]>
Commit | Line | Data |
---|---|---|
8cdea7c0 BS |
1 | /* memcontrol.c - Memory Controller |
2 | * | |
3 | * Copyright IBM Corporation, 2007 | |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> | |
5 | * | |
78fb7466 PE |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> | |
8 | * | |
2e72b634 KS |
9 | * Memory thresholds |
10 | * Copyright (C) 2009 Nokia Corporation | |
11 | * Author: Kirill A. Shutemov | |
12 | * | |
8cdea7c0 BS |
13 | * This program is free software; you can redistribute it and/or modify |
14 | * it under the terms of the GNU General Public License as published by | |
15 | * the Free Software Foundation; either version 2 of the License, or | |
16 | * (at your option) any later version. | |
17 | * | |
18 | * This program is distributed in the hope that it will be useful, | |
19 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
20 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
21 | * GNU General Public License for more details. | |
22 | */ | |
23 | ||
24 | #include <linux/res_counter.h> | |
25 | #include <linux/memcontrol.h> | |
26 | #include <linux/cgroup.h> | |
78fb7466 | 27 | #include <linux/mm.h> |
4ffef5fe | 28 | #include <linux/hugetlb.h> |
d13d1443 | 29 | #include <linux/pagemap.h> |
d52aa412 | 30 | #include <linux/smp.h> |
8a9f3ccd | 31 | #include <linux/page-flags.h> |
66e1707b | 32 | #include <linux/backing-dev.h> |
8a9f3ccd BS |
33 | #include <linux/bit_spinlock.h> |
34 | #include <linux/rcupdate.h> | |
e222432b | 35 | #include <linux/limits.h> |
8c7c6e34 | 36 | #include <linux/mutex.h> |
f64c3f54 | 37 | #include <linux/rbtree.h> |
b6ac57d5 | 38 | #include <linux/slab.h> |
66e1707b | 39 | #include <linux/swap.h> |
02491447 | 40 | #include <linux/swapops.h> |
66e1707b | 41 | #include <linux/spinlock.h> |
2e72b634 KS |
42 | #include <linux/eventfd.h> |
43 | #include <linux/sort.h> | |
66e1707b | 44 | #include <linux/fs.h> |
d2ceb9b7 | 45 | #include <linux/seq_file.h> |
33327948 | 46 | #include <linux/vmalloc.h> |
b69408e8 | 47 | #include <linux/mm_inline.h> |
52d4b9ac | 48 | #include <linux/page_cgroup.h> |
cdec2e42 | 49 | #include <linux/cpu.h> |
158e0a2d | 50 | #include <linux/oom.h> |
08e552c6 | 51 | #include "internal.h" |
8cdea7c0 | 52 | |
8697d331 BS |
53 | #include <asm/uaccess.h> |
54 | ||
cc8e970c KM |
55 | #include <trace/events/vmscan.h> |
56 | ||
a181b0e8 | 57 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 58 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 59 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 60 | |
c077719b | 61 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 62 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b | 63 | int do_swap_account __read_mostly; |
a42c390c MH |
64 | |
65 | /* for remember boot option*/ | |
66 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED | |
67 | static int really_do_swap_account __initdata = 1; | |
68 | #else | |
69 | static int really_do_swap_account __initdata = 0; | |
70 | #endif | |
71 | ||
c077719b KH |
72 | #else |
73 | #define do_swap_account (0) | |
74 | #endif | |
75 | ||
d2265e6f KH |
76 | /* |
77 | * Per memcg event counter is incremented at every pagein/pageout. This counter | |
78 | * is used for trigger some periodic events. This is straightforward and better | |
79 | * than using jiffies etc. to handle periodic memcg event. | |
80 | * | |
81 | * These values will be used as !((event) & ((1 <<(thresh)) - 1)) | |
82 | */ | |
83 | #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */ | |
84 | #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */ | |
c077719b | 85 | |
d52aa412 KH |
86 | /* |
87 | * Statistics for memory cgroup. | |
88 | */ | |
89 | enum mem_cgroup_stat_index { | |
90 | /* | |
91 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
92 | */ | |
93 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 94 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 95 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
55e462b0 BR |
96 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
97 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
0c3e73e8 | 98 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
711d3d2c KH |
99 | MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ |
100 | /* incremented at every pagein/pageout */ | |
101 | MEM_CGROUP_EVENTS = MEM_CGROUP_STAT_DATA, | |
32047e2a | 102 | MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */ |
d52aa412 KH |
103 | |
104 | MEM_CGROUP_STAT_NSTATS, | |
105 | }; | |
106 | ||
107 | struct mem_cgroup_stat_cpu { | |
108 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
d52aa412 KH |
109 | }; |
110 | ||
6d12e2d8 KH |
111 | /* |
112 | * per-zone information in memory controller. | |
113 | */ | |
6d12e2d8 | 114 | struct mem_cgroup_per_zone { |
072c56c1 KH |
115 | /* |
116 | * spin_lock to protect the per cgroup LRU | |
117 | */ | |
b69408e8 CL |
118 | struct list_head lists[NR_LRU_LISTS]; |
119 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
120 | |
121 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
122 | struct rb_node tree_node; /* RB tree node */ |
123 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
124 | /* the soft limit is exceeded*/ | |
125 | bool on_tree; | |
4e416953 BS |
126 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
127 | /* use container_of */ | |
6d12e2d8 KH |
128 | }; |
129 | /* Macro for accessing counter */ | |
130 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
131 | ||
132 | struct mem_cgroup_per_node { | |
133 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
134 | }; | |
135 | ||
136 | struct mem_cgroup_lru_info { | |
137 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
138 | }; | |
139 | ||
f64c3f54 BS |
140 | /* |
141 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
142 | * their hierarchy representation | |
143 | */ | |
144 | ||
145 | struct mem_cgroup_tree_per_zone { | |
146 | struct rb_root rb_root; | |
147 | spinlock_t lock; | |
148 | }; | |
149 | ||
150 | struct mem_cgroup_tree_per_node { | |
151 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
152 | }; | |
153 | ||
154 | struct mem_cgroup_tree { | |
155 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
156 | }; | |
157 | ||
158 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
159 | ||
2e72b634 KS |
160 | struct mem_cgroup_threshold { |
161 | struct eventfd_ctx *eventfd; | |
162 | u64 threshold; | |
163 | }; | |
164 | ||
9490ff27 | 165 | /* For threshold */ |
2e72b634 KS |
166 | struct mem_cgroup_threshold_ary { |
167 | /* An array index points to threshold just below usage. */ | |
5407a562 | 168 | int current_threshold; |
2e72b634 KS |
169 | /* Size of entries[] */ |
170 | unsigned int size; | |
171 | /* Array of thresholds */ | |
172 | struct mem_cgroup_threshold entries[0]; | |
173 | }; | |
2c488db2 KS |
174 | |
175 | struct mem_cgroup_thresholds { | |
176 | /* Primary thresholds array */ | |
177 | struct mem_cgroup_threshold_ary *primary; | |
178 | /* | |
179 | * Spare threshold array. | |
180 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
181 | * It must be able to store at least primary->size - 1 entries. | |
182 | */ | |
183 | struct mem_cgroup_threshold_ary *spare; | |
184 | }; | |
185 | ||
9490ff27 KH |
186 | /* for OOM */ |
187 | struct mem_cgroup_eventfd_list { | |
188 | struct list_head list; | |
189 | struct eventfd_ctx *eventfd; | |
190 | }; | |
2e72b634 | 191 | |
2e72b634 | 192 | static void mem_cgroup_threshold(struct mem_cgroup *mem); |
9490ff27 | 193 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem); |
2e72b634 | 194 | |
8cdea7c0 BS |
195 | /* |
196 | * The memory controller data structure. The memory controller controls both | |
197 | * page cache and RSS per cgroup. We would eventually like to provide | |
198 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
199 | * to help the administrator determine what knobs to tune. | |
200 | * | |
201 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
202 | * we hit the water mark. May be even add a low water mark, such that |
203 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
204 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
205 | */ |
206 | struct mem_cgroup { | |
207 | struct cgroup_subsys_state css; | |
208 | /* | |
209 | * the counter to account for memory usage | |
210 | */ | |
211 | struct res_counter res; | |
8c7c6e34 KH |
212 | /* |
213 | * the counter to account for mem+swap usage. | |
214 | */ | |
215 | struct res_counter memsw; | |
78fb7466 PE |
216 | /* |
217 | * Per cgroup active and inactive list, similar to the | |
218 | * per zone LRU lists. | |
78fb7466 | 219 | */ |
6d12e2d8 | 220 | struct mem_cgroup_lru_info info; |
072c56c1 | 221 | |
2733c06a KM |
222 | /* |
223 | protect against reclaim related member. | |
224 | */ | |
225 | spinlock_t reclaim_param_lock; | |
226 | ||
6d61ef40 | 227 | /* |
af901ca1 | 228 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 229 | * reclaimed from. |
6d61ef40 | 230 | */ |
04046e1a | 231 | int last_scanned_child; |
18f59ea7 BS |
232 | /* |
233 | * Should the accounting and control be hierarchical, per subtree? | |
234 | */ | |
235 | bool use_hierarchy; | |
867578cb | 236 | atomic_t oom_lock; |
8c7c6e34 | 237 | atomic_t refcnt; |
14797e23 | 238 | |
a7885eb8 | 239 | unsigned int swappiness; |
3c11ecf4 KH |
240 | /* OOM-Killer disable */ |
241 | int oom_kill_disable; | |
a7885eb8 | 242 | |
22a668d7 KH |
243 | /* set when res.limit == memsw.limit */ |
244 | bool memsw_is_minimum; | |
245 | ||
2e72b634 KS |
246 | /* protect arrays of thresholds */ |
247 | struct mutex thresholds_lock; | |
248 | ||
249 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 250 | struct mem_cgroup_thresholds thresholds; |
907860ed | 251 | |
2e72b634 | 252 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 253 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 254 | |
9490ff27 KH |
255 | /* For oom notifier event fd */ |
256 | struct list_head oom_notify; | |
257 | ||
7dc74be0 DN |
258 | /* |
259 | * Should we move charges of a task when a task is moved into this | |
260 | * mem_cgroup ? And what type of charges should we move ? | |
261 | */ | |
262 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 263 | /* |
c62b1a3b | 264 | * percpu counter. |
d52aa412 | 265 | */ |
c62b1a3b | 266 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
267 | /* |
268 | * used when a cpu is offlined or other synchronizations | |
269 | * See mem_cgroup_read_stat(). | |
270 | */ | |
271 | struct mem_cgroup_stat_cpu nocpu_base; | |
272 | spinlock_t pcp_counter_lock; | |
8cdea7c0 BS |
273 | }; |
274 | ||
7dc74be0 DN |
275 | /* Stuffs for move charges at task migration. */ |
276 | /* | |
277 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
278 | * left-shifted bitmap of these types. | |
279 | */ | |
280 | enum move_type { | |
4ffef5fe | 281 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 282 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
283 | NR_MOVE_TYPE, |
284 | }; | |
285 | ||
4ffef5fe DN |
286 | /* "mc" and its members are protected by cgroup_mutex */ |
287 | static struct move_charge_struct { | |
b1dd693e | 288 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
289 | struct mem_cgroup *from; |
290 | struct mem_cgroup *to; | |
291 | unsigned long precharge; | |
854ffa8d | 292 | unsigned long moved_charge; |
483c30b5 | 293 | unsigned long moved_swap; |
8033b97c DN |
294 | struct task_struct *moving_task; /* a task moving charges */ |
295 | wait_queue_head_t waitq; /* a waitq for other context */ | |
296 | } mc = { | |
2bd9bb20 | 297 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
298 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
299 | }; | |
4ffef5fe | 300 | |
90254a65 DN |
301 | static bool move_anon(void) |
302 | { | |
303 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
304 | &mc.to->move_charge_at_immigrate); | |
305 | } | |
306 | ||
87946a72 DN |
307 | static bool move_file(void) |
308 | { | |
309 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
310 | &mc.to->move_charge_at_immigrate); | |
311 | } | |
312 | ||
4e416953 BS |
313 | /* |
314 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
315 | * limit reclaim to prevent infinite loops, if they ever occur. | |
316 | */ | |
317 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
318 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
319 | ||
217bc319 KH |
320 | enum charge_type { |
321 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
322 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 323 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 324 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 325 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 326 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
327 | NR_CHARGE_TYPE, |
328 | }; | |
329 | ||
52d4b9ac KH |
330 | /* only for here (for easy reading.) */ |
331 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
332 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 333 | #define PCGF_LOCK (1UL << PCG_LOCK) |
4b3bde4c BS |
334 | /* Not used, but added here for completeness */ |
335 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
217bc319 | 336 | |
8c7c6e34 KH |
337 | /* for encoding cft->private value on file */ |
338 | #define _MEM (0) | |
339 | #define _MEMSWAP (1) | |
9490ff27 | 340 | #define _OOM_TYPE (2) |
8c7c6e34 KH |
341 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
342 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
343 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
344 | /* Used for OOM nofiier */ |
345 | #define OOM_CONTROL (0) | |
8c7c6e34 | 346 | |
75822b44 BS |
347 | /* |
348 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
349 | */ | |
350 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
351 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
352 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
353 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
354 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
355 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 356 | |
8c7c6e34 KH |
357 | static void mem_cgroup_get(struct mem_cgroup *mem); |
358 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 359 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 360 | static void drain_all_stock_async(void); |
8c7c6e34 | 361 | |
f64c3f54 BS |
362 | static struct mem_cgroup_per_zone * |
363 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
364 | { | |
365 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
366 | } | |
367 | ||
d324236b WF |
368 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
369 | { | |
370 | return &mem->css; | |
371 | } | |
372 | ||
f64c3f54 BS |
373 | static struct mem_cgroup_per_zone * |
374 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
375 | { | |
376 | struct mem_cgroup *mem = pc->mem_cgroup; | |
377 | int nid = page_cgroup_nid(pc); | |
378 | int zid = page_cgroup_zid(pc); | |
379 | ||
380 | if (!mem) | |
381 | return NULL; | |
382 | ||
383 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
384 | } | |
385 | ||
386 | static struct mem_cgroup_tree_per_zone * | |
387 | soft_limit_tree_node_zone(int nid, int zid) | |
388 | { | |
389 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
390 | } | |
391 | ||
392 | static struct mem_cgroup_tree_per_zone * | |
393 | soft_limit_tree_from_page(struct page *page) | |
394 | { | |
395 | int nid = page_to_nid(page); | |
396 | int zid = page_zonenum(page); | |
397 | ||
398 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
399 | } | |
400 | ||
401 | static void | |
4e416953 | 402 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 403 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
404 | struct mem_cgroup_tree_per_zone *mctz, |
405 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
406 | { |
407 | struct rb_node **p = &mctz->rb_root.rb_node; | |
408 | struct rb_node *parent = NULL; | |
409 | struct mem_cgroup_per_zone *mz_node; | |
410 | ||
411 | if (mz->on_tree) | |
412 | return; | |
413 | ||
ef8745c1 KH |
414 | mz->usage_in_excess = new_usage_in_excess; |
415 | if (!mz->usage_in_excess) | |
416 | return; | |
f64c3f54 BS |
417 | while (*p) { |
418 | parent = *p; | |
419 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
420 | tree_node); | |
421 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
422 | p = &(*p)->rb_left; | |
423 | /* | |
424 | * We can't avoid mem cgroups that are over their soft | |
425 | * limit by the same amount | |
426 | */ | |
427 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
428 | p = &(*p)->rb_right; | |
429 | } | |
430 | rb_link_node(&mz->tree_node, parent, p); | |
431 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
432 | mz->on_tree = true; | |
4e416953 BS |
433 | } |
434 | ||
435 | static void | |
436 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
437 | struct mem_cgroup_per_zone *mz, | |
438 | struct mem_cgroup_tree_per_zone *mctz) | |
439 | { | |
440 | if (!mz->on_tree) | |
441 | return; | |
442 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
443 | mz->on_tree = false; | |
444 | } | |
445 | ||
f64c3f54 BS |
446 | static void |
447 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
448 | struct mem_cgroup_per_zone *mz, | |
449 | struct mem_cgroup_tree_per_zone *mctz) | |
450 | { | |
451 | spin_lock(&mctz->lock); | |
4e416953 | 452 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
453 | spin_unlock(&mctz->lock); |
454 | } | |
455 | ||
f64c3f54 BS |
456 | |
457 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
458 | { | |
ef8745c1 | 459 | unsigned long long excess; |
f64c3f54 BS |
460 | struct mem_cgroup_per_zone *mz; |
461 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
462 | int nid = page_to_nid(page); |
463 | int zid = page_zonenum(page); | |
f64c3f54 BS |
464 | mctz = soft_limit_tree_from_page(page); |
465 | ||
466 | /* | |
4e649152 KH |
467 | * Necessary to update all ancestors when hierarchy is used. |
468 | * because their event counter is not touched. | |
f64c3f54 | 469 | */ |
4e649152 KH |
470 | for (; mem; mem = parent_mem_cgroup(mem)) { |
471 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 472 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
473 | /* |
474 | * We have to update the tree if mz is on RB-tree or | |
475 | * mem is over its softlimit. | |
476 | */ | |
ef8745c1 | 477 | if (excess || mz->on_tree) { |
4e649152 KH |
478 | spin_lock(&mctz->lock); |
479 | /* if on-tree, remove it */ | |
480 | if (mz->on_tree) | |
481 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
482 | /* | |
ef8745c1 KH |
483 | * Insert again. mz->usage_in_excess will be updated. |
484 | * If excess is 0, no tree ops. | |
4e649152 | 485 | */ |
ef8745c1 | 486 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
487 | spin_unlock(&mctz->lock); |
488 | } | |
f64c3f54 BS |
489 | } |
490 | } | |
491 | ||
492 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
493 | { | |
494 | int node, zone; | |
495 | struct mem_cgroup_per_zone *mz; | |
496 | struct mem_cgroup_tree_per_zone *mctz; | |
497 | ||
498 | for_each_node_state(node, N_POSSIBLE) { | |
499 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
500 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
501 | mctz = soft_limit_tree_node_zone(node, zone); | |
502 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
503 | } | |
504 | } | |
505 | } | |
506 | ||
4e416953 BS |
507 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
508 | { | |
509 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; | |
510 | } | |
511 | ||
512 | static struct mem_cgroup_per_zone * | |
513 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
514 | { | |
515 | struct rb_node *rightmost = NULL; | |
26251eaf | 516 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
517 | |
518 | retry: | |
26251eaf | 519 | mz = NULL; |
4e416953 BS |
520 | rightmost = rb_last(&mctz->rb_root); |
521 | if (!rightmost) | |
522 | goto done; /* Nothing to reclaim from */ | |
523 | ||
524 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
525 | /* | |
526 | * Remove the node now but someone else can add it back, | |
527 | * we will to add it back at the end of reclaim to its correct | |
528 | * position in the tree. | |
529 | */ | |
530 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
531 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
532 | !css_tryget(&mz->mem->css)) | |
533 | goto retry; | |
534 | done: | |
535 | return mz; | |
536 | } | |
537 | ||
538 | static struct mem_cgroup_per_zone * | |
539 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
540 | { | |
541 | struct mem_cgroup_per_zone *mz; | |
542 | ||
543 | spin_lock(&mctz->lock); | |
544 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
545 | spin_unlock(&mctz->lock); | |
546 | return mz; | |
547 | } | |
548 | ||
711d3d2c KH |
549 | /* |
550 | * Implementation Note: reading percpu statistics for memcg. | |
551 | * | |
552 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
553 | * synchronization to implement "quick" read. There are trade-off between | |
554 | * reading cost and precision of value. Then, we may have a chance to implement | |
555 | * a periodic synchronizion of counter in memcg's counter. | |
556 | * | |
557 | * But this _read() function is used for user interface now. The user accounts | |
558 | * memory usage by memory cgroup and he _always_ requires exact value because | |
559 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
560 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
561 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
562 | * | |
563 | * If there are kernel internal actions which can make use of some not-exact | |
564 | * value, and reading all cpu value can be performance bottleneck in some | |
565 | * common workload, threashold and synchonization as vmstat[] should be | |
566 | * implemented. | |
567 | */ | |
c62b1a3b KH |
568 | static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, |
569 | enum mem_cgroup_stat_index idx) | |
570 | { | |
571 | int cpu; | |
572 | s64 val = 0; | |
573 | ||
711d3d2c KH |
574 | get_online_cpus(); |
575 | for_each_online_cpu(cpu) | |
c62b1a3b | 576 | val += per_cpu(mem->stat->count[idx], cpu); |
711d3d2c KH |
577 | #ifdef CONFIG_HOTPLUG_CPU |
578 | spin_lock(&mem->pcp_counter_lock); | |
579 | val += mem->nocpu_base.count[idx]; | |
580 | spin_unlock(&mem->pcp_counter_lock); | |
581 | #endif | |
582 | put_online_cpus(); | |
c62b1a3b KH |
583 | return val; |
584 | } | |
585 | ||
586 | static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) | |
587 | { | |
588 | s64 ret; | |
589 | ||
590 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); | |
591 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); | |
592 | return ret; | |
593 | } | |
594 | ||
0c3e73e8 BS |
595 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
596 | bool charge) | |
597 | { | |
598 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 599 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
600 | } |
601 | ||
c05555b5 | 602 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
e401f176 | 603 | bool file, int nr_pages) |
d52aa412 | 604 | { |
c62b1a3b KH |
605 | preempt_disable(); |
606 | ||
e401f176 KH |
607 | if (file) |
608 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages); | |
d52aa412 | 609 | else |
e401f176 | 610 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages); |
