]>
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 | ||
76 | ||
d52aa412 KH |
77 | /* |
78 | * Statistics for memory cgroup. | |
79 | */ | |
80 | enum mem_cgroup_stat_index { | |
81 | /* | |
82 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
83 | */ | |
84 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 85 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 86 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
0c3e73e8 | 87 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
711d3d2c | 88 | MEM_CGROUP_STAT_DATA, /* end of data requires synchronization */ |
32047e2a | 89 | MEM_CGROUP_ON_MOVE, /* someone is moving account between groups */ |
d52aa412 KH |
90 | MEM_CGROUP_STAT_NSTATS, |
91 | }; | |
92 | ||
e9f8974f JW |
93 | enum mem_cgroup_events_index { |
94 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ | |
95 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ | |
96 | MEM_CGROUP_EVENTS_COUNT, /* # of pages paged in/out */ | |
97 | MEM_CGROUP_EVENTS_NSTATS, | |
98 | }; | |
7a159cc9 JW |
99 | /* |
100 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
101 | * it will be incremated by the number of pages. This counter is used for | |
102 | * for trigger some periodic events. This is straightforward and better | |
103 | * than using jiffies etc. to handle periodic memcg event. | |
104 | */ | |
105 | enum mem_cgroup_events_target { | |
106 | MEM_CGROUP_TARGET_THRESH, | |
107 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
108 | MEM_CGROUP_NTARGETS, | |
109 | }; | |
110 | #define THRESHOLDS_EVENTS_TARGET (128) | |
111 | #define SOFTLIMIT_EVENTS_TARGET (1024) | |
e9f8974f | 112 | |
d52aa412 | 113 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 114 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 115 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
7a159cc9 | 116 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
117 | }; |
118 | ||
6d12e2d8 KH |
119 | /* |
120 | * per-zone information in memory controller. | |
121 | */ | |
6d12e2d8 | 122 | struct mem_cgroup_per_zone { |
072c56c1 KH |
123 | /* |
124 | * spin_lock to protect the per cgroup LRU | |
125 | */ | |
b69408e8 CL |
126 | struct list_head lists[NR_LRU_LISTS]; |
127 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
128 | |
129 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
130 | struct rb_node tree_node; /* RB tree node */ |
131 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
132 | /* the soft limit is exceeded*/ | |
133 | bool on_tree; | |
4e416953 BS |
134 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
135 | /* use container_of */ | |
6d12e2d8 KH |
136 | }; |
137 | /* Macro for accessing counter */ | |
138 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
139 | ||
140 | struct mem_cgroup_per_node { | |
141 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
142 | }; | |
143 | ||
144 | struct mem_cgroup_lru_info { | |
145 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
146 | }; | |
147 | ||
f64c3f54 BS |
148 | /* |
149 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
150 | * their hierarchy representation | |
151 | */ | |
152 | ||
153 | struct mem_cgroup_tree_per_zone { | |
154 | struct rb_root rb_root; | |
155 | spinlock_t lock; | |
156 | }; | |
157 | ||
158 | struct mem_cgroup_tree_per_node { | |
159 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
160 | }; | |
161 | ||
162 | struct mem_cgroup_tree { | |
163 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
164 | }; | |
165 | ||
166 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
167 | ||
2e72b634 KS |
168 | struct mem_cgroup_threshold { |
169 | struct eventfd_ctx *eventfd; | |
170 | u64 threshold; | |
171 | }; | |
172 | ||
9490ff27 | 173 | /* For threshold */ |
2e72b634 KS |
174 | struct mem_cgroup_threshold_ary { |
175 | /* An array index points to threshold just below usage. */ | |
5407a562 | 176 | int current_threshold; |
2e72b634 KS |
177 | /* Size of entries[] */ |
178 | unsigned int size; | |
179 | /* Array of thresholds */ | |
180 | struct mem_cgroup_threshold entries[0]; | |
181 | }; | |
2c488db2 KS |
182 | |
183 | struct mem_cgroup_thresholds { | |
184 | /* Primary thresholds array */ | |
185 | struct mem_cgroup_threshold_ary *primary; | |
186 | /* | |
187 | * Spare threshold array. | |
188 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
189 | * It must be able to store at least primary->size - 1 entries. | |
190 | */ | |
191 | struct mem_cgroup_threshold_ary *spare; | |
192 | }; | |
193 | ||
9490ff27 KH |
194 | /* for OOM */ |
195 | struct mem_cgroup_eventfd_list { | |
196 | struct list_head list; | |
197 | struct eventfd_ctx *eventfd; | |
198 | }; | |
2e72b634 | 199 | |
2e72b634 | 200 | static void mem_cgroup_threshold(struct mem_cgroup *mem); |
9490ff27 | 201 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem); |
2e72b634 | 202 | |
8cdea7c0 BS |
203 | /* |
204 | * The memory controller data structure. The memory controller controls both | |
205 | * page cache and RSS per cgroup. We would eventually like to provide | |
206 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
207 | * to help the administrator determine what knobs to tune. | |
208 | * | |
209 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
210 | * we hit the water mark. May be even add a low water mark, such that |
211 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
212 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
213 | */ |
214 | struct mem_cgroup { | |
215 | struct cgroup_subsys_state css; | |
216 | /* | |
217 | * the counter to account for memory usage | |
218 | */ | |
219 | struct res_counter res; | |
8c7c6e34 KH |
220 | /* |
221 | * the counter to account for mem+swap usage. | |
222 | */ | |
223 | struct res_counter memsw; | |
78fb7466 PE |
224 | /* |
225 | * Per cgroup active and inactive list, similar to the | |
226 | * per zone LRU lists. | |
78fb7466 | 227 | */ |
6d12e2d8 | 228 | struct mem_cgroup_lru_info info; |
6d61ef40 | 229 | /* |
af901ca1 | 230 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 231 | * reclaimed from. |
6d61ef40 | 232 | */ |
04046e1a | 233 | int last_scanned_child; |
889976db YH |
234 | int last_scanned_node; |
235 | #if MAX_NUMNODES > 1 | |
236 | nodemask_t scan_nodes; | |
237 | unsigned long next_scan_node_update; | |
238 | #endif | |
18f59ea7 BS |
239 | /* |
240 | * Should the accounting and control be hierarchical, per subtree? | |
241 | */ | |
242 | bool use_hierarchy; | |
867578cb | 243 | atomic_t oom_lock; |
8c7c6e34 | 244 | atomic_t refcnt; |
14797e23 | 245 | |
a7885eb8 | 246 | unsigned int swappiness; |
3c11ecf4 KH |
247 | /* OOM-Killer disable */ |
248 | int oom_kill_disable; | |
a7885eb8 | 249 | |
22a668d7 KH |
250 | /* set when res.limit == memsw.limit */ |
251 | bool memsw_is_minimum; | |
252 | ||
2e72b634 KS |
253 | /* protect arrays of thresholds */ |
254 | struct mutex thresholds_lock; | |
255 | ||
256 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 257 | struct mem_cgroup_thresholds thresholds; |
907860ed | 258 | |
2e72b634 | 259 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 260 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 261 | |
9490ff27 KH |
262 | /* For oom notifier event fd */ |
263 | struct list_head oom_notify; | |
264 | ||
7dc74be0 DN |
265 | /* |
266 | * Should we move charges of a task when a task is moved into this | |
267 | * mem_cgroup ? And what type of charges should we move ? | |
268 | */ | |
269 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 270 | /* |
c62b1a3b | 271 | * percpu counter. |
d52aa412 | 272 | */ |
c62b1a3b | 273 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
274 | /* |
275 | * used when a cpu is offlined or other synchronizations | |
276 | * See mem_cgroup_read_stat(). | |
277 | */ | |
278 | struct mem_cgroup_stat_cpu nocpu_base; | |
279 | spinlock_t pcp_counter_lock; | |
8cdea7c0 BS |
280 | }; |
281 | ||
7dc74be0 DN |
282 | /* Stuffs for move charges at task migration. */ |
283 | /* | |
284 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
285 | * left-shifted bitmap of these types. | |
286 | */ | |
287 | enum move_type { | |
4ffef5fe | 288 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 289 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
290 | NR_MOVE_TYPE, |
291 | }; | |
292 | ||
4ffef5fe DN |
293 | /* "mc" and its members are protected by cgroup_mutex */ |
294 | static struct move_charge_struct { | |
b1dd693e | 295 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
296 | struct mem_cgroup *from; |
297 | struct mem_cgroup *to; | |
298 | unsigned long precharge; | |
854ffa8d | 299 | unsigned long moved_charge; |
483c30b5 | 300 | unsigned long moved_swap; |
8033b97c DN |
301 | struct task_struct *moving_task; /* a task moving charges */ |
302 | wait_queue_head_t waitq; /* a waitq for other context */ | |
303 | } mc = { | |
2bd9bb20 | 304 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
305 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
306 | }; | |
4ffef5fe | 307 | |
90254a65 DN |
308 | static bool move_anon(void) |
309 | { | |
310 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
311 | &mc.to->move_charge_at_immigrate); | |
312 | } | |
313 | ||
87946a72 DN |
314 | static bool move_file(void) |
315 | { | |
316 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
317 | &mc.to->move_charge_at_immigrate); | |
318 | } | |
319 | ||
4e416953 BS |
320 | /* |
321 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
322 | * limit reclaim to prevent infinite loops, if they ever occur. | |
323 | */ | |
324 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
325 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
326 | ||
217bc319 KH |
327 | enum charge_type { |
328 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
329 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 330 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 331 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 332 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 333 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
334 | NR_CHARGE_TYPE, |
335 | }; | |
336 | ||
8c7c6e34 KH |
337 | /* for encoding cft->private value on file */ |
338 | #define _MEM (0) | |
339 | #define _MEMSWAP (1) | |
9490ff27 | 340 | #define _OOM_TYPE (2) |
8c7c6e34 KH |
341 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
342 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
343 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
344 | /* Used for OOM nofiier */ |
345 | #define OOM_CONTROL (0) | |
8c7c6e34 | 346 | |
75822b44 BS |
347 | /* |
348 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
349 | */ | |
350 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
351 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
352 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
353 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
354 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
355 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 356 | |
8c7c6e34 KH |
357 | static void mem_cgroup_get(struct mem_cgroup *mem); |
358 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 359 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 360 | static void drain_all_stock_async(void); |
8c7c6e34 | 361 | |
f64c3f54 BS |
362 | static struct mem_cgroup_per_zone * |
363 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
364 | { | |
365 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
366 | } | |
367 | ||
d324236b WF |
368 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
369 | { | |
370 | return &mem->css; | |
371 | } | |
372 | ||
f64c3f54 | 373 | static struct mem_cgroup_per_zone * |
97a6c37b | 374 | page_cgroup_zoneinfo(struct mem_cgroup *mem, struct page *page) |
f64c3f54 | 375 | { |
97a6c37b JW |
376 | int nid = page_to_nid(page); |
377 | int zid = page_zonenum(page); | |
f64c3f54 | 378 | |
f64c3f54 BS |
379 | return mem_cgroup_zoneinfo(mem, nid, zid); |
380 | } | |
381 | ||
382 | static struct mem_cgroup_tree_per_zone * | |
383 | soft_limit_tree_node_zone(int nid, int zid) | |
384 | { | |
385 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
386 | } | |
387 | ||
388 | static struct mem_cgroup_tree_per_zone * | |
389 | soft_limit_tree_from_page(struct page *page) | |
390 | { | |
391 | int nid = page_to_nid(page); | |
392 | int zid = page_zonenum(page); | |
393 | ||
394 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
395 | } | |
396 | ||
397 | static void | |
4e416953 | 398 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 399 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
400 | struct mem_cgroup_tree_per_zone *mctz, |
401 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
402 | { |
403 | struct rb_node **p = &mctz->rb_root.rb_node; | |
404 | struct rb_node *parent = NULL; | |
405 | struct mem_cgroup_per_zone *mz_node; | |
406 | ||
407 | if (mz->on_tree) | |
408 | return; | |
409 | ||
ef8745c1 KH |
410 | mz->usage_in_excess = new_usage_in_excess; |
411 | if (!mz->usage_in_excess) | |
412 | return; | |
f64c3f54 BS |
413 | while (*p) { |
414 | parent = *p; | |
415 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
416 | tree_node); | |
417 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
418 | p = &(*p)->rb_left; | |
419 | /* | |
420 | * We can't avoid mem cgroups that are over their soft | |
421 | * limit by the same amount | |
422 | */ | |
423 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
424 | p = &(*p)->rb_right; | |
425 | } | |
426 | rb_link_node(&mz->tree_node, parent, p); | |
427 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
428 | mz->on_tree = true; | |
4e416953 BS |
429 | } |
430 | ||
431 | static void | |
432 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
433 | struct mem_cgroup_per_zone *mz, | |
434 | struct mem_cgroup_tree_per_zone *mctz) | |
435 | { | |
436 | if (!mz->on_tree) | |
437 | return; | |
438 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
439 | mz->on_tree = false; | |
440 | } | |
441 | ||
f64c3f54 BS |
442 | static void |
443 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
444 | struct mem_cgroup_per_zone *mz, | |
445 | struct mem_cgroup_tree_per_zone *mctz) | |
446 | { | |
447 | spin_lock(&mctz->lock); | |
4e416953 | 448 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
449 | spin_unlock(&mctz->lock); |
450 | } | |
451 | ||
f64c3f54 BS |
452 | |
453 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
454 | { | |
ef8745c1 | 455 | unsigned long long excess; |
f64c3f54 BS |
456 | struct mem_cgroup_per_zone *mz; |
457 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
458 | int nid = page_to_nid(page); |
459 | int zid = page_zonenum(page); | |
f64c3f54 BS |
460 | mctz = soft_limit_tree_from_page(page); |
461 | ||
462 | /* | |
4e649152 KH |
463 | * Necessary to update all ancestors when hierarchy is used. |
464 | * because their event counter is not touched. | |
f64c3f54 | 465 | */ |
4e649152 KH |
466 | for (; mem; mem = parent_mem_cgroup(mem)) { |
467 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 468 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
469 | /* |
470 | * We have to update the tree if mz is on RB-tree or | |
471 | * mem is over its softlimit. | |
472 | */ | |
ef8745c1 | 473 | if (excess || mz->on_tree) { |
4e649152 KH |
474 | spin_lock(&mctz->lock); |
475 | /* if on-tree, remove it */ | |
476 | if (mz->on_tree) | |
477 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
478 | /* | |
ef8745c1 KH |
479 | * Insert again. mz->usage_in_excess will be updated. |
480 | * If excess is 0, no tree ops. | |
4e649152 | 481 | */ |
ef8745c1 | 482 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
483 | spin_unlock(&mctz->lock); |
484 | } | |
f64c3f54 BS |
485 | } |
486 | } | |
487 | ||
488 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
489 | { | |
490 | int node, zone; | |
491 | struct mem_cgroup_per_zone *mz; | |
492 | struct mem_cgroup_tree_per_zone *mctz; | |
493 | ||
494 | for_each_node_state(node, N_POSSIBLE) { | |
495 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
496 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
497 | mctz = soft_limit_tree_node_zone(node, zone); | |
498 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
499 | } | |
500 | } | |
501 | } | |
502 | ||
4e416953 BS |
503 | static struct mem_cgroup_per_zone * |
504 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
505 | { | |
506 | struct rb_node *rightmost = NULL; | |
26251eaf | 507 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
508 | |
509 | retry: | |
26251eaf | 510 | mz = NULL; |
4e416953 BS |
511 | rightmost = rb_last(&mctz->rb_root); |
512 | if (!rightmost) | |
513 | goto done; /* Nothing to reclaim from */ | |
514 | ||
515 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
516 | /* | |
517 | * Remove the node now but someone else can add it back, | |
518 | * we will to add it back at the end of reclaim to its correct | |
519 | * position in the tree. | |
520 | */ | |
521 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
522 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
523 | !css_tryget(&mz->mem->css)) | |
524 | goto retry; | |
525 | done: | |
526 | return mz; | |
527 | } | |
528 | ||
529 | static struct mem_cgroup_per_zone * | |
530 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
531 | { | |
532 | struct mem_cgroup_per_zone *mz; | |
533 | ||
534 | spin_lock(&mctz->lock); | |
535 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
536 | spin_unlock(&mctz->lock); | |
537 | return mz; | |
538 | } | |
539 | ||
711d3d2c KH |
540 | /* |
541 | * Implementation Note: reading percpu statistics for memcg. | |
542 | * | |
543 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
544 | * synchronization to implement "quick" read. There are trade-off between | |
545 | * reading cost and precision of value. Then, we may have a chance to implement | |
546 | * a periodic synchronizion of counter in memcg's counter. | |
547 | * | |
548 | * But this _read() function is used for user interface now. The user accounts | |
549 | * memory usage by memory cgroup and he _always_ requires exact value because | |
550 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
551 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
552 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
553 | * | |
554 | * If there are kernel internal actions which can make use of some not-exact | |
555 | * value, and reading all cpu value can be performance bottleneck in some | |
556 | * common workload, threashold and synchonization as vmstat[] should be | |
557 | * implemented. | |
558 | */ | |
7a159cc9 JW |
559 | static long mem_cgroup_read_stat(struct mem_cgroup *mem, |
560 | enum mem_cgroup_stat_index idx) | |
c62b1a3b | 561 | { |
7a159cc9 | 562 | long val = 0; |
c62b1a3b | 563 | int cpu; |
c62b1a3b | 564 | |
711d3d2c KH |
565 | get_online_cpus(); |
566 | for_each_online_cpu(cpu) | |
c62b1a3b | 567 | val += per_cpu(mem->stat->count[idx], cpu); |
711d3d2c KH |
568 | #ifdef CONFIG_HOTPLUG_CPU |
569 | spin_lock(&mem->pcp_counter_lock); | |
570 | val += mem->nocpu_base.count[idx]; | |
571 | spin_unlock(&mem->pcp_counter_lock); | |
572 | #endif | |
573 | put_online_cpus(); | |
c62b1a3b KH |
574 | return val; |
575 | } | |
576 | ||
7a159cc9 | 577 | static long mem_cgroup_local_usage(struct mem_cgroup *mem) |
c62b1a3b | 578 | { |
7a159cc9 | 579 | long ret; |
c62b1a3b KH |
580 | |
581 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); | |
582 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); | |
583 | return ret; | |
584 | } | |
585 | ||
0c3e73e8 BS |
586 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
587 | bool charge) | |
588 | { | |
589 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 590 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
591 | } |
592 | ||
e9f8974f JW |
593 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *mem, |
594 | enum mem_cgroup_events_index idx) | |
595 | { | |
596 | unsigned long val = 0; | |
597 | int cpu; | |
598 | ||
599 | for_each_online_cpu(cpu) | |
600 | val += per_cpu(mem->stat->events[idx], cpu); | |
601 | #ifdef CONFIG_HOTPLUG_CPU | |
602 | spin_lock(&mem->pcp_counter_lock); | |
603 | val += mem->nocpu_base.events[idx]; | |
604 | spin_unlock(&mem->pcp_counter_lock); | |
605 | #endif | |
606 | return val; | |
607 | } | |
608 | ||
c05555b5 | 609 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
e401f176 | 610 | bool file, int nr_pages) |
d52aa412 | 611 | { |
c62b1a3b KH |
612 | preempt_disable(); |
613 | ||
e401f176 KH |
614 | if (file) |
615 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages); | |
d52aa412 | 616 | else |
e401f176 | 617 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages); |
55e462b0 | 618 | |
e401f176 KH |
619 | /* pagein of a big page is an event. So, ignore page size */ |
620 | if (nr_pages > 0) | |
e9f8974f | 621 | __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 622 | else { |
e9f8974f | 623 | __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
624 | nr_pages = -nr_pages; /* for event */ |
625 | } | |
e401f176 | 626 | |
