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; |
6d61ef40 | 221 | /* |
af901ca1 | 222 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 223 | * reclaimed from. |
6d61ef40 | 224 | */ |
04046e1a | 225 | int last_scanned_child; |
18f59ea7 BS |
226 | /* |
227 | * Should the accounting and control be hierarchical, per subtree? | |
228 | */ | |
229 | bool use_hierarchy; | |
867578cb | 230 | atomic_t oom_lock; |
8c7c6e34 | 231 | atomic_t refcnt; |
14797e23 | 232 | |
a7885eb8 | 233 | unsigned int swappiness; |
3c11ecf4 KH |
234 | /* OOM-Killer disable */ |
235 | int oom_kill_disable; | |
a7885eb8 | 236 | |
22a668d7 KH |
237 | /* set when res.limit == memsw.limit */ |
238 | bool memsw_is_minimum; | |
239 | ||
2e72b634 KS |
240 | /* protect arrays of thresholds */ |
241 | struct mutex thresholds_lock; | |
242 | ||
243 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 244 | struct mem_cgroup_thresholds thresholds; |
907860ed | 245 | |
2e72b634 | 246 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 247 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 248 | |
9490ff27 KH |
249 | /* For oom notifier event fd */ |
250 | struct list_head oom_notify; | |
251 | ||
7dc74be0 DN |
252 | /* |
253 | * Should we move charges of a task when a task is moved into this | |
254 | * mem_cgroup ? And what type of charges should we move ? | |
255 | */ | |
256 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 257 | /* |
c62b1a3b | 258 | * percpu counter. |
d52aa412 | 259 | */ |
c62b1a3b | 260 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
261 | /* |
262 | * used when a cpu is offlined or other synchronizations | |
263 | * See mem_cgroup_read_stat(). | |
264 | */ | |
265 | struct mem_cgroup_stat_cpu nocpu_base; | |
266 | spinlock_t pcp_counter_lock; | |
8cdea7c0 BS |
267 | }; |
268 | ||
7dc74be0 DN |
269 | /* Stuffs for move charges at task migration. */ |
270 | /* | |
271 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
272 | * left-shifted bitmap of these types. | |
273 | */ | |
274 | enum move_type { | |
4ffef5fe | 275 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 276 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
277 | NR_MOVE_TYPE, |
278 | }; | |
279 | ||
4ffef5fe DN |
280 | /* "mc" and its members are protected by cgroup_mutex */ |
281 | static struct move_charge_struct { | |
b1dd693e | 282 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
283 | struct mem_cgroup *from; |
284 | struct mem_cgroup *to; | |
285 | unsigned long precharge; | |
854ffa8d | 286 | unsigned long moved_charge; |
483c30b5 | 287 | unsigned long moved_swap; |
8033b97c DN |
288 | struct task_struct *moving_task; /* a task moving charges */ |
289 | wait_queue_head_t waitq; /* a waitq for other context */ | |
290 | } mc = { | |
2bd9bb20 | 291 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
292 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
293 | }; | |
4ffef5fe | 294 | |
90254a65 DN |
295 | static bool move_anon(void) |
296 | { | |
297 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
298 | &mc.to->move_charge_at_immigrate); | |
299 | } | |
300 | ||
87946a72 DN |
301 | static bool move_file(void) |
302 | { | |
303 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
304 | &mc.to->move_charge_at_immigrate); | |
305 | } | |
306 | ||
4e416953 BS |
307 | /* |
308 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
309 | * limit reclaim to prevent infinite loops, if they ever occur. | |
310 | */ | |
311 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
312 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
313 | ||
217bc319 KH |
314 | enum charge_type { |
315 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
316 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 317 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 318 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 319 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 320 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
321 | NR_CHARGE_TYPE, |
322 | }; | |
323 | ||
8c7c6e34 KH |
324 | /* for encoding cft->private value on file */ |
325 | #define _MEM (0) | |
326 | #define _MEMSWAP (1) | |
9490ff27 | 327 | #define _OOM_TYPE (2) |
8c7c6e34 KH |
328 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
329 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
330 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
331 | /* Used for OOM nofiier */ |
332 | #define OOM_CONTROL (0) | |
8c7c6e34 | 333 | |
75822b44 BS |
334 | /* |
335 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
336 | */ | |
337 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
338 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
339 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
340 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
341 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
342 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 343 | |
8c7c6e34 KH |
344 | static void mem_cgroup_get(struct mem_cgroup *mem); |
345 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 346 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 347 | static void drain_all_stock_async(void); |
8c7c6e34 | 348 | |
f64c3f54 BS |
349 | static struct mem_cgroup_per_zone * |
350 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
351 | { | |
352 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
353 | } | |
354 | ||
d324236b WF |
355 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
356 | { | |
357 | return &mem->css; | |
358 | } | |
359 | ||
f64c3f54 | 360 | static struct mem_cgroup_per_zone * |
97a6c37b | 361 | page_cgroup_zoneinfo(struct mem_cgroup *mem, struct page *page) |
f64c3f54 | 362 | { |
97a6c37b JW |
363 | int nid = page_to_nid(page); |
364 | int zid = page_zonenum(page); | |
f64c3f54 | 365 | |
f64c3f54 BS |
366 | return mem_cgroup_zoneinfo(mem, nid, zid); |
367 | } | |
368 | ||
369 | static struct mem_cgroup_tree_per_zone * | |
370 | soft_limit_tree_node_zone(int nid, int zid) | |
371 | { | |
372 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
373 | } | |
374 | ||
375 | static struct mem_cgroup_tree_per_zone * | |
376 | soft_limit_tree_from_page(struct page *page) | |
377 | { | |
378 | int nid = page_to_nid(page); | |
379 | int zid = page_zonenum(page); | |
380 | ||
381 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
382 | } | |
383 | ||
384 | static void | |
4e416953 | 385 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 386 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
387 | struct mem_cgroup_tree_per_zone *mctz, |
388 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
389 | { |
390 | struct rb_node **p = &mctz->rb_root.rb_node; | |
391 | struct rb_node *parent = NULL; | |
392 | struct mem_cgroup_per_zone *mz_node; | |
393 | ||
394 | if (mz->on_tree) | |
395 | return; | |
396 | ||
ef8745c1 KH |
397 | mz->usage_in_excess = new_usage_in_excess; |
398 | if (!mz->usage_in_excess) | |
399 | return; | |
f64c3f54 BS |
400 | while (*p) { |
401 | parent = *p; | |
402 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
403 | tree_node); | |
404 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
405 | p = &(*p)->rb_left; | |
406 | /* | |
407 | * We can't avoid mem cgroups that are over their soft | |
408 | * limit by the same amount | |
409 | */ | |
410 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
411 | p = &(*p)->rb_right; | |
412 | } | |
413 | rb_link_node(&mz->tree_node, parent, p); | |
414 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
415 | mz->on_tree = true; | |
4e416953 BS |
416 | } |
417 | ||
418 | static void | |
419 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
420 | struct mem_cgroup_per_zone *mz, | |
421 | struct mem_cgroup_tree_per_zone *mctz) | |
422 | { | |
423 | if (!mz->on_tree) | |
424 | return; | |
425 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
426 | mz->on_tree = false; | |
427 | } | |
428 | ||
f64c3f54 BS |
429 | static void |
430 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
431 | struct mem_cgroup_per_zone *mz, | |
432 | struct mem_cgroup_tree_per_zone *mctz) | |
433 | { | |
434 | spin_lock(&mctz->lock); | |
4e416953 | 435 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
436 | spin_unlock(&mctz->lock); |
437 | } | |
438 | ||
f64c3f54 BS |
439 | |
440 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
441 | { | |
ef8745c1 | 442 | unsigned long long excess; |
f64c3f54 BS |
443 | struct mem_cgroup_per_zone *mz; |
444 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
445 | int nid = page_to_nid(page); |
446 | int zid = page_zonenum(page); | |
f64c3f54 BS |
447 | mctz = soft_limit_tree_from_page(page); |
448 | ||
449 | /* | |
4e649152 KH |
450 | * Necessary to update all ancestors when hierarchy is used. |
451 | * because their event counter is not touched. | |
f64c3f54 | 452 | */ |
4e649152 KH |
453 | for (; mem; mem = parent_mem_cgroup(mem)) { |
454 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 455 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
456 | /* |
457 | * We have to update the tree if mz is on RB-tree or | |
458 | * mem is over its softlimit. | |
459 | */ | |
ef8745c1 | 460 | if (excess || mz->on_tree) { |
4e649152 KH |
461 | spin_lock(&mctz->lock); |
462 | /* if on-tree, remove it */ | |
463 | if (mz->on_tree) | |
464 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
465 | /* | |
ef8745c1 KH |
466 | * Insert again. mz->usage_in_excess will be updated. |
467 | * If excess is 0, no tree ops. | |
4e649152 | 468 | */ |
ef8745c1 | 469 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
470 | spin_unlock(&mctz->lock); |
471 | } | |
f64c3f54 BS |
472 | } |
473 | } | |
474 | ||
475 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
476 | { | |
477 | int node, zone; | |
478 | struct mem_cgroup_per_zone *mz; | |
479 | struct mem_cgroup_tree_per_zone *mctz; | |
480 | ||
481 | for_each_node_state(node, N_POSSIBLE) { | |
482 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
483 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
484 | mctz = soft_limit_tree_node_zone(node, zone); | |
485 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
486 | } | |
487 | } | |
488 | } | |
489 | ||
4e416953 BS |
490 | static struct mem_cgroup_per_zone * |
491 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
492 | { | |
493 | struct rb_node *rightmost = NULL; | |
26251eaf | 494 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
495 | |
496 | retry: | |
26251eaf | 497 | mz = NULL; |
4e416953 BS |
498 | rightmost = rb_last(&mctz->rb_root); |
499 | if (!rightmost) | |
500 | goto done; /* Nothing to reclaim from */ | |
501 | ||
502 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
503 | /* | |
504 | * Remove the node now but someone else can add it back, | |
505 | * we will to add it back at the end of reclaim to its correct | |
506 | * position in the tree. | |
507 | */ | |
508 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
509 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
510 | !css_tryget(&mz->mem->css)) | |
511 | goto retry; | |
512 | done: | |
513 | return mz; | |
514 | } | |
515 | ||
516 | static struct mem_cgroup_per_zone * | |
517 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
518 | { | |
519 | struct mem_cgroup_per_zone *mz; | |
520 | ||
521 | spin_lock(&mctz->lock); | |
522 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
523 | spin_unlock(&mctz->lock); | |
524 | return mz; | |
525 | } | |
526 | ||
711d3d2c KH |
527 | /* |
528 | * Implementation Note: reading percpu statistics for memcg. | |
529 | * | |
530 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
531 | * synchronization to implement "quick" read. There are trade-off between | |
532 | * reading cost and precision of value. Then, we may have a chance to implement | |
533 | * a periodic synchronizion of counter in memcg's counter. | |
534 | * | |
535 | * But this _read() function is used for user interface now. The user accounts | |
536 | * memory usage by memory cgroup and he _always_ requires exact value because | |
537 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
538 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
539 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
540 | * | |
541 | * If there are kernel internal actions which can make use of some not-exact | |
542 | * value, and reading all cpu value can be performance bottleneck in some | |
543 | * common workload, threashold and synchonization as vmstat[] should be | |
544 | * implemented. | |
545 | */ | |
c62b1a3b KH |
546 | static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, |
547 | enum mem_cgroup_stat_index idx) | |
548 | { | |
549 | int cpu; | |
550 | s64 val = 0; | |
551 | ||
711d3d2c KH |
552 | get_online_cpus(); |
553 | for_each_online_cpu(cpu) | |
c62b1a3b | 554 | val += per_cpu(mem->stat->count[idx], cpu); |
711d3d2c KH |
555 | #ifdef CONFIG_HOTPLUG_CPU |
556 | spin_lock(&mem->pcp_counter_lock); | |
557 | val += mem->nocpu_base.count[idx]; | |
558 | spin_unlock(&mem->pcp_counter_lock); | |
559 | #endif | |
560 | put_online_cpus(); | |
c62b1a3b KH |
561 | return val; |
562 | } | |
563 | ||
564 | static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) | |
565 | { | |
566 | s64 ret; | |
567 | ||
568 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); | |
569 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); | |
570 | return ret; | |
571 | } | |
572 | ||
0c3e73e8 BS |
573 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
574 | bool charge) | |
575 | { | |
576 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 577 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
578 | } |
579 | ||
c05555b5 | 580 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
e401f176 | 581 | bool file, int nr_pages) |
d52aa412 | 582 | { |
c62b1a3b KH |
583 | preempt_disable(); |
584 | ||
e401f176 KH |
585 | if (file) |
586 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages); | |
d52aa412 | 587 | else |
e401f176 | 588 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages); |
55e462b0 | 589 | |
e401f176 KH |
590 | /* pagein of a big page is an event. So, ignore page size */ |
591 | if (nr_pages > 0) | |
c62b1a3b | 592 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); |
3751d604 | 593 | else { |
c62b1a3b | 594 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); |
3751d604 KH |
595 | nr_pages = -nr_pages; /* for event */ |
596 | } | |
e401f176 KH |
597 | |
598 | __this_cpu_add(mem->stat->count[MEM_CGROUP_EVENTS], nr_pages); | |
2e72b634 | 599 | |
c62b1a3b | 600 | preempt_enable(); |
6d12e2d8 KH |
601 | } |
602 | ||
14067bb3 | 603 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 604 | enum lru_list idx) |
6d12e2d8 KH |
605 | { |
606 | int nid, zid; | |
607 | struct mem_cgroup_per_zone *mz; | |
608 | u64 total = 0; | |
609 | ||
610 | for_each_online_node(nid) | |
611 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
612 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
613 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
614 | } | |
615 | return total; | |
d52aa412 KH |
616 | } |
617 | ||
d2265e6f KH |
618 | static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift) |
619 | { | |
620 | s64 val; | |
621 | ||
622 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_EVENTS]); | |
623 | ||
624 | return !(val & ((1 << event_mask_shift) - 1)); | |
625 | } | |
626 | ||
627 | /* | |
628 | * Check events in order. | |
629 | * | |
630 | */ | |
631 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) | |
632 | { | |
633 | /* threshold event is triggered in finer grain than soft limit */ | |
634 | if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) { | |
635 | mem_cgroup_threshold(mem); | |
636 | if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH))) | |
637 | mem_cgroup_update_tree(mem, page); | |
638 | } | |
639 | } | |
640 | ||
d5b69e38 | 641 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
642 | { |
643 | return container_of(cgroup_subsys_state(cont, | |
644 | mem_cgroup_subsys_id), struct mem_cgroup, | |
645 | css); | |
646 | } | |
647 | ||
cf475ad2 | 648 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 649 | { |
31a78f23 BS |
650 | /* |
651 | * mm_update_next_owner() may clear mm->owner to NULL | |
652 | * if it races with swapoff, page migration, etc. | |
653 | * So this can be called with p == NULL. | |
654 | */ | |
655 | if (unlikely(!p)) | |
656 | return NULL; | |
657 | ||
78fb7466 PE |
658 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
659 | struct mem_cgroup, css); | |
660 | } | |
661 | ||
54595fe2 KH |
662 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
663 | { | |
664 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
665 | |
666 | if (!mm) | |
667 | return NULL; | |
54595fe2 KH |
668 | /* |
669 | * Because we have no locks, mm->owner's may be being moved to other | |
670 | * cgroup. We use css_tryget() here even if this looks | |
671 | * pessimistic (rather than adding locks here). | |
672 | */ | |
673 | rcu_read_lock(); | |
674 | do { | |
675 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
676 | if (unlikely(!mem)) | |
677 | break; | |
678 | } while (!css_tryget(&mem->css)); | |
679 | rcu_read_unlock(); | |
680 | return mem; | |
681 | } | |
682 | ||
7d74b06f KH |
683 | /* The caller has to guarantee "mem" exists before calling this */ |
684 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem) | |
14067bb3 | 685 | { |
711d3d2c KH |
686 | struct cgroup_subsys_state *css; |
687 | int found; | |
688 | ||
689 | if (!mem) /* ROOT cgroup has the smallest ID */ | |
690 | return root_mem_cgroup; /*css_put/get against root is ignored*/ | |
691 | if (!mem->use_hierarchy) { | |
692 | if (css_tryget(&mem->css)) | |
693 | return mem; | |
694 | return NULL; | |
695 | } | |
696 | rcu_read_lock(); | |
697 | /* | |
698 | * searching a memory cgroup which has the smallest ID under given | |
699 | * ROOT cgroup. (ID >= 1) | |
700 | */ | |
701 | css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found); | |
702 | if (css && css_tryget(css)) | |
703 | mem = container_of(css, struct mem_cgroup, css); | |
704 | else | |
705 | mem = NULL; | |
706 | rcu_read_unlock(); | |
707 | return mem; | |
7d74b06f KH |
708 | } |
709 | ||
710 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, | |
711 | struct mem_cgroup *root, | |
712 | bool cond) | |
713 | { | |
714 | int nextid = css_id(&iter->css) + 1; | |
715 | int found; | |
716 | int hierarchy_used; | |
14067bb3 | 717 | struct cgroup_subsys_state *css; |
14067bb3 | 718 | |
7d74b06f | 719 | hierarchy_used = iter->use_hierarchy; |
14067bb3 | 720 | |
7d74b06f | 721 | css_put(&iter->css); |
711d3d2c KH |
722 | /* If no ROOT, walk all, ignore hierarchy */ |
723 | if (!cond || (root && !hierarchy_used)) | |
7d74b06f | 724 | return NULL; |
14067bb3 | 725 | |
711d3d2c KH |
726 | if (!root) |
727 | root = root_mem_cgroup; | |
728 | ||
7d74b06f KH |
729 | do { |
730 | iter = NULL; | |
14067bb3 | 731 | rcu_read_lock(); |
7d74b06f KH |
732 | |
733 | css = css_get_next(&mem_cgroup_subsys, nextid, | |
734 | &root->css, &found); | |
14067bb3 | 735 | if (css && css_tryget(css)) |
7d74b06f | 736 | iter = container_of(css, struct mem_cgroup, css); |
14067bb3 | 737 | rcu_read_unlock(); |
7d74b06f | 738 | /* If css is NULL, no more cgroups will be found */ |
14067bb3 | 739 | nextid = found + 1; |
7d74b06f | 740 | } while (css && !iter); |
14067bb3 | 741 | |
7d74b06f | 742 | return iter; |
14067bb3 | 743 | } |
7d74b06f KH |
744 | /* |
745 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please | |
746 | * be careful that "break" loop is not allowed. We have reference count. | |
747 | * Instead of that modify "cond" to be false and "continue" to exit the loop. | |
748 | */ | |
749 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ | |
750 | for (iter = mem_cgroup_start_loop(root);\ | |
751 | iter != NULL;\ | |
752 | iter = mem_cgroup_get_next(iter, root, cond)) | |
753 | ||
754 | #define for_each_mem_cgroup_tree(iter, root) \ | |
755 | for_each_mem_cgroup_tree_cond(iter, root, true) | |
756 | ||
711d3d2c KH |
757 | #define for_each_mem_cgroup_all(iter) \ |
758 | for_each_mem_cgroup_tree_cond(iter, NULL, true) | |
759 | ||
14067bb3 | 760 | |
4b3bde4c BS |
761 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
762 | { | |
763 | return (mem == root_mem_cgroup); | |
764 | } | |
765 | ||
08e552c6 KH |
766 | /* |
767 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
768 | * Operations are called by routine of global LRU independently from memcg. | |
769 | * What we have to take care of here is validness of pc->mem_cgroup. | |
770 | * | |
771 | * Changes to pc->mem_cgroup happens when | |
772 | * 1. charge | |
773 | * 2. moving account | |
774 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
775 | * It is added to LRU before charge. | |
776 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
777 | * When moving account, the page is not on LRU. It's isolated. | |
778 | */ | |
4f98a2fe | 779 | |
08e552c6 KH |
780 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
781 | { | |
782 | struct page_cgroup *pc; | |
08e552c6 | 783 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 784 | |
f8d66542 | 785 | if (mem_cgroup_disabled()) |
08e552c6 KH |
786 | return; |
787 | pc = lookup_page_cgroup(page); | |
788 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 789 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 790 | return; |
4b3bde4c | 791 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
792 | /* |
793 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
794 | * removed from global LRU. | |
795 | */ | |
97a6c37b | 796 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
797 | /* huge page split is done under lru_lock. so, we have no races. */ |
798 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); | |
4b3bde4c BS |
799 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
800 | return; | |
801 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 | 802 | list_del_init(&pc->lru); |
6d12e2d8 KH |
803 | } |
804 | ||
08e552c6 | 805 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 806 | { |
08e552c6 KH |
807 | mem_cgroup_del_lru_list(page, page_lru(page)); |
808 | } | |
b69408e8 | 809 | |
3f58a829 MK |
810 | /* |
811 | * Writeback is about to end against a page which has been marked for immediate | |
812 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
813 | * inactive list. | |
814 | */ | |
815 | void mem_cgroup_rotate_reclaimable_page(struct page *page) | |
816 | { | |
817 | struct mem_cgroup_per_zone *mz; | |
818 | struct page_cgroup *pc; | |
819 | enum lru_list lru = page_lru(page); | |
820 | ||
821 | if (mem_cgroup_disabled()) | |
