3823 | .name = "stat", | |
c64745cf | 3824 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 3825 | }, |
c1e862c1 KH |
3826 | { |
3827 | .name = "force_empty", | |
3828 | .trigger = mem_cgroup_force_empty_write, | |
3829 | }, | |
18f59ea7 BS |
3830 | { |
3831 | .name = "use_hierarchy", | |
3832 | .write_u64 = mem_cgroup_hierarchy_write, | |
3833 | .read_u64 = mem_cgroup_hierarchy_read, | |
3834 | }, | |
a7885eb8 KM |
3835 | { |
3836 | .name = "swappiness", | |
3837 | .read_u64 = mem_cgroup_swappiness_read, | |
3838 | .write_u64 = mem_cgroup_swappiness_write, | |
3839 | }, | |
7dc74be0 DN |
3840 | { |
3841 | .name = "move_charge_at_immigrate", | |
3842 | .read_u64 = mem_cgroup_move_charge_read, | |
3843 | .write_u64 = mem_cgroup_move_charge_write, | |
3844 | }, | |
9490ff27 KH |
3845 | { |
3846 | .name = "oom_control", | |
3c11ecf4 KH |
3847 | .read_map = mem_cgroup_oom_control_read, |
3848 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
3849 | .register_event = mem_cgroup_oom_register_event, |
3850 | .unregister_event = mem_cgroup_oom_unregister_event, | |
3851 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
3852 | }, | |
8cdea7c0 BS |
3853 | }; |
3854 | ||
8c7c6e34 KH |
3855 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3856 | static struct cftype memsw_cgroup_files[] = { | |
3857 | { | |
3858 | .name = "memsw.usage_in_bytes", | |
3859 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
3860 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
3861 | .register_event = mem_cgroup_usage_register_event, |
3862 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
3863 | }, |
3864 | { | |
3865 | .name = "memsw.max_usage_in_bytes", | |
3866 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
3867 | .trigger = mem_cgroup_reset, | |
3868 | .read_u64 = mem_cgroup_read, | |
3869 | }, | |
3870 | { | |
3871 | .name = "memsw.limit_in_bytes", | |
3872 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
3873 | .write_string = mem_cgroup_write, | |
3874 | .read_u64 = mem_cgroup_read, | |
3875 | }, | |
3876 | { | |
3877 | .name = "memsw.failcnt", | |
3878 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
3879 | .trigger = mem_cgroup_reset, | |
3880 | .read_u64 = mem_cgroup_read, | |
3881 | }, | |
3882 | }; | |
3883 | ||
3884 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3885 | { | |
3886 | if (!do_swap_account) | |
3887 | return 0; | |
3888 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
3889 | ARRAY_SIZE(memsw_cgroup_files)); | |
3890 | }; | |
3891 | #else | |
3892 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3893 | { | |
3894 | return 0; | |
3895 | } | |
3896 | #endif | |
3897 | ||
6d12e2d8 KH |
3898 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3899 | { | |
3900 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 3901 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 3902 | enum lru_list l; |
41e3355d | 3903 | int zone, tmp = node; |
1ecaab2b KH |
3904 | /* |
3905 | * This routine is called against possible nodes. | |
3906 | * But it's BUG to call kmalloc() against offline node. | |
3907 | * | |
3908 | * TODO: this routine can waste much memory for nodes which will | |
3909 | * never be onlined. It's better to use memory hotplug callback | |
3910 | * function. | |
3911 | */ | |
41e3355d KH |
3912 | if (!node_state(node, N_NORMAL_MEMORY)) |
3913 | tmp = -1; | |
3914 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
3915 | if (!pn) |
3916 | return 1; | |
1ecaab2b | 3917 | |
6d12e2d8 KH |
3918 | mem->info.nodeinfo[node] = pn; |
3919 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
3920 | |
3921 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3922 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