e9f8974f | 627 | __this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
2e72b634 | 628 | |
c62b1a3b | 629 | preempt_enable(); |
6d12e2d8 KH |
630 | } |
631 | ||
889976db YH |
632 | static unsigned long |
633 | mem_cgroup_get_zonestat_node(struct mem_cgroup *mem, int nid, enum lru_list idx) | |
634 | { | |
635 | struct mem_cgroup_per_zone *mz; | |
636 | u64 total = 0; | |
637 | int zid; | |
638 | ||
639 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
640 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
641 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
642 | } | |
643 | return total; | |
644 | } | |
14067bb3 | 645 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 646 | enum lru_list idx) |
6d12e2d8 | 647 | { |
889976db | 648 | int nid; |
6d12e2d8 KH |
649 | u64 total = 0; |
650 | ||
651 | for_each_online_node(nid) | |
889976db | 652 | total += mem_cgroup_get_zonestat_node(mem, nid, idx); |
6d12e2d8 | 653 | return total; |
d52aa412 KH |
654 | } |
655 | ||
7a159cc9 JW |
656 | static bool __memcg_event_check(struct mem_cgroup *mem, int target) |
657 | { | |
658 | unsigned long val, next; | |
659 | ||
660 | val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]); | |
661 | next = this_cpu_read(mem->stat->targets[target]); | |
662 | /* from time_after() in jiffies.h */ | |
663 | return ((long)next - (long)val < 0); | |
664 | } | |
665 | ||
666 | static void __mem_cgroup_target_update(struct mem_cgroup *mem, int target) | |
d2265e6f | 667 | { |
7a159cc9 | 668 | unsigned long val, next; |
d2265e6f | 669 | |
e9f8974f | 670 | val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
d2265e6f | 671 | |
7a159cc9 JW |
672 | switch (target) { |
673 | case MEM_CGROUP_TARGET_THRESH: | |
674 | next = val + THRESHOLDS_EVENTS_TARGET; | |
675 | break; | |
676 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
677 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
678 | break; | |
679 | default: | |
680 | return; | |
681 | } | |
682 | ||
683 | this_cpu_write(mem->stat->targets[target], next); | |
d2265e6f KH |
684 | } |
685 | ||
686 | /* | |
687 | * Check events in order. | |
688 | * | |
689 | */ | |
690 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) | |
691 | { | |
692 | /* threshold event is triggered in finer grain than soft limit */ | |
7a159cc9 | 693 | if (unlikely(__memcg_event_check(mem, MEM_CGROUP_TARGET_THRESH))) { |
d2265e6f | 694 | mem_cgroup_threshold(mem); |
7a159cc9 JW |
695 | __mem_cgroup_target_update(mem, MEM_CGROUP_TARGET_THRESH); |
696 | if (unlikely(__memcg_event_check(mem, | |
697 | MEM_CGROUP_TARGET_SOFTLIMIT))){ | |
d2265e6f | 698 | mem_cgroup_update_tree(mem, page); |
7a159cc9 JW |
699 | __mem_cgroup_target_update(mem, |
700 | MEM_CGROUP_TARGET_SOFTLIMIT); | |
701 | } | |
d2265e6f KH |
702 | } |
703 | } | |
704 | ||
d5b69e38 | 705 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
706 | { |
707 | return container_of(cgroup_subsys_state(cont, | |
708 | mem_cgroup_subsys_id), struct mem_cgroup, | |
709 | css); | |
710 | } | |
711 | ||
cf475ad2 | 712 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 713 | { |
31a78f23 BS |
714 | /* |
715 | * mm_update_next_owner() may clear mm->owner to NULL | |
716 | * if it races with swapoff, page migration, etc. | |
717 | * So this can be called with p == NULL. | |
718 | */ | |
719 | if (unlikely(!p)) | |
720 | return NULL; | |
721 | ||
78fb7466 PE |
722 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
723 | struct mem_cgroup, css); | |
724 | } | |
725 | ||
54595fe2 KH |
726 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
727 | { | |
728 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
729 | |
730 | if (!mm) | |
731 | return NULL; | |
54595fe2 KH |
732 | /* |
733 | * Because we have no locks, mm->owner's may be being moved to other | |
734 | * cgroup. We use css_tryget() here even if this looks | |
735 | * pessimistic (rather than adding locks here). | |
736 | */ | |
737 | rcu_read_lock(); | |
738 | do { | |
739 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
740 | if (unlikely(!mem)) | |
741 | break; | |
742 | } while (!css_tryget(&mem->css)); | |
743 | rcu_read_unlock(); | |
744 | return mem; | |
745 | } | |
746 | ||
7d74b06f KH |
747 | /* The caller has to guarantee "mem" exists before calling this */ |
748 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem) | |
14067bb3 | 749 | { |
711d3d2c KH |
750 | struct cgroup_subsys_state *css; |
751 | int found; | |
752 | ||
753 | if (!mem) /* ROOT cgroup has the smallest ID */ | |
754 | return root_mem_cgroup; /*css_put/get against root is ignored*/ | |
755 | if (!mem->use_hierarchy) { | |
756 | if (css_tryget(&mem->css)) | |
757 | return mem; | |
758 | return NULL; | |
759 | } | |
760 | rcu_read_lock(); | |
761 | /* | |
762 | * searching a memory cgroup which has the smallest ID under given | |
763 | * ROOT cgroup. (ID >= 1) | |
764 | */ | |
765 | css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found); | |
766 | if (css && css_tryget(css)) | |
767 | mem = container_of(css, struct mem_cgroup, css); | |
768 | else | |
769 | mem = NULL; | |
770 | rcu_read_unlock(); | |
771 | return mem; | |
7d74b06f KH |
772 | } |
773 | ||
774 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, | |
775 | struct mem_cgroup *root, | |
776 | bool cond) | |
777 | { | |
778 | int nextid = css_id(&iter->css) + 1; | |
779 | int found; | |
780 | int hierarchy_used; | |
14067bb3 | 781 | struct cgroup_subsys_state *css; |
14067bb3 | 782 | |
7d74b06f | 783 | hierarchy_used = iter->use_hierarchy; |
14067bb3 | 784 | |
7d74b06f | 785 | css_put(&iter->css); |
711d3d2c KH |
786 | /* If no ROOT, walk all, ignore hierarchy */ |
787 | if (!cond || (root && !hierarchy_used)) | |
7d74b06f | 788 | return NULL; |
14067bb3 | 789 | |
711d3d2c KH |
790 | if (!root) |
791 | root = root_mem_cgroup; | |
792 | ||
7d74b06f KH |
793 | do { |
794 | iter = NULL; | |
14067bb3 | 795 | rcu_read_lock(); |
7d74b06f KH |
796 | |
797 | css = css_get_next(&mem_cgroup_subsys, nextid, | |
798 | &root->css, &found); | |
14067bb3 | 799 | if (css && css_tryget(css)) |
7d74b06f | 800 | iter = container_of(css, struct mem_cgroup, css); |
14067bb3 | 801 | rcu_read_unlock(); |
7d74b06f | 802 | /* If css is NULL, no more cgroups will be found */ |
14067bb3 | 803 | nextid = found + 1; |
7d74b06f | 804 | } while (css && !iter); |
14067bb3 | 805 | |
7d74b06f | 806 | return iter; |
14067bb3 | 807 | } |
7d74b06f KH |
808 | /* |
809 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please | |
810 | * be careful that "break" loop is not allowed. We have reference count. | |
811 | * Instead of that modify "cond" to be false and "continue" to exit the loop. | |
812 | */ | |
813 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ | |
814 | for (iter = mem_cgroup_start_loop(root);\ | |
815 | iter != NULL;\ | |
816 | iter = mem_cgroup_get_next(iter, root, cond)) | |
817 | ||
818 | #define for_each_mem_cgroup_tree(iter, root) \ | |
819 | for_each_mem_cgroup_tree_cond(iter, root, true) | |
820 | ||
711d3d2c KH |
821 | #define for_each_mem_cgroup_all(iter) \ |
822 | for_each_mem_cgroup_tree_cond(iter, NULL, true) | |
823 | ||
14067bb3 | 824 | |
4b3bde4c BS |
825 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
826 | { | |
827 | return (mem == root_mem_cgroup); | |
828 | } | |
829 | ||
08e552c6 KH |
830 | /* |
831 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
832 | * Operations are called by routine of global LRU independently from memcg. | |
833 | * What we have to take care of here is validness of pc->mem_cgroup. | |
834 | * | |
835 | * Changes to pc->mem_cgroup happens when | |
836 | * 1. charge | |
837 | * 2. moving account | |
838 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
839 | * It is added to LRU before charge. | |
840 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
841 | * When moving account, the page is not on LRU. It's isolated. | |
842 | */ | |
4f98a2fe | 843 | |
08e552c6 KH |
844 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
845 | { | |
846 | struct page_cgroup *pc; | |
08e552c6 | 847 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 848 | |
f8d66542 | 849 | if (mem_cgroup_disabled()) |
08e552c6 KH |
850 | return; |
851 | pc = lookup_page_cgroup(page); | |
852 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 853 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 854 | return; |
4b3bde4c | 855 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
856 | /* |
857 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
858 | * removed from global LRU. | |
859 | */ | |
97a6c37b | 860 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
861 | /* huge page split is done under lru_lock. so, we have no races. */ |
862 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); | |
4b3bde4c BS |
863 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
864 | return; | |
865 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 | 866 | list_del_init(&pc->lru); |
6d12e2d8 KH |
867 | } |
868 | ||
08e552c6 | 869 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 870 | { |
08e552c6 KH |
871 | mem_cgroup_del_lru_list(page, page_lru(page)); |
872 | } | |
b69408e8 | 873 | |
3f58a829 MK |
874 | /* |
875 | * Writeback is about to end against a page which has been marked for immediate | |
876 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
877 | * inactive list. | |
878 | */ | |
879 | void mem_cgroup_rotate_reclaimable_page(struct page *page) | |
880 | { | |
881 | struct mem_cgroup_per_zone *mz; | |
882 | struct page_cgroup *pc; | |
883 | enum lru_list lru = page_lru(page); | |
884 | ||
885 | if (mem_cgroup_disabled()) | |
886 | return; | |
887 | ||
888 | pc = lookup_page_cgroup(page); | |
889 | /* unused or root page is not rotated. */ | |
890 | if (!PageCgroupUsed(pc)) | |
891 | return; | |
892 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
893 | smp_rmb(); | |
894 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
895 | return; | |
97a6c37b | 896 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3f58a829 MK |
897 | list_move_tail(&pc->lru, &mz->lists[lru]); |
898 | } | |
899 | ||
08e552c6 KH |
900 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
901 | { | |
902 | struct mem_cgroup_per_zone *mz; | |
903 | struct page_cgroup *pc; | |
b69408e8 | 904 | |
f8d66542 | 905 | if (mem_cgroup_disabled()) |
08e552c6 | 906 | return; |
6d12e2d8 | 907 | |
08e552c6 | 908 | pc = lookup_page_cgroup(page); |
4b3bde4c | 909 | /* unused or root page is not rotated. */ |
713735b4 JW |
910 | if (!PageCgroupUsed(pc)) |
911 | return; | |
912 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
913 | smp_rmb(); | |
914 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 | 915 | return; |
97a6c37b | 916 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
08e552c6 | 917 | list_move(&pc->lru, &mz->lists[lru]); |
6d12e2d8 KH |
918 | } |
919 | ||
08e552c6 | 920 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 921 | { |
08e552c6 KH |
922 | struct page_cgroup *pc; |
923 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 924 | |
f8d66542 | 925 | if (mem_cgroup_disabled()) |
08e552c6 KH |
926 | return; |
927 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 928 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
08e552c6 | 929 | if (!PageCgroupUsed(pc)) |
894bc310 | 930 | return; |
713735b4 JW |
931 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
932 | smp_rmb(); | |
97a6c37b | 933 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
934 | /* huge page split is done under lru_lock. so, we have no races. */ |
935 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); | |
4b3bde4c BS |
936 | SetPageCgroupAcctLRU(pc); |
937 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
938 | return; | |
08e552c6 KH |
939 | list_add(&pc->lru, &mz->lists[lru]); |
940 | } | |
544122e5 | 941 | |
08e552c6 | 942 | /* |
5a6475a4 KH |
943 | * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed |
944 | * while it's linked to lru because the page may be reused after it's fully | |
945 | * uncharged. To handle that, unlink page_cgroup from LRU when charge it again. | |
946 | * It's done under lock_page and expected that zone->lru_lock isnever held. | |
08e552c6 | 947 | */ |
5a6475a4 | 948 | static void mem_cgroup_lru_del_before_commit(struct page *page) |
08e552c6 | 949 | { |
544122e5 KH |
950 | unsigned long flags; |
951 | struct zone *zone = page_zone(page); | |
952 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
953 | ||
5a6475a4 KH |
954 | /* |
955 | * Doing this check without taking ->lru_lock seems wrong but this | |
956 | * is safe. Because if page_cgroup's USED bit is unset, the page | |
957 | * will not be added to any memcg's LRU. If page_cgroup's USED bit is | |
958 | * set, the commit after this will fail, anyway. | |
959 | * This all charge/uncharge is done under some mutual execustion. | |
960 | * So, we don't need to taking care of changes in USED bit. | |
961 | */ | |
962 | if (likely(!PageLRU(page))) | |
963 | return; | |
964 | ||
544122e5 KH |
965 | spin_lock_irqsave(&zone->lru_lock, flags); |
966 | /* | |
967 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
968 | * is guarded by lock_page() because the page is SwapCache. | |
969 | */ | |
970 | if (!PageCgroupUsed(pc)) | |
971 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
972 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
973 | } |
974 | ||
5a6475a4 | 975 | static void mem_cgroup_lru_add_after_commit(struct page *page) |
544122e5 KH |
976 | { |
977 | unsigned long flags; | |
978 | struct zone *zone = page_zone(page); | |
979 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
980 | ||
5a6475a4 KH |
981 | /* taking care of that the page is added to LRU while we commit it */ |
982 | if (likely(!PageLRU(page))) | |
983 | return; | |
544122e5 KH |
984 | spin_lock_irqsave(&zone->lru_lock, flags); |
985 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 986 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
987 | mem_cgroup_add_lru_list(page, page_lru(page)); |
988 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
989 | } | |
990 | ||
991 | ||
08e552c6 KH |
992 | void mem_cgroup_move_lists(struct page *page, |
993 | enum lru_list from, enum lru_list to) | |
994 | { | |
f8d66542 | 995 | if (mem_cgroup_disabled()) |
08e552c6 KH |
996 | return; |
997 | mem_cgroup_del_lru_list(page, from); | |
998 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
999 | } |
1000 | ||
4c4a2214 DR |
1001 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
1002 | { | |
1003 | int ret; | |
0b7f569e | 1004 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1005 | struct task_struct *p; |
4c4a2214 | 1006 | |
158e0a2d KH |
1007 | p = find_lock_task_mm(task); |
1008 | if (!p) | |
1009 | return 0; | |
1010 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1011 | task_unlock(p); | |
0b7f569e KH |
1012 | if (!curr) |
1013 | return 0; | |
d31f56db DN |
1014 | /* |
1015 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
1016 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
1017 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
1018 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
1019 | */ | |
1020 | if (mem->use_hierarchy) | |
0b7f569e KH |
1021 | ret = css_is_ancestor(&curr->css, &mem->css); |
1022 | else | |
1023 | ret = (curr == mem); | |
1024 | css_put(&curr->css); | |
4c4a2214 DR |
1025 | return ret; |
1026 | } | |
1027 | ||
c772be93 | 1028 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
1029 | { |
1030 | unsigned long active; | |
1031 | unsigned long inactive; | |
c772be93 KM |
1032 | unsigned long gb; |
1033 | unsigned long inactive_ratio; | |
14797e23 | 1034 | |
14067bb3 KH |
1035 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
1036 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 1037 | |
c772be93 KM |
1038 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1039 | if (gb) | |
1040 | inactive_ratio = int_sqrt(10 * gb); | |
1041 | else | |
1042 | inactive_ratio = 1; | |
1043 | ||
1044 | if (present_pages) { | |
1045 | present_pages[0] = inactive; | |
1046 | present_pages[1] = active; | |
1047 | } | |
1048 | ||
1049 | return inactive_ratio; | |
1050 | } | |
1051 | ||
1052 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
1053 | { | |
1054 | unsigned long active; | |
1055 | unsigned long inactive; | |
1056 | unsigned long present_pages[2]; | |
1057 | unsigned long inactive_ratio; | |
1058 | ||
1059 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
1060 | ||
1061 | inactive = present_pages[0]; | |
1062 | active = present_pages[1]; | |
1063 | ||
1064 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
1065 | return 1; |
1066 | ||
1067 | return 0; | |
1068 | } | |
1069 | ||
56e49d21 RR |
1070 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
1071 | { | |
1072 | unsigned long active; | |
1073 | unsigned long inactive; | |
1074 | ||
1075 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
1076 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
1077 | ||
1078 | return (active > inactive); | |
1079 | } | |
1080 | ||
1bac180b YH |
1081 | unsigned long mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, |
1082 | struct zone *zone, | |
1083 | enum lru_list lru) | |
a3d8e054 | 1084 | { |
13d7e3a2 | 1085 | int nid = zone_to_nid(zone); |
a3d8e054 KM |
1086 | int zid = zone_idx(zone); |
1087 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1088 | ||
1089 | return MEM_CGROUP_ZSTAT(mz, lru); | |
1090 | } | |
1091 | ||
406eb0c9 YH |
1092 | #ifdef CONFIG_NUMA |
1093 | static unsigned long mem_cgroup_node_nr_file_lru_pages(struct mem_cgroup *memcg, | |
1094 | int nid) | |
1095 | { | |
1096 | unsigned long ret; | |
1097 | ||
1098 | ret = mem_cgroup_get_zonestat_node(memcg, nid, LRU_INACTIVE_FILE) + | |
1099 | mem_cgroup_get_zonestat_node(memcg, nid, LRU_ACTIVE_FILE); | |
1100 | ||
1101 | return ret; | |
1102 | } | |
1103 | ||
1104 | static unsigned long mem_cgroup_nr_file_lru_pages(struct mem_cgroup *memcg) | |
1105 | { | |
1106 | u64 total = 0; | |
1107 | int nid; | |
1108 | ||
1109 | for_each_node_state(nid, N_HIGH_MEMORY) | |
1110 | total += mem_cgroup_node_nr_file_lru_pages(memcg, nid); | |
1111 | ||
1112 | return total; | |
1113 | } | |
1114 | ||
1115 | static unsigned long mem_cgroup_node_nr_anon_lru_pages(struct mem_cgroup *memcg, | |
1116 | int nid) | |
1117 | { | |
1118 | unsigned long ret; | |
1119 | ||
1120 | ret = mem_cgroup_get_zonestat_node(memcg, nid, LRU_INACTIVE_ANON) + | |
1121 | mem_cgroup_get_zonestat_node(memcg, nid, LRU_ACTIVE_ANON); | |
1122 | ||
1123 | return ret; | |
1124 | } | |
1125 | ||
1126 | static unsigned long mem_cgroup_nr_anon_lru_pages(struct mem_cgroup *memcg) | |
1127 | { | |
1128 | u64 total = 0; | |
1129 | int nid; | |
1130 | ||
1131 | for_each_node_state(nid, N_HIGH_MEMORY) | |
1132 | total += mem_cgroup_node_nr_anon_lru_pages(memcg, nid); | |
1133 | ||
1134 | return total; | |
1135 | } | |
1136 | ||
1137 | static unsigned long | |
1138 | mem_cgroup_node_nr_unevictable_lru_pages(struct mem_cgroup *memcg, int nid) | |
1139 | { | |
1140 | return mem_cgroup_get_zonestat_node(memcg, nid, LRU_UNEVICTABLE); | |
1141 | } | |
1142 | ||
1143 | static unsigned long | |
1144 | mem_cgroup_nr_unevictable_lru_pages(struct mem_cgroup *memcg) | |
1145 | { | |
1146 | u64 total = 0; | |
1147 | int nid; | |
1148 | ||
1149 | for_each_node_state(nid, N_HIGH_MEMORY) | |
1150 | total += mem_cgroup_node_nr_unevictable_lru_pages(memcg, nid); | |
1151 | ||
1152 | return total; | |
1153 | } | |
1154 | ||
1155 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, | |
1156 | int nid) | |
1157 | { | |
1158 | enum lru_list l; | |
1159 | u64 total = 0; | |
1160 | ||
1161 | for_each_lru(l) | |
1162 | total += mem_cgroup_get_zonestat_node(memcg, nid, l); | |
1163 | ||
1164 | return total; | |
1165 | } | |
1166 | ||
1167 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg) | |
1168 | { | |
1169 | u64 total = 0; | |
1170 | int nid; | |
1171 | ||
1172 | for_each_node_state(nid, N_HIGH_MEMORY) | |
1173 | total += mem_cgroup_node_nr_lru_pages(memcg, nid); | |
1174 | ||
1175 | return total; | |
1176 | } | |
1177 | #endif /* CONFIG_NUMA */ | |
1178 | ||
3e2f41f1 KM |
1179 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1180 | struct zone *zone) | |
1181 | { | |
13d7e3a2 | 1182 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1183 | int zid = zone_idx(zone); |
1184 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1185 | ||
1186 | return &mz->reclaim_stat; | |
1187 | } | |
1188 | ||
1189 | struct zone_reclaim_stat * | |
1190 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1191 | { | |
1192 | struct page_cgroup *pc; | |
1193 | struct mem_cgroup_per_zone *mz; | |
1194 | ||
1195 | if (mem_cgroup_disabled()) | |
1196 | return NULL; | |
1197 | ||
1198 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1199 | if (!PageCgroupUsed(pc)) |
1200 | return NULL; | |
713735b4 JW |
1201 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1202 | smp_rmb(); | |
97a6c37b | 1203 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3e2f41f1 KM |
1204 | return &mz->reclaim_stat; |
1205 | } | |
1206 | ||
