]>
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 | * | |
8cdea7c0 BS |
9 | * This program is free software; you can redistribute it and/or modify |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | */ | |
19 | ||
20 | #include <linux/res_counter.h> | |
21 | #include <linux/memcontrol.h> | |
22 | #include <linux/cgroup.h> | |
78fb7466 | 23 | #include <linux/mm.h> |
4ffef5fe | 24 | #include <linux/hugetlb.h> |
d13d1443 | 25 | #include <linux/pagemap.h> |
d52aa412 | 26 | #include <linux/smp.h> |
8a9f3ccd | 27 | #include <linux/page-flags.h> |
66e1707b | 28 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
29 | #include <linux/bit_spinlock.h> |
30 | #include <linux/rcupdate.h> | |
e222432b | 31 | #include <linux/limits.h> |
8c7c6e34 | 32 | #include <linux/mutex.h> |
f64c3f54 | 33 | #include <linux/rbtree.h> |
b6ac57d5 | 34 | #include <linux/slab.h> |
66e1707b BS |
35 | #include <linux/swap.h> |
36 | #include <linux/spinlock.h> | |
37 | #include <linux/fs.h> | |
d2ceb9b7 | 38 | #include <linux/seq_file.h> |
33327948 | 39 | #include <linux/vmalloc.h> |
b69408e8 | 40 | #include <linux/mm_inline.h> |
52d4b9ac | 41 | #include <linux/page_cgroup.h> |
cdec2e42 | 42 | #include <linux/cpu.h> |
08e552c6 | 43 | #include "internal.h" |
8cdea7c0 | 44 | |
8697d331 BS |
45 | #include <asm/uaccess.h> |
46 | ||
a181b0e8 | 47 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 48 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 49 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 50 | |
c077719b | 51 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 52 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b KH |
53 | int do_swap_account __read_mostly; |
54 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
55 | #else | |
56 | #define do_swap_account (0) | |
57 | #endif | |
58 | ||
f64c3f54 | 59 | #define SOFTLIMIT_EVENTS_THRESH (1000) |
c077719b | 60 | |
d52aa412 KH |
61 | /* |
62 | * Statistics for memory cgroup. | |
63 | */ | |
64 | enum mem_cgroup_stat_index { | |
65 | /* | |
66 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
67 | */ | |
68 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 69 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 70 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
55e462b0 BR |
71 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
72 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
f64c3f54 | 73 | MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */ |
0c3e73e8 | 74 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
d52aa412 KH |
75 | |
76 | MEM_CGROUP_STAT_NSTATS, | |
77 | }; | |
78 | ||
79 | struct mem_cgroup_stat_cpu { | |
80 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
81 | } ____cacheline_aligned_in_smp; | |
82 | ||
83 | struct mem_cgroup_stat { | |
c8dad2bb | 84 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
85 | }; |
86 | ||
f64c3f54 BS |
87 | static inline void |
88 | __mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat, | |
89 | enum mem_cgroup_stat_index idx) | |
90 | { | |
91 | stat->count[idx] = 0; | |
92 | } | |
93 | ||
94 | static inline s64 | |
95 | __mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat, | |
96 | enum mem_cgroup_stat_index idx) | |
97 | { | |
98 | return stat->count[idx]; | |
99 | } | |
100 | ||
d52aa412 KH |
101 | /* |
102 | * For accounting under irq disable, no need for increment preempt count. | |
103 | */ | |
addb9efe | 104 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
105 | enum mem_cgroup_stat_index idx, int val) |
106 | { | |
addb9efe | 107 | stat->count[idx] += val; |
d52aa412 KH |
108 | } |
109 | ||
110 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
111 | enum mem_cgroup_stat_index idx) | |
112 | { | |
113 | int cpu; | |
114 | s64 ret = 0; | |
115 | for_each_possible_cpu(cpu) | |
116 | ret += stat->cpustat[cpu].count[idx]; | |
117 | return ret; | |
118 | } | |
119 | ||
04046e1a KH |
120 | static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat) |
121 | { | |
122 | s64 ret; | |
123 | ||
124 | ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE); | |
125 | ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS); | |
126 | return ret; | |
127 | } | |
128 | ||
6d12e2d8 KH |
129 | /* |
130 | * per-zone information in memory controller. | |
131 | */ | |
6d12e2d8 | 132 | struct mem_cgroup_per_zone { |
072c56c1 KH |
133 | /* |
134 | * spin_lock to protect the per cgroup LRU | |
135 | */ | |
b69408e8 CL |
136 | struct list_head lists[NR_LRU_LISTS]; |
137 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
138 | |
139 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
140 | struct rb_node tree_node; /* RB tree node */ |
141 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
142 | /* the soft limit is exceeded*/ | |
143 | bool on_tree; | |
4e416953 BS |
144 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
145 | /* use container_of */ | |
6d12e2d8 KH |
146 | }; |
147 | /* Macro for accessing counter */ | |
148 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
149 | ||
150 | struct mem_cgroup_per_node { | |
151 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
152 | }; | |
153 | ||
154 | struct mem_cgroup_lru_info { | |
155 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
156 | }; | |
157 | ||
f64c3f54 BS |
158 | /* |
159 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
160 | * their hierarchy representation | |
161 | */ | |
162 | ||
163 | struct mem_cgroup_tree_per_zone { | |
164 | struct rb_root rb_root; | |
165 | spinlock_t lock; | |
166 | }; | |
167 | ||
168 | struct mem_cgroup_tree_per_node { | |
169 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
170 | }; | |
171 | ||
172 | struct mem_cgroup_tree { | |
173 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
174 | }; | |
175 | ||
176 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
177 | ||
8cdea7c0 BS |
178 | /* |
179 | * The memory controller data structure. The memory controller controls both | |
180 | * page cache and RSS per cgroup. We would eventually like to provide | |
181 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
182 | * to help the administrator determine what knobs to tune. | |
183 | * | |
184 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
185 | * we hit the water mark. May be even add a low water mark, such that |
186 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
187 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
188 | */ |
189 | struct mem_cgroup { | |
190 | struct cgroup_subsys_state css; | |
191 | /* | |
192 | * the counter to account for memory usage | |
193 | */ | |
194 | struct res_counter res; | |
8c7c6e34 KH |
195 | /* |
196 | * the counter to account for mem+swap usage. | |
197 | */ | |
198 | struct res_counter memsw; | |
78fb7466 PE |
199 | /* |
200 | * Per cgroup active and inactive list, similar to the | |
201 | * per zone LRU lists. | |
78fb7466 | 202 | */ |
6d12e2d8 | 203 | struct mem_cgroup_lru_info info; |
072c56c1 | 204 | |
2733c06a KM |
205 | /* |
206 | protect against reclaim related member. | |
207 | */ | |
208 | spinlock_t reclaim_param_lock; | |
209 | ||
6c48a1d0 | 210 | int prev_priority; /* for recording reclaim priority */ |
6d61ef40 BS |
211 | |
212 | /* | |
af901ca1 | 213 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 214 | * reclaimed from. |
6d61ef40 | 215 | */ |
04046e1a | 216 | int last_scanned_child; |
18f59ea7 BS |
217 | /* |
218 | * Should the accounting and control be hierarchical, per subtree? | |
219 | */ | |
220 | bool use_hierarchy; | |
a636b327 | 221 | unsigned long last_oom_jiffies; |
8c7c6e34 | 222 | atomic_t refcnt; |
14797e23 | 223 | |
a7885eb8 KM |
224 | unsigned int swappiness; |
225 | ||
22a668d7 KH |
226 | /* set when res.limit == memsw.limit */ |
227 | bool memsw_is_minimum; | |
228 | ||
7dc74be0 DN |
229 | /* |
230 | * Should we move charges of a task when a task is moved into this | |
231 | * mem_cgroup ? And what type of charges should we move ? | |
232 | */ | |
233 | unsigned long move_charge_at_immigrate; | |
234 | ||
d52aa412 | 235 | /* |
c8dad2bb | 236 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
237 | */ |
238 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
239 | }; |
240 | ||
7dc74be0 DN |
241 | /* Stuffs for move charges at task migration. */ |
242 | /* | |
243 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
244 | * left-shifted bitmap of these types. | |
245 | */ | |
246 | enum move_type { | |
4ffef5fe | 247 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
7dc74be0 DN |
248 | NR_MOVE_TYPE, |
249 | }; | |
250 | ||
4ffef5fe DN |
251 | /* "mc" and its members are protected by cgroup_mutex */ |
252 | static struct move_charge_struct { | |
253 | struct mem_cgroup *from; | |
254 | struct mem_cgroup *to; | |
255 | unsigned long precharge; | |
854ffa8d | 256 | unsigned long moved_charge; |
8033b97c DN |
257 | struct task_struct *moving_task; /* a task moving charges */ |
258 | wait_queue_head_t waitq; /* a waitq for other context */ | |
259 | } mc = { | |
260 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), | |
261 | }; | |
4ffef5fe | 262 | |
4e416953 BS |
263 | /* |
264 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
265 | * limit reclaim to prevent infinite loops, if they ever occur. | |
266 | */ | |
267 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
268 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
269 | ||
217bc319 KH |
270 | enum charge_type { |
271 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
272 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 273 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 274 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 275 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 276 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
277 | NR_CHARGE_TYPE, |
278 | }; | |
279 | ||
52d4b9ac KH |
280 | /* only for here (for easy reading.) */ |
281 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
282 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 283 | #define PCGF_LOCK (1UL << PCG_LOCK) |
4b3bde4c BS |
284 | /* Not used, but added here for completeness */ |
285 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
217bc319 | 286 | |
8c7c6e34 KH |
287 | /* for encoding cft->private value on file */ |
288 | #define _MEM (0) | |
289 | #define _MEMSWAP (1) | |
290 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
291 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
292 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
293 | ||
75822b44 BS |
294 | /* |
295 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
296 | */ | |
297 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
298 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
299 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
300 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
301 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
302 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 303 | |
8c7c6e34 KH |
304 | static void mem_cgroup_get(struct mem_cgroup *mem); |
305 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 306 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 307 | static void drain_all_stock_async(void); |
8c7c6e34 | 308 | |
f64c3f54 BS |
309 | static struct mem_cgroup_per_zone * |
310 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
311 | { | |
312 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
313 | } | |
314 | ||
d324236b WF |
315 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
316 | { | |
317 | return &mem->css; | |
318 | } | |
319 | ||
f64c3f54 BS |
320 | static struct mem_cgroup_per_zone * |
321 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
322 | { | |
323 | struct mem_cgroup *mem = pc->mem_cgroup; | |
324 | int nid = page_cgroup_nid(pc); | |
325 | int zid = page_cgroup_zid(pc); | |
326 | ||
327 | if (!mem) | |
328 | return NULL; | |
329 | ||
330 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
331 | } | |
332 | ||
333 | static struct mem_cgroup_tree_per_zone * | |
334 | soft_limit_tree_node_zone(int nid, int zid) | |
335 | { | |
336 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
337 | } | |
338 | ||
339 | static struct mem_cgroup_tree_per_zone * | |
340 | soft_limit_tree_from_page(struct page *page) | |
341 | { | |
342 | int nid = page_to_nid(page); | |
343 | int zid = page_zonenum(page); | |
344 | ||
345 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
346 | } | |
347 | ||
348 | static void | |
4e416953 | 349 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 350 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
351 | struct mem_cgroup_tree_per_zone *mctz, |
352 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
353 | { |
354 | struct rb_node **p = &mctz->rb_root.rb_node; | |
355 | struct rb_node *parent = NULL; | |
356 | struct mem_cgroup_per_zone *mz_node; | |
357 | ||
358 | if (mz->on_tree) | |
359 | return; | |
360 | ||
ef8745c1 KH |
361 | mz->usage_in_excess = new_usage_in_excess; |
362 | if (!mz->usage_in_excess) | |
363 | return; | |
f64c3f54 BS |
364 | while (*p) { |
365 | parent = *p; | |
366 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
367 | tree_node); | |
368 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
369 | p = &(*p)->rb_left; | |
370 | /* | |
371 | * We can't avoid mem cgroups that are over their soft | |
372 | * limit by the same amount | |
373 | */ | |
374 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
375 | p = &(*p)->rb_right; | |
376 | } | |
377 | rb_link_node(&mz->tree_node, parent, p); | |
378 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
379 | mz->on_tree = true; | |
4e416953 BS |
380 | } |
381 | ||
382 | static void | |
383 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
384 | struct mem_cgroup_per_zone *mz, | |
385 | struct mem_cgroup_tree_per_zone *mctz) | |
386 | { | |
387 | if (!mz->on_tree) | |
388 | return; | |
389 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
390 | mz->on_tree = false; | |
391 | } | |
392 | ||
f64c3f54 BS |
393 | static void |
394 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
395 | struct mem_cgroup_per_zone *mz, | |
396 | struct mem_cgroup_tree_per_zone *mctz) | |
397 | { | |
398 | spin_lock(&mctz->lock); | |
4e416953 | 399 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
400 | spin_unlock(&mctz->lock); |
401 | } | |
402 | ||
403 | static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem) | |
404 | { | |
405 | bool ret = false; | |
406 | int cpu; | |
407 | s64 val; | |
408 | struct mem_cgroup_stat_cpu *cpustat; | |
409 | ||
410 | cpu = get_cpu(); | |
411 | cpustat = &mem->stat.cpustat[cpu]; | |
412 | val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS); | |
413 | if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) { | |
414 | __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS); | |
415 | ret = true; | |
416 | } | |
417 | put_cpu(); | |
418 | return ret; | |
419 | } | |
420 | ||
421 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
422 | { | |
ef8745c1 | 423 | unsigned long long excess; |
f64c3f54 BS |
424 | struct mem_cgroup_per_zone *mz; |
425 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
426 | int nid = page_to_nid(page); |
427 | int zid = page_zonenum(page); | |
f64c3f54 BS |
428 | mctz = soft_limit_tree_from_page(page); |
429 | ||
430 | /* | |
4e649152 KH |
431 | * Necessary to update all ancestors when hierarchy is used. |
432 | * because their event counter is not touched. | |
f64c3f54 | 433 | */ |
4e649152 KH |
434 | for (; mem; mem = parent_mem_cgroup(mem)) { |
435 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 436 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
437 | /* |
438 | * We have to update the tree if mz is on RB-tree or | |
439 | * mem is over its softlimit. | |
440 | */ | |
ef8745c1 | 441 | if (excess || mz->on_tree) { |
4e649152 KH |
442 | spin_lock(&mctz->lock); |
443 | /* if on-tree, remove it */ | |
444 | if (mz->on_tree) | |
445 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
446 | /* | |
ef8745c1 KH |
447 | * Insert again. mz->usage_in_excess will be updated. |
448 | * If excess is 0, no tree ops. | |
4e649152 | 449 | */ |
ef8745c1 | 450 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
451 | spin_unlock(&mctz->lock); |
452 | } | |
f64c3f54 BS |
453 | } |
454 | } | |
455 | ||
456 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
457 | { | |
458 | int node, zone; | |
459 | struct mem_cgroup_per_zone *mz; | |
460 | struct mem_cgroup_tree_per_zone *mctz; | |
461 | ||
462 | for_each_node_state(node, N_POSSIBLE) { | |
463 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
464 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
465 | mctz = soft_limit_tree_node_zone(node, zone); | |
466 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
467 | } | |
468 | } | |
469 | } | |
470 | ||
4e416953 BS |
471 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
472 | { | |
473 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; | |
474 | } | |
475 | ||
476 | static struct mem_cgroup_per_zone * | |
477 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
478 | { | |
479 | struct rb_node *rightmost = NULL; | |
26251eaf | 480 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
481 | |
482 | retry: | |
26251eaf | 483 | mz = NULL; |
4e416953 BS |
484 | rightmost = rb_last(&mctz->rb_root); |
485 | if (!rightmost) | |
486 | goto done; /* Nothing to reclaim from */ | |
487 | ||
488 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