55e462b0 | 611 | |
e401f176 KH |
612 | /* pagein of a big page is an event. So, ignore page size */ |
613 | if (nr_pages > 0) | |
c62b1a3b | 614 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); |
55e462b0 | 615 | else |
c62b1a3b | 616 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); |
e401f176 KH |
617 | |
618 | __this_cpu_add(mem->stat->count[MEM_CGROUP_EVENTS], nr_pages); | |
2e72b634 | 619 | |
c62b1a3b | 620 | preempt_enable(); |
6d12e2d8 KH |
621 | } |
622 | ||
14067bb3 | 623 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 624 | enum lru_list idx) |
6d12e2d8 KH |
625 | { |
626 | int nid, zid; | |
627 | struct mem_cgroup_per_zone *mz; | |
628 | u64 total = 0; | |
629 | ||
630 | for_each_online_node(nid) | |
631 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
632 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
633 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
634 | } | |
635 | return total; | |
d52aa412 KH |
636 | } |
637 | ||
d2265e6f KH |
638 | static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift) |
639 | { | |
640 | s64 val; | |
641 | ||
642 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]); | |
643 | ||
644 | return !(val & ((1 << event_mask_shift) - 1)); | |
645 | } | |
646 | ||
647 | /* | |
648 | * Check events in order. | |
649 | * | |
650 | */ | |
651 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) | |
652 | { | |
653 | /* threshold event is triggered in finer grain than soft limit */ | |
654 | if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) { | |
655 | mem_cgroup_threshold(mem); | |
656 | if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH))) | |
657 | mem_cgroup_update_tree(mem, page); | |
658 | } | |
659 | } | |
660 | ||
d5b69e38 | 661 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
662 | { |
663 | return container_of(cgroup_subsys_state(cont, | |
664 | mem_cgroup_subsys_id), struct mem_cgroup, | |
665 | css); | |
666 | } | |
667 | ||
cf475ad2 | 668 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 669 | { |
31a78f23 BS |
670 | /* |
671 | * mm_update_next_owner() may clear mm->owner to NULL | |
672 | * if it races with swapoff, page migration, etc. | |
673 | * So this can be called with p == NULL. | |
674 | */ | |
675 | if (unlikely(!p)) | |
676 | return NULL; | |
677 | ||
78fb7466 PE |
678 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
679 | struct mem_cgroup, css); | |
680 | } | |
681 | ||
54595fe2 KH |
682 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
683 | { | |
684 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
685 | |
686 | if (!mm) | |
687 | return NULL; | |
54595fe2 KH |
688 | /* |
689 | * Because we have no locks, mm->owner's may be being moved to other | |
690 | * cgroup. We use css_tryget() here even if this looks | |
691 | * pessimistic (rather than adding locks here). | |
692 | */ | |
693 | rcu_read_lock(); | |
694 | do { | |
695 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
696 | if (unlikely(!mem)) | |
697 | break; | |
698 | } while (!css_tryget(&mem->css)); | |
699 | rcu_read_unlock(); | |
700 | return mem; | |
701 | } | |
702 | ||
7d74b06f KH |
703 | /* The caller has to guarantee "mem" exists before calling this */ |
704 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem) | |
14067bb3 | 705 | { |
711d3d2c KH |
706 | struct cgroup_subsys_state *css; |
707 | int found; | |
708 | ||
709 | if (!mem) /* ROOT cgroup has the smallest ID */ | |
710 | return root_mem_cgroup; /*css_put/get against root is ignored*/ | |
711 | if (!mem->use_hierarchy) { | |
712 | if (css_tryget(&mem->css)) | |
713 | return mem; | |
714 | return NULL; | |
715 | } | |
716 | rcu_read_lock(); | |
717 | /* | |
718 | * searching a memory cgroup which has the smallest ID under given | |
719 | * ROOT cgroup. (ID >= 1) | |
720 | */ | |
721 | css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found); | |
722 | if (css && css_tryget(css)) | |
723 | mem = container_of(css, struct mem_cgroup, css); | |
724 | else | |
725 | mem = NULL; | |
726 | rcu_read_unlock(); | |
727 | return mem; | |
7d74b06f KH |
728 | } |
729 | ||
730 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, | |
731 | struct mem_cgroup *root, | |
732 | bool cond) | |
733 | { | |
734 | int nextid = css_id(&iter->css) + 1; | |
735 | int found; | |
736 | int hierarchy_used; | |
14067bb3 | 737 | struct cgroup_subsys_state *css; |
14067bb3 | 738 | |
7d74b06f | 739 | hierarchy_used = iter->use_hierarchy; |
14067bb3 | 740 | |
7d74b06f | 741 | css_put(&iter->css); |
711d3d2c KH |
742 | /* If no ROOT, walk all, ignore hierarchy */ |
743 | if (!cond || (root && !hierarchy_used)) | |
7d74b06f | 744 | return NULL; |
14067bb3 | 745 | |
711d3d2c KH |
746 | if (!root) |
747 | root = root_mem_cgroup; | |
748 | ||
7d74b06f KH |
749 | do { |
750 | iter = NULL; | |
14067bb3 | 751 | rcu_read_lock(); |
7d74b06f KH |
752 | |
753 | css = css_get_next(&mem_cgroup_subsys, nextid, | |
754 | &root->css, &found); | |
14067bb3 | 755 | if (css && css_tryget(css)) |
7d74b06f | 756 | iter = container_of(css, struct mem_cgroup, css); |
14067bb3 | 757 | rcu_read_unlock(); |
7d74b06f | 758 | /* If css is NULL, no more cgroups will be found */ |
14067bb3 | 759 | nextid = found + 1; |
7d74b06f | 760 | } while (css && !iter); |
14067bb3 | 761 | |
7d74b06f | 762 | return iter; |
14067bb3 | 763 | } |
7d74b06f KH |
764 | /* |
765 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please | |
766 | * be careful that "break" loop is not allowed. We have reference count. | |
767 | * Instead of that modify "cond" to be false and "continue" to exit the loop. | |
768 | */ | |
769 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ | |
770 | for (iter = mem_cgroup_start_loop(root);\ | |
771 | iter != NULL;\ | |
772 | iter = mem_cgroup_get_next(iter, root, cond)) | |
773 | ||
774 | #define for_each_mem_cgroup_tree(iter, root) \ | |
775 | for_each_mem_cgroup_tree_cond(iter, root, true) | |
776 | ||
711d3d2c KH |
777 | #define for_each_mem_cgroup_all(iter) \ |
778 | for_each_mem_cgroup_tree_cond(iter, NULL, true) | |
779 | ||
14067bb3 | 780 | |
4b3bde4c BS |
781 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
782 | { | |
783 | return (mem == root_mem_cgroup); | |
784 | } | |
785 | ||
08e552c6 KH |
786 | /* |
787 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
788 | * Operations are called by routine of global LRU independently from memcg. | |
789 | * What we have to take care of here is validness of pc->mem_cgroup. | |
790 | * | |
791 | * Changes to pc->mem_cgroup happens when | |
792 | * 1. charge | |
793 | * 2. moving account | |
794 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
795 | * It is added to LRU before charge. | |
796 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
797 | * When moving account, the page is not on LRU. It's isolated. | |
798 | */ | |
4f98a2fe | 799 | |
08e552c6 KH |
800 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
801 | { | |
802 | struct page_cgroup *pc; | |
08e552c6 | 803 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 804 | |
f8d66542 | 805 | if (mem_cgroup_disabled()) |
08e552c6 KH |
806 | return; |
807 | pc = lookup_page_cgroup(page); | |
808 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 809 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 810 | return; |
4b3bde4c | 811 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
812 | /* |
813 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
814 | * removed from global LRU. | |
815 | */ | |
08e552c6 | 816 | mz = page_cgroup_zoneinfo(pc); |
ece35ca8 KH |
817 | /* huge page split is done under lru_lock. so, we have no races. */ |
818 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); | |
4b3bde4c BS |
819 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
820 | return; | |
821 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 | 822 | list_del_init(&pc->lru); |
6d12e2d8 KH |
823 | } |
824 | ||
08e552c6 | 825 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 826 | { |
08e552c6 KH |
827 | mem_cgroup_del_lru_list(page, page_lru(page)); |
828 | } | |
b69408e8 | 829 | |
08e552c6 KH |
830 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
831 | { | |
832 | struct mem_cgroup_per_zone *mz; | |
833 | struct page_cgroup *pc; | |
b69408e8 | 834 | |
f8d66542 | 835 | if (mem_cgroup_disabled()) |
08e552c6 | 836 | return; |
6d12e2d8 | 837 | |
08e552c6 | 838 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
839 | /* |
840 | * Used bit is set without atomic ops but after smp_wmb(). | |
841 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
842 | */ | |
08e552c6 | 843 | smp_rmb(); |
4b3bde4c BS |
844 | /* unused or root page is not rotated. */ |
845 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 KH |
846 | return; |
847 | mz = page_cgroup_zoneinfo(pc); | |
848 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
849 | } |
850 | ||
08e552c6 | 851 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 852 | { |
08e552c6 KH |
853 | struct page_cgroup *pc; |
854 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 855 | |
f8d66542 | 856 | if (mem_cgroup_disabled()) |
08e552c6 KH |
857 | return; |
858 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 859 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
bd112db8 DN |
860 | /* |
861 | * Used bit is set without atomic ops but after smp_wmb(). | |
862 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
863 | */ | |
08e552c6 KH |
864 | smp_rmb(); |
865 | if (!PageCgroupUsed(pc)) | |
894bc310 | 866 | return; |
b69408e8 | 867 | |
08e552c6 | 868 | mz = page_cgroup_zoneinfo(pc); |
ece35ca8 KH |
869 | /* huge page split is done under lru_lock. so, we have no races. */ |
870 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); | |
4b3bde4c BS |
871 | SetPageCgroupAcctLRU(pc); |
872 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
873 | return; | |
08e552c6 KH |
874 | list_add(&pc->lru, &mz->lists[lru]); |
875 | } | |
544122e5 | 876 | |
08e552c6 | 877 | /* |
544122e5 KH |
878 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
879 | * lru because the page may.be reused after it's fully uncharged (because of | |
880 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
881 | * it again. This function is only used to charge SwapCache. It's done under | |
882 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 883 | */ |
544122e5 | 884 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 885 | { |
544122e5 KH |
886 | unsigned long flags; |
887 | struct zone *zone = page_zone(page); | |
888 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
889 | ||
890 | spin_lock_irqsave(&zone->lru_lock, flags); | |
891 | /* | |
892 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
893 | * is guarded by lock_page() because the page is SwapCache. | |
894 | */ | |
895 | if (!PageCgroupUsed(pc)) | |
896 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
897 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
898 | } |
899 | ||
544122e5 KH |
900 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
901 | { | |
902 | unsigned long flags; | |
903 | struct zone *zone = page_zone(page); | |
904 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
905 | ||
906 | spin_lock_irqsave(&zone->lru_lock, flags); | |
907 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 908 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
909 | mem_cgroup_add_lru_list(page, page_lru(page)); |
910 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
911 | } | |
912 | ||
913 | ||
08e552c6 KH |
914 | void mem_cgroup_move_lists(struct page *page, |
915 | enum lru_list from, enum lru_list to) | |
916 | { | |
f8d66542 | 917 | if (mem_cgroup_disabled()) |
08e552c6 KH |
918 | return; |
919 | mem_cgroup_del_lru_list(page, from); | |
920 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
921 | } |
922 | ||
4c4a2214 DR |
923 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
924 | { | |
925 | int ret; | |
0b7f569e | 926 | struct mem_cgroup *curr = NULL; |
158e0a2d | 927 | struct task_struct *p; |
4c4a2214 | 928 | |
158e0a2d KH |
929 | p = find_lock_task_mm(task); |
930 | if (!p) | |
931 | return 0; | |
932 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
933 | task_unlock(p); | |
0b7f569e KH |
934 | if (!curr) |
935 | return 0; | |
d31f56db DN |
936 | /* |
937 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
938 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
939 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
940 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
941 | */ | |
942 | if (mem->use_hierarchy) | |
0b7f569e KH |
943 | ret = css_is_ancestor(&curr->css, &mem->css); |
944 | else | |
945 | ret = (curr == mem); | |
946 | css_put(&curr->css); | |
4c4a2214 DR |
947 | return ret; |
948 | } | |
949 | ||
c772be93 | 950 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
951 | { |
952 | unsigned long active; | |
953 | unsigned long inactive; | |
c772be93 KM |
954 | unsigned long gb; |
955 | unsigned long inactive_ratio; | |
14797e23 | 956 | |
14067bb3 KH |
957 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
958 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 959 | |
c772be93 KM |
960 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
961 | if (gb) | |
962 | inactive_ratio = int_sqrt(10 * gb); | |
963 | else | |
964 | inactive_ratio = 1; | |
965 | ||
966 | if (present_pages) { | |
967 | present_pages[0] = inactive; | |
968 | present_pages[1] = active; | |
969 | } | |
970 | ||
971 | return inactive_ratio; | |
972 | } | |
973 | ||
974 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
975 | { | |
976 | unsigned long active; | |
977 | unsigned long inactive; | |
978 | unsigned long present_pages[2]; | |
979 | unsigned long inactive_ratio; | |
980 | ||
981 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
982 | ||
983 | inactive = present_pages[0]; | |
984 | active = present_pages[1]; | |
985 | ||
986 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
987 | return 1; |
988 | ||
989 | return 0; | |
990 | } | |
991 | ||
56e49d21 RR |
992 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
993 | { | |
994 | unsigned long active; | |
995 | unsigned long inactive; | |
996 | ||
997 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
998 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
999 | ||
1000 | return (active > inactive); | |
1001 | } | |
1002 | ||
a3d8e054 KM |
1003 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
1004 | struct zone *zone, | |
1005 | enum lru_list lru) | |
1006 | { | |
13d7e3a2 | 1007 | int nid = zone_to_nid(zone); |
a3d8e054 KM |
1008 | int zid = zone_idx(zone); |
1009 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1010 | ||
1011 | return MEM_CGROUP_ZSTAT(mz, lru); | |
1012 | } | |
1013 | ||
3e2f41f1 KM |
1014 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1015 | struct zone *zone) | |
1016 | { | |
13d7e3a2 | 1017 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1018 | int zid = zone_idx(zone); |
1019 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1020 | ||
1021 | return &mz->reclaim_stat; | |
1022 | } | |
1023 | ||
1024 | struct zone_reclaim_stat * | |
1025 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1026 | { | |
1027 | struct page_cgroup *pc; | |
1028 | struct mem_cgroup_per_zone *mz; | |
1029 | ||
1030 | if (mem_cgroup_disabled()) | |
1031 | return NULL; | |
1032 | ||
1033 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1034 | /* |
1035 | * Used bit is set without atomic ops but after smp_wmb(). | |
1036 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
1037 | */ | |
1038 | smp_rmb(); | |
1039 | if (!PageCgroupUsed(pc)) | |
1040 | return NULL; | |
1041 | ||
3e2f41f1 KM |
1042 | mz = page_cgroup_zoneinfo(pc); |
1043 | if (!mz) | |
1044 | return NULL; | |
1045 | ||
1046 | return &mz->reclaim_stat; | |
1047 | } | |
1048 | ||
66e1707b BS |
1049 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
1050 | struct list_head *dst, | |
1051 | unsigned long *scanned, int order, | |
1052 | int mode, struct zone *z, | |
1053 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 1054 | int active, int file) |
66e1707b BS |
1055 | { |
1056 | unsigned long nr_taken = 0; | |
1057 | struct page *page; | |
1058 | unsigned long scan; | |
1059 | LIST_HEAD(pc_list); | |
1060 | struct list_head *src; | |
ff7283fa | 1061 | struct page_cgroup *pc, *tmp; |
13d7e3a2 | 1062 | int nid = zone_to_nid(z); |
1ecaab2b KH |
1063 | int zid = zone_idx(z); |
1064 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 1065 | int lru = LRU_FILE * file + active; |
2ffebca6 | 1066 | int ret; |
66e1707b | 1067 | |
cf475ad2 | 1068 | BUG_ON(!mem_cont); |
1ecaab2b | 1069 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 1070 | src = &mz->lists[lru]; |
66e1707b | 1071 | |
ff7283fa KH |
1072 | scan = 0; |
1073 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 1074 | if (scan >= nr_to_scan) |
ff7283fa | 1075 | break; |
08e552c6 KH |
1076 | |
1077 | page = pc->page; | |
52d4b9ac KH |
1078 | if (unlikely(!PageCgroupUsed(pc))) |
1079 | continue; | |
436c6541 | 1080 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1081 | continue; |
ff7283fa | 1082 | |
436c6541 | 1083 | scan++; |
2ffebca6 KH |
1084 | ret = __isolate_lru_page(page, mode, file); |
1085 | switch (ret) { | |
1086 | case 0: | |
66e1707b | 1087 | list_move(&page->lru, dst); |
2ffebca6 | 1088 | mem_cgroup_del_lru(page); |
2c888cfb | 1089 | nr_taken += hpage_nr_pages(page); |
2ffebca6 KH |
1090 | break; |
1091 | case -EBUSY: | |
1092 | /* we don't affect global LRU but rotate in our LRU */ | |
1093 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1094 | break; | |
1095 | default: | |
1096 | break; | |
66e1707b BS |
1097 | } |
1098 | } | |
1099 | ||
66e1707b | 1100 | *scanned = scan; |
cc8e970c KM |
1101 | |
1102 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, | |
1103 | 0, 0, 0, mode); | |
1104 | ||
66e1707b BS |
1105 | return nr_taken; |
1106 | } | |
1107 | ||
6d61ef40 BS |
1108 | #define mem_cgroup_from_res_counter(counter, member) \ |
1109 | container_of(counter, struct mem_cgroup, member) | |
1110 | ||
b85a96c0 DN |
1111 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
1112 | { | |
1113 | if (do_swap_account) { | |
1114 | if (res_counter_check_under_limit(&mem->res) && | |
1115 | res_counter_check_under_limit(&mem->memsw)) | |
1116 | return true; | |
1117 | } else | |
1118 | if (res_counter_check_under_limit(&mem->res)) | |
1119 | return true; | |
1120 | return false; | |
1121 | } | |
1122 | ||
a7885eb8 KM |
1123 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1124 | { | |
1125 | struct cgroup *cgrp = memcg->css.cgroup; | |
1126 | unsigned int swappiness; | |
1127 | ||
1128 | /* root ? */ | |
1129 | if (cgrp->parent == NULL) | |
1130 | return vm_swappiness; | |
1131 | ||
1132 | spin_lock(&memcg->reclaim_param_lock); | |
1133 | swappiness = memcg->swappiness; | |
1134 | spin_unlock(&memcg->reclaim_param_lock); | |
1135 | ||
1136 | return swappiness; | |
1137 | } | |
1138 | ||
32047e2a KH |
1139 | static void mem_cgroup_start_move(struct mem_cgroup *mem) |
1140 | { | |
1141 | int cpu; | |
1489ebad KH |
1142 | |
1143 | get_online_cpus(); | |
1144 | spin_lock(&mem->pcp_counter_lock); | |
1145 | for_each_online_cpu(cpu) | |
32047e2a | 1146 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1489ebad KH |
1147 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; |
1148 | spin_unlock(&mem->pcp_counter_lock); | |
1149 | put_online_cpus(); | |
32047e2a KH |
1150 | |
1151 | synchronize_rcu(); | |
1152 | } | |
1153 | ||
1154 | static void mem_cgroup_end_move(struct mem_cgroup *mem) | |
1155 | { | |
1156 | int cpu; | |
1157 | ||
1158 | if (!mem) | |
1159 | return; | |
1489ebad KH |
1160 | get_online_cpus(); |
1161 | spin_lock(&mem->pcp_counter_lock); | |
1162 | for_each_online_cpu(cpu) | |
32047e2a | 1163 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1489ebad KH |
1164 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; |
1165 | spin_unlock(&mem->pcp_counter_lock); | |
1166 | put_online_cpus(); | |
32047e2a KH |
1167 | } |
1168 | /* | |
1169 | * 2 routines for checking "mem" is under move_account() or not. | |
1170 | * | |
1171 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1172 | * for avoiding race in accounting. If true, | |
1173 | * pc->mem_cgroup may be overwritten. | |
1174 | * | |
1175 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1176 | * under hierarchy of moving cgroups. This is for | |
1177 | * waiting at hith-memory prressure caused by "move". | |
1178 | */ | |
1179 | ||
1180 | static bool mem_cgroup_stealed(struct mem_cgroup *mem) | |
1181 | { | |
1182 | VM_BUG_ON(!rcu_read_lock_held()); | |
1183 | return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0; | |
1184 | } | |
4b534334 KH |
1185 | |
1186 | static bool mem_cgroup_under_move(struct mem_cgroup *mem) | |
1187 | { | |
2bd9bb20 KH |
1188 | struct mem_cgroup *from; |
1189 | struct mem_cgroup *to; | |
4b534334 | 1190 | bool ret = false; |
2bd9bb20 KH |
1191 | /* |
1192 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1193 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1194 | */ | |
1195 | spin_lock(&mc.lock); | |
1196 | from = mc.from; | |
1197 | to = mc.to; | |
1198 | if (!from) | |
1199 | goto unlock; | |
1200 | if (from == mem || to == mem | |
1201 | || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css)) | |
1202 | || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css))) | |
1203 | ret = true; | |
1204 | unlock: | |
1205 | spin_unlock(&mc.lock); | |
4b534334 KH |
1206 | return ret; |
1207 | } | |
1208 | ||
1209 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem) | |
1210 | { | |
1211 | if (mc.moving_task && current != mc.moving_task) { | |
1212 | if (mem_cgroup_under_move(mem)) { | |
1213 | DEFINE_WAIT(wait); | |
1214 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1215 | /* moving charge context might have finished. */ | |
1216 | if (mc.moving_task) | |
1217 | schedule(); | |
1218 | finish_wait(&mc.waitq, &wait); | |
1219 | return true; | |
1220 | } | |
1221 | } | |
1222 | return false; | |
1223 | } | |
1224 | ||
e222432b | 1225 | /** |
6a6135b6 | 1226 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1227 | * @memcg: The memory cgroup that went over limit |
1228 | * @p: Task that is going to be killed | |
1229 | * | |
1230 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1231 | * enabled | |
1232 | */ | |
1233 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1234 | { | |
1235 | struct cgroup *task_cgrp; | |
1236 | struct cgroup *mem_cgrp; | |
1237 | /* | |
1238 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1239 | * on the assumption that OOM is serialized for memory controller. | |
1240 | * If this assumption is broken, revisit this code. | |
1241 | */ | |
1242 | static char memcg_name[PATH_MAX]; | |
1243 | int ret; | |
1244 | ||
d31f56db | 1245 | if (!memcg || !p) |
e222432b BS |
1246 | return; |
1247 | ||
1248 | ||
1249 | rcu_read_lock(); | |
1250 | ||
1251 | mem_cgrp = memcg->css.cgroup; | |
1252 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1253 | ||
1254 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1255 | if (ret < 0) { | |
1256 | /* | |
1257 | * Unfortunately, we are unable to convert to a useful name | |
1258 | * But we'll still print out the usage information | |
