]>
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 | 35 | #include <linux/swap.h> |
02491447 | 36 | #include <linux/swapops.h> |
66e1707b BS |
37 | #include <linux/spinlock.h> |
38 | #include <linux/fs.h> | |
d2ceb9b7 | 39 | #include <linux/seq_file.h> |
33327948 | 40 | #include <linux/vmalloc.h> |
b69408e8 | 41 | #include <linux/mm_inline.h> |
52d4b9ac | 42 | #include <linux/page_cgroup.h> |
cdec2e42 | 43 | #include <linux/cpu.h> |
08e552c6 | 44 | #include "internal.h" |
8cdea7c0 | 45 | |
8697d331 BS |
46 | #include <asm/uaccess.h> |
47 | ||
a181b0e8 | 48 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 49 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 50 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 51 | |
c077719b | 52 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 53 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b KH |
54 | int do_swap_account __read_mostly; |
55 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
56 | #else | |
57 | #define do_swap_account (0) | |
58 | #endif | |
59 | ||
f64c3f54 | 60 | #define SOFTLIMIT_EVENTS_THRESH (1000) |
c077719b | 61 | |
d52aa412 KH |
62 | /* |
63 | * Statistics for memory cgroup. | |
64 | */ | |
65 | enum mem_cgroup_stat_index { | |
66 | /* | |
67 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
68 | */ | |
69 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 70 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 71 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
55e462b0 BR |
72 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
73 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
f64c3f54 | 74 | MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */ |
0c3e73e8 | 75 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
d52aa412 KH |
76 | |
77 | MEM_CGROUP_STAT_NSTATS, | |
78 | }; | |
79 | ||
80 | struct mem_cgroup_stat_cpu { | |
81 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
82 | } ____cacheline_aligned_in_smp; | |
83 | ||
84 | struct mem_cgroup_stat { | |
c8dad2bb | 85 | struct mem_cgroup_stat_cpu cpustat[0]; |
d52aa412 KH |
86 | }; |
87 | ||
f64c3f54 BS |
88 | static inline void |
89 | __mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat, | |
90 | enum mem_cgroup_stat_index idx) | |
91 | { | |
92 | stat->count[idx] = 0; | |
93 | } | |
94 | ||
95 | static inline s64 | |
96 | __mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat, | |
97 | enum mem_cgroup_stat_index idx) | |
98 | { | |
99 | return stat->count[idx]; | |
100 | } | |
101 | ||
d52aa412 KH |
102 | /* |
103 | * For accounting under irq disable, no need for increment preempt count. | |
104 | */ | |
addb9efe | 105 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
d52aa412 KH |
106 | enum mem_cgroup_stat_index idx, int val) |
107 | { | |
addb9efe | 108 | stat->count[idx] += val; |
d52aa412 KH |
109 | } |
110 | ||
111 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, | |
112 | enum mem_cgroup_stat_index idx) | |
113 | { | |
114 | int cpu; | |
115 | s64 ret = 0; | |
116 | for_each_possible_cpu(cpu) | |
117 | ret += stat->cpustat[cpu].count[idx]; | |
118 | return ret; | |
119 | } | |
120 | ||
04046e1a KH |
121 | static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat) |
122 | { | |
123 | s64 ret; | |
124 | ||
125 | ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE); | |
126 | ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS); | |
127 | return ret; | |
128 | } | |
129 | ||
6d12e2d8 KH |
130 | /* |
131 | * per-zone information in memory controller. | |
132 | */ | |
6d12e2d8 | 133 | struct mem_cgroup_per_zone { |
072c56c1 KH |
134 | /* |
135 | * spin_lock to protect the per cgroup LRU | |
136 | */ | |
b69408e8 CL |
137 | struct list_head lists[NR_LRU_LISTS]; |
138 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
139 | |
140 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
141 | struct rb_node tree_node; /* RB tree node */ |
142 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
143 | /* the soft limit is exceeded*/ | |
144 | bool on_tree; | |
4e416953 BS |
145 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
146 | /* use container_of */ | |
6d12e2d8 KH |
147 | }; |
148 | /* Macro for accessing counter */ | |
149 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
150 | ||
151 | struct mem_cgroup_per_node { | |
152 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
153 | }; | |
154 | ||
155 | struct mem_cgroup_lru_info { | |
156 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
157 | }; | |
158 | ||
f64c3f54 BS |
159 | /* |
160 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
161 | * their hierarchy representation | |
162 | */ | |
163 | ||
164 | struct mem_cgroup_tree_per_zone { | |
165 | struct rb_root rb_root; | |
166 | spinlock_t lock; | |
167 | }; | |
168 | ||
169 | struct mem_cgroup_tree_per_node { | |
170 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
171 | }; | |
172 | ||
173 | struct mem_cgroup_tree { | |
174 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
175 | }; | |
176 | ||
177 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
178 | ||
8cdea7c0 BS |
179 | /* |
180 | * The memory controller data structure. The memory controller controls both | |
181 | * page cache and RSS per cgroup. We would eventually like to provide | |
182 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
183 | * to help the administrator determine what knobs to tune. | |
184 | * | |
185 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
186 | * we hit the water mark. May be even add a low water mark, such that |
187 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
188 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
189 | */ |
190 | struct mem_cgroup { | |
191 | struct cgroup_subsys_state css; | |
192 | /* | |
193 | * the counter to account for memory usage | |
194 | */ | |
195 | struct res_counter res; | |
8c7c6e34 KH |
196 | /* |
197 | * the counter to account for mem+swap usage. | |
198 | */ | |
199 | struct res_counter memsw; | |
78fb7466 PE |
200 | /* |
201 | * Per cgroup active and inactive list, similar to the | |
202 | * per zone LRU lists. | |
78fb7466 | 203 | */ |
6d12e2d8 | 204 | struct mem_cgroup_lru_info info; |
072c56c1 | 205 | |
2733c06a KM |
206 | /* |
207 | protect against reclaim related member. | |
208 | */ | |
209 | spinlock_t reclaim_param_lock; | |
210 | ||
6c48a1d0 | 211 | int prev_priority; /* for recording reclaim priority */ |
6d61ef40 BS |
212 | |
213 | /* | |
af901ca1 | 214 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 215 | * reclaimed from. |
6d61ef40 | 216 | */ |
04046e1a | 217 | int last_scanned_child; |
18f59ea7 BS |
218 | /* |
219 | * Should the accounting and control be hierarchical, per subtree? | |
220 | */ | |
221 | bool use_hierarchy; | |
a636b327 | 222 | unsigned long last_oom_jiffies; |
8c7c6e34 | 223 | atomic_t refcnt; |
14797e23 | 224 | |
a7885eb8 KM |
225 | unsigned int swappiness; |
226 | ||
22a668d7 KH |
227 | /* set when res.limit == memsw.limit */ |
228 | bool memsw_is_minimum; | |
229 | ||
7dc74be0 DN |
230 | /* |
231 | * Should we move charges of a task when a task is moved into this | |
232 | * mem_cgroup ? And what type of charges should we move ? | |
233 | */ | |
234 | unsigned long move_charge_at_immigrate; | |
235 | ||
d52aa412 | 236 | /* |
c8dad2bb | 237 | * statistics. This must be placed at the end of memcg. |
d52aa412 KH |
238 | */ |
239 | struct mem_cgroup_stat stat; | |
8cdea7c0 BS |
240 | }; |
241 | ||
7dc74be0 DN |
242 | /* Stuffs for move charges at task migration. */ |
243 | /* | |
244 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
245 | * left-shifted bitmap of these types. | |
246 | */ | |
247 | enum move_type { | |
4ffef5fe | 248 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
7dc74be0 DN |
249 | NR_MOVE_TYPE, |
250 | }; | |
251 | ||
4ffef5fe DN |
252 | /* "mc" and its members are protected by cgroup_mutex */ |
253 | static struct move_charge_struct { | |
254 | struct mem_cgroup *from; | |
255 | struct mem_cgroup *to; | |
256 | unsigned long precharge; | |
854ffa8d | 257 | unsigned long moved_charge; |
8033b97c DN |
258 | struct task_struct *moving_task; /* a task moving charges */ |
259 | wait_queue_head_t waitq; /* a waitq for other context */ | |
260 | } mc = { | |
261 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), | |
262 | }; | |
4ffef5fe | 263 | |
4e416953 BS |
264 | /* |
265 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
266 | * limit reclaim to prevent infinite loops, if they ever occur. | |
267 | */ | |
268 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
269 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
270 | ||
217bc319 KH |
271 | enum charge_type { |
272 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
273 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 274 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 275 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 276 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 277 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
278 | NR_CHARGE_TYPE, |
279 | }; | |
280 | ||
52d4b9ac KH |
281 | /* only for here (for easy reading.) */ |
282 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
283 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 284 | #define PCGF_LOCK (1UL << PCG_LOCK) |
4b3bde4c BS |
285 | /* Not used, but added here for completeness */ |
286 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
217bc319 | 287 | |
8c7c6e34 KH |
288 | /* for encoding cft->private value on file */ |
289 | #define _MEM (0) | |
290 | #define _MEMSWAP (1) | |
291 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
292 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
293 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
294 | ||
75822b44 BS |
295 | /* |
296 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
297 | */ | |
298 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
299 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
300 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
301 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
302 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
303 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 304 | |
8c7c6e34 KH |
305 | static void mem_cgroup_get(struct mem_cgroup *mem); |
306 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 307 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 308 | static void drain_all_stock_async(void); |
8c7c6e34 | 309 | |
f64c3f54 BS |
310 | static struct mem_cgroup_per_zone * |
311 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
312 | { | |
313 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
314 | } | |
315 | ||
d324236b WF |
316 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
317 | { | |
318 | return &mem->css; | |
319 | } | |
320 | ||
f64c3f54 BS |
321 | static struct mem_cgroup_per_zone * |
322 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
323 | { | |
324 | struct mem_cgroup *mem = pc->mem_cgroup; | |
325 | int nid = page_cgroup_nid(pc); | |
326 | int zid = page_cgroup_zid(pc); | |
327 | ||
328 | if (!mem) | |
329 | return NULL; | |
330 | ||
331 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
332 | } | |
333 | ||
334 | static struct mem_cgroup_tree_per_zone * | |
335 | soft_limit_tree_node_zone(int nid, int zid) | |
336 | { | |
337 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
338 | } | |
339 | ||
340 | static struct mem_cgroup_tree_per_zone * | |
341 | soft_limit_tree_from_page(struct page *page) | |
342 | { | |
343 | int nid = page_to_nid(page); | |
344 | int zid = page_zonenum(page); | |
345 | ||
346 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
347 | } | |
348 | ||
349 | static void | |
4e416953 | 350 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 351 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
352 | struct mem_cgroup_tree_per_zone *mctz, |
353 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
354 | { |
355 | struct rb_node **p = &mctz->rb_root.rb_node; | |
356 | struct rb_node *parent = NULL; | |
357 | struct mem_cgroup_per_zone *mz_node; | |
358 | ||
359 | if (mz->on_tree) | |
360 | return; | |
361 | ||
ef8745c1 KH |
362 | mz->usage_in_excess = new_usage_in_excess; |
363 | if (!mz->usage_in_excess) | |
364 | return; | |
f64c3f54 BS |
365 | while (*p) { |
366 | parent = *p; | |
367 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
368 | tree_node); | |
369 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
370 | p = &(*p)->rb_left; | |
371 | /* | |
372 | * We can't avoid mem cgroups that are over their soft | |
373 | * limit by the same amount | |
374 | */ | |
375 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
376 | p = &(*p)->rb_right; | |
377 | } | |
378 | rb_link_node(&mz->tree_node, parent, p); | |
379 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
380 | mz->on_tree = true; | |
4e416953 BS |
381 | } |
382 | ||
383 | static void | |
384 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
385 | struct mem_cgroup_per_zone *mz, | |
386 | struct mem_cgroup_tree_per_zone *mctz) | |
387 | { | |
388 | if (!mz->on_tree) | |
389 | return; | |
390 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
391 | mz->on_tree = false; | |
392 | } | |
393 | ||
f64c3f54 BS |
394 | static void |
395 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
396 | struct mem_cgroup_per_zone *mz, | |
397 | struct mem_cgroup_tree_per_zone *mctz) | |
398 | { | |
399 | spin_lock(&mctz->lock); | |
4e416953 | 400 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
401 | spin_unlock(&mctz->lock); |
402 | } | |
403 | ||
404 | static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem) | |
405 | { | |
406 | bool ret = false; | |
407 | int cpu; | |
408 | s64 val; | |
409 | struct mem_cgroup_stat_cpu *cpustat; | |
410 | ||
411 | cpu = get_cpu(); | |
412 | cpustat = &mem->stat.cpustat[cpu]; | |
413 | val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS); | |
414 | if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) { | |
415 | __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS); | |
416 | ret = true; | |
417 | } | |
418 | put_cpu(); | |
419 | return ret; | |
420 | } | |
421 | ||
422 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
423 | { | |
ef8745c1 | 424 | unsigned long long excess; |
f64c3f54 BS |
425 | struct mem_cgroup_per_zone *mz; |
426 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
427 | int nid = page_to_nid(page); |
428 | int zid = page_zonenum(page); | |
f64c3f54 BS |
429 | mctz = soft_limit_tree_from_page(page); |
430 | ||
431 | /* | |
4e649152 KH |
432 | * Necessary to update all ancestors when hierarchy is used. |
433 | * because their event counter is not touched. | |
f64c3f54 | 434 | */ |
4e649152 KH |
435 | for (; mem; mem = parent_mem_cgroup(mem)) { |
436 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 437 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
438 | /* |
439 | * We have to update the tree if mz is on RB-tree or | |
440 | * mem is over its softlimit. | |
441 | */ | |
ef8745c1 | 442 | if (excess || mz->on_tree) { |
4e649152 KH |
443 | spin_lock(&mctz->lock); |
444 | /* if on-tree, remove it */ | |
445 | if (mz->on_tree) | |
446 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
447 | /* | |
ef8745c1 KH |
448 | * Insert again. mz->usage_in_excess will be updated. |
449 | * If excess is 0, no tree ops. | |
4e649152 | 450 | */ |
ef8745c1 | 451 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
452 | spin_unlock(&mctz->lock); |
453 | } | |
f64c3f54 BS |
454 | } |
455 | } | |
456 | ||
457 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
458 | { | |
459 | int node, zone; | |
460 | struct mem_cgroup_per_zone *mz; | |
461 | struct mem_cgroup_tree_per_zone *mctz; | |
462 | ||
463 | for_each_node_state(node, N_POSSIBLE) { | |
464 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
465 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
466 | mctz = soft_limit_tree_node_zone(node, zone); | |
467 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
468 | } | |
469 | } | |
470 | } | |
471 | ||
4e416953 BS |
472 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
473 | { | |
474 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; | |
475 | } | |
476 | ||
477 | static struct mem_cgroup_per_zone * | |
478 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
479 | { | |
480 | struct rb_node *rightmost = NULL; | |
26251eaf | 481 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
482 | |
483 | retry: | |
26251eaf | 484 | mz = NULL; |
4e416953 BS |
485 | rightmost = rb_last(&mctz->rb_root); |
486 | if (!rightmost) | |
487 | goto done; /* Nothing to reclaim from */ | |
488 | ||
489 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
490 | /* | |
491 | * Remove the node now but someone else can add it back, | |
492 | * we will to add it back at the end of reclaim to its correct | |
493 | * position in the tree. | |
494 | */ | |
495 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
496 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
497 | !css_tryget(&mz->mem->css)) | |
498 | goto retry; | |
499 | done: | |
500 | return mz; | |
501 | } | |
502 | ||
503 | static struct mem_cgroup_per_zone * | |
504 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
505 | { | |
506 | struct mem_cgroup_per_zone *mz; | |
507 | ||
508 | spin_lock(&mctz->lock); | |
509 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
510 | spin_unlock(&mctz->lock); | |
511 | return mz; | |
512 | } | |
513 | ||
0c3e73e8 BS |
514 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
515 | bool charge) | |
516 | { | |
517 | int val = (charge) ? 1 : -1; | |
518 | struct mem_cgroup_stat *stat = &mem->stat; | |
519 | struct mem_cgroup_stat_cpu *cpustat; | |
520 | int cpu = get_cpu(); | |
521 | ||
522 | cpustat = &stat->cpustat[cpu]; | |
523 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SWAPOUT, val); | |