3923 | for_each_lru(l) |
3924 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 3925 | mz->usage_in_excess = 0; |
4e416953 BS |
3926 | mz->on_tree = false; |
3927 | mz->mem = mem; | |
1ecaab2b | 3928 | } |
6d12e2d8 KH |
3929 | return 0; |
3930 | } | |
3931 | ||
1ecaab2b KH |
3932 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3933 | { | |
3934 | kfree(mem->info.nodeinfo[node]); | |
3935 | } | |
3936 | ||
33327948 KH |
3937 | static struct mem_cgroup *mem_cgroup_alloc(void) |
3938 | { | |
3939 | struct mem_cgroup *mem; | |
c62b1a3b | 3940 | int size = sizeof(struct mem_cgroup); |
33327948 | 3941 | |
c62b1a3b | 3942 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb JB |
3943 | if (size < PAGE_SIZE) |
3944 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 3945 | else |
c8dad2bb | 3946 | mem = vmalloc(size); |
33327948 | 3947 | |
e7bbcdf3 DC |
3948 | if (!mem) |
3949 | return NULL; | |
3950 | ||
3951 | memset(mem, 0, size); | |
c62b1a3b KH |
3952 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
3953 | if (!mem->stat) { | |
3954 | if (size < PAGE_SIZE) | |
3955 | kfree(mem); | |
3956 | else | |
3957 | vfree(mem); | |
3958 | mem = NULL; | |
3959 | } | |
33327948 KH |
3960 | return mem; |
3961 | } | |
3962 | ||
8c7c6e34 KH |
3963 | /* |
3964 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
3965 | * (scanning all at force_empty is too costly...) | |
3966 | * | |
3967 | * Instead of clearing all references at force_empty, we remember | |
3968 | * the number of reference from swap_cgroup and free mem_cgroup when | |
3969 | * it goes down to 0. | |
3970 | * | |
8c7c6e34 KH |
3971 | * Removal of cgroup itself succeeds regardless of refs from swap. |
3972 | */ | |
3973 | ||
a7ba0eef | 3974 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 3975 | { |
08e552c6 KH |
3976 | int node; |
3977 | ||
f64c3f54 | 3978 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
3979 | free_css_id(&mem_cgroup_subsys, &mem->css); |
3980 | ||
08e552c6 KH |
3981 | for_each_node_state(node, N_POSSIBLE) |
3982 | free_mem_cgroup_per_zone_info(mem, node); | |
3983 | ||
c62b1a3b KH |
3984 | free_percpu(mem->stat); |
3985 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
3986 | kfree(mem); |
3987 | else | |
3988 | vfree(mem); | |
3989 | } | |
3990 | ||
8c7c6e34 KH |
3991 | static void mem_cgroup_get(struct mem_cgroup *mem) |
3992 | { | |
3993 | atomic_inc(&mem->refcnt); | |
3994 | } | |
3995 | ||
483c30b5 | 3996 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 3997 | { |
483c30b5 | 3998 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 3999 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 4000 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
4001 | if (parent) |
4002 | mem_cgroup_put(parent); | |
4003 | } | |
8c7c6e34 KH |
4004 | } |
4005 | ||
483c30b5 DN |
4006 | static void mem_cgroup_put(struct mem_cgroup *mem) |
4007 | { | |
4008 | __mem_cgroup_put(mem, 1); | |
4009 | } | |
4010 | ||
7bcc1bb1 DN |
4011 | /* |
4012 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4013 | */ | |
4014 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
4015 | { | |
4016 | if (!mem->res.parent) | |
4017 | return NULL; | |
4018 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
4019 | } | |
33327948 | 4020 | |
c077719b KH |
4021 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4022 | static void __init enable_swap_cgroup(void) | |
4023 | { | |
f8d66542 | 4024 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4025 | do_swap_account = 1; |
4026 | } | |
4027 | #else | |
4028 | static void __init enable_swap_cgroup(void) | |
4029 | { | |
4030 | } | |
4031 | #endif | |
4032 | ||