489 | /* | |
490 | * Remove the node now but someone else can add it back, | |
491 | * we will to add it back at the end of reclaim to its correct | |
492 | * position in the tree. | |
493 | */ | |
494 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
495 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
496 | !css_tryget(&mz->mem->css)) | |
497 | goto retry; | |
498 | done: | |
499 | return mz; | |
500 | } | |
501 | ||
502 | static struct mem_cgroup_per_zone * | |
503 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
504 | { | |
505 | struct mem_cgroup_per_zone *mz; | |
506 | ||
507 | spin_lock(&mctz->lock); | |
508 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
509 | spin_unlock(&mctz->lock); | |
510 | return mz; | |
511 | } | |
512 | ||
0c3e73e8 BS |
513 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
514 | bool charge) | |
515 | { | |
516 | int val = (charge) ? 1 : -1; | |
517 | struct mem_cgroup_stat *stat = &mem->stat; | |
518 | struct mem_cgroup_stat_cpu *cpustat; | |
519 | int cpu = get_cpu(); | |
520 | ||
521 | cpustat = &stat->cpustat[cpu]; | |
522 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SWAPOUT, val); | |
523 | put_cpu(); | |
524 | } | |
525 | ||
c05555b5 KH |
526 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
527 | struct page_cgroup *pc, | |
528 | bool charge) | |
d52aa412 | 529 | { |
0c3e73e8 | 530 | int val = (charge) ? 1 : -1; |
d52aa412 | 531 | struct mem_cgroup_stat *stat = &mem->stat; |
addb9efe | 532 | struct mem_cgroup_stat_cpu *cpustat; |
08e552c6 | 533 | int cpu = get_cpu(); |
d52aa412 | 534 | |
08e552c6 | 535 | cpustat = &stat->cpustat[cpu]; |
c05555b5 | 536 | if (PageCgroupCache(pc)) |
addb9efe | 537 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 538 | else |
addb9efe | 539 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
540 | |
541 | if (charge) | |
addb9efe | 542 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
543 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
544 | else | |
addb9efe | 545 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 546 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
f64c3f54 | 547 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1); |
08e552c6 | 548 | put_cpu(); |
6d12e2d8 KH |
549 | } |
550 | ||
14067bb3 | 551 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 552 | enum lru_list idx) |
6d12e2d8 KH |
553 | { |
554 | int nid, zid; | |
555 | struct mem_cgroup_per_zone *mz; | |
556 | u64 total = 0; | |
557 | ||
558 | for_each_online_node(nid) | |
559 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
560 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
561 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
562 | } | |
563 | return total; | |
d52aa412 KH |
564 | } |
565 | ||
d5b69e38 | 566 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
567 | { |
568 | return container_of(cgroup_subsys_state(cont, | |
569 | mem_cgroup_subsys_id), struct mem_cgroup, | |
570 | css); | |
571 | } | |
572 | ||
cf475ad2 | 573 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 574 | { |
31a78f23 BS |
575 | /* |
576 | * mm_update_next_owner() may clear mm->owner to NULL | |
577 | * if it races with swapoff, page migration, etc. | |
578 | * So this can be called with p == NULL. | |
579 | */ | |
580 | if (unlikely(!p)) | |
581 | return NULL; | |
582 | ||
78fb7466 PE |
583 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
584 | struct mem_cgroup, css); | |
585 | } | |
586 | ||
54595fe2 KH |
587 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
588 | { | |
589 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
590 | |
591 | if (!mm) | |
592 | return NULL; | |
54595fe2 KH |
593 | /* |
594 | * Because we have no locks, mm->owner's may be being moved to other | |
595 | * cgroup. We use css_tryget() here even if this looks | |
596 | * pessimistic (rather than adding locks here). | |
597 | */ | |
598 | rcu_read_lock(); | |
599 | do { | |
600 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
601 | if (unlikely(!mem)) | |
602 | break; | |
603 | } while (!css_tryget(&mem->css)); | |
604 | rcu_read_unlock(); | |
605 | return mem; | |
606 | } | |
607 | ||
14067bb3 KH |
608 | /* |
609 | * Call callback function against all cgroup under hierarchy tree. | |
610 | */ | |
611 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
612 | int (*func)(struct mem_cgroup *, void *)) | |
613 | { | |
614 | int found, ret, nextid; | |
615 | struct cgroup_subsys_state *css; | |
616 | struct mem_cgroup *mem; | |
617 | ||
618 | if (!root->use_hierarchy) | |
619 | return (*func)(root, data); | |
620 | ||
621 | nextid = 1; | |
622 | do { | |
623 | ret = 0; | |
624 | mem = NULL; | |
625 | ||
626 | rcu_read_lock(); | |
627 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
628 | &found); | |
629 | if (css && css_tryget(css)) | |
630 | mem = container_of(css, struct mem_cgroup, css); | |
631 | rcu_read_unlock(); | |
632 | ||
633 | if (mem) { | |
634 | ret = (*func)(mem, data); | |
635 | css_put(&mem->css); | |
636 | } | |
637 | nextid = found + 1; | |
638 | } while (!ret && css); | |
639 | ||
640 | return ret; | |
641 | } | |
642 | ||
4b3bde4c BS |
643 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
644 | { | |
645 | return (mem == root_mem_cgroup); | |
646 | } | |
647 | ||
08e552c6 KH |
648 | /* |
649 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
650 | * Operations are called by routine of global LRU independently from memcg. | |
651 | * What we have to take care of here is validness of pc->mem_cgroup. | |
652 | * | |
653 | * Changes to pc->mem_cgroup happens when | |
654 | * 1. charge | |
655 | * 2. moving account | |
656 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
657 | * It is added to LRU before charge. | |
658 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
659 | * When moving account, the page is not on LRU. It's isolated. | |
660 | */ | |
4f98a2fe | 661 | |
08e552c6 KH |
662 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
663 | { | |
664 | struct page_cgroup *pc; | |
08e552c6 | 665 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 666 | |
f8d66542 | 667 | if (mem_cgroup_disabled()) |
08e552c6 KH |
668 | return; |
669 | pc = lookup_page_cgroup(page); | |
670 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 671 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 672 | return; |
4b3bde4c | 673 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
674 | /* |
675 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
676 | * removed from global LRU. | |
677 | */ | |
08e552c6 | 678 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 679 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
4b3bde4c BS |
680 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
681 | return; | |
682 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 KH |
683 | list_del_init(&pc->lru); |
684 | return; | |
6d12e2d8 KH |
685 | } |
686 | ||
08e552c6 | 687 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 688 | { |
08e552c6 KH |
689 | mem_cgroup_del_lru_list(page, page_lru(page)); |
690 | } | |
b69408e8 | 691 | |
08e552c6 KH |
692 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
693 | { | |
694 | struct mem_cgroup_per_zone *mz; | |
695 | struct page_cgroup *pc; | |
b69408e8 | 696 | |
f8d66542 | 697 | if (mem_cgroup_disabled()) |
08e552c6 | 698 | return; |
6d12e2d8 | 699 | |
08e552c6 | 700 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
701 | /* |
702 | * Used bit is set without atomic ops but after smp_wmb(). | |
703 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
704 | */ | |
08e552c6 | 705 | smp_rmb(); |
4b3bde4c BS |
706 | /* unused or root page is not rotated. */ |
707 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 KH |
708 | return; |
709 | mz = page_cgroup_zoneinfo(pc); | |
710 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
711 | } |
712 | ||
08e552c6 | 713 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 714 | { |
08e552c6 KH |
715 | struct page_cgroup *pc; |
716 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 717 | |
f8d66542 | 718 | if (mem_cgroup_disabled()) |
08e552c6 KH |
719 | return; |
720 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 721 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
bd112db8 DN |
722 | /* |
723 | * Used bit is set without atomic ops but after smp_wmb(). | |
724 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
725 | */ | |
08e552c6 KH |
726 | smp_rmb(); |
727 | if (!PageCgroupUsed(pc)) | |
894bc310 | 728 | return; |
b69408e8 | 729 | |
08e552c6 | 730 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 731 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
4b3bde4c BS |
732 | SetPageCgroupAcctLRU(pc); |
733 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
734 | return; | |
08e552c6 KH |
735 | list_add(&pc->lru, &mz->lists[lru]); |
736 | } | |
544122e5 | 737 | |
08e552c6 | 738 | /* |
544122e5 KH |
739 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
740 | * lru because the page may.be reused after it's fully uncharged (because of | |
741 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
742 | * it again. This function is only used to charge SwapCache. It's done under | |
743 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 744 | */ |
544122e5 | 745 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 746 | { |
544122e5 KH |
747 | unsigned long flags; |
748 | struct zone *zone = page_zone(page); | |
749 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
750 | ||
751 | spin_lock_irqsave(&zone->lru_lock, flags); | |
752 | /* | |
753 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
754 | * is guarded by lock_page() because the page is SwapCache. | |
755 | */ | |
756 | if (!PageCgroupUsed(pc)) | |
757 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
758 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
759 | } |
760 | ||
544122e5 KH |
761 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
762 | { | |
763 | unsigned long flags; | |
764 | struct zone *zone = page_zone(page); | |
765 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
766 | ||
767 | spin_lock_irqsave(&zone->lru_lock, flags); | |
768 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 769 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
770 | mem_cgroup_add_lru_list(page, page_lru(page)); |
771 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
772 | } | |
773 | ||
774 | ||
08e552c6 KH |
775 | void mem_cgroup_move_lists(struct page *page, |
776 | enum lru_list from, enum lru_list to) | |
777 | { | |
f8d66542 | 778 | if (mem_cgroup_disabled()) |
08e552c6 KH |
779 | return; |
780 | mem_cgroup_del_lru_list(page, from); | |
781 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
782 | } |
783 | ||
4c4a2214 DR |
784 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
785 | { | |
786 | int ret; | |
0b7f569e | 787 | struct mem_cgroup *curr = NULL; |
4c4a2214 DR |
788 | |
789 | task_lock(task); | |
0b7f569e KH |
790 | rcu_read_lock(); |
791 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
792 | rcu_read_unlock(); | |
4c4a2214 | 793 | task_unlock(task); |
0b7f569e KH |
794 | if (!curr) |
795 | return 0; | |
d31f56db DN |
796 | /* |
797 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
798 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
799 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
800 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
801 | */ | |
802 | if (mem->use_hierarchy) | |
0b7f569e KH |
803 | ret = css_is_ancestor(&curr->css, &mem->css); |
804 | else | |
805 | ret = (curr == mem); | |
806 | css_put(&curr->css); | |
4c4a2214 DR |
807 | return ret; |
808 | } | |
809 | ||
6c48a1d0 KH |
810 | /* |
811 | * prev_priority control...this will be used in memory reclaim path. | |
812 | */ | |
813 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
814 | { | |
2733c06a KM |
815 | int prev_priority; |
816 | ||
817 | spin_lock(&mem->reclaim_param_lock); | |
818 | prev_priority = mem->prev_priority; | |
819 | spin_unlock(&mem->reclaim_param_lock); | |
820 | ||
821 | return prev_priority; | |
6c48a1d0 KH |
822 | } |
823 | ||
824 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
825 | { | |
2733c06a | 826 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
827 | if (priority < mem->prev_priority) |
828 | mem->prev_priority = priority; | |
2733c06a | 829 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
830 | } |
831 | ||
832 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
833 | { | |
2733c06a | 834 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 835 | mem->prev_priority = priority; |
2733c06a | 836 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
837 | } |
838 | ||
c772be93 | 839 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
840 | { |
841 | unsigned long active; | |
842 | unsigned long inactive; | |
c772be93 KM |
843 | unsigned long gb; |
844 | unsigned long inactive_ratio; | |
14797e23 | 845 | |
14067bb3 KH |
846 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
847 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 848 | |
c772be93 KM |
849 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
850 | if (gb) | |
851 | inactive_ratio = int_sqrt(10 * gb); | |
852 | else | |
853 | inactive_ratio = 1; | |
854 | ||
855 | if (present_pages) { | |
856 | present_pages[0] = inactive; | |
857 | present_pages[1] = active; | |
858 | } | |
859 | ||
860 | return inactive_ratio; | |
861 | } | |
862 | ||
863 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
864 | { | |
865 | unsigned long active; | |
866 | unsigned long inactive; | |
867 | unsigned long present_pages[2]; | |
868 | unsigned long inactive_ratio; | |
869 | ||
870 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
871 | ||
872 | inactive = present_pages[0]; | |
873 | active = present_pages[1]; | |
874 | ||
875 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
876 | return 1; |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
56e49d21 RR |
881 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
882 | { | |
883 | unsigned long active; | |
884 | unsigned long inactive; | |
885 | ||
886 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
887 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
888 | ||
889 | return (active > inactive); | |
890 | } | |
891 | ||
a3d8e054 KM |
892 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
893 | struct zone *zone, | |
894 | enum lru_list lru) | |
895 | { | |
896 | int nid = zone->zone_pgdat->node_id; | |
897 | int zid = zone_idx(zone); | |
898 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
899 | ||
900 | return MEM_CGROUP_ZSTAT(mz, lru); | |
901 | } | |
902 | ||
3e2f41f1 KM |
903 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
904 | struct zone *zone) | |
905 | { | |
906 | int nid = zone->zone_pgdat->node_id; | |
907 | int zid = zone_idx(zone); | |
908 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
909 | ||
910 | return &mz->reclaim_stat; | |
911 | } | |
912 | ||
913 | struct zone_reclaim_stat * | |
914 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
915 | { | |
916 | struct page_cgroup *pc; | |
917 | struct mem_cgroup_per_zone *mz; | |
918 | ||
919 | if (mem_cgroup_disabled()) | |
920 | return NULL; | |
921 | ||
922 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
923 | /* |
924 | * Used bit is set without atomic ops but after smp_wmb(). | |
925 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
926 | */ | |
927 | smp_rmb(); | |
928 | if (!PageCgroupUsed(pc)) | |
929 | return NULL; | |
930 | ||
3e2f41f1 KM |
931 | mz = page_cgroup_zoneinfo(pc); |
932 | if (!mz) | |
933 | return NULL; | |
934 | ||
935 | return &mz->reclaim_stat; | |
936 | } | |
937 | ||
66e1707b BS |
938 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
939 | struct list_head *dst, | |
940 | unsigned long *scanned, int order, | |
941 | int mode, struct zone *z, | |
942 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 943 | int active, int file) |
66e1707b BS |
944 | { |
945 | unsigned long nr_taken = 0; | |
946 | struct page *page; | |
947 | unsigned long scan; | |
948 | LIST_HEAD(pc_list); | |
949 | struct list_head *src; | |
ff7283fa | 950 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
951 | int nid = z->zone_pgdat->node_id; |
952 | int zid = zone_idx(z); | |
953 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 954 | int lru = LRU_FILE * file + active; |
2ffebca6 | 955 | int ret; |
66e1707b | 956 | |
cf475ad2 | 957 | BUG_ON(!mem_cont); |
1ecaab2b | 958 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 959 | src = &mz->lists[lru]; |
66e1707b | 960 | |
ff7283fa KH |
961 | scan = 0; |
962 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 963 | if (scan >= nr_to_scan) |
ff7283fa | 964 | break; |
08e552c6 KH |
965 | |
966 | page = pc->page; | |
52d4b9ac KH |
967 | if (unlikely(!PageCgroupUsed(pc))) |
968 | continue; | |
436c6541 | 969 | if (unlikely(!PageLRU(page))) |
ff7283fa | 970 | continue; |
ff7283fa | 971 | |
436c6541 | 972 | scan++; |
2ffebca6 KH |
973 | ret = __isolate_lru_page(page, mode, file); |
974 | switch (ret) { | |
975 | case 0: | |
66e1707b | 976 | list_move(&page->lru, dst); |
2ffebca6 | 977 | mem_cgroup_del_lru(page); |
66e1707b | 978 | nr_taken++; |
2ffebca6 KH |
979 | break; |
980 | case -EBUSY: | |
981 | /* we don't affect global LRU but rotate in our LRU */ | |
982 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
983 | break; | |
984 | default: | |