524 | put_cpu(); | |
525 | } | |
526 | ||
c05555b5 KH |
527 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
528 | struct page_cgroup *pc, | |
529 | bool charge) | |
d52aa412 | 530 | { |
0c3e73e8 | 531 | int val = (charge) ? 1 : -1; |
d52aa412 | 532 | struct mem_cgroup_stat *stat = &mem->stat; |
addb9efe | 533 | struct mem_cgroup_stat_cpu *cpustat; |
08e552c6 | 534 | int cpu = get_cpu(); |
d52aa412 | 535 | |
08e552c6 | 536 | cpustat = &stat->cpustat[cpu]; |
c05555b5 | 537 | if (PageCgroupCache(pc)) |
addb9efe | 538 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
d52aa412 | 539 | else |
addb9efe | 540 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
55e462b0 BR |
541 | |
542 | if (charge) | |
addb9efe | 543 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 BR |
544 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
545 | else | |
addb9efe | 546 | __mem_cgroup_stat_add_safe(cpustat, |
55e462b0 | 547 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
f64c3f54 | 548 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1); |
08e552c6 | 549 | put_cpu(); |
6d12e2d8 KH |
550 | } |
551 | ||
14067bb3 | 552 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 553 | enum lru_list idx) |
6d12e2d8 KH |
554 | { |
555 | int nid, zid; | |
556 | struct mem_cgroup_per_zone *mz; | |
557 | u64 total = 0; | |
558 | ||
559 | for_each_online_node(nid) | |
560 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
561 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
562 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
563 | } | |
564 | return total; | |
d52aa412 KH |
565 | } |
566 | ||
d5b69e38 | 567 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
568 | { |
569 | return container_of(cgroup_subsys_state(cont, | |
570 | mem_cgroup_subsys_id), struct mem_cgroup, | |
571 | css); | |
572 | } | |
573 | ||
cf475ad2 | 574 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 575 | { |
31a78f23 BS |
576 | /* |
577 | * mm_update_next_owner() may clear mm->owner to NULL | |
578 | * if it races with swapoff, page migration, etc. | |
579 | * So this can be called with p == NULL. | |
580 | */ | |
581 | if (unlikely(!p)) | |
582 | return NULL; | |
583 | ||
78fb7466 PE |
584 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
585 | struct mem_cgroup, css); | |
586 | } | |
587 | ||
54595fe2 KH |
588 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
589 | { | |
590 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
591 | |
592 | if (!mm) | |
593 | return NULL; | |
54595fe2 KH |
594 | /* |
595 | * Because we have no locks, mm->owner's may be being moved to other | |
596 | * cgroup. We use css_tryget() here even if this looks | |
597 | * pessimistic (rather than adding locks here). | |
598 | */ | |
599 | rcu_read_lock(); | |
600 | do { | |
601 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
602 | if (unlikely(!mem)) | |
603 | break; | |
604 | } while (!css_tryget(&mem->css)); | |
605 | rcu_read_unlock(); | |
606 | return mem; | |
607 | } | |
608 | ||
14067bb3 KH |
609 | /* |
610 | * Call callback function against all cgroup under hierarchy tree. | |
611 | */ | |
612 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
613 | int (*func)(struct mem_cgroup *, void *)) | |
614 | { | |
615 | int found, ret, nextid; | |
616 | struct cgroup_subsys_state *css; | |
617 | struct mem_cgroup *mem; | |
618 | ||
619 | if (!root->use_hierarchy) | |
620 | return (*func)(root, data); | |
621 | ||
622 | nextid = 1; | |
623 | do { | |
624 | ret = 0; | |
625 | mem = NULL; | |
626 | ||
627 | rcu_read_lock(); | |
628 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
629 | &found); | |
630 | if (css && css_tryget(css)) | |
631 | mem = container_of(css, struct mem_cgroup, css); | |
632 | rcu_read_unlock(); | |
633 | ||
634 | if (mem) { | |
635 | ret = (*func)(mem, data); | |
636 | css_put(&mem->css); | |
637 | } | |
638 | nextid = found + 1; | |
639 | } while (!ret && css); | |
640 | ||
641 | return ret; | |
642 | } | |
643 | ||
4b3bde4c BS |
644 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
645 | { | |
646 | return (mem == root_mem_cgroup); | |
647 | } | |
648 | ||
08e552c6 KH |
649 | /* |
650 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
651 | * Operations are called by routine of global LRU independently from memcg. | |
652 | * What we have to take care of here is validness of pc->mem_cgroup. | |
653 | * | |
654 | * Changes to pc->mem_cgroup happens when | |
655 | * 1. charge | |
656 | * 2. moving account | |
657 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
658 | * It is added to LRU before charge. | |
659 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
660 | * When moving account, the page is not on LRU. It's isolated. | |
661 | */ | |
4f98a2fe | 662 | |
08e552c6 KH |
663 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
664 | { | |
665 | struct page_cgroup *pc; | |
08e552c6 | 666 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 667 | |
f8d66542 | 668 | if (mem_cgroup_disabled()) |
08e552c6 KH |
669 | return; |
670 | pc = lookup_page_cgroup(page); | |
671 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 672 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 673 | return; |
4b3bde4c | 674 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
675 | /* |
676 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
677 | * removed from global LRU. | |
678 | */ | |
08e552c6 | 679 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 680 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
4b3bde4c BS |
681 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
682 | return; | |
683 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 KH |
684 | list_del_init(&pc->lru); |
685 | return; | |
6d12e2d8 KH |
686 | } |
687 | ||
08e552c6 | 688 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 689 | { |
08e552c6 KH |
690 | mem_cgroup_del_lru_list(page, page_lru(page)); |
691 | } | |
b69408e8 | 692 | |
08e552c6 KH |
693 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
694 | { | |
695 | struct mem_cgroup_per_zone *mz; | |
696 | struct page_cgroup *pc; | |
b69408e8 | 697 | |
f8d66542 | 698 | if (mem_cgroup_disabled()) |
08e552c6 | 699 | return; |
6d12e2d8 | 700 | |
08e552c6 | 701 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
702 | /* |
703 | * Used bit is set without atomic ops but after smp_wmb(). | |
704 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
705 | */ | |
08e552c6 | 706 | smp_rmb(); |
4b3bde4c BS |
707 | /* unused or root page is not rotated. */ |
708 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 KH |
709 | return; |
710 | mz = page_cgroup_zoneinfo(pc); | |
711 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
712 | } |
713 | ||
08e552c6 | 714 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 715 | { |
08e552c6 KH |
716 | struct page_cgroup *pc; |
717 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 718 | |
f8d66542 | 719 | if (mem_cgroup_disabled()) |
08e552c6 KH |
720 | return; |
721 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 722 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
bd112db8 DN |
723 | /* |
724 | * Used bit is set without atomic ops but after smp_wmb(). | |
725 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
726 | */ | |
08e552c6 KH |
727 | smp_rmb(); |
728 | if (!PageCgroupUsed(pc)) | |
894bc310 | 729 | return; |
b69408e8 | 730 | |
08e552c6 | 731 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 732 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
4b3bde4c BS |
733 | SetPageCgroupAcctLRU(pc); |
734 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
735 | return; | |
08e552c6 KH |
736 | list_add(&pc->lru, &mz->lists[lru]); |
737 | } | |
544122e5 | 738 | |
08e552c6 | 739 | /* |
544122e5 KH |
740 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
741 | * lru because the page may.be reused after it's fully uncharged (because of | |
742 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
743 | * it again. This function is only used to charge SwapCache. It's done under | |
744 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 745 | */ |
544122e5 | 746 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 747 | { |
544122e5 KH |
748 | unsigned long flags; |
749 | struct zone *zone = page_zone(page); | |
750 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
751 | ||
752 | spin_lock_irqsave(&zone->lru_lock, flags); | |
753 | /* | |
754 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
755 | * is guarded by lock_page() because the page is SwapCache. | |
756 | */ | |
757 | if (!PageCgroupUsed(pc)) | |
758 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
759 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
760 | } |
761 | ||
544122e5 KH |
762 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
763 | { | |
764 | unsigned long flags; | |
765 | struct zone *zone = page_zone(page); | |
766 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
767 | ||
768 | spin_lock_irqsave(&zone->lru_lock, flags); | |
769 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 770 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
771 | mem_cgroup_add_lru_list(page, page_lru(page)); |
772 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
773 | } | |
774 | ||
775 | ||
08e552c6 KH |
776 | void mem_cgroup_move_lists(struct page *page, |
777 | enum lru_list from, enum lru_list to) | |
778 | { | |
f8d66542 | 779 | if (mem_cgroup_disabled()) |
08e552c6 KH |
780 | return; |
781 | mem_cgroup_del_lru_list(page, from); | |
782 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
783 | } |
784 | ||
4c4a2214 DR |
785 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
786 | { | |
787 | int ret; | |
0b7f569e | 788 | struct mem_cgroup *curr = NULL; |
4c4a2214 DR |
789 | |
790 | task_lock(task); | |
0b7f569e KH |
791 | rcu_read_lock(); |
792 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
793 | rcu_read_unlock(); | |
4c4a2214 | 794 | task_unlock(task); |
0b7f569e KH |
795 | if (!curr) |
796 | return 0; | |
d31f56db DN |
797 | /* |
798 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
799 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
800 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
801 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
802 | */ | |
803 | if (mem->use_hierarchy) | |
0b7f569e KH |
804 | ret = css_is_ancestor(&curr->css, &mem->css); |
805 | else | |
806 | ret = (curr == mem); | |
807 | css_put(&curr->css); | |
4c4a2214 DR |
808 | return ret; |
809 | } | |
810 | ||
6c48a1d0 KH |
811 | /* |
812 | * prev_priority control...this will be used in memory reclaim path. | |
813 | */ | |
814 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
815 | { | |
2733c06a KM |
816 | int prev_priority; |
817 | ||
818 | spin_lock(&mem->reclaim_param_lock); | |
819 | prev_priority = mem->prev_priority; | |
820 | spin_unlock(&mem->reclaim_param_lock); | |
821 | ||
822 | return prev_priority; | |
6c48a1d0 KH |
823 | } |
824 | ||
825 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
826 | { | |
2733c06a | 827 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
828 | if (priority < mem->prev_priority) |
829 | mem->prev_priority = priority; | |
2733c06a | 830 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
831 | } |
832 | ||
833 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
834 | { | |
2733c06a | 835 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 836 | mem->prev_priority = priority; |
2733c06a | 837 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
838 | } |
839 | ||
c772be93 | 840 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
841 | { |
842 | unsigned long active; | |
843 | unsigned long inactive; | |
c772be93 KM |
844 | unsigned long gb; |
845 | unsigned long inactive_ratio; | |
14797e23 | 846 | |
14067bb3 KH |
847 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
848 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 849 | |
c772be93 KM |
850 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
851 | if (gb) | |
852 | inactive_ratio = int_sqrt(10 * gb); | |
853 | else | |
854 | inactive_ratio = 1; | |
855 | ||
856 | if (present_pages) { | |
857 | present_pages[0] = inactive; | |
858 | present_pages[1] = active; | |
859 | } | |
860 | ||
861 | return inactive_ratio; | |
862 | } | |
863 | ||
864 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
865 | { | |
866 | unsigned long active; | |
867 | unsigned long inactive; | |
868 | unsigned long present_pages[2]; | |
869 | unsigned long inactive_ratio; | |
870 | ||
871 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
872 | ||
873 | inactive = present_pages[0]; | |
874 | active = present_pages[1]; | |
875 | ||
876 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
877 | return 1; |
878 | ||
879 | return 0; | |
880 | } | |
881 | ||
56e49d21 RR |
882 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
883 | { | |
884 | unsigned long active; | |
885 | unsigned long inactive; | |
886 | ||
887 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
888 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
889 | ||
890 | return (active > inactive); | |
891 | } | |
892 | ||
a3d8e054 KM |
893 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
894 | struct zone *zone, | |
895 | enum lru_list lru) | |
896 | { | |
897 | int nid = zone->zone_pgdat->node_id; | |
898 | int zid = zone_idx(zone); | |
899 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
900 | ||
901 | return MEM_CGROUP_ZSTAT(mz, lru); | |
902 | } | |
903 | ||
3e2f41f1 KM |
904 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
905 | struct zone *zone) | |
906 | { | |
907 | int nid = zone->zone_pgdat->node_id; | |
908 | int zid = zone_idx(zone); | |
909 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
910 | ||
911 | return &mz->reclaim_stat; | |
912 | } | |
913 | ||
914 | struct zone_reclaim_stat * | |
915 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
916 | { | |
917 | struct page_cgroup *pc; | |
918 | struct mem_cgroup_per_zone *mz; | |
919 | ||
920 | if (mem_cgroup_disabled()) | |
921 | return NULL; | |
922 | ||
923 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
924 | /* |
925 | * Used bit is set without atomic ops but after smp_wmb(). | |
926 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
927 | */ | |
928 | smp_rmb(); | |
929 | if (!PageCgroupUsed(pc)) | |
930 | return NULL; | |
931 | ||
3e2f41f1 KM |
932 | mz = page_cgroup_zoneinfo(pc); |
933 | if (!mz) | |
934 | return NULL; | |
935 | ||
936 | return &mz->reclaim_stat; | |
937 | } | |
938 | ||
66e1707b BS |
939 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
940 | struct list_head *dst, | |
941 | unsigned long *scanned, int order, | |
942 | int mode, struct zone *z, | |
943 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 944 | int active, int file) |
66e1707b BS |
945 | { |
946 | unsigned long nr_taken = 0; | |
947 | struct page *page; | |
948 | unsigned long scan; | |
949 | LIST_HEAD(pc_list); | |
950 | struct list_head *src; | |
ff7283fa | 951 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
952 | int nid = z->zone_pgdat->node_id; |
953 | int zid = zone_idx(z); | |
954 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 955 | int lru = LRU_FILE * file + active; |
2ffebca6 | 956 | int ret; |
66e1707b | 957 | |
cf475ad2 | 958 | BUG_ON(!mem_cont); |
1ecaab2b | 959 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 960 | src = &mz->lists[lru]; |
66e1707b | 961 | |
ff7283fa KH |
962 | scan = 0; |
963 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 964 | if (scan >= nr_to_scan) |
ff7283fa | 965 | break; |
08e552c6 KH |
966 | |
967 | page = pc->page; | |
52d4b9ac KH |
968 | if (unlikely(!PageCgroupUsed(pc))) |
969 | continue; | |
436c6541 | 970 | if (unlikely(!PageLRU(page))) |
ff7283fa | 971 | continue; |
ff7283fa | 972 | |
436c6541 | 973 | scan++; |
2ffebca6 KH |
974 | ret = __isolate_lru_page(page, mode, file); |
975 | switch (ret) { | |
976 | case 0: | |
66e1707b | 977 | list_move(&page->lru, dst); |
2ffebca6 | 978 | mem_cgroup_del_lru(page); |
66e1707b | 979 | nr_taken++; |
2ffebca6 KH |
980 | break; |
981 | case -EBUSY: | |
982 | /* we don't affect global LRU but rotate in our LRU */ | |
983 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
984 | break; | |
985 | default: | |
986 | break; | |
66e1707b BS |
987 | } |
988 | } | |
989 | ||
66e1707b BS |
990 | *scanned = scan; |
991 | return nr_taken; | |
992 | } | |
993 | ||
6d61ef40 BS |
994 | #define mem_cgroup_from_res_counter(counter, member) \ |
995 | container_of(counter, struct mem_cgroup, member) | |
996 | ||
b85a96c0 DN |
997 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
998 | { | |
999 | if (do_swap_account) { | |
1000 | if (res_counter_check_under_limit(&mem->res) && | |
1001 | res_counter_check_under_limit(&mem->memsw)) | |
1002 | return true; | |
1003 | } else | |
1004 | if (res_counter_check_under_limit(&mem->res)) | |
1005 | return true; | |
1006 | return false; | |
1007 | } | |
1008 | ||
a7885eb8 KM |
1009 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1010 | { | |
1011 | struct cgroup *cgrp = memcg->css.cgroup; | |
1012 | unsigned int swappiness; | |
1013 | ||
1014 | /* root ? */ | |
1015 | if (cgrp->parent == NULL) | |
1016 | return vm_swappiness; | |
1017 | ||
1018 | spin_lock(&memcg->reclaim_param_lock); | |
1019 | swappiness = memcg->swappiness; | |
1020 | spin_unlock(&memcg->reclaim_param_lock); | |
1021 | ||
1022 | return swappiness; | |
1023 | } | |
1024 | ||
81d39c20 KH |
1025 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
1026 | { | |
1027 | int *val = data; | |
1028 | (*val)++; | |
1029 | return 0; | |
1030 | } | |
e222432b BS |
1031 | |
1032 | /** | |
1033 | * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode. | |
1034 | * @memcg: The memory cgroup that went over limit | |
1035 | * @p: Task that is going to be killed | |