f64c3f54 BS |
4033 | static int mem_cgroup_soft_limit_tree_init(void) |
4034 | { | |
4035 | struct mem_cgroup_tree_per_node *rtpn; | |
4036 | struct mem_cgroup_tree_per_zone *rtpz; | |
4037 | int tmp, node, zone; | |
4038 | ||
4039 | for_each_node_state(node, N_POSSIBLE) { | |
4040 | tmp = node; | |
4041 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4042 | tmp = -1; | |
4043 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4044 | if (!rtpn) | |
4045 | return 1; | |
4046 | ||
4047 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4048 | ||
4049 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4050 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4051 | rtpz->rb_root = RB_ROOT; | |
4052 | spin_lock_init(&rtpz->lock); | |
4053 | } | |
4054 | } | |
4055 | return 0; | |
4056 | } | |
4057 | ||
0eb253e2 | 4058 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4059 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4060 | { | |
28dbc4b6 | 4061 | struct mem_cgroup *mem, *parent; |
04046e1a | 4062 | long error = -ENOMEM; |
6d12e2d8 | 4063 | int node; |
8cdea7c0 | 4064 | |
c8dad2bb JB |
4065 | mem = mem_cgroup_alloc(); |
4066 | if (!mem) | |
04046e1a | 4067 | return ERR_PTR(error); |
78fb7466 | 4068 | |
6d12e2d8 KH |
4069 | for_each_node_state(node, N_POSSIBLE) |
4070 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
4071 | goto free_out; | |
f64c3f54 | 4072 | |
c077719b | 4073 | /* root ? */ |
28dbc4b6 | 4074 | if (cont->parent == NULL) { |
cdec2e42 | 4075 | int cpu; |
c077719b | 4076 | enable_swap_cgroup(); |
28dbc4b6 | 4077 | parent = NULL; |
4b3bde4c | 4078 | root_mem_cgroup = mem; |
f64c3f54 BS |
4079 | if (mem_cgroup_soft_limit_tree_init()) |
4080 | goto free_out; | |
cdec2e42 KH |
4081 | for_each_possible_cpu(cpu) { |
4082 | struct memcg_stock_pcp *stock = | |
4083 | &per_cpu(memcg_stock, cpu); | |
4084 | INIT_WORK(&stock->work, drain_local_stock); | |
4085 | } | |
4086 | hotcpu_notifier(memcg_stock_cpu_callback, 0); | |
18f59ea7 | 4087 | } else { |
28dbc4b6 | 4088 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 | 4089 | mem->use_hierarchy = parent->use_hierarchy; |
3c11ecf4 | 4090 | mem->oom_kill_disable = parent->oom_kill_disable; |
18f59ea7 | 4091 | } |
28dbc4b6 | 4092 | |
18f59ea7 BS |
4093 | if (parent && parent->use_hierarchy) { |
4094 | res_counter_init(&mem->res, &parent->res); | |
4095 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4096 | /* |
4097 | * We increment refcnt of the parent to ensure that we can | |
4098 | * safely access it on res_counter_charge/uncharge. | |
4099 | * This refcnt will be decremented when freeing this | |
4100 | * mem_cgroup(see mem_cgroup_put). | |
4101 | */ | |
4102 | mem_cgroup_get(parent); | |
18f59ea7 BS |
4103 | } else { |
4104 | res_counter_init(&mem->res, NULL); | |
4105 | res_counter_init(&mem->memsw, NULL); | |
4106 | } | |
04046e1a | 4107 | mem->last_scanned_child = 0; |
2733c06a | 4108 | spin_lock_init(&mem->reclaim_param_lock); |
9490ff27 | 4109 | INIT_LIST_HEAD(&mem->oom_notify); |
6d61ef40 | 4110 | |
a7885eb8 KM |
4111 | if (parent) |
4112 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 4113 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 4114 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 4115 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 4116 | return &mem->css; |
6d12e2d8 | 4117 | free_out: |
a7ba0eef | 4118 | __mem_cgroup_free(mem); |
4b3bde4c | 4119 | root_mem_cgroup = NULL; |
04046e1a | 4120 | return ERR_PTR(error); |
8cdea7c0 BS |
4121 | } |
4122 | ||
ec64f515 | 4123 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4124 | struct cgroup *cont) |
4125 | { | |
4126 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