66e1707b BS |
1207 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
1208 | struct list_head *dst, | |
1209 | unsigned long *scanned, int order, | |
1210 | int mode, struct zone *z, | |
1211 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 1212 | int active, int file) |
66e1707b BS |
1213 | { |
1214 | unsigned long nr_taken = 0; | |
1215 | struct page *page; | |
1216 | unsigned long scan; | |
1217 | LIST_HEAD(pc_list); | |
1218 | struct list_head *src; | |
ff7283fa | 1219 | struct page_cgroup *pc, *tmp; |
13d7e3a2 | 1220 | int nid = zone_to_nid(z); |
1ecaab2b KH |
1221 | int zid = zone_idx(z); |
1222 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 1223 | int lru = LRU_FILE * file + active; |
2ffebca6 | 1224 | int ret; |
66e1707b | 1225 | |
cf475ad2 | 1226 | BUG_ON(!mem_cont); |
1ecaab2b | 1227 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 1228 | src = &mz->lists[lru]; |
66e1707b | 1229 | |
ff7283fa KH |
1230 | scan = 0; |
1231 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 1232 | if (scan >= nr_to_scan) |
ff7283fa | 1233 | break; |
08e552c6 | 1234 | |
52d4b9ac KH |
1235 | if (unlikely(!PageCgroupUsed(pc))) |
1236 | continue; | |
5564e88b | 1237 | |
6b3ae58e | 1238 | page = lookup_cgroup_page(pc); |
5564e88b | 1239 | |
436c6541 | 1240 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1241 | continue; |
ff7283fa | 1242 | |
436c6541 | 1243 | scan++; |
2ffebca6 KH |
1244 | ret = __isolate_lru_page(page, mode, file); |
1245 | switch (ret) { | |
1246 | case 0: | |
66e1707b | 1247 | list_move(&page->lru, dst); |
2ffebca6 | 1248 | mem_cgroup_del_lru(page); |
2c888cfb | 1249 | nr_taken += hpage_nr_pages(page); |
2ffebca6 KH |
1250 | break; |
1251 | case -EBUSY: | |
1252 | /* we don't affect global LRU but rotate in our LRU */ | |
1253 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1254 | break; | |
1255 | default: | |
1256 | break; | |
66e1707b BS |
1257 | } |
1258 | } | |
1259 | ||
66e1707b | 1260 | *scanned = scan; |
cc8e970c KM |
1261 | |
1262 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, | |
1263 | 0, 0, 0, mode); | |
1264 | ||
66e1707b BS |
1265 | return nr_taken; |
1266 | } | |
1267 | ||
6d61ef40 BS |
1268 | #define mem_cgroup_from_res_counter(counter, member) \ |
1269 | container_of(counter, struct mem_cgroup, member) | |
1270 | ||
19942822 | 1271 | /** |
9d11ea9f JW |
1272 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1273 | * @mem: the memory cgroup | |
19942822 | 1274 | * |
9d11ea9f | 1275 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1276 | * pages. |
19942822 | 1277 | */ |
7ec99d62 | 1278 | static unsigned long mem_cgroup_margin(struct mem_cgroup *mem) |
19942822 | 1279 | { |
9d11ea9f JW |
1280 | unsigned long long margin; |
1281 | ||
1282 | margin = res_counter_margin(&mem->res); | |
1283 | if (do_swap_account) | |
1284 | margin = min(margin, res_counter_margin(&mem->memsw)); | |
7ec99d62 | 1285 | return margin >> PAGE_SHIFT; |
19942822 JW |
1286 | } |
1287 | ||
a7885eb8 KM |
1288 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1289 | { | |
1290 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1291 | |
1292 | /* root ? */ | |
1293 | if (cgrp->parent == NULL) | |
1294 | return vm_swappiness; | |
1295 | ||
bf1ff263 | 1296 | return memcg->swappiness; |
a7885eb8 KM |
1297 | } |
1298 | ||
32047e2a KH |
1299 | static void mem_cgroup_start_move(struct mem_cgroup *mem) |
1300 | { | |
1301 | int cpu; | |
1489ebad KH |
1302 | |
1303 | get_online_cpus(); | |
1304 | spin_lock(&mem->pcp_counter_lock); | |
1305 | for_each_online_cpu(cpu) | |
32047e2a | 1306 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1489ebad KH |
1307 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; |
1308 | spin_unlock(&mem->pcp_counter_lock); | |
1309 | put_online_cpus(); | |
32047e2a KH |
1310 | |
1311 | synchronize_rcu(); | |
1312 | } | |
1313 | ||
1314 | static void mem_cgroup_end_move(struct mem_cgroup *mem) | |
1315 | { | |
1316 | int cpu; | |
1317 | ||
1318 | if (!mem) | |
1319 | return; | |
1489ebad KH |
1320 | get_online_cpus(); |
1321 | spin_lock(&mem->pcp_counter_lock); | |
1322 | for_each_online_cpu(cpu) | |
32047e2a | 1323 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1489ebad KH |
1324 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; |
1325 | spin_unlock(&mem->pcp_counter_lock); | |
1326 | put_online_cpus(); | |
32047e2a KH |
1327 | } |
1328 | /* | |
1329 | * 2 routines for checking "mem" is under move_account() or not. | |
1330 | * | |
1331 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1332 | * for avoiding race in accounting. If true, | |
1333 | * pc->mem_cgroup may be overwritten. | |
1334 | * | |
1335 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1336 | * under hierarchy of moving cgroups. This is for | |
1337 | * waiting at hith-memory prressure caused by "move". | |
1338 | */ | |
1339 | ||
1340 | static bool mem_cgroup_stealed(struct mem_cgroup *mem) | |
1341 | { | |
1342 | VM_BUG_ON(!rcu_read_lock_held()); | |
1343 | return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0; | |
1344 | } | |
4b534334 KH |
1345 | |
1346 | static bool mem_cgroup_under_move(struct mem_cgroup *mem) | |
1347 | { | |
2bd9bb20 KH |
1348 | struct mem_cgroup *from; |
1349 | struct mem_cgroup *to; | |
4b534334 | 1350 | bool ret = false; |
2bd9bb20 KH |
1351 | /* |
1352 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1353 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1354 | */ | |
1355 | spin_lock(&mc.lock); | |
1356 | from = mc.from; | |
1357 | to = mc.to; | |
1358 | if (!from) | |
1359 | goto unlock; | |
1360 | if (from == mem || to == mem | |
1361 | || (mem->use_hierarchy && css_is_ancestor(&from->css, &mem->css)) | |
1362 | || (mem->use_hierarchy && css_is_ancestor(&to->css, &mem->css))) | |
1363 | ret = true; | |
1364 | unlock: | |
1365 | spin_unlock(&mc.lock); | |
4b534334 KH |
1366 | return ret; |
1367 | } | |
1368 | ||
1369 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem) | |
1370 | { | |
1371 | if (mc.moving_task && current != mc.moving_task) { | |
1372 | if (mem_cgroup_under_move(mem)) { | |
1373 | DEFINE_WAIT(wait); | |
1374 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1375 | /* moving charge context might have finished. */ | |
1376 | if (mc.moving_task) | |
1377 | schedule(); | |
1378 | finish_wait(&mc.waitq, &wait); | |
1379 | return true; | |
1380 | } | |
1381 | } | |
1382 | return false; | |
1383 | } | |
1384 | ||
e222432b | 1385 | /** |
6a6135b6 | 1386 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1387 | * @memcg: The memory cgroup that went over limit |
1388 | * @p: Task that is going to be killed | |
1389 | * | |
1390 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1391 | * enabled | |
1392 | */ | |
1393 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1394 | { | |
1395 | struct cgroup *task_cgrp; | |
1396 | struct cgroup *mem_cgrp; | |
1397 | /* | |
1398 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1399 | * on the assumption that OOM is serialized for memory controller. | |
1400 | * If this assumption is broken, revisit this code. | |
1401 | */ | |
1402 | static char memcg_name[PATH_MAX]; | |
1403 | int ret; | |
1404 | ||
d31f56db | 1405 | if (!memcg || !p) |
e222432b BS |
1406 | return; |
1407 | ||
1408 | ||
1409 | rcu_read_lock(); | |
1410 | ||
1411 | mem_cgrp = memcg->css.cgroup; | |
1412 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1413 | ||
1414 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1415 | if (ret < 0) { | |
1416 | /* | |
1417 | * Unfortunately, we are unable to convert to a useful name | |
1418 | * But we'll still print out the usage information | |
1419 | */ | |
1420 | rcu_read_unlock(); | |
1421 | goto done; | |
1422 | } | |
1423 | rcu_read_unlock(); | |
1424 | ||
1425 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1426 | ||
1427 | rcu_read_lock(); | |
1428 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1429 | if (ret < 0) { | |
1430 | rcu_read_unlock(); | |
1431 | goto done; | |
1432 | } | |
1433 | rcu_read_unlock(); | |
1434 | ||
1435 | /* | |
1436 | * Continues from above, so we don't need an KERN_ level | |
1437 | */ | |
1438 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1439 | done: | |
1440 | ||
1441 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1442 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1443 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1444 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1445 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1446 | "failcnt %llu\n", | |
1447 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1448 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1449 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1450 | } | |
1451 | ||
81d39c20 KH |
1452 | /* |
1453 | * This function returns the number of memcg under hierarchy tree. Returns | |
1454 | * 1(self count) if no children. | |
1455 | */ | |
1456 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1457 | { | |
1458 | int num = 0; | |
7d74b06f KH |
1459 | struct mem_cgroup *iter; |
1460 | ||
1461 | for_each_mem_cgroup_tree(iter, mem) | |
1462 | num++; | |
81d39c20 KH |
1463 | return num; |
1464 | } | |
1465 | ||
a63d83f4 DR |
1466 | /* |
1467 | * Return the memory (and swap, if configured) limit for a memcg. | |
1468 | */ | |
1469 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1470 | { | |
1471 | u64 limit; | |
1472 | u64 memsw; | |
1473 | ||
f3e8eb70 JW |
1474 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1475 | limit += total_swap_pages << PAGE_SHIFT; | |
1476 | ||
a63d83f4 DR |
1477 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1478 | /* | |
1479 | * If memsw is finite and limits the amount of swap space available | |
1480 | * to this memcg, return that limit. | |
1481 | */ | |
1482 | return min(limit, memsw); | |
1483 | } | |
1484 | ||
6d61ef40 | 1485 | /* |
04046e1a KH |
1486 | * Visit the first child (need not be the first child as per the ordering |
1487 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1488 | * that to reclaim free pages from. | |
1489 | */ | |
1490 | static struct mem_cgroup * | |
1491 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1492 | { | |
1493 | struct mem_cgroup *ret = NULL; | |
1494 | struct cgroup_subsys_state *css; | |
1495 | int nextid, found; | |
1496 | ||
1497 | if (!root_mem->use_hierarchy) { | |
1498 | css_get(&root_mem->css); | |
1499 | ret = root_mem; | |
1500 | } | |
1501 | ||
1502 | while (!ret) { | |
1503 | rcu_read_lock(); | |
1504 | nextid = root_mem->last_scanned_child + 1; | |
1505 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1506 | &found); | |
1507 | if (css && css_tryget(css)) | |
1508 | ret = container_of(css, struct mem_cgroup, css); | |
1509 | ||
1510 | rcu_read_unlock(); | |
1511 | /* Updates scanning parameter */ | |
04046e1a KH |
1512 | if (!css) { |
1513 | /* this means start scan from ID:1 */ | |
1514 | root_mem->last_scanned_child = 0; | |
1515 | } else | |
1516 | root_mem->last_scanned_child = found; | |
04046e1a KH |
1517 | } |
1518 | ||
1519 | return ret; | |
1520 | } | |
1521 | ||
889976db YH |
1522 | #if MAX_NUMNODES > 1 |
1523 | ||
1524 | /* | |
1525 | * Always updating the nodemask is not very good - even if we have an empty | |
1526 | * list or the wrong list here, we can start from some node and traverse all | |
1527 | * nodes based on the zonelist. So update the list loosely once per 10 secs. | |
1528 | * | |
1529 | */ | |
1530 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *mem) | |
1531 | { | |
1532 | int nid; | |
1533 | ||
1534 | if (time_after(mem->next_scan_node_update, jiffies)) | |
1535 | return; | |
1536 | ||
1537 | mem->next_scan_node_update = jiffies + 10*HZ; | |
1538 | /* make a nodemask where this memcg uses memory from */ | |
1539 | mem->scan_nodes = node_states[N_HIGH_MEMORY]; | |
1540 | ||
1541 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1542 | ||
1543 | if (mem_cgroup_get_zonestat_node(mem, nid, LRU_INACTIVE_FILE) || | |
1544 | mem_cgroup_get_zonestat_node(mem, nid, LRU_ACTIVE_FILE)) | |
1545 | continue; | |
1546 | ||
1547 | if (total_swap_pages && | |
1548 | (mem_cgroup_get_zonestat_node(mem, nid, LRU_INACTIVE_ANON) || | |
1549 | mem_cgroup_get_zonestat_node(mem, nid, LRU_ACTIVE_ANON))) | |
1550 | continue; | |
1551 | node_clear(nid, mem->scan_nodes); | |
1552 | } | |
1553 | } | |
1554 | ||
1555 | /* | |
1556 | * Selecting a node where we start reclaim from. Because what we need is just | |
1557 | * reducing usage counter, start from anywhere is O,K. Considering | |
1558 | * memory reclaim from current node, there are pros. and cons. | |
1559 | * | |
1560 | * Freeing memory from current node means freeing memory from a node which | |
1561 | * we'll use or we've used. So, it may make LRU bad. And if several threads | |
1562 | * hit limits, it will see a contention on a node. But freeing from remote | |
1563 | * node means more costs for memory reclaim because of memory latency. | |
1564 | * | |
1565 | * Now, we use round-robin. Better algorithm is welcomed. | |
1566 | */ | |
1567 | int mem_cgroup_select_victim_node(struct mem_cgroup *mem) | |
1568 | { | |
1569 | int node; | |
1570 | ||
1571 | mem_cgroup_may_update_nodemask(mem); | |
1572 | node = mem->last_scanned_node; | |
1573 | ||
1574 | node = next_node(node, mem->scan_nodes); | |
1575 | if (node == MAX_NUMNODES) | |
1576 | node = first_node(mem->scan_nodes); | |
1577 | /* | |
1578 | * We call this when we hit limit, not when pages are added to LRU. | |
1579 | * No LRU may hold pages because all pages are UNEVICTABLE or | |
1580 | * memcg is too small and all pages are not on LRU. In that case, | |
1581 | * we use curret node. | |
1582 | */ | |
1583 | if (unlikely(node == MAX_NUMNODES)) | |
1584 | node = numa_node_id(); | |
1585 | ||
1586 | mem->last_scanned_node = node; | |
1587 | return node; | |
1588 | } | |
1589 | ||
1590 | #else | |
1591 | int mem_cgroup_select_victim_node(struct mem_cgroup *mem) | |
1592 | { | |
1593 | return 0; | |
1594 | } | |
1595 | #endif | |
1596 | ||
04046e1a KH |
1597 | /* |
1598 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1599 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1600 | * based on its position in the children list. | |
6d61ef40 BS |
1601 | * |
1602 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1603 | * |
1604 | * We give up and return to the caller when we visit root_mem twice. | |
1605 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1606 | * |
1607 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1608 | */ |
1609 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1610 | struct zone *zone, |
75822b44 | 1611 | gfp_t gfp_mask, |
0ae5e89c YH |
1612 | unsigned long reclaim_options, |
1613 | unsigned long *total_scanned) | |
6d61ef40 | 1614 | { |
04046e1a KH |
1615 | struct mem_cgroup *victim; |
1616 | int ret, total = 0; | |
1617 | int loop = 0; | |
75822b44 BS |
1618 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1619 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 | 1620 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
9d11ea9f | 1621 | unsigned long excess; |
0ae5e89c | 1622 | unsigned long nr_scanned; |
9d11ea9f JW |
1623 | |
1624 | excess = res_counter_soft_limit_excess(&root_mem->res) >> PAGE_SHIFT; | |
04046e1a | 1625 | |
22a668d7 KH |
1626 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1627 | if (root_mem->memsw_is_minimum) | |
1628 | noswap = true; | |
1629 | ||
4e416953 | 1630 | while (1) { |
04046e1a | 1631 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1632 | if (victim == root_mem) { |
04046e1a | 1633 | loop++; |
cdec2e42 KH |
1634 | if (loop >= 1) |
1635 | drain_all_stock_async(); | |
4e416953 BS |
1636 | if (loop >= 2) { |
1637 | /* | |
1638 | * If we have not been able to reclaim | |
1639 | * anything, it might because there are | |
1640 | * no reclaimable pages under this hierarchy | |
1641 | */ | |
1642 | if (!check_soft || !total) { | |
1643 | css_put(&victim->css); | |
1644 | break; | |
1645 | } | |
1646 | /* | |
25985edc | 1647 | * We want to do more targeted reclaim. |
4e416953 BS |
1648 | * excess >> 2 is not to excessive so as to |
1649 | * reclaim too much, nor too less that we keep | |
1650 | * coming back to reclaim from this cgroup | |
1651 | */ | |
1652 | if (total >= (excess >> 2) || | |
1653 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1654 | css_put(&victim->css); | |
1655 | break; | |
1656 | } | |
1657 | } | |
1658 | } | |
c62b1a3b | 1659 | if (!mem_cgroup_local_usage(victim)) { |
04046e1a KH |
1660 | /* this cgroup's local usage == 0 */ |
1661 | css_put(&victim->css); | |
6d61ef40 BS |
1662 | continue; |
1663 | } | |
04046e1a | 1664 | /* we use swappiness of local cgroup */ |
0ae5e89c | 1665 | if (check_soft) { |
4e416953 | 1666 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, |
0ae5e89c YH |
1667 | noswap, get_swappiness(victim), zone, |
1668 | &nr_scanned); | |
1669 | *total_scanned += nr_scanned; | |
1670 | } else | |
4e416953 BS |
1671 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, |
1672 | noswap, get_swappiness(victim)); | |
04046e1a | 1673 | css_put(&victim->css); |
81d39c20 KH |
1674 | /* |
1675 | * At shrinking usage, we can't check we should stop here or | |
1676 | * reclaim more. It's depends on callers. last_scanned_child | |
1677 | * will work enough for keeping fairness under tree. | |
1678 | */ | |
1679 | if (shrink) | |
1680 | return ret; | |
04046e1a | 1681 | total += ret; |
4e416953 | 1682 | if (check_soft) { |
9d11ea9f | 1683 | if (!res_counter_soft_limit_excess(&root_mem->res)) |
4e416953 | 1684 | return total; |
9d11ea9f | 1685 | } else if (mem_cgroup_margin(root_mem)) |
4fd14ebf | 1686 | return total; |
6d61ef40 | 1687 | } |
04046e1a | 1688 | return total; |
6d61ef40 BS |
1689 | } |
1690 | ||
867578cb KH |
1691 | /* |
1692 | * Check OOM-Killer is already running under our hierarchy. | |
1693 | * If someone is running, return false. | |
1694 | */ | |
1695 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) | |
1696 | { | |
7d74b06f KH |
1697 | int x, lock_count = 0; |
1698 | struct mem_cgroup *iter; | |
a636b327 | 1699 | |
7d74b06f KH |
1700 | for_each_mem_cgroup_tree(iter, mem) { |
1701 | x = atomic_inc_return(&iter->oom_lock); | |
1702 | lock_count = max(x, lock_count); | |
1703 | } | |
867578cb KH |
1704 | |
1705 | if (lock_count == 1) | |
1706 | return true; | |
1707 | return false; | |
a636b327 | 1708 | } |
0b7f569e | 1709 | |
7d74b06f | 1710 | static int mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
0b7f569e | 1711 | { |
7d74b06f KH |
1712 | struct mem_cgroup *iter; |
1713 | ||
867578cb KH |
1714 | /* |
1715 | * When a new child is created while the hierarchy is under oom, | |
1716 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1717 | * atomic_add_unless() here. | |
1718 | */ | |
7d74b06f KH |
1719 | for_each_mem_cgroup_tree(iter, mem) |
1720 | atomic_add_unless(&iter->oom_lock, -1, 0); | |
0b7f569e KH |
1721 | return 0; |
1722 | } | |
1723 | ||
867578cb KH |
1724 | |
1725 | static DEFINE_MUTEX(memcg_oom_mutex); | |
1726 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); | |
1727 | ||
dc98df5a KH |
1728 | struct oom_wait_info { |
1729 | struct mem_cgroup *mem; | |
1730 | wait_queue_t wait; | |
1731 | }; | |
1732 | ||
1733 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1734 | unsigned mode, int sync, void *arg) | |
1735 | { | |
1736 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg; | |
1737 | struct oom_wait_info *oom_wait_info; | |
1738 | ||
1739 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
1740 | ||
1741 | if (oom_wait_info->mem == wake_mem) | |
1742 | goto wakeup; | |
1743 | /* if no hierarchy, no match */ | |
1744 | if (!oom_wait_info->mem->use_hierarchy || !wake_mem->use_hierarchy) | |
1745 | return 0; | |
1746 | /* | |
1747 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1748 | * Then we can use css_is_ancestor without taking care of RCU. | |
1749 | */ | |
1750 | if (!css_is_ancestor(&oom_wait_info->mem->css, &wake_mem->css) && | |
1751 | !css_is_ancestor(&wake_mem->css, &oom_wait_info->mem->css)) | |
1752 | return 0; | |
1753 | ||
1754 | wakeup: | |
1755 | return autoremove_wake_function(wait, mode, sync, arg); | |
1756 | } | |
1757 | ||
1758 | static void memcg_wakeup_oom(struct mem_cgroup *mem) | |
1759 | { | |
1760 | /* for filtering, pass "mem" as argument. */ | |
1761 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); | |
1762 | } | |
1763 | ||
3c11ecf4 KH |
1764 | static void memcg_oom_recover(struct mem_cgroup *mem) |
1765 | { | |
2bd9bb20 | 1766 | if (mem && atomic_read(&mem->oom_lock)) |
3c11ecf4 KH |
1767 | memcg_wakeup_oom(mem); |
1768 | } | |
1769 | ||
867578cb KH |
1770 | /* |
1771 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1772 | */ | |
1773 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) | |
0b7f569e | 1774 | { |
dc98df5a | 1775 | struct oom_wait_info owait; |
3c11ecf4 | 1776 | bool locked, need_to_kill; |
867578cb | 1777 | |
dc98df5a KH |
1778 | owait.mem = mem; |
1779 | owait.wait.flags = 0; | |
1780 | owait.wait.func = memcg_oom_wake_function; | |