1036 | * | |
1037 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1038 | * enabled | |
1039 | */ | |
1040 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1041 | { | |
1042 | struct cgroup *task_cgrp; | |
1043 | struct cgroup *mem_cgrp; | |
1044 | /* | |
1045 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1046 | * on the assumption that OOM is serialized for memory controller. | |
1047 | * If this assumption is broken, revisit this code. | |
1048 | */ | |
1049 | static char memcg_name[PATH_MAX]; | |
1050 | int ret; | |
1051 | ||
d31f56db | 1052 | if (!memcg || !p) |
e222432b BS |
1053 | return; |
1054 | ||
1055 | ||
1056 | rcu_read_lock(); | |
1057 | ||
1058 | mem_cgrp = memcg->css.cgroup; | |
1059 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1060 | ||
1061 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1062 | if (ret < 0) { | |
1063 | /* | |
1064 | * Unfortunately, we are unable to convert to a useful name | |
1065 | * But we'll still print out the usage information | |
1066 | */ | |
1067 | rcu_read_unlock(); | |
1068 | goto done; | |
1069 | } | |
1070 | rcu_read_unlock(); | |
1071 | ||
1072 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1073 | ||
1074 | rcu_read_lock(); | |
1075 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1076 | if (ret < 0) { | |
1077 | rcu_read_unlock(); | |
1078 | goto done; | |
1079 | } | |
1080 | rcu_read_unlock(); | |
1081 | ||
1082 | /* | |
1083 | * Continues from above, so we don't need an KERN_ level | |
1084 | */ | |
1085 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1086 | done: | |
1087 | ||
1088 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1089 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1090 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1091 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1092 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1093 | "failcnt %llu\n", | |
1094 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1095 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1096 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1097 | } | |
1098 | ||
81d39c20 KH |
1099 | /* |
1100 | * This function returns the number of memcg under hierarchy tree. Returns | |
1101 | * 1(self count) if no children. | |
1102 | */ | |
1103 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1104 | { | |
1105 | int num = 0; | |
1106 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
1107 | return num; | |
1108 | } | |
1109 | ||
6d61ef40 | 1110 | /* |
04046e1a KH |
1111 | * Visit the first child (need not be the first child as per the ordering |
1112 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1113 | * that to reclaim free pages from. | |
1114 | */ | |
1115 | static struct mem_cgroup * | |
1116 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1117 | { | |
1118 | struct mem_cgroup *ret = NULL; | |
1119 | struct cgroup_subsys_state *css; | |
1120 | int nextid, found; | |
1121 | ||
1122 | if (!root_mem->use_hierarchy) { | |
1123 | css_get(&root_mem->css); | |
1124 | ret = root_mem; | |
1125 | } | |
1126 | ||
1127 | while (!ret) { | |
1128 | rcu_read_lock(); | |
1129 | nextid = root_mem->last_scanned_child + 1; | |
1130 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1131 | &found); | |
1132 | if (css && css_tryget(css)) | |
1133 | ret = container_of(css, struct mem_cgroup, css); | |
1134 | ||
1135 | rcu_read_unlock(); | |
1136 | /* Updates scanning parameter */ | |
1137 | spin_lock(&root_mem->reclaim_param_lock); | |
1138 | if (!css) { | |
1139 | /* this means start scan from ID:1 */ | |
1140 | root_mem->last_scanned_child = 0; | |
1141 | } else | |
1142 | root_mem->last_scanned_child = found; | |
1143 | spin_unlock(&root_mem->reclaim_param_lock); | |
1144 | } | |
1145 | ||
1146 | return ret; | |
1147 | } | |
1148 | ||
1149 | /* | |
1150 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1151 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1152 | * based on its position in the children list. | |
6d61ef40 BS |
1153 | * |
1154 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1155 | * |
1156 | * We give up and return to the caller when we visit root_mem twice. | |
1157 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1158 | * |
1159 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1160 | */ |
1161 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1162 | struct zone *zone, |
75822b44 BS |
1163 | gfp_t gfp_mask, |
1164 | unsigned long reclaim_options) | |
6d61ef40 | 1165 | { |
04046e1a KH |
1166 | struct mem_cgroup *victim; |
1167 | int ret, total = 0; | |
1168 | int loop = 0; | |
75822b44 BS |
1169 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1170 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 BS |
1171 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
1172 | unsigned long excess = mem_cgroup_get_excess(root_mem); | |
04046e1a | 1173 | |
22a668d7 KH |
1174 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1175 | if (root_mem->memsw_is_minimum) | |
1176 | noswap = true; | |
1177 | ||
4e416953 | 1178 | while (1) { |
04046e1a | 1179 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1180 | if (victim == root_mem) { |
04046e1a | 1181 | loop++; |
cdec2e42 KH |
1182 | if (loop >= 1) |
1183 | drain_all_stock_async(); | |
4e416953 BS |
1184 | if (loop >= 2) { |
1185 | /* | |
1186 | * If we have not been able to reclaim | |
1187 | * anything, it might because there are | |
1188 | * no reclaimable pages under this hierarchy | |
1189 | */ | |
1190 | if (!check_soft || !total) { | |
1191 | css_put(&victim->css); | |
1192 | break; | |
1193 | } | |
1194 | /* | |
1195 | * We want to do more targetted reclaim. | |
1196 | * excess >> 2 is not to excessive so as to | |
1197 | * reclaim too much, nor too less that we keep | |
1198 | * coming back to reclaim from this cgroup | |
1199 | */ | |
1200 | if (total >= (excess >> 2) || | |
1201 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1202 | css_put(&victim->css); | |
1203 | break; | |
1204 | } | |
1205 | } | |
1206 | } | |
04046e1a KH |
1207 | if (!mem_cgroup_local_usage(&victim->stat)) { |
1208 | /* this cgroup's local usage == 0 */ | |
1209 | css_put(&victim->css); | |
6d61ef40 BS |
1210 | continue; |
1211 | } | |
04046e1a | 1212 | /* we use swappiness of local cgroup */ |
4e416953 BS |
1213 | if (check_soft) |
1214 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
1215 | noswap, get_swappiness(victim), zone, | |
1216 | zone->zone_pgdat->node_id); | |
1217 | else | |
1218 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1219 | noswap, get_swappiness(victim)); | |
04046e1a | 1220 | css_put(&victim->css); |
81d39c20 KH |
1221 | /* |
1222 | * At shrinking usage, we can't check we should stop here or | |
1223 | * reclaim more. It's depends on callers. last_scanned_child | |
1224 | * will work enough for keeping fairness under tree. | |
1225 | */ | |
1226 | if (shrink) | |
1227 | return ret; | |
04046e1a | 1228 | total += ret; |
4e416953 BS |
1229 | if (check_soft) { |
1230 | if (res_counter_check_under_soft_limit(&root_mem->res)) | |
1231 | return total; | |
1232 | } else if (mem_cgroup_check_under_limit(root_mem)) | |
04046e1a | 1233 | return 1 + total; |
6d61ef40 | 1234 | } |
04046e1a | 1235 | return total; |
6d61ef40 BS |
1236 | } |
1237 | ||
a636b327 KH |
1238 | bool mem_cgroup_oom_called(struct task_struct *task) |
1239 | { | |
1240 | bool ret = false; | |
1241 | struct mem_cgroup *mem; | |
1242 | struct mm_struct *mm; | |
1243 | ||
1244 | rcu_read_lock(); | |
1245 | mm = task->mm; | |
1246 | if (!mm) | |
1247 | mm = &init_mm; | |
1248 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1249 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
1250 | ret = true; | |
1251 | rcu_read_unlock(); | |
1252 | return ret; | |
1253 | } | |
0b7f569e KH |
1254 | |
1255 | static int record_last_oom_cb(struct mem_cgroup *mem, void *data) | |
1256 | { | |
1257 | mem->last_oom_jiffies = jiffies; | |
1258 | return 0; | |
1259 | } | |
1260 | ||
1261 | static void record_last_oom(struct mem_cgroup *mem) | |
1262 | { | |
1263 | mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb); | |
1264 | } | |
1265 | ||
d69b042f BS |
1266 | /* |
1267 | * Currently used to update mapped file statistics, but the routine can be | |
1268 | * generalized to update other statistics as well. | |
1269 | */ | |
d8046582 | 1270 | void mem_cgroup_update_file_mapped(struct page *page, int val) |
d69b042f BS |
1271 | { |
1272 | struct mem_cgroup *mem; | |
1273 | struct mem_cgroup_stat *stat; | |
1274 | struct mem_cgroup_stat_cpu *cpustat; | |
1275 | int cpu; | |
1276 | struct page_cgroup *pc; | |
1277 | ||
d69b042f BS |
1278 | pc = lookup_page_cgroup(page); |
1279 | if (unlikely(!pc)) | |
1280 | return; | |
1281 | ||
1282 | lock_page_cgroup(pc); | |
1283 | mem = pc->mem_cgroup; | |
1284 | if (!mem) | |
1285 | goto done; | |
1286 | ||
1287 | if (!PageCgroupUsed(pc)) | |
1288 | goto done; | |
1289 | ||
1290 | /* | |
1291 | * Preemption is already disabled, we don't need get_cpu() | |
1292 | */ | |
1293 | cpu = smp_processor_id(); | |
1294 | stat = &mem->stat; | |
1295 | cpustat = &stat->cpustat[cpu]; | |
1296 | ||
d8046582 | 1297 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, val); |
d69b042f BS |
1298 | done: |
1299 | unlock_page_cgroup(pc); | |
1300 | } | |
0b7f569e | 1301 | |
cdec2e42 KH |
1302 | /* |
1303 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1304 | * TODO: maybe necessary to use big numbers in big irons. | |
1305 | */ | |
1306 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1307 | struct memcg_stock_pcp { | |
1308 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1309 | int charge; | |
1310 | struct work_struct work; | |
1311 | }; | |
1312 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1313 | static atomic_t memcg_drain_count; | |
1314 | ||
1315 | /* | |
1316 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1317 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1318 | * cgroup which is not current target, returns false. This stock will be | |
1319 | * refilled. | |
1320 | */ | |
1321 | static bool consume_stock(struct mem_cgroup *mem) | |
1322 | { | |
1323 | struct memcg_stock_pcp *stock; | |
1324 | bool ret = true; | |
1325 | ||
1326 | stock = &get_cpu_var(memcg_stock); | |
1327 | if (mem == stock->cached && stock->charge) | |
1328 | stock->charge -= PAGE_SIZE; | |
1329 | else /* need to call res_counter_charge */ | |
1330 | ret = false; | |
1331 | put_cpu_var(memcg_stock); | |
1332 | return ret; | |
1333 | } | |
1334 | ||
1335 | /* | |
1336 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1337 | */ | |
1338 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1339 | { | |
1340 | struct mem_cgroup *old = stock->cached; | |
1341 | ||
1342 | if (stock->charge) { | |
1343 | res_counter_uncharge(&old->res, stock->charge); | |
1344 | if (do_swap_account) | |
1345 | res_counter_uncharge(&old->memsw, stock->charge); | |
1346 | } | |
1347 | stock->cached = NULL; | |
1348 | stock->charge = 0; | |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * This must be called under preempt disabled or must be called by | |
1353 | * a thread which is pinned to local cpu. | |
1354 | */ | |
1355 | static void drain_local_stock(struct work_struct *dummy) | |
1356 | { | |
1357 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1358 | drain_stock(stock); | |
1359 | } | |
1360 | ||
1361 | /* | |
1362 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
1363 | * This will be consumed by consumt_stock() function, later. | |
1364 | */ | |
1365 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1366 | { | |
1367 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1368 | ||
1369 | if (stock->cached != mem) { /* reset if necessary */ | |
1370 | drain_stock(stock); | |
1371 | stock->cached = mem; | |
1372 | } | |
1373 | stock->charge += val; | |
1374 | put_cpu_var(memcg_stock); | |
1375 | } | |
1376 | ||
1377 | /* | |
1378 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1379 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1380 | * expects some charges will be back to res_counter later but cannot wait for | |
1381 | * it. | |
1382 | */ | |
1383 | static void drain_all_stock_async(void) | |
1384 | { | |
1385 | int cpu; | |
1386 | /* This function is for scheduling "drain" in asynchronous way. | |
1387 | * The result of "drain" is not directly handled by callers. Then, | |
1388 | * if someone is calling drain, we don't have to call drain more. | |
1389 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1390 | * there is a race. We just do loose check here. | |
1391 | */ | |
1392 | if (atomic_read(&memcg_drain_count)) | |
1393 | return; | |
1394 | /* Notify other cpus that system-wide "drain" is running */ | |
1395 | atomic_inc(&memcg_drain_count); | |
1396 | get_online_cpus(); | |
1397 | for_each_online_cpu(cpu) { | |
1398 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1399 | schedule_work_on(cpu, &stock->work); | |
1400 | } | |
1401 | put_online_cpus(); | |
1402 | atomic_dec(&memcg_drain_count); | |
1403 | /* We don't wait for flush_work */ | |
1404 | } | |
1405 | ||
1406 | /* This is a synchronous drain interface. */ | |
1407 | static void drain_all_stock_sync(void) | |
1408 | { | |
1409 | /* called when force_empty is called */ | |
1410 | atomic_inc(&memcg_drain_count); | |
1411 | schedule_on_each_cpu(drain_local_stock); | |
1412 | atomic_dec(&memcg_drain_count); | |
1413 | } | |
1414 | ||
1415 | static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb, | |
1416 | unsigned long action, | |
1417 | void *hcpu) | |
1418 | { | |
1419 | int cpu = (unsigned long)hcpu; | |
1420 | struct memcg_stock_pcp *stock; | |
1421 | ||
1422 | if (action != CPU_DEAD) | |
1423 | return NOTIFY_OK; | |
1424 | stock = &per_cpu(memcg_stock, cpu); | |
1425 | drain_stock(stock); | |
1426 | return NOTIFY_OK; | |
1427 | } | |
1428 | ||
f817ed48 KH |
1429 | /* |
1430 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1431 | * oom-killer can be invoked. | |
8a9f3ccd | 1432 | */ |
f817ed48 | 1433 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
8c7c6e34 | 1434 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
f64c3f54 | 1435 | bool oom, struct page *page) |
8a9f3ccd | 1436 | { |
4e649152 | 1437 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 1438 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
4e649152 | 1439 | struct res_counter *fail_res; |
cdec2e42 | 1440 | int csize = CHARGE_SIZE; |
a636b327 KH |
1441 | |
1442 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
1443 | /* Don't account this! */ | |
1444 | *memcg = NULL; | |
1445 | return 0; | |
1446 | } | |
1447 | ||
8a9f3ccd | 1448 | /* |
3be91277 HD |
1449 | * We always charge the cgroup the mm_struct belongs to. |
1450 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
1451 | * thread group leader migrates. It's possible that mm is not |
1452 | * set, if so charge the init_mm (happens for pagecache usage). | |
1453 | */ | |
54595fe2 KH |
1454 | mem = *memcg; |
1455 | if (likely(!mem)) { | |
1456 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 1457 | *memcg = mem; |
e8589cc1 | 1458 | } else { |
7a81b88c | 1459 | css_get(&mem->css); |
e8589cc1 | 1460 | } |
54595fe2 KH |
1461 | if (unlikely(!mem)) |
1462 | return 0; | |
1463 | ||
46f7e602 | 1464 | VM_BUG_ON(css_is_removed(&mem->css)); |
cdec2e42 KH |
1465 | if (mem_cgroup_is_root(mem)) |
1466 | goto done; | |
8a9f3ccd | 1467 | |
8c7c6e34 | 1468 | while (1) { |
0c3e73e8 | 1469 | int ret = 0; |
75822b44 | 1470 | unsigned long flags = 0; |
7a81b88c | 1471 | |
cdec2e42 KH |
1472 | if (consume_stock(mem)) |
1473 | goto charged; | |
1474 | ||
1475 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
8c7c6e34 KH |
1476 | if (likely(!ret)) { |
1477 | if (!do_swap_account) | |
1478 | break; | |
cdec2e42 | 1479 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); |
8c7c6e34 KH |
1480 | if (likely(!ret)) |
1481 | break; | |
1482 | /* mem+swap counter fails */ | |
cdec2e42 | 1483 | res_counter_uncharge(&mem->res, csize); |
75822b44 | 1484 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
6d61ef40 BS |
1485 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1486 | memsw); | |
1487 | } else | |
1488 | /* mem counter fails */ | |
1489 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1490 | res); | |
1491 | ||
cdec2e42 KH |
1492 | /* reduce request size and retry */ |
1493 | if (csize > PAGE_SIZE) { | |
1494 | csize = PAGE_SIZE; | |
1495 | continue; | |
1496 | } | |
3be91277 | 1497 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 1498 | goto nomem; |
e1a1cd59 | 1499 | |
4e416953 BS |
1500 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
1501 | gfp_mask, flags); | |
4d1c6273 DN |
1502 | if (ret) |
1503 | continue; | |
66e1707b BS |
1504 | |
1505 | /* | |
8869b8f6 HD |
1506 | * try_to_free_mem_cgroup_pages() might not give us a full |
1507 | * picture of reclaim. Some pages are reclaimed and might be | |
1508 | * moved to swap cache or just unmapped from the cgroup. | |
1509 | * Check the limit again to see if the reclaim reduced the | |
1510 | * current usage of the cgroup before giving up | |
8c7c6e34 | 1511 | * |
8869b8f6 | 1512 | */ |
b85a96c0 DN |
1513 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
1514 | continue; | |
3be91277 | 1515 | |
8033b97c DN |
1516 | /* try to avoid oom while someone is moving charge */ |
1517 | if (mc.moving_task && current != mc.moving_task) { | |
1518 | struct mem_cgroup *from, *to; | |
1519 | bool do_continue = false; | |
1520 | /* | |
1521 | * There is a small race that "from" or "to" can be | |
1522 | * freed by rmdir, so we use css_tryget(). | |
1523 | */ | |
1524 | rcu_read_lock(); | |
1525 | from = mc.from; | |
1526 | to = mc.to; | |
1527 | if (from && css_tryget(&from->css)) { | |
1528 | if (mem_over_limit->use_hierarchy) | |
1529 | do_continue = css_is_ancestor( | |