1259 | */ | |
1260 | rcu_read_unlock(); | |
1261 | goto done; | |
1262 | } | |
1263 | rcu_read_unlock(); | |
1264 | ||
1265 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1266 | ||
1267 | rcu_read_lock(); | |
1268 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1269 | if (ret < 0) { | |
1270 | rcu_read_unlock(); | |
1271 | goto done; | |
1272 | } | |
1273 | rcu_read_unlock(); | |
1274 | ||
1275 | /* | |
1276 | * Continues from above, so we don't need an KERN_ level | |
1277 | */ | |
1278 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1279 | done: | |
1280 | ||
1281 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1282 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1283 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1284 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1285 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1286 | "failcnt %llu\n", | |
1287 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1288 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1289 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1290 | } | |
1291 | ||
81d39c20 KH |
1292 | /* |
1293 | * This function returns the number of memcg under hierarchy tree. Returns | |
1294 | * 1(self count) if no children. | |
1295 | */ | |
1296 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1297 | { | |
1298 | int num = 0; | |
7d74b06f KH |
1299 | struct mem_cgroup *iter; |
1300 | ||
1301 | for_each_mem_cgroup_tree(iter, mem) | |
1302 | num++; | |
81d39c20 KH |
1303 | return num; |
1304 | } | |
1305 | ||
a63d83f4 DR |
1306 | /* |
1307 | * Return the memory (and swap, if configured) limit for a memcg. | |
1308 | */ | |
1309 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1310 | { | |
1311 | u64 limit; | |
1312 | u64 memsw; | |
1313 | ||
f3e8eb70 JW |
1314 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1315 | limit += total_swap_pages << PAGE_SHIFT; | |
1316 | ||
a63d83f4 DR |
1317 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1318 | /* | |
1319 | * If memsw is finite and limits the amount of swap space available | |
1320 | * to this memcg, return that limit. | |
1321 | */ | |
1322 | return min(limit, memsw); | |
1323 | } | |
1324 | ||
6d61ef40 | 1325 | /* |
04046e1a KH |
1326 | * Visit the first child (need not be the first child as per the ordering |
1327 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1328 | * that to reclaim free pages from. | |
1329 | */ | |
1330 | static struct mem_cgroup * | |
1331 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1332 | { | |
1333 | struct mem_cgroup *ret = NULL; | |
1334 | struct cgroup_subsys_state *css; | |
1335 | int nextid, found; | |
1336 | ||
1337 | if (!root_mem->use_hierarchy) { | |
1338 | css_get(&root_mem->css); | |
1339 | ret = root_mem; | |
1340 | } | |
1341 | ||
1342 | while (!ret) { | |
1343 | rcu_read_lock(); | |
1344 | nextid = root_mem->last_scanned_child + 1; | |
1345 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1346 | &found); | |
1347 | if (css && css_tryget(css)) | |
1348 | ret = container_of(css, struct mem_cgroup, css); | |
1349 | ||
1350 | rcu_read_unlock(); | |
1351 | /* Updates scanning parameter */ | |
1352 | spin_lock(&root_mem->reclaim_param_lock); | |
1353 | if (!css) { | |
1354 | /* this means start scan from ID:1 */ | |
1355 | root_mem->last_scanned_child = 0; | |
1356 | } else | |
1357 | root_mem->last_scanned_child = found; | |
1358 | spin_unlock(&root_mem->reclaim_param_lock); | |
1359 | } | |
1360 | ||
1361 | return ret; | |
1362 | } | |
1363 | ||
1364 | /* | |
1365 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1366 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1367 | * based on its position in the children list. | |
6d61ef40 BS |
1368 | * |
1369 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1370 | * |
1371 | * We give up and return to the caller when we visit root_mem twice. | |
1372 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1373 | * |
1374 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1375 | */ |
1376 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1377 | struct zone *zone, |
75822b44 BS |
1378 | gfp_t gfp_mask, |
1379 | unsigned long reclaim_options) | |
6d61ef40 | 1380 | { |
04046e1a KH |
1381 | struct mem_cgroup *victim; |
1382 | int ret, total = 0; | |
1383 | int loop = 0; | |
75822b44 BS |
1384 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1385 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 BS |
1386 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
1387 | unsigned long excess = mem_cgroup_get_excess(root_mem); | |
04046e1a | 1388 | |
22a668d7 KH |
1389 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1390 | if (root_mem->memsw_is_minimum) | |
1391 | noswap = true; | |
1392 | ||
4e416953 | 1393 | while (1) { |
04046e1a | 1394 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1395 | if (victim == root_mem) { |
04046e1a | 1396 | loop++; |
cdec2e42 KH |
1397 | if (loop >= 1) |
1398 | drain_all_stock_async(); | |
4e416953 BS |
1399 | if (loop >= 2) { |
1400 | /* | |
1401 | * If we have not been able to reclaim | |
1402 | * anything, it might because there are | |
1403 | * no reclaimable pages under this hierarchy | |
1404 | */ | |
1405 | if (!check_soft || !total) { | |
1406 | css_put(&victim->css); | |
1407 | break; | |
1408 | } | |
1409 | /* | |
1410 | * We want to do more targetted reclaim. | |
1411 | * excess >> 2 is not to excessive so as to | |
1412 | * reclaim too much, nor too less that we keep | |
1413 | * coming back to reclaim from this cgroup | |
1414 | */ | |
1415 | if (total >= (excess >> 2) || | |
1416 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1417 | css_put(&victim->css); | |
1418 | break; | |
1419 | } | |
1420 | } | |
1421 | } | |
c62b1a3b | 1422 | if (!mem_cgroup_local_usage(victim)) { |
04046e1a KH |
1423 | /* this cgroup's local usage == 0 */ |
1424 | css_put(&victim->css); | |
6d61ef40 BS |
1425 | continue; |
1426 | } | |
04046e1a | 1427 | /* we use swappiness of local cgroup */ |
4e416953 BS |
1428 | if (check_soft) |
1429 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
14fec796 | 1430 | noswap, get_swappiness(victim), zone); |
4e416953 BS |
1431 | else |
1432 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1433 | noswap, get_swappiness(victim)); | |
04046e1a | 1434 | css_put(&victim->css); |
81d39c20 KH |
1435 | /* |
1436 | * At shrinking usage, we can't check we should stop here or | |
1437 | * reclaim more. It's depends on callers. last_scanned_child | |
1438 | * will work enough for keeping fairness under tree. | |
1439 | */ | |
1440 | if (shrink) | |
1441 | return ret; | |
04046e1a | 1442 | total += ret; |
4e416953 BS |
1443 | if (check_soft) { |
1444 | if (res_counter_check_under_soft_limit(&root_mem->res)) | |
1445 | return total; | |
1446 | } else if (mem_cgroup_check_under_limit(root_mem)) | |
04046e1a | 1447 | return 1 + total; |
6d61ef40 | 1448 | } |
04046e1a | 1449 | return total; |
6d61ef40 BS |
1450 | } |
1451 | ||
867578cb KH |
1452 | /* |
1453 | * Check OOM-Killer is already running under our hierarchy. | |
1454 | * If someone is running, return false. | |
1455 | */ | |
1456 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) | |
1457 | { | |
7d74b06f KH |
1458 | int x, lock_count = 0; |
1459 | struct mem_cgroup *iter; | |
a636b327 | 1460 | |
7d74b06f KH |
1461 | for_each_mem_cgroup_tree(iter, mem) { |
1462 | x = atomic_inc_return(&iter->oom_lock); | |
1463 | lock_count = max(x, lock_count); | |
1464 | } | |
867578cb KH |
1465 | |
1466 | if (lock_count == 1) | |
1467 | return true; | |
1468 | return false; | |
a636b327 | 1469 | } |
0b7f569e | 1470 | |
7d74b06f | 1471 | static int mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
0b7f569e | 1472 | { |
7d74b06f KH |
1473 | struct mem_cgroup *iter; |
1474 | ||
867578cb KH |
1475 | /* |
1476 | * When a new child is created while the hierarchy is under oom, | |
1477 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1478 | * atomic_add_unless() here. | |
1479 | */ | |
7d74b06f KH |
1480 | for_each_mem_cgroup_tree(iter, mem) |
1481 | atomic_add_unless(&iter->oom_lock, -1, 0); | |
0b7f569e KH |
1482 | return 0; |
1483 | } | |
1484 | ||
867578cb KH |
1485 | |
1486 | static DEFINE_MUTEX(memcg_oom_mutex); | |
1487 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); | |
1488 | ||
dc98df5a KH |
1489 | struct oom_wait_info { |
1490 | struct mem_cgroup *mem; | |
1491 | wait_queue_t wait; | |
1492 | }; | |
1493 | ||
1494 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1495 | unsigned mode, int sync, void *arg) | |
1496 | { | |
1497 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; | |
1498 | struct oom_wait_info *oom_wait_info; | |
1499 | ||
1500 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
1501 | ||
1502 | if (oom_wait_info->mem == wake_mem) | |
1503 | goto wakeup; | |
1504 | /* if no hierarchy, no match */ | |
1505 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) | |
1506 | return 0; | |
1507 | /* | |
1508 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1509 | * Then we can use css_is_ancestor without taking care of RCU. | |
1510 | */ | |
1511 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && | |
1512 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) | |
1513 | return 0; | |
1514 | ||
1515 | wakeup: | |
1516 | return autoremove_wake_function(wait, mode, sync, arg); | |
1517 | } | |
1518 | ||
1519 | static void memcg_wakeup_oom(struct mem_cgroup *mem) | |
1520 | { | |
1521 | /* for filtering, pass "mem" as argument. */ | |
1522 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); | |
1523 | } | |
1524 | ||
3c11ecf4 KH |
1525 | static void memcg_oom_recover(struct mem_cgroup *mem) |
1526 | { | |
2bd9bb20 | 1527 | if (mem && atomic_read(&mem->oom_lock)) |
3c11ecf4 KH |
1528 | memcg_wakeup_oom(mem); |
1529 | } | |
1530 | ||
867578cb KH |
1531 | /* |
1532 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1533 | */ | |
1534 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) | |
0b7f569e | 1535 | { |
dc98df5a | 1536 | struct oom_wait_info owait; |
3c11ecf4 | 1537 | bool locked, need_to_kill; |
867578cb | 1538 | |
dc98df5a KH |
1539 | owait.mem = mem; |
1540 | owait.wait.flags = 0; | |
1541 | owait.wait.func = memcg_oom_wake_function; | |
1542 | owait.wait.private = current; | |
1543 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1544 | need_to_kill = true; |
867578cb KH |
1545 | /* At first, try to OOM lock hierarchy under mem.*/ |
1546 | mutex_lock(&memcg_oom_mutex); | |
1547 | locked = mem_cgroup_oom_lock(mem); | |
1548 | /* | |
1549 | * Even if signal_pending(), we can't quit charge() loop without | |
1550 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1551 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1552 | */ | |
3c11ecf4 KH |
1553 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
1554 | if (!locked || mem->oom_kill_disable) | |
1555 | need_to_kill = false; | |
1556 | if (locked) | |
9490ff27 | 1557 | mem_cgroup_oom_notify(mem); |
867578cb KH |
1558 | mutex_unlock(&memcg_oom_mutex); |
1559 | ||
3c11ecf4 KH |
1560 | if (need_to_kill) { |
1561 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
867578cb | 1562 | mem_cgroup_out_of_memory(mem, mask); |
3c11ecf4 | 1563 | } else { |
867578cb | 1564 | schedule(); |
dc98df5a | 1565 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb KH |
1566 | } |
1567 | mutex_lock(&memcg_oom_mutex); | |
1568 | mem_cgroup_oom_unlock(mem); | |
dc98df5a | 1569 | memcg_wakeup_oom(mem); |
867578cb KH |
1570 | mutex_unlock(&memcg_oom_mutex); |
1571 | ||
1572 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) | |
1573 | return false; | |
1574 | /* Give chance to dying process */ | |
1575 | schedule_timeout(1); | |
1576 | return true; | |
0b7f569e KH |
1577 | } |
1578 | ||
d69b042f BS |
1579 | /* |
1580 | * Currently used to update mapped file statistics, but the routine can be | |
1581 | * generalized to update other statistics as well. | |
32047e2a KH |
1582 | * |
1583 | * Notes: Race condition | |
1584 | * | |
1585 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1586 | * it tends to be costly. But considering some conditions, we doesn't need | |
1587 | * to do so _always_. | |
1588 | * | |
1589 | * Considering "charge", lock_page_cgroup() is not required because all | |
1590 | * file-stat operations happen after a page is attached to radix-tree. There | |
1591 | * are no race with "charge". | |
1592 | * | |
1593 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1594 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1595 | * if there are race with "uncharge". Statistics itself is properly handled | |
1596 | * by flags. | |
1597 | * | |
1598 | * Considering "move", this is an only case we see a race. To make the race | |
1599 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
1600 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 1601 | */ |
26174efd | 1602 | |
2a7106f2 GT |
1603 | void mem_cgroup_update_page_stat(struct page *page, |
1604 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f BS |
1605 | { |
1606 | struct mem_cgroup *mem; | |
32047e2a KH |
1607 | struct page_cgroup *pc = lookup_page_cgroup(page); |
1608 | bool need_unlock = false; | |
dbd4ea78 | 1609 | unsigned long uninitialized_var(flags); |
d69b042f | 1610 | |
d69b042f BS |
1611 | if (unlikely(!pc)) |
1612 | return; | |
1613 | ||
32047e2a | 1614 | rcu_read_lock(); |
d69b042f | 1615 | mem = pc->mem_cgroup; |
32047e2a KH |
1616 | if (unlikely(!mem || !PageCgroupUsed(pc))) |
1617 | goto out; | |
1618 | /* pc->mem_cgroup is unstable ? */ | |
ca3e0214 | 1619 | if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) { |
32047e2a | 1620 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 1621 | move_lock_page_cgroup(pc, &flags); |
32047e2a KH |
1622 | need_unlock = true; |
1623 | mem = pc->mem_cgroup; | |
1624 | if (!mem || !PageCgroupUsed(pc)) | |
1625 | goto out; | |
1626 | } | |
26174efd | 1627 | |
26174efd | 1628 | switch (idx) { |
2a7106f2 | 1629 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
1630 | if (val > 0) |
1631 | SetPageCgroupFileMapped(pc); | |
1632 | else if (!page_mapped(page)) | |
0c270f8f | 1633 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 1634 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1635 | break; |
1636 | default: | |
1637 | BUG(); | |
8725d541 | 1638 | } |
d69b042f | 1639 | |
2a7106f2 GT |
1640 | this_cpu_add(mem->stat->count[idx], val); |
1641 | ||
32047e2a KH |
1642 | out: |
1643 | if (unlikely(need_unlock)) | |
dbd4ea78 | 1644 | move_unlock_page_cgroup(pc, &flags); |
32047e2a KH |
1645 | rcu_read_unlock(); |
1646 | return; | |
d69b042f | 1647 | } |
2a7106f2 | 1648 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
26174efd | 1649 | |
cdec2e42 KH |
1650 | /* |
1651 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1652 | * TODO: maybe necessary to use big numbers in big irons. | |
1653 | */ | |
1654 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1655 | struct memcg_stock_pcp { | |
1656 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1657 | int charge; | |
1658 | struct work_struct work; | |
1659 | }; | |
1660 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1661 | static atomic_t memcg_drain_count; | |
1662 | ||
1663 | /* | |
1664 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1665 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1666 | * cgroup which is not current target, returns false. This stock will be | |
1667 | * refilled. | |
1668 | */ | |
1669 | static bool consume_stock(struct mem_cgroup *mem) | |
1670 | { | |
1671 | struct memcg_stock_pcp *stock; | |
1672 | bool ret = true; | |
1673 | ||
1674 | stock = &get_cpu_var(memcg_stock); | |
1675 | if (mem == stock->cached && stock->charge) | |
1676 | stock->charge -= PAGE_SIZE; | |
1677 | else /* need to call res_counter_charge */ | |
1678 | ret = false; | |
1679 | put_cpu_var(memcg_stock); | |
1680 | return ret; | |
1681 | } | |
1682 | ||
1683 | /* | |
1684 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1685 | */ | |
1686 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1687 | { | |
1688 | struct mem_cgroup *old = stock->cached; | |
1689 | ||
1690 | if (stock->charge) { | |
1691 | res_counter_uncharge(&old->res, stock->charge); | |
1692 | if (do_swap_account) | |
1693 | res_counter_uncharge(&old->memsw, stock->charge); | |
1694 | } | |
1695 | stock->cached = NULL; | |
1696 | stock->charge = 0; | |
1697 | } | |
1698 | ||
1699 | /* | |
1700 | * This must be called under preempt disabled or must be called by | |
1701 | * a thread which is pinned to local cpu. | |
1702 | */ | |
1703 | static void drain_local_stock(struct work_struct *dummy) | |
1704 | { | |
1705 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1706 | drain_stock(stock); | |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 1711 | * This will be consumed by consume_stock() function, later. |
cdec2e42 KH |
1712 | */ |
1713 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1714 | { | |
1715 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1716 | ||
1717 | if (stock->cached != mem) { /* reset if necessary */ | |
1718 | drain_stock(stock); | |
1719 | stock->cached = mem; | |
1720 | } | |
1721 | stock->charge += val; | |
1722 | put_cpu_var(memcg_stock); | |
1723 | } | |
1724 | ||
1725 | /* | |
1726 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1727 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1728 | * expects some charges will be back to res_counter later but cannot wait for | |
1729 | * it. | |
1730 | */ | |
1731 | static void drain_all_stock_async(void) | |
1732 | { | |
1733 | int cpu; | |
1734 | /* This function is for scheduling "drain" in asynchronous way. | |
1735 | * The result of "drain" is not directly handled by callers. Then, | |
1736 | * if someone is calling drain, we don't have to call drain more. | |
1737 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1738 | * there is a race. We just do loose check here. | |
1739 | */ | |
1740 | if (atomic_read(&memcg_drain_count)) | |
1741 | return; | |
1742 | /* Notify other cpus that system-wide "drain" is running */ | |
1743 | atomic_inc(&memcg_drain_count); | |
1744 | get_online_cpus(); | |
1745 | for_each_online_cpu(cpu) { | |
1746 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1747 | schedule_work_on(cpu, &stock->work); | |
1748 | } | |
1749 | put_online_cpus(); | |
1750 | atomic_dec(&memcg_drain_count); | |
1751 | /* We don't wait for flush_work */ | |
1752 | } | |
1753 | ||
1754 | /* This is a synchronous drain interface. */ | |
1755 | static void drain_all_stock_sync(void) | |
1756 | { | |
1757 | /* called when force_empty is called */ | |
1758 | atomic_inc(&memcg_drain_count); | |
1759 | schedule_on_each_cpu(drain_local_stock); | |
1760 | atomic_dec(&memcg_drain_count); | |
1761 | } | |
1762 | ||
711d3d2c KH |
1763 | /* |
1764 | * This function drains percpu counter value from DEAD cpu and | |
1765 | * move it to local cpu. Note that this function can be preempted. | |
1766 | */ | |
1767 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu) | |
1768 | { | |
1769 | int i; | |
1770 | ||
1771 | spin_lock(&mem->pcp_counter_lock); | |
1772 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { | |
1773 | s64 x = per_cpu(mem->stat->count[i], cpu); | |
1774 | ||
1775 | per_cpu(mem->stat->count[i], cpu) = 0; | |
1776 | mem->nocpu_base.count[i] += x; | |
1777 | } | |
1489ebad KH |
1778 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
1779 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; | |
1780 | spin_unlock(&mem->pcp_counter_lock); | |
1781 | } | |
1782 | ||
1783 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu) | |
1784 | { | |
1785 | int idx = MEM_CGROUP_ON_MOVE; | |
1786 | ||
1787 | spin_lock(&mem->pcp_counter_lock); | |
1788 | per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx]; | |
711d3d2c KH |
1789 | spin_unlock(&mem->pcp_counter_lock); |
1790 | } | |
1791 | ||
1792 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
1793 | unsigned long action, |
1794 | void *hcpu) | |
1795 | { | |
1796 | int cpu = (unsigned long)hcpu; | |
1797 | struct memcg_stock_pcp *stock; | |
711d3d2c | 1798 | struct mem_cgroup *iter; |
cdec2e42 | 1799 | |
1489ebad KH |
1800 | if ((action == CPU_ONLINE)) { |
1801 | for_each_mem_cgroup_all(iter) | |
1802 | synchronize_mem_cgroup_on_move(iter, cpu); | |
1803 | return NOTIFY_OK; | |
1804 | } | |
1805 | ||
711d3d2c | 1806 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 1807 | return NOTIFY_OK; |
711d3d2c KH |
1808 | |
1809 | for_each_mem_cgroup_all(iter) | |
1810 | mem_cgroup_drain_pcp_counter(iter, cpu); | |
1811 | ||
cdec2e42 KH |
1812 | stock = &per_cpu(memcg_stock, cpu); |
1813 | drain_stock(stock); | |
1814 | return NOTIFY_OK; | |
1815 | } | |
1816 | ||
4b534334 KH |
1817 | |
1818 | /* See __mem_cgroup_try_charge() for details */ | |
1819 | enum { | |
1820 | CHARGE_OK, /* success */ | |
1821 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
1822 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
1823 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
1824 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
1825 | }; | |
1826 | ||
1827 | static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask, | |
1828 | int csize, bool oom_check) | |
1829 | { | |
1830 | struct mem_cgroup *mem_over_limit; | |
1831 | struct res_counter *fail_res; | |
1832 | unsigned long flags = 0; | |
1833 | int ret; | |
1834 | ||
1835 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
1836 | ||
1837 | if (likely(!ret)) { | |
1838 | if (!do_swap_account) | |
1839 | return CHARGE_OK; | |
1840 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); | |
1841 | if (likely(!ret)) | |
1842 | return CHARGE_OK; | |
1843 | ||
1844 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); | |
1845 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
1846 | } else | |
1847 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
1848 | ||
1849 | if (csize > PAGE_SIZE) /* change csize and retry */ | |
1850 | return CHARGE_RETRY; | |
1851 | ||
1852 | if (!(gfp_mask & __GFP_WAIT)) | |
1853 | return CHARGE_WOULDBLOCK; | |
1854 | ||
1855 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, | |
1856 | gfp_mask, flags); | |
1857 | /* | |
1858 | * try_to_free_mem_cgroup_pages() might not give us a full | |
1859 | * picture of reclaim. Some pages are reclaimed and might be | |
1860 | * moved to swap cache or just unmapped from the cgroup. | |
1861 | * Check the limit again to see if the reclaim reduced the | |
1862 | * current usage of the cgroup before giving up | |
1863 | */ | |
1864 | if (ret || mem_cgroup_check_under_limit(mem_over_limit)) | |
1865 | return CHARGE_RETRY; | |
1866 | ||
1867 | /* | |
1868 | * At task move, charge accounts can be doubly counted. So, it's | |
1869 | * better to wait until the end of task_move if something is going on. | |
1870 | */ | |
1871 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
1872 | return CHARGE_RETRY; | |
1873 | ||
1874 | /* If we don't need to call oom-killer at el, return immediately */ | |
1875 | if (!oom_check) | |
1876 | return CHARGE_NOMEM; | |
1877 | /* check OOM */ | |
1878 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) | |
1879 | return CHARGE_OOM_DIE; | |
1880 | ||
1881 | return CHARGE_RETRY; | |
1882 | } | |