822 | return; | |
823 | ||
824 | pc = lookup_page_cgroup(page); | |
825 | /* unused or root page is not rotated. */ | |
826 | if (!PageCgroupUsed(pc)) | |
827 | return; | |
828 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
829 | smp_rmb(); | |
830 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
831 | return; | |
97a6c37b | 832 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3f58a829 MK |
833 | list_move_tail(&pc->lru, &mz->lists[lru]); |
834 | } | |
835 | ||
08e552c6 KH |
836 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
837 | { | |
838 | struct mem_cgroup_per_zone *mz; | |
839 | struct page_cgroup *pc; | |
b69408e8 | 840 | |
f8d66542 | 841 | if (mem_cgroup_disabled()) |
08e552c6 | 842 | return; |
6d12e2d8 | 843 | |
08e552c6 | 844 | pc = lookup_page_cgroup(page); |
4b3bde4c | 845 | /* unused or root page is not rotated. */ |
713735b4 JW |
846 | if (!PageCgroupUsed(pc)) |
847 | return; | |
848 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
849 | smp_rmb(); | |
850 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 | 851 | return; |
97a6c37b | 852 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
08e552c6 | 853 | list_move(&pc->lru, &mz->lists[lru]); |
6d12e2d8 KH |
854 | } |
855 | ||
08e552c6 | 856 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 857 | { |
08e552c6 KH |
858 | struct page_cgroup *pc; |
859 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 860 | |
f8d66542 | 861 | if (mem_cgroup_disabled()) |
08e552c6 KH |
862 | return; |
863 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 864 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
08e552c6 | 865 | if (!PageCgroupUsed(pc)) |
894bc310 | 866 | return; |
713735b4 JW |
867 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
868 | smp_rmb(); | |
97a6c37b | 869 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
870 | /* huge page split is done under lru_lock. so, we have no races. */ |
871 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); | |
4b3bde4c BS |
872 | SetPageCgroupAcctLRU(pc); |
873 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
874 | return; | |
08e552c6 KH |
875 | list_add(&pc->lru, &mz->lists[lru]); |
876 | } | |
544122e5 | 877 | |
08e552c6 | 878 | /* |
544122e5 KH |
879 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
880 | * lru because the page may.be reused after it's fully uncharged (because of | |
881 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
882 | * it again. This function is only used to charge SwapCache. It's done under | |
883 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 884 | */ |
544122e5 | 885 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 886 | { |
544122e5 KH |
887 | unsigned long flags; |
888 | struct zone *zone = page_zone(page); | |
889 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
890 | ||
891 | spin_lock_irqsave(&zone->lru_lock, flags); | |
892 | /* | |
893 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
894 | * is guarded by lock_page() because the page is SwapCache. | |
895 | */ | |
896 | if (!PageCgroupUsed(pc)) | |
897 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
898 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
899 | } |
900 | ||
544122e5 KH |
901 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
902 | { | |
903 | unsigned long flags; | |
904 | struct zone *zone = page_zone(page); | |
905 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
906 | ||
907 | spin_lock_irqsave(&zone->lru_lock, flags); | |
908 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 909 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
910 | mem_cgroup_add_lru_list(page, page_lru(page)); |
911 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
912 | } | |
913 | ||
914 | ||
08e552c6 KH |
915 | void mem_cgroup_move_lists(struct page *page, |
916 | enum lru_list from, enum lru_list to) | |
917 | { | |
f8d66542 | 918 | if (mem_cgroup_disabled()) |
08e552c6 KH |
919 | return; |
920 | mem_cgroup_del_lru_list(page, from); | |
921 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
922 | } |
923 | ||
4c4a2214 DR |
924 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
925 | { | |
926 | int ret; | |
0b7f569e | 927 | struct mem_cgroup *curr = NULL; |
158e0a2d | 928 | struct task_struct *p; |
4c4a2214 | 929 | |
158e0a2d KH |
930 | p = find_lock_task_mm(task); |
931 | if (!p) | |
932 | return 0; | |
933 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
934 | task_unlock(p); | |
0b7f569e KH |
935 | if (!curr) |
936 | return 0; | |
d31f56db DN |
937 | /* |
938 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
939 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
940 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
941 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
942 | */ | |
943 | if (mem->use_hierarchy) | |
0b7f569e KH |
944 | ret = css_is_ancestor(&curr->css, &mem->css); |
945 | else | |
946 | ret = (curr == mem); | |
947 | css_put(&curr->css); | |
4c4a2214 DR |
948 | return ret; |
949 | } | |
950 | ||
c772be93 | 951 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
952 | { |
953 | unsigned long active; | |
954 | unsigned long inactive; | |
c772be93 KM |
955 | unsigned long gb; |
956 | unsigned long inactive_ratio; | |
14797e23 | 957 | |
14067bb3 KH |
958 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
959 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 960 | |
c772be93 KM |
961 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
962 | if (gb) | |
963 | inactive_ratio = int_sqrt(10 * gb); | |
964 | else | |
965 | inactive_ratio = 1; | |
966 | ||
967 | if (present_pages) { | |
968 | present_pages[0] = inactive; | |
969 | present_pages[1] = active; | |
970 | } | |
971 | ||
972 | return inactive_ratio; | |
973 | } | |
974 | ||
975 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
976 | { | |
977 | unsigned long active; | |
978 | unsigned long inactive; | |
979 | unsigned long present_pages[2]; | |
980 | unsigned long inactive_ratio; | |
981 | ||
982 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
983 | ||
984 | inactive = present_pages[0]; | |
985 | active = present_pages[1]; | |
986 | ||
987 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
988 | return 1; |
989 | ||
990 | return 0; | |
991 | } | |
992 | ||
56e49d21 RR |
993 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
994 | { | |
995 | unsigned long active; | |
996 | unsigned long inactive; | |
997 | ||
998 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
999 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
1000 | ||
1001 | return (active > inactive); | |
1002 | } | |
1003 | ||
a3d8e054 KM |
1004 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
1005 | struct zone *zone, | |
1006 | enum lru_list lru) | |
1007 | { | |
13d7e3a2 | 1008 | int nid = zone_to_nid(zone); |
a3d8e054 KM |
1009 | int zid = zone_idx(zone); |
1010 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1011 | ||
1012 | return MEM_CGROUP_ZSTAT(mz, lru); | |
1013 | } | |
1014 | ||
3e2f41f1 KM |
1015 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1016 | struct zone *zone) | |
1017 | { | |
13d7e3a2 | 1018 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1019 | int zid = zone_idx(zone); |
1020 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1021 | ||
1022 | return &mz->reclaim_stat; | |
1023 | } | |
1024 | ||
1025 | struct zone_reclaim_stat * | |
1026 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1027 | { | |
1028 | struct page_cgroup *pc; | |
1029 | struct mem_cgroup_per_zone *mz; | |
1030 | ||
1031 | if (mem_cgroup_disabled()) | |
1032 | return NULL; | |
1033 | ||
1034 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1035 | if (!PageCgroupUsed(pc)) |
1036 | return NULL; | |
713735b4 JW |
1037 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1038 | smp_rmb(); | |
97a6c37b | 1039 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3e2f41f1 KM |
1040 | return &mz->reclaim_stat; |
1041 | } | |
1042 | ||
66e1707b BS |
1043 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
1044 | struct list_head *dst, | |
1045 | unsigned long *scanned, int order, | |
1046 | int mode, struct zone *z, | |
1047 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 1048 | int active, int file) |
66e1707b BS |
1049 | { |
1050 | unsigned long nr_taken = 0; | |
1051 | struct page *page; | |
1052 | unsigned long scan; | |
1053 | LIST_HEAD(pc_list); | |
1054 | struct list_head *src; | |
ff7283fa | 1055 | struct page_cgroup *pc, *tmp; |
13d7e3a2 | 1056 | int nid = zone_to_nid(z); |
1ecaab2b KH |
1057 | int zid = zone_idx(z); |
1058 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 1059 | int lru = LRU_FILE * file + active; |
2ffebca6 | 1060 | int ret; |
66e1707b | 1061 | |
cf475ad2 | 1062 | BUG_ON(!mem_cont); |
1ecaab2b | 1063 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 1064 | src = &mz->lists[lru]; |
66e1707b | 1065 | |
ff7283fa KH |
1066 | scan = 0; |
1067 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 1068 | if (scan >= nr_to_scan) |
ff7283fa | 1069 | break; |
08e552c6 | 1070 | |
52d4b9ac KH |
1071 | if (unlikely(!PageCgroupUsed(pc))) |
1072 | continue; | |
5564e88b | 1073 | |
6b3ae58e | 1074 | page = lookup_cgroup_page(pc); |
5564e88b | 1075 | |
436c6541 | 1076 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1077 | continue; |
ff7283fa | 1078 | |
436c6541 | 1079 | scan++; |
2ffebca6 KH |
1080 | ret = __isolate_lru_page(page, mode, file); |
1081 | switch (ret) { | |
1082 | case 0: | |
66e1707b | 1083 | list_move(&page->lru, dst); |
2ffebca6 | 1084 | mem_cgroup_del_lru(page); |
2c888cfb | 1085 | nr_taken += hpage_nr_pages(page); |
2ffebca6 KH |
1086 | break; |
1087 | case -EBUSY: | |
1088 | /* we don't affect global LRU but rotate in our LRU */ | |
1089 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1090 | break; | |
1091 | default: | |
1092 | break; | |
66e1707b BS |
1093 | } |
1094 | } | |
1095 | ||
66e1707b | 1096 | *scanned = scan; |
cc8e970c KM |
1097 | |
1098 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, | |
1099 | 0, 0, 0, mode); | |
1100 | ||
66e1707b BS |
1101 | return nr_taken; |
1102 | } | |
1103 | ||
6d61ef40 BS |
1104 | #define mem_cgroup_from_res_counter(counter, member) \ |
1105 | container_of(counter, struct mem_cgroup, member) | |
1106 | ||
19942822 | 1107 | /** |
9d11ea9f JW |
1108 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1109 | * @mem: the memory cgroup | |
19942822 | 1110 | * |
9d11ea9f JW |
1111 | * Returns the maximum amount of memory @mem can be charged with, in |
1112 | * bytes. | |
19942822 | 1113 | */ |
9d11ea9f | 1114 | static unsigned long long mem_cgroup_margin(struct mem_cgroup *mem) |
19942822 | 1115 | { |
9d11ea9f JW |
1116 | unsigned long long margin; |
1117 | ||
1118 | margin = res_counter_margin(&mem->res); | |
1119 | if (do_swap_account) | |
1120 | margin = min(margin, res_counter_margin(&mem->memsw)); | |
1121 | return margin; | |
19942822 JW |
1122 | } |
1123 | ||
a7885eb8 KM |
1124 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1125 | { | |
1126 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1127 | |
1128 | /* root ? */ | |
1129 | if (cgrp->parent == NULL) | |
1130 | return vm_swappiness; | |
1131 | ||
bf1ff263 | 1132 | return memcg->swappiness; |
a7885eb8 KM |
1133 | } |
1134 | ||
32047e2a KH |
1135 | static void mem_cgroup_start_move(struct mem_cgroup *mem) |
1136 | { | |
1137 | int cpu; | |
1489ebad KH |
1138 | |
1139 | get_online_cpus(); | |
1140 | spin_lock(&mem->pcp_counter_lock); | |
1141 | for_each_online_cpu(cpu) | |
32047e2a | 1142 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1489ebad KH |
1143 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; |
1144 | spin_unlock(&mem->pcp_counter_lock); | |
1145 | put_online_cpus(); | |
32047e2a KH |
1146 | |
1147 | synchronize_rcu(); | |
1148 | } | |
1149 | ||
1150 | static void mem_cgroup_end_move(struct mem_cgroup *mem) | |
1151 | { | |
1152 | int cpu; | |
1153 | ||
1154 | if (!mem) | |
1155 | return; | |
1489ebad KH |
1156 | get_online_cpus(); |
1157 | spin_lock(&mem->pcp_counter_lock); | |
1158 | for_each_online_cpu(cpu) | |
32047e2a | 1159 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1489ebad KH |
1160 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; |
1161 | spin_unlock(&mem->pcp_counter_lock); | |
1162 | put_online_cpus(); | |
32047e2a KH |
1163 | } |
1164 | /* | |
1165 | * 2 routines for checking "mem" is under move_account() or not. | |
1166 | * | |
1167 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1168 | * for avoiding race in accounting. If true, | |
1169 | * pc->mem_cgroup may be overwritten. | |
1170 | * | |
1171 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1172 | * under hierarchy of moving cgroups. This is for | |
1173 | * waiting at hith-memory prressure caused by "move". | |
1174 | */ | |
1175 | ||
1176 | static bool mem_cgroup_stealed(struct mem_cgroup *mem) | |
1177 | { | |
1178 | VM_BUG_ON(!rcu_read_lock_held()); | |
1179 | return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0; | |
1180 | } | |
4b534334 KH |
1181 | |
1182 | static bool mem_cgroup_under_move(struct mem_cgroup *mem) | |
1183 | { | |
2bd9bb20 KH |
1184 | struct mem_cgroup *from; |
1185 | struct mem_cgroup *to; | |
4b534334 | 1186 | bool ret = false; |
2bd9bb20 KH |
1187 | /* |
1188 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1189 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1190 | */ | |
1191 | spin_lock(&mc.lock); | |
1192 | from = mc.from; | |
1193 | to = mc.to; | |
1194 | if (!from) | |
1195 | goto unlock; | |
1196 | if (from == mem || to == mem | |
1197 | || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css)) | |
1198 | || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css))) | |
1199 | ret = true; | |
1200 | unlock: | |
1201 | spin_unlock(&mc.lock); | |
4b534334 KH |
1202 | return ret; |
1203 | } | |
1204 | ||
1205 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem) | |
1206 | { | |
1207 | if (mc.moving_task && current != mc.moving_task) { | |
1208 | if (mem_cgroup_under_move(mem)) { | |
1209 | DEFINE_WAIT(wait); | |
1210 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1211 | /* moving charge context might have finished. */ | |
1212 | if (mc.moving_task) | |
1213 | schedule(); | |
1214 | finish_wait(&mc.waitq, &wait); | |
1215 | return true; | |
1216 | } | |
1217 | } | |
1218 | return false; | |
1219 | } | |
1220 | ||
e222432b | 1221 | /** |
6a6135b6 | 1222 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1223 | * @memcg: The memory cgroup that went over limit |
1224 | * @p: Task that is going to be killed | |
1225 | * | |
1226 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1227 | * enabled | |
1228 | */ | |
1229 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1230 | { | |
1231 | struct cgroup *task_cgrp; | |
1232 | struct cgroup *mem_cgrp; | |
1233 | /* | |
1234 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1235 | * on the assumption that OOM is serialized for memory controller. | |
1236 | * If this assumption is broken, revisit this code. | |
1237 | */ | |
1238 | static char memcg_name[PATH_MAX]; | |
1239 | int ret; | |
1240 | ||
d31f56db | 1241 | if (!memcg || !p) |
e222432b BS |
1242 | return; |
1243 | ||
1244 | ||
1245 | rcu_read_lock(); | |
1246 | ||
1247 | mem_cgrp = memcg->css.cgroup; | |
1248 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1249 | ||
1250 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1251 | if (ret < 0) { | |
1252 | /* | |
1253 | * Unfortunately, we are unable to convert to a useful name | |
1254 | * But we'll still print out the usage information | |
1255 | */ | |
1256 | rcu_read_unlock(); | |
1257 | goto done; | |
1258 | } | |
1259 | rcu_read_unlock(); | |
1260 | ||
1261 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1262 | ||
1263 | rcu_read_lock(); | |
1264 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1265 | if (ret < 0) { | |
1266 | rcu_read_unlock(); | |
1267 | goto done; | |
1268 | } | |
1269 | rcu_read_unlock(); | |
1270 | ||
1271 | /* | |
1272 | * Continues from above, so we don't need an KERN_ level | |
1273 | */ | |
1274 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1275 | done: | |
1276 | ||
1277 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1278 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1279 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1280 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1281 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1282 | "failcnt %llu\n", | |
1283 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1284 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1285 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1286 | } | |
1287 | ||
81d39c20 KH |
1288 | /* |
1289 | * This function returns the number of memcg under hierarchy tree. Returns | |
1290 | * 1(self count) if no children. | |
1291 | */ | |
1292 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1293 | { | |
1294 | int num = 0; | |
7d74b06f KH |
1295 | struct mem_cgroup *iter; |
1296 | ||
1297 | for_each_mem_cgroup_tree(iter, mem) | |
1298 | num++; | |
81d39c20 KH |
1299 | return num; |
1300 | } | |
1301 | ||
a63d83f4 DR |
1302 | /* |
1303 | * Return the memory (and swap, if configured) limit for a memcg. | |
1304 | */ | |
1305 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1306 | { | |
1307 | u64 limit; | |
1308 | u64 memsw; | |
1309 | ||
f3e8eb70 JW |
1310 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1311 | limit += total_swap_pages << PAGE_SHIFT; | |
1312 | ||
a63d83f4 DR |
1313 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1314 | /* | |
1315 | * If memsw is finite and limits the amount of swap space available | |
1316 | * to this memcg, return that limit. | |
1317 | */ | |
1318 | return min(limit, memsw); | |
1319 | } | |
1320 | ||
6d61ef40 | 1321 | /* |
04046e1a KH |
1322 | * Visit the first child (need not be the first child as per the ordering |
1323 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1324 | * that to reclaim free pages from. | |
1325 | */ | |
1326 | static struct mem_cgroup * | |
1327 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1328 | { | |
1329 | struct mem_cgroup *ret = NULL; | |
1330 | struct cgroup_subsys_state *css; | |
1331 | int nextid, found; | |
1332 | ||
1333 | if (!root_mem->use_hierarchy) { | |
1334 | css_get(&root_mem->css); | |
1335 | ret = root_mem; | |
1336 | } | |
1337 | ||
1338 | while (!ret) { | |
1339 | rcu_read_lock(); | |
1340 | nextid = root_mem->last_scanned_child + 1; | |
1341 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1342 | &found); | |
1343 | if (css && css_tryget(css)) | |
1344 | ret = container_of(css, struct mem_cgroup, css); | |
1345 | ||
1346 | rcu_read_unlock(); | |
1347 | /* Updates scanning parameter */ | |
04046e1a KH |
1348 | if (!css) { |
1349 | /* this means start scan from ID:1 */ | |
1350 | root_mem->last_scanned_child = 0; | |
1351 | } else | |
1352 | root_mem->last_scanned_child = found; | |
04046e1a KH |
1353 | } |
1354 | ||
1355 | return ret; | |
1356 | } | |
1357 | ||
1358 | /* | |
1359 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1360 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1361 | * based on its position in the children list. | |
6d61ef40 BS |
1362 | * |
1363 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1364 | * |
1365 | * We give up and return to the caller when we visit root_mem twice. | |
1366 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1367 | * |
1368 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1369 | */ |
1370 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1371 | struct zone *zone, |
75822b44 BS |
1372 | gfp_t gfp_mask, |
1373 | unsigned long reclaim_options) | |
6d61ef40 | 1374 | { |
04046e1a KH |
1375 | struct mem_cgroup *victim; |
1376 | int ret, total = 0; | |
1377 | int loop = 0; | |
75822b44 BS |
1378 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1379 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 | 1380 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
9d11ea9f JW |
1381 | unsigned long excess; |
1382 | ||
1383 | excess = res_counter_soft_limit_excess(&root_mem->res) >> PAGE_SHIFT; | |
04046e1a | 1384 | |
22a668d7 KH |
1385 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1386 | if (root_mem->memsw_is_minimum) | |
1387 | noswap = true; | |
1388 | ||
4e416953 | 1389 | while (1) { |
04046e1a | 1390 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1391 | if (victim == root_mem) { |
04046e1a | 1392 | loop++; |
cdec2e42 KH |
1393 | if (loop >= 1) |
1394 | drain_all_stock_async(); | |
4e416953 BS |
1395 | if (loop >= 2) { |
1396 | /* | |
1397 | * If we have not been able to reclaim | |
1398 | * anything, it might because there are | |
1399 | * no reclaimable pages under this hierarchy | |
1400 | */ | |
1401 | if (!check_soft || !total) { | |
1402 | css_put(&victim->css); | |
1403 | break; | |
1404 | } | |
1405 | /* | |
1406 | * We want to do more targetted reclaim. | |
1407 | * excess >> 2 is not to excessive so as to | |
1408 | * reclaim too much, nor too less that we keep | |
1409 | * coming back to reclaim from this cgroup | |
1410 | */ | |
1411 | if (total >= (excess >> 2) || | |
1412 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1413 | css_put(&victim->css); | |
1414 | break; | |
1415 | } | |
1416 | } | |
1417 | } | |
c62b1a3b | 1418 | if (!mem_cgroup_local_usage(victim)) { |
04046e1a KH |
1419 | /* this cgroup's local usage == 0 */ |
1420 | css_put(&victim->css); | |
6d61ef40 BS |
1421 | continue; |
1422 | } | |
04046e1a | 1423 | /* we use swappiness of local cgroup */ |
4e416953 BS |
1424 | if (check_soft) |
1425 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
14fec796 | 1426 | noswap, get_swappiness(victim), zone); |
4e416953 BS |
1427 | else |
1428 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1429 | noswap, get_swappiness(victim)); | |
04046e1a | 1430 | css_put(&victim->css); |
81d39c20 KH |
1431 | /* |
1432 | * At shrinking usage, we can't check we should stop here or | |
1433 | * reclaim more. It's depends on callers. last_scanned_child | |
1434 | * will work enough for keeping fairness under tree. | |
1435 | */ | |
1436 | if (shrink) | |
1437 | return ret; | |
04046e1a | 1438 | total += ret; |
4e416953 | 1439 | if (check_soft) { |
9d11ea9f | 1440 | if (!res_counter_soft_limit_excess(&root_mem->res)) |
4e416953 | 1441 | return total; |
9d11ea9f | 1442 | } else if (mem_cgroup_margin(root_mem)) |
04046e1a | 1443 | return 1 + total; |
6d61ef40 | 1444 | } |
04046e1a | 1445 | return total; |
6d61ef40 BS |
1446 | } |
1447 | ||
867578cb KH |
1448 | /* |