4127 | |
4128 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
4129 | } |
4130 | ||
8cdea7c0 BS |
4131 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4132 | struct cgroup *cont) | |
4133 | { | |
c268e994 | 4134 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 4135 | |
c268e994 | 4136 | mem_cgroup_put(mem); |
8cdea7c0 BS |
4137 | } |
4138 | ||
4139 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4140 | struct cgroup *cont) | |
4141 | { | |
8c7c6e34 KH |
4142 | int ret; |
4143 | ||
4144 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4145 | ARRAY_SIZE(mem_cgroup_files)); | |
4146 | ||
4147 | if (!ret) | |
4148 | ret = register_memsw_files(cont, ss); | |
4149 | return ret; | |
8cdea7c0 BS |
4150 | } |
4151 | ||
02491447 | 4152 | #ifdef CONFIG_MMU |
7dc74be0 | 4153 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4154 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4155 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4156 | { |
854ffa8d DN |
4157 | int ret = 0; |
4158 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
4159 | struct mem_cgroup *mem = mc.to; |
4160 | ||
854ffa8d DN |
4161 | if (mem_cgroup_is_root(mem)) { |
4162 | mc.precharge += count; | |
4163 | /* we don't need css_get for root */ | |
4164 | return ret; | |
4165 | } | |
4166 | /* try to charge at once */ | |
4167 | if (count > 1) { | |
4168 | struct res_counter *dummy; | |
4169 | /* | |
4170 | * "mem" cannot be under rmdir() because we've already checked | |
4171 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
4172 | * are still under the same cgroup_mutex. So we can postpone | |
4173 | * css_get(). | |
4174 | */ | |
4175 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
4176 | goto one_by_one; | |
4177 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
4178 | PAGE_SIZE * count, &dummy)) { | |
4179 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
4180 | goto one_by_one; | |
4181 | } | |
4182 | mc.precharge += count; | |
4183 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
4184 | WARN_ON_ONCE(count > INT_MAX); | |
4185 | __css_get(&mem->css, (int)count); | |
4186 | return ret; | |
4187 | } | |
4188 | one_by_one: | |
4189 | /* fall back to one by one charge */ | |
4190 | while (count--) { | |
4191 | if (signal_pending(current)) { | |
4192 | ret = -EINTR; | |
4193 | break; | |
4194 | } | |
4195 | if (!batch_count--) { | |
4196 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4197 | cond_resched(); | |
4198 | } | |
430e4863 | 4199 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
854ffa8d DN |
4200 | if (ret || !mem) |
4201 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
4202 | return -ENOMEM; | |
4203 | mc.precharge++; | |
4204 | } | |
4ffef5fe DN |
4205 | return ret; |
4206 | } | |
4207 | ||
4208 | /** | |
4209 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
4210 | * @vma: the vma the pte to be checked belongs | |
4211 | * @addr: the address corresponding to the pte to be checked | |
4212 | * @ptent: the pte to be checked | |
02491447 | 4213 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4214 | * |
4215 | * Returns | |
4216 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4217 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4218 | * move charge. if @target is not NULL, the page is stored in target->page | |
4219 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4220 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4221 | * target for charge migration. if @target is not NULL, the entry is stored | |
4222 | * in target->ent. | |
4ffef5fe DN |
4223 | * |
4224 | * Called with pte lock held. | |
4225 | */ | |
4ffef5fe DN |
4226 | union mc_target { |