1883 | ||
f817ed48 KH |
1884 | /* |
1885 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1886 | * oom-killer can be invoked. | |
8a9f3ccd | 1887 | */ |
f817ed48 | 1888 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 AA |
1889 | gfp_t gfp_mask, |
1890 | struct mem_cgroup **memcg, bool oom, | |
1891 | int page_size) | |
8a9f3ccd | 1892 | { |
4b534334 KH |
1893 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
1894 | struct mem_cgroup *mem = NULL; | |
1895 | int ret; | |
ec168510 | 1896 | int csize = max(CHARGE_SIZE, (unsigned long) page_size); |
a636b327 | 1897 | |
867578cb KH |
1898 | /* |
1899 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
1900 | * in system level. So, allow to go ahead dying process in addition to | |
1901 | * MEMDIE process. | |
1902 | */ | |
1903 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
1904 | || fatal_signal_pending(current))) | |
1905 | goto bypass; | |
a636b327 | 1906 | |
8a9f3ccd | 1907 | /* |
3be91277 HD |
1908 | * We always charge the cgroup the mm_struct belongs to. |
1909 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
1910 | * thread group leader migrates. It's possible that mm is not |
1911 | * set, if so charge the init_mm (happens for pagecache usage). | |
1912 | */ | |
f75ca962 KH |
1913 | if (!*memcg && !mm) |
1914 | goto bypass; | |
1915 | again: | |
1916 | if (*memcg) { /* css should be a valid one */ | |
4b534334 | 1917 | mem = *memcg; |
f75ca962 KH |
1918 | VM_BUG_ON(css_is_removed(&mem->css)); |
1919 | if (mem_cgroup_is_root(mem)) | |
1920 | goto done; | |
ec168510 | 1921 | if (page_size == PAGE_SIZE && consume_stock(mem)) |
f75ca962 | 1922 | goto done; |
4b534334 KH |
1923 | css_get(&mem->css); |
1924 | } else { | |
f75ca962 | 1925 | struct task_struct *p; |
54595fe2 | 1926 | |
f75ca962 KH |
1927 | rcu_read_lock(); |
1928 | p = rcu_dereference(mm->owner); | |
f75ca962 | 1929 | /* |
ebb76ce1 KH |
1930 | * Because we don't have task_lock(), "p" can exit. |
1931 | * In that case, "mem" can point to root or p can be NULL with | |
1932 | * race with swapoff. Then, we have small risk of mis-accouning. | |
1933 | * But such kind of mis-account by race always happens because | |
1934 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
1935 | * small race, here. | |
1936 | * (*) swapoff at el will charge against mm-struct not against | |
1937 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 KH |
1938 | */ |
1939 | mem = mem_cgroup_from_task(p); | |
ebb76ce1 | 1940 | if (!mem || mem_cgroup_is_root(mem)) { |
f75ca962 KH |
1941 | rcu_read_unlock(); |
1942 | goto done; | |
1943 | } | |
ec168510 | 1944 | if (page_size == PAGE_SIZE && consume_stock(mem)) { |
f75ca962 KH |
1945 | /* |
1946 | * It seems dagerous to access memcg without css_get(). | |
1947 | * But considering how consume_stok works, it's not | |
1948 | * necessary. If consume_stock success, some charges | |
1949 | * from this memcg are cached on this cpu. So, we | |
1950 | * don't need to call css_get()/css_tryget() before | |
1951 | * calling consume_stock(). | |
1952 | */ | |
1953 | rcu_read_unlock(); | |
1954 | goto done; | |
1955 | } | |
1956 | /* after here, we may be blocked. we need to get refcnt */ | |
1957 | if (!css_tryget(&mem->css)) { | |
1958 | rcu_read_unlock(); | |
1959 | goto again; | |
1960 | } | |
1961 | rcu_read_unlock(); | |
1962 | } | |
8a9f3ccd | 1963 | |
4b534334 KH |
1964 | do { |
1965 | bool oom_check; | |
7a81b88c | 1966 | |
4b534334 | 1967 | /* If killed, bypass charge */ |
f75ca962 KH |
1968 | if (fatal_signal_pending(current)) { |
1969 | css_put(&mem->css); | |
4b534334 | 1970 | goto bypass; |
f75ca962 | 1971 | } |
6d61ef40 | 1972 | |
4b534334 KH |
1973 | oom_check = false; |
1974 | if (oom && !nr_oom_retries) { | |
1975 | oom_check = true; | |
1976 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 1977 | } |
66e1707b | 1978 | |
4b534334 | 1979 | ret = __mem_cgroup_do_charge(mem, gfp_mask, csize, oom_check); |
8033b97c | 1980 | |
4b534334 KH |
1981 | switch (ret) { |
1982 | case CHARGE_OK: | |
1983 | break; | |
1984 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
ec168510 | 1985 | csize = page_size; |
f75ca962 KH |
1986 | css_put(&mem->css); |
1987 | mem = NULL; | |
1988 | goto again; | |
4b534334 | 1989 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
f75ca962 | 1990 | css_put(&mem->css); |
4b534334 KH |
1991 | goto nomem; |
1992 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 KH |
1993 | if (!oom) { |
1994 | css_put(&mem->css); | |
867578cb | 1995 | goto nomem; |
f75ca962 | 1996 | } |
4b534334 KH |
1997 | /* If oom, we never return -ENOMEM */ |
1998 | nr_oom_retries--; | |
1999 | break; | |
2000 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
f75ca962 | 2001 | css_put(&mem->css); |
867578cb | 2002 | goto bypass; |
66e1707b | 2003 | } |
4b534334 KH |
2004 | } while (ret != CHARGE_OK); |
2005 | ||
ec168510 AA |
2006 | if (csize > page_size) |
2007 | refill_stock(mem, csize - page_size); | |
f75ca962 | 2008 | css_put(&mem->css); |
0c3e73e8 | 2009 | done: |
f75ca962 | 2010 | *memcg = mem; |
7a81b88c KH |
2011 | return 0; |
2012 | nomem: | |
f75ca962 | 2013 | *memcg = NULL; |
7a81b88c | 2014 | return -ENOMEM; |
867578cb KH |
2015 | bypass: |
2016 | *memcg = NULL; | |
2017 | return 0; | |
7a81b88c | 2018 | } |
8a9f3ccd | 2019 | |
a3032a2c DN |
2020 | /* |
2021 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2022 | * This function is for that and do uncharge, put css's refcnt. | |
2023 | * gotten by try_charge(). | |
2024 | */ | |
854ffa8d DN |
2025 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
2026 | unsigned long count) | |
a3032a2c DN |
2027 | { |
2028 | if (!mem_cgroup_is_root(mem)) { | |
854ffa8d | 2029 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
a3032a2c | 2030 | if (do_swap_account) |
854ffa8d | 2031 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
a3032a2c | 2032 | } |
854ffa8d DN |
2033 | } |
2034 | ||
ec168510 AA |
2035 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
2036 | int page_size) | |
854ffa8d | 2037 | { |
ec168510 | 2038 | __mem_cgroup_cancel_charge(mem, page_size >> PAGE_SHIFT); |
a3032a2c DN |
2039 | } |
2040 | ||
a3b2d692 KH |
2041 | /* |
2042 | * A helper function to get mem_cgroup from ID. must be called under | |
2043 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2044 | * it's concern. (dropping refcnt from swap can be called against removed | |
2045 | * memcg.) | |
2046 | */ | |
2047 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2048 | { | |
2049 | struct cgroup_subsys_state *css; | |
2050 | ||
2051 | /* ID 0 is unused ID */ | |
2052 | if (!id) | |
2053 | return NULL; | |
2054 | css = css_lookup(&mem_cgroup_subsys, id); | |
2055 | if (!css) | |
2056 | return NULL; | |
2057 | return container_of(css, struct mem_cgroup, css); | |
2058 | } | |
2059 | ||
e42d9d5d | 2060 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2061 | { |
e42d9d5d | 2062 | struct mem_cgroup *mem = NULL; |
3c776e64 | 2063 | struct page_cgroup *pc; |
a3b2d692 | 2064 | unsigned short id; |
b5a84319 KH |
2065 | swp_entry_t ent; |
2066 | ||
3c776e64 DN |
2067 | VM_BUG_ON(!PageLocked(page)); |
2068 | ||
3c776e64 | 2069 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2070 | lock_page_cgroup(pc); |
a3b2d692 | 2071 | if (PageCgroupUsed(pc)) { |
3c776e64 | 2072 | mem = pc->mem_cgroup; |
a3b2d692 KH |
2073 | if (mem && !css_tryget(&mem->css)) |
2074 | mem = NULL; | |
e42d9d5d | 2075 | } else if (PageSwapCache(page)) { |
3c776e64 | 2076 | ent.val = page_private(page); |
a3b2d692 KH |
2077 | id = lookup_swap_cgroup(ent); |
2078 | rcu_read_lock(); | |
2079 | mem = mem_cgroup_lookup(id); | |
2080 | if (mem && !css_tryget(&mem->css)) | |
2081 | mem = NULL; | |
2082 | rcu_read_unlock(); | |
3c776e64 | 2083 | } |
c0bd3f63 | 2084 | unlock_page_cgroup(pc); |
b5a84319 KH |
2085 | return mem; |
2086 | } | |
2087 | ||
ca3e0214 KH |
2088 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, |
2089 | struct page_cgroup *pc, | |
2090 | enum charge_type ctype, | |
2091 | int page_size) | |
7a81b88c | 2092 | { |
ca3e0214 KH |
2093 | int nr_pages = page_size >> PAGE_SHIFT; |
2094 | ||
2095 | /* try_charge() can return NULL to *memcg, taking care of it. */ | |
2096 | if (!mem) | |
2097 | return; | |
2098 | ||
2099 | lock_page_cgroup(pc); | |
2100 | if (unlikely(PageCgroupUsed(pc))) { | |
2101 | unlock_page_cgroup(pc); | |
2102 | mem_cgroup_cancel_charge(mem, page_size); | |
2103 | return; | |
2104 | } | |
2105 | /* | |
2106 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2107 | * accessed by any other context at this point. | |
2108 | */ | |
8a9f3ccd | 2109 | pc->mem_cgroup = mem; |
261fb61a KH |
2110 | /* |
2111 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2112 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2113 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2114 | * before USED bit, we need memory barrier here. | |
2115 | * See mem_cgroup_add_lru_list(), etc. | |
2116 | */ | |
08e552c6 | 2117 | smp_wmb(); |
4b3bde4c BS |
2118 | switch (ctype) { |
2119 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
2120 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
2121 | SetPageCgroupCache(pc); | |
2122 | SetPageCgroupUsed(pc); | |
2123 | break; | |
2124 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2125 | ClearPageCgroupCache(pc); | |
2126 | SetPageCgroupUsed(pc); | |
2127 | break; | |
2128 | default: | |
2129 | break; | |
2130 | } | |
3be91277 | 2131 | |
ca3e0214 | 2132 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages); |
52d4b9ac | 2133 | unlock_page_cgroup(pc); |
430e4863 KH |
2134 | /* |
2135 | * "charge_statistics" updated event counter. Then, check it. | |
2136 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2137 | * if they exceeds softlimit. | |
2138 | */ | |
d2265e6f | 2139 | memcg_check_events(mem, pc->page); |
7a81b88c | 2140 | } |
66e1707b | 2141 | |
ca3e0214 KH |
2142 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2143 | ||
2144 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
2145 | (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION)) | |
2146 | /* | |
2147 | * Because tail pages are not marked as "used", set it. We're under | |
2148 | * zone->lru_lock, 'splitting on pmd' and compund_lock. | |
2149 | */ | |
2150 | void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail) | |
2151 | { | |
2152 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
2153 | struct page_cgroup *tail_pc = lookup_page_cgroup(tail); | |
2154 | unsigned long flags; | |
2155 | ||
2156 | /* | |
ece35ca8 | 2157 | * We have no races with charge/uncharge but will have races with |
ca3e0214 KH |
2158 | * page state accounting. |
2159 | */ | |
2160 | move_lock_page_cgroup(head_pc, &flags); | |
2161 | ||
2162 | tail_pc->mem_cgroup = head_pc->mem_cgroup; | |
2163 | smp_wmb(); /* see __commit_charge() */ | |
ece35ca8 KH |
2164 | if (PageCgroupAcctLRU(head_pc)) { |
2165 | enum lru_list lru; | |
2166 | struct mem_cgroup_per_zone *mz; | |
2167 | ||
2168 | /* | |
2169 | * LRU flags cannot be copied because we need to add tail | |
2170 | *.page to LRU by generic call and our hook will be called. | |
2171 | * We hold lru_lock, then, reduce counter directly. | |
2172 | */ | |
2173 | lru = page_lru(head); | |
2174 | mz = page_cgroup_zoneinfo(head_pc); | |
2175 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; | |
2176 | } | |
ca3e0214 KH |
2177 | tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2178 | move_unlock_page_cgroup(head_pc, &flags); | |
2179 | } | |
2180 | #endif | |
2181 | ||
f817ed48 | 2182 | /** |
57f9fd7d | 2183 | * __mem_cgroup_move_account - move account of the page |
f817ed48 KH |
2184 | * @pc: page_cgroup of the page. |
2185 | * @from: mem_cgroup which the page is moved from. | |
2186 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2187 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2188 | * |
2189 | * The caller must confirm following. | |
08e552c6 | 2190 | * - page is not on LRU (isolate_page() is useful.) |
57f9fd7d | 2191 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
f817ed48 | 2192 | * |
854ffa8d DN |
2193 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
2194 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is | |
2195 | * true, this function does "uncharge" from old cgroup, but it doesn't if | |
2196 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 KH |
2197 | */ |
2198 | ||
57f9fd7d | 2199 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
854ffa8d | 2200 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
f817ed48 | 2201 | { |
f817ed48 | 2202 | VM_BUG_ON(from == to); |
08e552c6 | 2203 | VM_BUG_ON(PageLRU(pc->page)); |
112bc2e1 | 2204 | VM_BUG_ON(!page_is_cgroup_locked(pc)); |
57f9fd7d DN |
2205 | VM_BUG_ON(!PageCgroupUsed(pc)); |
2206 | VM_BUG_ON(pc->mem_cgroup != from); | |
f817ed48 | 2207 | |
8725d541 | 2208 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2209 | /* Update mapped_file data for mem_cgroup */ |
2210 | preempt_disable(); | |
2211 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2212 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2213 | preempt_enable(); | |
d69b042f | 2214 | } |
e401f176 | 2215 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -1); |
854ffa8d DN |
2216 | if (uncharge) |
2217 | /* This is not "cancel", but cancel_charge does all we need. */ | |
ec168510 | 2218 | mem_cgroup_cancel_charge(from, PAGE_SIZE); |
d69b042f | 2219 | |
854ffa8d | 2220 | /* caller should have done css_get */ |
08e552c6 | 2221 | pc->mem_cgroup = to; |
e401f176 | 2222 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), 1); |
88703267 KH |
2223 | /* |
2224 | * We charges against "to" which may not have any tasks. Then, "to" | |
2225 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe DN |
2226 | * this function is just force_empty() and move charge, so it's |
2227 | * garanteed that "to" is never removed. So, we don't check rmdir | |
2228 | * status here. | |
88703267 | 2229 | */ |
57f9fd7d DN |
2230 | } |
2231 | ||
2232 | /* | |
2233 | * check whether the @pc is valid for moving account and call | |
2234 | * __mem_cgroup_move_account() | |
2235 | */ | |
2236 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
854ffa8d | 2237 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
57f9fd7d DN |
2238 | { |
2239 | int ret = -EINVAL; | |
dbd4ea78 KH |
2240 | unsigned long flags; |
2241 | ||
57f9fd7d DN |
2242 | lock_page_cgroup(pc); |
2243 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { | |
dbd4ea78 | 2244 | move_lock_page_cgroup(pc, &flags); |
854ffa8d | 2245 | __mem_cgroup_move_account(pc, from, to, uncharge); |
dbd4ea78 | 2246 | move_unlock_page_cgroup(pc, &flags); |
57f9fd7d DN |
2247 | ret = 0; |
2248 | } | |
2249 | unlock_page_cgroup(pc); | |
d2265e6f KH |
2250 | /* |
2251 | * check events | |
2252 | */ | |
2253 | memcg_check_events(to, pc->page); | |
2254 | memcg_check_events(from, pc->page); | |
f817ed48 KH |
2255 | return ret; |
2256 | } | |
2257 | ||
2258 | /* | |
2259 | * move charges to its parent. | |
2260 | */ | |
2261 | ||
2262 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
2263 | struct mem_cgroup *child, | |
2264 | gfp_t gfp_mask) | |
2265 | { | |
08e552c6 | 2266 | struct page *page = pc->page; |
f817ed48 KH |
2267 | struct cgroup *cg = child->css.cgroup; |
2268 | struct cgroup *pcg = cg->parent; | |
2269 | struct mem_cgroup *parent; | |
f817ed48 KH |
2270 | int ret; |
2271 | ||
2272 | /* Is ROOT ? */ | |
2273 | if (!pcg) | |
2274 | return -EINVAL; | |
2275 | ||
57f9fd7d DN |
2276 | ret = -EBUSY; |
2277 | if (!get_page_unless_zero(page)) | |
2278 | goto out; | |
2279 | if (isolate_lru_page(page)) | |
2280 | goto put; | |
08e552c6 | 2281 | |
f817ed48 | 2282 | parent = mem_cgroup_from_cont(pcg); |
ec168510 AA |
2283 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, |
2284 | PAGE_SIZE); | |
a636b327 | 2285 | if (ret || !parent) |
57f9fd7d | 2286 | goto put_back; |
f817ed48 | 2287 | |
854ffa8d DN |
2288 | ret = mem_cgroup_move_account(pc, child, parent, true); |
2289 | if (ret) | |
ec168510 | 2290 | mem_cgroup_cancel_charge(parent, PAGE_SIZE); |
57f9fd7d | 2291 | put_back: |
08e552c6 | 2292 | putback_lru_page(page); |
57f9fd7d | 2293 | put: |
40d58138 | 2294 | put_page(page); |
57f9fd7d | 2295 | out: |
f817ed48 KH |
2296 | return ret; |
2297 | } | |
2298 | ||
7a81b88c KH |
2299 | /* |
2300 | * Charge the memory controller for page usage. | |
2301 | * Return | |
2302 | * 0 if the charge was successful | |
2303 | * < 0 if the cgroup is over its limit | |
2304 | */ | |
2305 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2306 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2307 | { |
73045c47 | 2308 | struct mem_cgroup *mem = NULL; |
7a81b88c KH |
2309 | struct page_cgroup *pc; |
2310 | int ret; | |
ec168510 AA |
2311 | int page_size = PAGE_SIZE; |
2312 | ||
37c2ac78 | 2313 | if (PageTransHuge(page)) { |
ec168510 | 2314 | page_size <<= compound_order(page); |
37c2ac78 AA |
2315 | VM_BUG_ON(!PageTransHuge(page)); |
2316 | } | |
7a81b88c KH |
2317 | |
2318 | pc = lookup_page_cgroup(page); | |
2319 | /* can happen at boot */ | |
2320 | if (unlikely(!pc)) | |
2321 | return 0; | |
2322 | prefetchw(pc); | |
2323 | ||
ec168510 | 2324 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page_size); |
a636b327 | 2325 | if (ret || !mem) |
7a81b88c KH |
2326 | return ret; |
2327 | ||
ec168510 | 2328 | __mem_cgroup_commit_charge(mem, pc, ctype, page_size); |
8a9f3ccd | 2329 | return 0; |
8a9f3ccd BS |
2330 | } |
2331 | ||
7a81b88c KH |
2332 | int mem_cgroup_newpage_charge(struct page *page, |
2333 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2334 | { |
f8d66542 | 2335 | if (mem_cgroup_disabled()) |
cede86ac | 2336 | return 0; |
69029cd5 KH |
2337 | /* |
2338 | * If already mapped, we don't have to account. | |
2339 | * If page cache, page->mapping has address_space. | |
2340 | * But page->mapping may have out-of-use anon_vma pointer, | |
2341 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
2342 | * is NULL. | |
2343 | */ | |
2344 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2345 | return 0; | |
2346 | if (unlikely(!mm)) | |
2347 | mm = &init_mm; | |
217bc319 | 2348 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
73045c47 | 2349 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2350 | } |
2351 | ||
83aae4c7 DN |
2352 | static void |
2353 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2354 | enum charge_type ctype); | |
2355 | ||
e1a1cd59 BS |
2356 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2357 | gfp_t gfp_mask) | |
8697d331 | 2358 | { |
b5a84319 KH |
2359 | int ret; |
2360 | ||
f8d66542 | 2361 | if (mem_cgroup_disabled()) |
cede86ac | 2362 | return 0; |
52d4b9ac KH |
2363 | if (PageCompound(page)) |
2364 | return 0; | |
accf163e KH |
2365 | /* |
2366 | * Corner case handling. This is called from add_to_page_cache() | |
2367 | * in usual. But some FS (shmem) precharges this page before calling it | |
2368 | * and call add_to_page_cache() with GFP_NOWAIT. | |
2369 | * | |
2370 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
2371 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
2372 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
2373 | * And when the page is SwapCache, it should take swap information |
2374 | * into account. This is under lock_page() now. | |
accf163e KH |
2375 | */ |
2376 | if (!(gfp_mask & __GFP_WAIT)) { | |
2377 | struct page_cgroup *pc; | |
2378 | ||
52d4b9ac KH |
2379 | pc = lookup_page_cgroup(page); |
2380 | if (!pc) | |
2381 | return 0; | |
2382 | lock_page_cgroup(pc); | |
2383 | if (PageCgroupUsed(pc)) { | |
2384 | unlock_page_cgroup(pc); | |
accf163e KH |
2385 | return 0; |
2386 | } | |
52d4b9ac | 2387 | unlock_page_cgroup(pc); |
accf163e KH |
2388 | } |
2389 | ||
73045c47 | 2390 | if (unlikely(!mm)) |
8697d331 | 2391 | mm = &init_mm; |
accf163e | 2392 | |
c05555b5 KH |
2393 | if (page_is_file_cache(page)) |
2394 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2395 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
b5a84319 | 2396 | |
83aae4c7 DN |
2397 | /* shmem */ |
2398 | if (PageSwapCache(page)) { | |
73045c47 DN |
2399 | struct mem_cgroup *mem = NULL; |
2400 | ||
83aae4c7 DN |
2401 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2402 | if (!ret) | |
2403 | __mem_cgroup_commit_charge_swapin(page, mem, | |
2404 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
2405 | } else | |
2406 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2407 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
b5a84319 | 2408 | |
b5a84319 | 2409 | return ret; |
e8589cc1 KH |
2410 | } |
2411 | ||
54595fe2 KH |
2412 | /* |
2413 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2414 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2415 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2416 | * "commit()" or removed by "cancel()" |
2417 | */ | |
8c7c6e34 KH |
2418 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2419 | struct page *page, | |
2420 | gfp_t mask, struct mem_cgroup **ptr) | |
2421 | { | |
2422 | struct mem_cgroup *mem; | |
54595fe2 | 2423 | int ret; |
8c7c6e34 | 2424 | |
f8d66542 | 2425 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2426 | return 0; |
2427 | ||
2428 | if (!do_swap_account) | |
2429 | goto charge_cur_mm; | |
8c7c6e34 KH |
2430 | /* |
2431 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2432 | * the pte, and even removed page from swap cache: in those cases |
2433 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2434 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2435 | */ |
2436 | if (!PageSwapCache(page)) | |
407f9c8b | 2437 | goto charge_cur_mm; |
e42d9d5d | 2438 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
2439 | if (!mem) |
2440 | goto charge_cur_mm; | |
8c7c6e34 | 2441 | *ptr = mem; |
ec168510 | 2442 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, PAGE_SIZE); |
54595fe2 KH |
2443 | css_put(&mem->css); |
2444 | return ret; | |
8c7c6e34 KH |
2445 | charge_cur_mm: |
2446 | if (unlikely(!mm)) | |
2447 | mm = &init_mm; | |
ec168510 | 2448 | return __mem_cgroup_try_charge(mm, mask, ptr, true, PAGE_SIZE); |
8c7c6e34 KH |
2449 | } |
2450 | ||
83aae4c7 DN |
2451 | static void |
2452 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2453 | enum charge_type ctype) | |
7a81b88c KH |
2454 | { |
2455 | struct page_cgroup *pc; | |
2456 | ||
f8d66542 | 2457 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2458 | return; |
2459 | if (!ptr) | |
2460 | return; | |
88703267 | 2461 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 2462 | pc = lookup_page_cgroup(page); |
544122e5 | 2463 | mem_cgroup_lru_del_before_commit_swapcache(page); |
ec168510 | 2464 | __mem_cgroup_commit_charge(ptr, pc, ctype, PAGE_SIZE); |
544122e5 | 2465 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
2466 | /* |
2467 | * Now swap is on-memory. This means this page may be | |
2468 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2469 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2470 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2471 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2472 | */ |
03f3c433 | 2473 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2474 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2475 | unsigned short id; |