985 | break; | |
66e1707b BS |
986 | } |
987 | } | |
988 | ||
66e1707b BS |
989 | *scanned = scan; |
990 | return nr_taken; | |
991 | } | |
992 | ||
6d61ef40 BS |
993 | #define mem_cgroup_from_res_counter(counter, member) \ |
994 | container_of(counter, struct mem_cgroup, member) | |
995 | ||
b85a96c0 DN |
996 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
997 | { | |
998 | if (do_swap_account) { | |
999 | if (res_counter_check_under_limit(&mem->res) && | |
1000 | res_counter_check_under_limit(&mem->memsw)) | |
1001 | return true; | |
1002 | } else | |
1003 | if (res_counter_check_under_limit(&mem->res)) | |
1004 | return true; | |
1005 | return false; | |
1006 | } | |
1007 | ||
a7885eb8 KM |
1008 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1009 | { | |
1010 | struct cgroup *cgrp = memcg->css.cgroup; | |
1011 | unsigned int swappiness; | |
1012 | ||
1013 | /* root ? */ | |
1014 | if (cgrp->parent == NULL) | |
1015 | return vm_swappiness; | |
1016 | ||
1017 | spin_lock(&memcg->reclaim_param_lock); | |
1018 | swappiness = memcg->swappiness; | |
1019 | spin_unlock(&memcg->reclaim_param_lock); | |
1020 | ||
1021 | return swappiness; | |
1022 | } | |
1023 | ||
81d39c20 KH |
1024 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
1025 | { | |
1026 | int *val = data; | |
1027 | (*val)++; | |
1028 | return 0; | |
1029 | } | |
e222432b BS |
1030 | |
1031 | /** | |
1032 | * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode. | |
1033 | * @memcg: The memory cgroup that went over limit | |
1034 | * @p: Task that is going to be killed | |
1035 | * | |
1036 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1037 | * enabled | |
1038 | */ | |
1039 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1040 | { | |
1041 | struct cgroup *task_cgrp; | |
1042 | struct cgroup *mem_cgrp; | |
1043 | /* | |
1044 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1045 | * on the assumption that OOM is serialized for memory controller. | |
1046 | * If this assumption is broken, revisit this code. | |
1047 | */ | |
1048 | static char memcg_name[PATH_MAX]; | |
1049 | int ret; | |
1050 | ||
d31f56db | 1051 | if (!memcg || !p) |
e222432b BS |
1052 | return; |
1053 | ||
1054 | ||
1055 | rcu_read_lock(); | |
1056 | ||
1057 | mem_cgrp = memcg->css.cgroup; | |
1058 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1059 | ||
1060 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1061 | if (ret < 0) { | |
1062 | /* | |
1063 | * Unfortunately, we are unable to convert to a useful name | |
1064 | * But we'll still print out the usage information | |
1065 | */ | |
1066 | rcu_read_unlock(); | |
1067 | goto done; | |
1068 | } | |
1069 | rcu_read_unlock(); | |
1070 | ||
1071 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1072 | ||
1073 | rcu_read_lock(); | |
1074 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1075 | if (ret < 0) { | |
1076 | rcu_read_unlock(); | |
1077 | goto done; | |
1078 | } | |
1079 | rcu_read_unlock(); | |
1080 | ||
1081 | /* | |
1082 | * Continues from above, so we don't need an KERN_ level | |
1083 | */ | |
1084 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1085 | done: | |
1086 | ||
1087 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1088 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1089 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1090 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1091 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1092 | "failcnt %llu\n", | |
1093 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1094 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1095 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1096 | } | |
1097 | ||
81d39c20 KH |
1098 | /* |
1099 | * This function returns the number of memcg under hierarchy tree. Returns | |
1100 | * 1(self count) if no children. | |
1101 | */ | |
1102 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1103 | { | |
1104 | int num = 0; | |
1105 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
1106 | return num; | |
1107 | } | |
1108 | ||
6d61ef40 | 1109 | /* |
04046e1a KH |
1110 | * Visit the first child (need not be the first child as per the ordering |
1111 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1112 | * that to reclaim free pages from. | |
1113 | */ | |
1114 | static struct mem_cgroup * | |
1115 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1116 | { | |
1117 | struct mem_cgroup *ret = NULL; | |
1118 | struct cgroup_subsys_state *css; | |
1119 | int nextid, found; | |
1120 | ||
1121 | if (!root_mem->use_hierarchy) { | |
1122 | css_get(&root_mem->css); | |
1123 | ret = root_mem; | |
1124 | } | |
1125 | ||
1126 | while (!ret) { | |
1127 | rcu_read_lock(); | |
1128 | nextid = root_mem->last_scanned_child + 1; | |
1129 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1130 | &found); | |
1131 | if (css && css_tryget(css)) | |
1132 | ret = container_of(css, struct mem_cgroup, css); | |
1133 | ||
1134 | rcu_read_unlock(); | |
1135 | /* Updates scanning parameter */ | |
1136 | spin_lock(&root_mem->reclaim_param_lock); | |
1137 | if (!css) { | |
1138 | /* this means start scan from ID:1 */ | |
1139 | root_mem->last_scanned_child = 0; | |
1140 | } else | |
1141 | root_mem->last_scanned_child = found; | |
1142 | spin_unlock(&root_mem->reclaim_param_lock); | |
1143 | } | |
1144 | ||
1145 | return ret; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1150 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1151 | * based on its position in the children list. | |
6d61ef40 BS |
1152 | * |
1153 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1154 | * |
1155 | * We give up and return to the caller when we visit root_mem twice. | |
1156 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1157 | * |
1158 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1159 | */ |
1160 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1161 | struct zone *zone, |
75822b44 BS |
1162 | gfp_t gfp_mask, |
1163 | unsigned long reclaim_options) | |
6d61ef40 | 1164 | { |
04046e1a KH |
1165 | struct mem_cgroup *victim; |
1166 | int ret, total = 0; | |
1167 | int loop = 0; | |
75822b44 BS |
1168 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1169 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 BS |
1170 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
1171 | unsigned long excess = mem_cgroup_get_excess(root_mem); | |
04046e1a | 1172 | |
22a668d7 KH |
1173 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1174 | if (root_mem->memsw_is_minimum) | |
1175 | noswap = true; | |
1176 | ||
4e416953 | 1177 | while (1) { |
04046e1a | 1178 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1179 | if (victim == root_mem) { |
04046e1a | 1180 | loop++; |
cdec2e42 KH |
1181 | if (loop >= 1) |
1182 | drain_all_stock_async(); | |
4e416953 BS |
1183 | if (loop >= 2) { |
1184 | /* | |
1185 | * If we have not been able to reclaim | |
1186 | * anything, it might because there are | |
1187 | * no reclaimable pages under this hierarchy | |
1188 | */ | |
1189 | if (!check_soft || !total) { | |
1190 | css_put(&victim->css); | |
1191 | break; | |
1192 | } | |
1193 | /* | |
1194 | * We want to do more targetted reclaim. | |
1195 | * excess >> 2 is not to excessive so as to | |
1196 | * reclaim too much, nor too less that we keep | |
1197 | * coming back to reclaim from this cgroup | |
1198 | */ | |
1199 | if (total >= (excess >> 2) || | |
1200 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1201 | css_put(&victim->css); | |
1202 | break; | |
1203 | } | |
1204 | } | |
1205 | } | |
04046e1a KH |
1206 | if (!mem_cgroup_local_usage(&victim->stat)) { |
1207 | /* this cgroup's local usage == 0 */ | |
1208 | css_put(&victim->css); | |
6d61ef40 BS |
1209 | continue; |
1210 | } | |
04046e1a | 1211 | /* we use swappiness of local cgroup */ |
4e416953 BS |
1212 | if (check_soft) |
1213 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
1214 | noswap, get_swappiness(victim), zone, | |
1215 | zone->zone_pgdat->node_id); | |
1216 | else | |
1217 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1218 | noswap, get_swappiness(victim)); | |
04046e1a | 1219 | css_put(&victim->css); |
81d39c20 KH |
1220 | /* |
1221 | * At shrinking usage, we can't check we should stop here or | |
1222 | * reclaim more. It's depends on callers. last_scanned_child | |
1223 | * will work enough for keeping fairness under tree. | |
1224 | */ | |
1225 | if (shrink) | |
1226 | return ret; | |
04046e1a | 1227 | total += ret; |
4e416953 BS |
1228 | if (check_soft) { |
1229 | if (res_counter_check_under_soft_limit(&root_mem->res)) | |
1230 | return total; | |
1231 | } else if (mem_cgroup_check_under_limit(root_mem)) | |
04046e1a | 1232 | return 1 + total; |
6d61ef40 | 1233 | } |
04046e1a | 1234 | return total; |
6d61ef40 BS |
1235 | } |
1236 | ||
a636b327 KH |
1237 | bool mem_cgroup_oom_called(struct task_struct *task) |
1238 | { | |
1239 | bool ret = false; | |
1240 | struct mem_cgroup *mem; | |
1241 | struct mm_struct *mm; | |
1242 | ||
1243 | rcu_read_lock(); | |
1244 | mm = task->mm; | |
1245 | if (!mm) | |
1246 | mm = &init_mm; | |
1247 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1248 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
1249 | ret = true; | |
1250 | rcu_read_unlock(); | |
1251 | return ret; | |
1252 | } | |
0b7f569e KH |
1253 | |
1254 | static int record_last_oom_cb(struct mem_cgroup *mem, void *data) | |
1255 | { | |
1256 | mem->last_oom_jiffies = jiffies; | |
1257 | return 0; | |
1258 | } | |
1259 | ||
1260 | static void record_last_oom(struct mem_cgroup *mem) | |
1261 | { | |
1262 | mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb); | |
1263 | } | |
1264 | ||
d69b042f BS |
1265 | /* |
1266 | * Currently used to update mapped file statistics, but the routine can be | |
1267 | * generalized to update other statistics as well. | |
1268 | */ | |
d8046582 | 1269 | void mem_cgroup_update_file_mapped(struct page *page, int val) |
d69b042f BS |
1270 | { |
1271 | struct mem_cgroup *mem; | |
1272 | struct mem_cgroup_stat *stat; | |
1273 | struct mem_cgroup_stat_cpu *cpustat; | |
1274 | int cpu; | |
1275 | struct page_cgroup *pc; | |
1276 | ||
d69b042f BS |
1277 | pc = lookup_page_cgroup(page); |
1278 | if (unlikely(!pc)) | |
1279 | return; | |
1280 | ||
1281 | lock_page_cgroup(pc); | |
1282 | mem = pc->mem_cgroup; | |
1283 | if (!mem) | |
1284 | goto done; | |
1285 | ||
1286 | if (!PageCgroupUsed(pc)) | |
1287 | goto done; | |
1288 | ||
1289 | /* | |
1290 | * Preemption is already disabled, we don't need get_cpu() | |
1291 | */ | |
1292 | cpu = smp_processor_id(); | |
1293 | stat = &mem->stat; | |
1294 | cpustat = &stat->cpustat[cpu]; | |
1295 | ||
d8046582 | 1296 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, val); |
d69b042f BS |
1297 | done: |
1298 | unlock_page_cgroup(pc); | |
1299 | } | |
0b7f569e | 1300 | |
cdec2e42 KH |
1301 | /* |
1302 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1303 | * TODO: maybe necessary to use big numbers in big irons. | |
1304 | */ | |
1305 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1306 | struct memcg_stock_pcp { | |
1307 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1308 | int charge; | |
1309 | struct work_struct work; | |
1310 | }; | |
1311 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1312 | static atomic_t memcg_drain_count; | |
1313 | ||
1314 | /* | |
1315 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1316 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1317 | * cgroup which is not current target, returns false. This stock will be | |
1318 | * refilled. | |
1319 | */ | |
1320 | static bool consume_stock(struct mem_cgroup *mem) | |
1321 | { | |
1322 | struct memcg_stock_pcp *stock; | |
1323 | bool ret = true; | |
1324 | ||
1325 | stock = &get_cpu_var(memcg_stock); | |
1326 | if (mem == stock->cached && stock->charge) | |
1327 | stock->charge -= PAGE_SIZE; | |
1328 | else /* need to call res_counter_charge */ | |
1329 | ret = false; | |
1330 | put_cpu_var(memcg_stock); | |
1331 | return ret; | |
1332 | } | |
1333 | ||
1334 | /* | |
1335 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1336 | */ | |
1337 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1338 | { | |
1339 | struct mem_cgroup *old = stock->cached; | |
1340 | ||
1341 | if (stock->charge) { | |
1342 | res_counter_uncharge(&old->res, stock->charge); | |
1343 | if (do_swap_account) | |
1344 | res_counter_uncharge(&old->memsw, stock->charge); | |
1345 | } | |
1346 | stock->cached = NULL; | |
1347 | stock->charge = 0; | |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * This must be called under preempt disabled or must be called by | |
1352 | * a thread which is pinned to local cpu. | |
1353 | */ | |
1354 | static void drain_local_stock(struct work_struct *dummy) | |
1355 | { | |
1356 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1357 | drain_stock(stock); | |
1358 | } | |
1359 | ||
1360 | /* | |
1361 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
1362 | * This will be consumed by consumt_stock() function, later. | |
1363 | */ | |
1364 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1365 | { | |
1366 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1367 | ||
1368 | if (stock->cached != mem) { /* reset if necessary */ | |
1369 | drain_stock(stock); | |
1370 | stock->cached = mem; | |
1371 | } | |
1372 | stock->charge += val; | |
1373 | put_cpu_var(memcg_stock); | |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1378 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1379 | * expects some charges will be back to res_counter later but cannot wait for | |
1380 | * it. | |
1381 | */ | |
1382 | static void drain_all_stock_async(void) | |
1383 | { | |
1384 | int cpu; | |
1385 | /* This function is for scheduling "drain" in asynchronous way. | |
1386 | * The result of "drain" is not directly handled by callers. Then, | |
1387 | * if someone is calling drain, we don't have to call drain more. | |
1388 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1389 | * there is a race. We just do loose check here. | |
1390 | */ | |
1391 | if (atomic_read(&memcg_drain_count)) | |
1392 | return; | |
1393 | /* Notify other cpus that system-wide "drain" is running */ | |
1394 | atomic_inc(&memcg_drain_count); | |
1395 | get_online_cpus(); | |
1396 | for_each_online_cpu(cpu) { | |
1397 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1398 | schedule_work_on(cpu, &stock->work); | |
1399 | } | |
1400 | put_online_cpus(); | |
1401 | atomic_dec(&memcg_drain_count); | |
1402 | /* We don't wait for flush_work */ | |
1403 | } | |
1404 | ||
1405 | /* This is a synchronous drain interface. */ | |
1406 | static void drain_all_stock_sync(void) | |
1407 | { | |
1408 | /* called when force_empty is called */ | |
1409 | atomic_inc(&memcg_drain_count); | |
1410 | schedule_on_each_cpu(drain_local_stock); | |
1411 | atomic_dec(&memcg_drain_count); | |
1412 | } | |
1413 | ||
1414 | static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb, | |
1415 | unsigned long action, | |
1416 | void *hcpu) | |
1417 | { | |
1418 | int cpu = (unsigned long)hcpu; | |
1419 | struct memcg_stock_pcp *stock; | |
1420 | ||
1421 | if (action != CPU_DEAD) | |
1422 | return NOTIFY_OK; | |
1423 | stock = &per_cpu(memcg_stock, cpu); | |
1424 | drain_stock(stock); | |
1425 | return NOTIFY_OK; | |
1426 | } | |
1427 | ||
f817ed48 KH |
1428 | /* |
1429 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1430 | * oom-killer can be invoked. | |
8a9f3ccd | 1431 | */ |
f817ed48 | 1432 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
8c7c6e34 | 1433 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
f64c3f54 | 1434 | bool oom, struct page *page) |
8a9f3ccd | 1435 | { |
4e649152 | 1436 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 1437 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
4e649152 | 1438 | struct res_counter *fail_res; |
cdec2e42 | 1439 | int csize = CHARGE_SIZE; |
a636b327 KH |
1440 | |
1441 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
1442 | /* Don't account this! */ | |
1443 | *memcg = NULL; | |
1444 | return 0; | |
1445 | } | |
1446 | ||
8a9f3ccd | 1447 | /* |
3be91277 HD |
1448 | * We always charge the cgroup the mm_struct belongs to. |
1449 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
1450 | * thread group leader migrates. It's possible that mm is not |
1451 | * set, if so charge the init_mm (happens for pagecache usage). | |
1452 | */ | |
54595fe2 KH |
1453 | mem = *memcg; |
1454 | if (likely(!mem)) { | |
1455 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 1456 | *memcg = mem; |
e8589cc1 | 1457 | } else { |
7a81b88c | 1458 | css_get(&mem->css); |
e8589cc1 | 1459 | } |
54595fe2 KH |
1460 | if (unlikely(!mem)) |
1461 | return 0; | |
1462 | ||
46f7e602 | 1463 | VM_BUG_ON(css_is_removed(&mem->css)); |
cdec2e42 KH |
1464 | if (mem_cgroup_is_root(mem)) |
1465 | goto done; | |
8a9f3ccd | 1466 | |
8c7c6e34 | 1467 | while (1) { |
0c3e73e8 | 1468 | int ret = 0; |
75822b44 | 1469 | unsigned long flags = 0; |
7a81b88c | 1470 | |
cdec2e42 KH |
1471 | if (consume_stock(mem)) |
1472 | goto charged; | |
1473 | ||
1474 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
8c7c6e34 KH |
1475 | if (likely(!ret)) { |
1476 | if (!do_swap_account) | |
1477 | break; | |
cdec2e42 | 1478 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); |
8c7c6e34 KH |
1479 | if (likely(!ret)) |
1480 | break; | |
1481 | /* mem+swap counter fails */ | |