1530 | &from->css, | |
1531 | &mem_over_limit->css); | |
1532 | else | |
1533 | do_continue = (from == mem_over_limit); | |
1534 | css_put(&from->css); | |
1535 | } | |
1536 | if (!do_continue && to && css_tryget(&to->css)) { | |
1537 | if (mem_over_limit->use_hierarchy) | |
1538 | do_continue = css_is_ancestor( | |
1539 | &to->css, | |
1540 | &mem_over_limit->css); | |
1541 | else | |
1542 | do_continue = (to == mem_over_limit); | |
1543 | css_put(&to->css); | |
1544 | } | |
1545 | rcu_read_unlock(); | |
1546 | if (do_continue) { | |
1547 | DEFINE_WAIT(wait); | |
1548 | prepare_to_wait(&mc.waitq, &wait, | |
1549 | TASK_INTERRUPTIBLE); | |
1550 | /* moving charge context might have finished. */ | |
1551 | if (mc.moving_task) | |
1552 | schedule(); | |
1553 | finish_wait(&mc.waitq, &wait); | |
1554 | continue; | |
1555 | } | |
1556 | } | |
1557 | ||
3be91277 | 1558 | if (!nr_retries--) { |
a636b327 | 1559 | if (oom) { |
88700756 | 1560 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
0b7f569e | 1561 | record_last_oom(mem_over_limit); |
a636b327 | 1562 | } |
7a81b88c | 1563 | goto nomem; |
66e1707b | 1564 | } |
8a9f3ccd | 1565 | } |
cdec2e42 KH |
1566 | if (csize > PAGE_SIZE) |
1567 | refill_stock(mem, csize - PAGE_SIZE); | |
1568 | charged: | |
f64c3f54 | 1569 | /* |
4e649152 KH |
1570 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. |
1571 | * if they exceeds softlimit. | |
f64c3f54 | 1572 | */ |
4ffef5fe | 1573 | if (page && mem_cgroup_soft_limit_check(mem)) |
4e649152 | 1574 | mem_cgroup_update_tree(mem, page); |
0c3e73e8 | 1575 | done: |
7a81b88c KH |
1576 | return 0; |
1577 | nomem: | |
1578 | css_put(&mem->css); | |
1579 | return -ENOMEM; | |
1580 | } | |
8a9f3ccd | 1581 | |
a3032a2c DN |
1582 | /* |
1583 | * Somemtimes we have to undo a charge we got by try_charge(). | |
1584 | * This function is for that and do uncharge, put css's refcnt. | |
1585 | * gotten by try_charge(). | |
1586 | */ | |
854ffa8d DN |
1587 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
1588 | unsigned long count) | |
a3032a2c DN |
1589 | { |
1590 | if (!mem_cgroup_is_root(mem)) { | |
854ffa8d | 1591 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
a3032a2c | 1592 | if (do_swap_account) |
854ffa8d DN |
1593 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
1594 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
1595 | WARN_ON_ONCE(count > INT_MAX); | |
1596 | __css_put(&mem->css, (int)count); | |
a3032a2c | 1597 | } |
854ffa8d DN |
1598 | /* we don't need css_put for root */ |
1599 | } | |
1600 | ||
1601 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem) | |
1602 | { | |
1603 | __mem_cgroup_cancel_charge(mem, 1); | |
a3032a2c DN |
1604 | } |
1605 | ||
a3b2d692 KH |
1606 | /* |
1607 | * A helper function to get mem_cgroup from ID. must be called under | |
1608 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
1609 | * it's concern. (dropping refcnt from swap can be called against removed | |
1610 | * memcg.) | |
1611 | */ | |
1612 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
1613 | { | |
1614 | struct cgroup_subsys_state *css; | |
1615 | ||
1616 | /* ID 0 is unused ID */ | |
1617 | if (!id) | |
1618 | return NULL; | |
1619 | css = css_lookup(&mem_cgroup_subsys, id); | |
1620 | if (!css) | |
1621 | return NULL; | |
1622 | return container_of(css, struct mem_cgroup, css); | |
1623 | } | |
1624 | ||
e42d9d5d | 1625 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 1626 | { |
e42d9d5d | 1627 | struct mem_cgroup *mem = NULL; |
3c776e64 | 1628 | struct page_cgroup *pc; |
a3b2d692 | 1629 | unsigned short id; |
b5a84319 KH |
1630 | swp_entry_t ent; |
1631 | ||
3c776e64 DN |
1632 | VM_BUG_ON(!PageLocked(page)); |
1633 | ||
3c776e64 | 1634 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 1635 | lock_page_cgroup(pc); |
a3b2d692 | 1636 | if (PageCgroupUsed(pc)) { |
3c776e64 | 1637 | mem = pc->mem_cgroup; |
a3b2d692 KH |
1638 | if (mem && !css_tryget(&mem->css)) |
1639 | mem = NULL; | |
e42d9d5d | 1640 | } else if (PageSwapCache(page)) { |
3c776e64 | 1641 | ent.val = page_private(page); |
a3b2d692 KH |
1642 | id = lookup_swap_cgroup(ent); |
1643 | rcu_read_lock(); | |
1644 | mem = mem_cgroup_lookup(id); | |
1645 | if (mem && !css_tryget(&mem->css)) | |
1646 | mem = NULL; | |
1647 | rcu_read_unlock(); | |
3c776e64 | 1648 | } |
c0bd3f63 | 1649 | unlock_page_cgroup(pc); |
b5a84319 KH |
1650 | return mem; |
1651 | } | |
1652 | ||
7a81b88c | 1653 | /* |
a5e924f5 | 1654 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
1655 | * USED state. If already USED, uncharge and return. |
1656 | */ | |
1657 | ||
1658 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
1659 | struct page_cgroup *pc, | |
1660 | enum charge_type ctype) | |
1661 | { | |
7a81b88c KH |
1662 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
1663 | if (!mem) | |
1664 | return; | |
52d4b9ac KH |
1665 | |
1666 | lock_page_cgroup(pc); | |
1667 | if (unlikely(PageCgroupUsed(pc))) { | |
1668 | unlock_page_cgroup(pc); | |
a3032a2c | 1669 | mem_cgroup_cancel_charge(mem); |
7a81b88c | 1670 | return; |
52d4b9ac | 1671 | } |
4b3bde4c | 1672 | |
8a9f3ccd | 1673 | pc->mem_cgroup = mem; |
261fb61a KH |
1674 | /* |
1675 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
1676 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
1677 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
1678 | * before USED bit, we need memory barrier here. | |
1679 | * See mem_cgroup_add_lru_list(), etc. | |
1680 | */ | |
08e552c6 | 1681 | smp_wmb(); |
4b3bde4c BS |
1682 | switch (ctype) { |
1683 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
1684 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
1685 | SetPageCgroupCache(pc); | |
1686 | SetPageCgroupUsed(pc); | |
1687 | break; | |
1688 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1689 | ClearPageCgroupCache(pc); | |
1690 | SetPageCgroupUsed(pc); | |
1691 | break; | |
1692 | default: | |
1693 | break; | |
1694 | } | |
3be91277 | 1695 | |
08e552c6 | 1696 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 1697 | |
52d4b9ac | 1698 | unlock_page_cgroup(pc); |
7a81b88c | 1699 | } |
66e1707b | 1700 | |
f817ed48 | 1701 | /** |
57f9fd7d | 1702 | * __mem_cgroup_move_account - move account of the page |
f817ed48 KH |
1703 | * @pc: page_cgroup of the page. |
1704 | * @from: mem_cgroup which the page is moved from. | |
1705 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 1706 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
1707 | * |
1708 | * The caller must confirm following. | |
08e552c6 | 1709 | * - page is not on LRU (isolate_page() is useful.) |
57f9fd7d | 1710 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
f817ed48 | 1711 | * |
854ffa8d DN |
1712 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
1713 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is | |
1714 | * true, this function does "uncharge" from old cgroup, but it doesn't if | |
1715 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 KH |
1716 | */ |
1717 | ||
57f9fd7d | 1718 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
854ffa8d | 1719 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
f817ed48 | 1720 | { |
d69b042f BS |
1721 | struct page *page; |
1722 | int cpu; | |
1723 | struct mem_cgroup_stat *stat; | |
1724 | struct mem_cgroup_stat_cpu *cpustat; | |
f817ed48 | 1725 | |
f817ed48 | 1726 | VM_BUG_ON(from == to); |
08e552c6 | 1727 | VM_BUG_ON(PageLRU(pc->page)); |
57f9fd7d DN |
1728 | VM_BUG_ON(!PageCgroupLocked(pc)); |
1729 | VM_BUG_ON(!PageCgroupUsed(pc)); | |
1730 | VM_BUG_ON(pc->mem_cgroup != from); | |
f817ed48 | 1731 | |
d69b042f | 1732 | page = pc->page; |
d8046582 | 1733 | if (page_mapped(page) && !PageAnon(page)) { |
d69b042f BS |
1734 | cpu = smp_processor_id(); |
1735 | /* Update mapped_file data for mem_cgroup "from" */ | |
1736 | stat = &from->stat; | |
1737 | cpustat = &stat->cpustat[cpu]; | |
d8046582 | 1738 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, |
d69b042f BS |
1739 | -1); |
1740 | ||
1741 | /* Update mapped_file data for mem_cgroup "to" */ | |
1742 | stat = &to->stat; | |
1743 | cpustat = &stat->cpustat[cpu]; | |
d8046582 | 1744 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_FILE_MAPPED, |
d69b042f BS |
1745 | 1); |
1746 | } | |
854ffa8d DN |
1747 | mem_cgroup_charge_statistics(from, pc, false); |
1748 | if (uncharge) | |
1749 | /* This is not "cancel", but cancel_charge does all we need. */ | |
1750 | mem_cgroup_cancel_charge(from); | |
d69b042f | 1751 | |
854ffa8d | 1752 | /* caller should have done css_get */ |
08e552c6 KH |
1753 | pc->mem_cgroup = to; |
1754 | mem_cgroup_charge_statistics(to, pc, true); | |
88703267 KH |
1755 | /* |
1756 | * We charges against "to" which may not have any tasks. Then, "to" | |
1757 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe DN |
1758 | * this function is just force_empty() and move charge, so it's |
1759 | * garanteed that "to" is never removed. So, we don't check rmdir | |
1760 | * status here. | |
88703267 | 1761 | */ |
57f9fd7d DN |
1762 | } |
1763 | ||
1764 | /* | |
1765 | * check whether the @pc is valid for moving account and call | |
1766 | * __mem_cgroup_move_account() | |
1767 | */ | |
1768 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
854ffa8d | 1769 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
57f9fd7d DN |
1770 | { |
1771 | int ret = -EINVAL; | |
1772 | lock_page_cgroup(pc); | |
1773 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { | |
854ffa8d | 1774 | __mem_cgroup_move_account(pc, from, to, uncharge); |
57f9fd7d DN |
1775 | ret = 0; |
1776 | } | |
1777 | unlock_page_cgroup(pc); | |
f817ed48 KH |
1778 | return ret; |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * move charges to its parent. | |
1783 | */ | |
1784 | ||
1785 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1786 | struct mem_cgroup *child, | |
1787 | gfp_t gfp_mask) | |
1788 | { | |
08e552c6 | 1789 | struct page *page = pc->page; |
f817ed48 KH |
1790 | struct cgroup *cg = child->css.cgroup; |
1791 | struct cgroup *pcg = cg->parent; | |
1792 | struct mem_cgroup *parent; | |
f817ed48 KH |
1793 | int ret; |
1794 | ||
1795 | /* Is ROOT ? */ | |
1796 | if (!pcg) | |
1797 | return -EINVAL; | |
1798 | ||
57f9fd7d DN |
1799 | ret = -EBUSY; |
1800 | if (!get_page_unless_zero(page)) | |
1801 | goto out; | |
1802 | if (isolate_lru_page(page)) | |
1803 | goto put; | |
08e552c6 | 1804 | |
f817ed48 | 1805 | parent = mem_cgroup_from_cont(pcg); |
f64c3f54 | 1806 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page); |
a636b327 | 1807 | if (ret || !parent) |
57f9fd7d | 1808 | goto put_back; |
f817ed48 | 1809 | |
854ffa8d DN |
1810 | ret = mem_cgroup_move_account(pc, child, parent, true); |
1811 | if (ret) | |
1812 | mem_cgroup_cancel_charge(parent); | |
57f9fd7d | 1813 | put_back: |
08e552c6 | 1814 | putback_lru_page(page); |
57f9fd7d | 1815 | put: |
40d58138 | 1816 | put_page(page); |
57f9fd7d | 1817 | out: |
f817ed48 KH |
1818 | return ret; |
1819 | } | |
1820 | ||
7a81b88c KH |
1821 | /* |
1822 | * Charge the memory controller for page usage. | |
1823 | * Return | |
1824 | * 0 if the charge was successful | |
1825 | * < 0 if the cgroup is over its limit | |
1826 | */ | |
1827 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1828 | gfp_t gfp_mask, enum charge_type ctype, | |
1829 | struct mem_cgroup *memcg) | |
1830 | { | |
1831 | struct mem_cgroup *mem; | |
1832 | struct page_cgroup *pc; | |
1833 | int ret; | |
1834 | ||
1835 | pc = lookup_page_cgroup(page); | |
1836 | /* can happen at boot */ | |
1837 | if (unlikely(!pc)) | |
1838 | return 0; | |
1839 | prefetchw(pc); | |
1840 | ||
1841 | mem = memcg; | |
f64c3f54 | 1842 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page); |
a636b327 | 1843 | if (ret || !mem) |
7a81b88c KH |
1844 | return ret; |
1845 | ||
1846 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1847 | return 0; |
8a9f3ccd BS |
1848 | } |
1849 | ||
7a81b88c KH |
1850 | int mem_cgroup_newpage_charge(struct page *page, |
1851 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1852 | { |
f8d66542 | 1853 | if (mem_cgroup_disabled()) |
cede86ac | 1854 | return 0; |
52d4b9ac KH |
1855 | if (PageCompound(page)) |
1856 | return 0; | |
69029cd5 KH |
1857 | /* |
1858 | * If already mapped, we don't have to account. | |
1859 | * If page cache, page->mapping has address_space. | |
1860 | * But page->mapping may have out-of-use anon_vma pointer, | |
1861 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1862 | * is NULL. | |
1863 | */ | |
1864 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1865 | return 0; | |
1866 | if (unlikely(!mm)) | |
1867 | mm = &init_mm; | |
217bc319 | 1868 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 1869 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
1870 | } |
1871 | ||
83aae4c7 DN |
1872 | static void |
1873 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1874 | enum charge_type ctype); | |
1875 | ||
e1a1cd59 BS |
1876 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
1877 | gfp_t gfp_mask) | |
8697d331 | 1878 | { |
b5a84319 KH |
1879 | struct mem_cgroup *mem = NULL; |
1880 | int ret; | |
1881 | ||
f8d66542 | 1882 | if (mem_cgroup_disabled()) |
cede86ac | 1883 | return 0; |
52d4b9ac KH |
1884 | if (PageCompound(page)) |
1885 | return 0; | |
accf163e KH |
1886 | /* |
1887 | * Corner case handling. This is called from add_to_page_cache() | |
1888 | * in usual. But some FS (shmem) precharges this page before calling it | |
1889 | * and call add_to_page_cache() with GFP_NOWAIT. | |
1890 | * | |
1891 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
1892 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
1893 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
1894 | * And when the page is SwapCache, it should take swap information |
1895 | * into account. This is under lock_page() now. | |
accf163e KH |
1896 | */ |
1897 | if (!(gfp_mask & __GFP_WAIT)) { | |
1898 | struct page_cgroup *pc; | |
1899 | ||
52d4b9ac KH |
1900 | |
1901 | pc = lookup_page_cgroup(page); | |
1902 | if (!pc) | |
1903 | return 0; | |
1904 | lock_page_cgroup(pc); | |
1905 | if (PageCgroupUsed(pc)) { | |
1906 | unlock_page_cgroup(pc); | |
accf163e KH |
1907 | return 0; |
1908 | } | |
52d4b9ac | 1909 | unlock_page_cgroup(pc); |
accf163e KH |
1910 | } |
1911 | ||
b5a84319 | 1912 | if (unlikely(!mm && !mem)) |
8697d331 | 1913 | mm = &init_mm; |
accf163e | 1914 | |
c05555b5 KH |
1915 | if (page_is_file_cache(page)) |
1916 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 1917 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 | 1918 | |
83aae4c7 DN |
1919 | /* shmem */ |
1920 | if (PageSwapCache(page)) { | |
1921 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
1922 | if (!ret) | |
1923 | __mem_cgroup_commit_charge_swapin(page, mem, | |
1924 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
1925 | } else | |
1926 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
1927 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
b5a84319 | 1928 | |
b5a84319 | 1929 | return ret; |
e8589cc1 KH |
1930 | } |
1931 | ||
54595fe2 KH |
1932 | /* |
1933 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
1934 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 1935 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
1936 | * "commit()" or removed by "cancel()" |
1937 | */ | |
8c7c6e34 KH |
1938 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1939 | struct page *page, | |
1940 | gfp_t mask, struct mem_cgroup **ptr) | |
1941 | { | |
1942 | struct mem_cgroup *mem; | |
54595fe2 | 1943 | int ret; |
8c7c6e34 | 1944 | |
f8d66542 | 1945 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
1946 | return 0; |
1947 | ||
1948 | if (!do_swap_account) | |
1949 | goto charge_cur_mm; | |
8c7c6e34 KH |
1950 | /* |
1951 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
1952 | * the pte, and even removed page from swap cache: in those cases |
1953 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
1954 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
1955 | */ |
1956 | if (!PageSwapCache(page)) | |
407f9c8b | 1957 | goto charge_cur_mm; |
e42d9d5d | 1958 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
1959 | if (!mem) |
1960 | goto charge_cur_mm; | |
8c7c6e34 | 1961 | *ptr = mem; |
f64c3f54 | 1962 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page); |
54595fe2 KH |
1963 | /* drop extra refcnt from tryget */ |
1964 | css_put(&mem->css); | |
1965 | return ret; | |
8c7c6e34 KH |
1966 | charge_cur_mm: |
1967 | if (unlikely(!mm)) | |
1968 | mm = &init_mm; | |
f64c3f54 | 1969 | return __mem_cgroup_try_charge(mm, mask, ptr, true, page); |
8c7c6e34 KH |
1970 | } |
1971 | ||
83aae4c7 DN |
1972 | static void |
1973 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1974 | enum charge_type ctype) | |
7a81b88c KH |
1975 | { |
1976 | struct page_cgroup *pc; | |
1977 | ||
f8d66542 | 1978 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1979 | return; |
1980 | if (!ptr) | |
1981 | return; | |
88703267 | 1982 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 1983 | pc = lookup_page_cgroup(page); |
544122e5 | 1984 | mem_cgroup_lru_del_before_commit_swapcache(page); |
83aae4c7 | 1985 | __mem_cgroup_commit_charge(ptr, pc, ctype); |
544122e5 | 1986 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
1987 | /* |
1988 | * Now swap is on-memory. This means this page may be | |
1989 | * counted both as mem and swap....double count. | |
03f3c433 KH |
1990 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
1991 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
1992 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 1993 | */ |
03f3c433 | 1994 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 1995 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 1996 | unsigned short id; |
8c7c6e34 | 1997 | struct mem_cgroup *memcg; |
a3b2d692 KH |
1998 | |
1999 | id = swap_cgroup_record(ent, 0); | |