1449 | * Check OOM-Killer is already running under our hierarchy. | |
1450 | * If someone is running, return false. | |
1451 | */ | |
1452 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) | |
1453 | { | |
7d74b06f KH |
1454 | int x, lock_count = 0; |
1455 | struct mem_cgroup *iter; | |
a636b327 | 1456 | |
7d74b06f KH |
1457 | for_each_mem_cgroup_tree(iter, mem) { |
1458 | x = atomic_inc_return(&iter->oom_lock); | |
1459 | lock_count = max(x, lock_count); | |
1460 | } | |
867578cb KH |
1461 | |
1462 | if (lock_count == 1) | |
1463 | return true; | |
1464 | return false; | |
a636b327 | 1465 | } |
0b7f569e | 1466 | |
7d74b06f | 1467 | static int mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
0b7f569e | 1468 | { |
7d74b06f KH |
1469 | struct mem_cgroup *iter; |
1470 | ||
867578cb KH |
1471 | /* |
1472 | * When a new child is created while the hierarchy is under oom, | |
1473 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1474 | * atomic_add_unless() here. | |
1475 | */ | |
7d74b06f KH |
1476 | for_each_mem_cgroup_tree(iter, mem) |
1477 | atomic_add_unless(&iter->oom_lock, -1, 0); | |
0b7f569e KH |
1478 | return 0; |
1479 | } | |
1480 | ||
867578cb KH |
1481 | |
1482 | static DEFINE_MUTEX(memcg_oom_mutex); | |
1483 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); | |
1484 | ||
dc98df5a KH |
1485 | struct oom_wait_info { |
1486 | struct mem_cgroup *mem; | |
1487 | wait_queue_t wait; | |
1488 | }; | |
1489 | ||
1490 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1491 | unsigned mode, int sync, void *arg) | |
1492 | { | |
1493 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; | |
1494 | struct oom_wait_info *oom_wait_info; | |
1495 | ||
1496 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
1497 | ||
1498 | if (oom_wait_info->mem == wake_mem) | |
1499 | goto wakeup; | |
1500 | /* if no hierarchy, no match */ | |
1501 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) | |
1502 | return 0; | |
1503 | /* | |
1504 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1505 | * Then we can use css_is_ancestor without taking care of RCU. | |
1506 | */ | |
1507 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && | |
1508 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) | |
1509 | return 0; | |
1510 | ||
1511 | wakeup: | |
1512 | return autoremove_wake_function(wait, mode, sync, arg); | |
1513 | } | |
1514 | ||
1515 | static void memcg_wakeup_oom(struct mem_cgroup *mem) | |
1516 | { | |
1517 | /* for filtering, pass "mem" as argument. */ | |
1518 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); | |
1519 | } | |
1520 | ||
3c11ecf4 KH |
1521 | static void memcg_oom_recover(struct mem_cgroup *mem) |
1522 | { | |
2bd9bb20 | 1523 | if (mem && atomic_read(&mem->oom_lock)) |
3c11ecf4 KH |
1524 | memcg_wakeup_oom(mem); |
1525 | } | |
1526 | ||
867578cb KH |
1527 | /* |
1528 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1529 | */ | |
1530 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) | |
0b7f569e | 1531 | { |
dc98df5a | 1532 | struct oom_wait_info owait; |
3c11ecf4 | 1533 | bool locked, need_to_kill; |
867578cb | 1534 | |
dc98df5a KH |
1535 | owait.mem = mem; |
1536 | owait.wait.flags = 0; | |
1537 | owait.wait.func = memcg_oom_wake_function; | |
1538 | owait.wait.private = current; | |
1539 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1540 | need_to_kill = true; |
867578cb KH |
1541 | /* At first, try to OOM lock hierarchy under mem.*/ |
1542 | mutex_lock(&memcg_oom_mutex); | |
1543 | locked = mem_cgroup_oom_lock(mem); | |
1544 | /* | |
1545 | * Even if signal_pending(), we can't quit charge() loop without | |
1546 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1547 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1548 | */ | |
3c11ecf4 KH |
1549 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
1550 | if (!locked || mem->oom_kill_disable) | |
1551 | need_to_kill = false; | |
1552 | if (locked) | |
9490ff27 | 1553 | mem_cgroup_oom_notify(mem); |
867578cb KH |
1554 | mutex_unlock(&memcg_oom_mutex); |
1555 | ||
3c11ecf4 KH |
1556 | if (need_to_kill) { |
1557 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
867578cb | 1558 | mem_cgroup_out_of_memory(mem, mask); |
3c11ecf4 | 1559 | } else { |
867578cb | 1560 | schedule(); |
dc98df5a | 1561 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb KH |
1562 | } |
1563 | mutex_lock(&memcg_oom_mutex); | |
1564 | mem_cgroup_oom_unlock(mem); | |
dc98df5a | 1565 | memcg_wakeup_oom(mem); |
867578cb KH |
1566 | mutex_unlock(&memcg_oom_mutex); |
1567 | ||
1568 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) | |
1569 | return false; | |
1570 | /* Give chance to dying process */ | |
1571 | schedule_timeout(1); | |
1572 | return true; | |
0b7f569e KH |
1573 | } |
1574 | ||
d69b042f BS |
1575 | /* |
1576 | * Currently used to update mapped file statistics, but the routine can be | |
1577 | * generalized to update other statistics as well. | |
32047e2a KH |
1578 | * |
1579 | * Notes: Race condition | |
1580 | * | |
1581 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1582 | * it tends to be costly. But considering some conditions, we doesn't need | |
1583 | * to do so _always_. | |
1584 | * | |
1585 | * Considering "charge", lock_page_cgroup() is not required because all | |
1586 | * file-stat operations happen after a page is attached to radix-tree. There | |
1587 | * are no race with "charge". | |
1588 | * | |
1589 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1590 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1591 | * if there are race with "uncharge". Statistics itself is properly handled | |
1592 | * by flags. | |
1593 | * | |
1594 | * Considering "move", this is an only case we see a race. To make the race | |
1595 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
1596 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 1597 | */ |
26174efd | 1598 | |
2a7106f2 GT |
1599 | void mem_cgroup_update_page_stat(struct page *page, |
1600 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f BS |
1601 | { |
1602 | struct mem_cgroup *mem; | |
32047e2a KH |
1603 | struct page_cgroup *pc = lookup_page_cgroup(page); |
1604 | bool need_unlock = false; | |
dbd4ea78 | 1605 | unsigned long uninitialized_var(flags); |
d69b042f | 1606 | |
d69b042f BS |
1607 | if (unlikely(!pc)) |
1608 | return; | |
1609 | ||
32047e2a | 1610 | rcu_read_lock(); |
d69b042f | 1611 | mem = pc->mem_cgroup; |
32047e2a KH |
1612 | if (unlikely(!mem || !PageCgroupUsed(pc))) |
1613 | goto out; | |
1614 | /* pc->mem_cgroup is unstable ? */ | |
ca3e0214 | 1615 | if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) { |
32047e2a | 1616 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 1617 | move_lock_page_cgroup(pc, &flags); |
32047e2a KH |
1618 | need_unlock = true; |
1619 | mem = pc->mem_cgroup; | |
1620 | if (!mem || !PageCgroupUsed(pc)) | |
1621 | goto out; | |
1622 | } | |
26174efd | 1623 | |
26174efd | 1624 | switch (idx) { |
2a7106f2 | 1625 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
1626 | if (val > 0) |
1627 | SetPageCgroupFileMapped(pc); | |
1628 | else if (!page_mapped(page)) | |
0c270f8f | 1629 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 1630 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1631 | break; |
1632 | default: | |
1633 | BUG(); | |
8725d541 | 1634 | } |
d69b042f | 1635 | |
2a7106f2 GT |
1636 | this_cpu_add(mem->stat->count[idx], val); |
1637 | ||
32047e2a KH |
1638 | out: |
1639 | if (unlikely(need_unlock)) | |
dbd4ea78 | 1640 | move_unlock_page_cgroup(pc, &flags); |
32047e2a KH |
1641 | rcu_read_unlock(); |
1642 | return; | |
d69b042f | 1643 | } |
2a7106f2 | 1644 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
26174efd | 1645 | |
cdec2e42 KH |
1646 | /* |
1647 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1648 | * TODO: maybe necessary to use big numbers in big irons. | |
1649 | */ | |
1650 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1651 | struct memcg_stock_pcp { | |
1652 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1653 | int charge; | |
1654 | struct work_struct work; | |
1655 | }; | |
1656 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1657 | static atomic_t memcg_drain_count; | |
1658 | ||
1659 | /* | |
1660 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1661 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1662 | * cgroup which is not current target, returns false. This stock will be | |
1663 | * refilled. | |
1664 | */ | |
1665 | static bool consume_stock(struct mem_cgroup *mem) | |
1666 | { | |
1667 | struct memcg_stock_pcp *stock; | |
1668 | bool ret = true; | |
1669 | ||
1670 | stock = &get_cpu_var(memcg_stock); | |
1671 | if (mem == stock->cached && stock->charge) | |
1672 | stock->charge -= PAGE_SIZE; | |
1673 | else /* need to call res_counter_charge */ | |
1674 | ret = false; | |
1675 | put_cpu_var(memcg_stock); | |
1676 | return ret; | |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1681 | */ | |
1682 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1683 | { | |
1684 | struct mem_cgroup *old = stock->cached; | |
1685 | ||
1686 | if (stock->charge) { | |
1687 | res_counter_uncharge(&old->res, stock->charge); | |
1688 | if (do_swap_account) | |
1689 | res_counter_uncharge(&old->memsw, stock->charge); | |
1690 | } | |
1691 | stock->cached = NULL; | |
1692 | stock->charge = 0; | |
1693 | } | |
1694 | ||
1695 | /* | |
1696 | * This must be called under preempt disabled or must be called by | |
1697 | * a thread which is pinned to local cpu. | |
1698 | */ | |
1699 | static void drain_local_stock(struct work_struct *dummy) | |
1700 | { | |
1701 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1702 | drain_stock(stock); | |
1703 | } | |
1704 | ||
1705 | /* | |
1706 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 1707 | * This will be consumed by consume_stock() function, later. |
cdec2e42 KH |
1708 | */ |
1709 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1710 | { | |
1711 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1712 | ||
1713 | if (stock->cached != mem) { /* reset if necessary */ | |
1714 | drain_stock(stock); | |
1715 | stock->cached = mem; | |
1716 | } | |
1717 | stock->charge += val; | |
1718 | put_cpu_var(memcg_stock); | |
1719 | } | |
1720 | ||
1721 | /* | |
1722 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1723 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1724 | * expects some charges will be back to res_counter later but cannot wait for | |
1725 | * it. | |
1726 | */ | |
1727 | static void drain_all_stock_async(void) | |
1728 | { | |
1729 | int cpu; | |
1730 | /* This function is for scheduling "drain" in asynchronous way. | |
1731 | * The result of "drain" is not directly handled by callers. Then, | |
1732 | * if someone is calling drain, we don't have to call drain more. | |
1733 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1734 | * there is a race. We just do loose check here. | |
1735 | */ | |
1736 | if (atomic_read(&memcg_drain_count)) | |
1737 | return; | |
1738 | /* Notify other cpus that system-wide "drain" is running */ | |
1739 | atomic_inc(&memcg_drain_count); | |
1740 | get_online_cpus(); | |
1741 | for_each_online_cpu(cpu) { | |
1742 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1743 | schedule_work_on(cpu, &stock->work); | |
1744 | } | |
1745 | put_online_cpus(); | |
1746 | atomic_dec(&memcg_drain_count); | |
1747 | /* We don't wait for flush_work */ | |
1748 | } | |
1749 | ||
1750 | /* This is a synchronous drain interface. */ | |
1751 | static void drain_all_stock_sync(void) | |
1752 | { | |
1753 | /* called when force_empty is called */ | |
1754 | atomic_inc(&memcg_drain_count); | |
1755 | schedule_on_each_cpu(drain_local_stock); | |
1756 | atomic_dec(&memcg_drain_count); | |
1757 | } | |
1758 | ||
711d3d2c KH |
1759 | /* |
1760 | * This function drains percpu counter value from DEAD cpu and | |
1761 | * move it to local cpu. Note that this function can be preempted. | |
1762 | */ | |
1763 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu) | |
1764 | { | |
1765 | int i; | |
1766 | ||
1767 | spin_lock(&mem->pcp_counter_lock); | |
1768 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { | |
1769 | s64 x = per_cpu(mem->stat->count[i], cpu); | |
1770 | ||
1771 | per_cpu(mem->stat->count[i], cpu) = 0; | |
1772 | mem->nocpu_base.count[i] += x; | |
1773 | } | |
1489ebad KH |
1774 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
1775 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; | |
1776 | spin_unlock(&mem->pcp_counter_lock); | |
1777 | } | |
1778 | ||
1779 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu) | |
1780 | { | |
1781 | int idx = MEM_CGROUP_ON_MOVE; | |
1782 | ||
1783 | spin_lock(&mem->pcp_counter_lock); | |
1784 | per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx]; | |
711d3d2c KH |
1785 | spin_unlock(&mem->pcp_counter_lock); |
1786 | } | |
1787 | ||
1788 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
1789 | unsigned long action, |
1790 | void *hcpu) | |
1791 | { | |
1792 | int cpu = (unsigned long)hcpu; | |
1793 | struct memcg_stock_pcp *stock; | |
711d3d2c | 1794 | struct mem_cgroup *iter; |
cdec2e42 | 1795 | |
1489ebad KH |
1796 | if ((action == CPU_ONLINE)) { |
1797 | for_each_mem_cgroup_all(iter) | |
1798 | synchronize_mem_cgroup_on_move(iter, cpu); | |
1799 | return NOTIFY_OK; | |
1800 | } | |
1801 | ||
711d3d2c | 1802 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 1803 | return NOTIFY_OK; |
711d3d2c KH |
1804 | |
1805 | for_each_mem_cgroup_all(iter) | |
1806 | mem_cgroup_drain_pcp_counter(iter, cpu); | |
1807 | ||
cdec2e42 KH |
1808 | stock = &per_cpu(memcg_stock, cpu); |
1809 | drain_stock(stock); | |
1810 | return NOTIFY_OK; | |
1811 | } | |
1812 | ||
4b534334 KH |
1813 | |
1814 | /* See __mem_cgroup_try_charge() for details */ | |
1815 | enum { | |
1816 | CHARGE_OK, /* success */ | |
1817 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
1818 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
1819 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
1820 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
1821 | }; | |
1822 | ||
1823 | static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask, | |
1824 | int csize, bool oom_check) | |
1825 | { | |
1826 | struct mem_cgroup *mem_over_limit; | |
1827 | struct res_counter *fail_res; | |
1828 | unsigned long flags = 0; | |
1829 | int ret; | |
1830 | ||
1831 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
1832 | ||
1833 | if (likely(!ret)) { | |
1834 | if (!do_swap_account) | |
1835 | return CHARGE_OK; | |
1836 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); | |
1837 | if (likely(!ret)) | |
1838 | return CHARGE_OK; | |
1839 | ||
01c88e2d | 1840 | res_counter_uncharge(&mem->res, csize); |
4b534334 KH |
1841 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
1842 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
1843 | } else | |
1844 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 JW |
1845 | /* |
1846 | * csize can be either a huge page (HPAGE_SIZE), a batch of | |
1847 | * regular pages (CHARGE_SIZE), or a single regular page | |
1848 | * (PAGE_SIZE). | |
1849 | * | |
1850 | * Never reclaim on behalf of optional batching, retry with a | |
1851 | * single page instead. | |
1852 | */ | |
1853 | if (csize == CHARGE_SIZE) | |
4b534334 KH |
1854 | return CHARGE_RETRY; |
1855 | ||
1856 | if (!(gfp_mask & __GFP_WAIT)) | |
1857 | return CHARGE_WOULDBLOCK; | |
1858 | ||
1859 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, | |
19942822 | 1860 | gfp_mask, flags); |
9d11ea9f | 1861 | if (mem_cgroup_margin(mem_over_limit) >= csize) |
19942822 | 1862 | return CHARGE_RETRY; |
4b534334 | 1863 | /* |
19942822 JW |
1864 | * Even though the limit is exceeded at this point, reclaim |
1865 | * may have been able to free some pages. Retry the charge | |
1866 | * before killing the task. | |
1867 | * | |
1868 | * Only for regular pages, though: huge pages are rather | |
1869 | * unlikely to succeed so close to the limit, and we fall back | |
1870 | * to regular pages anyway in case of failure. | |
4b534334 | 1871 | */ |
19942822 | 1872 | if (csize == PAGE_SIZE && ret) |
4b534334 KH |
1873 | return CHARGE_RETRY; |
1874 | ||
1875 | /* | |
1876 | * At task move, charge accounts can be doubly counted. So, it's | |
1877 | * better to wait until the end of task_move if something is going on. | |
1878 | */ | |
1879 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
1880 | return CHARGE_RETRY; | |
1881 | ||
1882 | /* If we don't need to call oom-killer at el, return immediately */ | |
1883 | if (!oom_check) | |
1884 | return CHARGE_NOMEM; | |
1885 | /* check OOM */ | |
1886 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) | |
1887 | return CHARGE_OOM_DIE; | |
1888 | ||
1889 | return CHARGE_RETRY; | |
1890 | } | |
1891 | ||
f817ed48 KH |
1892 | /* |
1893 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1894 | * oom-killer can be invoked. | |
8a9f3ccd | 1895 | */ |
f817ed48 | 1896 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 AA |
1897 | gfp_t gfp_mask, |
1898 | struct mem_cgroup **memcg, bool oom, | |
1899 | int page_size) | |
8a9f3ccd | 1900 | { |
4b534334 KH |
1901 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
1902 | struct mem_cgroup *mem = NULL; | |
1903 | int ret; | |
ec168510 | 1904 | int csize = max(CHARGE_SIZE, (unsigned long) page_size); |
a636b327 | 1905 | |
867578cb KH |
1906 | /* |
1907 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
1908 | * in system level. So, allow to go ahead dying process in addition to | |
1909 | * MEMDIE process. | |
1910 | */ | |
1911 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
1912 | || fatal_signal_pending(current))) | |
1913 | goto bypass; | |
a636b327 | 1914 | |
8a9f3ccd | 1915 | /* |
3be91277 HD |
1916 | * We always charge the cgroup the mm_struct belongs to. |
1917 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
1918 | * thread group leader migrates. It's possible that mm is not |
1919 | * set, if so charge the init_mm (happens for pagecache usage). | |
1920 | */ | |
f75ca962 KH |
1921 | if (!*memcg && !mm) |
1922 | goto bypass; | |
1923 | again: | |
1924 | if (*memcg) { /* css should be a valid one */ | |
4b534334 | 1925 | mem = *memcg; |
f75ca962 KH |
1926 | VM_BUG_ON(css_is_removed(&mem->css)); |
1927 | if (mem_cgroup_is_root(mem)) | |
1928 | goto done; | |
ec168510 | 1929 | if (page_size == PAGE_SIZE && consume_stock(mem)) |
f75ca962 | 1930 | goto done; |
4b534334 KH |
1931 | css_get(&mem->css); |
1932 | } else { | |
f75ca962 | 1933 | struct task_struct *p; |
54595fe2 | 1934 | |
f75ca962 KH |
1935 | rcu_read_lock(); |
1936 | p = rcu_dereference(mm->owner); | |
f75ca962 | 1937 | /* |
ebb76ce1 KH |
1938 | * Because we don't have task_lock(), "p" can exit. |
1939 | * In that case, "mem" can point to root or p can be NULL with | |
1940 | * race with swapoff. Then, we have small risk of mis-accouning. | |
1941 | * But such kind of mis-account by race always happens because | |
1942 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
1943 | * small race, here. | |
1944 | * (*) swapoff at el will charge against mm-struct not against | |
1945 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 KH |
1946 | */ |
1947 | mem = mem_cgroup_from_task(p); | |
ebb76ce1 | 1948 | if (!mem || mem_cgroup_is_root(mem)) { |
f75ca962 KH |
1949 | rcu_read_unlock(); |
1950 | goto done; | |
1951 | } | |
ec168510 | 1952 | if (page_size == PAGE_SIZE && consume_stock(mem)) { |
f75ca962 KH |
1953 | /* |
1954 | * It seems dagerous to access memcg without css_get(). | |
1955 | * But considering how consume_stok works, it's not | |
1956 | * necessary. If consume_stock success, some charges | |
1957 | * from this memcg are cached on this cpu. So, we | |
1958 | * don't need to call css_get()/css_tryget() before | |
1959 | * calling consume_stock(). | |
1960 | */ | |
1961 | rcu_read_unlock(); | |
1962 | goto done; | |
1963 | } | |
1964 | /* after here, we may be blocked. we need to get refcnt */ | |
1965 | if (!css_tryget(&mem->css)) { | |
1966 | rcu_read_unlock(); | |
1967 | goto again; | |
1968 | } | |
1969 | rcu_read_unlock(); | |
1970 | } | |
8a9f3ccd | 1971 | |
4b534334 KH |
1972 | do { |
1973 | bool oom_check; | |
7a81b88c | 1974 | |
4b534334 | 1975 | /* If killed, bypass charge */ |
f75ca962 KH |
1976 | if (fatal_signal_pending(current)) { |
1977 | css_put(&mem->css); | |
4b534334 | 1978 | goto bypass; |
f75ca962 | 1979 | } |
6d61ef40 | 1980 | |
4b534334 KH |
1981 | oom_check = false; |
1982 | if (oom && !nr_oom_retries) { | |
1983 | oom_check = true; | |
1984 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 1985 | } |
66e1707b | 1986 | |
4b534334 | 1987 | ret = __mem_cgroup_do_charge(mem, gfp_mask, csize, oom_check); |
8033b97c | 1988 | |
4b534334 KH |
1989 | switch (ret) { |
1990 | case CHARGE_OK: | |
1991 | break; | |
1992 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
ec168510 | 1993 | csize = page_size; |
f75ca962 KH |
1994 | css_put(&mem->css); |
1995 | mem = NULL; | |
1996 | goto again; | |
4b534334 | 1997 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
f75ca962 | 1998 | css_put(&mem->css); |
4b534334 KH |
1999 | goto nomem; |
2000 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 KH |
2001 | if (!oom) { |
2002 | css_put(&mem->css); | |
867578cb | 2003 | goto nomem; |
f75ca962 | 2004 | } |
4b534334 KH |
2005 | /* If oom, we never return -ENOMEM */ |
2006 | nr_oom_retries--; | |
2007 | break; | |
2008 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
f75ca962 | 2009 | css_put(&mem->css); |
867578cb | 2010 | goto bypass; |
66e1707b | 2011 | } |
4b534334 KH |
2012 | } while (ret != CHARGE_OK); |
2013 | ||
ec168510 AA |
2014 | if (csize > page_size) |
2015 | refill_stock(mem, csize - page_size); | |
f75ca962 | 2016 | css_put(&mem->css); |
0c3e73e8 | 2017 | done: |
f75ca962 | 2018 | *memcg = mem; |
7a81b88c KH |
2019 | return 0; |
2020 | nomem: | |
f75ca962 | 2021 | *memcg = NULL; |
7a81b88c | 2022 | return -ENOMEM; |
867578cb KH |
2023 | bypass: |
2024 | *memcg = NULL; | |
2025 | return 0; | |
7a81b88c | 2026 | } |
8a9f3ccd | 2027 | |
a3032a2c DN |
2028 | /* |
2029 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2030 | * This function is for that and do uncharge, put css's refcnt. | |
2031 | * gotten by try_charge(). | |
2032 | */ | |
854ffa8d DN |
2033 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
2034 | unsigned long count) | |
a3032a2c DN |
2035 | { |
2036 | if (!mem_cgroup_is_root(mem)) { | |
854ffa8d | 2037 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
a3032a2c | 2038 | if (do_swap_account) |
854ffa8d | 2039 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