1781 | owait.wait.private = current; | |
1782 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1783 | need_to_kill = true; |
867578cb KH |
1784 | /* At first, try to OOM lock hierarchy under mem.*/ |
1785 | mutex_lock(&memcg_oom_mutex); | |
1786 | locked = mem_cgroup_oom_lock(mem); | |
1787 | /* | |
1788 | * Even if signal_pending(), we can't quit charge() loop without | |
1789 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1790 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1791 | */ | |
3c11ecf4 KH |
1792 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
1793 | if (!locked || mem->oom_kill_disable) | |
1794 | need_to_kill = false; | |
1795 | if (locked) | |
9490ff27 | 1796 | mem_cgroup_oom_notify(mem); |
867578cb KH |
1797 | mutex_unlock(&memcg_oom_mutex); |
1798 | ||
3c11ecf4 KH |
1799 | if (need_to_kill) { |
1800 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
867578cb | 1801 | mem_cgroup_out_of_memory(mem, mask); |
3c11ecf4 | 1802 | } else { |
867578cb | 1803 | schedule(); |
dc98df5a | 1804 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb KH |
1805 | } |
1806 | mutex_lock(&memcg_oom_mutex); | |
1807 | mem_cgroup_oom_unlock(mem); | |
dc98df5a | 1808 | memcg_wakeup_oom(mem); |
867578cb KH |
1809 | mutex_unlock(&memcg_oom_mutex); |
1810 | ||
1811 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) | |
1812 | return false; | |
1813 | /* Give chance to dying process */ | |
1814 | schedule_timeout(1); | |
1815 | return true; | |
0b7f569e KH |
1816 | } |
1817 | ||
d69b042f BS |
1818 | /* |
1819 | * Currently used to update mapped file statistics, but the routine can be | |
1820 | * generalized to update other statistics as well. | |
32047e2a KH |
1821 | * |
1822 | * Notes: Race condition | |
1823 | * | |
1824 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1825 | * it tends to be costly. But considering some conditions, we doesn't need | |
1826 | * to do so _always_. | |
1827 | * | |
1828 | * Considering "charge", lock_page_cgroup() is not required because all | |
1829 | * file-stat operations happen after a page is attached to radix-tree. There | |
1830 | * are no race with "charge". | |
1831 | * | |
1832 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1833 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1834 | * if there are race with "uncharge". Statistics itself is properly handled | |
1835 | * by flags. | |
1836 | * | |
1837 | * Considering "move", this is an only case we see a race. To make the race | |
1838 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
1839 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 1840 | */ |
26174efd | 1841 | |
2a7106f2 GT |
1842 | void mem_cgroup_update_page_stat(struct page *page, |
1843 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f BS |
1844 | { |
1845 | struct mem_cgroup *mem; | |
32047e2a KH |
1846 | struct page_cgroup *pc = lookup_page_cgroup(page); |
1847 | bool need_unlock = false; | |
dbd4ea78 | 1848 | unsigned long uninitialized_var(flags); |
d69b042f | 1849 | |
d69b042f BS |
1850 | if (unlikely(!pc)) |
1851 | return; | |
1852 | ||
32047e2a | 1853 | rcu_read_lock(); |
d69b042f | 1854 | mem = pc->mem_cgroup; |
32047e2a KH |
1855 | if (unlikely(!mem || !PageCgroupUsed(pc))) |
1856 | goto out; | |
1857 | /* pc->mem_cgroup is unstable ? */ | |
ca3e0214 | 1858 | if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) { |
32047e2a | 1859 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 1860 | move_lock_page_cgroup(pc, &flags); |
32047e2a KH |
1861 | need_unlock = true; |
1862 | mem = pc->mem_cgroup; | |
1863 | if (!mem || !PageCgroupUsed(pc)) | |
1864 | goto out; | |
1865 | } | |
26174efd | 1866 | |
26174efd | 1867 | switch (idx) { |
2a7106f2 | 1868 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
1869 | if (val > 0) |
1870 | SetPageCgroupFileMapped(pc); | |
1871 | else if (!page_mapped(page)) | |
0c270f8f | 1872 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 1873 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1874 | break; |
1875 | default: | |
1876 | BUG(); | |
8725d541 | 1877 | } |
d69b042f | 1878 | |
2a7106f2 GT |
1879 | this_cpu_add(mem->stat->count[idx], val); |
1880 | ||
32047e2a KH |
1881 | out: |
1882 | if (unlikely(need_unlock)) | |
dbd4ea78 | 1883 | move_unlock_page_cgroup(pc, &flags); |
32047e2a KH |
1884 | rcu_read_unlock(); |
1885 | return; | |
d69b042f | 1886 | } |
2a7106f2 | 1887 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
26174efd | 1888 | |
cdec2e42 KH |
1889 | /* |
1890 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1891 | * TODO: maybe necessary to use big numbers in big irons. | |
1892 | */ | |
7ec99d62 | 1893 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1894 | struct memcg_stock_pcp { |
1895 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1896 | unsigned int nr_pages; |
cdec2e42 KH |
1897 | struct work_struct work; |
1898 | }; | |
1899 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1900 | static atomic_t memcg_drain_count; | |
1901 | ||
1902 | /* | |
11c9ea4e | 1903 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
1904 | * from local stock and true is returned. If the stock is 0 or charges from a |
1905 | * cgroup which is not current target, returns false. This stock will be | |
1906 | * refilled. | |
1907 | */ | |
1908 | static bool consume_stock(struct mem_cgroup *mem) | |
1909 | { | |
1910 | struct memcg_stock_pcp *stock; | |
1911 | bool ret = true; | |
1912 | ||
1913 | stock = &get_cpu_var(memcg_stock); | |
11c9ea4e JW |
1914 | if (mem == stock->cached && stock->nr_pages) |
1915 | stock->nr_pages--; | |
cdec2e42 KH |
1916 | else /* need to call res_counter_charge */ |
1917 | ret = false; | |
1918 | put_cpu_var(memcg_stock); | |
1919 | return ret; | |
1920 | } | |
1921 | ||
1922 | /* | |
1923 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1924 | */ | |
1925 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1926 | { | |
1927 | struct mem_cgroup *old = stock->cached; | |
1928 | ||
11c9ea4e JW |
1929 | if (stock->nr_pages) { |
1930 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
1931 | ||
1932 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 1933 | if (do_swap_account) |
11c9ea4e JW |
1934 | res_counter_uncharge(&old->memsw, bytes); |
1935 | stock->nr_pages = 0; | |
cdec2e42 KH |
1936 | } |
1937 | stock->cached = NULL; | |
cdec2e42 KH |
1938 | } |
1939 | ||
1940 | /* | |
1941 | * This must be called under preempt disabled or must be called by | |
1942 | * a thread which is pinned to local cpu. | |
1943 | */ | |
1944 | static void drain_local_stock(struct work_struct *dummy) | |
1945 | { | |
1946 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1947 | drain_stock(stock); | |
1948 | } | |
1949 | ||
1950 | /* | |
1951 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 1952 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 1953 | */ |
11c9ea4e | 1954 | static void refill_stock(struct mem_cgroup *mem, unsigned int nr_pages) |
cdec2e42 KH |
1955 | { |
1956 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1957 | ||
1958 | if (stock->cached != mem) { /* reset if necessary */ | |
1959 | drain_stock(stock); | |
1960 | stock->cached = mem; | |
1961 | } | |
11c9ea4e | 1962 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
1963 | put_cpu_var(memcg_stock); |
1964 | } | |
1965 | ||
1966 | /* | |
1967 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1968 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1969 | * expects some charges will be back to res_counter later but cannot wait for | |
1970 | * it. | |
1971 | */ | |
1972 | static void drain_all_stock_async(void) | |
1973 | { | |
1974 | int cpu; | |
1975 | /* This function is for scheduling "drain" in asynchronous way. | |
1976 | * The result of "drain" is not directly handled by callers. Then, | |
1977 | * if someone is calling drain, we don't have to call drain more. | |
1978 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1979 | * there is a race. We just do loose check here. | |
1980 | */ | |
1981 | if (atomic_read(&memcg_drain_count)) | |
1982 | return; | |
1983 | /* Notify other cpus that system-wide "drain" is running */ | |
1984 | atomic_inc(&memcg_drain_count); | |
1985 | get_online_cpus(); | |
1986 | for_each_online_cpu(cpu) { | |
1987 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1988 | schedule_work_on(cpu, &stock->work); | |
1989 | } | |
1990 | put_online_cpus(); | |
1991 | atomic_dec(&memcg_drain_count); | |
1992 | /* We don't wait for flush_work */ | |
1993 | } | |
1994 | ||
1995 | /* This is a synchronous drain interface. */ | |
1996 | static void drain_all_stock_sync(void) | |
1997 | { | |
1998 | /* called when force_empty is called */ | |
1999 | atomic_inc(&memcg_drain_count); | |
2000 | schedule_on_each_cpu(drain_local_stock); | |
2001 | atomic_dec(&memcg_drain_count); | |
2002 | } | |
2003 | ||
711d3d2c KH |
2004 | /* |
2005 | * This function drains percpu counter value from DEAD cpu and | |
2006 | * move it to local cpu. Note that this function can be preempted. | |
2007 | */ | |
2008 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu) | |
2009 | { | |
2010 | int i; | |
2011 | ||
2012 | spin_lock(&mem->pcp_counter_lock); | |
2013 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { | |
7a159cc9 | 2014 | long x = per_cpu(mem->stat->count[i], cpu); |
711d3d2c KH |
2015 | |
2016 | per_cpu(mem->stat->count[i], cpu) = 0; | |
2017 | mem->nocpu_base.count[i] += x; | |
2018 | } | |
e9f8974f JW |
2019 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
2020 | unsigned long x = per_cpu(mem->stat->events[i], cpu); | |
2021 | ||
2022 | per_cpu(mem->stat->events[i], cpu) = 0; | |
2023 | mem->nocpu_base.events[i] += x; | |
2024 | } | |
1489ebad KH |
2025 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
2026 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; | |
2027 | spin_unlock(&mem->pcp_counter_lock); | |
2028 | } | |
2029 | ||
2030 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu) | |
2031 | { | |
2032 | int idx = MEM_CGROUP_ON_MOVE; | |
2033 | ||
2034 | spin_lock(&mem->pcp_counter_lock); | |
2035 | per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx]; | |
711d3d2c KH |
2036 | spin_unlock(&mem->pcp_counter_lock); |
2037 | } | |
2038 | ||
2039 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2040 | unsigned long action, |
2041 | void *hcpu) | |
2042 | { | |
2043 | int cpu = (unsigned long)hcpu; | |
2044 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2045 | struct mem_cgroup *iter; |
cdec2e42 | 2046 | |
1489ebad KH |
2047 | if ((action == CPU_ONLINE)) { |
2048 | for_each_mem_cgroup_all(iter) | |
2049 | synchronize_mem_cgroup_on_move(iter, cpu); | |
2050 | return NOTIFY_OK; | |
2051 | } | |
2052 | ||
711d3d2c | 2053 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 2054 | return NOTIFY_OK; |
711d3d2c KH |
2055 | |
2056 | for_each_mem_cgroup_all(iter) | |
2057 | mem_cgroup_drain_pcp_counter(iter, cpu); | |
2058 | ||
cdec2e42 KH |
2059 | stock = &per_cpu(memcg_stock, cpu); |
2060 | drain_stock(stock); | |
2061 | return NOTIFY_OK; | |
2062 | } | |
2063 | ||
4b534334 KH |
2064 | |
2065 | /* See __mem_cgroup_try_charge() for details */ | |
2066 | enum { | |
2067 | CHARGE_OK, /* success */ | |
2068 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2069 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2070 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2071 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2072 | }; | |
2073 | ||
7ec99d62 JW |
2074 | static int mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask, |
2075 | unsigned int nr_pages, bool oom_check) | |
4b534334 | 2076 | { |
7ec99d62 | 2077 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2078 | struct mem_cgroup *mem_over_limit; |
2079 | struct res_counter *fail_res; | |
2080 | unsigned long flags = 0; | |
2081 | int ret; | |
2082 | ||
2083 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
2084 | ||
2085 | if (likely(!ret)) { | |
2086 | if (!do_swap_account) | |
2087 | return CHARGE_OK; | |
2088 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); | |
2089 | if (likely(!ret)) | |
2090 | return CHARGE_OK; | |
2091 | ||
01c88e2d | 2092 | res_counter_uncharge(&mem->res, csize); |
4b534334 KH |
2093 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2094 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2095 | } else | |
2096 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2097 | /* |
7ec99d62 JW |
2098 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2099 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2100 | * |
2101 | * Never reclaim on behalf of optional batching, retry with a | |
2102 | * single page instead. | |
2103 | */ | |
7ec99d62 | 2104 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2105 | return CHARGE_RETRY; |
2106 | ||
2107 | if (!(gfp_mask & __GFP_WAIT)) | |
2108 | return CHARGE_WOULDBLOCK; | |
2109 | ||
2110 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, | |
0ae5e89c | 2111 | gfp_mask, flags, NULL); |
7ec99d62 | 2112 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2113 | return CHARGE_RETRY; |
4b534334 | 2114 | /* |
19942822 JW |
2115 | * Even though the limit is exceeded at this point, reclaim |
2116 | * may have been able to free some pages. Retry the charge | |
2117 | * before killing the task. | |
2118 | * | |
2119 | * Only for regular pages, though: huge pages are rather | |
2120 | * unlikely to succeed so close to the limit, and we fall back | |
2121 | * to regular pages anyway in case of failure. | |
4b534334 | 2122 | */ |
7ec99d62 | 2123 | if (nr_pages == 1 && ret) |
4b534334 KH |
2124 | return CHARGE_RETRY; |
2125 | ||
2126 | /* | |
2127 | * At task move, charge accounts can be doubly counted. So, it's | |
2128 | * better to wait until the end of task_move if something is going on. | |
2129 | */ | |
2130 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2131 | return CHARGE_RETRY; | |
2132 | ||
2133 | /* If we don't need to call oom-killer at el, return immediately */ | |
2134 | if (!oom_check) | |
2135 | return CHARGE_NOMEM; | |
2136 | /* check OOM */ | |
2137 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) | |
2138 | return CHARGE_OOM_DIE; | |
2139 | ||
2140 | return CHARGE_RETRY; | |
2141 | } | |
2142 | ||
f817ed48 KH |
2143 | /* |
2144 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
2145 | * oom-killer can be invoked. | |
8a9f3ccd | 2146 | */ |
f817ed48 | 2147 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2148 | gfp_t gfp_mask, |
7ec99d62 JW |
2149 | unsigned int nr_pages, |
2150 | struct mem_cgroup **memcg, | |
2151 | bool oom) | |
8a9f3ccd | 2152 | { |
7ec99d62 | 2153 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 KH |
2154 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
2155 | struct mem_cgroup *mem = NULL; | |
2156 | int ret; | |
a636b327 | 2157 | |
867578cb KH |
2158 | /* |
2159 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2160 | * in system level. So, allow to go ahead dying process in addition to | |
2161 | * MEMDIE process. | |
2162 | */ | |
2163 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2164 | || fatal_signal_pending(current))) | |
2165 | goto bypass; | |
a636b327 | 2166 | |
8a9f3ccd | 2167 | /* |
3be91277 HD |
2168 | * We always charge the cgroup the mm_struct belongs to. |
2169 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2170 | * thread group leader migrates. It's possible that mm is not |
2171 | * set, if so charge the init_mm (happens for pagecache usage). | |
2172 | */ | |
f75ca962 KH |
2173 | if (!*memcg && !mm) |
2174 | goto bypass; | |
2175 | again: | |
2176 | if (*memcg) { /* css should be a valid one */ | |
4b534334 | 2177 | mem = *memcg; |
f75ca962 KH |
2178 | VM_BUG_ON(css_is_removed(&mem->css)); |
2179 | if (mem_cgroup_is_root(mem)) | |
2180 | goto done; | |
7ec99d62 | 2181 | if (nr_pages == 1 && consume_stock(mem)) |
f75ca962 | 2182 | goto done; |
4b534334 KH |
2183 | css_get(&mem->css); |
2184 | } else { | |
f75ca962 | 2185 | struct task_struct *p; |
54595fe2 | 2186 | |
f75ca962 KH |
2187 | rcu_read_lock(); |
2188 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2189 | /* |
ebb76ce1 KH |
2190 | * Because we don't have task_lock(), "p" can exit. |
2191 | * In that case, "mem" can point to root or p can be NULL with | |
2192 | * race with swapoff. Then, we have small risk of mis-accouning. | |
2193 | * But such kind of mis-account by race always happens because | |
2194 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2195 | * small race, here. | |
2196 | * (*) swapoff at el will charge against mm-struct not against | |
2197 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 KH |
2198 | */ |
2199 | mem = mem_cgroup_from_task(p); | |
ebb76ce1 | 2200 | if (!mem || mem_cgroup_is_root(mem)) { |
f75ca962 KH |
2201 | rcu_read_unlock(); |
2202 | goto done; | |
2203 | } | |
7ec99d62 | 2204 | if (nr_pages == 1 && consume_stock(mem)) { |
f75ca962 KH |
2205 | /* |
2206 | * It seems dagerous to access memcg without css_get(). | |
2207 | * But considering how consume_stok works, it's not | |
2208 | * necessary. If consume_stock success, some charges | |
2209 | * from this memcg are cached on this cpu. So, we | |
2210 | * don't need to call css_get()/css_tryget() before | |
2211 | * calling consume_stock(). | |
2212 | */ | |
2213 | rcu_read_unlock(); | |
2214 | goto done; | |
2215 | } | |
2216 | /* after here, we may be blocked. we need to get refcnt */ | |
2217 | if (!css_tryget(&mem->css)) { | |
2218 | rcu_read_unlock(); | |
2219 | goto again; | |
2220 | } | |
2221 | rcu_read_unlock(); | |
2222 | } | |
8a9f3ccd | 2223 | |
4b534334 KH |
2224 | do { |
2225 | bool oom_check; | |
7a81b88c | 2226 | |
4b534334 | 2227 | /* If killed, bypass charge */ |
f75ca962 KH |
2228 | if (fatal_signal_pending(current)) { |
2229 | css_put(&mem->css); | |
4b534334 | 2230 | goto bypass; |
f75ca962 | 2231 | } |
6d61ef40 | 2232 | |
4b534334 KH |
2233 | oom_check = false; |
2234 | if (oom && !nr_oom_retries) { | |
2235 | oom_check = true; | |
2236 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2237 | } |
66e1707b | 2238 | |
7ec99d62 | 2239 | ret = mem_cgroup_do_charge(mem, gfp_mask, batch, oom_check); |
4b534334 KH |
2240 | switch (ret) { |
2241 | case CHARGE_OK: | |
2242 | break; | |
2243 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2244 | batch = nr_pages; |
f75ca962 KH |
2245 | css_put(&mem->css); |
2246 | mem = NULL; | |
2247 | goto again; | |
4b534334 | 2248 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
f75ca962 | 2249 | css_put(&mem->css); |
4b534334 KH |
2250 | goto nomem; |
2251 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 KH |
2252 | if (!oom) { |
2253 | css_put(&mem->css); | |
867578cb | 2254 | goto nomem; |
f75ca962 | 2255 | } |
4b534334 KH |
2256 | /* If oom, we never return -ENOMEM */ |
2257 | nr_oom_retries--; | |
2258 | break; | |
2259 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
f75ca962 | 2260 | css_put(&mem->css); |
867578cb | 2261 | goto bypass; |
66e1707b | 2262 | } |
4b534334 KH |
2263 | } while (ret != CHARGE_OK); |
2264 | ||
7ec99d62 JW |
2265 | if (batch > nr_pages) |
2266 | refill_stock(mem, batch - nr_pages); | |
f75ca962 | 2267 | css_put(&mem->css); |
0c3e73e8 | 2268 | done: |
f75ca962 | 2269 | *memcg = mem; |
7a81b88c KH |
2270 | return 0; |
2271 | nomem: | |
f75ca962 | 2272 | *memcg = NULL; |
7a81b88c | 2273 | return -ENOMEM; |
867578cb KH |
2274 | bypass: |
2275 | *memcg = NULL; | |
2276 | return 0; | |
7a81b88c | 2277 | } |
8a9f3ccd | 2278 | |
a3032a2c DN |
2279 | /* |
2280 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2281 | * This function is for that and do uncharge, put css's refcnt. | |
2282 | * gotten by try_charge(). | |
2283 | */ | |
854ffa8d | 2284 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
e7018b8d | 2285 | unsigned int nr_pages) |
a3032a2c DN |
2286 | { |
2287 | if (!mem_cgroup_is_root(mem)) { | |
e7018b8d JW |
2288 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2289 | ||
2290 | res_counter_uncharge(&mem->res, bytes); | |
a3032a2c | 2291 | if (do_swap_account) |
e7018b8d | 2292 | res_counter_uncharge(&mem->memsw, bytes); |
a3032a2c | 2293 | } |
854ffa8d DN |
2294 | } |
2295 | ||
a3b2d692 KH |
2296 | /* |
2297 | * A helper function to get mem_cgroup from ID. must be called under | |
2298 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2299 | * it's concern. (dropping refcnt from swap can be called against removed | |
2300 | * memcg.) | |
2301 | */ | |
2302 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2303 | { | |
2304 | struct cgroup_subsys_state *css; | |
2305 | ||
2306 | /* ID 0 is unused ID */ | |
2307 | if (!id) | |
2308 | return NULL; | |
2309 | css = css_lookup(&mem_cgroup_subsys, id); | |
2310 | if (!css) | |
2311 | return NULL; | |
2312 | return container_of(css, struct mem_cgroup, css); | |
2313 | } | |
2314 | ||
e42d9d5d | 2315 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2316 | { |
e42d9d5d | 2317 | struct mem_cgroup *mem = NULL; |
3c776e64 | 2318 | struct page_cgroup *pc; |
a3b2d692 | 2319 | unsigned short id; |
b5a84319 KH |