2000 | rcu_read_lock(); | |
2001 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2002 | if (memcg) { |
a3b2d692 KH |
2003 | /* |
2004 | * This recorded memcg can be obsolete one. So, avoid | |
2005 | * calling css_tryget | |
2006 | */ | |
0c3e73e8 | 2007 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2008 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2009 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2010 | mem_cgroup_put(memcg); |
2011 | } | |
a3b2d692 | 2012 | rcu_read_unlock(); |
8c7c6e34 | 2013 | } |
88703267 KH |
2014 | /* |
2015 | * At swapin, we may charge account against cgroup which has no tasks. | |
2016 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2017 | * In that case, we need to call pre_destroy() again. check it here. | |
2018 | */ | |
2019 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2020 | } |
2021 | ||
83aae4c7 DN |
2022 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2023 | { | |
2024 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2025 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2026 | } | |
2027 | ||
7a81b88c KH |
2028 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2029 | { | |
f8d66542 | 2030 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2031 | return; |
2032 | if (!mem) | |
2033 | return; | |
a3032a2c | 2034 | mem_cgroup_cancel_charge(mem); |
7a81b88c KH |
2035 | } |
2036 | ||
569b846d KH |
2037 | static void |
2038 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype) | |
2039 | { | |
2040 | struct memcg_batch_info *batch = NULL; | |
2041 | bool uncharge_memsw = true; | |
2042 | /* If swapout, usage of swap doesn't decrease */ | |
2043 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2044 | uncharge_memsw = false; | |
2045 | /* | |
2046 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
2047 | * In those cases, all pages freed continously can be expected to be in | |
2048 | * the same cgroup and we have chance to coalesce uncharges. | |
2049 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2050 | * because we want to do uncharge as soon as possible. | |
2051 | */ | |
2052 | if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE)) | |
2053 | goto direct_uncharge; | |
2054 | ||
2055 | batch = ¤t->memcg_batch; | |
2056 | /* | |
2057 | * In usual, we do css_get() when we remember memcg pointer. | |
2058 | * But in this case, we keep res->usage until end of a series of | |
2059 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2060 | */ | |
2061 | if (!batch->memcg) | |
2062 | batch->memcg = mem; | |
2063 | /* | |
2064 | * In typical case, batch->memcg == mem. This means we can | |
2065 | * merge a series of uncharges to an uncharge of res_counter. | |
2066 | * If not, we uncharge res_counter ony by one. | |
2067 | */ | |
2068 | if (batch->memcg != mem) | |
2069 | goto direct_uncharge; | |
2070 | /* remember freed charge and uncharge it later */ | |
2071 | batch->bytes += PAGE_SIZE; | |
2072 | if (uncharge_memsw) | |
2073 | batch->memsw_bytes += PAGE_SIZE; | |
2074 | return; | |
2075 | direct_uncharge: | |
2076 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
2077 | if (uncharge_memsw) | |
2078 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
2079 | return; | |
2080 | } | |
7a81b88c | 2081 | |
8a9f3ccd | 2082 | /* |
69029cd5 | 2083 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2084 | */ |
8c7c6e34 | 2085 | static struct mem_cgroup * |
69029cd5 | 2086 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2087 | { |
8289546e | 2088 | struct page_cgroup *pc; |
8c7c6e34 | 2089 | struct mem_cgroup *mem = NULL; |
072c56c1 | 2090 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 2091 | |
f8d66542 | 2092 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2093 | return NULL; |
4077960e | 2094 | |
d13d1443 | 2095 | if (PageSwapCache(page)) |
8c7c6e34 | 2096 | return NULL; |
d13d1443 | 2097 | |
8697d331 | 2098 | /* |
3c541e14 | 2099 | * Check if our page_cgroup is valid |
8697d331 | 2100 | */ |
52d4b9ac KH |
2101 | pc = lookup_page_cgroup(page); |
2102 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2103 | return NULL; |
b9c565d5 | 2104 | |
52d4b9ac | 2105 | lock_page_cgroup(pc); |
d13d1443 | 2106 | |
8c7c6e34 KH |
2107 | mem = pc->mem_cgroup; |
2108 | ||
d13d1443 KH |
2109 | if (!PageCgroupUsed(pc)) |
2110 | goto unlock_out; | |
2111 | ||
2112 | switch (ctype) { | |
2113 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2114 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
d13d1443 KH |
2115 | if (page_mapped(page)) |
2116 | goto unlock_out; | |
2117 | break; | |
2118 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2119 | if (!PageAnon(page)) { /* Shared memory */ | |
2120 | if (page->mapping && !page_is_file_cache(page)) | |
2121 | goto unlock_out; | |
2122 | } else if (page_mapped(page)) /* Anon */ | |
2123 | goto unlock_out; | |
2124 | break; | |
2125 | default: | |
2126 | break; | |
52d4b9ac | 2127 | } |
d13d1443 | 2128 | |
569b846d KH |
2129 | if (!mem_cgroup_is_root(mem)) |
2130 | __do_uncharge(mem, ctype); | |
0c3e73e8 BS |
2131 | if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
2132 | mem_cgroup_swap_statistics(mem, true); | |
08e552c6 | 2133 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 2134 | |
52d4b9ac | 2135 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2136 | /* |
2137 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2138 | * freed from LRU. This is safe because uncharged page is expected not | |
2139 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2140 | * special functions. | |
2141 | */ | |
b9c565d5 | 2142 | |
69029cd5 | 2143 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 2144 | unlock_page_cgroup(pc); |
fb59e9f1 | 2145 | |
4e649152 | 2146 | if (mem_cgroup_soft_limit_check(mem)) |
f64c3f54 | 2147 | mem_cgroup_update_tree(mem, page); |
a7fe942e KH |
2148 | /* at swapout, this memcg will be accessed to record to swap */ |
2149 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2150 | css_put(&mem->css); | |
6d12e2d8 | 2151 | |
8c7c6e34 | 2152 | return mem; |
d13d1443 KH |
2153 | |
2154 | unlock_out: | |
2155 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2156 | return NULL; |
3c541e14 BS |
2157 | } |
2158 | ||
69029cd5 KH |
2159 | void mem_cgroup_uncharge_page(struct page *page) |
2160 | { | |
52d4b9ac KH |
2161 | /* early check. */ |
2162 | if (page_mapped(page)) | |
2163 | return; | |
2164 | if (page->mapping && !PageAnon(page)) | |
2165 | return; | |
69029cd5 KH |
2166 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2167 | } | |
2168 | ||
2169 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2170 | { | |
2171 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2172 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2173 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2174 | } | |
2175 | ||
569b846d KH |
2176 | /* |
2177 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2178 | * In that cases, pages are freed continuously and we can expect pages | |
2179 | * are in the same memcg. All these calls itself limits the number of | |
2180 | * pages freed at once, then uncharge_start/end() is called properly. | |
2181 | * This may be called prural(2) times in a context, | |
2182 | */ | |
2183 | ||
2184 | void mem_cgroup_uncharge_start(void) | |
2185 | { | |
2186 | current->memcg_batch.do_batch++; | |
2187 | /* We can do nest. */ | |
2188 | if (current->memcg_batch.do_batch == 1) { | |
2189 | current->memcg_batch.memcg = NULL; | |
2190 | current->memcg_batch.bytes = 0; | |
2191 | current->memcg_batch.memsw_bytes = 0; | |
2192 | } | |
2193 | } | |
2194 | ||
2195 | void mem_cgroup_uncharge_end(void) | |
2196 | { | |
2197 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2198 | ||
2199 | if (!batch->do_batch) | |
2200 | return; | |
2201 | ||
2202 | batch->do_batch--; | |
2203 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2204 | return; | |
2205 | ||
2206 | if (!batch->memcg) | |
2207 | return; | |
2208 | /* | |
2209 | * This "batch->memcg" is valid without any css_get/put etc... | |
2210 | * bacause we hide charges behind us. | |
2211 | */ | |
2212 | if (batch->bytes) | |
2213 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2214 | if (batch->memsw_bytes) | |
2215 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
2216 | /* forget this pointer (for sanity check) */ | |
2217 | batch->memcg = NULL; | |
2218 | } | |
2219 | ||
e767e056 | 2220 | #ifdef CONFIG_SWAP |
8c7c6e34 | 2221 | /* |
e767e056 | 2222 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
2223 | * memcg information is recorded to swap_cgroup of "ent" |
2224 | */ | |
8a9478ca KH |
2225 | void |
2226 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
2227 | { |
2228 | struct mem_cgroup *memcg; | |
8a9478ca KH |
2229 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
2230 | ||
2231 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2232 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2233 | ||
2234 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 2235 | |
8c7c6e34 | 2236 | /* record memcg information */ |
8a9478ca | 2237 | if (do_swap_account && swapout && memcg) { |
a3b2d692 | 2238 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 KH |
2239 | mem_cgroup_get(memcg); |
2240 | } | |
8a9478ca | 2241 | if (swapout && memcg) |
a7fe942e | 2242 | css_put(&memcg->css); |
8c7c6e34 | 2243 | } |
e767e056 | 2244 | #endif |
8c7c6e34 KH |
2245 | |
2246 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2247 | /* | |
2248 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2249 | * uncharge "memsw" account. | |
2250 | */ | |
2251 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 2252 | { |
8c7c6e34 | 2253 | struct mem_cgroup *memcg; |
a3b2d692 | 2254 | unsigned short id; |
8c7c6e34 KH |
2255 | |
2256 | if (!do_swap_account) | |
2257 | return; | |
2258 | ||
a3b2d692 KH |
2259 | id = swap_cgroup_record(ent, 0); |
2260 | rcu_read_lock(); | |
2261 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2262 | if (memcg) { |
a3b2d692 KH |
2263 | /* |
2264 | * We uncharge this because swap is freed. | |
2265 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2266 | */ | |
0c3e73e8 | 2267 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2268 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2269 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2270 | mem_cgroup_put(memcg); |
2271 | } | |
a3b2d692 | 2272 | rcu_read_unlock(); |
d13d1443 | 2273 | } |
02491447 DN |
2274 | |
2275 | /** | |
2276 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2277 | * @entry: swap entry to be moved | |
2278 | * @from: mem_cgroup which the entry is moved from | |
2279 | * @to: mem_cgroup which the entry is moved to | |
2280 | * | |
2281 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2282 | * as the mem_cgroup's id of @from. | |
2283 | * | |
2284 | * Returns 0 on success, -EINVAL on failure. | |
2285 | * | |
2286 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
2287 | * both res and memsw, and called css_get(). | |
2288 | */ | |
2289 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
2290 | struct mem_cgroup *from, struct mem_cgroup *to) | |
2291 | { | |
2292 | unsigned short old_id, new_id; | |
2293 | ||
2294 | old_id = css_id(&from->css); | |
2295 | new_id = css_id(&to->css); | |
2296 | ||
2297 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
2298 | if (!mem_cgroup_is_root(from)) | |
2299 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
2300 | mem_cgroup_swap_statistics(from, false); | |
2301 | mem_cgroup_put(from); | |
2302 | /* | |
2303 | * we charged both to->res and to->memsw, so we should uncharge | |
2304 | * to->res. | |
2305 | */ | |
2306 | if (!mem_cgroup_is_root(to)) | |
2307 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
2308 | mem_cgroup_swap_statistics(to, true); | |
2309 | mem_cgroup_get(to); | |
2310 | css_put(&to->css); | |
2311 | ||
2312 | return 0; | |
2313 | } | |
2314 | return -EINVAL; | |
2315 | } | |
2316 | #else | |
2317 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
2318 | struct mem_cgroup *from, struct mem_cgroup *to) | |
2319 | { | |
2320 | return -EINVAL; | |
2321 | } | |
8c7c6e34 | 2322 | #endif |
d13d1443 | 2323 | |
ae41be37 | 2324 | /* |
01b1ae63 KH |
2325 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
2326 | * page belongs to. | |
ae41be37 | 2327 | */ |
01b1ae63 | 2328 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
2329 | { |
2330 | struct page_cgroup *pc; | |
e8589cc1 | 2331 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 2332 | int ret = 0; |
8869b8f6 | 2333 | |
f8d66542 | 2334 | if (mem_cgroup_disabled()) |
4077960e BS |
2335 | return 0; |
2336 | ||
52d4b9ac KH |
2337 | pc = lookup_page_cgroup(page); |
2338 | lock_page_cgroup(pc); | |
2339 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
2340 | mem = pc->mem_cgroup; |
2341 | css_get(&mem->css); | |
e8589cc1 | 2342 | } |
52d4b9ac | 2343 | unlock_page_cgroup(pc); |
01b1ae63 | 2344 | |
e8589cc1 | 2345 | if (mem) { |
f64c3f54 BS |
2346 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, |
2347 | page); | |
e8589cc1 KH |
2348 | css_put(&mem->css); |
2349 | } | |
01b1ae63 | 2350 | *ptr = mem; |
e8589cc1 | 2351 | return ret; |
ae41be37 | 2352 | } |
8869b8f6 | 2353 | |
69029cd5 | 2354 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
2355 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
2356 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 2357 | { |
01b1ae63 KH |
2358 | struct page *target, *unused; |
2359 | struct page_cgroup *pc; | |
2360 | enum charge_type ctype; | |
2361 | ||
2362 | if (!mem) | |
2363 | return; | |
88703267 | 2364 | cgroup_exclude_rmdir(&mem->css); |
01b1ae63 KH |
2365 | /* at migration success, oldpage->mapping is NULL. */ |
2366 | if (oldpage->mapping) { | |
2367 | target = oldpage; | |
2368 | unused = NULL; | |
2369 | } else { | |
2370 | target = newpage; | |
2371 | unused = oldpage; | |
2372 | } | |
2373 | ||
2374 | if (PageAnon(target)) | |
2375 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2376 | else if (page_is_file_cache(target)) | |
2377 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2378 | else | |
2379 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
2380 | ||
2381 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 2382 | if (unused) |
01b1ae63 KH |
2383 | __mem_cgroup_uncharge_common(unused, ctype); |
2384 | ||
2385 | pc = lookup_page_cgroup(target); | |
69029cd5 | 2386 | /* |
01b1ae63 KH |
2387 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
2388 | * So, double-counting is effectively avoided. | |
2389 | */ | |
2390 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
2391 | ||
2392 | /* | |
2393 | * Both of oldpage and newpage are still under lock_page(). | |
2394 | * Then, we don't have to care about race in radix-tree. | |
2395 | * But we have to be careful that this page is unmapped or not. | |
2396 | * | |
2397 | * There is a case for !page_mapped(). At the start of | |
2398 | * migration, oldpage was mapped. But now, it's zapped. | |
2399 | * But we know *target* page is not freed/reused under us. | |
2400 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 2401 | */ |
01b1ae63 KH |
2402 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
2403 | mem_cgroup_uncharge_page(target); | |
88703267 KH |
2404 | /* |
2405 | * At migration, we may charge account against cgroup which has no tasks | |
2406 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2407 | * In that case, we need to call pre_destroy() again. check it here. | |
2408 | */ | |
2409 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 2410 | } |
78fb7466 | 2411 | |
c9b0ed51 | 2412 | /* |
ae3abae6 DN |
2413 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
2414 | * Calling hierarchical_reclaim is not enough because we should update | |
2415 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
2416 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
2417 | * not from the memcg which this page would be charged to. | |
2418 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 2419 | */ |
ae3abae6 | 2420 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
2421 | struct mm_struct *mm, |
2422 | gfp_t gfp_mask) | |
c9b0ed51 | 2423 | { |
b5a84319 | 2424 | struct mem_cgroup *mem = NULL; |
ae3abae6 | 2425 | int ret; |
c9b0ed51 | 2426 | |
f8d66542 | 2427 | if (mem_cgroup_disabled()) |
cede86ac | 2428 | return 0; |
c9b0ed51 | 2429 | |
ae3abae6 DN |
2430 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2431 | if (!ret) | |
2432 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 2433 | |
ae3abae6 | 2434 | return ret; |
c9b0ed51 KH |
2435 | } |
2436 | ||
8c7c6e34 KH |
2437 | static DEFINE_MUTEX(set_limit_mutex); |
2438 | ||
d38d2a75 | 2439 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 2440 | unsigned long long val) |
628f4235 | 2441 | { |
81d39c20 | 2442 | int retry_count; |
8c7c6e34 | 2443 | u64 memswlimit; |
628f4235 | 2444 | int ret = 0; |
81d39c20 KH |
2445 | int children = mem_cgroup_count_children(memcg); |
2446 | u64 curusage, oldusage; | |
2447 | ||
2448 | /* | |
2449 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2450 | * is depends on callers. We set our retry-count to be function | |
2451 | * of # of children which we should visit in this loop. | |
2452 | */ | |
2453 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
2454 | ||
2455 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 2456 | |
8c7c6e34 | 2457 | while (retry_count) { |
628f4235 KH |
2458 | if (signal_pending(current)) { |
2459 | ret = -EINTR; | |
2460 | break; | |
2461 | } | |
8c7c6e34 KH |
2462 | /* |
2463 | * Rather than hide all in some function, I do this in | |
2464 | * open coded manner. You see what this really does. | |
2465 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2466 | */ | |
2467 | mutex_lock(&set_limit_mutex); | |
2468 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2469 | if (memswlimit < val) { | |