8c7c6e34 | 2476 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2477 | |
2478 | id = swap_cgroup_record(ent, 0); | |
2479 | rcu_read_lock(); | |
2480 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2481 | if (memcg) { |
a3b2d692 KH |
2482 | /* |
2483 | * This recorded memcg can be obsolete one. So, avoid | |
2484 | * calling css_tryget | |
2485 | */ | |
0c3e73e8 | 2486 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2487 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2488 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2489 | mem_cgroup_put(memcg); |
2490 | } | |
a3b2d692 | 2491 | rcu_read_unlock(); |
8c7c6e34 | 2492 | } |
88703267 KH |
2493 | /* |
2494 | * At swapin, we may charge account against cgroup which has no tasks. | |
2495 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2496 | * In that case, we need to call pre_destroy() again. check it here. | |
2497 | */ | |
2498 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2499 | } |
2500 | ||
83aae4c7 DN |
2501 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2502 | { | |
2503 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2504 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2505 | } | |
2506 | ||
7a81b88c KH |
2507 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2508 | { | |
f8d66542 | 2509 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2510 | return; |
2511 | if (!mem) | |
2512 | return; | |
ec168510 | 2513 | mem_cgroup_cancel_charge(mem, PAGE_SIZE); |
7a81b88c KH |
2514 | } |
2515 | ||
569b846d | 2516 | static void |
ec168510 AA |
2517 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype, |
2518 | int page_size) | |
569b846d KH |
2519 | { |
2520 | struct memcg_batch_info *batch = NULL; | |
2521 | bool uncharge_memsw = true; | |
2522 | /* If swapout, usage of swap doesn't decrease */ | |
2523 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2524 | uncharge_memsw = false; | |
569b846d KH |
2525 | |
2526 | batch = ¤t->memcg_batch; | |
2527 | /* | |
2528 | * In usual, we do css_get() when we remember memcg pointer. | |
2529 | * But in this case, we keep res->usage until end of a series of | |
2530 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2531 | */ | |
2532 | if (!batch->memcg) | |
2533 | batch->memcg = mem; | |
3c11ecf4 KH |
2534 | /* |
2535 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
2536 | * In those cases, all pages freed continously can be expected to be in | |
2537 | * the same cgroup and we have chance to coalesce uncharges. | |
2538 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2539 | * because we want to do uncharge as soon as possible. | |
2540 | */ | |
2541 | ||
2542 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2543 | goto direct_uncharge; | |
2544 | ||
ec168510 AA |
2545 | if (page_size != PAGE_SIZE) |
2546 | goto direct_uncharge; | |
2547 | ||
569b846d KH |
2548 | /* |
2549 | * In typical case, batch->memcg == mem. This means we can | |
2550 | * merge a series of uncharges to an uncharge of res_counter. | |
2551 | * If not, we uncharge res_counter ony by one. | |
2552 | */ | |
2553 | if (batch->memcg != mem) | |
2554 | goto direct_uncharge; | |
2555 | /* remember freed charge and uncharge it later */ | |
2556 | batch->bytes += PAGE_SIZE; | |
2557 | if (uncharge_memsw) | |
2558 | batch->memsw_bytes += PAGE_SIZE; | |
2559 | return; | |
2560 | direct_uncharge: | |
ec168510 | 2561 | res_counter_uncharge(&mem->res, page_size); |
569b846d | 2562 | if (uncharge_memsw) |
ec168510 | 2563 | res_counter_uncharge(&mem->memsw, page_size); |
3c11ecf4 KH |
2564 | if (unlikely(batch->memcg != mem)) |
2565 | memcg_oom_recover(mem); | |
569b846d KH |
2566 | return; |
2567 | } | |
7a81b88c | 2568 | |
8a9f3ccd | 2569 | /* |
69029cd5 | 2570 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2571 | */ |
8c7c6e34 | 2572 | static struct mem_cgroup * |
69029cd5 | 2573 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2574 | { |
152c9ccb | 2575 | int count; |
8289546e | 2576 | struct page_cgroup *pc; |
8c7c6e34 | 2577 | struct mem_cgroup *mem = NULL; |
ec168510 | 2578 | int page_size = PAGE_SIZE; |
8a9f3ccd | 2579 | |
f8d66542 | 2580 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2581 | return NULL; |
4077960e | 2582 | |
d13d1443 | 2583 | if (PageSwapCache(page)) |
8c7c6e34 | 2584 | return NULL; |
d13d1443 | 2585 | |
37c2ac78 | 2586 | if (PageTransHuge(page)) { |
ec168510 | 2587 | page_size <<= compound_order(page); |
37c2ac78 AA |
2588 | VM_BUG_ON(!PageTransHuge(page)); |
2589 | } | |
ec168510 | 2590 | |
152c9ccb | 2591 | count = page_size >> PAGE_SHIFT; |
8697d331 | 2592 | /* |
3c541e14 | 2593 | * Check if our page_cgroup is valid |
8697d331 | 2594 | */ |
52d4b9ac KH |
2595 | pc = lookup_page_cgroup(page); |
2596 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2597 | return NULL; |
b9c565d5 | 2598 | |
52d4b9ac | 2599 | lock_page_cgroup(pc); |
d13d1443 | 2600 | |
8c7c6e34 KH |
2601 | mem = pc->mem_cgroup; |
2602 | ||
d13d1443 KH |
2603 | if (!PageCgroupUsed(pc)) |
2604 | goto unlock_out; | |
2605 | ||
2606 | switch (ctype) { | |
2607 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2608 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c AM |
2609 | /* See mem_cgroup_prepare_migration() */ |
2610 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2611 | goto unlock_out; |
2612 | break; | |
2613 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2614 | if (!PageAnon(page)) { /* Shared memory */ | |
2615 | if (page->mapping && !page_is_file_cache(page)) | |
2616 | goto unlock_out; | |
2617 | } else if (page_mapped(page)) /* Anon */ | |
2618 | goto unlock_out; | |
2619 | break; | |
2620 | default: | |
2621 | break; | |
52d4b9ac | 2622 | } |
d13d1443 | 2623 | |
ca3e0214 | 2624 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -count); |
04046e1a | 2625 | |
52d4b9ac | 2626 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2627 | /* |
2628 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2629 | * freed from LRU. This is safe because uncharged page is expected not | |
2630 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2631 | * special functions. | |
2632 | */ | |
b9c565d5 | 2633 | |
52d4b9ac | 2634 | unlock_page_cgroup(pc); |
f75ca962 KH |
2635 | /* |
2636 | * even after unlock, we have mem->res.usage here and this memcg | |
2637 | * will never be freed. | |
2638 | */ | |
d2265e6f | 2639 | memcg_check_events(mem, page); |
f75ca962 KH |
2640 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
2641 | mem_cgroup_swap_statistics(mem, true); | |
2642 | mem_cgroup_get(mem); | |
2643 | } | |
2644 | if (!mem_cgroup_is_root(mem)) | |
ec168510 | 2645 | __do_uncharge(mem, ctype, page_size); |
6d12e2d8 | 2646 | |
8c7c6e34 | 2647 | return mem; |
d13d1443 KH |
2648 | |
2649 | unlock_out: | |
2650 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2651 | return NULL; |
3c541e14 BS |
2652 | } |
2653 | ||
69029cd5 KH |
2654 | void mem_cgroup_uncharge_page(struct page *page) |
2655 | { | |
52d4b9ac KH |
2656 | /* early check. */ |
2657 | if (page_mapped(page)) | |
2658 | return; | |
2659 | if (page->mapping && !PageAnon(page)) | |
2660 | return; | |
69029cd5 KH |
2661 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2662 | } | |
2663 | ||
2664 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2665 | { | |
2666 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2667 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2668 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2669 | } | |
2670 | ||
569b846d KH |
2671 | /* |
2672 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2673 | * In that cases, pages are freed continuously and we can expect pages | |
2674 | * are in the same memcg. All these calls itself limits the number of | |
2675 | * pages freed at once, then uncharge_start/end() is called properly. | |
2676 | * This may be called prural(2) times in a context, | |
2677 | */ | |
2678 | ||
2679 | void mem_cgroup_uncharge_start(void) | |
2680 | { | |
2681 | current->memcg_batch.do_batch++; | |
2682 | /* We can do nest. */ | |
2683 | if (current->memcg_batch.do_batch == 1) { | |
2684 | current->memcg_batch.memcg = NULL; | |
2685 | current->memcg_batch.bytes = 0; | |
2686 | current->memcg_batch.memsw_bytes = 0; | |
2687 | } | |
2688 | } | |
2689 | ||
2690 | void mem_cgroup_uncharge_end(void) | |
2691 | { | |
2692 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2693 | ||
2694 | if (!batch->do_batch) | |
2695 | return; | |
2696 | ||
2697 | batch->do_batch--; | |
2698 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2699 | return; | |
2700 | ||
2701 | if (!batch->memcg) | |
2702 | return; | |
2703 | /* | |
2704 | * This "batch->memcg" is valid without any css_get/put etc... | |
2705 | * bacause we hide charges behind us. | |
2706 | */ | |
2707 | if (batch->bytes) | |
2708 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2709 | if (batch->memsw_bytes) | |
2710 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
3c11ecf4 | 2711 | memcg_oom_recover(batch->memcg); |
569b846d KH |
2712 | /* forget this pointer (for sanity check) */ |
2713 | batch->memcg = NULL; | |
2714 | } | |
2715 | ||
e767e056 | 2716 | #ifdef CONFIG_SWAP |
8c7c6e34 | 2717 | /* |
e767e056 | 2718 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
2719 | * memcg information is recorded to swap_cgroup of "ent" |
2720 | */ | |
8a9478ca KH |
2721 | void |
2722 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
2723 | { |
2724 | struct mem_cgroup *memcg; | |
8a9478ca KH |
2725 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
2726 | ||
2727 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2728 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2729 | ||
2730 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 2731 | |
f75ca962 KH |
2732 | /* |
2733 | * record memcg information, if swapout && memcg != NULL, | |
2734 | * mem_cgroup_get() was called in uncharge(). | |
2735 | */ | |
2736 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 2737 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 2738 | } |
e767e056 | 2739 | #endif |
8c7c6e34 KH |
2740 | |
2741 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2742 | /* | |
2743 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2744 | * uncharge "memsw" account. | |
2745 | */ | |
2746 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 2747 | { |
8c7c6e34 | 2748 | struct mem_cgroup *memcg; |
a3b2d692 | 2749 | unsigned short id; |
8c7c6e34 KH |
2750 | |
2751 | if (!do_swap_account) | |
2752 | return; | |
2753 | ||
a3b2d692 KH |
2754 | id = swap_cgroup_record(ent, 0); |
2755 | rcu_read_lock(); | |
2756 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2757 | if (memcg) { |
a3b2d692 KH |
2758 | /* |
2759 | * We uncharge this because swap is freed. | |
2760 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2761 | */ | |
0c3e73e8 | 2762 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2763 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2764 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2765 | mem_cgroup_put(memcg); |
2766 | } | |
a3b2d692 | 2767 | rcu_read_unlock(); |
d13d1443 | 2768 | } |
02491447 DN |
2769 | |
2770 | /** | |
2771 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2772 | * @entry: swap entry to be moved | |
2773 | * @from: mem_cgroup which the entry is moved from | |
2774 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 2775 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
2776 | * |
2777 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2778 | * as the mem_cgroup's id of @from. | |
2779 | * | |
2780 | * Returns 0 on success, -EINVAL on failure. | |
2781 | * | |
2782 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
2783 | * both res and memsw, and called css_get(). | |
2784 | */ | |
2785 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2786 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2787 | { |
2788 | unsigned short old_id, new_id; | |
2789 | ||
2790 | old_id = css_id(&from->css); | |
2791 | new_id = css_id(&to->css); | |
2792 | ||
2793 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2794 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2795 | mem_cgroup_swap_statistics(to, true); |
02491447 | 2796 | /* |
483c30b5 DN |
2797 | * This function is only called from task migration context now. |
2798 | * It postpones res_counter and refcount handling till the end | |
2799 | * of task migration(mem_cgroup_clear_mc()) for performance | |
2800 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
2801 | * because if the process that has been moved to @to does | |
2802 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 2803 | */ |
02491447 | 2804 | mem_cgroup_get(to); |
483c30b5 DN |
2805 | if (need_fixup) { |
2806 | if (!mem_cgroup_is_root(from)) | |
2807 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
2808 | mem_cgroup_put(from); | |
2809 | /* | |
2810 | * we charged both to->res and to->memsw, so we should | |
2811 | * uncharge to->res. | |
2812 | */ | |
2813 | if (!mem_cgroup_is_root(to)) | |
2814 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 2815 | } |
02491447 DN |
2816 | return 0; |
2817 | } | |
2818 | return -EINVAL; | |
2819 | } | |
2820 | #else | |
2821 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2822 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2823 | { |
2824 | return -EINVAL; | |
2825 | } | |
8c7c6e34 | 2826 | #endif |
d13d1443 | 2827 | |
ae41be37 | 2828 | /* |
01b1ae63 KH |
2829 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
2830 | * page belongs to. | |
ae41be37 | 2831 | */ |
ac39cf8c AM |
2832 | int mem_cgroup_prepare_migration(struct page *page, |
2833 | struct page *newpage, struct mem_cgroup **ptr) | |
ae41be37 KH |
2834 | { |
2835 | struct page_cgroup *pc; | |
e8589cc1 | 2836 | struct mem_cgroup *mem = NULL; |
ac39cf8c | 2837 | enum charge_type ctype; |
e8589cc1 | 2838 | int ret = 0; |
8869b8f6 | 2839 | |
ec168510 | 2840 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 2841 | if (mem_cgroup_disabled()) |
4077960e BS |
2842 | return 0; |
2843 | ||
52d4b9ac KH |
2844 | pc = lookup_page_cgroup(page); |
2845 | lock_page_cgroup(pc); | |
2846 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
2847 | mem = pc->mem_cgroup; |
2848 | css_get(&mem->css); | |
ac39cf8c AM |
2849 | /* |
2850 | * At migrating an anonymous page, its mapcount goes down | |
2851 | * to 0 and uncharge() will be called. But, even if it's fully | |
2852 | * unmapped, migration may fail and this page has to be | |
2853 | * charged again. We set MIGRATION flag here and delay uncharge | |
2854 | * until end_migration() is called | |
2855 | * | |
2856 | * Corner Case Thinking | |
2857 | * A) | |
2858 | * When the old page was mapped as Anon and it's unmap-and-freed | |
2859 | * while migration was ongoing. | |
2860 | * If unmap finds the old page, uncharge() of it will be delayed | |
2861 | * until end_migration(). If unmap finds a new page, it's | |
2862 | * uncharged when it make mapcount to be 1->0. If unmap code | |
2863 | * finds swap_migration_entry, the new page will not be mapped | |
2864 | * and end_migration() will find it(mapcount==0). | |
2865 | * | |
2866 | * B) | |
2867 | * When the old page was mapped but migraion fails, the kernel | |
2868 | * remaps it. A charge for it is kept by MIGRATION flag even | |
2869 | * if mapcount goes down to 0. We can do remap successfully | |
2870 | * without charging it again. | |
2871 | * | |
2872 | * C) | |
2873 | * The "old" page is under lock_page() until the end of | |
2874 | * migration, so, the old page itself will not be swapped-out. | |
2875 | * If the new page is swapped out before end_migraton, our | |
2876 | * hook to usual swap-out path will catch the event. | |
2877 | */ | |
2878 | if (PageAnon(page)) | |
2879 | SetPageCgroupMigration(pc); | |
e8589cc1 | 2880 | } |
52d4b9ac | 2881 | unlock_page_cgroup(pc); |
ac39cf8c AM |
2882 | /* |
2883 | * If the page is not charged at this point, | |
2884 | * we return here. | |
2885 | */ | |
2886 | if (!mem) | |
2887 | return 0; | |
01b1ae63 | 2888 | |
93d5c9be | 2889 | *ptr = mem; |
ec168510 | 2890 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, ptr, false, PAGE_SIZE); |
ac39cf8c AM |
2891 | css_put(&mem->css);/* drop extra refcnt */ |
2892 | if (ret || *ptr == NULL) { | |
2893 | if (PageAnon(page)) { | |
2894 | lock_page_cgroup(pc); | |
2895 | ClearPageCgroupMigration(pc); | |
2896 | unlock_page_cgroup(pc); | |
2897 | /* | |
2898 | * The old page may be fully unmapped while we kept it. | |
2899 | */ | |
2900 | mem_cgroup_uncharge_page(page); | |
2901 | } | |
2902 | return -ENOMEM; | |
e8589cc1 | 2903 | } |
ac39cf8c AM |
2904 | /* |
2905 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
2906 | * is called before end_migration, we can catch all events on this new | |
2907 | * page. In the case new page is migrated but not remapped, new page's | |
2908 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
2909 | */ | |
2910 | pc = lookup_page_cgroup(newpage); | |
2911 | if (PageAnon(page)) | |
2912 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2913 | else if (page_is_file_cache(page)) | |
2914 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2915 | else | |
2916 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
ec168510 | 2917 | __mem_cgroup_commit_charge(mem, pc, ctype, PAGE_SIZE); |
e8589cc1 | 2918 | return ret; |
ae41be37 | 2919 | } |
8869b8f6 | 2920 | |
69029cd5 | 2921 | /* remove redundant charge if migration failed*/ |
01b1ae63 | 2922 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
50de1dd9 | 2923 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 2924 | { |
ac39cf8c | 2925 | struct page *used, *unused; |
01b1ae63 | 2926 | struct page_cgroup *pc; |
01b1ae63 KH |
2927 | |
2928 | if (!mem) | |
2929 | return; | |
ac39cf8c | 2930 | /* blocks rmdir() */ |
88703267 | 2931 | cgroup_exclude_rmdir(&mem->css); |
50de1dd9 | 2932 | if (!migration_ok) { |
ac39cf8c AM |
2933 | used = oldpage; |
2934 | unused = newpage; | |
01b1ae63 | 2935 | } else { |
ac39cf8c | 2936 | used = newpage; |
01b1ae63 KH |
2937 | unused = oldpage; |
2938 | } | |
69029cd5 | 2939 | /* |
ac39cf8c AM |
2940 | * We disallowed uncharge of pages under migration because mapcount |
2941 | * of the page goes down to zero, temporarly. | |
2942 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 2943 | */ |
ac39cf8c AM |
2944 | pc = lookup_page_cgroup(oldpage); |
2945 | lock_page_cgroup(pc); | |
2946 | ClearPageCgroupMigration(pc); | |
2947 | unlock_page_cgroup(pc); | |
01b1ae63 | 2948 | |
ac39cf8c AM |
2949 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
2950 | ||
01b1ae63 | 2951 | /* |
ac39cf8c AM |
2952 | * If a page is a file cache, radix-tree replacement is very atomic |
2953 | * and we can skip this check. When it was an Anon page, its mapcount | |
2954 | * goes down to 0. But because we added MIGRATION flage, it's not | |
2955 | * uncharged yet. There are several case but page->mapcount check | |
2956 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
2957 | * check. (see prepare_charge() also) | |
69029cd5 | 2958 | */ |
ac39cf8c AM |
2959 | if (PageAnon(used)) |
2960 | mem_cgroup_uncharge_page(used); | |
88703267 | 2961 | /* |
ac39cf8c AM |
2962 | * At migration, we may charge account against cgroup which has no |
2963 | * tasks. | |
88703267 KH |
2964 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
2965 | * In that case, we need to call pre_destroy() again. check it here. | |
2966 | */ | |
2967 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 2968 | } |
78fb7466 | 2969 | |
c9b0ed51 | 2970 | /* |
ae3abae6 DN |
2971 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
2972 | * Calling hierarchical_reclaim is not enough because we should update | |
2973 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
2974 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
2975 | * not from the memcg which this page would be charged to. | |
2976 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 2977 | */ |
ae3abae6 | 2978 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
2979 | struct mm_struct *mm, |
2980 | gfp_t gfp_mask) | |
c9b0ed51 | 2981 | { |
b5a84319 | 2982 | struct mem_cgroup *mem = NULL; |
ae3abae6 | 2983 | int ret; |
c9b0ed51 | 2984 | |
f8d66542 | 2985 | if (mem_cgroup_disabled()) |
cede86ac | 2986 | return 0; |
c9b0ed51 | 2987 | |
ae3abae6 DN |
2988 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2989 | if (!ret) | |
2990 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 2991 | |
ae3abae6 | 2992 | return ret; |
c9b0ed51 KH |
2993 | } |
2994 | ||
8c7c6e34 KH |
2995 | static DEFINE_MUTEX(set_limit_mutex); |
2996 | ||
d38d2a75 | 2997 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 2998 | unsigned long long val) |
628f4235 | 2999 | { |
81d39c20 | 3000 | int retry_count; |
3c11ecf4 | 3001 | u64 memswlimit, memlimit; |
628f4235 | 3002 | int ret = 0; |
81d39c20 KH |
3003 | int children = mem_cgroup_count_children(memcg); |
3004 | u64 curusage, oldusage; | |
3c11ecf4 | 3005 | int enlarge; |
81d39c20 KH |
3006 | |
3007 | /* | |
3008 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3009 | * is depends on callers. We set our retry-count to be function | |
3010 | * of # of children which we should visit in this loop. | |
3011 | */ | |
3012 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3013 | ||
3014 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3015 | |
3c11ecf4 | 3016 | enlarge = 0; |
8c7c6e34 | 3017 | while (retry_count) { |
628f4235 KH |
3018 | if (signal_pending(current)) { |
3019 | ret = -EINTR; | |
3020 | break; | |
3021 | } | |
8c7c6e34 KH |
3022 | /* |
3023 | * Rather than hide all in some function, I do this in | |
3024 | * open coded manner. You see what this really does. | |
3025 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3026 | */ | |
3027 | mutex_lock(&set_limit_mutex); | |
3028 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3029 | if (memswlimit < val) { | |
3030 | ret = -EINVAL; | |
3031 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3032 | break; |
3033 | } | |
3c11ecf4 KH |
3034 | |
3035 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3036 | if (memlimit < val) | |
3037 | enlarge = 1; | |
3038 | ||
8c7c6e34 | 3039 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3040 | if (!ret) { |
3041 | if (memswlimit == val) | |
3042 | memcg->memsw_is_minimum = true; | |
3043 | else | |
3044 | memcg->memsw_is_minimum = false; | |
3045 | } | |
8c7c6e34 KH |
3046 | mutex_unlock(&set_limit_mutex); |
3047 | ||
3048 | if (!ret) | |
3049 | break; | |
3050 | ||
aa20d489 | 3051 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
4e416953 | 3052 | MEM_CGROUP_RECLAIM_SHRINK); |
81d39c20 KH |
3053 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3054 | /* Usage is reduced ? */ | |
3055 | if (curusage >= oldusage) | |
3056 | retry_count--; | |
3057 | else | |
3058 | oldusage = curusage; | |
8c7c6e34 | 3059 | } |
3c11ecf4 KH |
3060 | if (!ret && enlarge) |
3061 | memcg_oom_recover(memcg); | |
14797e23 | 3062 | |
8c7c6e34 KH |
3063 | return ret; |
3064 | } | |
3065 | ||
338c8431 LZ |
3066 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3067 | unsigned long long val) | |
8c7c6e34 | 3068 | { |
81d39c20 | 3069 | int retry_count; |
3c11ecf4 | 3070 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3071 | int children = mem_cgroup_count_children(memcg); |