2320 | swp_entry_t ent; |
2321 | ||
3c776e64 DN |
2322 | VM_BUG_ON(!PageLocked(page)); |
2323 | ||
3c776e64 | 2324 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2325 | lock_page_cgroup(pc); |
a3b2d692 | 2326 | if (PageCgroupUsed(pc)) { |
3c776e64 | 2327 | mem = pc->mem_cgroup; |
a3b2d692 KH |
2328 | if (mem && !css_tryget(&mem->css)) |
2329 | mem = NULL; | |
e42d9d5d | 2330 | } else if (PageSwapCache(page)) { |
3c776e64 | 2331 | ent.val = page_private(page); |
a3b2d692 KH |
2332 | id = lookup_swap_cgroup(ent); |
2333 | rcu_read_lock(); | |
2334 | mem = mem_cgroup_lookup(id); | |
2335 | if (mem && !css_tryget(&mem->css)) | |
2336 | mem = NULL; | |
2337 | rcu_read_unlock(); | |
3c776e64 | 2338 | } |
c0bd3f63 | 2339 | unlock_page_cgroup(pc); |
b5a84319 KH |
2340 | return mem; |
2341 | } | |
2342 | ||
ca3e0214 | 2343 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, |
5564e88b | 2344 | struct page *page, |
7ec99d62 | 2345 | unsigned int nr_pages, |
ca3e0214 | 2346 | struct page_cgroup *pc, |
7ec99d62 | 2347 | enum charge_type ctype) |
7a81b88c | 2348 | { |
ca3e0214 KH |
2349 | lock_page_cgroup(pc); |
2350 | if (unlikely(PageCgroupUsed(pc))) { | |
2351 | unlock_page_cgroup(pc); | |
e7018b8d | 2352 | __mem_cgroup_cancel_charge(mem, nr_pages); |
ca3e0214 KH |
2353 | return; |
2354 | } | |
2355 | /* | |
2356 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2357 | * accessed by any other context at this point. | |
2358 | */ | |
8a9f3ccd | 2359 | pc->mem_cgroup = mem; |
261fb61a KH |
2360 | /* |
2361 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2362 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2363 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2364 | * before USED bit, we need memory barrier here. | |
2365 | * See mem_cgroup_add_lru_list(), etc. | |
2366 | */ | |
08e552c6 | 2367 | smp_wmb(); |
4b3bde4c BS |
2368 | switch (ctype) { |
2369 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
2370 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
2371 | SetPageCgroupCache(pc); | |
2372 | SetPageCgroupUsed(pc); | |
2373 | break; | |
2374 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2375 | ClearPageCgroupCache(pc); | |
2376 | SetPageCgroupUsed(pc); | |
2377 | break; | |
2378 | default: | |
2379 | break; | |
2380 | } | |
3be91277 | 2381 | |
ca3e0214 | 2382 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages); |
52d4b9ac | 2383 | unlock_page_cgroup(pc); |
430e4863 KH |
2384 | /* |
2385 | * "charge_statistics" updated event counter. Then, check it. | |
2386 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2387 | * if they exceeds softlimit. | |
2388 | */ | |
5564e88b | 2389 | memcg_check_events(mem, page); |
7a81b88c | 2390 | } |
66e1707b | 2391 | |
ca3e0214 KH |
2392 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2393 | ||
2394 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
2395 | (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION)) | |
2396 | /* | |
2397 | * Because tail pages are not marked as "used", set it. We're under | |
2398 | * zone->lru_lock, 'splitting on pmd' and compund_lock. | |
2399 | */ | |
2400 | void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail) | |
2401 | { | |
2402 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
2403 | struct page_cgroup *tail_pc = lookup_page_cgroup(tail); | |
2404 | unsigned long flags; | |
2405 | ||
3d37c4a9 KH |
2406 | if (mem_cgroup_disabled()) |
2407 | return; | |
ca3e0214 | 2408 | /* |
ece35ca8 | 2409 | * We have no races with charge/uncharge but will have races with |
ca3e0214 KH |
2410 | * page state accounting. |
2411 | */ | |
2412 | move_lock_page_cgroup(head_pc, &flags); | |
2413 | ||
2414 | tail_pc->mem_cgroup = head_pc->mem_cgroup; | |
2415 | smp_wmb(); /* see __commit_charge() */ | |
ece35ca8 KH |
2416 | if (PageCgroupAcctLRU(head_pc)) { |
2417 | enum lru_list lru; | |
2418 | struct mem_cgroup_per_zone *mz; | |
2419 | ||
2420 | /* | |
2421 | * LRU flags cannot be copied because we need to add tail | |
2422 | *.page to LRU by generic call and our hook will be called. | |
2423 | * We hold lru_lock, then, reduce counter directly. | |
2424 | */ | |
2425 | lru = page_lru(head); | |
97a6c37b | 2426 | mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head); |
ece35ca8 KH |
2427 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
2428 | } | |
ca3e0214 KH |
2429 | tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2430 | move_unlock_page_cgroup(head_pc, &flags); | |
2431 | } | |
2432 | #endif | |
2433 | ||
f817ed48 | 2434 | /** |
de3638d9 | 2435 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2436 | * @page: the page |
7ec99d62 | 2437 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2438 | * @pc: page_cgroup of the page. |
2439 | * @from: mem_cgroup which the page is moved from. | |
2440 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2441 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2442 | * |
2443 | * The caller must confirm following. | |
08e552c6 | 2444 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2445 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2446 | * |
854ffa8d | 2447 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
25985edc | 2448 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
854ffa8d DN |
2449 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
2450 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2451 | */ |
7ec99d62 JW |
2452 | static int mem_cgroup_move_account(struct page *page, |
2453 | unsigned int nr_pages, | |
2454 | struct page_cgroup *pc, | |
2455 | struct mem_cgroup *from, | |
2456 | struct mem_cgroup *to, | |
2457 | bool uncharge) | |
f817ed48 | 2458 | { |
de3638d9 JW |
2459 | unsigned long flags; |
2460 | int ret; | |
987eba66 | 2461 | |
f817ed48 | 2462 | VM_BUG_ON(from == to); |
5564e88b | 2463 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2464 | /* |
2465 | * The page is isolated from LRU. So, collapse function | |
2466 | * will not handle this page. But page splitting can happen. | |
2467 | * Do this check under compound_page_lock(). The caller should | |
2468 | * hold it. | |
2469 | */ | |
2470 | ret = -EBUSY; | |
7ec99d62 | 2471 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2472 | goto out; |
2473 | ||
2474 | lock_page_cgroup(pc); | |
2475 | ||
2476 | ret = -EINVAL; | |
2477 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2478 | goto unlock; | |
2479 | ||
2480 | move_lock_page_cgroup(pc, &flags); | |
f817ed48 | 2481 | |
8725d541 | 2482 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2483 | /* Update mapped_file data for mem_cgroup */ |
2484 | preempt_disable(); | |
2485 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2486 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2487 | preempt_enable(); | |
d69b042f | 2488 | } |
987eba66 | 2489 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages); |
854ffa8d DN |
2490 | if (uncharge) |
2491 | /* This is not "cancel", but cancel_charge does all we need. */ | |
e7018b8d | 2492 | __mem_cgroup_cancel_charge(from, nr_pages); |
d69b042f | 2493 | |
854ffa8d | 2494 | /* caller should have done css_get */ |
08e552c6 | 2495 | pc->mem_cgroup = to; |
987eba66 | 2496 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages); |
88703267 KH |
2497 | /* |
2498 | * We charges against "to" which may not have any tasks. Then, "to" | |
2499 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2500 | * this function is just force_empty() and move charge, so it's |
25985edc | 2501 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2502 | * status here. |
88703267 | 2503 | */ |
de3638d9 JW |
2504 | move_unlock_page_cgroup(pc, &flags); |
2505 | ret = 0; | |
2506 | unlock: | |
57f9fd7d | 2507 | unlock_page_cgroup(pc); |
d2265e6f KH |
2508 | /* |
2509 | * check events | |
2510 | */ | |
5564e88b JW |
2511 | memcg_check_events(to, page); |
2512 | memcg_check_events(from, page); | |
de3638d9 | 2513 | out: |
f817ed48 KH |
2514 | return ret; |
2515 | } | |
2516 | ||
2517 | /* | |
2518 | * move charges to its parent. | |
2519 | */ | |
2520 | ||
5564e88b JW |
2521 | static int mem_cgroup_move_parent(struct page *page, |
2522 | struct page_cgroup *pc, | |
f817ed48 KH |
2523 | struct mem_cgroup *child, |
2524 | gfp_t gfp_mask) | |
2525 | { | |
2526 | struct cgroup *cg = child->css.cgroup; | |
2527 | struct cgroup *pcg = cg->parent; | |
2528 | struct mem_cgroup *parent; | |
7ec99d62 | 2529 | unsigned int nr_pages; |
4be4489f | 2530 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2531 | int ret; |
2532 | ||
2533 | /* Is ROOT ? */ | |
2534 | if (!pcg) | |
2535 | return -EINVAL; | |
2536 | ||
57f9fd7d DN |
2537 | ret = -EBUSY; |
2538 | if (!get_page_unless_zero(page)) | |
2539 | goto out; | |
2540 | if (isolate_lru_page(page)) | |
2541 | goto put; | |
52dbb905 | 2542 | |
7ec99d62 | 2543 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2544 | |
f817ed48 | 2545 | parent = mem_cgroup_from_cont(pcg); |
7ec99d62 | 2546 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
a636b327 | 2547 | if (ret || !parent) |
57f9fd7d | 2548 | goto put_back; |
f817ed48 | 2549 | |
7ec99d62 | 2550 | if (nr_pages > 1) |
987eba66 KH |
2551 | flags = compound_lock_irqsave(page); |
2552 | ||
7ec99d62 | 2553 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
854ffa8d | 2554 | if (ret) |
7ec99d62 | 2555 | __mem_cgroup_cancel_charge(parent, nr_pages); |
8dba474f | 2556 | |
7ec99d62 | 2557 | if (nr_pages > 1) |
987eba66 | 2558 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2559 | put_back: |
08e552c6 | 2560 | putback_lru_page(page); |
57f9fd7d | 2561 | put: |
40d58138 | 2562 | put_page(page); |
57f9fd7d | 2563 | out: |
f817ed48 KH |
2564 | return ret; |
2565 | } | |
2566 | ||
7a81b88c KH |
2567 | /* |
2568 | * Charge the memory controller for page usage. | |
2569 | * Return | |
2570 | * 0 if the charge was successful | |
2571 | * < 0 if the cgroup is over its limit | |
2572 | */ | |
2573 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2574 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2575 | { |
73045c47 | 2576 | struct mem_cgroup *mem = NULL; |
7ec99d62 | 2577 | unsigned int nr_pages = 1; |
7a81b88c | 2578 | struct page_cgroup *pc; |
8493ae43 | 2579 | bool oom = true; |
7a81b88c | 2580 | int ret; |
ec168510 | 2581 | |
37c2ac78 | 2582 | if (PageTransHuge(page)) { |
7ec99d62 | 2583 | nr_pages <<= compound_order(page); |
37c2ac78 | 2584 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2585 | /* |
2586 | * Never OOM-kill a process for a huge page. The | |
2587 | * fault handler will fall back to regular pages. | |
2588 | */ | |
2589 | oom = false; | |
37c2ac78 | 2590 | } |
7a81b88c KH |
2591 | |
2592 | pc = lookup_page_cgroup(page); | |
af4a6621 | 2593 | BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */ |
7a81b88c | 2594 | |
7ec99d62 | 2595 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &mem, oom); |
a636b327 | 2596 | if (ret || !mem) |
7a81b88c KH |
2597 | return ret; |
2598 | ||
7ec99d62 | 2599 | __mem_cgroup_commit_charge(mem, page, nr_pages, pc, ctype); |
8a9f3ccd | 2600 | return 0; |
8a9f3ccd BS |
2601 | } |
2602 | ||
7a81b88c KH |
2603 | int mem_cgroup_newpage_charge(struct page *page, |
2604 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2605 | { |
f8d66542 | 2606 | if (mem_cgroup_disabled()) |
cede86ac | 2607 | return 0; |
69029cd5 KH |
2608 | /* |
2609 | * If already mapped, we don't have to account. | |
2610 | * If page cache, page->mapping has address_space. | |
2611 | * But page->mapping may have out-of-use anon_vma pointer, | |
2612 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
2613 | * is NULL. | |
2614 | */ | |
2615 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2616 | return 0; | |
2617 | if (unlikely(!mm)) | |
2618 | mm = &init_mm; | |
217bc319 | 2619 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
73045c47 | 2620 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2621 | } |
2622 | ||
83aae4c7 DN |
2623 | static void |
2624 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2625 | enum charge_type ctype); | |
2626 | ||
5a6475a4 KH |
2627 | static void |
2628 | __mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *mem, | |
2629 | enum charge_type ctype) | |
2630 | { | |
2631 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2632 | /* | |
2633 | * In some case, SwapCache, FUSE(splice_buf->radixtree), the page | |
2634 | * is already on LRU. It means the page may on some other page_cgroup's | |
2635 | * LRU. Take care of it. | |
2636 | */ | |
2637 | mem_cgroup_lru_del_before_commit(page); | |
2638 | __mem_cgroup_commit_charge(mem, page, 1, pc, ctype); | |
2639 | mem_cgroup_lru_add_after_commit(page); | |
2640 | return; | |
2641 | } | |
2642 | ||
e1a1cd59 BS |
2643 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2644 | gfp_t gfp_mask) | |
8697d331 | 2645 | { |
5a6475a4 | 2646 | struct mem_cgroup *mem = NULL; |
b5a84319 KH |
2647 | int ret; |
2648 | ||
f8d66542 | 2649 | if (mem_cgroup_disabled()) |
cede86ac | 2650 | return 0; |
52d4b9ac KH |
2651 | if (PageCompound(page)) |
2652 | return 0; | |
accf163e KH |
2653 | /* |
2654 | * Corner case handling. This is called from add_to_page_cache() | |
2655 | * in usual. But some FS (shmem) precharges this page before calling it | |
2656 | * and call add_to_page_cache() with GFP_NOWAIT. | |
2657 | * | |
2658 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
2659 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
2660 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
2661 | * And when the page is SwapCache, it should take swap information |
2662 | * into account. This is under lock_page() now. | |
accf163e KH |
2663 | */ |
2664 | if (!(gfp_mask & __GFP_WAIT)) { | |
2665 | struct page_cgroup *pc; | |
2666 | ||
52d4b9ac KH |
2667 | pc = lookup_page_cgroup(page); |
2668 | if (!pc) | |
2669 | return 0; | |
2670 | lock_page_cgroup(pc); | |
2671 | if (PageCgroupUsed(pc)) { | |
2672 | unlock_page_cgroup(pc); | |
accf163e KH |
2673 | return 0; |
2674 | } | |
52d4b9ac | 2675 | unlock_page_cgroup(pc); |
accf163e KH |
2676 | } |
2677 | ||
73045c47 | 2678 | if (unlikely(!mm)) |
8697d331 | 2679 | mm = &init_mm; |
accf163e | 2680 | |
5a6475a4 KH |
2681 | if (page_is_file_cache(page)) { |
2682 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &mem, true); | |
2683 | if (ret || !mem) | |
2684 | return ret; | |
b5a84319 | 2685 | |
5a6475a4 KH |
2686 | /* |
2687 | * FUSE reuses pages without going through the final | |
2688 | * put that would remove them from the LRU list, make | |
2689 | * sure that they get relinked properly. | |
2690 | */ | |
2691 | __mem_cgroup_commit_charge_lrucare(page, mem, | |
2692 | MEM_CGROUP_CHARGE_TYPE_CACHE); | |
2693 | return ret; | |
2694 | } | |
83aae4c7 DN |
2695 | /* shmem */ |
2696 | if (PageSwapCache(page)) { | |
2697 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
2698 | if (!ret) | |
2699 | __mem_cgroup_commit_charge_swapin(page, mem, | |
2700 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
2701 | } else | |
2702 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2703 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
b5a84319 | 2704 | |
b5a84319 | 2705 | return ret; |
e8589cc1 KH |
2706 | } |
2707 | ||
54595fe2 KH |
2708 | /* |
2709 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2710 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2711 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2712 | * "commit()" or removed by "cancel()" |
2713 | */ | |
8c7c6e34 KH |
2714 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2715 | struct page *page, | |
2716 | gfp_t mask, struct mem_cgroup **ptr) | |
2717 | { | |
2718 | struct mem_cgroup *mem; | |
54595fe2 | 2719 | int ret; |
8c7c6e34 | 2720 | |
56039efa KH |
2721 | *ptr = NULL; |
2722 | ||
f8d66542 | 2723 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2724 | return 0; |
2725 | ||
2726 | if (!do_swap_account) | |
2727 | goto charge_cur_mm; | |
8c7c6e34 KH |
2728 | /* |
2729 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2730 | * the pte, and even removed page from swap cache: in those cases |
2731 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2732 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2733 | */ |
2734 | if (!PageSwapCache(page)) | |
407f9c8b | 2735 | goto charge_cur_mm; |
e42d9d5d | 2736 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
2737 | if (!mem) |
2738 | goto charge_cur_mm; | |
8c7c6e34 | 2739 | *ptr = mem; |
7ec99d62 | 2740 | ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true); |
54595fe2 KH |
2741 | css_put(&mem->css); |
2742 | return ret; | |
8c7c6e34 KH |
2743 | charge_cur_mm: |
2744 | if (unlikely(!mm)) | |
2745 | mm = &init_mm; | |
7ec99d62 | 2746 | return __mem_cgroup_try_charge(mm, mask, 1, ptr, true); |
8c7c6e34 KH |
2747 | } |
2748 | ||
83aae4c7 DN |
2749 | static void |
2750 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2751 | enum charge_type ctype) | |
7a81b88c | 2752 | { |
f8d66542 | 2753 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2754 | return; |
2755 | if (!ptr) | |
2756 | return; | |
88703267 | 2757 | cgroup_exclude_rmdir(&ptr->css); |
5a6475a4 KH |
2758 | |
2759 | __mem_cgroup_commit_charge_lrucare(page, ptr, ctype); | |
8c7c6e34 KH |
2760 | /* |
2761 | * Now swap is on-memory. This means this page may be | |
2762 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2763 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2764 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2765 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2766 | */ |
03f3c433 | 2767 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2768 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2769 | unsigned short id; |
8c7c6e34 | 2770 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2771 | |
2772 | id = swap_cgroup_record(ent, 0); | |
2773 | rcu_read_lock(); | |
2774 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2775 | if (memcg) { |
a3b2d692 KH |
2776 | /* |
2777 | * This recorded memcg can be obsolete one. So, avoid | |
2778 | * calling css_tryget | |
2779 | */ | |
0c3e73e8 | 2780 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2781 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2782 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2783 | mem_cgroup_put(memcg); |
2784 | } | |
a3b2d692 | 2785 | rcu_read_unlock(); |
8c7c6e34 | 2786 | } |
88703267 KH |
2787 | /* |
2788 | * At swapin, we may charge account against cgroup which has no tasks. | |
2789 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2790 | * In that case, we need to call pre_destroy() again. check it here. | |
2791 | */ | |
2792 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2793 | } |
2794 | ||
83aae4c7 DN |
2795 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2796 | { | |
2797 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2798 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2799 | } | |
2800 | ||
7a81b88c KH |
2801 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2802 | { | |
f8d66542 | 2803 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2804 | return; |
2805 | if (!mem) | |
2806 | return; | |
e7018b8d | 2807 | __mem_cgroup_cancel_charge(mem, 1); |
7a81b88c KH |
2808 | } |
2809 | ||
7ec99d62 JW |
2810 | static void mem_cgroup_do_uncharge(struct mem_cgroup *mem, |
2811 | unsigned int nr_pages, | |
2812 | const enum charge_type ctype) | |
569b846d KH |
2813 | { |
2814 | struct memcg_batch_info *batch = NULL; | |
2815 | bool uncharge_memsw = true; | |
7ec99d62 | 2816 | |
569b846d KH |
2817 | /* If swapout, usage of swap doesn't decrease */ |
2818 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2819 | uncharge_memsw = false; | |
569b846d KH |
2820 | |
2821 | batch = ¤t->memcg_batch; | |
2822 | /* | |
2823 | * In usual, we do css_get() when we remember memcg pointer. | |
2824 | * But in this case, we keep res->usage until end of a series of | |
2825 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2826 | */ | |
2827 | if (!batch->memcg) | |
2828 | batch->memcg = mem; | |
3c11ecf4 KH |
2829 | /* |
2830 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2831 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2832 | * the same cgroup and we have chance to coalesce uncharges. |
2833 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2834 | * because we want to do uncharge as soon as possible. | |
2835 | */ | |
2836 | ||
2837 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2838 | goto direct_uncharge; | |
2839 | ||
7ec99d62 | 2840 | if (nr_pages > 1) |
ec168510 AA |
2841 | goto direct_uncharge; |
2842 | ||
569b846d KH |
2843 | /* |
2844 | * In typical case, batch->memcg == mem. This means we can | |
2845 | * merge a series of uncharges to an uncharge of res_counter. | |
2846 | * If not, we uncharge res_counter ony by one. | |
2847 | */ | |
2848 | if (batch->memcg != mem) | |
2849 | goto direct_uncharge; | |
2850 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2851 | batch->nr_pages++; |
569b846d | 2852 | if (uncharge_memsw) |
7ffd4ca7 | 2853 | batch->memsw_nr_pages++; |
569b846d KH |
2854 | return; |
2855 | direct_uncharge: | |