2470 | ret = -EINVAL; | |
2471 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
2472 | break; |
2473 | } | |
8c7c6e34 | 2474 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
2475 | if (!ret) { |
2476 | if (memswlimit == val) | |
2477 | memcg->memsw_is_minimum = true; | |
2478 | else | |
2479 | memcg->memsw_is_minimum = false; | |
2480 | } | |
8c7c6e34 KH |
2481 | mutex_unlock(&set_limit_mutex); |
2482 | ||
2483 | if (!ret) | |
2484 | break; | |
2485 | ||
aa20d489 | 2486 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
4e416953 | 2487 | MEM_CGROUP_RECLAIM_SHRINK); |
81d39c20 KH |
2488 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
2489 | /* Usage is reduced ? */ | |
2490 | if (curusage >= oldusage) | |
2491 | retry_count--; | |
2492 | else | |
2493 | oldusage = curusage; | |
8c7c6e34 | 2494 | } |
14797e23 | 2495 | |
8c7c6e34 KH |
2496 | return ret; |
2497 | } | |
2498 | ||
338c8431 LZ |
2499 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
2500 | unsigned long long val) | |
8c7c6e34 | 2501 | { |
81d39c20 | 2502 | int retry_count; |
8c7c6e34 | 2503 | u64 memlimit, oldusage, curusage; |
81d39c20 KH |
2504 | int children = mem_cgroup_count_children(memcg); |
2505 | int ret = -EBUSY; | |
8c7c6e34 | 2506 | |
81d39c20 KH |
2507 | /* see mem_cgroup_resize_res_limit */ |
2508 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
2509 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
2510 | while (retry_count) { |
2511 | if (signal_pending(current)) { | |
2512 | ret = -EINTR; | |
2513 | break; | |
2514 | } | |
2515 | /* | |
2516 | * Rather than hide all in some function, I do this in | |
2517 | * open coded manner. You see what this really does. | |
2518 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2519 | */ | |
2520 | mutex_lock(&set_limit_mutex); | |
2521 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2522 | if (memlimit > val) { | |
2523 | ret = -EINVAL; | |
2524 | mutex_unlock(&set_limit_mutex); | |
2525 | break; | |
2526 | } | |
2527 | ret = res_counter_set_limit(&memcg->memsw, val); | |
22a668d7 KH |
2528 | if (!ret) { |
2529 | if (memlimit == val) | |
2530 | memcg->memsw_is_minimum = true; | |
2531 | else | |
2532 | memcg->memsw_is_minimum = false; | |
2533 | } | |
8c7c6e34 KH |
2534 | mutex_unlock(&set_limit_mutex); |
2535 | ||
2536 | if (!ret) | |
2537 | break; | |
2538 | ||
4e416953 | 2539 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 BS |
2540 | MEM_CGROUP_RECLAIM_NOSWAP | |
2541 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 2542 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 2543 | /* Usage is reduced ? */ |
8c7c6e34 | 2544 | if (curusage >= oldusage) |
628f4235 | 2545 | retry_count--; |
81d39c20 KH |
2546 | else |
2547 | oldusage = curusage; | |
628f4235 KH |
2548 | } |
2549 | return ret; | |
2550 | } | |
2551 | ||
4e416953 BS |
2552 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2553 | gfp_t gfp_mask, int nid, | |
2554 | int zid) | |
2555 | { | |
2556 | unsigned long nr_reclaimed = 0; | |
2557 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2558 | unsigned long reclaimed; | |
2559 | int loop = 0; | |
2560 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 2561 | unsigned long long excess; |
4e416953 BS |
2562 | |
2563 | if (order > 0) | |
2564 | return 0; | |
2565 | ||
2566 | mctz = soft_limit_tree_node_zone(nid, zid); | |
2567 | /* | |
2568 | * This loop can run a while, specially if mem_cgroup's continuously | |
2569 | * keep exceeding their soft limit and putting the system under | |
2570 | * pressure | |
2571 | */ | |
2572 | do { | |
2573 | if (next_mz) | |
2574 | mz = next_mz; | |
2575 | else | |
2576 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2577 | if (!mz) | |
2578 | break; | |
2579 | ||
2580 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
2581 | gfp_mask, | |
2582 | MEM_CGROUP_RECLAIM_SOFT); | |
2583 | nr_reclaimed += reclaimed; | |
2584 | spin_lock(&mctz->lock); | |
2585 | ||
2586 | /* | |
2587 | * If we failed to reclaim anything from this memory cgroup | |
2588 | * it is time to move on to the next cgroup | |
2589 | */ | |
2590 | next_mz = NULL; | |
2591 | if (!reclaimed) { | |
2592 | do { | |
2593 | /* | |
2594 | * Loop until we find yet another one. | |
2595 | * | |
2596 | * By the time we get the soft_limit lock | |
2597 | * again, someone might have aded the | |
2598 | * group back on the RB tree. Iterate to | |
2599 | * make sure we get a different mem. | |
2600 | * mem_cgroup_largest_soft_limit_node returns | |
2601 | * NULL if no other cgroup is present on | |
2602 | * the tree | |
2603 | */ | |
2604 | next_mz = | |
2605 | __mem_cgroup_largest_soft_limit_node(mctz); | |
2606 | if (next_mz == mz) { | |
2607 | css_put(&next_mz->mem->css); | |
2608 | next_mz = NULL; | |
2609 | } else /* next_mz == NULL or other memcg */ | |
2610 | break; | |
2611 | } while (1); | |
2612 | } | |
4e416953 | 2613 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 2614 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
2615 | /* |
2616 | * One school of thought says that we should not add | |
2617 | * back the node to the tree if reclaim returns 0. | |
2618 | * But our reclaim could return 0, simply because due | |
2619 | * to priority we are exposing a smaller subset of | |
2620 | * memory to reclaim from. Consider this as a longer | |
2621 | * term TODO. | |
2622 | */ | |
ef8745c1 KH |
2623 | /* If excess == 0, no tree ops */ |
2624 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
2625 | spin_unlock(&mctz->lock); |
2626 | css_put(&mz->mem->css); | |
2627 | loop++; | |
2628 | /* | |
2629 | * Could not reclaim anything and there are no more | |
2630 | * mem cgroups to try or we seem to be looping without | |
2631 | * reclaiming anything. | |
2632 | */ | |
2633 | if (!nr_reclaimed && | |
2634 | (next_mz == NULL || | |
2635 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2636 | break; | |
2637 | } while (!nr_reclaimed); | |
2638 | if (next_mz) | |
2639 | css_put(&next_mz->mem->css); | |
2640 | return nr_reclaimed; | |
2641 | } | |
2642 | ||
cc847582 KH |
2643 | /* |
2644 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
2645 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
2646 | */ | |
f817ed48 | 2647 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 2648 | int node, int zid, enum lru_list lru) |
cc847582 | 2649 | { |
08e552c6 KH |
2650 | struct zone *zone; |
2651 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 2652 | struct page_cgroup *pc, *busy; |
08e552c6 | 2653 | unsigned long flags, loop; |
072c56c1 | 2654 | struct list_head *list; |
f817ed48 | 2655 | int ret = 0; |
072c56c1 | 2656 | |
08e552c6 KH |
2657 | zone = &NODE_DATA(node)->node_zones[zid]; |
2658 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 2659 | list = &mz->lists[lru]; |
cc847582 | 2660 | |
f817ed48 KH |
2661 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
2662 | /* give some margin against EBUSY etc...*/ | |
2663 | loop += 256; | |
2664 | busy = NULL; | |
2665 | while (loop--) { | |
2666 | ret = 0; | |
08e552c6 | 2667 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 2668 | if (list_empty(list)) { |
08e552c6 | 2669 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 2670 | break; |
f817ed48 KH |
2671 | } |
2672 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
2673 | if (busy == pc) { | |
2674 | list_move(&pc->lru, list); | |
648bcc77 | 2675 | busy = NULL; |
08e552c6 | 2676 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
2677 | continue; |
2678 | } | |
08e552c6 | 2679 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 2680 | |
2c26fdd7 | 2681 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 2682 | if (ret == -ENOMEM) |
52d4b9ac | 2683 | break; |
f817ed48 KH |
2684 | |
2685 | if (ret == -EBUSY || ret == -EINVAL) { | |
2686 | /* found lock contention or "pc" is obsolete. */ | |
2687 | busy = pc; | |
2688 | cond_resched(); | |
2689 | } else | |
2690 | busy = NULL; | |
cc847582 | 2691 | } |
08e552c6 | 2692 | |
f817ed48 KH |
2693 | if (!ret && !list_empty(list)) |
2694 | return -EBUSY; | |
2695 | return ret; | |
cc847582 KH |
2696 | } |
2697 | ||
2698 | /* | |
2699 | * make mem_cgroup's charge to be 0 if there is no task. | |
2700 | * This enables deleting this mem_cgroup. | |
2701 | */ | |
c1e862c1 | 2702 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 2703 | { |
f817ed48 KH |
2704 | int ret; |
2705 | int node, zid, shrink; | |
2706 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 2707 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 2708 | |
cc847582 | 2709 | css_get(&mem->css); |
f817ed48 KH |
2710 | |
2711 | shrink = 0; | |
c1e862c1 KH |
2712 | /* should free all ? */ |
2713 | if (free_all) | |
2714 | goto try_to_free; | |
f817ed48 | 2715 | move_account: |
fce66477 | 2716 | do { |
f817ed48 | 2717 | ret = -EBUSY; |
c1e862c1 KH |
2718 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
2719 | goto out; | |
2720 | ret = -EINTR; | |
2721 | if (signal_pending(current)) | |
cc847582 | 2722 | goto out; |
52d4b9ac KH |
2723 | /* This is for making all *used* pages to be on LRU. */ |
2724 | lru_add_drain_all(); | |
cdec2e42 | 2725 | drain_all_stock_sync(); |
f817ed48 | 2726 | ret = 0; |
299b4eaa | 2727 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 2728 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 2729 | enum lru_list l; |
f817ed48 KH |
2730 | for_each_lru(l) { |
2731 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 2732 | node, zid, l); |
f817ed48 KH |
2733 | if (ret) |
2734 | break; | |
2735 | } | |
1ecaab2b | 2736 | } |
f817ed48 KH |
2737 | if (ret) |
2738 | break; | |
2739 | } | |
2740 | /* it seems parent cgroup doesn't have enough mem */ | |
2741 | if (ret == -ENOMEM) | |
2742 | goto try_to_free; | |
52d4b9ac | 2743 | cond_resched(); |
fce66477 DN |
2744 | /* "ret" should also be checked to ensure all lists are empty. */ |
2745 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
2746 | out: |
2747 | css_put(&mem->css); | |
2748 | return ret; | |
f817ed48 KH |
2749 | |
2750 | try_to_free: | |
c1e862c1 KH |
2751 | /* returns EBUSY if there is a task or if we come here twice. */ |
2752 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
2753 | ret = -EBUSY; |
2754 | goto out; | |
2755 | } | |
c1e862c1 KH |
2756 | /* we call try-to-free pages for make this cgroup empty */ |
2757 | lru_add_drain_all(); | |
f817ed48 KH |
2758 | /* try to free all pages in this cgroup */ |
2759 | shrink = 1; | |
2760 | while (nr_retries && mem->res.usage > 0) { | |
2761 | int progress; | |
c1e862c1 KH |
2762 | |
2763 | if (signal_pending(current)) { | |
2764 | ret = -EINTR; | |
2765 | goto out; | |
2766 | } | |
a7885eb8 KM |
2767 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
2768 | false, get_swappiness(mem)); | |
c1e862c1 | 2769 | if (!progress) { |
f817ed48 | 2770 | nr_retries--; |
c1e862c1 | 2771 | /* maybe some writeback is necessary */ |
8aa7e847 | 2772 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2773 | } |
f817ed48 KH |
2774 | |
2775 | } | |
08e552c6 | 2776 | lru_add_drain(); |
f817ed48 | 2777 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 2778 | goto move_account; |
cc847582 KH |
2779 | } |
2780 | ||
c1e862c1 KH |
2781 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
2782 | { | |
2783 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
2784 | } | |
2785 | ||
2786 | ||
18f59ea7 BS |
2787 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
2788 | { | |
2789 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
2790 | } | |
2791 | ||
2792 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
2793 | u64 val) | |
2794 | { | |
2795 | int retval = 0; | |
2796 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
2797 | struct cgroup *parent = cont->parent; | |
2798 | struct mem_cgroup *parent_mem = NULL; | |
2799 | ||
2800 | if (parent) | |
2801 | parent_mem = mem_cgroup_from_cont(parent); | |
2802 | ||
2803 | cgroup_lock(); | |
2804 | /* | |
af901ca1 | 2805 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2806 | * in the child subtrees. If it is unset, then the change can |
2807 | * occur, provided the current cgroup has no children. | |
2808 | * | |
2809 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2810 | * set if there are no children. | |
2811 | */ | |
2812 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
2813 | (val == 1 || val == 0)) { | |
2814 | if (list_empty(&cont->children)) | |
2815 | mem->use_hierarchy = val; | |
2816 | else | |
2817 | retval = -EBUSY; | |
2818 | } else | |
2819 | retval = -EINVAL; | |
2820 | cgroup_unlock(); | |
2821 | ||
2822 | return retval; | |
2823 | } | |
2824 | ||
0c3e73e8 BS |
2825 | struct mem_cgroup_idx_data { |
2826 | s64 val; | |
2827 | enum mem_cgroup_stat_index idx; | |
2828 | }; | |
2829 | ||
2830 | static int | |
2831 | mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) | |
2832 | { | |
2833 | struct mem_cgroup_idx_data *d = data; | |
2834 | d->val += mem_cgroup_read_stat(&mem->stat, d->idx); | |
2835 | return 0; | |
2836 | } | |
2837 | ||
2838 | static void | |
2839 | mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, | |
2840 | enum mem_cgroup_stat_index idx, s64 *val) | |
2841 | { | |
2842 | struct mem_cgroup_idx_data d; | |
2843 | d.idx = idx; | |
2844 | d.val = 0; | |
2845 | mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat); | |
2846 | *val = d.val; | |
2847 | } | |
2848 | ||
2c3daa72 | 2849 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 2850 | { |
8c7c6e34 | 2851 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
0c3e73e8 | 2852 | u64 idx_val, val; |
8c7c6e34 KH |
2853 | int type, name; |
2854 | ||
2855 | type = MEMFILE_TYPE(cft->private); | |
2856 | name = MEMFILE_ATTR(cft->private); | |
2857 | switch (type) { | |
2858 | case _MEM: | |
0c3e73e8 BS |
2859 | if (name == RES_USAGE && mem_cgroup_is_root(mem)) { |
2860 | mem_cgroup_get_recursive_idx_stat(mem, | |
2861 | MEM_CGROUP_STAT_CACHE, &idx_val); | |
2862 | val = idx_val; | |
2863 | mem_cgroup_get_recursive_idx_stat(mem, | |
2864 | MEM_CGROUP_STAT_RSS, &idx_val); | |
2865 | val += idx_val; | |
2866 | val <<= PAGE_SHIFT; | |
2867 | } else | |
2868 | val = res_counter_read_u64(&mem->res, name); | |
8c7c6e34 KH |
2869 | break; |
2870 | case _MEMSWAP: | |
0c3e73e8 BS |
2871 | if (name == RES_USAGE && mem_cgroup_is_root(mem)) { |
2872 | mem_cgroup_get_recursive_idx_stat(mem, | |
2873 | MEM_CGROUP_STAT_CACHE, &idx_val); | |
2874 | val = idx_val; | |
2875 | mem_cgroup_get_recursive_idx_stat(mem, | |
2876 | MEM_CGROUP_STAT_RSS, &idx_val); | |
2877 | val += idx_val; | |
2878 | mem_cgroup_get_recursive_idx_stat(mem, | |
2879 | MEM_CGROUP_STAT_SWAPOUT, &idx_val); | |
cd9b45b7 | 2880 | val += idx_val; |
0c3e73e8 BS |
2881 | val <<= PAGE_SHIFT; |
2882 | } else | |
2883 | val = res_counter_read_u64(&mem->memsw, name); | |
8c7c6e34 KH |
2884 | break; |
2885 | default: | |
2886 | BUG(); | |
2887 | break; | |
2888 | } | |
2889 | return val; | |
8cdea7c0 | 2890 | } |
628f4235 KH |
2891 | /* |
2892 | * The user of this function is... | |
2893 | * RES_LIMIT. | |
2894 | */ | |
856c13aa PM |
2895 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
2896 | const char *buffer) | |
8cdea7c0 | 2897 | { |
628f4235 | 2898 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 2899 | int type, name; |
628f4235 KH |
2900 | unsigned long long val; |
2901 | int ret; | |
2902 | ||
8c7c6e34 KH |
2903 | type = MEMFILE_TYPE(cft->private); |
2904 | name = MEMFILE_ATTR(cft->private); | |
2905 | switch (name) { | |
628f4235 | 2906 | case RES_LIMIT: |
4b3bde4c BS |
2907 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2908 | ret = -EINVAL; | |
2909 | break; | |
2910 | } | |
628f4235 KH |
2911 | /* This function does all necessary parse...reuse it */ |
2912 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
2913 | if (ret) |
2914 | break; | |
2915 | if (type == _MEM) | |
628f4235 | 2916 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
2917 | else |
2918 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 2919 | break; |
296c81d8 BS |
2920 | case RES_SOFT_LIMIT: |
2921 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
2922 | if (ret) | |
2923 | break; | |
2924 | /* | |
2925 | * For memsw, soft limits are hard to implement in terms | |
2926 | * of semantics, for now, we support soft limits for | |
2927 | * control without swap | |
2928 | */ | |
2929 | if (type == _MEM) | |
2930 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
2931 | else | |
2932 | ret = -EINVAL; | |
2933 | break; | |
628f4235 KH |
2934 | default: |
2935 | ret = -EINVAL; /* should be BUG() ? */ | |
2936 | break; | |
2937 | } | |
2938 | return ret; | |
8cdea7c0 BS |
2939 | } |
2940 | ||
fee7b548 KH |
2941 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
2942 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
2943 | { | |
2944 | struct cgroup *cgroup; | |
2945 | unsigned long long min_limit, min_memsw_limit, tmp; | |
2946 | ||
2947 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2948 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2949 | cgroup = memcg->css.cgroup; | |
2950 | if (!memcg->use_hierarchy) | |
2951 | goto out; | |
2952 | ||
2953 | while (cgroup->parent) { | |
2954 | cgroup = cgroup->parent; | |
2955 | memcg = mem_cgroup_from_cont(cgroup); | |
2956 | if (!memcg->use_hierarchy) | |
2957 | break; | |
2958 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2959 | min_limit = min(min_limit, tmp); | |
2960 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2961 | min_memsw_limit = min(min_memsw_limit, tmp); | |
2962 | } | |
2963 | out: | |
2964 | *mem_limit = min_limit; | |
2965 | *memsw_limit = min_memsw_limit; | |
2966 | return; | |
2967 | } | |
2968 | ||
29f2a4da | 2969 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
2970 | { |
2971 | struct mem_cgroup *mem; | |
8c7c6e34 | 2972 | int type, name; |
c84872e1 PE |