3072 | int ret = -EBUSY; | |
3c11ecf4 | 3073 | int enlarge = 0; |
8c7c6e34 | 3074 | |
81d39c20 KH |
3075 | /* see mem_cgroup_resize_res_limit */ |
3076 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3077 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3078 | while (retry_count) { |
3079 | if (signal_pending(current)) { | |
3080 | ret = -EINTR; | |
3081 | break; | |
3082 | } | |
3083 | /* | |
3084 | * Rather than hide all in some function, I do this in | |
3085 | * open coded manner. You see what this really does. | |
3086 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3087 | */ | |
3088 | mutex_lock(&set_limit_mutex); | |
3089 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3090 | if (memlimit > val) { | |
3091 | ret = -EINVAL; | |
3092 | mutex_unlock(&set_limit_mutex); | |
3093 | break; | |
3094 | } | |
3c11ecf4 KH |
3095 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3096 | if (memswlimit < val) | |
3097 | enlarge = 1; | |
8c7c6e34 | 3098 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3099 | if (!ret) { |
3100 | if (memlimit == val) | |
3101 | memcg->memsw_is_minimum = true; | |
3102 | else | |
3103 | memcg->memsw_is_minimum = false; | |
3104 | } | |
8c7c6e34 KH |
3105 | mutex_unlock(&set_limit_mutex); |
3106 | ||
3107 | if (!ret) | |
3108 | break; | |
3109 | ||
4e416953 | 3110 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 BS |
3111 | MEM_CGROUP_RECLAIM_NOSWAP | |
3112 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3113 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3114 | /* Usage is reduced ? */ |
8c7c6e34 | 3115 | if (curusage >= oldusage) |
628f4235 | 3116 | retry_count--; |
81d39c20 KH |
3117 | else |
3118 | oldusage = curusage; | |
628f4235 | 3119 | } |
3c11ecf4 KH |
3120 | if (!ret && enlarge) |
3121 | memcg_oom_recover(memcg); | |
628f4235 KH |
3122 | return ret; |
3123 | } | |
3124 | ||
4e416953 | 3125 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
00918b6a | 3126 | gfp_t gfp_mask) |
4e416953 BS |
3127 | { |
3128 | unsigned long nr_reclaimed = 0; | |
3129 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3130 | unsigned long reclaimed; | |
3131 | int loop = 0; | |
3132 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3133 | unsigned long long excess; |
4e416953 BS |
3134 | |
3135 | if (order > 0) | |
3136 | return 0; | |
3137 | ||
00918b6a | 3138 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3139 | /* |
3140 | * This loop can run a while, specially if mem_cgroup's continuously | |
3141 | * keep exceeding their soft limit and putting the system under | |
3142 | * pressure | |
3143 | */ | |
3144 | do { | |
3145 | if (next_mz) | |
3146 | mz = next_mz; | |
3147 | else | |
3148 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3149 | if (!mz) | |
3150 | break; | |
3151 | ||
3152 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
3153 | gfp_mask, | |
3154 | MEM_CGROUP_RECLAIM_SOFT); | |
3155 | nr_reclaimed += reclaimed; | |
3156 | spin_lock(&mctz->lock); | |
3157 | ||
3158 | /* | |
3159 | * If we failed to reclaim anything from this memory cgroup | |
3160 | * it is time to move on to the next cgroup | |
3161 | */ | |
3162 | next_mz = NULL; | |
3163 | if (!reclaimed) { | |
3164 | do { | |
3165 | /* | |
3166 | * Loop until we find yet another one. | |
3167 | * | |
3168 | * By the time we get the soft_limit lock | |
3169 | * again, someone might have aded the | |
3170 | * group back on the RB tree. Iterate to | |
3171 | * make sure we get a different mem. | |
3172 | * mem_cgroup_largest_soft_limit_node returns | |
3173 | * NULL if no other cgroup is present on | |
3174 | * the tree | |
3175 | */ | |
3176 | next_mz = | |
3177 | __mem_cgroup_largest_soft_limit_node(mctz); | |
3178 | if (next_mz == mz) { | |
3179 | css_put(&next_mz->mem->css); | |
3180 | next_mz = NULL; | |
3181 | } else /* next_mz == NULL or other memcg */ | |
3182 | break; | |
3183 | } while (1); | |
3184 | } | |
4e416953 | 3185 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 3186 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
3187 | /* |
3188 | * One school of thought says that we should not add | |
3189 | * back the node to the tree if reclaim returns 0. | |
3190 | * But our reclaim could return 0, simply because due | |
3191 | * to priority we are exposing a smaller subset of | |
3192 | * memory to reclaim from. Consider this as a longer | |
3193 | * term TODO. | |
3194 | */ | |
ef8745c1 KH |
3195 | /* If excess == 0, no tree ops */ |
3196 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
3197 | spin_unlock(&mctz->lock); |
3198 | css_put(&mz->mem->css); | |
3199 | loop++; | |
3200 | /* | |
3201 | * Could not reclaim anything and there are no more | |
3202 | * mem cgroups to try or we seem to be looping without | |
3203 | * reclaiming anything. | |
3204 | */ | |
3205 | if (!nr_reclaimed && | |
3206 | (next_mz == NULL || | |
3207 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3208 | break; | |
3209 | } while (!nr_reclaimed); | |
3210 | if (next_mz) | |
3211 | css_put(&next_mz->mem->css); | |
3212 | return nr_reclaimed; | |
3213 | } | |
3214 | ||
cc847582 KH |
3215 | /* |
3216 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3217 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3218 | */ | |
f817ed48 | 3219 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 3220 | int node, int zid, enum lru_list lru) |
cc847582 | 3221 | { |
08e552c6 KH |
3222 | struct zone *zone; |
3223 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 3224 | struct page_cgroup *pc, *busy; |
08e552c6 | 3225 | unsigned long flags, loop; |
072c56c1 | 3226 | struct list_head *list; |
f817ed48 | 3227 | int ret = 0; |
072c56c1 | 3228 | |
08e552c6 KH |
3229 | zone = &NODE_DATA(node)->node_zones[zid]; |
3230 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 3231 | list = &mz->lists[lru]; |
cc847582 | 3232 | |
f817ed48 KH |
3233 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
3234 | /* give some margin against EBUSY etc...*/ | |
3235 | loop += 256; | |
3236 | busy = NULL; | |
3237 | while (loop--) { | |
3238 | ret = 0; | |
08e552c6 | 3239 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3240 | if (list_empty(list)) { |
08e552c6 | 3241 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3242 | break; |
f817ed48 KH |
3243 | } |
3244 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
3245 | if (busy == pc) { | |
3246 | list_move(&pc->lru, list); | |
648bcc77 | 3247 | busy = NULL; |
08e552c6 | 3248 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3249 | continue; |
3250 | } | |
08e552c6 | 3251 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3252 | |
2c26fdd7 | 3253 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 3254 | if (ret == -ENOMEM) |
52d4b9ac | 3255 | break; |
f817ed48 KH |
3256 | |
3257 | if (ret == -EBUSY || ret == -EINVAL) { | |
3258 | /* found lock contention or "pc" is obsolete. */ | |
3259 | busy = pc; | |
3260 | cond_resched(); | |
3261 | } else | |
3262 | busy = NULL; | |
cc847582 | 3263 | } |
08e552c6 | 3264 | |
f817ed48 KH |
3265 | if (!ret && !list_empty(list)) |
3266 | return -EBUSY; | |
3267 | return ret; | |
cc847582 KH |
3268 | } |
3269 | ||
3270 | /* | |
3271 | * make mem_cgroup's charge to be 0 if there is no task. | |
3272 | * This enables deleting this mem_cgroup. | |
3273 | */ | |
c1e862c1 | 3274 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 3275 | { |
f817ed48 KH |
3276 | int ret; |
3277 | int node, zid, shrink; | |
3278 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 3279 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 3280 | |
cc847582 | 3281 | css_get(&mem->css); |
f817ed48 KH |
3282 | |
3283 | shrink = 0; | |
c1e862c1 KH |
3284 | /* should free all ? */ |
3285 | if (free_all) | |
3286 | goto try_to_free; | |
f817ed48 | 3287 | move_account: |
fce66477 | 3288 | do { |
f817ed48 | 3289 | ret = -EBUSY; |
c1e862c1 KH |
3290 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3291 | goto out; | |
3292 | ret = -EINTR; | |
3293 | if (signal_pending(current)) | |
cc847582 | 3294 | goto out; |
52d4b9ac KH |
3295 | /* This is for making all *used* pages to be on LRU. */ |
3296 | lru_add_drain_all(); | |
cdec2e42 | 3297 | drain_all_stock_sync(); |
f817ed48 | 3298 | ret = 0; |
32047e2a | 3299 | mem_cgroup_start_move(mem); |
299b4eaa | 3300 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3301 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 3302 | enum lru_list l; |
f817ed48 KH |
3303 | for_each_lru(l) { |
3304 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 3305 | node, zid, l); |
f817ed48 KH |
3306 | if (ret) |
3307 | break; | |
3308 | } | |
1ecaab2b | 3309 | } |
f817ed48 KH |
3310 | if (ret) |
3311 | break; | |
3312 | } | |
32047e2a | 3313 | mem_cgroup_end_move(mem); |
3c11ecf4 | 3314 | memcg_oom_recover(mem); |
f817ed48 KH |
3315 | /* it seems parent cgroup doesn't have enough mem */ |
3316 | if (ret == -ENOMEM) | |
3317 | goto try_to_free; | |
52d4b9ac | 3318 | cond_resched(); |
fce66477 DN |
3319 | /* "ret" should also be checked to ensure all lists are empty. */ |
3320 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
3321 | out: |
3322 | css_put(&mem->css); | |
3323 | return ret; | |
f817ed48 KH |
3324 | |
3325 | try_to_free: | |
c1e862c1 KH |
3326 | /* returns EBUSY if there is a task or if we come here twice. */ |
3327 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3328 | ret = -EBUSY; |
3329 | goto out; | |
3330 | } | |
c1e862c1 KH |
3331 | /* we call try-to-free pages for make this cgroup empty */ |
3332 | lru_add_drain_all(); | |
f817ed48 KH |
3333 | /* try to free all pages in this cgroup */ |
3334 | shrink = 1; | |
3335 | while (nr_retries && mem->res.usage > 0) { | |
3336 | int progress; | |
c1e862c1 KH |
3337 | |
3338 | if (signal_pending(current)) { | |
3339 | ret = -EINTR; | |
3340 | goto out; | |
3341 | } | |
a7885eb8 KM |
3342 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
3343 | false, get_swappiness(mem)); | |
c1e862c1 | 3344 | if (!progress) { |
f817ed48 | 3345 | nr_retries--; |
c1e862c1 | 3346 | /* maybe some writeback is necessary */ |
8aa7e847 | 3347 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3348 | } |
f817ed48 KH |
3349 | |
3350 | } | |
08e552c6 | 3351 | lru_add_drain(); |
f817ed48 | 3352 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3353 | goto move_account; |
cc847582 KH |
3354 | } |
3355 | ||
c1e862c1 KH |
3356 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3357 | { | |
3358 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3359 | } | |
3360 | ||
3361 | ||
18f59ea7 BS |
3362 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3363 | { | |
3364 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3365 | } | |
3366 | ||
3367 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3368 | u64 val) | |
3369 | { | |
3370 | int retval = 0; | |
3371 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3372 | struct cgroup *parent = cont->parent; | |
3373 | struct mem_cgroup *parent_mem = NULL; | |
3374 | ||
3375 | if (parent) | |
3376 | parent_mem = mem_cgroup_from_cont(parent); | |
3377 | ||
3378 | cgroup_lock(); | |
3379 | /* | |
af901ca1 | 3380 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3381 | * in the child subtrees. If it is unset, then the change can |
3382 | * occur, provided the current cgroup has no children. | |
3383 | * | |
3384 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3385 | * set if there are no children. | |
3386 | */ | |
3387 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
3388 | (val == 1 || val == 0)) { | |
3389 | if (list_empty(&cont->children)) | |
3390 | mem->use_hierarchy = val; | |
3391 | else | |
3392 | retval = -EBUSY; | |
3393 | } else | |
3394 | retval = -EINVAL; | |
3395 | cgroup_unlock(); | |
3396 | ||
3397 | return retval; | |
3398 | } | |
3399 | ||
0c3e73e8 | 3400 | |
7d74b06f KH |
3401 | static u64 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, |
3402 | enum mem_cgroup_stat_index idx) | |
0c3e73e8 | 3403 | { |
7d74b06f KH |
3404 | struct mem_cgroup *iter; |
3405 | s64 val = 0; | |
0c3e73e8 | 3406 | |
7d74b06f KH |
3407 | /* each per cpu's value can be minus.Then, use s64 */ |
3408 | for_each_mem_cgroup_tree(iter, mem) | |
3409 | val += mem_cgroup_read_stat(iter, idx); | |
3410 | ||
3411 | if (val < 0) /* race ? */ | |
3412 | val = 0; | |
3413 | return val; | |
0c3e73e8 BS |
3414 | } |
3415 | ||
104f3928 KS |
3416 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
3417 | { | |
7d74b06f | 3418 | u64 val; |
104f3928 KS |
3419 | |
3420 | if (!mem_cgroup_is_root(mem)) { | |
3421 | if (!swap) | |
3422 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
3423 | else | |
3424 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
3425 | } | |
3426 | ||
7d74b06f KH |
3427 | val = mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE); |
3428 | val += mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3429 | |
7d74b06f KH |
3430 | if (swap) |
3431 | val += mem_cgroup_get_recursive_idx_stat(mem, | |
3432 | MEM_CGROUP_STAT_SWAPOUT); | |
104f3928 KS |
3433 | |
3434 | return val << PAGE_SHIFT; | |
3435 | } | |
3436 | ||
2c3daa72 | 3437 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3438 | { |
8c7c6e34 | 3439 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 3440 | u64 val; |
8c7c6e34 KH |
3441 | int type, name; |
3442 | ||
3443 | type = MEMFILE_TYPE(cft->private); | |
3444 | name = MEMFILE_ATTR(cft->private); | |
3445 | switch (type) { | |
3446 | case _MEM: | |
104f3928 KS |
3447 | if (name == RES_USAGE) |
3448 | val = mem_cgroup_usage(mem, false); | |
3449 | else | |
0c3e73e8 | 3450 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
3451 | break; |
3452 | case _MEMSWAP: | |
104f3928 KS |
3453 | if (name == RES_USAGE) |
3454 | val = mem_cgroup_usage(mem, true); | |
3455 | else | |
0c3e73e8 | 3456 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
3457 | break; |
3458 | default: | |
3459 | BUG(); | |
3460 | break; | |
3461 | } | |
3462 | return val; | |
8cdea7c0 | 3463 | } |
628f4235 KH |
3464 | /* |
3465 | * The user of this function is... | |
3466 | * RES_LIMIT. | |
3467 | */ | |
856c13aa PM |
3468 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3469 | const char *buffer) | |
8cdea7c0 | 3470 | { |
628f4235 | 3471 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3472 | int type, name; |
628f4235 KH |
3473 | unsigned long long val; |
3474 | int ret; | |
3475 | ||
8c7c6e34 KH |
3476 | type = MEMFILE_TYPE(cft->private); |
3477 | name = MEMFILE_ATTR(cft->private); | |
3478 | switch (name) { | |
628f4235 | 3479 | case RES_LIMIT: |
4b3bde4c BS |
3480 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3481 | ret = -EINVAL; | |
3482 | break; | |
3483 | } | |
628f4235 KH |
3484 | /* This function does all necessary parse...reuse it */ |
3485 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3486 | if (ret) |
3487 | break; | |
3488 | if (type == _MEM) | |
628f4235 | 3489 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3490 | else |
3491 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3492 | break; |
296c81d8 BS |
3493 | case RES_SOFT_LIMIT: |
3494 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3495 | if (ret) | |
3496 | break; | |
3497 | /* | |
3498 | * For memsw, soft limits are hard to implement in terms | |
3499 | * of semantics, for now, we support soft limits for | |
3500 | * control without swap | |
3501 | */ | |
3502 | if (type == _MEM) | |
3503 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3504 | else | |
3505 | ret = -EINVAL; | |
3506 | break; | |
628f4235 KH |
3507 | default: |
3508 | ret = -EINVAL; /* should be BUG() ? */ | |
3509 | break; | |
3510 | } | |
3511 | return ret; | |
8cdea7c0 BS |
3512 | } |
3513 | ||
fee7b548 KH |
3514 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3515 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3516 | { | |
3517 | struct cgroup *cgroup; | |
3518 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3519 | ||
3520 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3521 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3522 | cgroup = memcg->css.cgroup; | |
3523 | if (!memcg->use_hierarchy) | |
3524 | goto out; | |
3525 | ||
3526 | while (cgroup->parent) { | |
3527 | cgroup = cgroup->parent; | |
3528 | memcg = mem_cgroup_from_cont(cgroup); | |
3529 | if (!memcg->use_hierarchy) | |
3530 | break; | |
3531 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3532 | min_limit = min(min_limit, tmp); | |
3533 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3534 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3535 | } | |
3536 | out: | |
3537 | *mem_limit = min_limit; | |
3538 | *memsw_limit = min_memsw_limit; | |
3539 | return; | |
3540 | } | |
3541 | ||
29f2a4da | 3542 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
3543 | { |
3544 | struct mem_cgroup *mem; | |
8c7c6e34 | 3545 | int type, name; |
c84872e1 PE |
3546 | |
3547 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
3548 | type = MEMFILE_TYPE(event); |
3549 | name = MEMFILE_ATTR(event); | |
3550 | switch (name) { | |
29f2a4da | 3551 | case RES_MAX_USAGE: |
8c7c6e34 KH |
3552 | if (type == _MEM) |
3553 | res_counter_reset_max(&mem->res); | |
3554 | else | |
3555 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
3556 | break; |
3557 | case RES_FAILCNT: | |
8c7c6e34 KH |
3558 | if (type == _MEM) |
3559 | res_counter_reset_failcnt(&mem->res); | |
3560 | else | |
3561 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
3562 | break; |
3563 | } | |
f64c3f54 | 3564 | |
85cc59db | 3565 | return 0; |
c84872e1 PE |
3566 | } |
3567 | ||
7dc74be0 DN |
3568 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3569 | struct cftype *cft) | |
3570 | { | |
3571 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3572 | } | |
3573 | ||
02491447 | 3574 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3575 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3576 | struct cftype *cft, u64 val) | |
3577 | { | |
3578 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3579 | ||
3580 | if (val >= (1 << NR_MOVE_TYPE)) | |
3581 | return -EINVAL; | |
3582 | /* | |
3583 | * We check this value several times in both in can_attach() and | |
3584 | * attach(), so we need cgroup lock to prevent this value from being | |
3585 | * inconsistent. | |
3586 | */ | |
3587 | cgroup_lock(); | |
3588 | mem->move_charge_at_immigrate = val; | |
3589 | cgroup_unlock(); | |
3590 | ||
3591 | return 0; | |
3592 | } | |
02491447 DN |
3593 | #else |
3594 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3595 | struct cftype *cft, u64 val) | |
3596 | { | |
3597 | return -ENOSYS; | |
3598 | } | |
3599 | #endif | |
7dc74be0 | 3600 | |
14067bb3 KH |
3601 | |
3602 | /* For read statistics */ | |
3603 | enum { | |
3604 | MCS_CACHE, | |
3605 | MCS_RSS, | |
d8046582 | 3606 | MCS_FILE_MAPPED, |
14067bb3 KH |
3607 | MCS_PGPGIN, |
3608 | MCS_PGPGOUT, | |
1dd3a273 | 3609 | MCS_SWAP, |
14067bb3 KH |
3610 | MCS_INACTIVE_ANON, |
3611 | MCS_ACTIVE_ANON, | |
3612 | MCS_INACTIVE_FILE, | |
3613 | MCS_ACTIVE_FILE, | |
3614 | MCS_UNEVICTABLE, | |
3615 | NR_MCS_STAT, | |
3616 | }; | |
3617 | ||
3618 | struct mcs_total_stat { | |
3619 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
3620 | }; |
3621 | ||
14067bb3 KH |
3622 | struct { |
3623 | char *local_name; | |
3624 | char *total_name; | |
3625 | } memcg_stat_strings[NR_MCS_STAT] = { | |
3626 | {"cache", "total_cache"}, | |
3627 | {"rss", "total_rss"}, | |
d69b042f | 3628 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
3629 | {"pgpgin", "total_pgpgin"}, |
3630 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3631 | {"swap", "total_swap"}, |
14067bb3 KH |
3632 | {"inactive_anon", "total_inactive_anon"}, |
3633 | {"active_anon", "total_active_anon"}, | |
3634 | {"inactive_file", "total_inactive_file"}, | |
3635 | {"active_file", "total_active_file"}, | |
3636 | {"unevictable", "total_unevictable"} | |
3637 | }; | |
3638 | ||
3639 | ||
7d74b06f KH |
3640 | static void |
3641 | mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
14067bb3 | 3642 | { |
14067bb3 KH |
3643 | s64 val; |
3644 | ||
3645 | /* per cpu stat */ | |
c62b1a3b | 3646 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 3647 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 3648 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 3649 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 3650 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 3651 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c62b1a3b | 3652 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); |
14067bb3 | 3653 | s->stat[MCS_PGPGIN] += val; |
c62b1a3b | 3654 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
14067bb3 | 3655 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 3656 | if (do_swap_account) { |
c62b1a3b | 3657 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
3658 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
3659 | } | |
14067bb3 KH |
3660 | |
3661 | /* per zone stat */ | |
3662 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
3663 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
3664 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
3665 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
3666 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
3667 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
3668 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
3669 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
3670 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
3671 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
14067bb3 KH |
3672 | } |
3673 | ||
3674 | static void | |
3675 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
3676 | { | |
7d74b06f KH |
3677 | struct mem_cgroup *iter; |
3678 | ||
3679 | for_each_mem_cgroup_tree(iter, mem) | |
3680 | mem_cgroup_get_local_stat(iter, s); | |
14067bb3 KH |
3681 | } |
3682 | ||
c64745cf PM |
3683 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
3684 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 3685 | { |
d2ceb9b7 | 3686 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 3687 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
3688 | int i; |
3689 | ||
14067bb3 KH |
3690 | memset(&mystat, 0, sizeof(mystat)); |
3691 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 3692 | |
1dd3a273 DN |
3693 | for (i = 0; i < NR_MCS_STAT; i++) { |
3694 | if (i == MCS_SWAP && !do_swap_account) | |
3695 | continue; | |
14067bb3 | 3696 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 3697 | } |
7b854121 | 3698 | |
14067bb3 | 3699 | /* Hierarchical information */ |
fee7b548 KH |
3700 | { |
3701 | unsigned long long limit, memsw_limit; | |
3702 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
3703 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
3704 | if (do_swap_account) | |
3705 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
3706 | } | |
7f016ee8 | 3707 | |
14067bb3 KH |
3708 | memset(&mystat, 0, sizeof(mystat)); |
3709 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
3710 | for (i = 0; i < NR_MCS_STAT; i++) { |
3711 | if (i == MCS_SWAP && !do_swap_account) | |
3712 | continue; | |
14067bb3 | 3713 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 3714 | } |
14067bb3 | 3715 | |
7f016ee8 | 3716 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 3717 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