cdec2e42 | 1482 | res_counter_uncharge(&mem->res, csize); |
75822b44 | 1483 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
6d61ef40 BS |
1484 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1485 | memsw); | |
1486 | } else | |
1487 | /* mem counter fails */ | |
1488 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1489 | res); | |
1490 | ||
cdec2e42 KH |
1491 | /* reduce request size and retry */ |
1492 | if (csize > PAGE_SIZE) { | |
1493 | csize = PAGE_SIZE; | |
1494 | continue; | |
1495 | } | |
3be91277 | 1496 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 1497 | goto nomem; |
e1a1cd59 | 1498 | |
4e416953 BS |
1499 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
1500 | gfp_mask, flags); | |
4d1c6273 DN |
1501 | if (ret) |
1502 | continue; | |
66e1707b BS |
1503 | |
1504 | /* | |
8869b8f6 HD |
1505 | * try_to_free_mem_cgroup_pages() might not give us a full |
1506 | * picture of reclaim. Some pages are reclaimed and might be | |
1507 | * moved to swap cache or just unmapped from the cgroup. | |
1508 | * Check the limit again to see if the reclaim reduced the | |
1509 | * current usage of the cgroup before giving up | |
8c7c6e34 | 1510 | * |
8869b8f6 | 1511 | */ |
b85a96c0 DN |
1512 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
1513 | continue; | |
3be91277 | 1514 | |
8033b97c DN |
1515 | /* try to avoid oom while someone is moving charge */ |
1516 | if (mc.moving_task && current != mc.moving_task) { | |
1517 | struct mem_cgroup *from, *to; | |
1518 | bool do_continue = false; | |
1519 | /* | |
1520 | * There is a small race that "from" or "to" can be | |
1521 | * freed by rmdir, so we use css_tryget(). | |
1522 | */ | |
1523 | rcu_read_lock(); | |
1524 | from = mc.from; | |
1525 | to = mc.to; | |
1526 | if (from && css_tryget(&from->css)) { | |
1527 | if (mem_over_limit->use_hierarchy) | |
1528 | do_continue = css_is_ancestor( | |
1529 | &from->css, | |
1530 | &mem_over_limit->css); | |
1531 | else | |
1532 | do_continue = (from == mem_over_limit); | |
1533 | css_put(&from->css); | |
1534 | } | |
1535 | if (!do_continue && to && css_tryget(&to->css)) { | |
1536 | if (mem_over_limit->use_hierarchy) | |
1537 | do_continue = css_is_ancestor( | |
1538 | &to->css, | |
1539 | &mem_over_limit->css); | |
1540 | else | |
1541 | do_continue = (to == mem_over_limit); | |
1542 | css_put(&to->css); | |
1543 | } | |
1544 | rcu_read_unlock(); | |
1545 | if (do_continue) { | |
1546 | DEFINE_WAIT(wait); | |
1547 | prepare_to_wait(&mc.waitq, &wait, | |
1548 | TASK_INTERRUPTIBLE); | |
1549 | /* moving charge context might have finished. */ | |
1550 | if (mc.moving_task) | |
1551 | schedule(); | |
1552 | finish_wait(&mc.waitq, &wait); | |
1553 | continue; | |
1554 | } | |
1555 | } | |
1556 | ||
3be91277 | 1557 | if (!nr_retries--) { |
a636b327 | 1558 | if (oom) { |
88700756 | 1559 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
0b7f569e | 1560 | record_last_oom(mem_over_limit); |
a636b327 | 1561 | } |
7a81b88c | 1562 | goto nomem; |
66e1707b | 1563 | } |
8a9f3ccd | 1564 | } |
cdec2e42 KH |
1565 | if (csize > PAGE_SIZE) |
1566 | refill_stock(mem, csize - PAGE_SIZE); | |
1567 | charged: | |
f64c3f54 | 1568 | /* |
4e649152 KH |
1569 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. |
1570 | * if they exceeds softlimit. | |
f64c3f54 | 1571 | */ |
4ffef5fe | 1572 | if (page && mem_cgroup_soft_limit_check(mem)) |
4e649152 | 1573 | mem_cgroup_update_tree(mem, page); |
0c3e73e8 | 1574 | done: |
7a81b88c KH |
1575 | return 0; |
1576 | nomem: | |
1577 | css_put(&mem->css); | |
1578 | return -ENOMEM; | |
1579 | } | |
8a9f3ccd | 1580 | |
a3032a2c DN |
1581 | /* |
1582 | * Somemtimes we have to undo a charge we got by try_charge(). | |
1583 | * This function is for that and do uncharge, put css's refcnt. | |
1584 | * gotten by try_charge(). | |
1585 | */ | |
854ffa8d DN |
1586 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
1587 | unsigned long count) | |
a3032a2c DN |
1588 | { |
1589 | if (!mem_cgroup_is_root(mem)) { | |
854ffa8d | 1590 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
a3032a2c | 1591 | if (do_swap_account) |
854ffa8d DN |
1592 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
1593 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
1594 | WARN_ON_ONCE(count > INT_MAX); | |
1595 | __css_put(&mem->css, (int)count); | |
a3032a2c | 1596 | } |
854ffa8d DN |
1597 | /* we don't need css_put for root */ |
1598 | } | |
1599 | ||
1600 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem) | |
1601 | { | |
1602 | __mem_cgroup_cancel_charge(mem, 1); | |
a3032a2c DN |
1603 | } |
1604 | ||
a3b2d692 KH |
1605 | /* |
1606 | * A helper function to get mem_cgroup from ID. must be called under | |
1607 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
1608 | * it's concern. (dropping refcnt from swap can be called against removed | |
1609 | * memcg.) | |
1610 | */ | |
1611 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
1612 | { | |
1613 | struct cgroup_subsys_state *css; | |
1614 | ||
1615 | /* ID 0 is unused ID */ | |
1616 | if (!id) | |
1617 | return NULL; | |
1618 | css = css_lookup(&mem_cgroup_subsys, id); | |
1619 | if (!css) | |
1620 | return NULL; | |
1621 | return container_of(css, struct mem_cgroup, css); | |
1622 | } | |
1623 | ||
e42d9d5d | 1624 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 1625 | { |
e42d9d5d | 1626 | struct mem_cgroup *mem = NULL; |
3c776e64 | 1627 | struct page_cgroup *pc; |
a3b2d692 | 1628 | unsigned short id; |
b5a84319 KH |
1629 | swp_entry_t ent; |
1630 | ||
3c776e64 DN |
1631 | VM_BUG_ON(!PageLocked(page)); |
1632 | ||
3c776e64 | 1633 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 1634 | lock_page_cgroup(pc); |
a3b2d692 | 1635 | if (PageCgroupUsed(pc)) { |
3c776e64 | 1636 | mem = pc->mem_cgroup; |
a3b2d692 KH |
1637 | if (mem && !css_tryget(&mem->css)) |
1638 | mem = NULL; | |
e42d9d5d | 1639 | } else if (PageSwapCache(page)) { |
3c776e64 | 1640 | ent.val = page_private(page); |
a3b2d692 KH |
1641 | id = lookup_swap_cgroup(ent); |
1642 | rcu_read_lock(); | |
1643 | mem = mem_cgroup_lookup(id); | |
1644 | if (mem && !css_tryget(&mem->css)) | |
1645 | mem = NULL; | |
1646 | rcu_read_unlock(); | |
3c776e64 | 1647 | } |
c0bd3f63 | 1648 | unlock_page_cgroup(pc); |
b5a84319 KH |
1649 | return mem; |
1650 | } | |
1651 | ||
7a81b88c | 1652 | /* |
a5e924f5 | 1653 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
1654 | * USED state. If already USED, uncharge and return. |
1655 | */ | |
1656 | ||
1657 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
1658 | struct page_cgroup *pc, | |
1659 | enum charge_type ctype) | |
1660 | { | |
7a81b88c KH |
1661 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
1662 | if (!mem) | |
1663 | return; | |
52d4b9ac KH |
1664 | |
1665 | lock_page_cgroup(pc); | |
1666 | if (unlikely(PageCgroupUsed(pc))) { | |
1667 | unlock_page_cgroup(pc); | |
a3032a2c | 1668 | mem_cgroup_cancel_charge(mem); |
7a81b88c | 1669 | return; |
52d4b9ac | 1670 | } |
4b3bde4c | 1671 | |
8a9f3ccd | 1672 | pc->mem_cgroup = mem; |
261fb61a KH |
1673 | /* |
1674 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
1675 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
1676 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
1677 | * before USED bit, we need memory barrier here. | |
1678 | * See mem_cgroup_add_lru_list(), etc. | |
1679 | */ | |
08e552c6 | 1680 | smp_wmb(); |
4b3bde4c BS |
1681 | switch (ctype) { |
1682 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
1683 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
1684 | SetPageCgroupCache(pc); | |
1685 | SetPageCgroupUsed(pc); | |
1686 | break; | |
1687 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1688 | ClearPageCgroupCache(pc); | |
1689 | SetPageCgroupUsed(pc); | |
1690 | break; | |
1691 | default: | |
1692 | break; | |
1693 | } | |
3be91277 | 1694 | |
08e552c6 | 1695 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 1696 | |
52d4b9ac | 1697 | unlock_page_cgroup(pc); |
7a81b88c | 1698 | } |
66e1707b | 1699 | |
f817ed48 | 1700 | /** |
57f9fd7d | 1701 | * __mem_cgroup_move_account - move account of the page |
f817ed48 KH |
1702 | * @pc: page_cgroup of the page. |
1703 | * @from: mem_cgroup which the page is moved from. | |
1704 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 1705 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
1706 | * |
1707 | * The caller must confirm following. | |
08e552c6 | 1708 | * - page is not on LRU (isolate_page() is useful.) |
57f9fd7d | 1709 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
f817ed48 | 1710 | * |
854ffa8d DN |
1711 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
1712 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is | |
1713 | * true, this function does "uncharge" from old cgroup, but it doesn't if | |
1714 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 KH |
1715 | */ |
1716 | ||
57f9fd7d | 1717 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
854ffa8d | 1718 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
f817ed48 | 1719 | { |
d69b042f BS |
1720 | struct page *page; |
1721 | int cpu; | |
1722 | struct mem_cgroup_stat *stat; | |
1723 | struct mem_cgroup_stat_cpu *cpustat; | |
f817ed48 | 1724 | |
f817ed48 | 1725 | VM_BUG_ON(from == to); |
08e552c6 | 1726 | VM_BUG_ON(PageLRU(pc->page)); |
57f9fd7d DN |
1727 | VM_BUG_ON(!PageCgroupLocked(pc)); |
1728 | VM_BUG_ON(!PageCgroupUsed(pc)); | |
1729 | VM_BUG_ON(pc->mem_cgroup != from); | |
f817ed48 | 1730 | |
d69b042f | 1731 | page = pc->page; |
d8046582 | 1732 | if (page_mapped(page) && !PageAnon(page)) { |
d69b042f BS |
1733 | cpu = smp_processor_id(); |
1734 | /* Update mapped_file data for mem_cgroup "from" */ | |
1735 | stat = &from->stat; | |
1736 | cpustat = &stat->cpustat[cpu]; | |
d8046582 | 1737 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, |
d69b042f BS |
1738 | -1); |
1739 | ||
1740 | /* Update mapped_file data for mem_cgroup "to" */ | |
1741 | stat = &to->stat; | |
1742 | cpustat = &stat->cpustat[cpu]; | |
d8046582 | 1743 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, |
d69b042f BS |
1744 | 1); |
1745 | } | |
854ffa8d DN |
1746 | mem_cgroup_charge_statistics(from, pc, false); |
1747 | if (uncharge) | |
1748 | /* This is not "cancel", but cancel_charge does all we need. */ | |
1749 | mem_cgroup_cancel_charge(from); | |
d69b042f | 1750 | |
854ffa8d | 1751 | /* caller should have done css_get */ |
08e552c6 KH |
1752 | pc->mem_cgroup = to; |
1753 | mem_cgroup_charge_statistics(to, pc, true); | |
88703267 KH |
1754 | /* |
1755 | * We charges against "to" which may not have any tasks. Then, "to" | |
1756 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe DN |
1757 | * this function is just force_empty() and move charge, so it's |
1758 | * garanteed that "to" is never removed. So, we don't check rmdir | |
1759 | * status here. | |
88703267 | 1760 | */ |
57f9fd7d DN |
1761 | } |
1762 | ||
1763 | /* | |
1764 | * check whether the @pc is valid for moving account and call | |
1765 | * __mem_cgroup_move_account() | |
1766 | */ | |
1767 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
854ffa8d | 1768 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
57f9fd7d DN |
1769 | { |
1770 | int ret = -EINVAL; | |
1771 | lock_page_cgroup(pc); | |
1772 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { | |
854ffa8d | 1773 | __mem_cgroup_move_account(pc, from, to, uncharge); |
57f9fd7d DN |
1774 | ret = 0; |
1775 | } | |
1776 | unlock_page_cgroup(pc); | |
f817ed48 KH |
1777 | return ret; |
1778 | } | |
1779 | ||
1780 | /* | |
1781 | * move charges to its parent. | |
1782 | */ | |
1783 | ||
1784 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1785 | struct mem_cgroup *child, | |
1786 | gfp_t gfp_mask) | |
1787 | { | |
08e552c6 | 1788 | struct page *page = pc->page; |
f817ed48 KH |
1789 | struct cgroup *cg = child->css.cgroup; |
1790 | struct cgroup *pcg = cg->parent; | |
1791 | struct mem_cgroup *parent; | |
f817ed48 KH |
1792 | int ret; |
1793 | ||
1794 | /* Is ROOT ? */ | |
1795 | if (!pcg) | |
1796 | return -EINVAL; | |
1797 | ||
57f9fd7d DN |
1798 | ret = -EBUSY; |
1799 | if (!get_page_unless_zero(page)) | |
1800 | goto out; | |
1801 | if (isolate_lru_page(page)) | |
1802 | goto put; | |
08e552c6 | 1803 | |
f817ed48 | 1804 | parent = mem_cgroup_from_cont(pcg); |
f64c3f54 | 1805 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page); |
a636b327 | 1806 | if (ret || !parent) |
57f9fd7d | 1807 | goto put_back; |
f817ed48 | 1808 | |
854ffa8d DN |
1809 | ret = mem_cgroup_move_account(pc, child, parent, true); |
1810 | if (ret) | |
1811 | mem_cgroup_cancel_charge(parent); | |
57f9fd7d | 1812 | put_back: |
08e552c6 | 1813 | putback_lru_page(page); |
57f9fd7d | 1814 | put: |
40d58138 | 1815 | put_page(page); |
57f9fd7d | 1816 | out: |
f817ed48 KH |
1817 | return ret; |
1818 | } | |
1819 | ||
7a81b88c KH |
1820 | /* |
1821 | * Charge the memory controller for page usage. | |
1822 | * Return | |
1823 | * 0 if the charge was successful | |
1824 | * < 0 if the cgroup is over its limit | |
1825 | */ | |
1826 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1827 | gfp_t gfp_mask, enum charge_type ctype, | |
1828 | struct mem_cgroup *memcg) | |
1829 | { | |
1830 | struct mem_cgroup *mem; | |
1831 | struct page_cgroup *pc; | |
1832 | int ret; | |
1833 | ||
1834 | pc = lookup_page_cgroup(page); | |
1835 | /* can happen at boot */ | |
1836 | if (unlikely(!pc)) | |
1837 | return 0; | |
1838 | prefetchw(pc); | |
1839 | ||
1840 | mem = memcg; | |
f64c3f54 | 1841 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page); |
a636b327 | 1842 | if (ret || !mem) |
7a81b88c KH |
1843 | return ret; |
1844 | ||
1845 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1846 | return 0; |
8a9f3ccd BS |
1847 | } |
1848 | ||
7a81b88c KH |
1849 | int mem_cgroup_newpage_charge(struct page *page, |
1850 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1851 | { |
f8d66542 | 1852 | if (mem_cgroup_disabled()) |
cede86ac | 1853 | return 0; |
52d4b9ac KH |
1854 | if (PageCompound(page)) |
1855 | return 0; | |
69029cd5 KH |
1856 | /* |
1857 | * If already mapped, we don't have to account. | |
1858 | * If page cache, page->mapping has address_space. | |
1859 | * But page->mapping may have out-of-use anon_vma pointer, | |
1860 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1861 | * is NULL. | |
1862 | */ | |
1863 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1864 | return 0; | |
1865 | if (unlikely(!mm)) | |
1866 | mm = &init_mm; | |
217bc319 | 1867 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 1868 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
1869 | } |
1870 | ||
83aae4c7 DN |
1871 | static void |
1872 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1873 | enum charge_type ctype); | |
1874 | ||
e1a1cd59 BS |
1875 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
1876 | gfp_t gfp_mask) | |
8697d331 | 1877 | { |
b5a84319 KH |
1878 | struct mem_cgroup *mem = NULL; |
1879 | int ret; | |
1880 | ||
f8d66542 | 1881 | if (mem_cgroup_disabled()) |
cede86ac | 1882 | return 0; |
52d4b9ac KH |
1883 | if (PageCompound(page)) |
1884 | return 0; | |
accf163e KH |
1885 | /* |
1886 | * Corner case handling. This is called from add_to_page_cache() | |
1887 | * in usual. But some FS (shmem) precharges this page before calling it | |
1888 | * and call add_to_page_cache() with GFP_NOWAIT. | |
1889 | * | |
1890 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
1891 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
1892 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
1893 | * And when the page is SwapCache, it should take swap information |
1894 | * into account. This is under lock_page() now. | |
accf163e KH |
1895 | */ |
1896 | if (!(gfp_mask & __GFP_WAIT)) { | |
1897 | struct page_cgroup *pc; | |
1898 | ||
52d4b9ac KH |
1899 | |
1900 | pc = lookup_page_cgroup(page); | |
1901 | if (!pc) | |
1902 | return 0; | |
1903 | lock_page_cgroup(pc); | |
1904 | if (PageCgroupUsed(pc)) { | |
1905 | unlock_page_cgroup(pc); | |
accf163e KH |
1906 | return 0; |
1907 | } | |
52d4b9ac | 1908 | unlock_page_cgroup(pc); |
accf163e KH |
1909 | } |
1910 | ||
b5a84319 | 1911 | if (unlikely(!mm && !mem)) |
8697d331 | 1912 | mm = &init_mm; |
accf163e | 1913 | |
c05555b5 KH |
1914 | if (page_is_file_cache(page)) |
1915 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 1916 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 | 1917 | |
83aae4c7 DN |
1918 | /* shmem */ |
1919 | if (PageSwapCache(page)) { | |
1920 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
1921 | if (!ret) | |
1922 | __mem_cgroup_commit_charge_swapin(page, mem, | |
1923 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
1924 | } else | |
1925 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
1926 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
b5a84319 | 1927 | |
b5a84319 | 1928 | return ret; |
e8589cc1 KH |
1929 | } |
1930 | ||
54595fe2 KH |
1931 | /* |
1932 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
1933 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 1934 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
1935 | * "commit()" or removed by "cancel()" |
1936 | */ | |
8c7c6e34 KH |
1937 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1938 | struct page *page, | |
1939 | gfp_t mask, struct mem_cgroup **ptr) | |
1940 | { | |
1941 | struct mem_cgroup *mem; | |
54595fe2 | 1942 | int ret; |