7ec99d62 | 2856 | res_counter_uncharge(&mem->res, nr_pages * PAGE_SIZE); |
569b846d | 2857 | if (uncharge_memsw) |
7ec99d62 | 2858 | res_counter_uncharge(&mem->memsw, nr_pages * PAGE_SIZE); |
3c11ecf4 KH |
2859 | if (unlikely(batch->memcg != mem)) |
2860 | memcg_oom_recover(mem); | |
569b846d KH |
2861 | return; |
2862 | } | |
7a81b88c | 2863 | |
8a9f3ccd | 2864 | /* |
69029cd5 | 2865 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2866 | */ |
8c7c6e34 | 2867 | static struct mem_cgroup * |
69029cd5 | 2868 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2869 | { |
8c7c6e34 | 2870 | struct mem_cgroup *mem = NULL; |
7ec99d62 JW |
2871 | unsigned int nr_pages = 1; |
2872 | struct page_cgroup *pc; | |
8a9f3ccd | 2873 | |
f8d66542 | 2874 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2875 | return NULL; |
4077960e | 2876 | |
d13d1443 | 2877 | if (PageSwapCache(page)) |
8c7c6e34 | 2878 | return NULL; |
d13d1443 | 2879 | |
37c2ac78 | 2880 | if (PageTransHuge(page)) { |
7ec99d62 | 2881 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2882 | VM_BUG_ON(!PageTransHuge(page)); |
2883 | } | |
8697d331 | 2884 | /* |
3c541e14 | 2885 | * Check if our page_cgroup is valid |
8697d331 | 2886 | */ |
52d4b9ac KH |
2887 | pc = lookup_page_cgroup(page); |
2888 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2889 | return NULL; |
b9c565d5 | 2890 | |
52d4b9ac | 2891 | lock_page_cgroup(pc); |
d13d1443 | 2892 | |
8c7c6e34 KH |
2893 | mem = pc->mem_cgroup; |
2894 | ||
d13d1443 KH |
2895 | if (!PageCgroupUsed(pc)) |
2896 | goto unlock_out; | |
2897 | ||
2898 | switch (ctype) { | |
2899 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2900 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c AM |
2901 | /* See mem_cgroup_prepare_migration() */ |
2902 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2903 | goto unlock_out; |
2904 | break; | |
2905 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2906 | if (!PageAnon(page)) { /* Shared memory */ | |
2907 | if (page->mapping && !page_is_file_cache(page)) | |
2908 | goto unlock_out; | |
2909 | } else if (page_mapped(page)) /* Anon */ | |
2910 | goto unlock_out; | |
2911 | break; | |
2912 | default: | |
2913 | break; | |
52d4b9ac | 2914 | } |
d13d1443 | 2915 | |
7ec99d62 | 2916 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -nr_pages); |
04046e1a | 2917 | |
52d4b9ac | 2918 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2919 | /* |
2920 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2921 | * freed from LRU. This is safe because uncharged page is expected not | |
2922 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2923 | * special functions. | |
2924 | */ | |
b9c565d5 | 2925 | |
52d4b9ac | 2926 | unlock_page_cgroup(pc); |
f75ca962 KH |
2927 | /* |
2928 | * even after unlock, we have mem->res.usage here and this memcg | |
2929 | * will never be freed. | |
2930 | */ | |
d2265e6f | 2931 | memcg_check_events(mem, page); |
f75ca962 KH |
2932 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
2933 | mem_cgroup_swap_statistics(mem, true); | |
2934 | mem_cgroup_get(mem); | |
2935 | } | |
2936 | if (!mem_cgroup_is_root(mem)) | |
7ec99d62 | 2937 | mem_cgroup_do_uncharge(mem, nr_pages, ctype); |
6d12e2d8 | 2938 | |
8c7c6e34 | 2939 | return mem; |
d13d1443 KH |
2940 | |
2941 | unlock_out: | |
2942 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2943 | return NULL; |
3c541e14 BS |
2944 | } |
2945 | ||
69029cd5 KH |
2946 | void mem_cgroup_uncharge_page(struct page *page) |
2947 | { | |
52d4b9ac KH |
2948 | /* early check. */ |
2949 | if (page_mapped(page)) | |
2950 | return; | |
2951 | if (page->mapping && !PageAnon(page)) | |
2952 | return; | |
69029cd5 KH |
2953 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2954 | } | |
2955 | ||
2956 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2957 | { | |
2958 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2959 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2960 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2961 | } | |
2962 | ||
569b846d KH |
2963 | /* |
2964 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2965 | * In that cases, pages are freed continuously and we can expect pages | |
2966 | * are in the same memcg. All these calls itself limits the number of | |
2967 | * pages freed at once, then uncharge_start/end() is called properly. | |
2968 | * This may be called prural(2) times in a context, | |
2969 | */ | |
2970 | ||
2971 | void mem_cgroup_uncharge_start(void) | |
2972 | { | |
2973 | current->memcg_batch.do_batch++; | |
2974 | /* We can do nest. */ | |
2975 | if (current->memcg_batch.do_batch == 1) { | |
2976 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
2977 | current->memcg_batch.nr_pages = 0; |
2978 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
2979 | } |
2980 | } | |
2981 | ||
2982 | void mem_cgroup_uncharge_end(void) | |
2983 | { | |
2984 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2985 | ||
2986 | if (!batch->do_batch) | |
2987 | return; | |
2988 | ||
2989 | batch->do_batch--; | |
2990 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2991 | return; | |
2992 | ||
2993 | if (!batch->memcg) | |
2994 | return; | |
2995 | /* | |
2996 | * This "batch->memcg" is valid without any css_get/put etc... | |
2997 | * bacause we hide charges behind us. | |
2998 | */ | |
7ffd4ca7 JW |
2999 | if (batch->nr_pages) |
3000 | res_counter_uncharge(&batch->memcg->res, | |
3001 | batch->nr_pages * PAGE_SIZE); | |
3002 | if (batch->memsw_nr_pages) | |
3003 | res_counter_uncharge(&batch->memcg->memsw, | |
3004 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3005 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3006 | /* forget this pointer (for sanity check) */ |
3007 | batch->memcg = NULL; | |
3008 | } | |
3009 | ||
e767e056 | 3010 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3011 | /* |
e767e056 | 3012 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3013 | * memcg information is recorded to swap_cgroup of "ent" |
3014 | */ | |
8a9478ca KH |
3015 | void |
3016 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3017 | { |
3018 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3019 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3020 | ||
3021 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3022 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3023 | ||
3024 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3025 | |
f75ca962 KH |
3026 | /* |
3027 | * record memcg information, if swapout && memcg != NULL, | |
3028 | * mem_cgroup_get() was called in uncharge(). | |
3029 | */ | |
3030 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3031 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3032 | } |
e767e056 | 3033 | #endif |
8c7c6e34 KH |
3034 | |
3035 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3036 | /* | |
3037 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3038 | * uncharge "memsw" account. | |
3039 | */ | |
3040 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3041 | { |
8c7c6e34 | 3042 | struct mem_cgroup *memcg; |
a3b2d692 | 3043 | unsigned short id; |
8c7c6e34 KH |
3044 | |
3045 | if (!do_swap_account) | |
3046 | return; | |
3047 | ||
a3b2d692 KH |
3048 | id = swap_cgroup_record(ent, 0); |
3049 | rcu_read_lock(); | |
3050 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3051 | if (memcg) { |
a3b2d692 KH |
3052 | /* |
3053 | * We uncharge this because swap is freed. | |
3054 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3055 | */ | |
0c3e73e8 | 3056 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3057 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3058 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3059 | mem_cgroup_put(memcg); |
3060 | } | |
a3b2d692 | 3061 | rcu_read_unlock(); |
d13d1443 | 3062 | } |
02491447 DN |
3063 | |
3064 | /** | |
3065 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3066 | * @entry: swap entry to be moved | |
3067 | * @from: mem_cgroup which the entry is moved from | |
3068 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 3069 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
3070 | * |
3071 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3072 | * as the mem_cgroup's id of @from. | |
3073 | * | |
3074 | * Returns 0 on success, -EINVAL on failure. | |
3075 | * | |
3076 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3077 | * both res and memsw, and called css_get(). | |
3078 | */ | |
3079 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3080 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3081 | { |
3082 | unsigned short old_id, new_id; | |
3083 | ||
3084 | old_id = css_id(&from->css); | |
3085 | new_id = css_id(&to->css); | |
3086 | ||
3087 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3088 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3089 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3090 | /* |
483c30b5 DN |
3091 | * This function is only called from task migration context now. |
3092 | * It postpones res_counter and refcount handling till the end | |
3093 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3094 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3095 | * because if the process that has been moved to @to does | |
3096 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3097 | */ |
02491447 | 3098 | mem_cgroup_get(to); |
483c30b5 DN |
3099 | if (need_fixup) { |
3100 | if (!mem_cgroup_is_root(from)) | |
3101 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
3102 | mem_cgroup_put(from); | |
3103 | /* | |
3104 | * we charged both to->res and to->memsw, so we should | |
3105 | * uncharge to->res. | |
3106 | */ | |
3107 | if (!mem_cgroup_is_root(to)) | |
3108 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 3109 | } |
02491447 DN |
3110 | return 0; |
3111 | } | |
3112 | return -EINVAL; | |
3113 | } | |
3114 | #else | |
3115 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3116 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3117 | { |
3118 | return -EINVAL; | |
3119 | } | |
8c7c6e34 | 3120 | #endif |
d13d1443 | 3121 | |
ae41be37 | 3122 | /* |
01b1ae63 KH |
3123 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3124 | * page belongs to. | |
ae41be37 | 3125 | */ |
ac39cf8c | 3126 | int mem_cgroup_prepare_migration(struct page *page, |
ef6a3c63 | 3127 | struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask) |
ae41be37 | 3128 | { |
e8589cc1 | 3129 | struct mem_cgroup *mem = NULL; |
7ec99d62 | 3130 | struct page_cgroup *pc; |
ac39cf8c | 3131 | enum charge_type ctype; |
e8589cc1 | 3132 | int ret = 0; |
8869b8f6 | 3133 | |
56039efa KH |
3134 | *ptr = NULL; |
3135 | ||
ec168510 | 3136 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3137 | if (mem_cgroup_disabled()) |
4077960e BS |
3138 | return 0; |
3139 | ||
52d4b9ac KH |
3140 | pc = lookup_page_cgroup(page); |
3141 | lock_page_cgroup(pc); | |
3142 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
3143 | mem = pc->mem_cgroup; |
3144 | css_get(&mem->css); | |
ac39cf8c AM |
3145 | /* |
3146 | * At migrating an anonymous page, its mapcount goes down | |
3147 | * to 0 and uncharge() will be called. But, even if it's fully | |
3148 | * unmapped, migration may fail and this page has to be | |
3149 | * charged again. We set MIGRATION flag here and delay uncharge | |
3150 | * until end_migration() is called | |
3151 | * | |
3152 | * Corner Case Thinking | |
3153 | * A) | |
3154 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3155 | * while migration was ongoing. | |
3156 | * If unmap finds the old page, uncharge() of it will be delayed | |
3157 | * until end_migration(). If unmap finds a new page, it's | |
3158 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3159 | * finds swap_migration_entry, the new page will not be mapped | |
3160 | * and end_migration() will find it(mapcount==0). | |
3161 | * | |
3162 | * B) | |
3163 | * When the old page was mapped but migraion fails, the kernel | |
3164 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3165 | * if mapcount goes down to 0. We can do remap successfully | |
3166 | * without charging it again. | |
3167 | * | |
3168 | * C) | |
3169 | * The "old" page is under lock_page() until the end of | |
3170 | * migration, so, the old page itself will not be swapped-out. | |
3171 | * If the new page is swapped out before end_migraton, our | |
3172 | * hook to usual swap-out path will catch the event. | |
3173 | */ | |
3174 | if (PageAnon(page)) | |
3175 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3176 | } |
52d4b9ac | 3177 | unlock_page_cgroup(pc); |
ac39cf8c AM |
3178 | /* |
3179 | * If the page is not charged at this point, | |
3180 | * we return here. | |
3181 | */ | |
3182 | if (!mem) | |
3183 | return 0; | |
01b1ae63 | 3184 | |
93d5c9be | 3185 | *ptr = mem; |
7ec99d62 | 3186 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false); |
ac39cf8c AM |
3187 | css_put(&mem->css);/* drop extra refcnt */ |
3188 | if (ret || *ptr == NULL) { | |
3189 | if (PageAnon(page)) { | |
3190 | lock_page_cgroup(pc); | |
3191 | ClearPageCgroupMigration(pc); | |
3192 | unlock_page_cgroup(pc); | |
3193 | /* | |
3194 | * The old page may be fully unmapped while we kept it. | |
3195 | */ | |
3196 | mem_cgroup_uncharge_page(page); | |
3197 | } | |
3198 | return -ENOMEM; | |
e8589cc1 | 3199 | } |
ac39cf8c AM |
3200 | /* |
3201 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3202 | * is called before end_migration, we can catch all events on this new | |
3203 | * page. In the case new page is migrated but not remapped, new page's | |
3204 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3205 | */ | |
3206 | pc = lookup_page_cgroup(newpage); | |
3207 | if (PageAnon(page)) | |
3208 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3209 | else if (page_is_file_cache(page)) | |
3210 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3211 | else | |
3212 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
7ec99d62 | 3213 | __mem_cgroup_commit_charge(mem, page, 1, pc, ctype); |
e8589cc1 | 3214 | return ret; |
ae41be37 | 3215 | } |
8869b8f6 | 3216 | |
69029cd5 | 3217 | /* remove redundant charge if migration failed*/ |
01b1ae63 | 3218 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
50de1dd9 | 3219 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3220 | { |
ac39cf8c | 3221 | struct page *used, *unused; |
01b1ae63 | 3222 | struct page_cgroup *pc; |
01b1ae63 KH |
3223 | |
3224 | if (!mem) | |
3225 | return; | |
ac39cf8c | 3226 | /* blocks rmdir() */ |
88703267 | 3227 | cgroup_exclude_rmdir(&mem->css); |
50de1dd9 | 3228 | if (!migration_ok) { |
ac39cf8c AM |
3229 | used = oldpage; |
3230 | unused = newpage; | |
01b1ae63 | 3231 | } else { |
ac39cf8c | 3232 | used = newpage; |
01b1ae63 KH |
3233 | unused = oldpage; |
3234 | } | |
69029cd5 | 3235 | /* |
ac39cf8c AM |
3236 | * We disallowed uncharge of pages under migration because mapcount |
3237 | * of the page goes down to zero, temporarly. | |
3238 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3239 | */ |
ac39cf8c AM |
3240 | pc = lookup_page_cgroup(oldpage); |
3241 | lock_page_cgroup(pc); | |
3242 | ClearPageCgroupMigration(pc); | |
3243 | unlock_page_cgroup(pc); | |
01b1ae63 | 3244 | |
ac39cf8c AM |
3245 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
3246 | ||
01b1ae63 | 3247 | /* |
ac39cf8c AM |
3248 | * If a page is a file cache, radix-tree replacement is very atomic |
3249 | * and we can skip this check. When it was an Anon page, its mapcount | |
3250 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3251 | * uncharged yet. There are several case but page->mapcount check | |
3252 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3253 | * check. (see prepare_charge() also) | |
69029cd5 | 3254 | */ |
ac39cf8c AM |
3255 | if (PageAnon(used)) |
3256 | mem_cgroup_uncharge_page(used); | |
88703267 | 3257 | /* |
ac39cf8c AM |
3258 | * At migration, we may charge account against cgroup which has no |
3259 | * tasks. | |
88703267 KH |
3260 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3261 | * In that case, we need to call pre_destroy() again. check it here. | |
3262 | */ | |
3263 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 3264 | } |
78fb7466 | 3265 | |
c9b0ed51 | 3266 | /* |
ae3abae6 DN |
3267 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
3268 | * Calling hierarchical_reclaim is not enough because we should update | |
3269 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
3270 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
3271 | * not from the memcg which this page would be charged to. | |
3272 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 3273 | */ |
ae3abae6 | 3274 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
3275 | struct mm_struct *mm, |
3276 | gfp_t gfp_mask) | |
c9b0ed51 | 3277 | { |
56039efa | 3278 | struct mem_cgroup *mem; |
ae3abae6 | 3279 | int ret; |
c9b0ed51 | 3280 | |
f8d66542 | 3281 | if (mem_cgroup_disabled()) |
cede86ac | 3282 | return 0; |
c9b0ed51 | 3283 | |
ae3abae6 DN |
3284 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
3285 | if (!ret) | |
3286 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 3287 | |
ae3abae6 | 3288 | return ret; |
c9b0ed51 KH |
3289 | } |
3290 | ||
f212ad7c DN |
3291 | #ifdef CONFIG_DEBUG_VM |
3292 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3293 | { | |
3294 | struct page_cgroup *pc; | |
3295 | ||
3296 | pc = lookup_page_cgroup(page); | |
3297 | if (likely(pc) && PageCgroupUsed(pc)) | |
3298 | return pc; | |
3299 | return NULL; | |
3300 | } | |
3301 | ||
3302 | bool mem_cgroup_bad_page_check(struct page *page) | |
3303 | { | |
3304 | if (mem_cgroup_disabled()) | |
3305 | return false; | |
3306 | ||
3307 | return lookup_page_cgroup_used(page) != NULL; | |
3308 | } | |
3309 | ||
3310 | void mem_cgroup_print_bad_page(struct page *page) | |
3311 | { | |
3312 | struct page_cgroup *pc; | |
3313 | ||
3314 | pc = lookup_page_cgroup_used(page); | |
3315 | if (pc) { | |
3316 | int ret = -1; | |
3317 | char *path; | |
3318 | ||
3319 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p", | |
3320 | pc, pc->flags, pc->mem_cgroup); | |
3321 | ||
3322 | path = kmalloc(PATH_MAX, GFP_KERNEL); | |
3323 | if (path) { | |
3324 | rcu_read_lock(); | |
3325 | ret = cgroup_path(pc->mem_cgroup->css.cgroup, | |
3326 | path, PATH_MAX); | |
3327 | rcu_read_unlock(); | |
3328 | } | |
3329 | ||
3330 | printk(KERN_CONT "(%s)\n", | |
3331 | (ret < 0) ? "cannot get the path" : path); | |
3332 | kfree(path); | |
3333 | } | |
3334 | } | |
3335 | #endif | |
3336 | ||
8c7c6e34 KH |
3337 | static DEFINE_MUTEX(set_limit_mutex); |
3338 | ||
d38d2a75 | 3339 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3340 | unsigned long long val) |
628f4235 | 3341 | { |
81d39c20 | 3342 | int retry_count; |
3c11ecf4 | 3343 | u64 memswlimit, memlimit; |
628f4235 | 3344 | int ret = 0; |
81d39c20 KH |
3345 | int children = mem_cgroup_count_children(memcg); |
3346 | u64 curusage, oldusage; | |
3c11ecf4 | 3347 | int enlarge; |
81d39c20 KH |
3348 | |
3349 | /* | |
3350 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3351 | * is depends on callers. We set our retry-count to be function | |
3352 | * of # of children which we should visit in this loop. | |
3353 | */ | |
3354 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3355 | ||
3356 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3357 | |
3c11ecf4 | 3358 | enlarge = 0; |
8c7c6e34 | 3359 | while (retry_count) { |
628f4235 KH |
3360 | if (signal_pending(current)) { |
3361 | ret = -EINTR; | |
3362 | break; | |
3363 | } | |
8c7c6e34 KH |
3364 | /* |
3365 | * Rather than hide all in some function, I do this in | |
3366 | * open coded manner. You see what this really does. | |
3367 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3368 | */ | |
3369 | mutex_lock(&set_limit_mutex); | |
3370 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3371 | if (memswlimit < val) { | |
3372 | ret = -EINVAL; | |
3373 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3374 | break; |
3375 | } | |
3c11ecf4 KH |
3376 | |
3377 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3378 | if (memlimit < val) | |
3379 | enlarge = 1; | |
3380 | ||
8c7c6e34 | 3381 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3382 | if (!ret) { |
3383 | if (memswlimit == val) | |
3384 | memcg->memsw_is_minimum = true; | |
3385 | else | |
3386 | memcg->memsw_is_minimum = false; | |
3387 | } | |
8c7c6e34 KH |
3388 | mutex_unlock(&set_limit_mutex); |
3389 | ||
3390 | if (!ret) | |
3391 | break; | |
3392 | ||
aa20d489 | 3393 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
0ae5e89c YH |
3394 | MEM_CGROUP_RECLAIM_SHRINK, |
3395 | NULL); | |
81d39c20 KH |
3396 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3397 | /* Usage is reduced ? */ | |
3398 | if (curusage >= oldusage) | |
3399 | retry_count--; | |
3400 | else | |
3401 | oldusage = curusage; | |
8c7c6e34 | 3402 | } |
3c11ecf4 KH |
3403 | if (!ret && enlarge) |
3404 | memcg_oom_recover(memcg); | |
14797e23 | 3405 | |
8c7c6e34 KH |
3406 | return ret; |
3407 | } | |
3408 | ||
338c8431 LZ |