3718 | |
3719 | { | |
3720 | int nid, zid; | |
3721 | struct mem_cgroup_per_zone *mz; | |
3722 | unsigned long recent_rotated[2] = {0, 0}; | |
3723 | unsigned long recent_scanned[2] = {0, 0}; | |
3724 | ||
3725 | for_each_online_node(nid) | |
3726 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3727 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3728 | ||
3729 | recent_rotated[0] += | |
3730 | mz->reclaim_stat.recent_rotated[0]; | |
3731 | recent_rotated[1] += | |
3732 | mz->reclaim_stat.recent_rotated[1]; | |
3733 | recent_scanned[0] += | |
3734 | mz->reclaim_stat.recent_scanned[0]; | |
3735 | recent_scanned[1] += | |
3736 | mz->reclaim_stat.recent_scanned[1]; | |
3737 | } | |
3738 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3739 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3740 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3741 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3742 | } | |
3743 | #endif | |
3744 | ||
d2ceb9b7 KH |
3745 | return 0; |
3746 | } | |
3747 | ||
a7885eb8 KM |
3748 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
3749 | { | |
3750 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3751 | ||
3752 | return get_swappiness(memcg); | |
3753 | } | |
3754 | ||
3755 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3756 | u64 val) | |
3757 | { | |
3758 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3759 | struct mem_cgroup *parent; | |
068b38c1 | 3760 | |
a7885eb8 KM |
3761 | if (val > 100) |
3762 | return -EINVAL; | |
3763 | ||
3764 | if (cgrp->parent == NULL) | |
3765 | return -EINVAL; | |
3766 | ||
3767 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
3768 | |
3769 | cgroup_lock(); | |
3770 | ||
a7885eb8 KM |
3771 | /* If under hierarchy, only empty-root can set this value */ |
3772 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
3773 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3774 | cgroup_unlock(); | |
a7885eb8 | 3775 | return -EINVAL; |
068b38c1 | 3776 | } |
a7885eb8 KM |
3777 | |
3778 | spin_lock(&memcg->reclaim_param_lock); | |
3779 | memcg->swappiness = val; | |
3780 | spin_unlock(&memcg->reclaim_param_lock); | |
3781 | ||
068b38c1 LZ |
3782 | cgroup_unlock(); |
3783 | ||
a7885eb8 KM |
3784 | return 0; |
3785 | } | |
3786 | ||
2e72b634 KS |
3787 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3788 | { | |
3789 | struct mem_cgroup_threshold_ary *t; | |
3790 | u64 usage; | |
3791 | int i; | |
3792 | ||
3793 | rcu_read_lock(); | |
3794 | if (!swap) | |
2c488db2 | 3795 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3796 | else |
2c488db2 | 3797 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3798 | |
3799 | if (!t) | |
3800 | goto unlock; | |
3801 | ||
3802 | usage = mem_cgroup_usage(memcg, swap); | |
3803 | ||
3804 | /* | |
3805 | * current_threshold points to threshold just below usage. | |
3806 | * If it's not true, a threshold was crossed after last | |
3807 | * call of __mem_cgroup_threshold(). | |
3808 | */ | |
5407a562 | 3809 | i = t->current_threshold; |
2e72b634 KS |
3810 | |
3811 | /* | |
3812 | * Iterate backward over array of thresholds starting from | |
3813 | * current_threshold and check if a threshold is crossed. | |
3814 | * If none of thresholds below usage is crossed, we read | |
3815 | * only one element of the array here. | |
3816 | */ | |
3817 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3818 | eventfd_signal(t->entries[i].eventfd, 1); | |
3819 | ||
3820 | /* i = current_threshold + 1 */ | |
3821 | i++; | |
3822 | ||
3823 | /* | |
3824 | * Iterate forward over array of thresholds starting from | |
3825 | * current_threshold+1 and check if a threshold is crossed. | |
3826 | * If none of thresholds above usage is crossed, we read | |
3827 | * only one element of the array here. | |
3828 | */ | |
3829 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3830 | eventfd_signal(t->entries[i].eventfd, 1); | |
3831 | ||
3832 | /* Update current_threshold */ | |
5407a562 | 3833 | t->current_threshold = i - 1; |
2e72b634 KS |
3834 | unlock: |
3835 | rcu_read_unlock(); | |
3836 | } | |
3837 | ||
3838 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3839 | { | |
ad4ca5f4 KS |
3840 | while (memcg) { |
3841 | __mem_cgroup_threshold(memcg, false); | |
3842 | if (do_swap_account) | |
3843 | __mem_cgroup_threshold(memcg, true); | |
3844 | ||
3845 | memcg = parent_mem_cgroup(memcg); | |
3846 | } | |
2e72b634 KS |
3847 | } |
3848 | ||
3849 | static int compare_thresholds(const void *a, const void *b) | |
3850 | { | |
3851 | const struct mem_cgroup_threshold *_a = a; | |
3852 | const struct mem_cgroup_threshold *_b = b; | |
3853 | ||
3854 | return _a->threshold - _b->threshold; | |
3855 | } | |
3856 | ||
7d74b06f | 3857 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem) |
9490ff27 KH |
3858 | { |
3859 | struct mem_cgroup_eventfd_list *ev; | |
3860 | ||
3861 | list_for_each_entry(ev, &mem->oom_notify, list) | |
3862 | eventfd_signal(ev->eventfd, 1); | |
3863 | return 0; | |
3864 | } | |
3865 | ||
3866 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) | |
3867 | { | |
7d74b06f KH |
3868 | struct mem_cgroup *iter; |
3869 | ||
3870 | for_each_mem_cgroup_tree(iter, mem) | |
3871 | mem_cgroup_oom_notify_cb(iter); | |
9490ff27 KH |
3872 | } |
3873 | ||
3874 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
3875 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
3876 | { |
3877 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
3878 | struct mem_cgroup_thresholds *thresholds; |
3879 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
3880 | int type = MEMFILE_TYPE(cft->private); |
3881 | u64 threshold, usage; | |
2c488db2 | 3882 | int i, size, ret; |
2e72b634 KS |
3883 | |
3884 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
3885 | if (ret) | |
3886 | return ret; | |
3887 | ||
3888 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3889 | |
2e72b634 | 3890 | if (type == _MEM) |
2c488db2 | 3891 | thresholds = &memcg->thresholds; |
2e72b634 | 3892 | else if (type == _MEMSWAP) |
2c488db2 | 3893 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
3894 | else |
3895 | BUG(); | |
3896 | ||
3897 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3898 | ||
3899 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 3900 | if (thresholds->primary) |
2e72b634 KS |
3901 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3902 | ||
2c488db2 | 3903 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3904 | |
3905 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3906 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3907 | GFP_KERNEL); |
2c488db2 | 3908 | if (!new) { |
2e72b634 KS |
3909 | ret = -ENOMEM; |
3910 | goto unlock; | |
3911 | } | |
2c488db2 | 3912 | new->size = size; |
2e72b634 KS |
3913 | |
3914 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3915 | if (thresholds->primary) { |
3916 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3917 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3918 | } |
3919 | ||
2e72b634 | 3920 | /* Add new threshold */ |
2c488db2 KS |
3921 | new->entries[size - 1].eventfd = eventfd; |
3922 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
3923 | |
3924 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 3925 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
3926 | compare_thresholds, NULL); |
3927 | ||
3928 | /* Find current threshold */ | |
2c488db2 | 3929 | new->current_threshold = -1; |
2e72b634 | 3930 | for (i = 0; i < size; i++) { |
2c488db2 | 3931 | if (new->entries[i].threshold < usage) { |
2e72b634 | 3932 | /* |
2c488db2 KS |
3933 | * new->current_threshold will not be used until |
3934 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
3935 | * it here. |
3936 | */ | |
2c488db2 | 3937 | ++new->current_threshold; |
2e72b634 KS |
3938 | } |
3939 | } | |
3940 | ||
2c488db2 KS |
3941 | /* Free old spare buffer and save old primary buffer as spare */ |
3942 | kfree(thresholds->spare); | |
3943 | thresholds->spare = thresholds->primary; | |
3944 | ||
3945 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 3946 | |
907860ed | 3947 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
3948 | synchronize_rcu(); |
3949 | ||
2e72b634 KS |
3950 | unlock: |
3951 | mutex_unlock(&memcg->thresholds_lock); | |
3952 | ||
3953 | return ret; | |
3954 | } | |
3955 | ||
907860ed | 3956 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 3957 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
3958 | { |
3959 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
3960 | struct mem_cgroup_thresholds *thresholds; |
3961 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
3962 | int type = MEMFILE_TYPE(cft->private); |
3963 | u64 usage; | |
2c488db2 | 3964 | int i, j, size; |
2e72b634 KS |
3965 | |
3966 | mutex_lock(&memcg->thresholds_lock); | |
3967 | if (type == _MEM) | |
2c488db2 | 3968 | thresholds = &memcg->thresholds; |
2e72b634 | 3969 | else if (type == _MEMSWAP) |
2c488db2 | 3970 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
3971 | else |
3972 | BUG(); | |
3973 | ||
3974 | /* | |
3975 | * Something went wrong if we trying to unregister a threshold | |
3976 | * if we don't have thresholds | |
3977 | */ | |
3978 | BUG_ON(!thresholds); | |
3979 | ||
3980 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3981 | ||
3982 | /* Check if a threshold crossed before removing */ | |
3983 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3984 | ||
3985 | /* Calculate new number of threshold */ | |
2c488db2 KS |
3986 | size = 0; |
3987 | for (i = 0; i < thresholds->primary->size; i++) { | |
3988 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
3989 | size++; |
3990 | } | |
3991 | ||
2c488db2 | 3992 | new = thresholds->spare; |
907860ed | 3993 | |
2e72b634 KS |
3994 | /* Set thresholds array to NULL if we don't have thresholds */ |
3995 | if (!size) { | |
2c488db2 KS |
3996 | kfree(new); |
3997 | new = NULL; | |
907860ed | 3998 | goto swap_buffers; |
2e72b634 KS |
3999 | } |
4000 | ||
2c488db2 | 4001 | new->size = size; |
2e72b634 KS |
4002 | |
4003 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4004 | new->current_threshold = -1; |
4005 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4006 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4007 | continue; |
4008 | ||
2c488db2 KS |
4009 | new->entries[j] = thresholds->primary->entries[i]; |
4010 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4011 | /* |
2c488db2 | 4012 | * new->current_threshold will not be used |
2e72b634 KS |
4013 | * until rcu_assign_pointer(), so it's safe to increment |
4014 | * it here. | |
4015 | */ | |
2c488db2 | 4016 | ++new->current_threshold; |
2e72b634 KS |
4017 | } |
4018 | j++; | |
4019 | } | |
4020 | ||
907860ed | 4021 | swap_buffers: |
2c488db2 KS |
4022 | /* Swap primary and spare array */ |
4023 | thresholds->spare = thresholds->primary; | |
4024 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4025 | |
907860ed | 4026 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4027 | synchronize_rcu(); |
4028 | ||
2e72b634 | 4029 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4030 | } |
c1e862c1 | 4031 | |
9490ff27 KH |
4032 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4033 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4034 | { | |
4035 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4036 | struct mem_cgroup_eventfd_list *event; | |
4037 | int type = MEMFILE_TYPE(cft->private); | |
4038 | ||
4039 | BUG_ON(type != _OOM_TYPE); | |
4040 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4041 | if (!event) | |
4042 | return -ENOMEM; | |
4043 | ||
4044 | mutex_lock(&memcg_oom_mutex); | |
4045 | ||
4046 | event->eventfd = eventfd; | |
4047 | list_add(&event->list, &memcg->oom_notify); | |
4048 | ||
4049 | /* already in OOM ? */ | |
4050 | if (atomic_read(&memcg->oom_lock)) | |
4051 | eventfd_signal(eventfd, 1); | |
4052 | mutex_unlock(&memcg_oom_mutex); | |
4053 | ||
4054 | return 0; | |
4055 | } | |
4056 | ||
907860ed | 4057 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4058 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4059 | { | |
4060 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4061 | struct mem_cgroup_eventfd_list *ev, *tmp; | |
4062 | int type = MEMFILE_TYPE(cft->private); | |
4063 | ||
4064 | BUG_ON(type != _OOM_TYPE); | |
4065 | ||
4066 | mutex_lock(&memcg_oom_mutex); | |
4067 | ||
4068 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { | |
4069 | if (ev->eventfd == eventfd) { | |
4070 | list_del(&ev->list); | |
4071 | kfree(ev); | |
4072 | } | |
4073 | } | |
4074 | ||
4075 | mutex_unlock(&memcg_oom_mutex); | |
9490ff27 KH |
4076 | } |
4077 | ||
3c11ecf4 KH |
4078 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4079 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4080 | { | |
4081 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4082 | ||
4083 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); | |
4084 | ||
4085 | if (atomic_read(&mem->oom_lock)) | |
4086 | cb->fill(cb, "under_oom", 1); | |
4087 | else | |
4088 | cb->fill(cb, "under_oom", 0); | |
4089 | return 0; | |
4090 | } | |
4091 | ||
3c11ecf4 KH |
4092 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4093 | struct cftype *cft, u64 val) | |
4094 | { | |
4095 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4096 | struct mem_cgroup *parent; | |
4097 | ||
4098 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4099 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4100 | return -EINVAL; | |
4101 | ||
4102 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4103 | ||
4104 | cgroup_lock(); | |
4105 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4106 | if ((parent->use_hierarchy) || | |
4107 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { | |
4108 | cgroup_unlock(); | |
4109 | return -EINVAL; | |
4110 | } | |
4111 | mem->oom_kill_disable = val; | |
4d845ebf KH |
4112 | if (!val) |
4113 | memcg_oom_recover(mem); | |
3c11ecf4 KH |
4114 | cgroup_unlock(); |
4115 | return 0; | |
4116 | } | |
4117 | ||
8cdea7c0 BS |
4118 | static struct cftype mem_cgroup_files[] = { |
4119 | { | |
0eea1030 | 4120 | .name = "usage_in_bytes", |
8c7c6e34 | 4121 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4122 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4123 | .register_event = mem_cgroup_usage_register_event, |
4124 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4125 | }, |
c84872e1 PE |
4126 | { |
4127 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4128 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4129 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4130 | .read_u64 = mem_cgroup_read, |
4131 | }, | |
8cdea7c0 | 4132 | { |
0eea1030 | 4133 | .name = "limit_in_bytes", |
8c7c6e34 | 4134 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4135 | .write_string = mem_cgroup_write, |
2c3daa72 | 4136 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4137 | }, |
296c81d8 BS |
4138 | { |
4139 | .name = "soft_limit_in_bytes", | |
4140 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4141 | .write_string = mem_cgroup_write, | |
4142 | .read_u64 = mem_cgroup_read, | |
4143 | }, | |
8cdea7c0 BS |
4144 | { |
4145 | .name = "failcnt", | |
8c7c6e34 | 4146 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4147 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4148 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4149 | }, |
d2ceb9b7 KH |
4150 | { |
4151 | .name = "stat", | |
c64745cf | 4152 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4153 | }, |
c1e862c1 KH |
4154 | { |
4155 | .name = "force_empty", | |
4156 | .trigger = mem_cgroup_force_empty_write, | |
4157 | }, | |
18f59ea7 BS |
4158 | { |
4159 | .name = "use_hierarchy", | |
4160 | .write_u64 = mem_cgroup_hierarchy_write, | |
4161 | .read_u64 = mem_cgroup_hierarchy_read, | |
4162 | }, | |
a7885eb8 KM |
4163 | { |
4164 | .name = "swappiness", | |
4165 | .read_u64 = mem_cgroup_swappiness_read, | |
4166 | .write_u64 = mem_cgroup_swappiness_write, | |
4167 | }, | |
7dc74be0 DN |
4168 | { |
4169 | .name = "move_charge_at_immigrate", | |
4170 | .read_u64 = mem_cgroup_move_charge_read, | |
4171 | .write_u64 = mem_cgroup_move_charge_write, | |
4172 | }, | |
9490ff27 KH |
4173 | { |
4174 | .name = "oom_control", | |
3c11ecf4 KH |
4175 | .read_map = mem_cgroup_oom_control_read, |
4176 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4177 | .register_event = mem_cgroup_oom_register_event, |
4178 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4179 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4180 | }, | |
8cdea7c0 BS |
4181 | }; |
4182 | ||
8c7c6e34 KH |
4183 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4184 | static struct cftype memsw_cgroup_files[] = { | |
4185 | { | |
4186 | .name = "memsw.usage_in_bytes", | |
4187 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4188 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4189 | .register_event = mem_cgroup_usage_register_event, |
4190 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4191 | }, |
4192 | { | |
4193 | .name = "memsw.max_usage_in_bytes", | |
4194 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4195 | .trigger = mem_cgroup_reset, | |
4196 | .read_u64 = mem_cgroup_read, | |
4197 | }, | |
4198 | { | |
4199 | .name = "memsw.limit_in_bytes", | |
4200 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4201 | .write_string = mem_cgroup_write, | |
4202 | .read_u64 = mem_cgroup_read, | |
4203 | }, | |
4204 | { | |
4205 | .name = "memsw.failcnt", | |
4206 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4207 | .trigger = mem_cgroup_reset, | |
4208 | .read_u64 = mem_cgroup_read, | |
4209 | }, | |
4210 | }; | |
4211 | ||
4212 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4213 | { | |
4214 | if (!do_swap_account) | |
4215 | return 0; | |
4216 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4217 | ARRAY_SIZE(memsw_cgroup_files)); | |
4218 | }; | |
4219 | #else | |
4220 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4221 | { | |
4222 | return 0; | |
4223 | } | |
4224 | #endif | |
4225 | ||
6d12e2d8 KH |
4226 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4227 | { | |
4228 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4229 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 4230 | enum lru_list l; |
41e3355d | 4231 | int zone, tmp = node; |
1ecaab2b KH |
4232 | /* |
4233 | * This routine is called against possible nodes. | |
4234 | * But it's BUG to call kmalloc() against offline node. | |
4235 | * | |
4236 | * TODO: this routine can waste much memory for nodes which will | |
4237 | * never be onlined. It's better to use memory hotplug callback | |
4238 | * function. | |
4239 | */ | |
41e3355d KH |
4240 | if (!node_state(node, N_NORMAL_MEMORY)) |
4241 | tmp = -1; | |
17295c88 | 4242 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4243 | if (!pn) |
4244 | return 1; | |
1ecaab2b | 4245 | |
6d12e2d8 | 4246 | mem->info.nodeinfo[node] = pn; |
1ecaab2b KH |
4247 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4248 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
4249 | for_each_lru(l) |
4250 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 4251 | mz->usage_in_excess = 0; |
4e416953 BS |
4252 | mz->on_tree = false; |
4253 | mz->mem = mem; | |
1ecaab2b | 4254 | } |
6d12e2d8 KH |
4255 | return 0; |
4256 | } | |
4257 | ||
1ecaab2b KH |
4258 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4259 | { | |
4260 | kfree(mem->info.nodeinfo[node]); | |
4261 | } | |
4262 | ||
33327948 KH |
4263 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4264 | { | |
4265 | struct mem_cgroup *mem; | |
c62b1a3b | 4266 | int size = sizeof(struct mem_cgroup); |
33327948 | 4267 | |
c62b1a3b | 4268 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4269 | if (size < PAGE_SIZE) |
17295c88 | 4270 | mem = kzalloc(size, GFP_KERNEL); |
33327948 | 4271 | else |
17295c88 | 4272 | mem = vzalloc(size); |
33327948 | 4273 | |
e7bbcdf3 DC |
4274 | if (!mem) |
4275 | return NULL; | |
4276 | ||
c62b1a3b | 4277 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
d2e61b8d DC |
4278 | if (!mem->stat) |
4279 | goto out_free; | |
711d3d2c | 4280 | spin_lock_init(&mem->pcp_counter_lock); |
33327948 | 4281 | return mem; |
d2e61b8d DC |
4282 | |
4283 | out_free: | |
4284 | if (size < PAGE_SIZE) | |
4285 | kfree(mem); | |
4286 | else | |
4287 | vfree(mem); | |
4288 | return NULL; | |
33327948 KH |
4289 | } |
4290 | ||
8c7c6e34 KH |
4291 | /* |
4292 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4293 | * (scanning all at force_empty is too costly...) | |
4294 | * | |
4295 | * Instead of clearing all references at force_empty, we remember | |
4296 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4297 | * it goes down to 0. | |
4298 | * | |
8c7c6e34 KH |
4299 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4300 | */ | |
4301 | ||
a7ba0eef | 4302 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 4303 | { |
08e552c6 KH |
4304 | int node; |
4305 | ||
f64c3f54 | 4306 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
4307 | free_css_id(&mem_cgroup_subsys, &mem->css); |
4308 | ||
08e552c6 KH |
4309 | for_each_node_state(node, N_POSSIBLE) |
4310 | free_mem_cgroup_per_zone_info(mem, node); | |
4311 | ||
c62b1a3b KH |
4312 | free_percpu(mem->stat); |
4313 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
4314 | kfree(mem); |
4315 | else | |
4316 | vfree(mem); | |
4317 | } | |
4318 | ||
8c7c6e34 KH |
4319 | static void mem_cgroup_get(struct mem_cgroup *mem) |
4320 | { | |
4321 | atomic_inc(&mem->refcnt); | |
4322 | } | |
4323 | ||
483c30b5 | 4324 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 4325 | { |
483c30b5 | 4326 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 4327 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 4328 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
4329 | if (parent) |
4330 | mem_cgroup_put(parent); | |
4331 | } | |
8c7c6e34 KH |
4332 | } |
4333 | ||
483c30b5 DN |
4334 | static void mem_cgroup_put(struct mem_cgroup *mem) |
4335 | { | |
4336 | __mem_cgroup_put(mem, 1); | |
4337 | } | |
4338 | ||
7bcc1bb1 DN |
4339 | /* |
4340 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4341 | */ | |
4342 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
4343 | { | |
4344 | if (!mem->res.parent) | |
4345 | return NULL; | |
4346 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
4347 | } | |
33327948 | 4348 | |
c077719b KH |
4349 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4350 | static void __init enable_swap_cgroup(void) | |
4351 | { | |
f8d66542 | 4352 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4353 | do_swap_account = 1; |
4354 | } | |
4355 | #else | |
4356 | static void __init enable_swap_cgroup(void) | |
4357 | { | |
4358 | } | |
4359 | #endif | |
4360 | ||
f64c3f54 BS |
4361 | static int mem_cgroup_soft_limit_tree_init(void) |
4362 | { | |
4363 | struct mem_cgroup_tree_per_node *rtpn; | |
4364 | struct mem_cgroup_tree_per_zone *rtpz; | |
4365 | int tmp, node, zone; | |
4366 | ||
4367 | for_each_node_state(node, N_POSSIBLE) { | |
4368 | tmp = node; | |
4369 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4370 | tmp = -1; | |
4371 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4372 | if (!rtpn) | |
4373 | return 1; | |
4374 | ||