4227 | struct page *page; | |
02491447 | 4228 | swp_entry_t ent; |
4ffef5fe DN |
4229 | }; |
4230 | ||
4ffef5fe DN |
4231 | enum mc_target_type { |
4232 | MC_TARGET_NONE, /* not used */ | |
4233 | MC_TARGET_PAGE, | |
02491447 | 4234 | MC_TARGET_SWAP, |
4ffef5fe DN |
4235 | }; |
4236 | ||
90254a65 DN |
4237 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4238 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4239 | { |
90254a65 | 4240 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4241 | |
90254a65 DN |
4242 | if (!page || !page_mapped(page)) |
4243 | return NULL; | |
4244 | if (PageAnon(page)) { | |
4245 | /* we don't move shared anon */ | |
4246 | if (!move_anon() || page_mapcount(page) > 2) | |
4247 | return NULL; | |
87946a72 DN |
4248 | } else if (!move_file()) |
4249 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
4250 | return NULL; |
4251 | if (!get_page_unless_zero(page)) | |
4252 | return NULL; | |
4253 | ||
4254 | return page; | |
4255 | } | |
4256 | ||
4257 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4258 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4259 | { | |
4260 | int usage_count; | |
4261 | struct page *page = NULL; | |
4262 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4263 | ||
4264 | if (!move_anon() || non_swap_entry(ent)) | |
4265 | return NULL; | |
4266 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
4267 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
4268 | if (page) |
4269 | put_page(page); | |
90254a65 | 4270 | return NULL; |
02491447 | 4271 | } |
90254a65 DN |
4272 | if (do_swap_account) |
4273 | entry->val = ent.val; | |
4274 | ||
4275 | return page; | |
4276 | } | |
4277 | ||
87946a72 DN |
4278 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4279 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4280 | { | |
4281 | struct page *page = NULL; | |
4282 | struct inode *inode; | |
4283 | struct address_space *mapping; | |
4284 | pgoff_t pgoff; | |
4285 | ||
4286 | if (!vma->vm_file) /* anonymous vma */ | |
4287 | return NULL; | |
4288 | if (!move_file()) | |
4289 | return NULL; | |
4290 | ||
4291 | inode = vma->vm_file->f_path.dentry->d_inode; | |
4292 | mapping = vma->vm_file->f_mapping; | |
4293 | if (pte_none(ptent)) | |
4294 | pgoff = linear_page_index(vma, addr); | |
4295 | else /* pte_file(ptent) is true */ | |
4296 | pgoff = pte_to_pgoff(ptent); | |
4297 | ||
4298 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
4299 | if (!mapping_cap_swap_backed(mapping)) { /* normal file */ | |
4300 | page = find_get_page(mapping, pgoff); | |
4301 | } else { /* shmem/tmpfs file. we should take account of swap too. */ | |
4302 | swp_entry_t ent; | |
4303 | mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent); | |
4304 | if (do_swap_account) | |
4305 | entry->val = ent.val; | |
4306 | } | |
4307 | ||
4308 | return page; | |
4309 | } | |
4310 | ||
90254a65 DN |
4311 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
4312 | unsigned long addr, pte_t ptent, union mc_target *target) | |
4313 | { | |
4314 | struct page *page = NULL; | |
4315 | struct page_cgroup *pc; | |
4316 | int ret = 0; | |
4317 | swp_entry_t ent = { .val = 0 }; | |
4318 | ||
4319 | if (pte_present(ptent)) | |
4320 | page = mc_handle_present_pte(vma, addr, ptent); | |
4321 | else if (is_swap_pte(ptent)) | |
4322 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
4323 | else if (pte_none(ptent) || pte_file(ptent)) |
4324 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
4325 | |
4326 | if (!page && !ent.val) | |
4327 | return 0; | |
02491447 DN |
4328 | if (page) { |
4329 | pc = lookup_page_cgroup(page); | |
4330 | /* | |
4331 | * Do only loose check w/o page_cgroup lock. | |