a3032a2c | 2040 | } |
854ffa8d DN |
2041 | } |
2042 | ||
ec168510 AA |
2043 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
2044 | int page_size) | |
854ffa8d | 2045 | { |
ec168510 | 2046 | __mem_cgroup_cancel_charge(mem, page_size >> PAGE_SHIFT); |
a3032a2c DN |
2047 | } |
2048 | ||
a3b2d692 KH |
2049 | /* |
2050 | * A helper function to get mem_cgroup from ID. must be called under | |
2051 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2052 | * it's concern. (dropping refcnt from swap can be called against removed | |
2053 | * memcg.) | |
2054 | */ | |
2055 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2056 | { | |
2057 | struct cgroup_subsys_state *css; | |
2058 | ||
2059 | /* ID 0 is unused ID */ | |
2060 | if (!id) | |
2061 | return NULL; | |
2062 | css = css_lookup(&mem_cgroup_subsys, id); | |
2063 | if (!css) | |
2064 | return NULL; | |
2065 | return container_of(css, struct mem_cgroup, css); | |
2066 | } | |
2067 | ||
e42d9d5d | 2068 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2069 | { |
e42d9d5d | 2070 | struct mem_cgroup *mem = NULL; |
3c776e64 | 2071 | struct page_cgroup *pc; |
a3b2d692 | 2072 | unsigned short id; |
b5a84319 KH |
2073 | swp_entry_t ent; |
2074 | ||
3c776e64 DN |
2075 | VM_BUG_ON(!PageLocked(page)); |
2076 | ||
3c776e64 | 2077 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2078 | lock_page_cgroup(pc); |
a3b2d692 | 2079 | if (PageCgroupUsed(pc)) { |
3c776e64 | 2080 | mem = pc->mem_cgroup; |
a3b2d692 KH |
2081 | if (mem && !css_tryget(&mem->css)) |
2082 | mem = NULL; | |
e42d9d5d | 2083 | } else if (PageSwapCache(page)) { |
3c776e64 | 2084 | ent.val = page_private(page); |
a3b2d692 KH |
2085 | id = lookup_swap_cgroup(ent); |
2086 | rcu_read_lock(); | |
2087 | mem = mem_cgroup_lookup(id); | |
2088 | if (mem && !css_tryget(&mem->css)) | |
2089 | mem = NULL; | |
2090 | rcu_read_unlock(); | |
3c776e64 | 2091 | } |
c0bd3f63 | 2092 | unlock_page_cgroup(pc); |
b5a84319 KH |
2093 | return mem; |
2094 | } | |
2095 | ||
ca3e0214 | 2096 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, |
5564e88b | 2097 | struct page *page, |
ca3e0214 KH |
2098 | struct page_cgroup *pc, |
2099 | enum charge_type ctype, | |
2100 | int page_size) | |
7a81b88c | 2101 | { |
ca3e0214 KH |
2102 | int nr_pages = page_size >> PAGE_SHIFT; |
2103 | ||
ca3e0214 KH |
2104 | lock_page_cgroup(pc); |
2105 | if (unlikely(PageCgroupUsed(pc))) { | |
2106 | unlock_page_cgroup(pc); | |
2107 | mem_cgroup_cancel_charge(mem, page_size); | |
2108 | return; | |
2109 | } | |
2110 | /* | |
2111 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2112 | * accessed by any other context at this point. | |
2113 | */ | |
8a9f3ccd | 2114 | pc->mem_cgroup = mem; |
261fb61a KH |
2115 | /* |
2116 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2117 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2118 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2119 | * before USED bit, we need memory barrier here. | |
2120 | * See mem_cgroup_add_lru_list(), etc. | |
2121 | */ | |
08e552c6 | 2122 | smp_wmb(); |
4b3bde4c BS |
2123 | switch (ctype) { |
2124 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
2125 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
2126 | SetPageCgroupCache(pc); | |
2127 | SetPageCgroupUsed(pc); | |
2128 | break; | |
2129 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2130 | ClearPageCgroupCache(pc); | |
2131 | SetPageCgroupUsed(pc); | |
2132 | break; | |
2133 | default: | |
2134 | break; | |
2135 | } | |
3be91277 | 2136 | |
ca3e0214 | 2137 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages); |
52d4b9ac | 2138 | unlock_page_cgroup(pc); |
430e4863 KH |
2139 | /* |
2140 | * "charge_statistics" updated event counter. Then, check it. | |
2141 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2142 | * if they exceeds softlimit. | |
2143 | */ | |
5564e88b | 2144 | memcg_check_events(mem, page); |
7a81b88c | 2145 | } |
66e1707b | 2146 | |
ca3e0214 KH |
2147 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2148 | ||
2149 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
2150 | (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION)) | |
2151 | /* | |
2152 | * Because tail pages are not marked as "used", set it. We're under | |
2153 | * zone->lru_lock, 'splitting on pmd' and compund_lock. | |
2154 | */ | |
2155 | void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail) | |
2156 | { | |
2157 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
2158 | struct page_cgroup *tail_pc = lookup_page_cgroup(tail); | |
2159 | unsigned long flags; | |
2160 | ||
3d37c4a9 KH |
2161 | if (mem_cgroup_disabled()) |
2162 | return; | |
ca3e0214 | 2163 | /* |
ece35ca8 | 2164 | * We have no races with charge/uncharge but will have races with |
ca3e0214 KH |
2165 | * page state accounting. |
2166 | */ | |
2167 | move_lock_page_cgroup(head_pc, &flags); | |
2168 | ||
2169 | tail_pc->mem_cgroup = head_pc->mem_cgroup; | |
2170 | smp_wmb(); /* see __commit_charge() */ | |
ece35ca8 KH |
2171 | if (PageCgroupAcctLRU(head_pc)) { |
2172 | enum lru_list lru; | |
2173 | struct mem_cgroup_per_zone *mz; | |
2174 | ||
2175 | /* | |
2176 | * LRU flags cannot be copied because we need to add tail | |
2177 | *.page to LRU by generic call and our hook will be called. | |
2178 | * We hold lru_lock, then, reduce counter directly. | |
2179 | */ | |
2180 | lru = page_lru(head); | |
97a6c37b | 2181 | mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head); |
ece35ca8 KH |
2182 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
2183 | } | |
ca3e0214 KH |
2184 | tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2185 | move_unlock_page_cgroup(head_pc, &flags); | |
2186 | } | |
2187 | #endif | |
2188 | ||
f817ed48 | 2189 | /** |
de3638d9 | 2190 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2191 | * @page: the page |
f817ed48 KH |
2192 | * @pc: page_cgroup of the page. |
2193 | * @from: mem_cgroup which the page is moved from. | |
2194 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2195 | * @uncharge: whether we should call uncharge and css_put against @from. |
de3638d9 | 2196 | * @charge_size: number of bytes to charge (regular or huge page) |
f817ed48 KH |
2197 | * |
2198 | * The caller must confirm following. | |
08e552c6 | 2199 | * - page is not on LRU (isolate_page() is useful.) |
de3638d9 | 2200 | * - compound_lock is held when charge_size > PAGE_SIZE |
f817ed48 | 2201 | * |
854ffa8d DN |
2202 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
2203 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is | |
2204 | * true, this function does "uncharge" from old cgroup, but it doesn't if | |
2205 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2206 | */ |
5564e88b | 2207 | static int mem_cgroup_move_account(struct page *page, struct page_cgroup *pc, |
de3638d9 JW |
2208 | struct mem_cgroup *from, struct mem_cgroup *to, |
2209 | bool uncharge, int charge_size) | |
f817ed48 | 2210 | { |
987eba66 | 2211 | int nr_pages = charge_size >> PAGE_SHIFT; |
de3638d9 JW |
2212 | unsigned long flags; |
2213 | int ret; | |
987eba66 | 2214 | |
f817ed48 | 2215 | VM_BUG_ON(from == to); |
5564e88b | 2216 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2217 | /* |
2218 | * The page is isolated from LRU. So, collapse function | |
2219 | * will not handle this page. But page splitting can happen. | |
2220 | * Do this check under compound_page_lock(). The caller should | |
2221 | * hold it. | |
2222 | */ | |
2223 | ret = -EBUSY; | |
5564e88b | 2224 | if (charge_size > PAGE_SIZE && !PageTransHuge(page)) |
de3638d9 JW |
2225 | goto out; |
2226 | ||
2227 | lock_page_cgroup(pc); | |
2228 | ||
2229 | ret = -EINVAL; | |
2230 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2231 | goto unlock; | |
2232 | ||
2233 | move_lock_page_cgroup(pc, &flags); | |
f817ed48 | 2234 | |
8725d541 | 2235 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2236 | /* Update mapped_file data for mem_cgroup */ |
2237 | preempt_disable(); | |
2238 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2239 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2240 | preempt_enable(); | |
d69b042f | 2241 | } |
987eba66 | 2242 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages); |
854ffa8d DN |
2243 | if (uncharge) |
2244 | /* This is not "cancel", but cancel_charge does all we need. */ | |
987eba66 | 2245 | mem_cgroup_cancel_charge(from, charge_size); |
d69b042f | 2246 | |
854ffa8d | 2247 | /* caller should have done css_get */ |
08e552c6 | 2248 | pc->mem_cgroup = to; |
987eba66 | 2249 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages); |
88703267 KH |
2250 | /* |
2251 | * We charges against "to" which may not have any tasks. Then, "to" | |
2252 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe DN |
2253 | * this function is just force_empty() and move charge, so it's |
2254 | * garanteed that "to" is never removed. So, we don't check rmdir | |
2255 | * status here. | |
88703267 | 2256 | */ |
de3638d9 JW |
2257 | move_unlock_page_cgroup(pc, &flags); |
2258 | ret = 0; | |
2259 | unlock: | |
57f9fd7d | 2260 | unlock_page_cgroup(pc); |
d2265e6f KH |
2261 | /* |
2262 | * check events | |
2263 | */ | |
5564e88b JW |
2264 | memcg_check_events(to, page); |
2265 | memcg_check_events(from, page); | |
de3638d9 | 2266 | out: |
f817ed48 KH |
2267 | return ret; |
2268 | } | |
2269 | ||
2270 | /* | |
2271 | * move charges to its parent. | |
2272 | */ | |
2273 | ||
5564e88b JW |
2274 | static int mem_cgroup_move_parent(struct page *page, |
2275 | struct page_cgroup *pc, | |
f817ed48 KH |
2276 | struct mem_cgroup *child, |
2277 | gfp_t gfp_mask) | |
2278 | { | |
2279 | struct cgroup *cg = child->css.cgroup; | |
2280 | struct cgroup *pcg = cg->parent; | |
2281 | struct mem_cgroup *parent; | |
52dbb905 | 2282 | int page_size = PAGE_SIZE; |
987eba66 | 2283 | unsigned long flags; |
f817ed48 KH |
2284 | int ret; |
2285 | ||
2286 | /* Is ROOT ? */ | |
2287 | if (!pcg) | |
2288 | return -EINVAL; | |
2289 | ||
57f9fd7d DN |
2290 | ret = -EBUSY; |
2291 | if (!get_page_unless_zero(page)) | |
2292 | goto out; | |
2293 | if (isolate_lru_page(page)) | |
2294 | goto put; | |
52dbb905 KH |
2295 | |
2296 | if (PageTransHuge(page)) | |
2297 | page_size = HPAGE_SIZE; | |
08e552c6 | 2298 | |
f817ed48 | 2299 | parent = mem_cgroup_from_cont(pcg); |
52dbb905 KH |
2300 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, |
2301 | &parent, false, page_size); | |
a636b327 | 2302 | if (ret || !parent) |
57f9fd7d | 2303 | goto put_back; |
f817ed48 | 2304 | |
52dbb905 | 2305 | if (page_size > PAGE_SIZE) |
987eba66 KH |
2306 | flags = compound_lock_irqsave(page); |
2307 | ||
5564e88b | 2308 | ret = mem_cgroup_move_account(page, pc, child, parent, true, page_size); |
854ffa8d | 2309 | if (ret) |
52dbb905 | 2310 | mem_cgroup_cancel_charge(parent, page_size); |
8dba474f | 2311 | |
52dbb905 | 2312 | if (page_size > PAGE_SIZE) |
987eba66 | 2313 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2314 | put_back: |
08e552c6 | 2315 | putback_lru_page(page); |
57f9fd7d | 2316 | put: |
40d58138 | 2317 | put_page(page); |
57f9fd7d | 2318 | out: |
f817ed48 KH |
2319 | return ret; |
2320 | } | |
2321 | ||
7a81b88c KH |
2322 | /* |
2323 | * Charge the memory controller for page usage. | |
2324 | * Return | |
2325 | * 0 if the charge was successful | |
2326 | * < 0 if the cgroup is over its limit | |
2327 | */ | |
2328 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2329 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2330 | { |
73045c47 | 2331 | struct mem_cgroup *mem = NULL; |
8493ae43 | 2332 | int page_size = PAGE_SIZE; |
7a81b88c | 2333 | struct page_cgroup *pc; |
8493ae43 | 2334 | bool oom = true; |
7a81b88c | 2335 | int ret; |
ec168510 | 2336 | |
37c2ac78 | 2337 | if (PageTransHuge(page)) { |
ec168510 | 2338 | page_size <<= compound_order(page); |
37c2ac78 | 2339 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2340 | /* |
2341 | * Never OOM-kill a process for a huge page. The | |
2342 | * fault handler will fall back to regular pages. | |
2343 | */ | |
2344 | oom = false; | |
37c2ac78 | 2345 | } |
7a81b88c KH |
2346 | |
2347 | pc = lookup_page_cgroup(page); | |
af4a6621 | 2348 | BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */ |
7a81b88c | 2349 | |
8493ae43 | 2350 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, oom, page_size); |
a636b327 | 2351 | if (ret || !mem) |
7a81b88c KH |
2352 | return ret; |
2353 | ||
5564e88b | 2354 | __mem_cgroup_commit_charge(mem, page, pc, ctype, page_size); |
8a9f3ccd | 2355 | return 0; |
8a9f3ccd BS |
2356 | } |
2357 | ||
7a81b88c KH |
2358 | int mem_cgroup_newpage_charge(struct page *page, |
2359 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2360 | { |
f8d66542 | 2361 | if (mem_cgroup_disabled()) |
cede86ac | 2362 | return 0; |
69029cd5 KH |
2363 | /* |
2364 | * If already mapped, we don't have to account. | |
2365 | * If page cache, page->mapping has address_space. | |
2366 | * But page->mapping may have out-of-use anon_vma pointer, | |
2367 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
2368 | * is NULL. | |
2369 | */ | |
2370 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2371 | return 0; | |
2372 | if (unlikely(!mm)) | |
2373 | mm = &init_mm; | |
217bc319 | 2374 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
73045c47 | 2375 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2376 | } |
2377 | ||
83aae4c7 DN |
2378 | static void |
2379 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2380 | enum charge_type ctype); | |
2381 | ||
e1a1cd59 BS |
2382 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2383 | gfp_t gfp_mask) | |
8697d331 | 2384 | { |
b5a84319 KH |
2385 | int ret; |
2386 | ||
f8d66542 | 2387 | if (mem_cgroup_disabled()) |
cede86ac | 2388 | return 0; |
52d4b9ac KH |
2389 | if (PageCompound(page)) |
2390 | return 0; | |
accf163e KH |
2391 | /* |
2392 | * Corner case handling. This is called from add_to_page_cache() | |
2393 | * in usual. But some FS (shmem) precharges this page before calling it | |
2394 | * and call add_to_page_cache() with GFP_NOWAIT. | |
2395 | * | |
2396 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
2397 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
2398 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
2399 | * And when the page is SwapCache, it should take swap information |
2400 | * into account. This is under lock_page() now. | |
accf163e KH |
2401 | */ |
2402 | if (!(gfp_mask & __GFP_WAIT)) { | |
2403 | struct page_cgroup *pc; | |
2404 | ||
52d4b9ac KH |
2405 | pc = lookup_page_cgroup(page); |
2406 | if (!pc) | |
2407 | return 0; | |
2408 | lock_page_cgroup(pc); | |
2409 | if (PageCgroupUsed(pc)) { | |
2410 | unlock_page_cgroup(pc); | |
accf163e KH |
2411 | return 0; |
2412 | } | |
52d4b9ac | 2413 | unlock_page_cgroup(pc); |
accf163e KH |
2414 | } |
2415 | ||
73045c47 | 2416 | if (unlikely(!mm)) |
8697d331 | 2417 | mm = &init_mm; |
accf163e | 2418 | |
c05555b5 KH |
2419 | if (page_is_file_cache(page)) |
2420 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2421 | MEM_CGROUP_CHARGE_TYPE_CACHE); |
b5a84319 | 2422 | |
83aae4c7 DN |
2423 | /* shmem */ |
2424 | if (PageSwapCache(page)) { | |
56039efa | 2425 | struct mem_cgroup *mem; |
73045c47 | 2426 | |
83aae4c7 DN |
2427 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2428 | if (!ret) | |
2429 | __mem_cgroup_commit_charge_swapin(page, mem, | |
2430 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
2431 | } else | |
2432 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2433 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
b5a84319 | 2434 | |
b5a84319 | 2435 | return ret; |
e8589cc1 KH |
2436 | } |
2437 | ||
54595fe2 KH |
2438 | /* |
2439 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2440 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2441 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2442 | * "commit()" or removed by "cancel()" |
2443 | */ | |
8c7c6e34 KH |
2444 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2445 | struct page *page, | |
2446 | gfp_t mask, struct mem_cgroup **ptr) | |
2447 | { | |
2448 | struct mem_cgroup *mem; | |
54595fe2 | 2449 | int ret; |
8c7c6e34 | 2450 | |
56039efa KH |
2451 | *ptr = NULL; |
2452 | ||
f8d66542 | 2453 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2454 | return 0; |
2455 | ||
2456 | if (!do_swap_account) | |
2457 | goto charge_cur_mm; | |
8c7c6e34 KH |
2458 | /* |
2459 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2460 | * the pte, and even removed page from swap cache: in those cases |
2461 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2462 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2463 | */ |
2464 | if (!PageSwapCache(page)) | |
407f9c8b | 2465 | goto charge_cur_mm; |
e42d9d5d | 2466 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
2467 | if (!mem) |
2468 | goto charge_cur_mm; | |
8c7c6e34 | 2469 | *ptr = mem; |
ec168510 | 2470 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, PAGE_SIZE); |
54595fe2 KH |
2471 | css_put(&mem->css); |
2472 | return ret; | |
8c7c6e34 KH |
2473 | charge_cur_mm: |
2474 | if (unlikely(!mm)) | |
2475 | mm = &init_mm; | |
ec168510 | 2476 | return __mem_cgroup_try_charge(mm, mask, ptr, true, PAGE_SIZE); |
8c7c6e34 KH |
2477 | } |
2478 | ||
83aae4c7 DN |
2479 | static void |
2480 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2481 | enum charge_type ctype) | |
7a81b88c KH |
2482 | { |
2483 | struct page_cgroup *pc; | |
2484 | ||
f8d66542 | 2485 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2486 | return; |
2487 | if (!ptr) | |
2488 | return; | |
88703267 | 2489 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 2490 | pc = lookup_page_cgroup(page); |
544122e5 | 2491 | mem_cgroup_lru_del_before_commit_swapcache(page); |
5564e88b | 2492 | __mem_cgroup_commit_charge(ptr, page, pc, ctype, PAGE_SIZE); |
544122e5 | 2493 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
2494 | /* |
2495 | * Now swap is on-memory. This means this page may be | |
2496 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2497 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2498 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2499 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2500 | */ |
03f3c433 | 2501 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2502 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2503 | unsigned short id; |
8c7c6e34 | 2504 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2505 | |
2506 | id = swap_cgroup_record(ent, 0); | |
2507 | rcu_read_lock(); | |
2508 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2509 | if (memcg) { |
a3b2d692 KH |
2510 | /* |
2511 | * This recorded memcg can be obsolete one. So, avoid | |
2512 | * calling css_tryget | |
2513 | */ | |
0c3e73e8 | 2514 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2515 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2516 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2517 | mem_cgroup_put(memcg); |
2518 | } | |
a3b2d692 | 2519 | rcu_read_unlock(); |
8c7c6e34 | 2520 | } |
88703267 KH |
2521 | /* |
2522 | * At swapin, we may charge account against cgroup which has no tasks. | |
2523 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2524 | * In that case, we need to call pre_destroy() again. check it here. | |
2525 | */ | |
2526 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2527 | } |
2528 | ||
83aae4c7 DN |
2529 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2530 | { | |
2531 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2532 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2533 | } | |
2534 | ||
7a81b88c KH |
2535 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2536 | { | |
f8d66542 | 2537 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2538 | return; |
2539 | if (!mem) | |
2540 | return; | |
ec168510 | 2541 | mem_cgroup_cancel_charge(mem, PAGE_SIZE); |
7a81b88c KH |
2542 | } |
2543 | ||
569b846d | 2544 | static void |
ec168510 AA |
2545 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype, |
2546 | int page_size) | |
569b846d KH |
2547 | { |
2548 | struct memcg_batch_info *batch = NULL; | |
2549 | bool uncharge_memsw = true; | |
2550 | /* If swapout, usage of swap doesn't decrease */ | |
2551 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2552 | uncharge_memsw = false; | |
569b846d KH |
2553 | |
2554 | batch = ¤t->memcg_batch; | |
2555 | /* | |
2556 | * In usual, we do css_get() when we remember memcg pointer. | |
2557 | * But in this case, we keep res->usage until end of a series of | |
2558 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2559 | */ | |
2560 | if (!batch->memcg) | |
2561 | batch->memcg = mem; | |
3c11ecf4 KH |
2562 | /* |
2563 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
2564 | * In those cases, all pages freed continously can be expected to be in | |
2565 | * the same cgroup and we have chance to coalesce uncharges. | |
2566 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2567 | * because we want to do uncharge as soon as possible. | |
2568 | */ | |
2569 | ||
2570 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2571 | goto direct_uncharge; | |
2572 | ||
ec168510 AA |
2573 | if (page_size != PAGE_SIZE) |
2574 | goto direct_uncharge; | |
2575 | ||
569b846d KH |
2576 | /* |
2577 | * In typical case, batch->memcg == mem. This means we can | |
2578 | * merge a series of uncharges to an uncharge of res_counter. | |
2579 | * If not, we uncharge res_counter ony by one. | |
2580 | */ | |
2581 | if (batch->memcg != mem) | |
2582 | goto direct_uncharge; | |
2583 | /* remember freed charge and uncharge it later */ | |
2584 | batch->bytes += PAGE_SIZE; | |
2585 | if (uncharge_memsw) | |
2586 | batch->memsw_bytes += PAGE_SIZE; | |
2587 | return; | |
2588 | direct_uncharge: | |
ec168510 | 2589 | res_counter_uncharge(&mem->res, page_size); |
569b846d | 2590 | if (uncharge_memsw) |
ec168510 | 2591 | res_counter_uncharge(&mem->memsw, page_size); |
3c11ecf4 KH |
2592 | if (unlikely(batch->memcg != mem)) |
2593 | memcg_oom_recover(mem); | |
569b846d KH |
2594 | return; |
2595 | } | |
7a81b88c | 2596 | |
8a9f3ccd | 2597 | /* |
69029cd5 | 2598 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2599 | */ |
8c7c6e34 | 2600 | static struct mem_cgroup * |
69029cd5 | 2601 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2602 | { |
152c9ccb | 2603 | int count; |
8289546e | 2604 | struct page_cgroup *pc; |
8c7c6e34 | 2605 | struct mem_cgroup *mem = NULL; |
ec168510 | 2606 | int page_size = PAGE_SIZE; |
8a9f3ccd | 2607 | |
f8d66542 | 2608 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2609 | return NULL; |