8c7c6e34 | 1943 | |
f8d66542 | 1944 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
1945 | return 0; |
1946 | ||
1947 | if (!do_swap_account) | |
1948 | goto charge_cur_mm; | |
8c7c6e34 KH |
1949 | /* |
1950 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
1951 | * the pte, and even removed page from swap cache: in those cases |
1952 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
1953 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
1954 | */ |
1955 | if (!PageSwapCache(page)) | |
407f9c8b | 1956 | goto charge_cur_mm; |
e42d9d5d | 1957 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
1958 | if (!mem) |
1959 | goto charge_cur_mm; | |
8c7c6e34 | 1960 | *ptr = mem; |
f64c3f54 | 1961 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page); |
54595fe2 KH |
1962 | /* drop extra refcnt from tryget */ |
1963 | css_put(&mem->css); | |
1964 | return ret; | |
8c7c6e34 KH |
1965 | charge_cur_mm: |
1966 | if (unlikely(!mm)) | |
1967 | mm = &init_mm; | |
f64c3f54 | 1968 | return __mem_cgroup_try_charge(mm, mask, ptr, true, page); |
8c7c6e34 KH |
1969 | } |
1970 | ||
83aae4c7 DN |
1971 | static void |
1972 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1973 | enum charge_type ctype) | |
7a81b88c KH |
1974 | { |
1975 | struct page_cgroup *pc; | |
1976 | ||
f8d66542 | 1977 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1978 | return; |
1979 | if (!ptr) | |
1980 | return; | |
88703267 | 1981 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 1982 | pc = lookup_page_cgroup(page); |
544122e5 | 1983 | mem_cgroup_lru_del_before_commit_swapcache(page); |
83aae4c7 | 1984 | __mem_cgroup_commit_charge(ptr, pc, ctype); |
544122e5 | 1985 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
1986 | /* |
1987 | * Now swap is on-memory. This means this page may be | |
1988 | * counted both as mem and swap....double count. | |
03f3c433 KH |
1989 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
1990 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
1991 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 1992 | */ |
03f3c433 | 1993 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 1994 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 1995 | unsigned short id; |
8c7c6e34 | 1996 | struct mem_cgroup *memcg; |
a3b2d692 KH |
1997 | |
1998 | id = swap_cgroup_record(ent, 0); | |
1999 | rcu_read_lock(); | |
2000 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2001 | if (memcg) { |
a3b2d692 KH |
2002 | /* |
2003 | * This recorded memcg can be obsolete one. So, avoid | |
2004 | * calling css_tryget | |
2005 | */ | |
0c3e73e8 | 2006 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2007 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2008 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2009 | mem_cgroup_put(memcg); |
2010 | } | |
a3b2d692 | 2011 | rcu_read_unlock(); |
8c7c6e34 | 2012 | } |
88703267 KH |
2013 | /* |
2014 | * At swapin, we may charge account against cgroup which has no tasks. | |
2015 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2016 | * In that case, we need to call pre_destroy() again. check it here. | |
2017 | */ | |
2018 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2019 | } |
2020 | ||
83aae4c7 DN |
2021 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2022 | { | |
2023 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2024 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2025 | } | |
2026 | ||
7a81b88c KH |
2027 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2028 | { | |
f8d66542 | 2029 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2030 | return; |
2031 | if (!mem) | |
2032 | return; | |
a3032a2c | 2033 | mem_cgroup_cancel_charge(mem); |
7a81b88c KH |
2034 | } |
2035 | ||
569b846d KH |
2036 | static void |
2037 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype) | |
2038 | { | |
2039 | struct memcg_batch_info *batch = NULL; | |
2040 | bool uncharge_memsw = true; | |
2041 | /* If swapout, usage of swap doesn't decrease */ | |
2042 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2043 | uncharge_memsw = false; | |
2044 | /* | |
2045 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
2046 | * In those cases, all pages freed continously can be expected to be in | |
2047 | * the same cgroup and we have chance to coalesce uncharges. | |
2048 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2049 | * because we want to do uncharge as soon as possible. | |
2050 | */ | |
2051 | if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE)) | |
2052 | goto direct_uncharge; | |
2053 | ||
2054 | batch = ¤t->memcg_batch; | |
2055 | /* | |
2056 | * In usual, we do css_get() when we remember memcg pointer. | |
2057 | * But in this case, we keep res->usage until end of a series of | |
2058 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2059 | */ | |
2060 | if (!batch->memcg) | |
2061 | batch->memcg = mem; | |
2062 | /* | |
2063 | * In typical case, batch->memcg == mem. This means we can | |
2064 | * merge a series of uncharges to an uncharge of res_counter. | |
2065 | * If not, we uncharge res_counter ony by one. | |
2066 | */ | |
2067 | if (batch->memcg != mem) | |
2068 | goto direct_uncharge; | |
2069 | /* remember freed charge and uncharge it later */ | |
2070 | batch->bytes += PAGE_SIZE; | |
2071 | if (uncharge_memsw) | |
2072 | batch->memsw_bytes += PAGE_SIZE; | |
2073 | return; | |
2074 | direct_uncharge: | |
2075 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
2076 | if (uncharge_memsw) | |
2077 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
2078 | return; | |
2079 | } | |
7a81b88c | 2080 | |
8a9f3ccd | 2081 | /* |
69029cd5 | 2082 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2083 | */ |
8c7c6e34 | 2084 | static struct mem_cgroup * |
69029cd5 | 2085 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2086 | { |
8289546e | 2087 | struct page_cgroup *pc; |
8c7c6e34 | 2088 | struct mem_cgroup *mem = NULL; |
072c56c1 | 2089 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 2090 | |
f8d66542 | 2091 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2092 | return NULL; |
4077960e | 2093 | |
d13d1443 | 2094 | if (PageSwapCache(page)) |
8c7c6e34 | 2095 | return NULL; |
d13d1443 | 2096 | |
8697d331 | 2097 | /* |
3c541e14 | 2098 | * Check if our page_cgroup is valid |
8697d331 | 2099 | */ |
52d4b9ac KH |
2100 | pc = lookup_page_cgroup(page); |
2101 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2102 | return NULL; |
b9c565d5 | 2103 | |
52d4b9ac | 2104 | lock_page_cgroup(pc); |
d13d1443 | 2105 | |
8c7c6e34 KH |
2106 | mem = pc->mem_cgroup; |
2107 | ||
d13d1443 KH |
2108 | if (!PageCgroupUsed(pc)) |
2109 | goto unlock_out; | |
2110 | ||
2111 | switch (ctype) { | |
2112 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2113 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
d13d1443 KH |
2114 | if (page_mapped(page)) |
2115 | goto unlock_out; | |
2116 | break; | |
2117 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2118 | if (!PageAnon(page)) { /* Shared memory */ | |
2119 | if (page->mapping && !page_is_file_cache(page)) | |
2120 | goto unlock_out; | |
2121 | } else if (page_mapped(page)) /* Anon */ | |
2122 | goto unlock_out; | |
2123 | break; | |
2124 | default: | |
2125 | break; | |
52d4b9ac | 2126 | } |
d13d1443 | 2127 | |
569b846d KH |
2128 | if (!mem_cgroup_is_root(mem)) |
2129 | __do_uncharge(mem, ctype); | |
0c3e73e8 BS |
2130 | if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
2131 | mem_cgroup_swap_statistics(mem, true); | |
08e552c6 | 2132 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 2133 | |
52d4b9ac | 2134 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2135 | /* |
2136 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2137 | * freed from LRU. This is safe because uncharged page is expected not | |
2138 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2139 | * special functions. | |
2140 | */ | |
b9c565d5 | 2141 | |
69029cd5 | 2142 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 2143 | unlock_page_cgroup(pc); |
fb59e9f1 | 2144 | |
4e649152 | 2145 | if (mem_cgroup_soft_limit_check(mem)) |
f64c3f54 | 2146 | mem_cgroup_update_tree(mem, page); |
a7fe942e KH |
2147 | /* at swapout, this memcg will be accessed to record to swap */ |
2148 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2149 | css_put(&mem->css); | |
6d12e2d8 | 2150 | |
8c7c6e34 | 2151 | return mem; |
d13d1443 KH |
2152 | |
2153 | unlock_out: | |
2154 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2155 | return NULL; |
3c541e14 BS |
2156 | } |
2157 | ||
69029cd5 KH |
2158 | void mem_cgroup_uncharge_page(struct page *page) |
2159 | { | |
52d4b9ac KH |
2160 | /* early check. */ |
2161 | if (page_mapped(page)) | |
2162 | return; | |
2163 | if (page->mapping && !PageAnon(page)) | |
2164 | return; | |
69029cd5 KH |
2165 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2166 | } | |
2167 | ||
2168 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2169 | { | |
2170 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2171 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2172 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2173 | } | |
2174 | ||
569b846d KH |
2175 | /* |
2176 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2177 | * In that cases, pages are freed continuously and we can expect pages | |
2178 | * are in the same memcg. All these calls itself limits the number of | |
2179 | * pages freed at once, then uncharge_start/end() is called properly. | |
2180 | * This may be called prural(2) times in a context, | |
2181 | */ | |
2182 | ||
2183 | void mem_cgroup_uncharge_start(void) | |
2184 | { | |
2185 | current->memcg_batch.do_batch++; | |
2186 | /* We can do nest. */ | |
2187 | if (current->memcg_batch.do_batch == 1) { | |
2188 | current->memcg_batch.memcg = NULL; | |
2189 | current->memcg_batch.bytes = 0; | |
2190 | current->memcg_batch.memsw_bytes = 0; | |
2191 | } | |
2192 | } | |
2193 | ||
2194 | void mem_cgroup_uncharge_end(void) | |
2195 | { | |
2196 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2197 | ||
2198 | if (!batch->do_batch) | |
2199 | return; | |
2200 | ||
2201 | batch->do_batch--; | |
2202 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2203 | return; | |
2204 | ||
2205 | if (!batch->memcg) | |
2206 | return; | |
2207 | /* | |
2208 | * This "batch->memcg" is valid without any css_get/put etc... | |
2209 | * bacause we hide charges behind us. | |
2210 | */ | |
2211 | if (batch->bytes) | |
2212 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2213 | if (batch->memsw_bytes) | |
2214 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
2215 | /* forget this pointer (for sanity check) */ | |
2216 | batch->memcg = NULL; | |
2217 | } | |
2218 | ||
e767e056 | 2219 | #ifdef CONFIG_SWAP |
8c7c6e34 | 2220 | /* |
e767e056 | 2221 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
2222 | * memcg information is recorded to swap_cgroup of "ent" |
2223 | */ | |
8a9478ca KH |
2224 | void |
2225 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
2226 | { |
2227 | struct mem_cgroup *memcg; | |
8a9478ca KH |
2228 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
2229 | ||
2230 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2231 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2232 | ||
2233 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 2234 | |
8c7c6e34 | 2235 | /* record memcg information */ |
8a9478ca | 2236 | if (do_swap_account && swapout && memcg) { |
a3b2d692 | 2237 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 KH |
2238 | mem_cgroup_get(memcg); |
2239 | } | |
8a9478ca | 2240 | if (swapout && memcg) |
a7fe942e | 2241 | css_put(&memcg->css); |
8c7c6e34 | 2242 | } |
e767e056 | 2243 | #endif |
8c7c6e34 KH |
2244 | |
2245 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2246 | /* | |
2247 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2248 | * uncharge "memsw" account. | |
2249 | */ | |
2250 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 2251 | { |
8c7c6e34 | 2252 | struct mem_cgroup *memcg; |
a3b2d692 | 2253 | unsigned short id; |
8c7c6e34 KH |
2254 | |
2255 | if (!do_swap_account) | |
2256 | return; | |
2257 | ||
a3b2d692 KH |
2258 | id = swap_cgroup_record(ent, 0); |
2259 | rcu_read_lock(); | |
2260 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2261 | if (memcg) { |
a3b2d692 KH |
2262 | /* |
2263 | * We uncharge this because swap is freed. | |
2264 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2265 | */ | |
0c3e73e8 | 2266 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2267 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2268 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2269 | mem_cgroup_put(memcg); |
2270 | } | |
a3b2d692 | 2271 | rcu_read_unlock(); |
d13d1443 | 2272 | } |
8c7c6e34 | 2273 | #endif |
d13d1443 | 2274 | |
ae41be37 | 2275 | /* |
01b1ae63 KH |
2276 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
2277 | * page belongs to. | |
ae41be37 | 2278 | */ |
01b1ae63 | 2279 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
2280 | { |
2281 | struct page_cgroup *pc; | |
e8589cc1 | 2282 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 2283 | int ret = 0; |
8869b8f6 | 2284 | |
f8d66542 | 2285 | if (mem_cgroup_disabled()) |
4077960e BS |
2286 | return 0; |
2287 | ||
52d4b9ac KH |
2288 | pc = lookup_page_cgroup(page); |
2289 | lock_page_cgroup(pc); | |
2290 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
2291 | mem = pc->mem_cgroup; |
2292 | css_get(&mem->css); | |
e8589cc1 | 2293 | } |
52d4b9ac | 2294 | unlock_page_cgroup(pc); |
01b1ae63 | 2295 | |
e8589cc1 | 2296 | if (mem) { |
f64c3f54 BS |
2297 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, |
2298 | page); | |
e8589cc1 KH |
2299 | css_put(&mem->css); |
2300 | } | |
01b1ae63 | 2301 | *ptr = mem; |
e8589cc1 | 2302 | return ret; |
ae41be37 | 2303 | } |
8869b8f6 | 2304 | |
69029cd5 | 2305 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
2306 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
2307 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 2308 | { |
01b1ae63 KH |
2309 | struct page *target, *unused; |
2310 | struct page_cgroup *pc; | |
2311 | enum charge_type ctype; | |
2312 | ||
2313 | if (!mem) | |
2314 | return; | |
88703267 | 2315 | cgroup_exclude_rmdir(&mem->css); |
01b1ae63 KH |
2316 | /* at migration success, oldpage->mapping is NULL. */ |
2317 | if (oldpage->mapping) { | |
2318 | target = oldpage; | |
2319 | unused = NULL; | |
2320 | } else { | |
2321 | target = newpage; | |
2322 | unused = oldpage; | |
2323 | } | |
2324 | ||
2325 | if (PageAnon(target)) | |
2326 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2327 | else if (page_is_file_cache(target)) | |
2328 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2329 | else | |
2330 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
2331 | ||
2332 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 2333 | if (unused) |
01b1ae63 KH |
2334 | __mem_cgroup_uncharge_common(unused, ctype); |
2335 | ||
2336 | pc = lookup_page_cgroup(target); | |
69029cd5 | 2337 | /* |
01b1ae63 KH |
2338 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
2339 | * So, double-counting is effectively avoided. | |
2340 | */ | |
2341 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
2342 | ||
2343 | /* | |
2344 | * Both of oldpage and newpage are still under lock_page(). | |
2345 | * Then, we don't have to care about race in radix-tree. | |
2346 | * But we have to be careful that this page is unmapped or not. | |
2347 | * | |
2348 | * There is a case for !page_mapped(). At the start of | |
2349 | * migration, oldpage was mapped. But now, it's zapped. | |
2350 | * But we know *target* page is not freed/reused under us. | |
2351 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 2352 | */ |
01b1ae63 KH |
2353 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
2354 | mem_cgroup_uncharge_page(target); | |
88703267 KH |
2355 | /* |
2356 | * At migration, we may charge account against cgroup which has no tasks | |
2357 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2358 | * In that case, we need to call pre_destroy() again. check it here. | |
2359 | */ | |
2360 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 2361 | } |
78fb7466 | 2362 | |
c9b0ed51 | 2363 | /* |
ae3abae6 DN |
2364 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
2365 | * Calling hierarchical_reclaim is not enough because we should update | |
2366 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
2367 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
2368 | * not from the memcg which this page would be charged to. | |
2369 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 2370 | */ |
ae3abae6 | 2371 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
2372 | struct mm_struct *mm, |
2373 | gfp_t gfp_mask) | |
c9b0ed51 | 2374 | { |
b5a84319 | 2375 | struct mem_cgroup *mem = NULL; |
ae3abae6 | 2376 | int ret; |
c9b0ed51 | 2377 | |
f8d66542 | 2378 | if (mem_cgroup_disabled()) |
cede86ac | 2379 | return 0; |
c9b0ed51 | 2380 | |
ae3abae6 DN |
2381 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2382 | if (!ret) | |
2383 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 2384 | |
ae3abae6 | 2385 | return ret; |
c9b0ed51 KH |
2386 | } |
2387 | ||
8c7c6e34 KH |
2388 | static DEFINE_MUTEX(set_limit_mutex); |
2389 | ||
d38d2a75 | 2390 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 2391 | unsigned long long val) |