4332 | * mem_cgroup_move_account() checks the pc is valid or not under | |
4333 | * the lock. | |
4334 | */ | |
4335 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
4336 | ret = MC_TARGET_PAGE; | |
4337 | if (target) | |
4338 | target->page = page; | |
4339 | } | |
4340 | if (!ret || !target) | |
4341 | put_page(page); | |
4342 | } | |
90254a65 DN |
4343 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4344 | if (ent.val && !ret && | |
7f0f1546 KH |
4345 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
4346 | ret = MC_TARGET_SWAP; | |
4347 | if (target) | |
4348 | target->ent = ent; | |
4ffef5fe | 4349 | } |
4ffef5fe DN |
4350 | return ret; |
4351 | } | |
4352 | ||
4353 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
4354 | unsigned long addr, unsigned long end, | |
4355 | struct mm_walk *walk) | |
4356 | { | |
4357 | struct vm_area_struct *vma = walk->private; | |
4358 | pte_t *pte; | |
4359 | spinlock_t *ptl; | |
4360 | ||
4361 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4362 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
4363 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
4364 | mc.precharge++; /* increment precharge temporarily */ | |
4365 | pte_unmap_unlock(pte - 1, ptl); | |
4366 | cond_resched(); | |
4367 | ||
7dc74be0 DN |
4368 | return 0; |
4369 | } | |
4370 | ||
4ffef5fe DN |
4371 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4372 | { | |
4373 | unsigned long precharge; | |
4374 | struct vm_area_struct *vma; | |
4375 | ||
4376 | down_read(&mm->mmap_sem); | |
4377 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
4378 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
4379 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4380 | .mm = mm, | |
4381 | .private = vma, | |
4382 | }; | |
4383 | if (is_vm_hugetlb_page(vma)) | |
4384 | continue; | |
4ffef5fe DN |
4385 | walk_page_range(vma->vm_start, vma->vm_end, |
4386 | &mem_cgroup_count_precharge_walk); | |
4387 | } | |
4388 | up_read(&mm->mmap_sem); | |
4389 | ||
4390 | precharge = mc.precharge; | |
4391 | mc.precharge = 0; | |
4392 | ||
4393 | return precharge; | |
4394 | } | |
4395 | ||
4ffef5fe DN |
4396 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4397 | { | |
854ffa8d | 4398 | return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm)); |
4ffef5fe DN |
4399 | } |
4400 | ||
4401 | static void mem_cgroup_clear_mc(void) | |
4402 | { | |
4403 | /* we must uncharge all the leftover precharges from mc.to */ | |
854ffa8d DN |
4404 | if (mc.precharge) { |
4405 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
4406 | mc.precharge = 0; | |
3c11ecf4 | 4407 | memcg_oom_recover(mc.to); |
854ffa8d DN |
4408 | } |
4409 | /* | |
4410 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4411 | * we must uncharge here. | |
4412 | */ | |
4413 | if (mc.moved_charge) { | |
4414 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
4415 | mc.moved_charge = 0; | |
3c11ecf4 | 4416 | memcg_oom_recover(mc.from); |
4ffef5fe | 4417 | } |
483c30b5 DN |
4418 | /* we must fixup refcnts and charges */ |
4419 | if (mc.moved_swap) { | |
4420 | WARN_ON_ONCE(mc.moved_swap > INT_MAX); | |
4421 | /* uncharge swap account from the old cgroup */ | |
4422 | if (!mem_cgroup_is_root(mc.from)) | |
4423 | res_counter_uncharge(&mc.from->memsw, | |
4424 | PAGE_SIZE * mc.moved_swap); | |
4425 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
4426 | ||
4427 | if (!mem_cgroup_is_root(mc.to)) { | |
4428 | /* | |
4429 | * we charged both to->res and to->memsw, so we should | |
4430 | * uncharge to->res. | |
4431 | */ | |
4432 | res_counter_uncharge(&mc.to->res, | |
4433 | PAGE_SIZE * mc.moved_swap); | |