2973 | |
2974 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
2975 | type = MEMFILE_TYPE(event); |
2976 | name = MEMFILE_ATTR(event); | |
2977 | switch (name) { | |
29f2a4da | 2978 | case RES_MAX_USAGE: |
8c7c6e34 KH |
2979 | if (type == _MEM) |
2980 | res_counter_reset_max(&mem->res); | |
2981 | else | |
2982 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
2983 | break; |
2984 | case RES_FAILCNT: | |
8c7c6e34 KH |
2985 | if (type == _MEM) |
2986 | res_counter_reset_failcnt(&mem->res); | |
2987 | else | |
2988 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
2989 | break; |
2990 | } | |
f64c3f54 | 2991 | |
85cc59db | 2992 | return 0; |
c84872e1 PE |
2993 | } |
2994 | ||
7dc74be0 DN |
2995 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
2996 | struct cftype *cft) | |
2997 | { | |
2998 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
2999 | } | |
3000 | ||
02491447 | 3001 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3002 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3003 | struct cftype *cft, u64 val) | |
3004 | { | |
3005 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3006 | ||
3007 | if (val >= (1 << NR_MOVE_TYPE)) | |
3008 | return -EINVAL; | |
3009 | /* | |
3010 | * We check this value several times in both in can_attach() and | |
3011 | * attach(), so we need cgroup lock to prevent this value from being | |
3012 | * inconsistent. | |
3013 | */ | |
3014 | cgroup_lock(); | |
3015 | mem->move_charge_at_immigrate = val; | |
3016 | cgroup_unlock(); | |
3017 | ||
3018 | return 0; | |
3019 | } | |
02491447 DN |
3020 | #else |
3021 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3022 | struct cftype *cft, u64 val) | |
3023 | { | |
3024 | return -ENOSYS; | |
3025 | } | |
3026 | #endif | |
7dc74be0 | 3027 | |
14067bb3 KH |
3028 | |
3029 | /* For read statistics */ | |
3030 | enum { | |
3031 | MCS_CACHE, | |
3032 | MCS_RSS, | |
d8046582 | 3033 | MCS_FILE_MAPPED, |
14067bb3 KH |
3034 | MCS_PGPGIN, |
3035 | MCS_PGPGOUT, | |
1dd3a273 | 3036 | MCS_SWAP, |
14067bb3 KH |
3037 | MCS_INACTIVE_ANON, |
3038 | MCS_ACTIVE_ANON, | |
3039 | MCS_INACTIVE_FILE, | |
3040 | MCS_ACTIVE_FILE, | |
3041 | MCS_UNEVICTABLE, | |
3042 | NR_MCS_STAT, | |
3043 | }; | |
3044 | ||
3045 | struct mcs_total_stat { | |
3046 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
3047 | }; |
3048 | ||
14067bb3 KH |
3049 | struct { |
3050 | char *local_name; | |
3051 | char *total_name; | |
3052 | } memcg_stat_strings[NR_MCS_STAT] = { | |
3053 | {"cache", "total_cache"}, | |
3054 | {"rss", "total_rss"}, | |
d69b042f | 3055 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
3056 | {"pgpgin", "total_pgpgin"}, |
3057 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3058 | {"swap", "total_swap"}, |
14067bb3 KH |
3059 | {"inactive_anon", "total_inactive_anon"}, |
3060 | {"active_anon", "total_active_anon"}, | |
3061 | {"inactive_file", "total_inactive_file"}, | |
3062 | {"active_file", "total_active_file"}, | |
3063 | {"unevictable", "total_unevictable"} | |
3064 | }; | |
3065 | ||
3066 | ||
3067 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
3068 | { | |
3069 | struct mcs_total_stat *s = data; | |
3070 | s64 val; | |
3071 | ||
3072 | /* per cpu stat */ | |
3073 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE); | |
3074 | s->stat[MCS_CACHE] += val * PAGE_SIZE; | |
3075 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); | |
3076 | s->stat[MCS_RSS] += val * PAGE_SIZE; | |
d8046582 KH |
3077 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_FILE_MAPPED); |
3078 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; | |
14067bb3 KH |
3079 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT); |
3080 | s->stat[MCS_PGPGIN] += val; | |
3081 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); | |
3082 | s->stat[MCS_PGPGOUT] += val; | |
1dd3a273 DN |
3083 | if (do_swap_account) { |
3084 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_SWAPOUT); | |
3085 | s->stat[MCS_SWAP] += val * PAGE_SIZE; | |
3086 | } | |
14067bb3 KH |
3087 | |
3088 | /* per zone stat */ | |
3089 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
3090 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
3091 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
3092 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
3093 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
3094 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
3095 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
3096 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
3097 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
3098 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
3099 | return 0; | |
3100 | } | |
3101 | ||
3102 | static void | |
3103 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
3104 | { | |
3105 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
3106 | } | |
3107 | ||
c64745cf PM |
3108 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
3109 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 3110 | { |
d2ceb9b7 | 3111 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 3112 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
3113 | int i; |
3114 | ||
14067bb3 KH |
3115 | memset(&mystat, 0, sizeof(mystat)); |
3116 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 3117 | |
1dd3a273 DN |
3118 | for (i = 0; i < NR_MCS_STAT; i++) { |
3119 | if (i == MCS_SWAP && !do_swap_account) | |
3120 | continue; | |
14067bb3 | 3121 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 3122 | } |
7b854121 | 3123 | |
14067bb3 | 3124 | /* Hierarchical information */ |
fee7b548 KH |
3125 | { |
3126 | unsigned long long limit, memsw_limit; | |
3127 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
3128 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
3129 | if (do_swap_account) | |
3130 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
3131 | } | |
7f016ee8 | 3132 | |
14067bb3 KH |
3133 | memset(&mystat, 0, sizeof(mystat)); |
3134 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
3135 | for (i = 0; i < NR_MCS_STAT; i++) { |
3136 | if (i == MCS_SWAP && !do_swap_account) | |
3137 | continue; | |
14067bb3 | 3138 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 3139 | } |
14067bb3 | 3140 | |
7f016ee8 | 3141 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 3142 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
3143 | |
3144 | { | |
3145 | int nid, zid; | |
3146 | struct mem_cgroup_per_zone *mz; | |
3147 | unsigned long recent_rotated[2] = {0, 0}; | |
3148 | unsigned long recent_scanned[2] = {0, 0}; | |
3149 | ||
3150 | for_each_online_node(nid) | |
3151 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3152 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3153 | ||
3154 | recent_rotated[0] += | |
3155 | mz->reclaim_stat.recent_rotated[0]; | |
3156 | recent_rotated[1] += | |
3157 | mz->reclaim_stat.recent_rotated[1]; | |
3158 | recent_scanned[0] += | |
3159 | mz->reclaim_stat.recent_scanned[0]; | |
3160 | recent_scanned[1] += | |
3161 | mz->reclaim_stat.recent_scanned[1]; | |
3162 | } | |
3163 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3164 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3165 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3166 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3167 | } | |
3168 | #endif | |
3169 | ||
d2ceb9b7 KH |
3170 | return 0; |
3171 | } | |
3172 | ||
a7885eb8 KM |
3173 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
3174 | { | |
3175 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3176 | ||
3177 | return get_swappiness(memcg); | |
3178 | } | |
3179 | ||
3180 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3181 | u64 val) | |
3182 | { | |
3183 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3184 | struct mem_cgroup *parent; | |
068b38c1 | 3185 | |
a7885eb8 KM |
3186 | if (val > 100) |
3187 | return -EINVAL; | |
3188 | ||
3189 | if (cgrp->parent == NULL) | |
3190 | return -EINVAL; | |
3191 | ||
3192 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
3193 | |
3194 | cgroup_lock(); | |
3195 | ||
a7885eb8 KM |
3196 | /* If under hierarchy, only empty-root can set this value */ |
3197 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
3198 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3199 | cgroup_unlock(); | |
a7885eb8 | 3200 | return -EINVAL; |
068b38c1 | 3201 | } |
a7885eb8 KM |
3202 | |
3203 | spin_lock(&memcg->reclaim_param_lock); | |
3204 | memcg->swappiness = val; | |
3205 | spin_unlock(&memcg->reclaim_param_lock); | |
3206 | ||
068b38c1 LZ |
3207 | cgroup_unlock(); |
3208 | ||
a7885eb8 KM |
3209 | return 0; |
3210 | } | |
3211 | ||
c1e862c1 | 3212 | |
8cdea7c0 BS |
3213 | static struct cftype mem_cgroup_files[] = { |
3214 | { | |
0eea1030 | 3215 | .name = "usage_in_bytes", |
8c7c6e34 | 3216 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 3217 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3218 | }, |
c84872e1 PE |
3219 | { |
3220 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3221 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 3222 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
3223 | .read_u64 = mem_cgroup_read, |
3224 | }, | |
8cdea7c0 | 3225 | { |
0eea1030 | 3226 | .name = "limit_in_bytes", |
8c7c6e34 | 3227 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 3228 | .write_string = mem_cgroup_write, |
2c3daa72 | 3229 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3230 | }, |
296c81d8 BS |
3231 | { |
3232 | .name = "soft_limit_in_bytes", | |
3233 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
3234 | .write_string = mem_cgroup_write, | |
3235 | .read_u64 = mem_cgroup_read, | |
3236 | }, | |
8cdea7c0 BS |
3237 | { |
3238 | .name = "failcnt", | |
8c7c6e34 | 3239 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 3240 | .trigger = mem_cgroup_reset, |
2c3daa72 | 3241 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3242 | }, |
d2ceb9b7 KH |
3243 | { |
3244 | .name = "stat", | |
c64745cf | 3245 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 3246 | }, |
c1e862c1 KH |
3247 | { |
3248 | .name = "force_empty", | |
3249 | .trigger = mem_cgroup_force_empty_write, | |
3250 | }, | |
18f59ea7 BS |
3251 | { |
3252 | .name = "use_hierarchy", | |
3253 | .write_u64 = mem_cgroup_hierarchy_write, | |
3254 | .read_u64 = mem_cgroup_hierarchy_read, | |
3255 | }, | |
a7885eb8 KM |
3256 | { |
3257 | .name = "swappiness", | |
3258 | .read_u64 = mem_cgroup_swappiness_read, | |
3259 | .write_u64 = mem_cgroup_swappiness_write, | |
3260 | }, | |
7dc74be0 DN |
3261 | { |
3262 | .name = "move_charge_at_immigrate", | |
3263 | .read_u64 = mem_cgroup_move_charge_read, | |
3264 | .write_u64 = mem_cgroup_move_charge_write, | |
3265 | }, | |
8cdea7c0 BS |
3266 | }; |
3267 | ||
8c7c6e34 KH |
3268 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3269 | static struct cftype memsw_cgroup_files[] = { | |
3270 | { | |
3271 | .name = "memsw.usage_in_bytes", | |
3272 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
3273 | .read_u64 = mem_cgroup_read, | |
3274 | }, | |
3275 | { | |
3276 | .name = "memsw.max_usage_in_bytes", | |
3277 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
3278 | .trigger = mem_cgroup_reset, | |
3279 | .read_u64 = mem_cgroup_read, | |
3280 | }, | |
3281 | { | |
3282 | .name = "memsw.limit_in_bytes", | |
3283 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
3284 | .write_string = mem_cgroup_write, | |
3285 | .read_u64 = mem_cgroup_read, | |
3286 | }, | |
3287 | { | |
3288 | .name = "memsw.failcnt", | |
3289 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
3290 | .trigger = mem_cgroup_reset, | |
3291 | .read_u64 = mem_cgroup_read, | |
3292 | }, | |
3293 | }; | |
3294 | ||
3295 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3296 | { | |
3297 | if (!do_swap_account) | |
3298 | return 0; | |
3299 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
3300 | ARRAY_SIZE(memsw_cgroup_files)); | |
3301 | }; | |
3302 | #else | |
3303 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3304 | { | |
3305 | return 0; | |
3306 | } | |
3307 | #endif | |
3308 | ||
6d12e2d8 KH |
3309 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3310 | { | |
3311 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 3312 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 3313 | enum lru_list l; |
41e3355d | 3314 | int zone, tmp = node; |
1ecaab2b KH |
3315 | /* |
3316 | * This routine is called against possible nodes. | |
3317 | * But it's BUG to call kmalloc() against offline node. | |
3318 | * | |
3319 | * TODO: this routine can waste much memory for nodes which will | |
3320 | * never be onlined. It's better to use memory hotplug callback | |
3321 | * function. | |
3322 | */ | |
41e3355d KH |
3323 | if (!node_state(node, N_NORMAL_MEMORY)) |
3324 | tmp = -1; | |
3325 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
3326 | if (!pn) |
3327 | return 1; | |
1ecaab2b | 3328 | |
6d12e2d8 KH |
3329 | mem->info.nodeinfo[node] = pn; |
3330 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
3331 | |
3332 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3333 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
3334 | for_each_lru(l) |
3335 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 3336 | mz->usage_in_excess = 0; |
4e416953 BS |
3337 | mz->on_tree = false; |
3338 | mz->mem = mem; | |
1ecaab2b | 3339 | } |
6d12e2d8 KH |
3340 | return 0; |
3341 | } | |
3342 | ||
1ecaab2b KH |
3343 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3344 | { | |
3345 | kfree(mem->info.nodeinfo[node]); | |
3346 | } | |
3347 | ||
c8dad2bb JB |
3348 | static int mem_cgroup_size(void) |
3349 | { | |
3350 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); | |
3351 | return sizeof(struct mem_cgroup) + cpustat_size; | |
3352 | } | |
3353 | ||
33327948 KH |
3354 | static struct mem_cgroup *mem_cgroup_alloc(void) |
3355 | { | |
3356 | struct mem_cgroup *mem; | |
c8dad2bb | 3357 | int size = mem_cgroup_size(); |
33327948 | 3358 | |
c8dad2bb JB |
3359 | if (size < PAGE_SIZE) |
3360 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 3361 | else |
c8dad2bb | 3362 | mem = vmalloc(size); |
33327948 KH |
3363 | |
3364 | if (mem) | |
c8dad2bb | 3365 | memset(mem, 0, size); |
33327948 KH |
3366 | return mem; |
3367 | } | |
3368 | ||
8c7c6e34 KH |
3369 | /* |
3370 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
3371 | * (scanning all at force_empty is too costly...) | |
3372 | * | |
3373 | * Instead of clearing all references at force_empty, we remember | |
3374 | * the number of reference from swap_cgroup and free mem_cgroup when | |
3375 | * it goes down to 0. | |
3376 | * | |
8c7c6e34 KH |
3377 | * Removal of cgroup itself succeeds regardless of refs from swap. |
3378 | */ | |
3379 | ||
a7ba0eef | 3380 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 3381 | { |
08e552c6 KH |
3382 | int node; |
3383 | ||
f64c3f54 | 3384 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
3385 | free_css_id(&mem_cgroup_subsys, &mem->css); |
3386 | ||
08e552c6 KH |
3387 | for_each_node_state(node, N_POSSIBLE) |
3388 | free_mem_cgroup_per_zone_info(mem, node); | |
3389 | ||
c8dad2bb | 3390 | if (mem_cgroup_size() < PAGE_SIZE) |
33327948 KH |
3391 | kfree(mem); |
3392 | else | |
3393 | vfree(mem); | |
3394 | } | |
3395 | ||
8c7c6e34 KH |
3396 | static void mem_cgroup_get(struct mem_cgroup *mem) |
3397 | { | |
3398 | atomic_inc(&mem->refcnt); | |
3399 | } | |
3400 | ||
3401 | static void mem_cgroup_put(struct mem_cgroup *mem) | |
3402 | { | |
7bcc1bb1 DN |
3403 | if (atomic_dec_and_test(&mem->refcnt)) { |
3404 | struct mem_cgroup *parent = parent_mem_cgroup(mem); | |
a7ba0eef | 3405 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
3406 | if (parent) |
3407 | mem_cgroup_put(parent); | |
3408 | } | |
8c7c6e34 KH |
3409 | } |
3410 | ||
7bcc1bb1 DN |
3411 | /* |
3412 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
3413 | */ | |
3414 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
3415 | { | |
3416 | if (!mem->res.parent) | |
3417 | return NULL; | |
3418 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
3419 | } | |
33327948 | 3420 | |
c077719b KH |
3421 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3422 | static void __init enable_swap_cgroup(void) | |
3423 | { | |
f8d66542 | 3424 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
3425 | do_swap_account = 1; |
3426 | } | |
3427 | #else | |
3428 | static void __init enable_swap_cgroup(void) | |
3429 | { | |
3430 | } | |
3431 | #endif | |
3432 | ||
f64c3f54 BS |
3433 | static int mem_cgroup_soft_limit_tree_init(void) |
3434 | { | |
3435 | struct mem_cgroup_tree_per_node *rtpn; | |
3436 | struct mem_cgroup_tree_per_zone *rtpz; | |
3437 | int tmp, node, zone; | |
3438 | ||
3439 | for_each_node_state(node, N_POSSIBLE) { | |
3440 | tmp = node; | |
3441 | if (!node_state(node, N_NORMAL_MEMORY)) | |
3442 | tmp = -1; | |
3443 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
3444 | if (!rtpn) | |
3445 | return 1; | |
3446 | ||
3447 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
3448 | ||
3449 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3450 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
3451 | rtpz->rb_root = RB_ROOT; | |
3452 | spin_lock_init(&rtpz->lock); | |
3453 | } | |
3454 | } | |
3455 | return 0; | |
3456 | } | |
3457 | ||
0eb253e2 | 3458 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
3459 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
3460 | { | |
28dbc4b6 | 3461 | struct mem_cgroup *mem, *parent; |
04046e1a | 3462 | long error = -ENOMEM; |
6d12e2d8 | 3463 | int node; |
8cdea7c0 | 3464 | |
c8dad2bb JB |
3465 | mem = mem_cgroup_alloc(); |
3466 | if (!mem) | |
04046e1a | 3467 | return ERR_PTR(error); |
78fb7466 | 3468 | |
6d12e2d8 KH |
3469 | for_each_node_state(node, N_POSSIBLE) |
3470 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
3471 | goto free_out; | |