628f4235 | 2392 | { |
81d39c20 | 2393 | int retry_count; |
8c7c6e34 | 2394 | u64 memswlimit; |
628f4235 | 2395 | int ret = 0; |
81d39c20 KH |
2396 | int children = mem_cgroup_count_children(memcg); |
2397 | u64 curusage, oldusage; | |
2398 | ||
2399 | /* | |
2400 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2401 | * is depends on callers. We set our retry-count to be function | |
2402 | * of # of children which we should visit in this loop. | |
2403 | */ | |
2404 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
2405 | ||
2406 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 2407 | |
8c7c6e34 | 2408 | while (retry_count) { |
628f4235 KH |
2409 | if (signal_pending(current)) { |
2410 | ret = -EINTR; | |
2411 | break; | |
2412 | } | |
8c7c6e34 KH |
2413 | /* |
2414 | * Rather than hide all in some function, I do this in | |
2415 | * open coded manner. You see what this really does. | |
2416 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2417 | */ | |
2418 | mutex_lock(&set_limit_mutex); | |
2419 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2420 | if (memswlimit < val) { | |
2421 | ret = -EINVAL; | |
2422 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
2423 | break; |
2424 | } | |
8c7c6e34 | 2425 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
2426 | if (!ret) { |
2427 | if (memswlimit == val) | |
2428 | memcg->memsw_is_minimum = true; | |
2429 | else | |
2430 | memcg->memsw_is_minimum = false; | |
2431 | } | |
8c7c6e34 KH |
2432 | mutex_unlock(&set_limit_mutex); |
2433 | ||
2434 | if (!ret) | |
2435 | break; | |
2436 | ||
aa20d489 | 2437 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
4e416953 | 2438 | MEM_CGROUP_RECLAIM_SHRINK); |
81d39c20 KH |
2439 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
2440 | /* Usage is reduced ? */ | |
2441 | if (curusage >= oldusage) | |
2442 | retry_count--; | |
2443 | else | |
2444 | oldusage = curusage; | |
8c7c6e34 | 2445 | } |
14797e23 | 2446 | |
8c7c6e34 KH |
2447 | return ret; |
2448 | } | |
2449 | ||
338c8431 LZ |
2450 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
2451 | unsigned long long val) | |
8c7c6e34 | 2452 | { |
81d39c20 | 2453 | int retry_count; |
8c7c6e34 | 2454 | u64 memlimit, oldusage, curusage; |
81d39c20 KH |
2455 | int children = mem_cgroup_count_children(memcg); |
2456 | int ret = -EBUSY; | |
8c7c6e34 | 2457 | |
81d39c20 KH |
2458 | /* see mem_cgroup_resize_res_limit */ |
2459 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
2460 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
2461 | while (retry_count) { |
2462 | if (signal_pending(current)) { | |
2463 | ret = -EINTR; | |
2464 | break; | |
2465 | } | |
2466 | /* | |
2467 | * Rather than hide all in some function, I do this in | |
2468 | * open coded manner. You see what this really does. | |
2469 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2470 | */ | |
2471 | mutex_lock(&set_limit_mutex); | |
2472 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2473 | if (memlimit > val) { | |
2474 | ret = -EINVAL; | |
2475 | mutex_unlock(&set_limit_mutex); | |
2476 | break; | |
2477 | } | |
2478 | ret = res_counter_set_limit(&memcg->memsw, val); | |
22a668d7 KH |
2479 | if (!ret) { |
2480 | if (memlimit == val) | |
2481 | memcg->memsw_is_minimum = true; | |
2482 | else | |
2483 | memcg->memsw_is_minimum = false; | |
2484 | } | |
8c7c6e34 KH |
2485 | mutex_unlock(&set_limit_mutex); |
2486 | ||
2487 | if (!ret) | |
2488 | break; | |
2489 | ||
4e416953 | 2490 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 BS |
2491 | MEM_CGROUP_RECLAIM_NOSWAP | |
2492 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 2493 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 2494 | /* Usage is reduced ? */ |
8c7c6e34 | 2495 | if (curusage >= oldusage) |
628f4235 | 2496 | retry_count--; |
81d39c20 KH |
2497 | else |
2498 | oldusage = curusage; | |
628f4235 KH |
2499 | } |
2500 | return ret; | |
2501 | } | |
2502 | ||
4e416953 BS |
2503 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2504 | gfp_t gfp_mask, int nid, | |
2505 | int zid) | |
2506 | { | |
2507 | unsigned long nr_reclaimed = 0; | |
2508 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2509 | unsigned long reclaimed; | |
2510 | int loop = 0; | |
2511 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 2512 | unsigned long long excess; |
4e416953 BS |
2513 | |
2514 | if (order > 0) | |
2515 | return 0; | |
2516 | ||
2517 | mctz = soft_limit_tree_node_zone(nid, zid); | |
2518 | /* | |
2519 | * This loop can run a while, specially if mem_cgroup's continuously | |
2520 | * keep exceeding their soft limit and putting the system under | |
2521 | * pressure | |
2522 | */ | |
2523 | do { | |
2524 | if (next_mz) | |
2525 | mz = next_mz; | |
2526 | else | |
2527 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2528 | if (!mz) | |
2529 | break; | |
2530 | ||
2531 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
2532 | gfp_mask, | |
2533 | MEM_CGROUP_RECLAIM_SOFT); | |
2534 | nr_reclaimed += reclaimed; | |
2535 | spin_lock(&mctz->lock); | |
2536 | ||
2537 | /* | |
2538 | * If we failed to reclaim anything from this memory cgroup | |
2539 | * it is time to move on to the next cgroup | |
2540 | */ | |
2541 | next_mz = NULL; | |
2542 | if (!reclaimed) { | |
2543 | do { | |
2544 | /* | |
2545 | * Loop until we find yet another one. | |
2546 | * | |
2547 | * By the time we get the soft_limit lock | |
2548 | * again, someone might have aded the | |
2549 | * group back on the RB tree. Iterate to | |
2550 | * make sure we get a different mem. | |
2551 | * mem_cgroup_largest_soft_limit_node returns | |
2552 | * NULL if no other cgroup is present on | |
2553 | * the tree | |
2554 | */ | |
2555 | next_mz = | |
2556 | __mem_cgroup_largest_soft_limit_node(mctz); | |
2557 | if (next_mz == mz) { | |
2558 | css_put(&next_mz->mem->css); | |
2559 | next_mz = NULL; | |
2560 | } else /* next_mz == NULL or other memcg */ | |
2561 | break; | |
2562 | } while (1); | |
2563 | } | |
4e416953 | 2564 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 2565 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
2566 | /* |
2567 | * One school of thought says that we should not add | |
2568 | * back the node to the tree if reclaim returns 0. | |
2569 | * But our reclaim could return 0, simply because due | |
2570 | * to priority we are exposing a smaller subset of | |
2571 | * memory to reclaim from. Consider this as a longer | |
2572 | * term TODO. | |
2573 | */ | |
ef8745c1 KH |
2574 | /* If excess == 0, no tree ops */ |
2575 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
2576 | spin_unlock(&mctz->lock); |
2577 | css_put(&mz->mem->css); | |
2578 | loop++; | |
2579 | /* | |
2580 | * Could not reclaim anything and there are no more | |
2581 | * mem cgroups to try or we seem to be looping without | |
2582 | * reclaiming anything. | |
2583 | */ | |
2584 | if (!nr_reclaimed && | |
2585 | (next_mz == NULL || | |
2586 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2587 | break; | |
2588 | } while (!nr_reclaimed); | |
2589 | if (next_mz) | |
2590 | css_put(&next_mz->mem->css); | |
2591 | return nr_reclaimed; | |
2592 | } | |
2593 | ||
cc847582 KH |
2594 | /* |
2595 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
2596 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
2597 | */ | |
f817ed48 | 2598 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 2599 | int node, int zid, enum lru_list lru) |
cc847582 | 2600 | { |
08e552c6 KH |
2601 | struct zone *zone; |
2602 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 2603 | struct page_cgroup *pc, *busy; |
08e552c6 | 2604 | unsigned long flags, loop; |
072c56c1 | 2605 | struct list_head *list; |
f817ed48 | 2606 | int ret = 0; |
072c56c1 | 2607 | |
08e552c6 KH |
2608 | zone = &NODE_DATA(node)->node_zones[zid]; |
2609 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 2610 | list = &mz->lists[lru]; |
cc847582 | 2611 | |
f817ed48 KH |
2612 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
2613 | /* give some margin against EBUSY etc...*/ | |
2614 | loop += 256; | |
2615 | busy = NULL; | |
2616 | while (loop--) { | |
2617 | ret = 0; | |
08e552c6 | 2618 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 2619 | if (list_empty(list)) { |
08e552c6 | 2620 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 2621 | break; |
f817ed48 KH |
2622 | } |
2623 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
2624 | if (busy == pc) { | |
2625 | list_move(&pc->lru, list); | |
648bcc77 | 2626 | busy = NULL; |
08e552c6 | 2627 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
2628 | continue; |
2629 | } | |
08e552c6 | 2630 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 2631 | |
2c26fdd7 | 2632 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 2633 | if (ret == -ENOMEM) |
52d4b9ac | 2634 | break; |
f817ed48 KH |
2635 | |
2636 | if (ret == -EBUSY || ret == -EINVAL) { | |
2637 | /* found lock contention or "pc" is obsolete. */ | |
2638 | busy = pc; | |
2639 | cond_resched(); | |
2640 | } else | |
2641 | busy = NULL; | |
cc847582 | 2642 | } |
08e552c6 | 2643 | |
f817ed48 KH |
2644 | if (!ret && !list_empty(list)) |
2645 | return -EBUSY; | |
2646 | return ret; | |
cc847582 KH |
2647 | } |
2648 | ||
2649 | /* | |
2650 | * make mem_cgroup's charge to be 0 if there is no task. | |
2651 | * This enables deleting this mem_cgroup. | |
2652 | */ | |
c1e862c1 | 2653 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 2654 | { |
f817ed48 KH |
2655 | int ret; |
2656 | int node, zid, shrink; | |
2657 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 2658 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 2659 | |
cc847582 | 2660 | css_get(&mem->css); |
f817ed48 KH |
2661 | |
2662 | shrink = 0; | |
c1e862c1 KH |
2663 | /* should free all ? */ |
2664 | if (free_all) | |
2665 | goto try_to_free; | |
f817ed48 | 2666 | move_account: |
fce66477 | 2667 | do { |
f817ed48 | 2668 | ret = -EBUSY; |
c1e862c1 KH |
2669 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
2670 | goto out; | |
2671 | ret = -EINTR; | |
2672 | if (signal_pending(current)) | |
cc847582 | 2673 | goto out; |
52d4b9ac KH |
2674 | /* This is for making all *used* pages to be on LRU. */ |
2675 | lru_add_drain_all(); | |
cdec2e42 | 2676 | drain_all_stock_sync(); |
f817ed48 | 2677 | ret = 0; |
299b4eaa | 2678 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 2679 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 2680 | enum lru_list l; |
f817ed48 KH |
2681 | for_each_lru(l) { |
2682 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 2683 | node, zid, l); |
f817ed48 KH |
2684 | if (ret) |
2685 | break; | |
2686 | } | |
1ecaab2b | 2687 | } |
f817ed48 KH |
2688 | if (ret) |
2689 | break; | |
2690 | } | |
2691 | /* it seems parent cgroup doesn't have enough mem */ | |
2692 | if (ret == -ENOMEM) | |
2693 | goto try_to_free; | |
52d4b9ac | 2694 | cond_resched(); |
fce66477 DN |
2695 | /* "ret" should also be checked to ensure all lists are empty. */ |
2696 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
2697 | out: |
2698 | css_put(&mem->css); | |
2699 | return ret; | |
f817ed48 KH |
2700 | |
2701 | try_to_free: | |
c1e862c1 KH |
2702 | /* returns EBUSY if there is a task or if we come here twice. */ |
2703 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
2704 | ret = -EBUSY; |
2705 | goto out; | |
2706 | } | |
c1e862c1 KH |
2707 | /* we call try-to-free pages for make this cgroup empty */ |
2708 | lru_add_drain_all(); | |
f817ed48 KH |
2709 | /* try to free all pages in this cgroup */ |
2710 | shrink = 1; | |
2711 | while (nr_retries && mem->res.usage > 0) { | |
2712 | int progress; | |
c1e862c1 KH |
2713 | |
2714 | if (signal_pending(current)) { | |
2715 | ret = -EINTR; | |
2716 | goto out; | |
2717 | } | |
a7885eb8 KM |
2718 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
2719 | false, get_swappiness(mem)); | |
c1e862c1 | 2720 | if (!progress) { |
f817ed48 | 2721 | nr_retries--; |
c1e862c1 | 2722 | /* maybe some writeback is necessary */ |
8aa7e847 | 2723 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2724 | } |
f817ed48 KH |
2725 | |
2726 | } | |
08e552c6 | 2727 | lru_add_drain(); |
f817ed48 | 2728 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 2729 | goto move_account; |
cc847582 KH |
2730 | } |
2731 | ||
c1e862c1 KH |
2732 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
2733 | { | |
2734 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
2735 | } | |
2736 | ||
2737 | ||
18f59ea7 BS |
2738 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
2739 | { | |
2740 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
2741 | } | |
2742 | ||
2743 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
2744 | u64 val) | |
2745 | { | |
2746 | int retval = 0; | |
2747 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
2748 | struct cgroup *parent = cont->parent; | |
2749 | struct mem_cgroup *parent_mem = NULL; | |
2750 | ||
2751 | if (parent) | |
2752 | parent_mem = mem_cgroup_from_cont(parent); | |
2753 | ||
2754 | cgroup_lock(); | |
2755 | /* | |
af901ca1 | 2756 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2757 | * in the child subtrees. If it is unset, then the change can |
2758 | * occur, provided the current cgroup has no children. | |
2759 | * | |
2760 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2761 | * set if there are no children. | |
2762 | */ | |
2763 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
2764 | (val == 1 || val == 0)) { | |
2765 | if (list_empty(&cont->children)) | |
2766 | mem->use_hierarchy = val; | |
2767 | else | |
2768 | retval = -EBUSY; | |
2769 | } else | |
2770 | retval = -EINVAL; | |
2771 | cgroup_unlock(); | |
2772 | ||
2773 | return retval; | |
2774 | } | |
2775 | ||
0c3e73e8 BS |
2776 | struct mem_cgroup_idx_data { |
2777 | s64 val; | |
2778 | enum mem_cgroup_stat_index idx; | |
2779 | }; | |
2780 | ||
2781 | static int | |
2782 | mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) | |
2783 | { | |
2784 | struct mem_cgroup_idx_data *d = data; | |
2785 | d->val += mem_cgroup_read_stat(&mem->stat, d->idx); | |
2786 | return 0; | |
2787 | } | |
2788 | ||
2789 | static void | |
2790 | mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, | |
2791 | enum mem_cgroup_stat_index idx, s64 *val) | |
2792 | { | |
2793 | struct mem_cgroup_idx_data d; | |
2794 | d.idx = idx; | |
2795 | d.val = 0; | |
2796 | mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat); | |
2797 | *val = d.val; | |
2798 | } | |
2799 | ||
2c3daa72 | 2800 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 2801 | { |
8c7c6e34 | 2802 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
0c3e73e8 | 2803 | u64 idx_val, val; |
8c7c6e34 KH |
2804 | int type, name; |
2805 | ||
2806 | type = MEMFILE_TYPE(cft->private); | |
2807 | name = MEMFILE_ATTR(cft->private); | |
2808 | switch (type) { | |
2809 | case _MEM: | |
0c3e73e8 BS |
2810 | if (name == RES_USAGE && mem_cgroup_is_root(mem)) { |
2811 | mem_cgroup_get_recursive_idx_stat(mem, | |
2812 | MEM_CGROUP_STAT_CACHE, &idx_val); | |
2813 | val = idx_val; | |
2814 | mem_cgroup_get_recursive_idx_stat(mem, | |
2815 | MEM_CGROUP_STAT_RSS, &idx_val); | |
2816 | val += idx_val; | |
2817 | val <<= PAGE_SHIFT; | |
2818 | } else | |
2819 | val = res_counter_read_u64(&mem->res, name); | |
8c7c6e34 KH |
2820 | break; |
2821 | case _MEMSWAP: | |
0c3e73e8 BS |
2822 | if (name == RES_USAGE && mem_cgroup_is_root(mem)) { |
2823 | mem_cgroup_get_recursive_idx_stat(mem, | |
2824 | MEM_CGROUP_STAT_CACHE, &idx_val); | |
2825 | val = idx_val; | |
2826 | mem_cgroup_get_recursive_idx_stat(mem, | |
2827 | MEM_CGROUP_STAT_RSS, &idx_val); | |
2828 | val += idx_val; | |
2829 | mem_cgroup_get_recursive_idx_stat(mem, | |
2830 | MEM_CGROUP_STAT_SWAPOUT, &idx_val); | |
cd9b45b7 | 2831 | val += idx_val; |
0c3e73e8 BS |
2832 | val <<= PAGE_SHIFT; |
2833 | } else | |
2834 | val = res_counter_read_u64(&mem->memsw, name); | |
8c7c6e34 KH |
2835 | break; |
2836 | default: | |
2837 | BUG(); | |
2838 | break; | |
2839 | } | |
2840 | return val; | |
8cdea7c0 | 2841 | } |
628f4235 KH |
2842 | /* |
2843 | * The user of this function is... | |
2844 | * RES_LIMIT. | |
2845 | */ | |
856c13aa PM |
2846 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
2847 | const char *buffer) | |
8cdea7c0 | 2848 | { |
628f4235 | 2849 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 2850 | int type, name; |
628f4235 KH |
2851 | unsigned long long val; |
2852 | int ret; | |
2853 | ||
8c7c6e34 KH |
2854 | type = MEMFILE_TYPE(cft->private); |
2855 | name = MEMFILE_ATTR(cft->private); | |
2856 | switch (name) { | |
628f4235 | 2857 | case RES_LIMIT: |
4b3bde4c BS |
2858 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2859 | ret = -EINVAL; | |
2860 | break; | |
2861 | } | |
628f4235 KH |
2862 | /* This function does all necessary parse...reuse it */ |
2863 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
2864 | if (ret) |
2865 | break; | |
2866 | if (type == _MEM) | |
628f4235 | 2867 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
2868 | else |
2869 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 2870 | break; |
296c81d8 BS |
2871 | case RES_SOFT_LIMIT: |
2872 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
2873 | if (ret) | |
2874 | break; | |
2875 | /* | |
2876 | * For memsw, soft limits are hard to implement in terms | |
2877 | * of semantics, for now, we support soft limits for | |
2878 | * control without swap | |