4077960e | 2610 | |
d13d1443 | 2611 | if (PageSwapCache(page)) |
8c7c6e34 | 2612 | return NULL; |
d13d1443 | 2613 | |
37c2ac78 | 2614 | if (PageTransHuge(page)) { |
ec168510 | 2615 | page_size <<= compound_order(page); |
37c2ac78 AA |
2616 | VM_BUG_ON(!PageTransHuge(page)); |
2617 | } | |
ec168510 | 2618 | |
152c9ccb | 2619 | count = page_size >> PAGE_SHIFT; |
8697d331 | 2620 | /* |
3c541e14 | 2621 | * Check if our page_cgroup is valid |
8697d331 | 2622 | */ |
52d4b9ac KH |
2623 | pc = lookup_page_cgroup(page); |
2624 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2625 | return NULL; |
b9c565d5 | 2626 | |
52d4b9ac | 2627 | lock_page_cgroup(pc); |
d13d1443 | 2628 | |
8c7c6e34 KH |
2629 | mem = pc->mem_cgroup; |
2630 | ||
d13d1443 KH |
2631 | if (!PageCgroupUsed(pc)) |
2632 | goto unlock_out; | |
2633 | ||
2634 | switch (ctype) { | |
2635 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2636 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2637 | /* See mem_cgroup_prepare_migration() */ |
2638 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2639 | goto unlock_out; |
2640 | break; | |
2641 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2642 | if (!PageAnon(page)) { /* Shared memory */ | |
2643 | if (page->mapping && !page_is_file_cache(page)) | |
2644 | goto unlock_out; | |
2645 | } else if (page_mapped(page)) /* Anon */ | |
2646 | goto unlock_out; | |
2647 | break; | |
2648 | default: | |
2649 | break; | |
52d4b9ac | 2650 | } |
d13d1443 | 2651 | |
ca3e0214 | 2652 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -count); |
04046e1a | 2653 | |
52d4b9ac | 2654 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2655 | /* |
2656 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2657 | * freed from LRU. This is safe because uncharged page is expected not | |
2658 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2659 | * special functions. | |
2660 | */ | |
b9c565d5 | 2661 | |
52d4b9ac | 2662 | unlock_page_cgroup(pc); |
f75ca962 KH |
2663 | /* |
2664 | * even after unlock, we have mem->res.usage here and this memcg | |
2665 | * will never be freed. | |
2666 | */ | |
d2265e6f | 2667 | memcg_check_events(mem, page); |
f75ca962 KH |
2668 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
2669 | mem_cgroup_swap_statistics(mem, true); | |
2670 | mem_cgroup_get(mem); | |
2671 | } | |
2672 | if (!mem_cgroup_is_root(mem)) | |
ec168510 | 2673 | __do_uncharge(mem, ctype, page_size); |
6d12e2d8 | 2674 | |
8c7c6e34 | 2675 | return mem; |
d13d1443 KH |
2676 | |
2677 | unlock_out: | |
2678 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2679 | return NULL; |
3c541e14 BS |
2680 | } |
2681 | ||
69029cd5 KH |
2682 | void mem_cgroup_uncharge_page(struct page *page) |
2683 | { | |
52d4b9ac KH |
2684 | /* early check. */ |
2685 | if (page_mapped(page)) | |
2686 | return; | |
2687 | if (page->mapping && !PageAnon(page)) | |
2688 | return; | |
69029cd5 KH |
2689 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2690 | } | |
2691 | ||
2692 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2693 | { | |
2694 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2695 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2696 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2697 | } | |
2698 | ||
569b846d KH |
2699 | /* |
2700 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2701 | * In that cases, pages are freed continuously and we can expect pages | |
2702 | * are in the same memcg. All these calls itself limits the number of | |
2703 | * pages freed at once, then uncharge_start/end() is called properly. | |
2704 | * This may be called prural(2) times in a context, | |
2705 | */ | |
2706 | ||
2707 | void mem_cgroup_uncharge_start(void) | |
2708 | { | |
2709 | current->memcg_batch.do_batch++; | |
2710 | /* We can do nest. */ | |
2711 | if (current->memcg_batch.do_batch == 1) { | |
2712 | current->memcg_batch.memcg = NULL; | |
2713 | current->memcg_batch.bytes = 0; | |
2714 | current->memcg_batch.memsw_bytes = 0; | |
2715 | } | |
2716 | } | |
2717 | ||
2718 | void mem_cgroup_uncharge_end(void) | |
2719 | { | |
2720 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2721 | ||
2722 | if (!batch->do_batch) | |
2723 | return; | |
2724 | ||
2725 | batch->do_batch--; | |
2726 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2727 | return; | |
2728 | ||
2729 | if (!batch->memcg) | |
2730 | return; | |
2731 | /* | |
2732 | * This "batch->memcg" is valid without any css_get/put etc... | |
2733 | * bacause we hide charges behind us. | |
2734 | */ | |
2735 | if (batch->bytes) | |
2736 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2737 | if (batch->memsw_bytes) | |
2738 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
3c11ecf4 | 2739 | memcg_oom_recover(batch->memcg); |
569b846d KH |
2740 | /* forget this pointer (for sanity check) */ |
2741 | batch->memcg = NULL; | |
2742 | } | |
2743 | ||
e767e056 | 2744 | #ifdef CONFIG_SWAP |
8c7c6e34 | 2745 | /* |
e767e056 | 2746 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
2747 | * memcg information is recorded to swap_cgroup of "ent" |
2748 | */ | |
8a9478ca KH |
2749 | void |
2750 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
2751 | { |
2752 | struct mem_cgroup *memcg; | |
8a9478ca KH |
2753 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
2754 | ||
2755 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2756 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2757 | ||
2758 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 2759 | |
f75ca962 KH |
2760 | /* |
2761 | * record memcg information, if swapout && memcg != NULL, | |
2762 | * mem_cgroup_get() was called in uncharge(). | |
2763 | */ | |
2764 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 2765 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 2766 | } |
e767e056 | 2767 | #endif |
8c7c6e34 KH |
2768 | |
2769 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2770 | /* | |
2771 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2772 | * uncharge "memsw" account. | |
2773 | */ | |
2774 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 2775 | { |
8c7c6e34 | 2776 | struct mem_cgroup *memcg; |
a3b2d692 | 2777 | unsigned short id; |
8c7c6e34 KH |
2778 | |
2779 | if (!do_swap_account) | |
2780 | return; | |
2781 | ||
a3b2d692 KH |
2782 | id = swap_cgroup_record(ent, 0); |
2783 | rcu_read_lock(); | |
2784 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2785 | if (memcg) { |
a3b2d692 KH |
2786 | /* |
2787 | * We uncharge this because swap is freed. | |
2788 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2789 | */ | |
0c3e73e8 | 2790 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2791 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2792 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2793 | mem_cgroup_put(memcg); |
2794 | } | |
a3b2d692 | 2795 | rcu_read_unlock(); |
d13d1443 | 2796 | } |
02491447 DN |
2797 | |
2798 | /** | |
2799 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2800 | * @entry: swap entry to be moved | |
2801 | * @from: mem_cgroup which the entry is moved from | |
2802 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 2803 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
2804 | * |
2805 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2806 | * as the mem_cgroup's id of @from. | |
2807 | * | |
2808 | * Returns 0 on success, -EINVAL on failure. | |
2809 | * | |
2810 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
2811 | * both res and memsw, and called css_get(). | |
2812 | */ | |
2813 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2814 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2815 | { |
2816 | unsigned short old_id, new_id; | |
2817 | ||
2818 | old_id = css_id(&from->css); | |
2819 | new_id = css_id(&to->css); | |
2820 | ||
2821 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2822 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2823 | mem_cgroup_swap_statistics(to, true); |
02491447 | 2824 | /* |
483c30b5 DN |
2825 | * This function is only called from task migration context now. |
2826 | * It postpones res_counter and refcount handling till the end | |
2827 | * of task migration(mem_cgroup_clear_mc()) for performance | |
2828 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
2829 | * because if the process that has been moved to @to does | |
2830 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 2831 | */ |
02491447 | 2832 | mem_cgroup_get(to); |
483c30b5 DN |
2833 | if (need_fixup) { |
2834 | if (!mem_cgroup_is_root(from)) | |
2835 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
2836 | mem_cgroup_put(from); | |
2837 | /* | |
2838 | * we charged both to->res and to->memsw, so we should | |
2839 | * uncharge to->res. | |
2840 | */ | |
2841 | if (!mem_cgroup_is_root(to)) | |
2842 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 2843 | } |
02491447 DN |
2844 | return 0; |
2845 | } | |
2846 | return -EINVAL; | |
2847 | } | |
2848 | #else | |
2849 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2850 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2851 | { |
2852 | return -EINVAL; | |
2853 | } | |
8c7c6e34 | 2854 | #endif |
d13d1443 | 2855 | |
ae41be37 | 2856 | /* |
01b1ae63 KH |
2857 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
2858 | * page belongs to. | |
ae41be37 | 2859 | */ |
ac39cf8c | 2860 | int mem_cgroup_prepare_migration(struct page *page, |
ef6a3c63 | 2861 | struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask) |
ae41be37 KH |
2862 | { |
2863 | struct page_cgroup *pc; | |
e8589cc1 | 2864 | struct mem_cgroup *mem = NULL; |
ac39cf8c | 2865 | enum charge_type ctype; |
e8589cc1 | 2866 | int ret = 0; |
8869b8f6 | 2867 | |
56039efa KH |
2868 | *ptr = NULL; |
2869 | ||
ec168510 | 2870 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 2871 | if (mem_cgroup_disabled()) |
4077960e BS |
2872 | return 0; |
2873 | ||
52d4b9ac KH |
2874 | pc = lookup_page_cgroup(page); |
2875 | lock_page_cgroup(pc); | |
2876 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
2877 | mem = pc->mem_cgroup; |
2878 | css_get(&mem->css); | |
ac39cf8c | 2879 | /* |
2880 | * At migrating an anonymous page, its mapcount goes down | |
2881 | * to 0 and uncharge() will be called. But, even if it's fully | |
2882 | * unmapped, migration may fail and this page has to be | |
2883 | * charged again. We set MIGRATION flag here and delay uncharge | |
2884 | * until end_migration() is called | |
2885 | * | |
2886 | * Corner Case Thinking | |
2887 | * A) | |
2888 | * When the old page was mapped as Anon and it's unmap-and-freed | |
2889 | * while migration was ongoing. | |
2890 | * If unmap finds the old page, uncharge() of it will be delayed | |
2891 | * until end_migration(). If unmap finds a new page, it's | |
2892 | * uncharged when it make mapcount to be 1->0. If unmap code | |
2893 | * finds swap_migration_entry, the new page will not be mapped | |
2894 | * and end_migration() will find it(mapcount==0). | |
2895 | * | |
2896 | * B) | |
2897 | * When the old page was mapped but migraion fails, the kernel | |
2898 | * remaps it. A charge for it is kept by MIGRATION flag even | |
2899 | * if mapcount goes down to 0. We can do remap successfully | |
2900 | * without charging it again. | |
2901 | * | |
2902 | * C) | |
2903 | * The "old" page is under lock_page() until the end of | |
2904 | * migration, so, the old page itself will not be swapped-out. | |
2905 | * If the new page is swapped out before end_migraton, our | |
2906 | * hook to usual swap-out path will catch the event. | |
2907 | */ | |
2908 | if (PageAnon(page)) | |
2909 | SetPageCgroupMigration(pc); | |
e8589cc1 | 2910 | } |
52d4b9ac | 2911 | unlock_page_cgroup(pc); |
ac39cf8c | 2912 | /* |
2913 | * If the page is not charged at this point, | |
2914 | * we return here. | |
2915 | */ | |
2916 | if (!mem) | |
2917 | return 0; | |
01b1ae63 | 2918 | |
93d5c9be | 2919 | *ptr = mem; |
ef6a3c63 | 2920 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, ptr, false, PAGE_SIZE); |
ac39cf8c | 2921 | css_put(&mem->css);/* drop extra refcnt */ |
2922 | if (ret || *ptr == NULL) { | |
2923 | if (PageAnon(page)) { | |
2924 | lock_page_cgroup(pc); | |
2925 | ClearPageCgroupMigration(pc); | |
2926 | unlock_page_cgroup(pc); | |
2927 | /* | |
2928 | * The old page may be fully unmapped while we kept it. | |
2929 | */ | |
2930 | mem_cgroup_uncharge_page(page); | |
2931 | } | |
2932 | return -ENOMEM; | |
e8589cc1 | 2933 | } |
ac39cf8c | 2934 | /* |
2935 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
2936 | * is called before end_migration, we can catch all events on this new | |
2937 | * page. In the case new page is migrated but not remapped, new page's | |
2938 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
2939 | */ | |
2940 | pc = lookup_page_cgroup(newpage); | |
2941 | if (PageAnon(page)) | |
2942 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2943 | else if (page_is_file_cache(page)) | |
2944 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2945 | else | |
2946 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
5564e88b | 2947 | __mem_cgroup_commit_charge(mem, page, pc, ctype, PAGE_SIZE); |
e8589cc1 | 2948 | return ret; |
ae41be37 | 2949 | } |
8869b8f6 | 2950 | |
69029cd5 | 2951 | /* remove redundant charge if migration failed*/ |
01b1ae63 | 2952 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
50de1dd9 | 2953 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 2954 | { |
ac39cf8c | 2955 | struct page *used, *unused; |
01b1ae63 | 2956 | struct page_cgroup *pc; |
01b1ae63 KH |
2957 | |
2958 | if (!mem) | |
2959 | return; | |
ac39cf8c | 2960 | /* blocks rmdir() */ |
88703267 | 2961 | cgroup_exclude_rmdir(&mem->css); |
50de1dd9 | 2962 | if (!migration_ok) { |
ac39cf8c | 2963 | used = oldpage; |
2964 | unused = newpage; | |
01b1ae63 | 2965 | } else { |
ac39cf8c | 2966 | used = newpage; |
01b1ae63 KH |
2967 | unused = oldpage; |
2968 | } | |
69029cd5 | 2969 | /* |
ac39cf8c | 2970 | * We disallowed uncharge of pages under migration because mapcount |
2971 | * of the page goes down to zero, temporarly. | |
2972 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 2973 | */ |
ac39cf8c | 2974 | pc = lookup_page_cgroup(oldpage); |
2975 | lock_page_cgroup(pc); | |
2976 | ClearPageCgroupMigration(pc); | |
2977 | unlock_page_cgroup(pc); | |
01b1ae63 | 2978 | |
ac39cf8c | 2979 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
2980 | ||
01b1ae63 | 2981 | /* |
ac39cf8c | 2982 | * If a page is a file cache, radix-tree replacement is very atomic |
2983 | * and we can skip this check. When it was an Anon page, its mapcount | |
2984 | * goes down to 0. But because we added MIGRATION flage, it's not | |
2985 | * uncharged yet. There are several case but page->mapcount check | |
2986 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
2987 | * check. (see prepare_charge() also) | |
69029cd5 | 2988 | */ |
ac39cf8c | 2989 | if (PageAnon(used)) |
2990 | mem_cgroup_uncharge_page(used); | |
88703267 | 2991 | /* |
ac39cf8c | 2992 | * At migration, we may charge account against cgroup which has no |
2993 | * tasks. | |
88703267 KH |
2994 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
2995 | * In that case, we need to call pre_destroy() again. check it here. | |
2996 | */ | |
2997 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 2998 | } |
78fb7466 | 2999 | |
c9b0ed51 | 3000 | /* |
ae3abae6 DN |
3001 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
3002 | * Calling hierarchical_reclaim is not enough because we should update | |
3003 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
3004 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
3005 | * not from the memcg which this page would be charged to. | |
3006 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 3007 | */ |
ae3abae6 | 3008 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
3009 | struct mm_struct *mm, |
3010 | gfp_t gfp_mask) | |
c9b0ed51 | 3011 | { |
56039efa | 3012 | struct mem_cgroup *mem; |
ae3abae6 | 3013 | int ret; |
c9b0ed51 | 3014 | |
f8d66542 | 3015 | if (mem_cgroup_disabled()) |
cede86ac | 3016 | return 0; |
c9b0ed51 | 3017 | |
ae3abae6 DN |
3018 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
3019 | if (!ret) | |
3020 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 3021 | |
ae3abae6 | 3022 | return ret; |
c9b0ed51 KH |
3023 | } |
3024 | ||
f212ad7c DN |
3025 | #ifdef CONFIG_DEBUG_VM |
3026 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3027 | { | |
3028 | struct page_cgroup *pc; | |
3029 | ||
3030 | pc = lookup_page_cgroup(page); | |
3031 | if (likely(pc) && PageCgroupUsed(pc)) | |
3032 | return pc; | |
3033 | return NULL; | |
3034 | } | |
3035 | ||
3036 | bool mem_cgroup_bad_page_check(struct page *page) | |
3037 | { | |
3038 | if (mem_cgroup_disabled()) | |
3039 | return false; | |
3040 | ||
3041 | return lookup_page_cgroup_used(page) != NULL; | |
3042 | } | |
3043 | ||
3044 | void mem_cgroup_print_bad_page(struct page *page) | |
3045 | { | |
3046 | struct page_cgroup *pc; | |
3047 | ||
3048 | pc = lookup_page_cgroup_used(page); | |
3049 | if (pc) { | |
3050 | int ret = -1; | |
3051 | char *path; | |
3052 | ||
3053 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p", | |
3054 | pc, pc->flags, pc->mem_cgroup); | |
3055 | ||
3056 | path = kmalloc(PATH_MAX, GFP_KERNEL); | |
3057 | if (path) { | |
3058 | rcu_read_lock(); | |
3059 | ret = cgroup_path(pc->mem_cgroup->css.cgroup, | |
3060 | path, PATH_MAX); | |
3061 | rcu_read_unlock(); | |
3062 | } | |
3063 | ||
3064 | printk(KERN_CONT "(%s)\n", | |
3065 | (ret < 0) ? "cannot get the path" : path); | |
3066 | kfree(path); | |
3067 | } | |
3068 | } | |
3069 | #endif | |
3070 | ||
8c7c6e34 KH |
3071 | static DEFINE_MUTEX(set_limit_mutex); |
3072 | ||
d38d2a75 | 3073 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3074 | unsigned long long val) |
628f4235 | 3075 | { |
81d39c20 | 3076 | int retry_count; |
3c11ecf4 | 3077 | u64 memswlimit, memlimit; |
628f4235 | 3078 | int ret = 0; |
81d39c20 KH |
3079 | int children = mem_cgroup_count_children(memcg); |
3080 | u64 curusage, oldusage; | |
3c11ecf4 | 3081 | int enlarge; |
81d39c20 KH |
3082 | |
3083 | /* | |
3084 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3085 | * is depends on callers. We set our retry-count to be function | |
3086 | * of # of children which we should visit in this loop. | |
3087 | */ | |
3088 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3089 | ||
3090 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3091 | |
3c11ecf4 | 3092 | enlarge = 0; |
8c7c6e34 | 3093 | while (retry_count) { |
628f4235 KH |
3094 | if (signal_pending(current)) { |
3095 | ret = -EINTR; | |
3096 | break; | |
3097 | } | |
8c7c6e34 KH |
3098 | /* |
3099 | * Rather than hide all in some function, I do this in | |
3100 | * open coded manner. You see what this really does. | |
3101 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3102 | */ | |
3103 | mutex_lock(&set_limit_mutex); | |
3104 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3105 | if (memswlimit < val) { | |
3106 | ret = -EINVAL; | |
3107 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3108 | break; |
3109 | } | |
3c11ecf4 KH |
3110 | |
3111 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3112 | if (memlimit < val) | |
3113 | enlarge = 1; | |
3114 | ||
8c7c6e34 | 3115 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3116 | if (!ret) { |
3117 | if (memswlimit == val) | |
3118 | memcg->memsw_is_minimum = true; | |
3119 | else | |
3120 | memcg->memsw_is_minimum = false; | |
3121 | } | |
8c7c6e34 KH |
3122 | mutex_unlock(&set_limit_mutex); |
3123 | ||
3124 | if (!ret) | |
3125 | break; | |
3126 | ||
aa20d489 | 3127 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
4e416953 | 3128 | MEM_CGROUP_RECLAIM_SHRINK); |
81d39c20 KH |
3129 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3130 | /* Usage is reduced ? */ | |
3131 | if (curusage >= oldusage) | |
3132 | retry_count--; | |
3133 | else | |
3134 | oldusage = curusage; | |
8c7c6e34 | 3135 | } |
3c11ecf4 KH |
3136 | if (!ret && enlarge) |
3137 | memcg_oom_recover(memcg); | |
14797e23 | 3138 | |
8c7c6e34 KH |
3139 | return ret; |
3140 | } | |
3141 | ||
338c8431 LZ |
3142 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3143 | unsigned long long val) | |
8c7c6e34 | 3144 | { |
81d39c20 | 3145 | int retry_count; |
3c11ecf4 | 3146 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3147 | int children = mem_cgroup_count_children(memcg); |
3148 | int ret = -EBUSY; | |
3c11ecf4 | 3149 | int enlarge = 0; |
8c7c6e34 | 3150 | |
81d39c20 KH |
3151 | /* see mem_cgroup_resize_res_limit */ |
3152 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3153 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3154 | while (retry_count) { |
3155 | if (signal_pending(current)) { | |
3156 | ret = -EINTR; | |
3157 | break; | |
3158 | } | |
3159 | /* | |
3160 | * Rather than hide all in some function, I do this in | |
3161 | * open coded manner. You see what this really does. | |
3162 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3163 | */ | |
3164 | mutex_lock(&set_limit_mutex); | |
3165 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3166 | if (memlimit > val) { | |
3167 | ret = -EINVAL; | |
3168 | mutex_unlock(&set_limit_mutex); | |
3169 | break; | |
3170 | } | |
3c11ecf4 KH |
3171 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3172 | if (memswlimit < val) | |
3173 | enlarge = 1; | |
8c7c6e34 | 3174 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3175 | if (!ret) { |
3176 | if (memlimit == val) | |
3177 | memcg->memsw_is_minimum = true; | |
3178 | else | |
3179 | memcg->memsw_is_minimum = false; | |
3180 | } | |
8c7c6e34 KH |
3181 | mutex_unlock(&set_limit_mutex); |
3182 | ||
3183 | if (!ret) | |
3184 | break; | |
3185 | ||
4e416953 | 3186 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 BS |
3187 | MEM_CGROUP_RECLAIM_NOSWAP | |
3188 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 3189 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3190 | /* Usage is reduced ? */ |
8c7c6e34 | 3191 | if (curusage >= oldusage) |
628f4235 | 3192 | retry_count--; |
81d39c20 KH |
3193 | else |
3194 | oldusage = curusage; | |
628f4235 | 3195 | } |
3c11ecf4 KH |
3196 | if (!ret && enlarge) |
3197 | memcg_oom_recover(memcg); | |
628f4235 KH |
3198 | return ret; |
3199 | } | |
3200 | ||
4e416953 | 3201 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
00918b6a | 3202 | gfp_t gfp_mask) |
4e416953 BS |
3203 | { |
3204 | unsigned long nr_reclaimed = 0; | |
3205 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3206 | unsigned long reclaimed; | |
3207 | int loop = 0; | |