f64c3f54 | 3472 | |
c077719b | 3473 | /* root ? */ |
28dbc4b6 | 3474 | if (cont->parent == NULL) { |
cdec2e42 | 3475 | int cpu; |
c077719b | 3476 | enable_swap_cgroup(); |
28dbc4b6 | 3477 | parent = NULL; |
4b3bde4c | 3478 | root_mem_cgroup = mem; |
f64c3f54 BS |
3479 | if (mem_cgroup_soft_limit_tree_init()) |
3480 | goto free_out; | |
cdec2e42 KH |
3481 | for_each_possible_cpu(cpu) { |
3482 | struct memcg_stock_pcp *stock = | |
3483 | &per_cpu(memcg_stock, cpu); | |
3484 | INIT_WORK(&stock->work, drain_local_stock); | |
3485 | } | |
3486 | hotcpu_notifier(memcg_stock_cpu_callback, 0); | |
18f59ea7 | 3487 | } else { |
28dbc4b6 | 3488 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 BS |
3489 | mem->use_hierarchy = parent->use_hierarchy; |
3490 | } | |
28dbc4b6 | 3491 | |
18f59ea7 BS |
3492 | if (parent && parent->use_hierarchy) { |
3493 | res_counter_init(&mem->res, &parent->res); | |
3494 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
3495 | /* |
3496 | * We increment refcnt of the parent to ensure that we can | |
3497 | * safely access it on res_counter_charge/uncharge. | |
3498 | * This refcnt will be decremented when freeing this | |
3499 | * mem_cgroup(see mem_cgroup_put). | |
3500 | */ | |
3501 | mem_cgroup_get(parent); | |
18f59ea7 BS |
3502 | } else { |
3503 | res_counter_init(&mem->res, NULL); | |
3504 | res_counter_init(&mem->memsw, NULL); | |
3505 | } | |
04046e1a | 3506 | mem->last_scanned_child = 0; |
2733c06a | 3507 | spin_lock_init(&mem->reclaim_param_lock); |
6d61ef40 | 3508 | |
a7885eb8 KM |
3509 | if (parent) |
3510 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 3511 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 3512 | mem->move_charge_at_immigrate = 0; |
8cdea7c0 | 3513 | return &mem->css; |
6d12e2d8 | 3514 | free_out: |
a7ba0eef | 3515 | __mem_cgroup_free(mem); |
4b3bde4c | 3516 | root_mem_cgroup = NULL; |
04046e1a | 3517 | return ERR_PTR(error); |
8cdea7c0 BS |
3518 | } |
3519 | ||
ec64f515 | 3520 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
3521 | struct cgroup *cont) |
3522 | { | |
3523 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
3524 | |
3525 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
3526 | } |
3527 | ||
8cdea7c0 BS |
3528 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
3529 | struct cgroup *cont) | |
3530 | { | |
c268e994 | 3531 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 3532 | |
c268e994 | 3533 | mem_cgroup_put(mem); |
8cdea7c0 BS |
3534 | } |
3535 | ||
3536 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
3537 | struct cgroup *cont) | |
3538 | { | |
8c7c6e34 KH |
3539 | int ret; |
3540 | ||
3541 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
3542 | ARRAY_SIZE(mem_cgroup_files)); | |
3543 | ||
3544 | if (!ret) | |
3545 | ret = register_memsw_files(cont, ss); | |
3546 | return ret; | |
8cdea7c0 BS |
3547 | } |
3548 | ||
02491447 | 3549 | #ifdef CONFIG_MMU |
7dc74be0 | 3550 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
3551 | #define PRECHARGE_COUNT_AT_ONCE 256 |
3552 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 3553 | { |
854ffa8d DN |
3554 | int ret = 0; |
3555 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
3556 | struct mem_cgroup *mem = mc.to; |
3557 | ||
854ffa8d DN |
3558 | if (mem_cgroup_is_root(mem)) { |
3559 | mc.precharge += count; | |
3560 | /* we don't need css_get for root */ | |
3561 | return ret; | |
3562 | } | |
3563 | /* try to charge at once */ | |
3564 | if (count > 1) { | |
3565 | struct res_counter *dummy; | |
3566 | /* | |
3567 | * "mem" cannot be under rmdir() because we've already checked | |
3568 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
3569 | * are still under the same cgroup_mutex. So we can postpone | |
3570 | * css_get(). | |
3571 | */ | |
3572 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
3573 | goto one_by_one; | |
3574 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
3575 | PAGE_SIZE * count, &dummy)) { | |
3576 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
3577 | goto one_by_one; | |
3578 | } | |
3579 | mc.precharge += count; | |
3580 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
3581 | WARN_ON_ONCE(count > INT_MAX); | |
3582 | __css_get(&mem->css, (int)count); | |
3583 | return ret; | |
3584 | } | |
3585 | one_by_one: | |
3586 | /* fall back to one by one charge */ | |
3587 | while (count--) { | |
3588 | if (signal_pending(current)) { | |
3589 | ret = -EINTR; | |
3590 | break; | |
3591 | } | |
3592 | if (!batch_count--) { | |
3593 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
3594 | cond_resched(); | |
3595 | } | |
3596 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, | |
3597 | false, NULL); | |
3598 | if (ret || !mem) | |
3599 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
3600 | return -ENOMEM; | |
3601 | mc.precharge++; | |
3602 | } | |
4ffef5fe DN |
3603 | return ret; |
3604 | } | |
02491447 DN |
3605 | #else /* !CONFIG_MMU */ |
3606 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
3607 | struct cgroup *cgroup, | |
3608 | struct task_struct *p, | |
3609 | bool threadgroup) | |
3610 | { | |
3611 | return 0; | |
3612 | } | |
3613 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
3614 | struct cgroup *cgroup, | |
3615 | struct task_struct *p, | |
3616 | bool threadgroup) | |
3617 | { | |
3618 | } | |
3619 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
3620 | struct cgroup *cont, | |
3621 | struct cgroup *old_cont, | |
3622 | struct task_struct *p, | |
3623 | bool threadgroup) | |
3624 | { | |
3625 | } | |
3626 | #endif | |
4ffef5fe DN |
3627 | |
3628 | /** | |
3629 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
3630 | * @vma: the vma the pte to be checked belongs | |
3631 | * @addr: the address corresponding to the pte to be checked | |
3632 | * @ptent: the pte to be checked | |
02491447 | 3633 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
3634 | * |
3635 | * Returns | |
3636 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
3637 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
3638 | * move charge. if @target is not NULL, the page is stored in target->page | |
3639 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
3640 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
3641 | * target for charge migration. if @target is not NULL, the entry is stored | |
3642 | * in target->ent. | |
4ffef5fe DN |
3643 | * |
3644 | * Called with pte lock held. | |
3645 | */ | |
4ffef5fe DN |
3646 | union mc_target { |
3647 | struct page *page; | |
02491447 | 3648 | swp_entry_t ent; |
4ffef5fe DN |
3649 | }; |
3650 | ||
4ffef5fe DN |
3651 | enum mc_target_type { |
3652 | MC_TARGET_NONE, /* not used */ | |
3653 | MC_TARGET_PAGE, | |
02491447 | 3654 | MC_TARGET_SWAP, |
4ffef5fe DN |
3655 | }; |
3656 | ||
3657 | static int is_target_pte_for_mc(struct vm_area_struct *vma, | |
3658 | unsigned long addr, pte_t ptent, union mc_target *target) | |
3659 | { | |
02491447 | 3660 | struct page *page = NULL; |
4ffef5fe DN |
3661 | struct page_cgroup *pc; |
3662 | int ret = 0; | |
02491447 DN |
3663 | swp_entry_t ent = { .val = 0 }; |
3664 | int usage_count = 0; | |
4ffef5fe DN |
3665 | bool move_anon = test_bit(MOVE_CHARGE_TYPE_ANON, |
3666 | &mc.to->move_charge_at_immigrate); | |
3667 | ||
02491447 DN |
3668 | if (!pte_present(ptent)) { |
3669 | /* TODO: handle swap of shmes/tmpfs */ | |
3670 | if (pte_none(ptent) || pte_file(ptent)) | |
3671 | return 0; | |
3672 | else if (is_swap_pte(ptent)) { | |
3673 | ent = pte_to_swp_entry(ptent); | |
3674 | if (!move_anon || non_swap_entry(ent)) | |
3675 | return 0; | |
3676 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
3677 | } | |
3678 | } else { | |
3679 | page = vm_normal_page(vma, addr, ptent); | |
3680 | if (!page || !page_mapped(page)) | |
3681 | return 0; | |
3682 | /* | |
3683 | * TODO: We don't move charges of file(including shmem/tmpfs) | |
3684 | * pages for now. | |
3685 | */ | |
3686 | if (!move_anon || !PageAnon(page)) | |
3687 | return 0; | |
3688 | if (!get_page_unless_zero(page)) | |
3689 | return 0; | |
3690 | usage_count = page_mapcount(page); | |
3691 | } | |
3692 | if (usage_count > 1) { | |
3693 | /* | |
3694 | * TODO: We don't move charges of shared(used by multiple | |
3695 | * processes) pages for now. | |
3696 | */ | |
3697 | if (page) | |
3698 | put_page(page); | |
4ffef5fe | 3699 | return 0; |
02491447 DN |
3700 | } |
3701 | if (page) { | |
3702 | pc = lookup_page_cgroup(page); | |
3703 | /* | |
3704 | * Do only loose check w/o page_cgroup lock. | |
3705 | * mem_cgroup_move_account() checks the pc is valid or not under | |
3706 | * the lock. | |
3707 | */ | |
3708 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
3709 | ret = MC_TARGET_PAGE; | |
3710 | if (target) | |
3711 | target->page = page; | |
3712 | } | |
3713 | if (!ret || !target) | |
3714 | put_page(page); | |
3715 | } | |
3716 | /* throught */ | |
3717 | if (ent.val && do_swap_account && !ret && | |
3718 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { | |
3719 | ret = MC_TARGET_SWAP; | |
4ffef5fe | 3720 | if (target) |
02491447 | 3721 | target->ent = ent; |
4ffef5fe | 3722 | } |
4ffef5fe DN |
3723 | return ret; |
3724 | } | |
3725 | ||
3726 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
3727 | unsigned long addr, unsigned long end, | |
3728 | struct mm_walk *walk) | |
3729 | { | |
3730 | struct vm_area_struct *vma = walk->private; | |
3731 | pte_t *pte; | |
3732 | spinlock_t *ptl; | |
3733 | ||
3734 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
3735 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
3736 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
3737 | mc.precharge++; /* increment precharge temporarily */ | |
3738 | pte_unmap_unlock(pte - 1, ptl); | |
3739 | cond_resched(); | |
3740 | ||
7dc74be0 DN |
3741 | return 0; |
3742 | } | |
3743 | ||
4ffef5fe DN |
3744 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
3745 | { | |
3746 | unsigned long precharge; | |
3747 | struct vm_area_struct *vma; | |
3748 | ||
3749 | down_read(&mm->mmap_sem); | |
3750 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3751 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
3752 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
3753 | .mm = mm, | |
3754 | .private = vma, | |
3755 | }; | |
3756 | if (is_vm_hugetlb_page(vma)) | |
3757 | continue; | |
3758 | /* TODO: We don't move charges of shmem/tmpfs pages for now. */ | |
3759 | if (vma->vm_flags & VM_SHARED) | |
3760 | continue; | |
3761 | walk_page_range(vma->vm_start, vma->vm_end, | |
3762 | &mem_cgroup_count_precharge_walk); | |
3763 | } | |
3764 | up_read(&mm->mmap_sem); | |
3765 | ||
3766 | precharge = mc.precharge; | |
3767 | mc.precharge = 0; | |
3768 | ||
3769 | return precharge; | |
3770 | } | |
3771 | ||
4ffef5fe DN |
3772 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
3773 | { | |
854ffa8d | 3774 | return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm)); |
4ffef5fe DN |
3775 | } |
3776 | ||
3777 | static void mem_cgroup_clear_mc(void) | |
3778 | { | |
3779 | /* we must uncharge all the leftover precharges from mc.to */ | |
854ffa8d DN |
3780 | if (mc.precharge) { |
3781 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
3782 | mc.precharge = 0; | |
3783 | } | |
3784 | /* | |
3785 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
3786 | * we must uncharge here. | |
3787 | */ | |
3788 | if (mc.moved_charge) { | |
3789 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
3790 | mc.moved_charge = 0; | |
4ffef5fe DN |
3791 | } |
3792 | mc.from = NULL; | |
3793 | mc.to = NULL; | |
8033b97c DN |
3794 | mc.moving_task = NULL; |
3795 | wake_up_all(&mc.waitq); | |
4ffef5fe DN |
3796 | } |
3797 | ||
7dc74be0 DN |
3798 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
3799 | struct cgroup *cgroup, | |
3800 | struct task_struct *p, | |
3801 | bool threadgroup) | |
3802 | { | |
3803 | int ret = 0; | |
3804 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
3805 | ||
3806 | if (mem->move_charge_at_immigrate) { | |
3807 | struct mm_struct *mm; | |
3808 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
3809 | ||
3810 | VM_BUG_ON(from == mem); | |
3811 | ||
3812 | mm = get_task_mm(p); | |
3813 | if (!mm) | |
3814 | return 0; | |
7dc74be0 | 3815 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
3816 | if (mm->owner == p) { |
3817 | VM_BUG_ON(mc.from); | |
3818 | VM_BUG_ON(mc.to); | |
3819 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 3820 | VM_BUG_ON(mc.moved_charge); |
8033b97c | 3821 | VM_BUG_ON(mc.moving_task); |
4ffef5fe DN |
3822 | mc.from = from; |
3823 | mc.to = mem; | |
3824 | mc.precharge = 0; | |
854ffa8d | 3825 | mc.moved_charge = 0; |
8033b97c | 3826 | mc.moving_task = current; |
4ffef5fe DN |
3827 | |
3828 | ret = mem_cgroup_precharge_mc(mm); | |
3829 | if (ret) | |
3830 | mem_cgroup_clear_mc(); | |
3831 | } | |
7dc74be0 DN |
3832 | mmput(mm); |
3833 | } | |
3834 | return ret; | |
3835 | } | |
3836 | ||
3837 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
3838 | struct cgroup *cgroup, | |
3839 | struct task_struct *p, | |
3840 | bool threadgroup) | |
3841 | { | |
4ffef5fe | 3842 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
3843 | } |
3844 | ||
4ffef5fe DN |
3845 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
3846 | unsigned long addr, unsigned long end, | |
3847 | struct mm_walk *walk) | |
7dc74be0 | 3848 | { |
4ffef5fe DN |
3849 | int ret = 0; |
3850 | struct vm_area_struct *vma = walk->private; | |
3851 | pte_t *pte; | |
3852 | spinlock_t *ptl; | |
3853 | ||
3854 | retry: | |
3855 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
3856 | for (; addr != end; addr += PAGE_SIZE) { | |
3857 | pte_t ptent = *(pte++); | |
3858 | union mc_target target; | |
3859 | int type; | |
3860 | struct page *page; | |
3861 | struct page_cgroup *pc; | |
02491447 | 3862 | swp_entry_t ent; |
4ffef5fe DN |
3863 | |
3864 | if (!mc.precharge) | |
3865 | break; | |
3866 | ||
3867 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
3868 | switch (type) { | |
3869 | case MC_TARGET_PAGE: | |
3870 | page = target.page; | |
3871 | if (isolate_lru_page(page)) | |
3872 | goto put; | |
3873 | pc = lookup_page_cgroup(page); | |
854ffa8d DN |
3874 | if (!mem_cgroup_move_account(pc, |
3875 | mc.from, mc.to, false)) { | |
4ffef5fe | 3876 | mc.precharge--; |
854ffa8d DN |
3877 | /* we uncharge from mc.from later. */ |
3878 | mc.moved_charge++; | |
4ffef5fe DN |
3879 | } |
3880 | putback_lru_page(page); | |
3881 | put: /* is_target_pte_for_mc() gets the page */ | |
3882 | put_page(page); | |
3883 | break; | |
02491447 DN |
3884 | case MC_TARGET_SWAP: |
3885 | ent = target.ent; | |
3886 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) | |
3887 | mc.precharge--; | |
3888 | break; | |
4ffef5fe DN |
3889 | default: |
3890 | break; | |
3891 | } | |
3892 | } | |
3893 | pte_unmap_unlock(pte - 1, ptl); | |
3894 | cond_resched(); | |
3895 | ||
3896 | if (addr != end) { | |
3897 | /* | |
3898 | * We have consumed all precharges we got in can_attach(). | |
3899 | * We try charge one by one, but don't do any additional | |
3900 | * charges to mc.to if we have failed in charge once in attach() | |
3901 | * phase. | |
3902 | */ | |
854ffa8d | 3903 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
3904 | if (!ret) |
3905 | goto retry; | |
3906 | } | |
3907 | ||
3908 | return ret; | |
3909 | } | |
3910 | ||
3911 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
3912 | { | |
3913 | struct vm_area_struct *vma; | |
3914 | ||
3915 | lru_add_drain_all(); | |
3916 | down_read(&mm->mmap_sem); | |
3917 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
3918 | int ret; | |
3919 | struct mm_walk mem_cgroup_move_charge_walk = { | |
3920 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
3921 | .mm = mm, | |
3922 | .private = vma, | |
3923 | }; | |
3924 | if (is_vm_hugetlb_page(vma)) | |
3925 | continue; | |
3926 | /* TODO: We don't move charges of shmem/tmpfs pages for now. */ | |
3927 | if (vma->vm_flags & VM_SHARED) | |
3928 | continue; | |
3929 | ret = walk_page_range(vma->vm_start, vma->vm_end, | |
3930 | &mem_cgroup_move_charge_walk); | |
3931 | if (ret) | |
3932 | /* | |
3933 | * means we have consumed all precharges and failed in | |
3934 | * doing additional charge. Just abandon here. | |
3935 | */ | |
3936 | break; | |
3937 | } | |
3938 | up_read(&mm->mmap_sem); | |
7dc74be0 DN |
3939 | } |
3940 | ||
67e465a7 BS |
3941 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
3942 | struct cgroup *cont, | |
3943 | struct cgroup *old_cont, | |
be367d09 BB |
3944 | struct task_struct *p, |
3945 | bool threadgroup) | |
67e465a7 | 3946 | { |
4ffef5fe DN |
3947 | struct mm_struct *mm; |
3948 | ||
3949 | if (!mc.to) | |
3950 | /* no need to move charge */ | |
3951 | return; | |
3952 | ||
3953 | mm = get_task_mm(p); | |
3954 | if (mm) { | |
3955 | mem_cgroup_move_charge(mm); | |
3956 | mmput(mm); | |
3957 | } | |
3958 | mem_cgroup_clear_mc(); | |
67e465a7 BS |
3959 | } |
3960 | ||
8cdea7c0 BS |
3961 | struct cgroup_subsys mem_cgroup_subsys = { |
3962 | .name = "memory", | |
3963 | .subsys_id = mem_cgroup_subsys_id, | |
3964 | .create = mem_cgroup_create, | |
df878fb0 | 3965 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
3966 | .destroy = mem_cgroup_destroy, |
3967 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
3968 | .can_attach = mem_cgroup_can_attach, |
3969 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 3970 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 3971 | .early_init = 0, |
04046e1a | 3972 | .use_id = 1, |
8cdea7c0 | 3973 | }; |
c077719b KH |
3974 | |
3975 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3976 | ||
3977 | static int __init disable_swap_account(char *s) | |
3978 | { | |
3979 | really_do_swap_account = 0; | |
3980 | return 1; | |
3981 | } | |
3982 | __setup("noswapaccount", disable_swap_account); | |
3983 | #endif |