4434 | VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags)); | |
4435 | __css_put(&mc.to->css, mc.moved_swap); | |
4436 | } | |
4437 | /* we've already done mem_cgroup_get(mc.to) */ | |
4438 | ||
4439 | mc.moved_swap = 0; | |
4440 | } | |
4ffef5fe DN |
4441 | mc.from = NULL; |
4442 | mc.to = NULL; | |
8033b97c DN |
4443 | mc.moving_task = NULL; |
4444 | wake_up_all(&mc.waitq); | |
4ffef5fe DN |
4445 | } |
4446 | ||
7dc74be0 DN |
4447 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
4448 | struct cgroup *cgroup, | |
4449 | struct task_struct *p, | |
4450 | bool threadgroup) | |
4451 | { | |
4452 | int ret = 0; | |
4453 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
4454 | ||
4455 | if (mem->move_charge_at_immigrate) { | |
4456 | struct mm_struct *mm; | |
4457 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
4458 | ||
4459 | VM_BUG_ON(from == mem); | |
4460 | ||
4461 | mm = get_task_mm(p); | |
4462 | if (!mm) | |
4463 | return 0; | |
7dc74be0 | 4464 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
4465 | if (mm->owner == p) { |
4466 | VM_BUG_ON(mc.from); | |
4467 | VM_BUG_ON(mc.to); | |
4468 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 4469 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 4470 | VM_BUG_ON(mc.moved_swap); |
8033b97c | 4471 | VM_BUG_ON(mc.moving_task); |
4ffef5fe DN |
4472 | mc.from = from; |
4473 | mc.to = mem; | |
4474 | mc.precharge = 0; | |
854ffa8d | 4475 | mc.moved_charge = 0; |
483c30b5 | 4476 | mc.moved_swap = 0; |
8033b97c | 4477 | mc.moving_task = current; |
4ffef5fe DN |
4478 | |
4479 | ret = mem_cgroup_precharge_mc(mm); | |
4480 | if (ret) | |
4481 | mem_cgroup_clear_mc(); | |
4482 | } | |
7dc74be0 DN |
4483 | mmput(mm); |
4484 | } | |
4485 | return ret; | |
4486 | } | |
4487 | ||
4488 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4489 | struct cgroup *cgroup, | |
4490 | struct task_struct *p, | |
4491 | bool threadgroup) | |
4492 | { | |
4ffef5fe | 4493 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
4494 | } |
4495 | ||
4ffef5fe DN |
4496 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4497 | unsigned long addr, unsigned long end, | |
4498 | struct mm_walk *walk) | |
7dc74be0 | 4499 | { |
4ffef5fe DN |
4500 | int ret = 0; |
4501 | struct vm_area_struct *vma = walk->private; | |
4502 | pte_t *pte; | |
4503 | spinlock_t *ptl; | |
4504 | ||
4505 | retry: | |
4506 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4507 | for (; addr != end; addr += PAGE_SIZE) { | |
4508 | pte_t ptent = *(pte++); | |
4509 | union mc_target target; | |
4510 | int type; | |
4511 | struct page *page; | |
4512 | struct page_cgroup *pc; | |
02491447 | 4513 | swp_entry_t ent; |
4ffef5fe DN |
4514 | |
4515 | if (!mc.precharge) | |
4516 | break; | |
4517 | ||
4518 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
4519 | switch (type) { | |
4520 | case MC_TARGET_PAGE: | |
4521 | page = target.page; | |
4522 | if (isolate_lru_page(page)) | |
4523 | goto put; | |
4524 | pc = lookup_page_cgroup(page); | |
854ffa8d DN |
4525 | if (!mem_cgroup_move_account(pc, |
4526 | mc.from, mc.to, false)) { | |
4ffef5fe | 4527 | mc.precharge--; |
854ffa8d DN |
4528 | /* we uncharge from mc.from later. */ |
4529 | mc.moved_charge++; | |
4ffef5fe DN |
4530 | } |
4531 | putback_lru_page(page); | |
4532 | put: /* is_target_pte_for_mc() gets the page */ | |
4533 | put_page(page); | |
4534 | break; | |
02491447 DN |
4535 | case MC_TARGET_SWAP: |
4536 | ent = target.ent; | |
483c30b5 DN |
4537 | if (!mem_cgroup_move_swap_account(ent, |
4538 | mc.from, mc.to, false)) { | |
02491447 | 4539 | mc.precharge--; |
483c30b5 DN |
4540 | /* we fixup refcnts and charges later. */ |
4541 | mc.moved_swap++; | |
4542 | } | |