3208 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3209 | unsigned long long excess; |
4e416953 BS |
3210 | |
3211 | if (order > 0) | |
3212 | return 0; | |
3213 | ||
00918b6a | 3214 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3215 | /* |
3216 | * This loop can run a while, specially if mem_cgroup's continuously | |
3217 | * keep exceeding their soft limit and putting the system under | |
3218 | * pressure | |
3219 | */ | |
3220 | do { | |
3221 | if (next_mz) | |
3222 | mz = next_mz; | |
3223 | else | |
3224 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3225 | if (!mz) | |
3226 | break; | |
3227 | ||
3228 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
3229 | gfp_mask, | |
3230 | MEM_CGROUP_RECLAIM_SOFT); | |
3231 | nr_reclaimed += reclaimed; | |
3232 | spin_lock(&mctz->lock); | |
3233 | ||
3234 | /* | |
3235 | * If we failed to reclaim anything from this memory cgroup | |
3236 | * it is time to move on to the next cgroup | |
3237 | */ | |
3238 | next_mz = NULL; | |
3239 | if (!reclaimed) { | |
3240 | do { | |
3241 | /* | |
3242 | * Loop until we find yet another one. | |
3243 | * | |
3244 | * By the time we get the soft_limit lock | |
3245 | * again, someone might have aded the | |
3246 | * group back on the RB tree. Iterate to | |
3247 | * make sure we get a different mem. | |
3248 | * mem_cgroup_largest_soft_limit_node returns | |
3249 | * NULL if no other cgroup is present on | |
3250 | * the tree | |
3251 | */ | |
3252 | next_mz = | |
3253 | __mem_cgroup_largest_soft_limit_node(mctz); | |
3254 | if (next_mz == mz) { | |
3255 | css_put(&next_mz->mem->css); | |
3256 | next_mz = NULL; | |
3257 | } else /* next_mz == NULL or other memcg */ | |
3258 | break; | |
3259 | } while (1); | |
3260 | } | |
4e416953 | 3261 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 3262 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
3263 | /* |
3264 | * One school of thought says that we should not add | |
3265 | * back the node to the tree if reclaim returns 0. | |
3266 | * But our reclaim could return 0, simply because due | |
3267 | * to priority we are exposing a smaller subset of | |
3268 | * memory to reclaim from. Consider this as a longer | |
3269 | * term TODO. | |
3270 | */ | |
ef8745c1 KH |
3271 | /* If excess == 0, no tree ops */ |
3272 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
3273 | spin_unlock(&mctz->lock); |
3274 | css_put(&mz->mem->css); | |
3275 | loop++; | |
3276 | /* | |
3277 | * Could not reclaim anything and there are no more | |
3278 | * mem cgroups to try or we seem to be looping without | |
3279 | * reclaiming anything. | |
3280 | */ | |
3281 | if (!nr_reclaimed && | |
3282 | (next_mz == NULL || | |
3283 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3284 | break; | |
3285 | } while (!nr_reclaimed); | |
3286 | if (next_mz) | |
3287 | css_put(&next_mz->mem->css); | |
3288 | return nr_reclaimed; | |
3289 | } | |
3290 | ||
cc847582 KH |
3291 | /* |
3292 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3293 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3294 | */ | |
f817ed48 | 3295 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 3296 | int node, int zid, enum lru_list lru) |
cc847582 | 3297 | { |
08e552c6 KH |
3298 | struct zone *zone; |
3299 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 3300 | struct page_cgroup *pc, *busy; |
08e552c6 | 3301 | unsigned long flags, loop; |
072c56c1 | 3302 | struct list_head *list; |
f817ed48 | 3303 | int ret = 0; |
072c56c1 | 3304 | |
08e552c6 KH |
3305 | zone = &NODE_DATA(node)->node_zones[zid]; |
3306 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 3307 | list = &mz->lists[lru]; |
cc847582 | 3308 | |
f817ed48 KH |
3309 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
3310 | /* give some margin against EBUSY etc...*/ | |
3311 | loop += 256; | |
3312 | busy = NULL; | |
3313 | while (loop--) { | |
5564e88b JW |
3314 | struct page *page; |
3315 | ||
f817ed48 | 3316 | ret = 0; |
08e552c6 | 3317 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3318 | if (list_empty(list)) { |
08e552c6 | 3319 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3320 | break; |
f817ed48 KH |
3321 | } |
3322 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
3323 | if (busy == pc) { | |
3324 | list_move(&pc->lru, list); | |
648bcc77 | 3325 | busy = NULL; |
08e552c6 | 3326 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3327 | continue; |
3328 | } | |
08e552c6 | 3329 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3330 | |
6b3ae58e | 3331 | page = lookup_cgroup_page(pc); |
5564e88b JW |
3332 | |
3333 | ret = mem_cgroup_move_parent(page, pc, mem, GFP_KERNEL); | |
f817ed48 | 3334 | if (ret == -ENOMEM) |
52d4b9ac | 3335 | break; |
f817ed48 KH |
3336 | |
3337 | if (ret == -EBUSY || ret == -EINVAL) { | |
3338 | /* found lock contention or "pc" is obsolete. */ | |
3339 | busy = pc; | |
3340 | cond_resched(); | |
3341 | } else | |
3342 | busy = NULL; | |
cc847582 | 3343 | } |
08e552c6 | 3344 | |
f817ed48 KH |
3345 | if (!ret && !list_empty(list)) |
3346 | return -EBUSY; | |
3347 | return ret; | |
cc847582 KH |
3348 | } |
3349 | ||
3350 | /* | |
3351 | * make mem_cgroup's charge to be 0 if there is no task. | |
3352 | * This enables deleting this mem_cgroup. | |
3353 | */ | |
c1e862c1 | 3354 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 3355 | { |
f817ed48 KH |
3356 | int ret; |
3357 | int node, zid, shrink; | |
3358 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 3359 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 3360 | |
cc847582 | 3361 | css_get(&mem->css); |
f817ed48 KH |
3362 | |
3363 | shrink = 0; | |
c1e862c1 KH |
3364 | /* should free all ? */ |
3365 | if (free_all) | |
3366 | goto try_to_free; | |
f817ed48 | 3367 | move_account: |
fce66477 | 3368 | do { |
f817ed48 | 3369 | ret = -EBUSY; |
c1e862c1 KH |
3370 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3371 | goto out; | |
3372 | ret = -EINTR; | |
3373 | if (signal_pending(current)) | |
cc847582 | 3374 | goto out; |
52d4b9ac KH |
3375 | /* This is for making all *used* pages to be on LRU. */ |
3376 | lru_add_drain_all(); | |
cdec2e42 | 3377 | drain_all_stock_sync(); |
f817ed48 | 3378 | ret = 0; |
32047e2a | 3379 | mem_cgroup_start_move(mem); |
299b4eaa | 3380 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3381 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 3382 | enum lru_list l; |
f817ed48 KH |
3383 | for_each_lru(l) { |
3384 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 3385 | node, zid, l); |
f817ed48 KH |
3386 | if (ret) |
3387 | break; | |
3388 | } | |
1ecaab2b | 3389 | } |
f817ed48 KH |
3390 | if (ret) |
3391 | break; | |
3392 | } | |
32047e2a | 3393 | mem_cgroup_end_move(mem); |
3c11ecf4 | 3394 | memcg_oom_recover(mem); |
f817ed48 KH |
3395 | /* it seems parent cgroup doesn't have enough mem */ |
3396 | if (ret == -ENOMEM) | |
3397 | goto try_to_free; | |
52d4b9ac | 3398 | cond_resched(); |
fce66477 DN |
3399 | /* "ret" should also be checked to ensure all lists are empty. */ |
3400 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
3401 | out: |
3402 | css_put(&mem->css); | |
3403 | return ret; | |
f817ed48 KH |
3404 | |
3405 | try_to_free: | |
c1e862c1 KH |
3406 | /* returns EBUSY if there is a task or if we come here twice. */ |
3407 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3408 | ret = -EBUSY; |
3409 | goto out; | |
3410 | } | |
c1e862c1 KH |
3411 | /* we call try-to-free pages for make this cgroup empty */ |
3412 | lru_add_drain_all(); | |
f817ed48 KH |
3413 | /* try to free all pages in this cgroup */ |
3414 | shrink = 1; | |
3415 | while (nr_retries && mem->res.usage > 0) { | |
3416 | int progress; | |
c1e862c1 KH |
3417 | |
3418 | if (signal_pending(current)) { | |
3419 | ret = -EINTR; | |
3420 | goto out; | |
3421 | } | |
a7885eb8 KM |
3422 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
3423 | false, get_swappiness(mem)); | |
c1e862c1 | 3424 | if (!progress) { |
f817ed48 | 3425 | nr_retries--; |
c1e862c1 | 3426 | /* maybe some writeback is necessary */ |
8aa7e847 | 3427 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3428 | } |
f817ed48 KH |
3429 | |
3430 | } | |
08e552c6 | 3431 | lru_add_drain(); |
f817ed48 | 3432 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3433 | goto move_account; |
cc847582 KH |
3434 | } |
3435 | ||
c1e862c1 KH |
3436 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3437 | { | |
3438 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3439 | } | |
3440 | ||
3441 | ||
18f59ea7 BS |
3442 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3443 | { | |
3444 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3445 | } | |
3446 | ||
3447 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3448 | u64 val) | |
3449 | { | |
3450 | int retval = 0; | |
3451 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3452 | struct cgroup *parent = cont->parent; | |
3453 | struct mem_cgroup *parent_mem = NULL; | |
3454 | ||
3455 | if (parent) | |
3456 | parent_mem = mem_cgroup_from_cont(parent); | |
3457 | ||
3458 | cgroup_lock(); | |
3459 | /* | |
af901ca1 | 3460 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3461 | * in the child subtrees. If it is unset, then the change can |
3462 | * occur, provided the current cgroup has no children. | |
3463 | * | |
3464 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3465 | * set if there are no children. | |
3466 | */ | |
3467 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
3468 | (val == 1 || val == 0)) { | |
3469 | if (list_empty(&cont->children)) | |
3470 | mem->use_hierarchy = val; | |
3471 | else | |
3472 | retval = -EBUSY; | |
3473 | } else | |
3474 | retval = -EINVAL; | |
3475 | cgroup_unlock(); | |
3476 | ||
3477 | return retval; | |
3478 | } | |
3479 | ||
0c3e73e8 | 3480 | |
7d74b06f KH |
3481 | static u64 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, |
3482 | enum mem_cgroup_stat_index idx) | |
0c3e73e8 | 3483 | { |
7d74b06f KH |
3484 | struct mem_cgroup *iter; |
3485 | s64 val = 0; | |
0c3e73e8 | 3486 | |
7d74b06f KH |
3487 | /* each per cpu's value can be minus.Then, use s64 */ |
3488 | for_each_mem_cgroup_tree(iter, mem) | |
3489 | val += mem_cgroup_read_stat(iter, idx); | |
3490 | ||
3491 | if (val < 0) /* race ? */ | |
3492 | val = 0; | |
3493 | return val; | |
0c3e73e8 BS |
3494 | } |
3495 | ||
104f3928 KS |
3496 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
3497 | { | |
7d74b06f | 3498 | u64 val; |
104f3928 KS |
3499 | |
3500 | if (!mem_cgroup_is_root(mem)) { | |
3501 | if (!swap) | |
3502 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
3503 | else | |
3504 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
3505 | } | |
3506 | ||
7d74b06f KH |
3507 | val = mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE); |
3508 | val += mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3509 | |
7d74b06f KH |
3510 | if (swap) |
3511 | val += mem_cgroup_get_recursive_idx_stat(mem, | |
3512 | MEM_CGROUP_STAT_SWAPOUT); | |
104f3928 KS |
3513 | |
3514 | return val << PAGE_SHIFT; | |
3515 | } | |
3516 | ||
2c3daa72 | 3517 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3518 | { |
8c7c6e34 | 3519 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 3520 | u64 val; |
8c7c6e34 KH |
3521 | int type, name; |
3522 | ||
3523 | type = MEMFILE_TYPE(cft->private); | |
3524 | name = MEMFILE_ATTR(cft->private); | |
3525 | switch (type) { | |
3526 | case _MEM: | |
104f3928 KS |
3527 | if (name == RES_USAGE) |
3528 | val = mem_cgroup_usage(mem, false); | |
3529 | else | |
0c3e73e8 | 3530 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
3531 | break; |
3532 | case _MEMSWAP: | |
104f3928 KS |
3533 | if (name == RES_USAGE) |
3534 | val = mem_cgroup_usage(mem, true); | |
3535 | else | |
0c3e73e8 | 3536 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
3537 | break; |
3538 | default: | |
3539 | BUG(); | |
3540 | break; | |
3541 | } | |
3542 | return val; | |
8cdea7c0 | 3543 | } |
628f4235 KH |
3544 | /* |
3545 | * The user of this function is... | |
3546 | * RES_LIMIT. | |
3547 | */ | |
856c13aa PM |
3548 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3549 | const char *buffer) | |
8cdea7c0 | 3550 | { |
628f4235 | 3551 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3552 | int type, name; |
628f4235 KH |
3553 | unsigned long long val; |
3554 | int ret; | |
3555 | ||
8c7c6e34 KH |
3556 | type = MEMFILE_TYPE(cft->private); |
3557 | name = MEMFILE_ATTR(cft->private); | |
3558 | switch (name) { | |
628f4235 | 3559 | case RES_LIMIT: |
4b3bde4c BS |
3560 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3561 | ret = -EINVAL; | |
3562 | break; | |
3563 | } | |
628f4235 KH |
3564 | /* This function does all necessary parse...reuse it */ |
3565 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3566 | if (ret) |
3567 | break; | |
3568 | if (type == _MEM) | |
628f4235 | 3569 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3570 | else |
3571 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3572 | break; |
296c81d8 BS |
3573 | case RES_SOFT_LIMIT: |
3574 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3575 | if (ret) | |
3576 | break; | |
3577 | /* | |
3578 | * For memsw, soft limits are hard to implement in terms | |
3579 | * of semantics, for now, we support soft limits for | |
3580 | * control without swap | |
3581 | */ | |
3582 | if (type == _MEM) | |
3583 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3584 | else | |
3585 | ret = -EINVAL; | |
3586 | break; | |
628f4235 KH |
3587 | default: |
3588 | ret = -EINVAL; /* should be BUG() ? */ | |
3589 | break; | |
3590 | } | |
3591 | return ret; | |
8cdea7c0 BS |
3592 | } |
3593 | ||
fee7b548 KH |
3594 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3595 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3596 | { | |
3597 | struct cgroup *cgroup; | |
3598 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3599 | ||
3600 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3601 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3602 | cgroup = memcg->css.cgroup; | |
3603 | if (!memcg->use_hierarchy) | |
3604 | goto out; | |
3605 | ||
3606 | while (cgroup->parent) { | |
3607 | cgroup = cgroup->parent; | |
3608 | memcg = mem_cgroup_from_cont(cgroup); | |
3609 | if (!memcg->use_hierarchy) | |
3610 | break; | |
3611 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3612 | min_limit = min(min_limit, tmp); | |
3613 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3614 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3615 | } | |
3616 | out: | |
3617 | *mem_limit = min_limit; | |
3618 | *memsw_limit = min_memsw_limit; | |
3619 | return; | |
3620 | } | |
3621 | ||
29f2a4da | 3622 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
3623 | { |
3624 | struct mem_cgroup *mem; | |
8c7c6e34 | 3625 | int type, name; |
c84872e1 PE |
3626 | |
3627 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
3628 | type = MEMFILE_TYPE(event); |
3629 | name = MEMFILE_ATTR(event); | |
3630 | switch (name) { | |
29f2a4da | 3631 | case RES_MAX_USAGE: |
8c7c6e34 KH |
3632 | if (type == _MEM) |
3633 | res_counter_reset_max(&mem->res); | |
3634 | else | |
3635 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
3636 | break; |
3637 | case RES_FAILCNT: | |
8c7c6e34 KH |
3638 | if (type == _MEM) |
3639 | res_counter_reset_failcnt(&mem->res); | |
3640 | else | |
3641 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
3642 | break; |
3643 | } | |
f64c3f54 | 3644 | |
85cc59db | 3645 | return 0; |
c84872e1 PE |
3646 | } |
3647 | ||
7dc74be0 DN |
3648 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3649 | struct cftype *cft) | |
3650 | { | |
3651 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3652 | } | |
3653 | ||
02491447 | 3654 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3655 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3656 | struct cftype *cft, u64 val) | |
3657 | { | |
3658 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3659 | ||
3660 | if (val >= (1 << NR_MOVE_TYPE)) | |
3661 | return -EINVAL; | |
3662 | /* | |
3663 | * We check this value several times in both in can_attach() and | |
3664 | * attach(), so we need cgroup lock to prevent this value from being | |
3665 | * inconsistent. | |
3666 | */ | |
3667 | cgroup_lock(); | |
3668 | mem->move_charge_at_immigrate = val; | |
3669 | cgroup_unlock(); | |
3670 | ||
3671 | return 0; | |
3672 | } | |
02491447 DN |
3673 | #else |
3674 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3675 | struct cftype *cft, u64 val) | |
3676 | { | |
3677 | return -ENOSYS; | |
3678 | } | |
3679 | #endif | |
7dc74be0 | 3680 | |
14067bb3 KH |
3681 | |
3682 | /* For read statistics */ | |
3683 | enum { | |
3684 | MCS_CACHE, | |
3685 | MCS_RSS, | |
d8046582 | 3686 | MCS_FILE_MAPPED, |
14067bb3 KH |
3687 | MCS_PGPGIN, |
3688 | MCS_PGPGOUT, | |
1dd3a273 | 3689 | MCS_SWAP, |
14067bb3 KH |
3690 | MCS_INACTIVE_ANON, |
3691 | MCS_ACTIVE_ANON, | |
3692 | MCS_INACTIVE_FILE, | |
3693 | MCS_ACTIVE_FILE, | |
3694 | MCS_UNEVICTABLE, | |
3695 | NR_MCS_STAT, | |
3696 | }; | |
3697 | ||
3698 | struct mcs_total_stat { | |
3699 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
3700 | }; |
3701 | ||
14067bb3 KH |
3702 | struct { |
3703 | char *local_name; | |
3704 | char *total_name; | |
3705 | } memcg_stat_strings[NR_MCS_STAT] = { | |
3706 | {"cache", "total_cache"}, | |
3707 | {"rss", "total_rss"}, | |
d69b042f | 3708 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
3709 | {"pgpgin", "total_pgpgin"}, |
3710 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3711 | {"swap", "total_swap"}, |
14067bb3 KH |
3712 | {"inactive_anon", "total_inactive_anon"}, |
3713 | {"active_anon", "total_active_anon"}, | |
3714 | {"inactive_file", "total_inactive_file"}, | |
3715 | {"active_file", "total_active_file"}, | |
3716 | {"unevictable", "total_unevictable"} | |
3717 | }; | |
3718 | ||
3719 | ||
7d74b06f KH |
3720 | static void |
3721 | mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
14067bb3 | 3722 | { |
14067bb3 KH |
3723 | s64 val; |
3724 | ||
3725 | /* per cpu stat */ | |
c62b1a3b | 3726 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 3727 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 3728 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 3729 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 3730 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 3731 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c62b1a3b | 3732 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); |
14067bb3 | 3733 | s->stat[MCS_PGPGIN] += val; |
c62b1a3b | 3734 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
14067bb3 | 3735 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 3736 | if (do_swap_account) { |
c62b1a3b | 3737 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
3738 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
3739 | } | |
14067bb3 KH |
3740 | |
3741 | /* per zone stat */ | |
3742 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
3743 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
3744 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
3745 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
3746 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
3747 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
3748 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
3749 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
3750 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
3751 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
14067bb3 KH |
3752 | } |
3753 | ||
3754 | static void | |
3755 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
3756 | { | |
7d74b06f KH |
3757 | struct mem_cgroup *iter; |
3758 | ||
3759 | for_each_mem_cgroup_tree(iter, mem) | |
3760 | mem_cgroup_get_local_stat(iter, s); | |
14067bb3 KH |
3761 | } |
3762 | ||
c64745cf PM |
3763 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
3764 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 3765 | { |
d2ceb9b7 | 3766 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 3767 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
3768 | int i; |
3769 | ||
14067bb3 KH |
3770 | memset(&mystat, 0, sizeof(mystat)); |
3771 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 3772 | |
1dd3a273 DN |
3773 | for (i = 0; i < NR_MCS_STAT; i++) { |
3774 | if (i == MCS_SWAP && !do_swap_account) | |
3775 | continue; | |
14067bb3 | 3776 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 3777 | } |
7b854121 | 3778 | |
14067bb3 | 3779 | /* Hierarchical information */ |
fee7b548 KH |
3780 | { |
3781 | unsigned long long limit, memsw_limit; | |
3782 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
3783 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
3784 | if (do_swap_account) | |
3785 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
3786 | } | |
7f016ee8 | 3787 | |
14067bb3 KH |
3788 | memset(&mystat, 0, sizeof(mystat)); |
3789 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
3790 | for (i = 0; i < NR_MCS_STAT; i++) { |
3791 | if (i == MCS_SWAP && !do_swap_account) | |
3792 | continue; | |
14067bb3 | 3793 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 3794 | } |
14067bb3 | 3795 | |
7f016ee8 | 3796 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 3797 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
3798 | |
3799 | { | |
3800 | int nid, zid; | |
3801 | struct mem_cgroup_per_zone *mz; | |
3802 | unsigned long recent_rotated[2] = {0, 0}; | |
3803 | unsigned long recent_scanned[2] = {0, 0}; | |
3804 | ||
3805 | for_each_online_node(nid) | |
3806 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3807 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3808 | ||
3809 | recent_rotated[0] += | |
3810 | mz->reclaim_stat.recent_rotated[0]; | |
3811 | recent_rotated[1] += | |
3812 | mz->reclaim_stat.recent_rotated[1]; | |
3813 | recent_scanned[0] += | |
3814 | mz->reclaim_stat.recent_scanned[0]; | |
3815 | recent_scanned[1] += | |
3816 | mz->reclaim_stat.recent_scanned[1]; | |
3817 | } | |
3818 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3819 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3820 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3821 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3822 | } | |
3823 | #endif | |
3824 | ||
d2ceb9b7 KH |
3825 | return 0; |
3826 | } | |
3827 | ||
a7885eb8 KM |
3828 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
3829 | { | |
3830 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3831 | ||
3832 | return get_swappiness(memcg); | |
3833 | } | |
3834 | ||
3835 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3836 | u64 val) | |
3837 | { | |