3409 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3410 | unsigned long long val) | |
8c7c6e34 | 3411 | { |
81d39c20 | 3412 | int retry_count; |
3c11ecf4 | 3413 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3414 | int children = mem_cgroup_count_children(memcg); |
3415 | int ret = -EBUSY; | |
3c11ecf4 | 3416 | int enlarge = 0; |
8c7c6e34 | 3417 | |
81d39c20 KH |
3418 | /* see mem_cgroup_resize_res_limit */ |
3419 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3420 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3421 | while (retry_count) { |
3422 | if (signal_pending(current)) { | |
3423 | ret = -EINTR; | |
3424 | break; | |
3425 | } | |
3426 | /* | |
3427 | * Rather than hide all in some function, I do this in | |
3428 | * open coded manner. You see what this really does. | |
3429 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3430 | */ | |
3431 | mutex_lock(&set_limit_mutex); | |
3432 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3433 | if (memlimit > val) { | |
3434 | ret = -EINVAL; | |
3435 | mutex_unlock(&set_limit_mutex); | |
3436 | break; | |
3437 | } | |
3c11ecf4 KH |
3438 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3439 | if (memswlimit < val) | |
3440 | enlarge = 1; | |
8c7c6e34 | 3441 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3442 | if (!ret) { |
3443 | if (memlimit == val) | |
3444 | memcg->memsw_is_minimum = true; | |
3445 | else | |
3446 | memcg->memsw_is_minimum = false; | |
3447 | } | |
8c7c6e34 KH |
3448 | mutex_unlock(&set_limit_mutex); |
3449 | ||
3450 | if (!ret) | |
3451 | break; | |
3452 | ||
4e416953 | 3453 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 | 3454 | MEM_CGROUP_RECLAIM_NOSWAP | |
0ae5e89c YH |
3455 | MEM_CGROUP_RECLAIM_SHRINK, |
3456 | NULL); | |
8c7c6e34 | 3457 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3458 | /* Usage is reduced ? */ |
8c7c6e34 | 3459 | if (curusage >= oldusage) |
628f4235 | 3460 | retry_count--; |
81d39c20 KH |
3461 | else |
3462 | oldusage = curusage; | |
628f4235 | 3463 | } |
3c11ecf4 KH |
3464 | if (!ret && enlarge) |
3465 | memcg_oom_recover(memcg); | |
628f4235 KH |
3466 | return ret; |
3467 | } | |
3468 | ||
4e416953 | 3469 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3470 | gfp_t gfp_mask, |
3471 | unsigned long *total_scanned) | |
4e416953 BS |
3472 | { |
3473 | unsigned long nr_reclaimed = 0; | |
3474 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3475 | unsigned long reclaimed; | |
3476 | int loop = 0; | |
3477 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3478 | unsigned long long excess; |
0ae5e89c | 3479 | unsigned long nr_scanned; |
4e416953 BS |
3480 | |
3481 | if (order > 0) | |
3482 | return 0; | |
3483 | ||
00918b6a | 3484 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3485 | /* |
3486 | * This loop can run a while, specially if mem_cgroup's continuously | |
3487 | * keep exceeding their soft limit and putting the system under | |
3488 | * pressure | |
3489 | */ | |
3490 | do { | |
3491 | if (next_mz) | |
3492 | mz = next_mz; | |
3493 | else | |
3494 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3495 | if (!mz) | |
3496 | break; | |
3497 | ||
0ae5e89c | 3498 | nr_scanned = 0; |
4e416953 BS |
3499 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, |
3500 | gfp_mask, | |
0ae5e89c YH |
3501 | MEM_CGROUP_RECLAIM_SOFT, |
3502 | &nr_scanned); | |
4e416953 | 3503 | nr_reclaimed += reclaimed; |
0ae5e89c | 3504 | *total_scanned += nr_scanned; |
4e416953 BS |
3505 | spin_lock(&mctz->lock); |
3506 | ||
3507 | /* | |
3508 | * If we failed to reclaim anything from this memory cgroup | |
3509 | * it is time to move on to the next cgroup | |
3510 | */ | |
3511 | next_mz = NULL; | |
3512 | if (!reclaimed) { | |
3513 | do { | |
3514 | /* | |
3515 | * Loop until we find yet another one. | |
3516 | * | |
3517 | * By the time we get the soft_limit lock | |
3518 | * again, someone might have aded the | |
3519 | * group back on the RB tree. Iterate to | |
3520 | * make sure we get a different mem. | |
3521 | * mem_cgroup_largest_soft_limit_node returns | |
3522 | * NULL if no other cgroup is present on | |
3523 | * the tree | |
3524 | */ | |
3525 | next_mz = | |
3526 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3527 | if (next_mz == mz) |
4e416953 | 3528 | css_put(&next_mz->mem->css); |
39cc98f1 | 3529 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3530 | break; |
3531 | } while (1); | |
3532 | } | |
4e416953 | 3533 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 3534 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
3535 | /* |
3536 | * One school of thought says that we should not add | |
3537 | * back the node to the tree if reclaim returns 0. | |
3538 | * But our reclaim could return 0, simply because due | |
3539 | * to priority we are exposing a smaller subset of | |
3540 | * memory to reclaim from. Consider this as a longer | |
3541 | * term TODO. | |
3542 | */ | |
ef8745c1 KH |
3543 | /* If excess == 0, no tree ops */ |
3544 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
3545 | spin_unlock(&mctz->lock); |
3546 | css_put(&mz->mem->css); | |
3547 | loop++; | |
3548 | /* | |
3549 | * Could not reclaim anything and there are no more | |
3550 | * mem cgroups to try or we seem to be looping without | |
3551 | * reclaiming anything. | |
3552 | */ | |
3553 | if (!nr_reclaimed && | |
3554 | (next_mz == NULL || | |
3555 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3556 | break; | |
3557 | } while (!nr_reclaimed); | |
3558 | if (next_mz) | |
3559 | css_put(&next_mz->mem->css); | |
3560 | return nr_reclaimed; | |
3561 | } | |
3562 | ||
cc847582 KH |
3563 | /* |
3564 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3565 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3566 | */ | |
f817ed48 | 3567 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 3568 | int node, int zid, enum lru_list lru) |
cc847582 | 3569 | { |
08e552c6 KH |
3570 | struct zone *zone; |
3571 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 3572 | struct page_cgroup *pc, *busy; |
08e552c6 | 3573 | unsigned long flags, loop; |
072c56c1 | 3574 | struct list_head *list; |
f817ed48 | 3575 | int ret = 0; |
072c56c1 | 3576 | |
08e552c6 KH |
3577 | zone = &NODE_DATA(node)->node_zones[zid]; |
3578 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 3579 | list = &mz->lists[lru]; |
cc847582 | 3580 | |
f817ed48 KH |
3581 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
3582 | /* give some margin against EBUSY etc...*/ | |
3583 | loop += 256; | |
3584 | busy = NULL; | |
3585 | while (loop--) { | |
5564e88b JW |
3586 | struct page *page; |
3587 | ||
f817ed48 | 3588 | ret = 0; |
08e552c6 | 3589 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3590 | if (list_empty(list)) { |
08e552c6 | 3591 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3592 | break; |
f817ed48 KH |
3593 | } |
3594 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
3595 | if (busy == pc) { | |
3596 | list_move(&pc->lru, list); | |
648bcc77 | 3597 | busy = NULL; |
08e552c6 | 3598 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3599 | continue; |
3600 | } | |
08e552c6 | 3601 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3602 | |
6b3ae58e | 3603 | page = lookup_cgroup_page(pc); |
5564e88b JW |
3604 | |
3605 | ret = mem_cgroup_move_parent(page, pc, mem, GFP_KERNEL); | |
f817ed48 | 3606 | if (ret == -ENOMEM) |
52d4b9ac | 3607 | break; |
f817ed48 KH |
3608 | |
3609 | if (ret == -EBUSY || ret == -EINVAL) { | |
3610 | /* found lock contention or "pc" is obsolete. */ | |
3611 | busy = pc; | |
3612 | cond_resched(); | |
3613 | } else | |
3614 | busy = NULL; | |
cc847582 | 3615 | } |
08e552c6 | 3616 | |
f817ed48 KH |
3617 | if (!ret && !list_empty(list)) |
3618 | return -EBUSY; | |
3619 | return ret; | |
cc847582 KH |
3620 | } |
3621 | ||
3622 | /* | |
3623 | * make mem_cgroup's charge to be 0 if there is no task. | |
3624 | * This enables deleting this mem_cgroup. | |
3625 | */ | |
c1e862c1 | 3626 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 3627 | { |
f817ed48 KH |
3628 | int ret; |
3629 | int node, zid, shrink; | |
3630 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 3631 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 3632 | |
cc847582 | 3633 | css_get(&mem->css); |
f817ed48 KH |
3634 | |
3635 | shrink = 0; | |
c1e862c1 KH |
3636 | /* should free all ? */ |
3637 | if (free_all) | |
3638 | goto try_to_free; | |
f817ed48 | 3639 | move_account: |
fce66477 | 3640 | do { |
f817ed48 | 3641 | ret = -EBUSY; |
c1e862c1 KH |
3642 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3643 | goto out; | |
3644 | ret = -EINTR; | |
3645 | if (signal_pending(current)) | |
cc847582 | 3646 | goto out; |
52d4b9ac KH |
3647 | /* This is for making all *used* pages to be on LRU. */ |
3648 | lru_add_drain_all(); | |
cdec2e42 | 3649 | drain_all_stock_sync(); |
f817ed48 | 3650 | ret = 0; |
32047e2a | 3651 | mem_cgroup_start_move(mem); |
299b4eaa | 3652 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3653 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 3654 | enum lru_list l; |
f817ed48 KH |
3655 | for_each_lru(l) { |
3656 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 3657 | node, zid, l); |
f817ed48 KH |
3658 | if (ret) |
3659 | break; | |
3660 | } | |
1ecaab2b | 3661 | } |
f817ed48 KH |
3662 | if (ret) |
3663 | break; | |
3664 | } | |
32047e2a | 3665 | mem_cgroup_end_move(mem); |
3c11ecf4 | 3666 | memcg_oom_recover(mem); |
f817ed48 KH |
3667 | /* it seems parent cgroup doesn't have enough mem */ |
3668 | if (ret == -ENOMEM) | |
3669 | goto try_to_free; | |
52d4b9ac | 3670 | cond_resched(); |
fce66477 DN |
3671 | /* "ret" should also be checked to ensure all lists are empty. */ |
3672 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
3673 | out: |
3674 | css_put(&mem->css); | |
3675 | return ret; | |
f817ed48 KH |
3676 | |
3677 | try_to_free: | |
c1e862c1 KH |
3678 | /* returns EBUSY if there is a task or if we come here twice. */ |
3679 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3680 | ret = -EBUSY; |
3681 | goto out; | |
3682 | } | |
c1e862c1 KH |
3683 | /* we call try-to-free pages for make this cgroup empty */ |
3684 | lru_add_drain_all(); | |
f817ed48 KH |
3685 | /* try to free all pages in this cgroup */ |
3686 | shrink = 1; | |
3687 | while (nr_retries && mem->res.usage > 0) { | |
3688 | int progress; | |
c1e862c1 KH |
3689 | |
3690 | if (signal_pending(current)) { | |
3691 | ret = -EINTR; | |
3692 | goto out; | |
3693 | } | |
a7885eb8 KM |
3694 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
3695 | false, get_swappiness(mem)); | |
c1e862c1 | 3696 | if (!progress) { |
f817ed48 | 3697 | nr_retries--; |
c1e862c1 | 3698 | /* maybe some writeback is necessary */ |
8aa7e847 | 3699 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3700 | } |
f817ed48 KH |
3701 | |
3702 | } | |
08e552c6 | 3703 | lru_add_drain(); |
f817ed48 | 3704 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3705 | goto move_account; |
cc847582 KH |
3706 | } |
3707 | ||
c1e862c1 KH |
3708 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3709 | { | |
3710 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3711 | } | |
3712 | ||
3713 | ||
18f59ea7 BS |
3714 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3715 | { | |
3716 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3717 | } | |
3718 | ||
3719 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3720 | u64 val) | |
3721 | { | |
3722 | int retval = 0; | |
3723 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3724 | struct cgroup *parent = cont->parent; | |
3725 | struct mem_cgroup *parent_mem = NULL; | |
3726 | ||
3727 | if (parent) | |
3728 | parent_mem = mem_cgroup_from_cont(parent); | |
3729 | ||
3730 | cgroup_lock(); | |
3731 | /* | |
af901ca1 | 3732 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3733 | * in the child subtrees. If it is unset, then the change can |
3734 | * occur, provided the current cgroup has no children. | |
3735 | * | |
3736 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3737 | * set if there are no children. | |
3738 | */ | |
3739 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
3740 | (val == 1 || val == 0)) { | |
3741 | if (list_empty(&cont->children)) | |
3742 | mem->use_hierarchy = val; | |
3743 | else | |
3744 | retval = -EBUSY; | |
3745 | } else | |
3746 | retval = -EINVAL; | |
3747 | cgroup_unlock(); | |
3748 | ||
3749 | return retval; | |
3750 | } | |
3751 | ||
0c3e73e8 | 3752 | |
7a159cc9 JW |
3753 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *mem, |
3754 | enum mem_cgroup_stat_index idx) | |
0c3e73e8 | 3755 | { |
7d74b06f | 3756 | struct mem_cgroup *iter; |
7a159cc9 | 3757 | long val = 0; |
0c3e73e8 | 3758 | |
7a159cc9 | 3759 | /* Per-cpu values can be negative, use a signed accumulator */ |
7d74b06f KH |
3760 | for_each_mem_cgroup_tree(iter, mem) |
3761 | val += mem_cgroup_read_stat(iter, idx); | |
3762 | ||
3763 | if (val < 0) /* race ? */ | |
3764 | val = 0; | |
3765 | return val; | |
0c3e73e8 BS |
3766 | } |
3767 | ||
104f3928 KS |
3768 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
3769 | { | |
7d74b06f | 3770 | u64 val; |
104f3928 KS |
3771 | |
3772 | if (!mem_cgroup_is_root(mem)) { | |
3773 | if (!swap) | |
3774 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
3775 | else | |
3776 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
3777 | } | |
3778 | ||
7a159cc9 JW |
3779 | val = mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_CACHE); |
3780 | val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3781 | |
7d74b06f | 3782 | if (swap) |
7a159cc9 | 3783 | val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3784 | |
3785 | return val << PAGE_SHIFT; | |
3786 | } | |
3787 | ||
2c3daa72 | 3788 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3789 | { |
8c7c6e34 | 3790 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 3791 | u64 val; |
8c7c6e34 KH |
3792 | int type, name; |
3793 | ||
3794 | type = MEMFILE_TYPE(cft->private); | |
3795 | name = MEMFILE_ATTR(cft->private); | |
3796 | switch (type) { | |
3797 | case _MEM: | |
104f3928 KS |
3798 | if (name == RES_USAGE) |
3799 | val = mem_cgroup_usage(mem, false); | |
3800 | else | |
0c3e73e8 | 3801 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
3802 | break; |
3803 | case _MEMSWAP: | |
104f3928 KS |
3804 | if (name == RES_USAGE) |
3805 | val = mem_cgroup_usage(mem, true); | |
3806 | else | |
0c3e73e8 | 3807 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
3808 | break; |
3809 | default: | |
3810 | BUG(); | |
3811 | break; | |
3812 | } | |
3813 | return val; | |
8cdea7c0 | 3814 | } |
628f4235 KH |
3815 | /* |
3816 | * The user of this function is... | |
3817 | * RES_LIMIT. | |
3818 | */ | |
856c13aa PM |
3819 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3820 | const char *buffer) | |
8cdea7c0 | 3821 | { |
628f4235 | 3822 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3823 | int type, name; |
628f4235 KH |
3824 | unsigned long long val; |
3825 | int ret; | |
3826 | ||
8c7c6e34 KH |
3827 | type = MEMFILE_TYPE(cft->private); |
3828 | name = MEMFILE_ATTR(cft->private); | |
3829 | switch (name) { | |
628f4235 | 3830 | case RES_LIMIT: |
4b3bde4c BS |
3831 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3832 | ret = -EINVAL; | |
3833 | break; | |
3834 | } | |
628f4235 KH |
3835 | /* This function does all necessary parse...reuse it */ |
3836 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3837 | if (ret) |
3838 | break; | |
3839 | if (type == _MEM) | |
628f4235 | 3840 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3841 | else |
3842 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3843 | break; |
296c81d8 BS |
3844 | case RES_SOFT_LIMIT: |
3845 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3846 | if (ret) | |
3847 | break; | |
3848 | /* | |
3849 | * For memsw, soft limits are hard to implement in terms | |
3850 | * of semantics, for now, we support soft limits for | |
3851 | * control without swap | |
3852 | */ | |
3853 | if (type == _MEM) | |
3854 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3855 | else | |
3856 | ret = -EINVAL; | |
3857 | break; | |
628f4235 KH |
3858 | default: |
3859 | ret = -EINVAL; /* should be BUG() ? */ | |
3860 | break; | |
3861 | } | |
3862 | return ret; | |
8cdea7c0 BS |
3863 | } |
3864 | ||
fee7b548 KH |
3865 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3866 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3867 | { | |
3868 | struct cgroup *cgroup; | |
3869 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3870 | ||
3871 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3872 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3873 | cgroup = memcg->css.cgroup; | |
3874 | if (!memcg->use_hierarchy) | |
3875 | goto out; | |
3876 | ||
3877 | while (cgroup->parent) { | |
3878 | cgroup = cgroup->parent; | |
3879 | memcg = mem_cgroup_from_cont(cgroup); | |
3880 | if (!memcg->use_hierarchy) | |
3881 | break; | |
3882 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3883 | min_limit = min(min_limit, tmp); | |
3884 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3885 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3886 | } | |
3887 | out: | |
3888 | *mem_limit = min_limit; | |
3889 | *memsw_limit = min_memsw_limit; | |
3890 | return; | |
3891 | } | |
3892 | ||
29f2a4da | 3893 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
3894 | { |
3895 | struct mem_cgroup *mem; | |
8c7c6e34 | 3896 | int type, name; |
c84872e1 PE |
3897 | |
3898 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
3899 | type = MEMFILE_TYPE(event); |
3900 | name = MEMFILE_ATTR(event); | |
3901 | switch (name) { | |
29f2a4da | 3902 | case RES_MAX_USAGE: |
8c7c6e34 KH |
3903 | if (type == _MEM) |
3904 | res_counter_reset_max(&mem->res); | |
3905 | else | |
3906 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
3907 | break; |
3908 | case RES_FAILCNT: | |
8c7c6e34 KH |
3909 | if (type == _MEM) |
3910 | res_counter_reset_failcnt(&mem->res); | |
3911 | else | |
3912 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
3913 | break; |
3914 | } | |
f64c3f54 | 3915 | |
85cc59db | 3916 | return 0; |
c84872e1 PE |
3917 | } |
3918 | ||
7dc74be0 DN |
3919 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3920 | struct cftype *cft) | |
3921 | { | |
3922 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3923 | } | |
3924 | ||
02491447 | 3925 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3926 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3927 | struct cftype *cft, u64 val) | |
3928 | { | |
3929 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3930 | ||
3931 | if (val >= (1 << NR_MOVE_TYPE)) | |
3932 | return -EINVAL; | |
3933 | /* | |
3934 | * We check this value several times in both in can_attach() and | |
3935 | * attach(), so we need cgroup lock to prevent this value from being | |
3936 | * inconsistent. | |
3937 | */ | |
3938 | cgroup_lock(); | |
3939 | mem->move_charge_at_immigrate = val; | |
3940 | cgroup_unlock(); | |
3941 | ||
3942 | return 0; | |
3943 | } | |
02491447 DN |
3944 | #else |
3945 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3946 | struct cftype *cft, u64 val) | |
3947 | { | |
3948 | return -ENOSYS; | |
3949 | } | |
3950 | #endif | |
7dc74be0 | 3951 | |
14067bb3 KH |
3952 | |
3953 | /* For read statistics */ | |
3954 | enum { | |
3955 | MCS_CACHE, | |
3956 | MCS_RSS, | |
d8046582 | 3957 | MCS_FILE_MAPPED, |
14067bb3 KH |
3958 | MCS_PGPGIN, |
3959 | MCS_PGPGOUT, | |
1dd3a273 | 3960 | MCS_SWAP, |
14067bb3 KH |
3961 | MCS_INACTIVE_ANON, |
3962 | MCS_ACTIVE_ANON, | |
3963 | MCS_INACTIVE_FILE, | |
3964 | MCS_ACTIVE_FILE, | |
3965 | MCS_UNEVICTABLE, | |
3966 | NR_MCS_STAT, | |
3967 | }; | |
3968 | ||
3969 | struct mcs_total_stat { | |
3970 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
3971 | }; |
3972 | ||
14067bb3 KH |
3973 | struct { |
3974 | char *local_name; | |
3975 | char *total_name; | |
3976 | } memcg_stat_strings[NR_MCS_STAT] = { | |
3977 | {"cache", "total_cache"}, | |
3978 | {"rss", "total_rss"}, | |
d69b042f | 3979 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
3980 | {"pgpgin", "total_pgpgin"}, |
3981 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3982 | {"swap", "total_swap"}, |
14067bb3 KH |
3983 | {"inactive_anon", "total_inactive_anon"}, |
3984 | {"active_anon", "total_active_anon"}, | |
3985 | {"inactive_file", "total_inactive_file"}, | |
3986 | {"active_file", "total_active_file"}, | |
3987 | {"unevictable", "total_unevictable"} | |
3988 | }; | |
3989 | ||
3990 | ||
7d74b06f KH |
3991 | static void |
3992 | mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
14067bb3 | 3993 | { |
14067bb3 KH |
3994 | s64 val; |
3995 | ||
3996 | /* per cpu stat */ | |
c62b1a3b | 3997 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 3998 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 3999 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 4000 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 4001 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 4002 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