4375 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4376 | ||
4377 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4378 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4379 | rtpz->rb_root = RB_ROOT; | |
4380 | spin_lock_init(&rtpz->lock); | |
4381 | } | |
4382 | } | |
4383 | return 0; | |
4384 | } | |
4385 | ||
0eb253e2 | 4386 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4387 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4388 | { | |
28dbc4b6 | 4389 | struct mem_cgroup *mem, *parent; |
04046e1a | 4390 | long error = -ENOMEM; |
6d12e2d8 | 4391 | int node; |
8cdea7c0 | 4392 | |
c8dad2bb JB |
4393 | mem = mem_cgroup_alloc(); |
4394 | if (!mem) | |
04046e1a | 4395 | return ERR_PTR(error); |
78fb7466 | 4396 | |
6d12e2d8 KH |
4397 | for_each_node_state(node, N_POSSIBLE) |
4398 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
4399 | goto free_out; | |
f64c3f54 | 4400 | |
c077719b | 4401 | /* root ? */ |
28dbc4b6 | 4402 | if (cont->parent == NULL) { |
cdec2e42 | 4403 | int cpu; |
c077719b | 4404 | enable_swap_cgroup(); |
28dbc4b6 | 4405 | parent = NULL; |
4b3bde4c | 4406 | root_mem_cgroup = mem; |
f64c3f54 BS |
4407 | if (mem_cgroup_soft_limit_tree_init()) |
4408 | goto free_out; | |
cdec2e42 KH |
4409 | for_each_possible_cpu(cpu) { |
4410 | struct memcg_stock_pcp *stock = | |
4411 | &per_cpu(memcg_stock, cpu); | |
4412 | INIT_WORK(&stock->work, drain_local_stock); | |
4413 | } | |
711d3d2c | 4414 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4415 | } else { |
28dbc4b6 | 4416 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 | 4417 | mem->use_hierarchy = parent->use_hierarchy; |
3c11ecf4 | 4418 | mem->oom_kill_disable = parent->oom_kill_disable; |
18f59ea7 | 4419 | } |
28dbc4b6 | 4420 | |
18f59ea7 BS |
4421 | if (parent && parent->use_hierarchy) { |
4422 | res_counter_init(&mem->res, &parent->res); | |
4423 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4424 | /* |
4425 | * We increment refcnt of the parent to ensure that we can | |
4426 | * safely access it on res_counter_charge/uncharge. | |
4427 | * This refcnt will be decremented when freeing this | |
4428 | * mem_cgroup(see mem_cgroup_put). | |
4429 | */ | |
4430 | mem_cgroup_get(parent); | |
18f59ea7 BS |
4431 | } else { |
4432 | res_counter_init(&mem->res, NULL); | |
4433 | res_counter_init(&mem->memsw, NULL); | |
4434 | } | |
04046e1a | 4435 | mem->last_scanned_child = 0; |
2733c06a | 4436 | spin_lock_init(&mem->reclaim_param_lock); |
9490ff27 | 4437 | INIT_LIST_HEAD(&mem->oom_notify); |
6d61ef40 | 4438 | |
a7885eb8 KM |
4439 | if (parent) |
4440 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 4441 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 4442 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 4443 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 4444 | return &mem->css; |
6d12e2d8 | 4445 | free_out: |
a7ba0eef | 4446 | __mem_cgroup_free(mem); |
4b3bde4c | 4447 | root_mem_cgroup = NULL; |
04046e1a | 4448 | return ERR_PTR(error); |
8cdea7c0 BS |
4449 | } |
4450 | ||
ec64f515 | 4451 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4452 | struct cgroup *cont) |
4453 | { | |
4454 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
4455 | |
4456 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
4457 | } |
4458 | ||
8cdea7c0 BS |
4459 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4460 | struct cgroup *cont) | |
4461 | { | |
c268e994 | 4462 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 4463 | |
c268e994 | 4464 | mem_cgroup_put(mem); |
8cdea7c0 BS |
4465 | } |
4466 | ||
4467 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4468 | struct cgroup *cont) | |
4469 | { | |
8c7c6e34 KH |
4470 | int ret; |
4471 | ||
4472 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4473 | ARRAY_SIZE(mem_cgroup_files)); | |
4474 | ||
4475 | if (!ret) | |
4476 | ret = register_memsw_files(cont, ss); | |
4477 | return ret; | |
8cdea7c0 BS |
4478 | } |
4479 | ||
02491447 | 4480 | #ifdef CONFIG_MMU |
7dc74be0 | 4481 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4482 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4483 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4484 | { |
854ffa8d DN |
4485 | int ret = 0; |
4486 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
4487 | struct mem_cgroup *mem = mc.to; |
4488 | ||
854ffa8d DN |
4489 | if (mem_cgroup_is_root(mem)) { |
4490 | mc.precharge += count; | |
4491 | /* we don't need css_get for root */ | |
4492 | return ret; | |
4493 | } | |
4494 | /* try to charge at once */ | |
4495 | if (count > 1) { | |
4496 | struct res_counter *dummy; | |
4497 | /* | |
4498 | * "mem" cannot be under rmdir() because we've already checked | |
4499 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
4500 | * are still under the same cgroup_mutex. So we can postpone | |
4501 | * css_get(). | |
4502 | */ | |
4503 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
4504 | goto one_by_one; | |
4505 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
4506 | PAGE_SIZE * count, &dummy)) { | |
4507 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
4508 | goto one_by_one; | |
4509 | } | |
4510 | mc.precharge += count; | |
854ffa8d DN |
4511 | return ret; |
4512 | } | |
4513 | one_by_one: | |
4514 | /* fall back to one by one charge */ | |
4515 | while (count--) { | |
4516 | if (signal_pending(current)) { | |
4517 | ret = -EINTR; | |
4518 | break; | |
4519 | } | |
4520 | if (!batch_count--) { | |
4521 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4522 | cond_resched(); | |
4523 | } | |
ec168510 AA |
4524 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, |
4525 | PAGE_SIZE); | |
854ffa8d DN |
4526 | if (ret || !mem) |
4527 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
4528 | return -ENOMEM; | |
4529 | mc.precharge++; | |
4530 | } | |
4ffef5fe DN |
4531 | return ret; |
4532 | } | |
4533 | ||
4534 | /** | |
4535 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
4536 | * @vma: the vma the pte to be checked belongs | |
4537 | * @addr: the address corresponding to the pte to be checked | |
4538 | * @ptent: the pte to be checked | |
02491447 | 4539 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4540 | * |
4541 | * Returns | |
4542 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4543 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4544 | * move charge. if @target is not NULL, the page is stored in target->page | |
4545 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4546 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4547 | * target for charge migration. if @target is not NULL, the entry is stored | |
4548 | * in target->ent. | |
4ffef5fe DN |
4549 | * |
4550 | * Called with pte lock held. | |
4551 | */ | |
4ffef5fe DN |
4552 | union mc_target { |
4553 | struct page *page; | |
02491447 | 4554 | swp_entry_t ent; |
4ffef5fe DN |
4555 | }; |
4556 | ||
4ffef5fe DN |
4557 | enum mc_target_type { |
4558 | MC_TARGET_NONE, /* not used */ | |
4559 | MC_TARGET_PAGE, | |
02491447 | 4560 | MC_TARGET_SWAP, |
4ffef5fe DN |
4561 | }; |
4562 | ||
90254a65 DN |
4563 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4564 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4565 | { |
90254a65 | 4566 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4567 | |
90254a65 DN |
4568 | if (!page || !page_mapped(page)) |
4569 | return NULL; | |
4570 | if (PageAnon(page)) { | |
4571 | /* we don't move shared anon */ | |
4572 | if (!move_anon() || page_mapcount(page) > 2) | |
4573 | return NULL; | |
87946a72 DN |
4574 | } else if (!move_file()) |
4575 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
4576 | return NULL; |
4577 | if (!get_page_unless_zero(page)) | |
4578 | return NULL; | |
4579 | ||
4580 | return page; | |
4581 | } | |
4582 | ||
4583 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4584 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4585 | { | |
4586 | int usage_count; | |
4587 | struct page *page = NULL; | |
4588 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4589 | ||
4590 | if (!move_anon() || non_swap_entry(ent)) | |
4591 | return NULL; | |
4592 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
4593 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
4594 | if (page) |
4595 | put_page(page); | |
90254a65 | 4596 | return NULL; |
02491447 | 4597 | } |
90254a65 DN |
4598 | if (do_swap_account) |
4599 | entry->val = ent.val; | |
4600 | ||
4601 | return page; | |
4602 | } | |
4603 | ||
87946a72 DN |
4604 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4605 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4606 | { | |
4607 | struct page *page = NULL; | |
4608 | struct inode *inode; | |
4609 | struct address_space *mapping; | |
4610 | pgoff_t pgoff; | |
4611 | ||
4612 | if (!vma->vm_file) /* anonymous vma */ | |
4613 | return NULL; | |
4614 | if (!move_file()) | |
4615 | return NULL; | |
4616 | ||
4617 | inode = vma->vm_file->f_path.dentry->d_inode; | |
4618 | mapping = vma->vm_file->f_mapping; | |
4619 | if (pte_none(ptent)) | |
4620 | pgoff = linear_page_index(vma, addr); | |
4621 | else /* pte_file(ptent) is true */ | |
4622 | pgoff = pte_to_pgoff(ptent); | |
4623 | ||
4624 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
4625 | if (!mapping_cap_swap_backed(mapping)) { /* normal file */ | |
4626 | page = find_get_page(mapping, pgoff); | |
4627 | } else { /* shmem/tmpfs file. we should take account of swap too. */ | |
4628 | swp_entry_t ent; | |
4629 | mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent); | |
4630 | if (do_swap_account) | |
4631 | entry->val = ent.val; | |
4632 | } | |
4633 | ||
4634 | return page; | |
4635 | } | |
4636 | ||
90254a65 DN |
4637 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
4638 | unsigned long addr, pte_t ptent, union mc_target *target) | |
4639 | { | |
4640 | struct page *page = NULL; | |
4641 | struct page_cgroup *pc; | |
4642 | int ret = 0; | |
4643 | swp_entry_t ent = { .val = 0 }; | |
4644 | ||
4645 | if (pte_present(ptent)) | |
4646 | page = mc_handle_present_pte(vma, addr, ptent); | |
4647 | else if (is_swap_pte(ptent)) | |
4648 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
4649 | else if (pte_none(ptent) || pte_file(ptent)) |
4650 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
4651 | |
4652 | if (!page && !ent.val) | |
4653 | return 0; | |
02491447 DN |
4654 | if (page) { |
4655 | pc = lookup_page_cgroup(page); | |
4656 | /* | |
4657 | * Do only loose check w/o page_cgroup lock. | |
4658 | * mem_cgroup_move_account() checks the pc is valid or not under | |
4659 | * the lock. | |
4660 | */ | |
4661 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
4662 | ret = MC_TARGET_PAGE; | |
4663 | if (target) | |
4664 | target->page = page; | |
4665 | } | |
4666 | if (!ret || !target) | |
4667 | put_page(page); | |
4668 | } | |
90254a65 DN |
4669 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4670 | if (ent.val && !ret && | |
7f0f1546 KH |
4671 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
4672 | ret = MC_TARGET_SWAP; | |
4673 | if (target) | |
4674 | target->ent = ent; | |
4ffef5fe | 4675 | } |
4ffef5fe DN |
4676 | return ret; |
4677 | } | |
4678 | ||
4679 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
4680 | unsigned long addr, unsigned long end, | |
4681 | struct mm_walk *walk) | |
4682 | { | |
4683 | struct vm_area_struct *vma = walk->private; | |
4684 | pte_t *pte; | |
4685 | spinlock_t *ptl; | |
4686 | ||
ec168510 | 4687 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
4ffef5fe DN |
4688 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4689 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
4690 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
4691 | mc.precharge++; /* increment precharge temporarily */ | |
4692 | pte_unmap_unlock(pte - 1, ptl); | |
4693 | cond_resched(); | |
4694 | ||
7dc74be0 DN |
4695 | return 0; |
4696 | } | |
4697 | ||
4ffef5fe DN |
4698 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4699 | { | |
4700 | unsigned long precharge; | |
4701 | struct vm_area_struct *vma; | |
4702 | ||
dfe076b0 | 4703 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
4704 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
4705 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
4706 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4707 | .mm = mm, | |
4708 | .private = vma, | |
4709 | }; | |
4710 | if (is_vm_hugetlb_page(vma)) | |
4711 | continue; | |
4ffef5fe DN |
4712 | walk_page_range(vma->vm_start, vma->vm_end, |
4713 | &mem_cgroup_count_precharge_walk); | |
4714 | } | |
dfe076b0 | 4715 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4716 | |
4717 | precharge = mc.precharge; | |
4718 | mc.precharge = 0; | |
4719 | ||
4720 | return precharge; | |
4721 | } | |
4722 | ||
4ffef5fe DN |
4723 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4724 | { | |
dfe076b0 DN |
4725 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4726 | ||
4727 | VM_BUG_ON(mc.moving_task); | |
4728 | mc.moving_task = current; | |
4729 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4730 | } |
4731 | ||
dfe076b0 DN |
4732 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4733 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4734 | { |
2bd9bb20 KH |
4735 | struct mem_cgroup *from = mc.from; |
4736 | struct mem_cgroup *to = mc.to; | |
4737 | ||
4ffef5fe | 4738 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
4739 | if (mc.precharge) { |
4740 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
4741 | mc.precharge = 0; | |
4742 | } | |
4743 | /* | |
4744 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4745 | * we must uncharge here. | |
4746 | */ | |
4747 | if (mc.moved_charge) { | |
4748 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
4749 | mc.moved_charge = 0; | |
4ffef5fe | 4750 | } |
483c30b5 DN |
4751 | /* we must fixup refcnts and charges */ |
4752 | if (mc.moved_swap) { | |
483c30b5 DN |
4753 | /* uncharge swap account from the old cgroup */ |
4754 | if (!mem_cgroup_is_root(mc.from)) | |
4755 | res_counter_uncharge(&mc.from->memsw, | |
4756 | PAGE_SIZE * mc.moved_swap); | |
4757 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
4758 | ||
4759 | if (!mem_cgroup_is_root(mc.to)) { | |
4760 | /* | |
4761 | * we charged both to->res and to->memsw, so we should | |
4762 | * uncharge to->res. | |
4763 | */ | |
4764 | res_counter_uncharge(&mc.to->res, | |
4765 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
4766 | } |
4767 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
4768 | mc.moved_swap = 0; |
4769 | } | |
dfe076b0 DN |
4770 | memcg_oom_recover(from); |
4771 | memcg_oom_recover(to); | |
4772 | wake_up_all(&mc.waitq); | |
4773 | } | |
4774 | ||
4775 | static void mem_cgroup_clear_mc(void) | |
4776 | { | |
4777 | struct mem_cgroup *from = mc.from; | |
4778 | ||
4779 | /* | |
4780 | * we must clear moving_task before waking up waiters at the end of | |
4781 | * task migration. | |
4782 | */ | |
4783 | mc.moving_task = NULL; | |
4784 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4785 | spin_lock(&mc.lock); |
4ffef5fe DN |
4786 | mc.from = NULL; |
4787 | mc.to = NULL; | |
2bd9bb20 | 4788 | spin_unlock(&mc.lock); |
32047e2a | 4789 | mem_cgroup_end_move(from); |
4ffef5fe DN |
4790 | } |
4791 | ||
7dc74be0 DN |
4792 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
4793 | struct cgroup *cgroup, | |
4794 | struct task_struct *p, | |
4795 | bool threadgroup) | |
4796 | { | |
4797 | int ret = 0; | |
4798 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
4799 | ||
4800 | if (mem->move_charge_at_immigrate) { | |
4801 | struct mm_struct *mm; | |
4802 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
4803 | ||
4804 | VM_BUG_ON(from == mem); | |
4805 | ||
4806 | mm = get_task_mm(p); | |
4807 | if (!mm) | |
4808 | return 0; | |
7dc74be0 | 4809 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
4810 | if (mm->owner == p) { |
4811 | VM_BUG_ON(mc.from); | |
4812 | VM_BUG_ON(mc.to); | |
4813 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 4814 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 4815 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 4816 | mem_cgroup_start_move(from); |
2bd9bb20 | 4817 | spin_lock(&mc.lock); |
4ffef5fe DN |
4818 | mc.from = from; |
4819 | mc.to = mem; | |
2bd9bb20 | 4820 | spin_unlock(&mc.lock); |
dfe076b0 | 4821 | /* We set mc.moving_task later */ |
4ffef5fe DN |
4822 | |
4823 | ret = mem_cgroup_precharge_mc(mm); | |
4824 | if (ret) | |
4825 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
4826 | } |
4827 | mmput(mm); | |
7dc74be0 DN |
4828 | } |
4829 | return ret; | |
4830 | } | |
4831 | ||
4832 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4833 | struct cgroup *cgroup, | |
4834 | struct task_struct *p, | |
4835 | bool threadgroup) | |
4836 | { | |
4ffef5fe | 4837 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
4838 | } |
4839 | ||
4ffef5fe DN |
4840 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4841 | unsigned long addr, unsigned long end, | |
4842 | struct mm_walk *walk) | |
7dc74be0 | 4843 | { |
4ffef5fe DN |
4844 | int ret = 0; |
4845 | struct vm_area_struct *vma = walk->private; | |
4846 | pte_t *pte; | |
4847 | spinlock_t *ptl; | |
4848 | ||
4849 | retry: | |
ec168510 | 4850 | VM_BUG_ON(pmd_trans_huge(*pmd)); |
4ffef5fe DN |
4851 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4852 | for (; addr != end; addr += PAGE_SIZE) { | |
4853 | pte_t ptent = *(pte++); | |
4854 | union mc_target target; | |
4855 | int type; | |
4856 | struct page *page; | |
4857 | struct page_cgroup *pc; | |
02491447 | 4858 | swp_entry_t ent; |
4ffef5fe DN |
4859 | |
4860 | if (!mc.precharge) | |
4861 | break; | |
4862 | ||
4863 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
4864 | switch (type) { | |
4865 | case MC_TARGET_PAGE: | |
4866 | page = target.page; | |
4867 | if (isolate_lru_page(page)) | |
4868 | goto put; | |
4869 | pc = lookup_page_cgroup(page); | |
854ffa8d DN |
4870 | if (!mem_cgroup_move_account(pc, |
4871 | mc.from, mc.to, false)) { | |
4ffef5fe | 4872 | mc.precharge--; |
854ffa8d DN |
4873 | /* we uncharge from mc.from later. */ |
4874 | mc.moved_charge++; | |
4ffef5fe DN |
4875 | } |
4876 | putback_lru_page(page); | |
4877 | put: /* is_target_pte_for_mc() gets the page */ | |
4878 | put_page(page); | |
4879 | break; | |
02491447 DN |
4880 | case MC_TARGET_SWAP: |
4881 | ent = target.ent; | |
483c30b5 DN |
4882 | if (!mem_cgroup_move_swap_account(ent, |
4883 | mc.from, mc.to, false)) { | |
02491447 | 4884 | mc.precharge--; |
483c30b5 DN |
4885 | /* we fixup refcnts and charges later. */ |
4886 | mc.moved_swap++; | |
4887 | } | |
02491447 | 4888 | break; |
4ffef5fe DN |
4889 | default: |
4890 | break; | |
4891 | } | |
4892 | } | |
4893 | pte_unmap_unlock(pte - 1, ptl); | |
4894 | cond_resched(); | |
4895 | ||
4896 | if (addr != end) { | |
4897 | /* | |
4898 | * We have consumed all precharges we got in can_attach(). | |
4899 | * We try charge one by one, but don't do any additional | |
4900 | * charges to mc.to if we have failed in charge once in attach() | |
4901 | * phase. | |
4902 | */ | |
854ffa8d | 4903 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4904 | if (!ret) |
4905 | goto retry; | |
4906 | } | |
4907 | ||
4908 | return ret; | |
4909 | } | |
4910 | ||
4911 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4912 | { | |
4913 | struct vm_area_struct *vma; | |
4914 | ||
4915 | lru_add_drain_all(); | |
dfe076b0 DN |
4916 | retry: |
4917 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
4918 | /* | |
4919 | * Someone who are holding the mmap_sem might be waiting in | |
4920 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4921 | * and retry. Because we cancel precharges, we might not be able | |
4922 | * to move enough charges, but moving charge is a best-effort | |
4923 | * feature anyway, so it wouldn't be a big problem. | |
4924 | */ | |
4925 | __mem_cgroup_clear_mc(); | |
4926 | cond_resched(); | |
4927 | goto retry; | |
4928 | } | |
4ffef5fe DN |
4929 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
4930 | int ret; | |
4931 | struct mm_walk mem_cgroup_move_charge_walk = { | |
4932 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4933 | .mm = mm, | |
4934 | .private = vma, | |
4935 | }; | |
4936 | if (is_vm_hugetlb_page(vma)) | |
4937 | continue; | |
4ffef5fe DN |
4938 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
4939 | &mem_cgroup_move_charge_walk); | |
4940 | if (ret) | |
4941 | /* | |
4942 | * means we have consumed all precharges and failed in | |
4943 | * doing additional charge. Just abandon here. | |
4944 | */ | |
4945 | break; | |
4946 | } | |
dfe076b0 | 4947 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
4948 | } |
4949 | ||
67e465a7 BS |
4950 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
4951 | struct cgroup *cont, | |
4952 | struct cgroup *old_cont, | |
be367d09 BB |
4953 | struct task_struct *p, |
4954 | bool threadgroup) | |
67e465a7 | 4955 | { |
dfe076b0 DN |
4956 | struct mm_struct *mm; |
4957 | ||
4958 | if (!mc.to) | |
4ffef5fe DN |
4959 | /* no need to move charge */ |
4960 | return; | |
4961 | ||
dfe076b0 DN |
4962 | mm = get_task_mm(p); |
4963 | if (mm) { | |
4964 | mem_cgroup_move_charge(mm); | |
4965 | mmput(mm); | |
4966 | } | |
4ffef5fe | 4967 | mem_cgroup_clear_mc(); |
67e465a7 | 4968 | } |
5cfb80a7 DN |
4969 | #else /* !CONFIG_MMU */ |
4970 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
4971 | struct cgroup *cgroup, | |
4972 | struct task_struct *p, | |
4973 | bool threadgroup) | |
4974 | { | |
4975 | return 0; | |
4976 | } | |
4977 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4978 | struct cgroup *cgroup, | |
4979 | struct task_struct *p, | |
4980 | bool threadgroup) | |
4981 | { | |
4982 | } | |
4983 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
4984 | struct cgroup *cont, | |
4985 | struct cgroup *old_cont, | |
4986 | struct task_struct *p, | |
4987 | bool threadgroup) | |
4988 | { | |
4989 | } | |
4990 | #endif | |
67e465a7 | 4991 | |
8cdea7c0 BS |
4992 | struct cgroup_subsys mem_cgroup_subsys = { |
4993 | .name = "memory", | |
4994 | .subsys_id = mem_cgroup_subsys_id, | |
4995 | .create = mem_cgroup_create, | |
df878fb0 | 4996 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
4997 | .destroy = mem_cgroup_destroy, |
4998 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
4999 | .can_attach = mem_cgroup_can_attach, |
5000 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5001 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5002 | .early_init = 0, |
04046e1a | 5003 | .use_id = 1, |
8cdea7c0 | 5004 | }; |
c077719b KH |
5005 | |
5006 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5007 | static int __init enable_swap_account(char *s) |
5008 | { | |
5009 | /* consider enabled if no parameter or 1 is given */ | |
5010 | if (!s || !strcmp(s, "1")) | |
5011 | really_do_swap_account = 1; | |
5012 | else if (!strcmp(s, "0")) | |
5013 | really_do_swap_account = 0; | |
5014 | return 1; | |
5015 | } | |
5016 | __setup("swapaccount", enable_swap_account); | |
c077719b KH |
5017 | |
5018 | static int __init disable_swap_account(char *s) | |
5019 | { | |
a42c390c | 5020 | enable_swap_account("0"); |
c077719b KH |
5021 | return 1; |
5022 | } | |
5023 | __setup("noswapaccount", disable_swap_account); | |
5024 | #endif |