3838 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3839 | struct mem_cgroup *parent; | |
068b38c1 | 3840 | |
a7885eb8 KM |
3841 | if (val > 100) |
3842 | return -EINVAL; | |
3843 | ||
3844 | if (cgrp->parent == NULL) | |
3845 | return -EINVAL; | |
3846 | ||
3847 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
3848 | |
3849 | cgroup_lock(); | |
3850 | ||
a7885eb8 KM |
3851 | /* If under hierarchy, only empty-root can set this value */ |
3852 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
3853 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3854 | cgroup_unlock(); | |
a7885eb8 | 3855 | return -EINVAL; |
068b38c1 | 3856 | } |
a7885eb8 | 3857 | |
a7885eb8 | 3858 | memcg->swappiness = val; |
a7885eb8 | 3859 | |
068b38c1 LZ |
3860 | cgroup_unlock(); |
3861 | ||
a7885eb8 KM |
3862 | return 0; |
3863 | } | |
3864 | ||
2e72b634 KS |
3865 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
3866 | { | |
3867 | struct mem_cgroup_threshold_ary *t; | |
3868 | u64 usage; | |
3869 | int i; | |
3870 | ||
3871 | rcu_read_lock(); | |
3872 | if (!swap) | |
2c488db2 | 3873 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 3874 | else |
2c488db2 | 3875 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
3876 | |
3877 | if (!t) | |
3878 | goto unlock; | |
3879 | ||
3880 | usage = mem_cgroup_usage(memcg, swap); | |
3881 | ||
3882 | /* | |
3883 | * current_threshold points to threshold just below usage. | |
3884 | * If it's not true, a threshold was crossed after last | |
3885 | * call of __mem_cgroup_threshold(). | |
3886 | */ | |
5407a562 | 3887 | i = t->current_threshold; |
2e72b634 KS |
3888 | |
3889 | /* | |
3890 | * Iterate backward over array of thresholds starting from | |
3891 | * current_threshold and check if a threshold is crossed. | |
3892 | * If none of thresholds below usage is crossed, we read | |
3893 | * only one element of the array here. | |
3894 | */ | |
3895 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3896 | eventfd_signal(t->entries[i].eventfd, 1); | |
3897 | ||
3898 | /* i = current_threshold + 1 */ | |
3899 | i++; | |
3900 | ||
3901 | /* | |
3902 | * Iterate forward over array of thresholds starting from | |
3903 | * current_threshold+1 and check if a threshold is crossed. | |
3904 | * If none of thresholds above usage is crossed, we read | |
3905 | * only one element of the array here. | |
3906 | */ | |
3907 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3908 | eventfd_signal(t->entries[i].eventfd, 1); | |
3909 | ||
3910 | /* Update current_threshold */ | |
5407a562 | 3911 | t->current_threshold = i - 1; |
2e72b634 KS |
3912 | unlock: |
3913 | rcu_read_unlock(); | |
3914 | } | |
3915 | ||
3916 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3917 | { | |
ad4ca5f4 KS |
3918 | while (memcg) { |
3919 | __mem_cgroup_threshold(memcg, false); | |
3920 | if (do_swap_account) | |
3921 | __mem_cgroup_threshold(memcg, true); | |
3922 | ||
3923 | memcg = parent_mem_cgroup(memcg); | |
3924 | } | |
2e72b634 KS |
3925 | } |
3926 | ||
3927 | static int compare_thresholds(const void *a, const void *b) | |
3928 | { | |
3929 | const struct mem_cgroup_threshold *_a = a; | |
3930 | const struct mem_cgroup_threshold *_b = b; | |
3931 | ||
3932 | return _a->threshold - _b->threshold; | |
3933 | } | |
3934 | ||
7d74b06f | 3935 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem) |
9490ff27 KH |
3936 | { |
3937 | struct mem_cgroup_eventfd_list *ev; | |
3938 | ||
3939 | list_for_each_entry(ev, &mem->oom_notify, list) | |
3940 | eventfd_signal(ev->eventfd, 1); | |
3941 | return 0; | |
3942 | } | |
3943 | ||
3944 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) | |
3945 | { | |
7d74b06f KH |
3946 | struct mem_cgroup *iter; |
3947 | ||
3948 | for_each_mem_cgroup_tree(iter, mem) | |
3949 | mem_cgroup_oom_notify_cb(iter); | |
9490ff27 KH |
3950 | } |
3951 | ||
3952 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
3953 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
3954 | { |
3955 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
3956 | struct mem_cgroup_thresholds *thresholds; |
3957 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
3958 | int type = MEMFILE_TYPE(cft->private); |
3959 | u64 threshold, usage; | |
2c488db2 | 3960 | int i, size, ret; |
2e72b634 KS |
3961 | |
3962 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
3963 | if (ret) | |
3964 | return ret; | |
3965 | ||
3966 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 3967 | |
2e72b634 | 3968 | if (type == _MEM) |
2c488db2 | 3969 | thresholds = &memcg->thresholds; |
2e72b634 | 3970 | else if (type == _MEMSWAP) |
2c488db2 | 3971 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
3972 | else |
3973 | BUG(); | |
3974 | ||
3975 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3976 | ||
3977 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 3978 | if (thresholds->primary) |
2e72b634 KS |
3979 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
3980 | ||
2c488db2 | 3981 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
3982 | |
3983 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 3984 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 3985 | GFP_KERNEL); |
2c488db2 | 3986 | if (!new) { |
2e72b634 KS |
3987 | ret = -ENOMEM; |
3988 | goto unlock; | |
3989 | } | |
2c488db2 | 3990 | new->size = size; |
2e72b634 KS |
3991 | |
3992 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
3993 | if (thresholds->primary) { |
3994 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 3995 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
3996 | } |
3997 | ||
2e72b634 | 3998 | /* Add new threshold */ |
2c488db2 KS |
3999 | new->entries[size - 1].eventfd = eventfd; |
4000 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4001 | |
4002 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4003 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4004 | compare_thresholds, NULL); |
4005 | ||
4006 | /* Find current threshold */ | |
2c488db2 | 4007 | new->current_threshold = -1; |
2e72b634 | 4008 | for (i = 0; i < size; i++) { |
2c488db2 | 4009 | if (new->entries[i].threshold < usage) { |
2e72b634 | 4010 | /* |
2c488db2 KS |
4011 | * new->current_threshold will not be used until |
4012 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4013 | * it here. |
4014 | */ | |
2c488db2 | 4015 | ++new->current_threshold; |
2e72b634 KS |
4016 | } |
4017 | } | |
4018 | ||
2c488db2 KS |
4019 | /* Free old spare buffer and save old primary buffer as spare */ |
4020 | kfree(thresholds->spare); | |
4021 | thresholds->spare = thresholds->primary; | |
4022 | ||
4023 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4024 | |
907860ed | 4025 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4026 | synchronize_rcu(); |
4027 | ||
2e72b634 KS |
4028 | unlock: |
4029 | mutex_unlock(&memcg->thresholds_lock); | |
4030 | ||
4031 | return ret; | |
4032 | } | |
4033 | ||
907860ed | 4034 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4035 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4036 | { |
4037 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4038 | struct mem_cgroup_thresholds *thresholds; |
4039 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4040 | int type = MEMFILE_TYPE(cft->private); |
4041 | u64 usage; | |
2c488db2 | 4042 | int i, j, size; |
2e72b634 KS |
4043 | |
4044 | mutex_lock(&memcg->thresholds_lock); | |
4045 | if (type == _MEM) | |
2c488db2 | 4046 | thresholds = &memcg->thresholds; |
2e72b634 | 4047 | else if (type == _MEMSWAP) |
2c488db2 | 4048 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4049 | else |
4050 | BUG(); | |
4051 | ||
4052 | /* | |
4053 | * Something went wrong if we trying to unregister a threshold | |
4054 | * if we don't have thresholds | |
4055 | */ | |
4056 | BUG_ON(!thresholds); | |
4057 | ||
4058 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4059 | ||
4060 | /* Check if a threshold crossed before removing */ | |
4061 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4062 | ||
4063 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4064 | size = 0; |
4065 | for (i = 0; i < thresholds->primary->size; i++) { | |
4066 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4067 | size++; |
4068 | } | |
4069 | ||
2c488db2 | 4070 | new = thresholds->spare; |
907860ed | 4071 | |
2e72b634 KS |
4072 | /* Set thresholds array to NULL if we don't have thresholds */ |
4073 | if (!size) { | |
2c488db2 KS |
4074 | kfree(new); |
4075 | new = NULL; | |
907860ed | 4076 | goto swap_buffers; |
2e72b634 KS |
4077 | } |
4078 | ||
2c488db2 | 4079 | new->size = size; |
2e72b634 KS |
4080 | |
4081 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4082 | new->current_threshold = -1; |
4083 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4084 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4085 | continue; |
4086 | ||
2c488db2 KS |
4087 | new->entries[j] = thresholds->primary->entries[i]; |
4088 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4089 | /* |
2c488db2 | 4090 | * new->current_threshold will not be used |
2e72b634 KS |
4091 | * until rcu_assign_pointer(), so it's safe to increment |
4092 | * it here. | |
4093 | */ | |
2c488db2 | 4094 | ++new->current_threshold; |
2e72b634 KS |
4095 | } |
4096 | j++; | |
4097 | } | |
4098 | ||
907860ed | 4099 | swap_buffers: |
2c488db2 KS |
4100 | /* Swap primary and spare array */ |
4101 | thresholds->spare = thresholds->primary; | |
4102 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4103 | |
907860ed | 4104 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4105 | synchronize_rcu(); |
4106 | ||
2e72b634 | 4107 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4108 | } |
c1e862c1 | 4109 | |
9490ff27 KH |
4110 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4111 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4112 | { | |
4113 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4114 | struct mem_cgroup_eventfd_list *event; | |
4115 | int type = MEMFILE_TYPE(cft->private); | |
4116 | ||
4117 | BUG_ON(type != _OOM_TYPE); | |
4118 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4119 | if (!event) | |
4120 | return -ENOMEM; | |
4121 | ||
4122 | mutex_lock(&memcg_oom_mutex); | |
4123 | ||
4124 | event->eventfd = eventfd; | |
4125 | list_add(&event->list, &memcg->oom_notify); | |
4126 | ||
4127 | /* already in OOM ? */ | |
4128 | if (atomic_read(&memcg->oom_lock)) | |
4129 | eventfd_signal(eventfd, 1); | |
4130 | mutex_unlock(&memcg_oom_mutex); | |
4131 | ||
4132 | return 0; | |
4133 | } | |
4134 | ||
907860ed | 4135 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4136 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4137 | { | |
4138 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4139 | struct mem_cgroup_eventfd_list *ev, *tmp; | |
4140 | int type = MEMFILE_TYPE(cft->private); | |
4141 | ||
4142 | BUG_ON(type != _OOM_TYPE); | |
4143 | ||
4144 | mutex_lock(&memcg_oom_mutex); | |
4145 | ||
4146 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { | |
4147 | if (ev->eventfd == eventfd) { | |
4148 | list_del(&ev->list); | |
4149 | kfree(ev); | |
4150 | } | |
4151 | } | |
4152 | ||
4153 | mutex_unlock(&memcg_oom_mutex); | |
9490ff27 KH |
4154 | } |
4155 | ||
3c11ecf4 KH |
4156 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4157 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4158 | { | |
4159 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4160 | ||
4161 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); | |
4162 | ||
4163 | if (atomic_read(&mem->oom_lock)) | |
4164 | cb->fill(cb, "under_oom", 1); | |
4165 | else | |
4166 | cb->fill(cb, "under_oom", 0); | |
4167 | return 0; | |
4168 | } | |
4169 | ||
3c11ecf4 KH |
4170 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4171 | struct cftype *cft, u64 val) | |
4172 | { | |
4173 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4174 | struct mem_cgroup *parent; | |
4175 | ||
4176 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4177 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4178 | return -EINVAL; | |
4179 | ||
4180 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4181 | ||
4182 | cgroup_lock(); | |
4183 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4184 | if ((parent->use_hierarchy) || | |
4185 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { | |
4186 | cgroup_unlock(); | |
4187 | return -EINVAL; | |
4188 | } | |
4189 | mem->oom_kill_disable = val; | |
4d845ebf KH |
4190 | if (!val) |
4191 | memcg_oom_recover(mem); | |
3c11ecf4 KH |
4192 | cgroup_unlock(); |
4193 | return 0; | |
4194 | } | |
4195 | ||
8cdea7c0 BS |
4196 | static struct cftype mem_cgroup_files[] = { |
4197 | { | |
0eea1030 | 4198 | .name = "usage_in_bytes", |
8c7c6e34 | 4199 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4200 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4201 | .register_event = mem_cgroup_usage_register_event, |
4202 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4203 | }, |
c84872e1 PE |
4204 | { |
4205 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4206 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4207 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4208 | .read_u64 = mem_cgroup_read, |
4209 | }, | |
8cdea7c0 | 4210 | { |
0eea1030 | 4211 | .name = "limit_in_bytes", |
8c7c6e34 | 4212 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4213 | .write_string = mem_cgroup_write, |
2c3daa72 | 4214 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4215 | }, |
296c81d8 BS |
4216 | { |
4217 | .name = "soft_limit_in_bytes", | |
4218 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4219 | .write_string = mem_cgroup_write, | |
4220 | .read_u64 = mem_cgroup_read, | |
4221 | }, | |
8cdea7c0 BS |
4222 | { |
4223 | .name = "failcnt", | |
8c7c6e34 | 4224 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4225 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4226 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4227 | }, |
d2ceb9b7 KH |
4228 | { |
4229 | .name = "stat", | |
c64745cf | 4230 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4231 | }, |
c1e862c1 KH |
4232 | { |
4233 | .name = "force_empty", | |
4234 | .trigger = mem_cgroup_force_empty_write, | |
4235 | }, | |
18f59ea7 BS |
4236 | { |
4237 | .name = "use_hierarchy", | |
4238 | .write_u64 = mem_cgroup_hierarchy_write, | |
4239 | .read_u64 = mem_cgroup_hierarchy_read, | |
4240 | }, | |
a7885eb8 KM |
4241 | { |
4242 | .name = "swappiness", | |
4243 | .read_u64 = mem_cgroup_swappiness_read, | |
4244 | .write_u64 = mem_cgroup_swappiness_write, | |
4245 | }, | |
7dc74be0 DN |
4246 | { |
4247 | .name = "move_charge_at_immigrate", | |
4248 | .read_u64 = mem_cgroup_move_charge_read, | |
4249 | .write_u64 = mem_cgroup_move_charge_write, | |
4250 | }, | |
9490ff27 KH |
4251 | { |
4252 | .name = "oom_control", | |
3c11ecf4 KH |
4253 | .read_map = mem_cgroup_oom_control_read, |
4254 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4255 | .register_event = mem_cgroup_oom_register_event, |
4256 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4257 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4258 | }, | |
8cdea7c0 BS |
4259 | }; |
4260 | ||
8c7c6e34 KH |
4261 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4262 | static struct cftype memsw_cgroup_files[] = { | |
4263 | { | |
4264 | .name = "memsw.usage_in_bytes", | |
4265 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4266 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4267 | .register_event = mem_cgroup_usage_register_event, |
4268 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4269 | }, |
4270 | { | |
4271 | .name = "memsw.max_usage_in_bytes", | |
4272 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4273 | .trigger = mem_cgroup_reset, | |
4274 | .read_u64 = mem_cgroup_read, | |
4275 | }, | |
4276 | { | |
4277 | .name = "memsw.limit_in_bytes", | |
4278 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4279 | .write_string = mem_cgroup_write, | |
4280 | .read_u64 = mem_cgroup_read, | |
4281 | }, | |
4282 | { | |
4283 | .name = "memsw.failcnt", | |
4284 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4285 | .trigger = mem_cgroup_reset, | |
4286 | .read_u64 = mem_cgroup_read, | |
4287 | }, | |
4288 | }; | |
4289 | ||
4290 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4291 | { | |
4292 | if (!do_swap_account) | |
4293 | return 0; | |
4294 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4295 | ARRAY_SIZE(memsw_cgroup_files)); | |
4296 | }; | |
4297 | #else | |
4298 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4299 | { | |
4300 | return 0; | |
4301 | } | |
4302 | #endif | |
4303 | ||
6d12e2d8 KH |
4304 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4305 | { | |
4306 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4307 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 4308 | enum lru_list l; |
41e3355d | 4309 | int zone, tmp = node; |
1ecaab2b KH |
4310 | /* |
4311 | * This routine is called against possible nodes. | |
4312 | * But it's BUG to call kmalloc() against offline node. | |
4313 | * | |
4314 | * TODO: this routine can waste much memory for nodes which will | |
4315 | * never be onlined. It's better to use memory hotplug callback | |
4316 | * function. | |
4317 | */ | |
41e3355d KH |
4318 | if (!node_state(node, N_NORMAL_MEMORY)) |
4319 | tmp = -1; | |
17295c88 | 4320 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4321 | if (!pn) |
4322 | return 1; | |
1ecaab2b | 4323 | |
6d12e2d8 | 4324 | mem->info.nodeinfo[node] = pn; |
1ecaab2b KH |
4325 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4326 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
4327 | for_each_lru(l) |
4328 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 4329 | mz->usage_in_excess = 0; |
4e416953 BS |
4330 | mz->on_tree = false; |
4331 | mz->mem = mem; | |
1ecaab2b | 4332 | } |
6d12e2d8 KH |
4333 | return 0; |
4334 | } | |
4335 | ||
1ecaab2b KH |
4336 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4337 | { | |
4338 | kfree(mem->info.nodeinfo[node]); | |
4339 | } | |
4340 | ||
33327948 KH |
4341 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4342 | { | |
4343 | struct mem_cgroup *mem; | |
c62b1a3b | 4344 | int size = sizeof(struct mem_cgroup); |
33327948 | 4345 | |
c62b1a3b | 4346 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4347 | if (size < PAGE_SIZE) |
17295c88 | 4348 | mem = kzalloc(size, GFP_KERNEL); |
33327948 | 4349 | else |
17295c88 | 4350 | mem = vzalloc(size); |
33327948 | 4351 | |
e7bbcdf3 DC |
4352 | if (!mem) |
4353 | return NULL; | |
4354 | ||
c62b1a3b | 4355 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
d2e61b8d DC |
4356 | if (!mem->stat) |
4357 | goto out_free; | |
711d3d2c | 4358 | spin_lock_init(&mem->pcp_counter_lock); |
33327948 | 4359 | return mem; |
d2e61b8d DC |
4360 | |
4361 | out_free: | |
4362 | if (size < PAGE_SIZE) | |
4363 | kfree(mem); | |
4364 | else | |
4365 | vfree(mem); | |
4366 | return NULL; | |
33327948 KH |
4367 | } |
4368 | ||
8c7c6e34 KH |
4369 | /* |
4370 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4371 | * (scanning all at force_empty is too costly...) | |
4372 | * | |
4373 | * Instead of clearing all references at force_empty, we remember | |
4374 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4375 | * it goes down to 0. | |
4376 | * | |
8c7c6e34 KH |
4377 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4378 | */ | |
4379 | ||
a7ba0eef | 4380 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 4381 | { |
08e552c6 KH |
4382 | int node; |
4383 | ||
f64c3f54 | 4384 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
4385 | free_css_id(&mem_cgroup_subsys, &mem->css); |
4386 | ||
08e552c6 KH |
4387 | for_each_node_state(node, N_POSSIBLE) |
4388 | free_mem_cgroup_per_zone_info(mem, node); | |
4389 | ||
c62b1a3b KH |
4390 | free_percpu(mem->stat); |
4391 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
4392 | kfree(mem); |
4393 | else | |
4394 | vfree(mem); | |
4395 | } | |
4396 | ||
8c7c6e34 KH |
4397 | static void mem_cgroup_get(struct mem_cgroup *mem) |
4398 | { | |
4399 | atomic_inc(&mem->refcnt); | |
4400 | } | |
4401 | ||
483c30b5 | 4402 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 4403 | { |
483c30b5 | 4404 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 4405 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 4406 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
4407 | if (parent) |
4408 | mem_cgroup_put(parent); | |
4409 | } | |
8c7c6e34 KH |
4410 | } |
4411 | ||
483c30b5 DN |
4412 | static void mem_cgroup_put(struct mem_cgroup *mem) |
4413 | { | |
4414 | __mem_cgroup_put(mem, 1); | |
4415 | } | |
4416 | ||
7bcc1bb1 DN |
4417 | /* |
4418 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4419 | */ | |
4420 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
4421 | { | |
4422 | if (!mem->res.parent) | |
4423 | return NULL; | |
4424 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
4425 | } | |
33327948 | 4426 | |
c077719b KH |
4427 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4428 | static void __init enable_swap_cgroup(void) | |
4429 | { | |
f8d66542 | 4430 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4431 | do_swap_account = 1; |
4432 | } | |
4433 | #else | |
4434 | static void __init enable_swap_cgroup(void) | |
4435 | { | |
4436 | } | |
4437 | #endif | |
4438 | ||
f64c3f54 BS |
4439 | static int mem_cgroup_soft_limit_tree_init(void) |
4440 | { | |
4441 | struct mem_cgroup_tree_per_node *rtpn; | |
4442 | struct mem_cgroup_tree_per_zone *rtpz; | |
4443 | int tmp, node, zone; | |
4444 | ||
4445 | for_each_node_state(node, N_POSSIBLE) { | |
4446 | tmp = node; | |
4447 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4448 | tmp = -1; | |
4449 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4450 | if (!rtpn) | |
4451 | return 1; | |
4452 | ||
4453 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4454 | ||
4455 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4456 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4457 | rtpz->rb_root = RB_ROOT; | |
4458 | spin_lock_init(&rtpz->lock); | |
4459 | } | |
4460 | } | |
4461 | return 0; | |
4462 | } | |
4463 | ||
0eb253e2 | 4464 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4465 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4466 | { | |
28dbc4b6 | 4467 | struct mem_cgroup *mem, *parent; |
04046e1a | 4468 | long error = -ENOMEM; |
6d12e2d8 | 4469 | int node; |
8cdea7c0 | 4470 | |
c8dad2bb JB |
4471 | mem = mem_cgroup_alloc(); |
4472 | if (!mem) | |
04046e1a | 4473 | return ERR_PTR(error); |
78fb7466 | 4474 | |
6d12e2d8 KH |
4475 | for_each_node_state(node, N_POSSIBLE) |