02491447 | 4543 | break; |
4ffef5fe DN |
4544 | default: |
4545 | break; | |
4546 | } | |
4547 | } | |
4548 | pte_unmap_unlock(pte - 1, ptl); | |
4549 | cond_resched(); | |
4550 | ||
4551 | if (addr != end) { | |
4552 | /* | |
4553 | * We have consumed all precharges we got in can_attach(). | |
4554 | * We try charge one by one, but don't do any additional | |
4555 | * charges to mc.to if we have failed in charge once in attach() | |
4556 | * phase. | |
4557 | */ | |
854ffa8d | 4558 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4559 | if (!ret) |
4560 | goto retry; | |
4561 | } | |
4562 | ||
4563 | return ret; | |
4564 | } | |
4565 | ||
4566 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4567 | { | |
4568 | struct vm_area_struct *vma; | |
4569 | ||
4570 | lru_add_drain_all(); | |
4571 | down_read(&mm->mmap_sem); | |
4572 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
4573 | int ret; | |
4574 | struct mm_walk mem_cgroup_move_charge_walk = { | |
4575 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4576 | .mm = mm, | |
4577 | .private = vma, | |
4578 | }; | |
4579 | if (is_vm_hugetlb_page(vma)) | |
4580 | continue; | |
4ffef5fe DN |
4581 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
4582 | &mem_cgroup_move_charge_walk); | |
4583 | if (ret) | |
4584 | /* | |
4585 | * means we have consumed all precharges and failed in | |
4586 | * doing additional charge. Just abandon here. | |
4587 | */ | |
4588 | break; | |
4589 | } | |
4590 | up_read(&mm->mmap_sem); | |
7dc74be0 DN |
4591 | } |
4592 | ||
67e465a7 BS |
4593 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
4594 | struct cgroup *cont, | |
4595 | struct cgroup *old_cont, | |
be367d09 BB |
4596 | struct task_struct *p, |
4597 | bool threadgroup) | |
67e465a7 | 4598 | { |
4ffef5fe DN |
4599 | struct mm_struct *mm; |
4600 | ||
4601 | if (!mc.to) | |
4602 | /* no need to move charge */ | |
4603 | return; | |
4604 | ||
4605 | mm = get_task_mm(p); | |
4606 | if (mm) { | |
4607 | mem_cgroup_move_charge(mm); | |
4608 | mmput(mm); | |
4609 | } | |
4610 | mem_cgroup_clear_mc(); | |
67e465a7 | 4611 | } |
5cfb80a7 DN |
4612 | #else /* !CONFIG_MMU */ |
4613 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
4614 | struct cgroup *cgroup, | |
4615 | struct task_struct *p, | |
4616 | bool threadgroup) | |
4617 | { | |
4618 | return 0; | |
4619 | } | |
4620 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4621 | struct cgroup *cgroup, | |
4622 | struct task_struct *p, | |
4623 | bool threadgroup) | |
4624 | { | |
4625 | } | |
4626 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
4627 | struct cgroup *cont, | |
4628 | struct cgroup *old_cont, | |
4629 | struct task_struct *p, | |
4630 | bool threadgroup) | |
4631 | { | |
4632 | } | |
4633 | #endif | |
67e465a7 | 4634 | |
8cdea7c0 BS |
4635 | struct cgroup_subsys mem_cgroup_subsys = { |
4636 | .name = "memory", | |
4637 | .subsys_id = mem_cgroup_subsys_id, | |
4638 | .create = mem_cgroup_create, | |
df878fb0 | 4639 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
4640 | .destroy = mem_cgroup_destroy, |
4641 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
4642 | .can_attach = mem_cgroup_can_attach, |
4643 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 4644 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 4645 | .early_init = 0, |
04046e1a | 4646 | .use_id = 1, |
8cdea7c0 | 4647 | }; |
c077719b KH |
4648 | |
4649 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
4650 | ||
4651 | static int __init disable_swap_account(char *s) | |
4652 | { | |
4653 | really_do_swap_account = 0; | |
4654 | return 1; | |
4655 | } | |
4656 | __setup("noswapaccount", disable_swap_account); | |
4657 | #endif |