2879 | */ | |
2880 | if (type == _MEM) | |
2881 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
2882 | else | |
2883 | ret = -EINVAL; | |
2884 | break; | |
628f4235 KH |
2885 | default: |
2886 | ret = -EINVAL; /* should be BUG() ? */ | |
2887 | break; | |
2888 | } | |
2889 | return ret; | |
8cdea7c0 BS |
2890 | } |
2891 | ||
fee7b548 KH |
2892 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
2893 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
2894 | { | |
2895 | struct cgroup *cgroup; | |
2896 | unsigned long long min_limit, min_memsw_limit, tmp; | |
2897 | ||
2898 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2899 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2900 | cgroup = memcg->css.cgroup; | |
2901 | if (!memcg->use_hierarchy) | |
2902 | goto out; | |
2903 | ||
2904 | while (cgroup->parent) { | |
2905 | cgroup = cgroup->parent; | |
2906 | memcg = mem_cgroup_from_cont(cgroup); | |
2907 | if (!memcg->use_hierarchy) | |
2908 | break; | |
2909 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2910 | min_limit = min(min_limit, tmp); | |
2911 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2912 | min_memsw_limit = min(min_memsw_limit, tmp); | |
2913 | } | |
2914 | out: | |
2915 | *mem_limit = min_limit; | |
2916 | *memsw_limit = min_memsw_limit; | |
2917 | return; | |
2918 | } | |
2919 | ||
29f2a4da | 2920 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
2921 | { |
2922 | struct mem_cgroup *mem; | |
8c7c6e34 | 2923 | int type, name; |
c84872e1 PE |
2924 | |
2925 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
2926 | type = MEMFILE_TYPE(event); |
2927 | name = MEMFILE_ATTR(event); | |
2928 | switch (name) { | |
29f2a4da | 2929 | case RES_MAX_USAGE: |
8c7c6e34 KH |
2930 | if (type == _MEM) |
2931 | res_counter_reset_max(&mem->res); | |
2932 | else | |
2933 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
2934 | break; |
2935 | case RES_FAILCNT: | |
8c7c6e34 KH |
2936 | if (type == _MEM) |
2937 | res_counter_reset_failcnt(&mem->res); | |
2938 | else | |
2939 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
2940 | break; |
2941 | } | |
f64c3f54 | 2942 | |
85cc59db | 2943 | return 0; |
c84872e1 PE |
2944 | } |
2945 | ||
7dc74be0 DN |
2946 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
2947 | struct cftype *cft) | |
2948 | { | |
2949 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
2950 | } | |
2951 | ||
2952 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
2953 | struct cftype *cft, u64 val) | |
2954 | { | |
2955 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
2956 | ||
2957 | if (val >= (1 << NR_MOVE_TYPE)) | |
2958 | return -EINVAL; | |
2959 | /* | |
2960 | * We check this value several times in both in can_attach() and | |
2961 | * attach(), so we need cgroup lock to prevent this value from being | |
2962 | * inconsistent. | |
2963 | */ | |
2964 | cgroup_lock(); | |
2965 | mem->move_charge_at_immigrate = val; | |
2966 | cgroup_unlock(); | |
2967 | ||
2968 | return 0; | |
2969 | } | |
2970 | ||
14067bb3 KH |
2971 | |
2972 | /* For read statistics */ | |
2973 | enum { | |
2974 | MCS_CACHE, | |
2975 | MCS_RSS, | |
d8046582 | 2976 | MCS_FILE_MAPPED, |
14067bb3 KH |
2977 | MCS_PGPGIN, |
2978 | MCS_PGPGOUT, | |
1dd3a273 | 2979 | MCS_SWAP, |
14067bb3 KH |
2980 | MCS_INACTIVE_ANON, |
2981 | MCS_ACTIVE_ANON, | |
2982 | MCS_INACTIVE_FILE, | |
2983 | MCS_ACTIVE_FILE, | |
2984 | MCS_UNEVICTABLE, | |
2985 | NR_MCS_STAT, | |
2986 | }; | |
2987 | ||
2988 | struct mcs_total_stat { | |
2989 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
2990 | }; |
2991 | ||
14067bb3 KH |
2992 | struct { |
2993 | char *local_name; | |
2994 | char *total_name; | |
2995 | } memcg_stat_strings[NR_MCS_STAT] = { | |
2996 | {"cache", "total_cache"}, | |
2997 | {"rss", "total_rss"}, | |
d69b042f | 2998 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
2999 | {"pgpgin", "total_pgpgin"}, |
3000 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3001 | {"swap", "total_swap"}, |
14067bb3 KH |
3002 | {"inactive_anon", "total_inactive_anon"}, |
3003 | {"active_anon", "total_active_anon"}, | |
3004 | {"inactive_file", "total_inactive_file"}, | |
3005 | {"active_file", "total_active_file"}, | |
3006 | {"unevictable", "total_unevictable"} | |
3007 | }; | |
3008 | ||
3009 | ||
3010 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
3011 | { | |
3012 | struct mcs_total_stat *s = data; | |
3013 | s64 val; | |
3014 | ||
3015 | /* per cpu stat */ | |
3016 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE); | |
3017 | s->stat[MCS_CACHE] += val * PAGE_SIZE; | |
3018 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
3019 | s->stat[MCS_RSS] += val * PAGE_SIZE; | |
d8046582 KH |
3020 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_FILE_MAPPED); |
3021 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; | |
14067bb3 KH |
3022 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT); |
3023 | s->stat[MCS_PGPGIN] += val; | |
3024 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); | |
3025 | s->stat[MCS_PGPGOUT] += val; | |
1dd3a273 DN |
3026 | if (do_swap_account) { |
3027 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_SWAPOUT); | |
3028 | s->stat[MCS_SWAP] += val * PAGE_SIZE; | |
3029 | } | |
14067bb3 KH |
3030 | |
3031 | /* per zone stat */ | |
3032 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
3033 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
3034 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
3035 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
3036 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
3037 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
3038 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
3039 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
3040 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
3041 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
3042 | return 0; | |
3043 | } | |
3044 | ||
3045 | static void | |
3046 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
3047 | { | |
3048 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
3049 | } | |
3050 | ||
c64745cf PM |
3051 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
3052 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 3053 | { |
d2ceb9b7 | 3054 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 3055 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
3056 | int i; |
3057 | ||
14067bb3 KH |
3058 | memset(&mystat, 0, sizeof(mystat)); |
3059 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 3060 | |
1dd3a273 DN |
3061 | for (i = 0; i < NR_MCS_STAT; i++) { |
3062 | if (i == MCS_SWAP && !do_swap_account) | |
3063 | continue; | |
14067bb3 | 3064 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 3065 | } |
7b854121 | 3066 | |
14067bb3 | 3067 | /* Hierarchical information */ |
fee7b548 KH |
3068 | { |
3069 | unsigned long long limit, memsw_limit; | |
3070 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
3071 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
3072 | if (do_swap_account) | |
3073 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
3074 | } | |
7f016ee8 | 3075 | |
14067bb3 KH |
3076 | memset(&mystat, 0, sizeof(mystat)); |
3077 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
3078 | for (i = 0; i < NR_MCS_STAT; i++) { |
3079 | if (i == MCS_SWAP && !do_swap_account) | |
3080 | continue; | |
14067bb3 | 3081 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 3082 | } |
14067bb3 | 3083 | |
7f016ee8 | 3084 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 3085 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
3086 | |
3087 | { | |
3088 | int nid, zid; | |
3089 | struct mem_cgroup_per_zone *mz; | |
3090 | unsigned long recent_rotated[2] = {0, 0}; | |
3091 | unsigned long recent_scanned[2] = {0, 0}; | |
3092 | ||
3093 | for_each_online_node(nid) | |
3094 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3095 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3096 | ||
3097 | recent_rotated[0] += | |
3098 | mz->reclaim_stat.recent_rotated[0]; | |
3099 | recent_rotated[1] += | |
3100 | mz->reclaim_stat.recent_rotated[1]; | |
3101 | recent_scanned[0] += | |
3102 | mz->reclaim_stat.recent_scanned[0]; | |
3103 | recent_scanned[1] += | |
3104 | mz->reclaim_stat.recent_scanned[1]; | |
3105 | } | |
3106 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3107 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3108 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3109 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3110 | } | |
3111 | #endif | |
3112 | ||
d2ceb9b7 KH |
3113 | return 0; |
3114 | } | |
3115 | ||
a7885eb8 KM |
3116 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
3117 | { | |
3118 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3119 | ||
3120 | return get_swappiness(memcg); | |
3121 | } | |
3122 | ||
3123 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3124 | u64 val) | |
3125 | { | |
3126 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3127 | struct mem_cgroup *parent; | |
068b38c1 | 3128 | |
a7885eb8 KM |
3129 | if (val > 100) |
3130 | return -EINVAL; | |
3131 | ||
3132 | if (cgrp->parent == NULL) | |
3133 | return -EINVAL; | |
3134 | ||
3135 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
3136 | |
3137 | cgroup_lock(); | |
3138 | ||
a7885eb8 KM |
3139 | /* If under hierarchy, only empty-root can set this value */ |
3140 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
3141 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3142 | cgroup_unlock(); | |
a7885eb8 | 3143 | return -EINVAL; |
068b38c1 | 3144 | } |
a7885eb8 KM |
3145 | |
3146 | spin_lock(&memcg->reclaim_param_lock); | |
3147 | memcg->swappiness = val; | |
3148 | spin_unlock(&memcg->reclaim_param_lock); | |
3149 | ||
068b38c1 LZ |
3150 | cgroup_unlock(); |
3151 | ||
a7885eb8 KM |
3152 | return 0; |
3153 | } | |
3154 | ||
c1e862c1 | 3155 | |
8cdea7c0 BS |
3156 | static struct cftype mem_cgroup_files[] = { |
3157 | { | |
0eea1030 | 3158 | .name = "usage_in_bytes", |
8c7c6e34 | 3159 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 3160 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3161 | }, |
c84872e1 PE |
3162 | { |
3163 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3164 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 3165 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
3166 | .read_u64 = mem_cgroup_read, |
3167 | }, | |
8cdea7c0 | 3168 | { |
0eea1030 | 3169 | .name = "limit_in_bytes", |
8c7c6e34 | 3170 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 3171 | .write_string = mem_cgroup_write, |
2c3daa72 | 3172 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3173 | }, |
296c81d8 BS |
3174 | { |
3175 | .name = "soft_limit_in_bytes", | |
3176 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
3177 | .write_string = mem_cgroup_write, | |
3178 | .read_u64 = mem_cgroup_read, | |
3179 | }, | |
8cdea7c0 BS |
3180 | { |
3181 | .name = "failcnt", | |
8c7c6e34 | 3182 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 3183 | .trigger = mem_cgroup_reset, |
2c3daa72 | 3184 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3185 | }, |
d2ceb9b7 KH |
3186 | { |
3187 | .name = "stat", | |
c64745cf | 3188 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 3189 | }, |
c1e862c1 KH |
3190 | { |
3191 | .name = "force_empty", | |
3192 | .trigger = mem_cgroup_force_empty_write, | |
3193 | }, | |
18f59ea7 BS |
3194 | { |
3195 | .name = "use_hierarchy", | |
3196 | .write_u64 = mem_cgroup_hierarchy_write, | |
3197 | .read_u64 = mem_cgroup_hierarchy_read, | |
3198 | }, | |
a7885eb8 KM |
3199 | { |
3200 | .name = "swappiness", | |
3201 | .read_u64 = mem_cgroup_swappiness_read, | |
3202 | .write_u64 = mem_cgroup_swappiness_write, | |
3203 | }, | |
7dc74be0 DN |
3204 | { |
3205 | .name = "move_charge_at_immigrate", | |
3206 | .read_u64 = mem_cgroup_move_charge_read, | |
3207 | .write_u64 = mem_cgroup_move_charge_write, | |
3208 | }, | |
8cdea7c0 BS |
3209 | }; |
3210 | ||
8c7c6e34 KH |
3211 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3212 | static struct cftype memsw_cgroup_files[] = { | |
3213 | { | |
3214 | .name = "memsw.usage_in_bytes", | |
3215 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
3216 | .read_u64 = mem_cgroup_read, | |
3217 | }, | |
3218 | { | |
3219 | .name = "memsw.max_usage_in_bytes", | |
3220 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
3221 | .trigger = mem_cgroup_reset, | |
3222 | .read_u64 = mem_cgroup_read, | |
3223 | }, | |
3224 | { | |
3225 | .name = "memsw.limit_in_bytes", | |
3226 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
3227 | .write_string = mem_cgroup_write, | |
3228 | .read_u64 = mem_cgroup_read, | |
3229 | }, | |
3230 | { | |
3231 | .name = "memsw.failcnt", | |
3232 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
3233 | .trigger = mem_cgroup_reset, | |
3234 | .read_u64 = mem_cgroup_read, | |
3235 | }, | |
3236 | }; | |
3237 | ||
3238 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3239 | { | |
3240 | if (!do_swap_account) | |
3241 | return 0; | |
3242 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
3243 | ARRAY_SIZE(memsw_cgroup_files)); | |
3244 | }; | |
3245 | #else | |
3246 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3247 | { | |
3248 | return 0; | |
3249 | } | |
3250 | #endif | |
3251 | ||
6d12e2d8 KH |
3252 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3253 | { | |
3254 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 3255 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 3256 | enum lru_list l; |
41e3355d | 3257 | int zone, tmp = node; |
1ecaab2b KH |
3258 | /* |
3259 | * This routine is called against possible nodes. | |
3260 | * But it's BUG to call kmalloc() against offline node. | |
3261 | * | |
3262 | * TODO: this routine can waste much memory for nodes which will | |
3263 | * never be onlined. It's better to use memory hotplug callback | |
3264 | * function. | |
3265 | */ | |
41e3355d KH |
3266 | if (!node_state(node, N_NORMAL_MEMORY)) |
3267 | tmp = -1; | |
3268 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
3269 | if (!pn) |
3270 | return 1; | |
1ecaab2b | 3271 | |
6d12e2d8 KH |
3272 | mem->info.nodeinfo[node] = pn; |
3273 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
3274 | |
3275 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3276 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
3277 | for_each_lru(l) |
3278 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 3279 | mz->usage_in_excess = 0; |
4e416953 BS |
3280 | mz->on_tree = false; |
3281 | mz->mem = mem; | |
1ecaab2b | 3282 | } |
6d12e2d8 KH |
3283 | return 0; |
3284 | } | |
3285 | ||
1ecaab2b KH |
3286 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3287 | { | |
3288 | kfree(mem->info.nodeinfo[node]); | |
3289 | } | |
3290 | ||
c8dad2bb JB |
3291 | static int mem_cgroup_size(void) |
3292 | { | |
3293 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
3294 | return sizeof(struct mem_cgroup) + cpustat_size; | |
3295 | } | |
3296 | ||
33327948 KH |
3297 | static struct mem_cgroup *mem_cgroup_alloc(void) |
3298 | { | |
3299 | struct mem_cgroup *mem; | |
c8dad2bb | 3300 | int size = mem_cgroup_size(); |
33327948 | 3301 | |
c8dad2bb JB |
3302 | if (size < PAGE_SIZE) |
3303 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 3304 | else |
c8dad2bb | 3305 | mem = vmalloc(size); |
33327948 KH |
3306 | |
3307 | if (mem) | |
c8dad2bb | 3308 | memset(mem, 0, size); |
33327948 KH |
3309 | return mem; |
3310 | } | |
3311 | ||
8c7c6e34 KH |
3312 | /* |
3313 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
3314 | * (scanning all at force_empty is too costly...) | |
3315 | * | |
3316 | * Instead of clearing all references at force_empty, we remember | |
3317 | * the number of reference from swap_cgroup and free mem_cgroup when | |
3318 | * it goes down to 0. | |
3319 | * | |
8c7c6e34 KH |
3320 | * Removal of cgroup itself succeeds regardless of refs from swap. |
3321 | */ | |
3322 | ||
a7ba0eef | 3323 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 3324 | { |
08e552c6 KH |
3325 | int node; |
3326 | ||
f64c3f54 | 3327 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
3328 | free_css_id(&mem_cgroup_subsys, &mem->css); |
3329 | ||
08e552c6 KH |
3330 | for_each_node_state(node, N_POSSIBLE) |
3331 | free_mem_cgroup_per_zone_info(mem, node); | |
3332 | ||
c8dad2bb | 3333 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
3334 | kfree(mem); |
3335 | else | |
3336 | vfree(mem); | |
3337 | } | |
3338 | ||
8c7c6e34 KH |
3339 | static void mem_cgroup_get(struct mem_cgroup *mem) |
3340 | { | |
3341 | atomic_inc(&mem->refcnt); | |
3342 | } | |
3343 | ||
3344 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
3345 | { | |
7bcc1bb1 DN |
3346 | if (atomic_dec_and_test(&mem->refcnt)) { |
3347 | struct mem_cgroup *parent = parent_mem_cgroup(mem); | |
a7ba0eef | 3348 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
3349 | if (parent) |
3350 | mem_cgroup_put(parent); | |
3351 | } | |
8c7c6e34 KH |
3352 | } |
3353 | ||
7bcc1bb1 DN |
3354 | /* |
3355 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
3356 | */ | |
3357 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
3358 | { | |
3359 | if (!mem->res.parent) | |
3360 | return NULL; | |
3361 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
3362 | } | |
33327948 | 3363 | |
c077719b KH |
3364 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3365 | static void __init enable_swap_cgroup(void) | |