4476 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
4477 | goto free_out; | |
f64c3f54 | 4478 | |
c077719b | 4479 | /* root ? */ |
28dbc4b6 | 4480 | if (cont->parent == NULL) { |
cdec2e42 | 4481 | int cpu; |
c077719b | 4482 | enable_swap_cgroup(); |
28dbc4b6 | 4483 | parent = NULL; |
4b3bde4c | 4484 | root_mem_cgroup = mem; |
f64c3f54 BS |
4485 | if (mem_cgroup_soft_limit_tree_init()) |
4486 | goto free_out; | |
cdec2e42 KH |
4487 | for_each_possible_cpu(cpu) { |
4488 | struct memcg_stock_pcp *stock = | |
4489 | &per_cpu(memcg_stock, cpu); | |
4490 | INIT_WORK(&stock->work, drain_local_stock); | |
4491 | } | |
711d3d2c | 4492 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4493 | } else { |
28dbc4b6 | 4494 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 | 4495 | mem->use_hierarchy = parent->use_hierarchy; |
3c11ecf4 | 4496 | mem->oom_kill_disable = parent->oom_kill_disable; |
18f59ea7 | 4497 | } |
28dbc4b6 | 4498 | |
18f59ea7 BS |
4499 | if (parent && parent->use_hierarchy) { |
4500 | res_counter_init(&mem->res, &parent->res); | |
4501 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4502 | /* |
4503 | * We increment refcnt of the parent to ensure that we can | |
4504 | * safely access it on res_counter_charge/uncharge. | |
4505 | * This refcnt will be decremented when freeing this | |
4506 | * mem_cgroup(see mem_cgroup_put). | |
4507 | */ | |
4508 | mem_cgroup_get(parent); | |
18f59ea7 BS |
4509 | } else { |
4510 | res_counter_init(&mem->res, NULL); | |
4511 | res_counter_init(&mem->memsw, NULL); | |
4512 | } | |
04046e1a | 4513 | mem->last_scanned_child = 0; |
9490ff27 | 4514 | INIT_LIST_HEAD(&mem->oom_notify); |
6d61ef40 | 4515 | |
a7885eb8 KM |
4516 | if (parent) |
4517 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 4518 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 4519 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 4520 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 4521 | return &mem->css; |
6d12e2d8 | 4522 | free_out: |
a7ba0eef | 4523 | __mem_cgroup_free(mem); |
4b3bde4c | 4524 | root_mem_cgroup = NULL; |
04046e1a | 4525 | return ERR_PTR(error); |
8cdea7c0 BS |
4526 | } |
4527 | ||
ec64f515 | 4528 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4529 | struct cgroup *cont) |
4530 | { | |
4531 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
4532 | |
4533 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
4534 | } |
4535 | ||
8cdea7c0 BS |
4536 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4537 | struct cgroup *cont) | |
4538 | { | |
c268e994 | 4539 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 4540 | |
c268e994 | 4541 | mem_cgroup_put(mem); |
8cdea7c0 BS |
4542 | } |
4543 | ||
4544 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4545 | struct cgroup *cont) | |
4546 | { | |
8c7c6e34 KH |
4547 | int ret; |
4548 | ||
4549 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4550 | ARRAY_SIZE(mem_cgroup_files)); | |
4551 | ||
4552 | if (!ret) | |
4553 | ret = register_memsw_files(cont, ss); | |
4554 | return ret; | |
8cdea7c0 BS |
4555 | } |
4556 | ||
02491447 | 4557 | #ifdef CONFIG_MMU |
7dc74be0 | 4558 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4559 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4560 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4561 | { |
854ffa8d DN |
4562 | int ret = 0; |
4563 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
4564 | struct mem_cgroup *mem = mc.to; |
4565 | ||
854ffa8d DN |
4566 | if (mem_cgroup_is_root(mem)) { |
4567 | mc.precharge += count; | |
4568 | /* we don't need css_get for root */ | |
4569 | return ret; | |
4570 | } | |
4571 | /* try to charge at once */ | |
4572 | if (count > 1) { | |
4573 | struct res_counter *dummy; | |
4574 | /* | |
4575 | * "mem" cannot be under rmdir() because we've already checked | |
4576 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
4577 | * are still under the same cgroup_mutex. So we can postpone | |
4578 | * css_get(). | |
4579 | */ | |
4580 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
4581 | goto one_by_one; | |
4582 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
4583 | PAGE_SIZE * count, &dummy)) { | |
4584 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
4585 | goto one_by_one; | |
4586 | } | |
4587 | mc.precharge += count; | |
854ffa8d DN |
4588 | return ret; |
4589 | } | |
4590 | one_by_one: | |
4591 | /* fall back to one by one charge */ | |
4592 | while (count--) { | |
4593 | if (signal_pending(current)) { | |
4594 | ret = -EINTR; | |
4595 | break; | |
4596 | } | |
4597 | if (!batch_count--) { | |
4598 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4599 | cond_resched(); | |
4600 | } | |
ec168510 AA |
4601 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, |
4602 | PAGE_SIZE); | |
854ffa8d DN |
4603 | if (ret || !mem) |
4604 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
4605 | return -ENOMEM; | |
4606 | mc.precharge++; | |
4607 | } | |
4ffef5fe DN |
4608 | return ret; |
4609 | } | |
4610 | ||
4611 | /** | |
4612 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
4613 | * @vma: the vma the pte to be checked belongs | |
4614 | * @addr: the address corresponding to the pte to be checked | |
4615 | * @ptent: the pte to be checked | |
02491447 | 4616 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4617 | * |
4618 | * Returns | |
4619 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4620 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4621 | * move charge. if @target is not NULL, the page is stored in target->page | |
4622 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4623 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4624 | * target for charge migration. if @target is not NULL, the entry is stored | |
4625 | * in target->ent. | |
4ffef5fe DN |
4626 | * |
4627 | * Called with pte lock held. | |
4628 | */ | |
4ffef5fe DN |
4629 | union mc_target { |
4630 | struct page *page; | |
02491447 | 4631 | swp_entry_t ent; |
4ffef5fe DN |
4632 | }; |
4633 | ||
4ffef5fe DN |
4634 | enum mc_target_type { |
4635 | MC_TARGET_NONE, /* not used */ | |
4636 | MC_TARGET_PAGE, | |
02491447 | 4637 | MC_TARGET_SWAP, |
4ffef5fe DN |
4638 | }; |
4639 | ||
90254a65 DN |
4640 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4641 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4642 | { |
90254a65 | 4643 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4644 | |
90254a65 DN |
4645 | if (!page || !page_mapped(page)) |
4646 | return NULL; | |
4647 | if (PageAnon(page)) { | |
4648 | /* we don't move shared anon */ | |
4649 | if (!move_anon() || page_mapcount(page) > 2) | |
4650 | return NULL; | |
87946a72 DN |
4651 | } else if (!move_file()) |
4652 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
4653 | return NULL; |
4654 | if (!get_page_unless_zero(page)) | |
4655 | return NULL; | |
4656 | ||
4657 | return page; | |
4658 | } | |
4659 | ||
4660 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
4661 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4662 | { | |
4663 | int usage_count; | |
4664 | struct page *page = NULL; | |
4665 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
4666 | ||
4667 | if (!move_anon() || non_swap_entry(ent)) | |
4668 | return NULL; | |
4669 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
4670 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
4671 | if (page) |
4672 | put_page(page); | |
90254a65 | 4673 | return NULL; |
02491447 | 4674 | } |
90254a65 DN |
4675 | if (do_swap_account) |
4676 | entry->val = ent.val; | |
4677 | ||
4678 | return page; | |
4679 | } | |
4680 | ||
87946a72 DN |
4681 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
4682 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
4683 | { | |
4684 | struct page *page = NULL; | |
4685 | struct inode *inode; | |
4686 | struct address_space *mapping; | |
4687 | pgoff_t pgoff; | |
4688 | ||
4689 | if (!vma->vm_file) /* anonymous vma */ | |
4690 | return NULL; | |
4691 | if (!move_file()) | |
4692 | return NULL; | |
4693 | ||
4694 | inode = vma->vm_file->f_path.dentry->d_inode; | |
4695 | mapping = vma->vm_file->f_mapping; | |
4696 | if (pte_none(ptent)) | |
4697 | pgoff = linear_page_index(vma, addr); | |
4698 | else /* pte_file(ptent) is true */ | |
4699 | pgoff = pte_to_pgoff(ptent); | |
4700 | ||
4701 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
4702 | if (!mapping_cap_swap_backed(mapping)) { /* normal file */ | |
4703 | page = find_get_page(mapping, pgoff); | |
4704 | } else { /* shmem/tmpfs file. we should take account of swap too. */ | |
4705 | swp_entry_t ent; | |
4706 | mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent); | |
4707 | if (do_swap_account) | |
4708 | entry->val = ent.val; | |
4709 | } | |
4710 | ||
4711 | return page; | |
4712 | } | |
4713 | ||
90254a65 DN |
4714 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
4715 | unsigned long addr, pte_t ptent, union mc_target *target) | |
4716 | { | |
4717 | struct page *page = NULL; | |
4718 | struct page_cgroup *pc; | |
4719 | int ret = 0; | |
4720 | swp_entry_t ent = { .val = 0 }; | |
4721 | ||
4722 | if (pte_present(ptent)) | |
4723 | page = mc_handle_present_pte(vma, addr, ptent); | |
4724 | else if (is_swap_pte(ptent)) | |
4725 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
4726 | else if (pte_none(ptent) || pte_file(ptent)) |
4727 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
4728 | |
4729 | if (!page && !ent.val) | |
4730 | return 0; | |
02491447 DN |
4731 | if (page) { |
4732 | pc = lookup_page_cgroup(page); | |
4733 | /* | |
4734 | * Do only loose check w/o page_cgroup lock. | |
4735 | * mem_cgroup_move_account() checks the pc is valid or not under | |
4736 | * the lock. | |
4737 | */ | |
4738 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
4739 | ret = MC_TARGET_PAGE; | |
4740 | if (target) | |
4741 | target->page = page; | |
4742 | } | |
4743 | if (!ret || !target) | |
4744 | put_page(page); | |
4745 | } | |
90254a65 DN |
4746 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
4747 | if (ent.val && !ret && | |
7f0f1546 KH |
4748 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
4749 | ret = MC_TARGET_SWAP; | |
4750 | if (target) | |
4751 | target->ent = ent; | |
4ffef5fe | 4752 | } |
4ffef5fe DN |
4753 | return ret; |
4754 | } | |
4755 | ||
4756 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
4757 | unsigned long addr, unsigned long end, | |
4758 | struct mm_walk *walk) | |
4759 | { | |
4760 | struct vm_area_struct *vma = walk->private; | |
4761 | pte_t *pte; | |
4762 | spinlock_t *ptl; | |
4763 | ||
03319327 DH |
4764 | split_huge_page_pmd(walk->mm, pmd); |
4765 | ||
4ffef5fe DN |
4766 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
4767 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
4768 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
4769 | mc.precharge++; /* increment precharge temporarily */ | |
4770 | pte_unmap_unlock(pte - 1, ptl); | |
4771 | cond_resched(); | |
4772 | ||
7dc74be0 DN |
4773 | return 0; |
4774 | } | |
4775 | ||
4ffef5fe DN |
4776 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4777 | { | |
4778 | unsigned long precharge; | |
4779 | struct vm_area_struct *vma; | |
4780 | ||
dfe076b0 | 4781 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
4782 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
4783 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
4784 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4785 | .mm = mm, | |
4786 | .private = vma, | |
4787 | }; | |
4788 | if (is_vm_hugetlb_page(vma)) | |
4789 | continue; | |
4ffef5fe DN |
4790 | walk_page_range(vma->vm_start, vma->vm_end, |
4791 | &mem_cgroup_count_precharge_walk); | |
4792 | } | |
dfe076b0 | 4793 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
4794 | |
4795 | precharge = mc.precharge; | |
4796 | mc.precharge = 0; | |
4797 | ||
4798 | return precharge; | |
4799 | } | |
4800 | ||
4ffef5fe DN |
4801 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4802 | { | |
dfe076b0 DN |
4803 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
4804 | ||
4805 | VM_BUG_ON(mc.moving_task); | |
4806 | mc.moving_task = current; | |
4807 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
4808 | } |
4809 | ||
dfe076b0 DN |
4810 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
4811 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 4812 | { |
2bd9bb20 KH |
4813 | struct mem_cgroup *from = mc.from; |
4814 | struct mem_cgroup *to = mc.to; | |
4815 | ||
4ffef5fe | 4816 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
4817 | if (mc.precharge) { |
4818 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
4819 | mc.precharge = 0; | |
4820 | } | |
4821 | /* | |
4822 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4823 | * we must uncharge here. | |
4824 | */ | |
4825 | if (mc.moved_charge) { | |
4826 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
4827 | mc.moved_charge = 0; | |
4ffef5fe | 4828 | } |
483c30b5 DN |
4829 | /* we must fixup refcnts and charges */ |
4830 | if (mc.moved_swap) { | |
483c30b5 DN |
4831 | /* uncharge swap account from the old cgroup */ |
4832 | if (!mem_cgroup_is_root(mc.from)) | |
4833 | res_counter_uncharge(&mc.from->memsw, | |
4834 | PAGE_SIZE * mc.moved_swap); | |
4835 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
4836 | ||
4837 | if (!mem_cgroup_is_root(mc.to)) { | |
4838 | /* | |
4839 | * we charged both to->res and to->memsw, so we should | |
4840 | * uncharge to->res. | |
4841 | */ | |
4842 | res_counter_uncharge(&mc.to->res, | |
4843 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
4844 | } |
4845 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
4846 | mc.moved_swap = 0; |
4847 | } | |
dfe076b0 DN |
4848 | memcg_oom_recover(from); |
4849 | memcg_oom_recover(to); | |
4850 | wake_up_all(&mc.waitq); | |
4851 | } | |
4852 | ||
4853 | static void mem_cgroup_clear_mc(void) | |
4854 | { | |
4855 | struct mem_cgroup *from = mc.from; | |
4856 | ||
4857 | /* | |
4858 | * we must clear moving_task before waking up waiters at the end of | |
4859 | * task migration. | |
4860 | */ | |
4861 | mc.moving_task = NULL; | |
4862 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 4863 | spin_lock(&mc.lock); |
4ffef5fe DN |
4864 | mc.from = NULL; |
4865 | mc.to = NULL; | |
2bd9bb20 | 4866 | spin_unlock(&mc.lock); |
32047e2a | 4867 | mem_cgroup_end_move(from); |
4ffef5fe DN |
4868 | } |
4869 | ||
7dc74be0 DN |
4870 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
4871 | struct cgroup *cgroup, | |
4872 | struct task_struct *p, | |
4873 | bool threadgroup) | |
4874 | { | |
4875 | int ret = 0; | |
4876 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
4877 | ||
4878 | if (mem->move_charge_at_immigrate) { | |
4879 | struct mm_struct *mm; | |
4880 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
4881 | ||
4882 | VM_BUG_ON(from == mem); | |
4883 | ||
4884 | mm = get_task_mm(p); | |
4885 | if (!mm) | |
4886 | return 0; | |
7dc74be0 | 4887 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
4888 | if (mm->owner == p) { |
4889 | VM_BUG_ON(mc.from); | |
4890 | VM_BUG_ON(mc.to); | |
4891 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 4892 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 4893 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 4894 | mem_cgroup_start_move(from); |
2bd9bb20 | 4895 | spin_lock(&mc.lock); |
4ffef5fe DN |
4896 | mc.from = from; |
4897 | mc.to = mem; | |
2bd9bb20 | 4898 | spin_unlock(&mc.lock); |
dfe076b0 | 4899 | /* We set mc.moving_task later */ |
4ffef5fe DN |
4900 | |
4901 | ret = mem_cgroup_precharge_mc(mm); | |
4902 | if (ret) | |
4903 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
4904 | } |
4905 | mmput(mm); | |
7dc74be0 DN |
4906 | } |
4907 | return ret; | |
4908 | } | |
4909 | ||
4910 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4911 | struct cgroup *cgroup, | |
4912 | struct task_struct *p, | |
4913 | bool threadgroup) | |
4914 | { | |
4ffef5fe | 4915 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
4916 | } |
4917 | ||
4ffef5fe DN |
4918 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4919 | unsigned long addr, unsigned long end, | |
4920 | struct mm_walk *walk) | |
7dc74be0 | 4921 | { |
4ffef5fe DN |
4922 | int ret = 0; |
4923 | struct vm_area_struct *vma = walk->private; | |
4924 | pte_t *pte; | |
4925 | spinlock_t *ptl; | |
4926 | ||
03319327 | 4927 | split_huge_page_pmd(walk->mm, pmd); |
4ffef5fe DN |
4928 | retry: |
4929 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4930 | for (; addr != end; addr += PAGE_SIZE) { | |
4931 | pte_t ptent = *(pte++); | |
4932 | union mc_target target; | |
4933 | int type; | |
4934 | struct page *page; | |
4935 | struct page_cgroup *pc; | |
02491447 | 4936 | swp_entry_t ent; |
4ffef5fe DN |
4937 | |
4938 | if (!mc.precharge) | |
4939 | break; | |
4940 | ||
4941 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
4942 | switch (type) { | |
4943 | case MC_TARGET_PAGE: | |
4944 | page = target.page; | |
4945 | if (isolate_lru_page(page)) | |
4946 | goto put; | |
4947 | pc = lookup_page_cgroup(page); | |
5564e88b | 4948 | if (!mem_cgroup_move_account(page, pc, |
987eba66 | 4949 | mc.from, mc.to, false, PAGE_SIZE)) { |
4ffef5fe | 4950 | mc.precharge--; |
854ffa8d DN |
4951 | /* we uncharge from mc.from later. */ |
4952 | mc.moved_charge++; | |
4ffef5fe DN |
4953 | } |
4954 | putback_lru_page(page); | |
4955 | put: /* is_target_pte_for_mc() gets the page */ | |
4956 | put_page(page); | |
4957 | break; | |
02491447 DN |
4958 | case MC_TARGET_SWAP: |
4959 | ent = target.ent; | |
483c30b5 DN |
4960 | if (!mem_cgroup_move_swap_account(ent, |
4961 | mc.from, mc.to, false)) { | |
02491447 | 4962 | mc.precharge--; |
483c30b5 DN |
4963 | /* we fixup refcnts and charges later. */ |
4964 | mc.moved_swap++; | |
4965 | } | |
02491447 | 4966 | break; |
4ffef5fe DN |
4967 | default: |
4968 | break; | |
4969 | } | |
4970 | } | |
4971 | pte_unmap_unlock(pte - 1, ptl); | |
4972 | cond_resched(); | |
4973 | ||
4974 | if (addr != end) { | |
4975 | /* | |
4976 | * We have consumed all precharges we got in can_attach(). | |
4977 | * We try charge one by one, but don't do any additional | |
4978 | * charges to mc.to if we have failed in charge once in attach() | |
4979 | * phase. | |
4980 | */ | |
854ffa8d | 4981 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4982 | if (!ret) |
4983 | goto retry; | |
4984 | } | |
4985 | ||
4986 | return ret; | |
4987 | } | |
4988 | ||
4989 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4990 | { | |
4991 | struct vm_area_struct *vma; | |
4992 | ||
4993 | lru_add_drain_all(); | |
dfe076b0 DN |
4994 | retry: |
4995 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
4996 | /* | |
4997 | * Someone who are holding the mmap_sem might be waiting in | |
4998 | * waitq. So we cancel all extra charges, wake up all waiters, | |
4999 | * and retry. Because we cancel precharges, we might not be able | |
5000 | * to move enough charges, but moving charge is a best-effort | |
5001 | * feature anyway, so it wouldn't be a big problem. | |
5002 | */ | |
5003 | __mem_cgroup_clear_mc(); | |
5004 | cond_resched(); | |
5005 | goto retry; | |
5006 | } | |
4ffef5fe DN |
5007 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5008 | int ret; | |
5009 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5010 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5011 | .mm = mm, | |
5012 | .private = vma, | |
5013 | }; | |
5014 | if (is_vm_hugetlb_page(vma)) | |
5015 | continue; | |
4ffef5fe DN |
5016 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5017 | &mem_cgroup_move_charge_walk); | |
5018 | if (ret) | |
5019 | /* | |
5020 | * means we have consumed all precharges and failed in | |
5021 | * doing additional charge. Just abandon here. | |
5022 | */ | |
5023 | break; | |
5024 | } | |
dfe076b0 | 5025 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5026 | } |
5027 | ||
67e465a7 BS |
5028 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
5029 | struct cgroup *cont, | |
5030 | struct cgroup *old_cont, | |
be367d09 BB |
5031 | struct task_struct *p, |
5032 | bool threadgroup) | |
67e465a7 | 5033 | { |
dfe076b0 DN |
5034 | struct mm_struct *mm; |
5035 | ||
5036 | if (!mc.to) | |
4ffef5fe DN |
5037 | /* no need to move charge */ |
5038 | return; | |
5039 | ||
dfe076b0 DN |
5040 | mm = get_task_mm(p); |
5041 | if (mm) { | |
5042 | mem_cgroup_move_charge(mm); | |
5043 | mmput(mm); | |
5044 | } | |
4ffef5fe | 5045 | mem_cgroup_clear_mc(); |
67e465a7 | 5046 | } |
5cfb80a7 DN |
5047 | #else /* !CONFIG_MMU */ |
5048 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
5049 | struct cgroup *cgroup, | |
5050 | struct task_struct *p, | |
5051 | bool threadgroup) | |
5052 | { | |
5053 | return 0; | |
5054 | } | |
5055 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5056 | struct cgroup *cgroup, | |
5057 | struct task_struct *p, | |
5058 | bool threadgroup) | |
5059 | { | |
5060 | } | |
5061 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
5062 | struct cgroup *cont, | |
5063 | struct cgroup *old_cont, | |
5064 | struct task_struct *p, | |
5065 | bool threadgroup) | |
5066 | { | |
5067 | } | |
5068 | #endif | |
67e465a7 | 5069 | |
8cdea7c0 BS |
5070 | struct cgroup_subsys mem_cgroup_subsys = { |
5071 | .name = "memory", | |
5072 | .subsys_id = mem_cgroup_subsys_id, | |
5073 | .create = mem_cgroup_create, | |
df878fb0 | 5074 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
5075 | .destroy = mem_cgroup_destroy, |
5076 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
5077 | .can_attach = mem_cgroup_can_attach, |
5078 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5079 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5080 | .early_init = 0, |
04046e1a | 5081 | .use_id = 1, |
8cdea7c0 | 5082 | }; |
c077719b KH |
5083 | |
5084 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5085 | static int __init enable_swap_account(char *s) |
5086 | { | |
5087 | /* consider enabled if no parameter or 1 is given */ | |
fceda1bf | 5088 | if (!(*s) || !strcmp(s, "=1")) |
a42c390c | 5089 | really_do_swap_account = 1; |
fceda1bf | 5090 | else if (!strcmp(s, "=0")) |
a42c390c MH |
5091 | really_do_swap_account = 0; |
5092 | return 1; | |
5093 | } | |
5094 | __setup("swapaccount", enable_swap_account); | |
c077719b KH |
5095 | |
5096 | static int __init disable_swap_account(char *s) | |
5097 | { | |
552b372b | 5098 | printk_once("noswapaccount is deprecated and will be removed in 2.6.40. Use swapaccount=0 instead\n"); |
fceda1bf | 5099 | enable_swap_account("=0"); |
c077719b KH |
5100 | return 1; |
5101 | } | |
5102 | __setup("noswapaccount", disable_swap_account); | |
5103 | #endif |