3366 | { | |
f8d66542 | 3367 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
3368 | do_swap_account = 1; |
3369 | } | |
3370 | #else | |
3371 | static void __init enable_swap_cgroup(void) | |
3372 | { | |
3373 | } | |
3374 | #endif | |
3375 | ||
f64c3f54 BS |
3376 | static int mem_cgroup_soft_limit_tree_init(void) |
3377 | { | |
3378 | struct mem_cgroup_tree_per_node *rtpn; | |
3379 | struct mem_cgroup_tree_per_zone *rtpz; | |
3380 | int tmp, node, zone; | |
3381 | ||
3382 | for_each_node_state(node, N_POSSIBLE) { | |
3383 | tmp = node; | |
3384 | if (!node_state(node, N_NORMAL_MEMORY)) | |
3385 | tmp = -1; | |
3386 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
3387 | if (!rtpn) | |
3388 | return 1; | |
3389 | ||
3390 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
3391 | ||
3392 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3393 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
3394 | rtpz->rb_root = RB_ROOT; | |
3395 | spin_lock_init(&rtpz->lock); | |
3396 | } | |
3397 | } | |
3398 | return 0; | |
3399 | } | |
3400 | ||
0eb253e2 | 3401 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
3402 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
3403 | { | |
28dbc4b6 | 3404 | struct mem_cgroup *mem, *parent; |
04046e1a | 3405 | long error = -ENOMEM; |
6d12e2d8 | 3406 | int node; |
8cdea7c0 | 3407 | |
c8dad2bb JB |
3408 | mem = mem_cgroup_alloc(); |
3409 | if (!mem) | |
04046e1a | 3410 | return ERR_PTR(error); |
78fb7466 | 3411 | |
6d12e2d8 KH |
3412 | for_each_node_state(node, N_POSSIBLE) |
3413 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
3414 | goto free_out; | |
f64c3f54 | 3415 | |
c077719b | 3416 | /* root ? */ |
28dbc4b6 | 3417 | if (cont->parent == NULL) { |
cdec2e42 | 3418 | int cpu; |
c077719b | 3419 | enable_swap_cgroup(); |
28dbc4b6 | 3420 | parent = NULL; |
4b3bde4c | 3421 | root_mem_cgroup = mem; |
f64c3f54 BS |
3422 | if (mem_cgroup_soft_limit_tree_init()) |
3423 | goto free_out; | |
cdec2e42 KH |
3424 | for_each_possible_cpu(cpu) { |
3425 | struct memcg_stock_pcp *stock = | |
3426 | &per_cpu(memcg_stock, cpu); | |
3427 | INIT_WORK(&stock->work, drain_local_stock); | |
3428 | } | |
3429 | hotcpu_notifier(memcg_stock_cpu_callback, 0); | |
18f59ea7 | 3430 | } else { |
28dbc4b6 | 3431 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 BS |
3432 | mem->use_hierarchy = parent->use_hierarchy; |
3433 | } | |
28dbc4b6 | 3434 | |
18f59ea7 BS |
3435 | if (parent && parent->use_hierarchy) { |
3436 | res_counter_init(&mem->res, &parent->res); | |
3437 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
3438 | /* |
3439 | * We increment refcnt of the parent to ensure that we can | |
3440 | * safely access it on res_counter_charge/uncharge. | |
3441 | * This refcnt will be decremented when freeing this | |
3442 | * mem_cgroup(see mem_cgroup_put). | |
3443 | */ | |
3444 | mem_cgroup_get(parent); | |
18f59ea7 BS |
3445 | } else { |
3446 | res_counter_init(&mem->res, NULL); | |
3447 | res_counter_init(&mem->memsw, NULL); | |
3448 | } | |
04046e1a | 3449 | mem->last_scanned_child = 0; |
2733c06a | 3450 | spin_lock_init(&mem->reclaim_param_lock); |
6d61ef40 | 3451 | |
a7885eb8 KM |
3452 | if (parent) |
3453 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 3454 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 3455 | mem->move_charge_at_immigrate = 0; |
8cdea7c0 | 3456 | return &mem->css; |
6d12e2d8 | 3457 | free_out: |
a7ba0eef | 3458 | __mem_cgroup_free(mem); |
4b3bde4c | 3459 | root_mem_cgroup = NULL; |
04046e1a | 3460 | return ERR_PTR(error); |
8cdea7c0 BS |
3461 | } |
3462 | ||
ec64f515 | 3463 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
3464 | struct cgroup *cont) |
3465 | { | |
3466 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
3467 | |
3468 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
3469 | } |
3470 | ||
8cdea7c0 BS |
3471 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
3472 | struct cgroup *cont) | |
3473 | { | |
c268e994 | 3474 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 3475 | |
c268e994 | 3476 | mem_cgroup_put(mem); |
8cdea7c0 BS |
3477 | } |
3478 | ||
3479 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
3480 | struct cgroup *cont) | |
3481 | { | |
8c7c6e34 KH |
3482 | int ret; |
3483 | ||
3484 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
3485 | ARRAY_SIZE(mem_cgroup_files)); | |
3486 | ||
3487 | if (!ret) | |
3488 | ret = register_memsw_files(cont, ss); | |
3489 | return ret; | |
8cdea7c0 BS |
3490 | } |
3491 | ||
7dc74be0 | 3492 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
3493 | #define PRECHARGE_COUNT_AT_ONCE 256 |
3494 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 3495 | { |
854ffa8d DN |
3496 | int ret = 0; |
3497 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
3498 | struct mem_cgroup *mem = mc.to; |
3499 | ||
854ffa8d DN |
3500 | if (mem_cgroup_is_root(mem)) { |
3501 | mc.precharge += count; | |
3502 | /* we don't need css_get for root */ | |
3503 | return ret; | |
3504 | } | |
3505 | /* try to charge at once */ | |
3506 | if (count > 1) { | |
3507 | struct res_counter *dummy; | |
3508 | /* | |
3509 | * "mem" cannot be under rmdir() because we've already checked | |
3510 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
3511 | * are still under the same cgroup_mutex. So we can postpone | |
3512 | * css_get(). | |
3513 | */ | |
3514 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
3515 | goto one_by_one; | |
3516 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
3517 | PAGE_SIZE * count, &dummy)) { | |
3518 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
3519 | goto one_by_one; | |
3520 | } | |
3521 | mc.precharge += count; | |
3522 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
3523 | WARN_ON_ONCE(count > INT_MAX); | |
3524 | __css_get(&mem->css, (int)count); | |
3525 | return ret; | |
3526 | } | |
3527 | one_by_one: | |
3528 | /* fall back to one by one charge */ | |
3529 | while (count--) { | |
3530 | if (signal_pending(current)) { | |
3531 | ret = -EINTR; | |
3532 | break; | |
3533 | } | |
3534 | if (!batch_count--) { | |
3535 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
3536 | cond_resched(); | |
3537 | } | |
3538 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, | |
3539 | false, NULL); | |
3540 | if (ret || !mem) | |
3541 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
3542 | return -ENOMEM; | |
3543 | mc.precharge++; | |
3544 | } | |
4ffef5fe DN |
3545 | return ret; |
3546 | } | |
3547 | ||
3548 | /** | |
3549 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
3550 | * @vma: the vma the pte to be checked belongs | |
3551 | * @addr: the address corresponding to the pte to be checked | |
3552 | * @ptent: the pte to be checked | |
3553 | * @target: the pointer the target page will be stored(can be NULL) | |
3554 | * | |
3555 | * Returns | |
3556 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
3557 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
3558 | * move charge. if @target is not NULL, the page is stored in target->page | |
3559 | * with extra refcnt got(Callers should handle it). | |
3560 | * | |
3561 | * Called with pte lock held. | |
3562 | */ | |
3563 | /* We add a new member later. */ | |
3564 | union mc_target { | |
3565 | struct page *page; | |
3566 | }; | |
3567 | ||
3568 | /* We add a new type later. */ | |
3569 | enum mc_target_type { | |
3570 | MC_TARGET_NONE, /* not used */ | |
3571 | MC_TARGET_PAGE, | |
3572 | }; | |
3573 | ||
3574 | static int is_target_pte_for_mc(struct vm_area_struct *vma, | |
3575 | unsigned long addr, pte_t ptent, union mc_target *target) | |
3576 | { | |
3577 | struct page *page; | |
3578 | struct page_cgroup *pc; | |
3579 | int ret = 0; | |
3580 | bool move_anon = test_bit(MOVE_CHARGE_TYPE_ANON, | |
3581 | &mc.to->move_charge_at_immigrate); | |
3582 | ||
3583 | if (!pte_present(ptent)) | |
3584 | return 0; | |
3585 | ||
3586 | page = vm_normal_page(vma, addr, ptent); | |
3587 | if (!page || !page_mapped(page)) | |
3588 | return 0; | |
3589 | /* | |
3590 | * TODO: We don't move charges of file(including shmem/tmpfs) pages for | |
3591 | * now. | |
3592 | */ | |
3593 | if (!move_anon || !PageAnon(page)) | |
3594 | return 0; | |
3595 | /* | |
3596 | * TODO: We don't move charges of shared(used by multiple processes) | |
3597 | * pages for now. | |
3598 | */ | |
3599 | if (page_mapcount(page) > 1) | |
3600 | return 0; | |
3601 | if (!get_page_unless_zero(page)) | |
3602 | return 0; | |
3603 | ||
3604 | pc = lookup_page_cgroup(page); | |
3605 | /* | |
3606 | * Do only loose check w/o page_cgroup lock. mem_cgroup_move_account() | |
3607 | * checks the pc is valid or not under the lock. | |
3608 | */ | |
3609 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
3610 | ret = MC_TARGET_PAGE; | |
3611 | if (target) | |
3612 | target->page = page; | |
3613 | } | |
3614 | ||
3615 | if (!ret || !target) | |
3616 | put_page(page); | |
3617 | ||
3618 | return ret; | |
3619 | } | |
3620 | ||
3621 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
3622 | unsigned long addr, unsigned long end, | |
3623 | struct mm_walk *walk) | |
3624 | { | |
3625 | struct vm_area_struct *vma = walk->private; | |
3626 | pte_t *pte; | |
3627 | spinlock_t *ptl; | |
3628 | ||
3629 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
3630 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
3631 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
3632 | mc.precharge++; /* increment precharge temporarily */ | |
3633 | pte_unmap_unlock(pte - 1, ptl); | |
3634 | cond_resched(); | |
3635 | ||
7dc74be0 DN |
3636 | return 0; |
3637 | } | |
3638 | ||
4ffef5fe DN |
3639 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
3640 | { | |
3641 | unsigned long precharge; | |
3642 | struct vm_area_struct *vma; | |
3643 | ||
3644 | down_read(&mm->mmap_sem); | |
3645 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3646 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
3647 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
3648 | .mm = mm, | |
3649 | .private = vma, | |
3650 | }; | |
3651 | if (is_vm_hugetlb_page(vma)) | |
3652 | continue; | |
3653 | /* TODO: We don't move charges of shmem/tmpfs pages for now. */ | |
3654 | if (vma->vm_flags & VM_SHARED) | |
3655 | continue; | |
3656 | walk_page_range(vma->vm_start, vma->vm_end, | |
3657 | &mem_cgroup_count_precharge_walk); | |
3658 | } | |
3659 | up_read(&mm->mmap_sem); | |
3660 | ||
3661 | precharge = mc.precharge; | |
3662 | mc.precharge = 0; | |
3663 | ||
3664 | return precharge; | |
3665 | } | |
3666 | ||
4ffef5fe DN |
3667 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
3668 | { | |
854ffa8d | 3669 | return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm)); |
4ffef5fe DN |
3670 | } |
3671 | ||
3672 | static void mem_cgroup_clear_mc(void) | |
3673 | { | |
3674 | /* we must uncharge all the leftover precharges from mc.to */ | |
854ffa8d DN |
3675 | if (mc.precharge) { |
3676 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
3677 | mc.precharge = 0; | |
3678 | } | |
3679 | /* | |
3680 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
3681 | * we must uncharge here. | |
3682 | */ | |
3683 | if (mc.moved_charge) { | |
3684 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
3685 | mc.moved_charge = 0; | |
4ffef5fe DN |
3686 | } |
3687 | mc.from = NULL; | |
3688 | mc.to = NULL; | |
8033b97c DN |
3689 | mc.moving_task = NULL; |
3690 | wake_up_all(&mc.waitq); | |
4ffef5fe DN |
3691 | } |
3692 | ||
7dc74be0 DN |
3693 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
3694 | struct cgroup *cgroup, | |
3695 | struct task_struct *p, | |
3696 | bool threadgroup) | |
3697 | { | |
3698 | int ret = 0; | |
3699 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
3700 | ||
3701 | if (mem->move_charge_at_immigrate) { | |
3702 | struct mm_struct *mm; | |
3703 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
3704 | ||
3705 | VM_BUG_ON(from == mem); | |
3706 | ||
3707 | mm = get_task_mm(p); | |
3708 | if (!mm) | |
3709 | return 0; | |
7dc74be0 | 3710 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
3711 | if (mm->owner == p) { |
3712 | VM_BUG_ON(mc.from); | |
3713 | VM_BUG_ON(mc.to); | |
3714 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 3715 | VM_BUG_ON(mc.moved_charge); |
8033b97c | 3716 | VM_BUG_ON(mc.moving_task); |
4ffef5fe DN |
3717 | mc.from = from; |
3718 | mc.to = mem; | |
3719 | mc.precharge = 0; | |
854ffa8d | 3720 | mc.moved_charge = 0; |
8033b97c | 3721 | mc.moving_task = current; |
4ffef5fe DN |
3722 | |
3723 | ret = mem_cgroup_precharge_mc(mm); | |
3724 | if (ret) | |
3725 | mem_cgroup_clear_mc(); | |
3726 | } | |
7dc74be0 DN |
3727 | mmput(mm); |
3728 | } | |
3729 | return ret; | |
3730 | } | |
3731 | ||
3732 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
3733 | struct cgroup *cgroup, | |
3734 | struct task_struct *p, | |
3735 | bool threadgroup) | |
3736 | { | |
4ffef5fe | 3737 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
3738 | } |
3739 | ||
4ffef5fe DN |
3740 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
3741 | unsigned long addr, unsigned long end, | |
3742 | struct mm_walk *walk) | |
7dc74be0 | 3743 | { |
4ffef5fe DN |
3744 | int ret = 0; |
3745 | struct vm_area_struct *vma = walk->private; | |
3746 | pte_t *pte; | |
3747 | spinlock_t *ptl; | |
3748 | ||
3749 | retry: | |
3750 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
3751 | for (; addr != end; addr += PAGE_SIZE) { | |
3752 | pte_t ptent = *(pte++); | |
3753 | union mc_target target; | |
3754 | int type; | |
3755 | struct page *page; | |
3756 | struct page_cgroup *pc; | |
3757 | ||
3758 | if (!mc.precharge) | |
3759 | break; | |
3760 | ||
3761 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
3762 | switch (type) { | |
3763 | case MC_TARGET_PAGE: | |
3764 | page = target.page; | |
3765 | if (isolate_lru_page(page)) | |
3766 | goto put; | |
3767 | pc = lookup_page_cgroup(page); | |
854ffa8d DN |
3768 | if (!mem_cgroup_move_account(pc, |
3769 | mc.from, mc.to, false)) { | |
4ffef5fe | 3770 | mc.precharge--; |
854ffa8d DN |
3771 | /* we uncharge from mc.from later. */ |
3772 | mc.moved_charge++; | |
4ffef5fe DN |
3773 | } |
3774 | putback_lru_page(page); | |
3775 | put: /* is_target_pte_for_mc() gets the page */ | |
3776 | put_page(page); | |
3777 | break; | |
3778 | default: | |
3779 | break; | |
3780 | } | |
3781 | } | |
3782 | pte_unmap_unlock(pte - 1, ptl); | |
3783 | cond_resched(); | |
3784 | ||
3785 | if (addr != end) { | |
3786 | /* | |
3787 | * We have consumed all precharges we got in can_attach(). | |
3788 | * We try charge one by one, but don't do any additional | |
3789 | * charges to mc.to if we have failed in charge once in attach() | |
3790 | * phase. | |
3791 | */ | |
854ffa8d | 3792 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
3793 | if (!ret) |
3794 | goto retry; | |
3795 | } | |
3796 | ||
3797 | return ret; | |
3798 | } | |
3799 | ||
3800 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
3801 | { | |
3802 | struct vm_area_struct *vma; | |
3803 | ||
3804 | lru_add_drain_all(); | |
3805 | down_read(&mm->mmap_sem); | |
3806 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3807 | int ret; | |
3808 | struct mm_walk mem_cgroup_move_charge_walk = { | |
3809 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
3810 | .mm = mm, | |
3811 | .private = vma, | |
3812 | }; | |
3813 | if (is_vm_hugetlb_page(vma)) | |
3814 | continue; | |
3815 | /* TODO: We don't move charges of shmem/tmpfs pages for now. */ | |
3816 | if (vma->vm_flags & VM_SHARED) | |
3817 | continue; | |
3818 | ret = walk_page_range(vma->vm_start, vma->vm_end, | |
3819 | &mem_cgroup_move_charge_walk); | |
3820 | if (ret) | |
3821 | /* | |
3822 | * means we have consumed all precharges and failed in | |
3823 | * doing additional charge. Just abandon here. | |
3824 | */ | |
3825 | break; | |
3826 | } | |
3827 | up_read(&mm->mmap_sem); | |
7dc74be0 DN |
3828 | } |
3829 | ||
67e465a7 BS |
3830 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
3831 | struct cgroup *cont, | |
3832 | struct cgroup *old_cont, | |
be367d09 BB |
3833 | struct task_struct *p, |
3834 | bool threadgroup) | |
67e465a7 | 3835 | { |
4ffef5fe DN |
3836 | struct mm_struct *mm; |
3837 | ||
3838 | if (!mc.to) | |
3839 | /* no need to move charge */ | |
3840 | return; | |
3841 | ||
3842 | mm = get_task_mm(p); | |
3843 | if (mm) { | |
3844 | mem_cgroup_move_charge(mm); | |
3845 | mmput(mm); | |
3846 | } | |
3847 | mem_cgroup_clear_mc(); | |
67e465a7 BS |
3848 | } |
3849 | ||
8cdea7c0 BS |
3850 | struct cgroup_subsys mem_cgroup_subsys = { |
3851 | .name = "memory", | |
3852 | .subsys_id = mem_cgroup_subsys_id, | |
3853 | .create = mem_cgroup_create, | |
df878fb0 | 3854 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
3855 | .destroy = mem_cgroup_destroy, |
3856 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
3857 | .can_attach = mem_cgroup_can_attach, |
3858 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 3859 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 3860 | .early_init = 0, |
04046e1a | 3861 | .use_id = 1, |
8cdea7c0 | 3862 | }; |
c077719b KH |
3863 | |
3864 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3865 | ||
3866 | static int __init disable_swap_account(char *s) | |
3867 | { | |
3868 | really_do_swap_account = 0; | |
3869 | return 1; | |
3870 | } | |
3871 | __setup("noswapaccount", disable_swap_account); | |
3872 | #endif |