e9f8974f | 4003 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGPGIN); |
14067bb3 | 4004 | s->stat[MCS_PGPGIN] += val; |
e9f8974f | 4005 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGPGOUT); |
14067bb3 | 4006 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 4007 | if (do_swap_account) { |
c62b1a3b | 4008 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
4009 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
4010 | } | |
14067bb3 KH |
4011 | |
4012 | /* per zone stat */ | |
4013 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
4014 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
4015 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
4016 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
4017 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
4018 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
4019 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
4020 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
4021 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
4022 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
14067bb3 KH |
4023 | } |
4024 | ||
4025 | static void | |
4026 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
4027 | { | |
7d74b06f KH |
4028 | struct mem_cgroup *iter; |
4029 | ||
4030 | for_each_mem_cgroup_tree(iter, mem) | |
4031 | mem_cgroup_get_local_stat(iter, s); | |
14067bb3 KH |
4032 | } |
4033 | ||
406eb0c9 YH |
4034 | #ifdef CONFIG_NUMA |
4035 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) | |
4036 | { | |
4037 | int nid; | |
4038 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4039 | unsigned long node_nr; | |
4040 | struct cgroup *cont = m->private; | |
4041 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
4042 | ||
4043 | total_nr = mem_cgroup_nr_lru_pages(mem_cont); | |
4044 | seq_printf(m, "total=%lu", total_nr); | |
4045 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4046 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid); | |
4047 | seq_printf(m, " N%d=%lu", nid, node_nr); | |
4048 | } | |
4049 | seq_putc(m, '\n'); | |
4050 | ||
4051 | file_nr = mem_cgroup_nr_file_lru_pages(mem_cont); | |
4052 | seq_printf(m, "file=%lu", file_nr); | |
4053 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4054 | node_nr = mem_cgroup_node_nr_file_lru_pages(mem_cont, nid); | |
4055 | seq_printf(m, " N%d=%lu", nid, node_nr); | |
4056 | } | |
4057 | seq_putc(m, '\n'); | |
4058 | ||
4059 | anon_nr = mem_cgroup_nr_anon_lru_pages(mem_cont); | |
4060 | seq_printf(m, "anon=%lu", anon_nr); | |
4061 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4062 | node_nr = mem_cgroup_node_nr_anon_lru_pages(mem_cont, nid); | |
4063 | seq_printf(m, " N%d=%lu", nid, node_nr); | |
4064 | } | |
4065 | seq_putc(m, '\n'); | |
4066 | ||
4067 | unevictable_nr = mem_cgroup_nr_unevictable_lru_pages(mem_cont); | |
4068 | seq_printf(m, "unevictable=%lu", unevictable_nr); | |
4069 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
4070 | node_nr = mem_cgroup_node_nr_unevictable_lru_pages(mem_cont, | |
4071 | nid); | |
4072 | seq_printf(m, " N%d=%lu", nid, node_nr); | |
4073 | } | |
4074 | seq_putc(m, '\n'); | |
4075 | return 0; | |
4076 | } | |
4077 | #endif /* CONFIG_NUMA */ | |
4078 | ||
c64745cf PM |
4079 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
4080 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 4081 | { |
d2ceb9b7 | 4082 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 4083 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
4084 | int i; |
4085 | ||
14067bb3 KH |
4086 | memset(&mystat, 0, sizeof(mystat)); |
4087 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 4088 | |
406eb0c9 | 4089 | |
1dd3a273 DN |
4090 | for (i = 0; i < NR_MCS_STAT; i++) { |
4091 | if (i == MCS_SWAP && !do_swap_account) | |
4092 | continue; | |
14067bb3 | 4093 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 4094 | } |
7b854121 | 4095 | |
14067bb3 | 4096 | /* Hierarchical information */ |
fee7b548 KH |
4097 | { |
4098 | unsigned long long limit, memsw_limit; | |
4099 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
4100 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
4101 | if (do_swap_account) | |
4102 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
4103 | } | |
7f016ee8 | 4104 | |
14067bb3 KH |
4105 | memset(&mystat, 0, sizeof(mystat)); |
4106 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
4107 | for (i = 0; i < NR_MCS_STAT; i++) { |
4108 | if (i == MCS_SWAP && !do_swap_account) | |
4109 | continue; | |
14067bb3 | 4110 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 4111 | } |
14067bb3 | 4112 | |
7f016ee8 | 4113 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 4114 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
4115 | |
4116 | { | |
4117 | int nid, zid; | |
4118 | struct mem_cgroup_per_zone *mz; | |
4119 | unsigned long recent_rotated[2] = {0, 0}; | |
4120 | unsigned long recent_scanned[2] = {0, 0}; | |
4121 | ||
4122 | for_each_online_node(nid) | |
4123 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
4124 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
4125 | ||
4126 | recent_rotated[0] += | |
4127 | mz->reclaim_stat.recent_rotated[0]; | |
4128 | recent_rotated[1] += | |
4129 | mz->reclaim_stat.recent_rotated[1]; | |
4130 | recent_scanned[0] += | |
4131 | mz->reclaim_stat.recent_scanned[0]; | |
4132 | recent_scanned[1] += | |
4133 | mz->reclaim_stat.recent_scanned[1]; | |
4134 | } | |
4135 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
4136 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
4137 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
4138 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
4139 | } | |
4140 | #endif | |
4141 | ||
d2ceb9b7 KH |
4142 | return 0; |
4143 | } | |
4144 | ||
a7885eb8 KM |
4145 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4146 | { | |
4147 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4148 | ||
4149 | return get_swappiness(memcg); | |
4150 | } | |
4151 | ||
4152 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4153 | u64 val) | |
4154 | { | |
4155 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4156 | struct mem_cgroup *parent; | |
068b38c1 | 4157 | |
a7885eb8 KM |
4158 | if (val > 100) |
4159 | return -EINVAL; | |
4160 | ||
4161 | if (cgrp->parent == NULL) | |
4162 | return -EINVAL; | |
4163 | ||
4164 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4165 | |
4166 | cgroup_lock(); | |
4167 | ||
a7885eb8 KM |
4168 | /* If under hierarchy, only empty-root can set this value */ |
4169 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4170 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4171 | cgroup_unlock(); | |
a7885eb8 | 4172 | return -EINVAL; |
068b38c1 | 4173 | } |
a7885eb8 | 4174 | |
a7885eb8 | 4175 | memcg->swappiness = val; |
a7885eb8 | 4176 | |
068b38c1 LZ |
4177 | cgroup_unlock(); |
4178 | ||
a7885eb8 KM |
4179 | return 0; |
4180 | } | |
4181 | ||
2e72b634 KS |
4182 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4183 | { | |
4184 | struct mem_cgroup_threshold_ary *t; | |
4185 | u64 usage; | |
4186 | int i; | |
4187 | ||
4188 | rcu_read_lock(); | |
4189 | if (!swap) | |
2c488db2 | 4190 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4191 | else |
2c488db2 | 4192 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4193 | |
4194 | if (!t) | |
4195 | goto unlock; | |
4196 | ||
4197 | usage = mem_cgroup_usage(memcg, swap); | |
4198 | ||
4199 | /* | |
4200 | * current_threshold points to threshold just below usage. | |
4201 | * If it's not true, a threshold was crossed after last | |
4202 | * call of __mem_cgroup_threshold(). | |
4203 | */ | |
5407a562 | 4204 | i = t->current_threshold; |
2e72b634 KS |
4205 | |
4206 | /* | |
4207 | * Iterate backward over array of thresholds starting from | |
4208 | * current_threshold and check if a threshold is crossed. | |
4209 | * If none of thresholds below usage is crossed, we read | |
4210 | * only one element of the array here. | |
4211 | */ | |
4212 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4213 | eventfd_signal(t->entries[i].eventfd, 1); | |
4214 | ||
4215 | /* i = current_threshold + 1 */ | |
4216 | i++; | |
4217 | ||
4218 | /* | |
4219 | * Iterate forward over array of thresholds starting from | |
4220 | * current_threshold+1 and check if a threshold is crossed. | |
4221 | * If none of thresholds above usage is crossed, we read | |
4222 | * only one element of the array here. | |
4223 | */ | |
4224 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4225 | eventfd_signal(t->entries[i].eventfd, 1); | |
4226 | ||
4227 | /* Update current_threshold */ | |
5407a562 | 4228 | t->current_threshold = i - 1; |
2e72b634 KS |
4229 | unlock: |
4230 | rcu_read_unlock(); | |
4231 | } | |
4232 | ||
4233 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4234 | { | |
ad4ca5f4 KS |
4235 | while (memcg) { |
4236 | __mem_cgroup_threshold(memcg, false); | |
4237 | if (do_swap_account) | |
4238 | __mem_cgroup_threshold(memcg, true); | |
4239 | ||
4240 | memcg = parent_mem_cgroup(memcg); | |
4241 | } | |
2e72b634 KS |
4242 | } |
4243 | ||
4244 | static int compare_thresholds(const void *a, const void *b) | |
4245 | { | |
4246 | const struct mem_cgroup_threshold *_a = a; | |
4247 | const struct mem_cgroup_threshold *_b = b; | |
4248 | ||
4249 | return _a->threshold - _b->threshold; | |
4250 | } | |
4251 | ||
7d74b06f | 4252 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem) |
9490ff27 KH |
4253 | { |
4254 | struct mem_cgroup_eventfd_list *ev; | |
4255 | ||
4256 | list_for_each_entry(ev, &mem->oom_notify, list) | |
4257 | eventfd_signal(ev->eventfd, 1); | |
4258 | return 0; | |
4259 | } | |
4260 | ||
4261 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) | |
4262 | { | |
7d74b06f KH |
4263 | struct mem_cgroup *iter; |
4264 | ||
4265 | for_each_mem_cgroup_tree(iter, mem) | |
4266 | mem_cgroup_oom_notify_cb(iter); | |
9490ff27 KH |
4267 | } |
4268 | ||
4269 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4270 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4271 | { |
4272 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4273 | struct mem_cgroup_thresholds *thresholds; |
4274 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4275 | int type = MEMFILE_TYPE(cft->private); |
4276 | u64 threshold, usage; | |
2c488db2 | 4277 | int i, size, ret; |
2e72b634 KS |
4278 | |
4279 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4280 | if (ret) | |
4281 | return ret; | |
4282 | ||
4283 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4284 | |
2e72b634 | 4285 | if (type == _MEM) |
2c488db2 | 4286 | thresholds = &memcg->thresholds; |
2e72b634 | 4287 | else if (type == _MEMSWAP) |
2c488db2 | 4288 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4289 | else |
4290 | BUG(); | |
4291 | ||
4292 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4293 | ||
4294 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4295 | if (thresholds->primary) |
2e72b634 KS |
4296 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4297 | ||
2c488db2 | 4298 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4299 | |
4300 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4301 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4302 | GFP_KERNEL); |
2c488db2 | 4303 | if (!new) { |
2e72b634 KS |
4304 | ret = -ENOMEM; |
4305 | goto unlock; | |
4306 | } | |
2c488db2 | 4307 | new->size = size; |
2e72b634 KS |
4308 | |
4309 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4310 | if (thresholds->primary) { |
4311 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4312 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4313 | } |
4314 | ||
2e72b634 | 4315 | /* Add new threshold */ |
2c488db2 KS |
4316 | new->entries[size - 1].eventfd = eventfd; |
4317 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4318 | |
4319 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4320 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4321 | compare_thresholds, NULL); |
4322 | ||
4323 | /* Find current threshold */ | |
2c488db2 | 4324 | new->current_threshold = -1; |
2e72b634 | 4325 | for (i = 0; i < size; i++) { |
2c488db2 | 4326 | if (new->entries[i].threshold < usage) { |
2e72b634 | 4327 | /* |
2c488db2 KS |
4328 | * new->current_threshold will not be used until |
4329 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4330 | * it here. |
4331 | */ | |
2c488db2 | 4332 | ++new->current_threshold; |
2e72b634 KS |
4333 | } |
4334 | } | |
4335 | ||
2c488db2 KS |
4336 | /* Free old spare buffer and save old primary buffer as spare */ |
4337 | kfree(thresholds->spare); | |
4338 | thresholds->spare = thresholds->primary; | |
4339 | ||
4340 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4341 | |
907860ed | 4342 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4343 | synchronize_rcu(); |
4344 | ||
2e72b634 KS |
4345 | unlock: |
4346 | mutex_unlock(&memcg->thresholds_lock); | |
4347 | ||
4348 | return ret; | |
4349 | } | |
4350 | ||
907860ed | 4351 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4352 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4353 | { |
4354 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4355 | struct mem_cgroup_thresholds *thresholds; |
4356 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4357 | int type = MEMFILE_TYPE(cft->private); |
4358 | u64 usage; | |
2c488db2 | 4359 | int i, j, size; |
2e72b634 KS |
4360 | |
4361 | mutex_lock(&memcg->thresholds_lock); | |
4362 | if (type == _MEM) | |
2c488db2 | 4363 | thresholds = &memcg->thresholds; |
2e72b634 | 4364 | else if (type == _MEMSWAP) |
2c488db2 | 4365 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4366 | else |
4367 | BUG(); | |
4368 | ||
4369 | /* | |
4370 | * Something went wrong if we trying to unregister a threshold | |
4371 | * if we don't have thresholds | |
4372 | */ | |
4373 | BUG_ON(!thresholds); | |
4374 | ||
4375 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4376 | ||
4377 | /* Check if a threshold crossed before removing */ | |
4378 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4379 | ||
4380 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4381 | size = 0; |
4382 | for (i = 0; i < thresholds->primary->size; i++) { | |
4383 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4384 | size++; |
4385 | } | |
4386 | ||
2c488db2 | 4387 | new = thresholds->spare; |
907860ed | 4388 | |
2e72b634 KS |
4389 | /* Set thresholds array to NULL if we don't have thresholds */ |
4390 | if (!size) { | |
2c488db2 KS |
4391 | kfree(new); |
4392 | new = NULL; | |
907860ed | 4393 | goto swap_buffers; |
2e72b634 KS |
4394 | } |
4395 | ||
2c488db2 | 4396 | new->size = size; |
2e72b634 KS |
4397 | |
4398 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4399 | new->current_threshold = -1; |
4400 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4401 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4402 | continue; |
4403 | ||
2c488db2 KS |
4404 | new->entries[j] = thresholds->primary->entries[i]; |
4405 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4406 | /* |
2c488db2 | 4407 | * new->current_threshold will not be used |
2e72b634 KS |
4408 | * until rcu_assign_pointer(), so it's safe to increment |
4409 | * it here. | |
4410 | */ | |
2c488db2 | 4411 | ++new->current_threshold; |
2e72b634 KS |
4412 | } |
4413 | j++; | |
4414 | } | |
4415 | ||
907860ed | 4416 | swap_buffers: |
2c488db2 KS |
4417 | /* Swap primary and spare array */ |
4418 | thresholds->spare = thresholds->primary; | |
4419 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4420 | |
907860ed | 4421 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4422 | synchronize_rcu(); |
4423 | ||
2e72b634 | 4424 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4425 | } |
c1e862c1 | 4426 | |
9490ff27 KH |
4427 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4428 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4429 | { | |
4430 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4431 | struct mem_cgroup_eventfd_list *event; | |
4432 | int type = MEMFILE_TYPE(cft->private); | |
4433 | ||
4434 | BUG_ON(type != _OOM_TYPE); | |
4435 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4436 | if (!event) | |
4437 | return -ENOMEM; | |
4438 | ||
4439 | mutex_lock(&memcg_oom_mutex); | |
4440 | ||
4441 | event->eventfd = eventfd; | |
4442 | list_add(&event->list, &memcg->oom_notify); | |
4443 | ||
4444 | /* already in OOM ? */ | |
4445 | if (atomic_read(&memcg->oom_lock)) | |
4446 | eventfd_signal(eventfd, 1); | |
4447 | mutex_unlock(&memcg_oom_mutex); | |
4448 | ||
4449 | return 0; | |
4450 | } | |
4451 | ||
907860ed | 4452 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4453 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4454 | { | |
4455 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4456 | struct mem_cgroup_eventfd_list *ev, *tmp; | |
4457 | int type = MEMFILE_TYPE(cft->private); | |
4458 | ||
4459 | BUG_ON(type != _OOM_TYPE); | |
4460 | ||
4461 | mutex_lock(&memcg_oom_mutex); | |
4462 | ||
4463 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { | |
4464 | if (ev->eventfd == eventfd) { | |
4465 | list_del(&ev->list); | |
4466 | kfree(ev); | |
4467 | } | |
4468 | } | |
4469 | ||
4470 | mutex_unlock(&memcg_oom_mutex); | |
9490ff27 KH |
4471 | } |
4472 | ||
3c11ecf4 KH |
4473 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4474 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4475 | { | |
4476 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4477 | ||
4478 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); | |
4479 | ||
4480 | if (atomic_read(&mem->oom_lock)) | |
4481 | cb->fill(cb, "under_oom", 1); | |
4482 | else | |
4483 | cb->fill(cb, "under_oom", 0); | |
4484 | return 0; | |
4485 | } | |
4486 | ||
3c11ecf4 KH |
4487 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4488 | struct cftype *cft, u64 val) | |
4489 | { | |
4490 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4491 | struct mem_cgroup *parent; | |
4492 | ||
4493 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4494 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4495 | return -EINVAL; | |
4496 | ||
4497 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4498 | ||
4499 | cgroup_lock(); | |
4500 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4501 | if ((parent->use_hierarchy) || | |
4502 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { | |
4503 | cgroup_unlock(); | |
4504 | return -EINVAL; | |
4505 | } | |
4506 | mem->oom_kill_disable = val; | |
4d845ebf KH |
4507 | if (!val) |
4508 | memcg_oom_recover(mem); | |
3c11ecf4 KH |
4509 | cgroup_unlock(); |
4510 | return 0; | |
4511 | } | |
4512 | ||
406eb0c9 YH |
4513 | #ifdef CONFIG_NUMA |
4514 | static const struct file_operations mem_control_numa_stat_file_operations = { | |
4515 | .read = seq_read, | |
4516 | .llseek = seq_lseek, | |
4517 | .release = single_release, | |
4518 | }; | |
4519 | ||
4520 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) | |
4521 | { | |
4522 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
4523 | ||
4524 | file->f_op = &mem_control_numa_stat_file_operations; | |
4525 | return single_open(file, mem_control_numa_stat_show, cont); | |
4526 | } | |
4527 | #endif /* CONFIG_NUMA */ | |
4528 | ||
8cdea7c0 BS |
4529 | static struct cftype mem_cgroup_files[] = { |
4530 | { | |
0eea1030 | 4531 | .name = "usage_in_bytes", |
8c7c6e34 | 4532 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4533 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4534 | .register_event = mem_cgroup_usage_register_event, |
4535 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4536 | }, |
c84872e1 PE |
4537 | { |
4538 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4539 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4540 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4541 | .read_u64 = mem_cgroup_read, |
4542 | }, | |
8cdea7c0 | 4543 | { |
0eea1030 | 4544 | .name = "limit_in_bytes", |
8c7c6e34 | 4545 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4546 | .write_string = mem_cgroup_write, |
2c3daa72 | 4547 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4548 | }, |
296c81d8 BS |
4549 | { |
4550 | .name = "soft_limit_in_bytes", | |
4551 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4552 | .write_string = mem_cgroup_write, | |
4553 | .read_u64 = mem_cgroup_read, | |
4554 | }, | |
8cdea7c0 BS |
4555 | { |
4556 | .name = "failcnt", | |
8c7c6e34 | 4557 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4558 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4559 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4560 | }, |
d2ceb9b7 KH |
4561 | { |
4562 | .name = "stat", | |
c64745cf | 4563 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4564 | }, |
c1e862c1 KH |
4565 | { |
4566 | .name = "force_empty", | |
4567 | .trigger = mem_cgroup_force_empty_write, | |
4568 | }, | |
18f59ea7 BS |
4569 | { |
4570 | .name = "use_hierarchy", | |
4571 | .write_u64 = mem_cgroup_hierarchy_write, | |
4572 | .read_u64 = mem_cgroup_hierarchy_read, | |
4573 | }, | |
a7885eb8 KM |
4574 | { |
4575 | .name = "swappiness", | |
4576 | .read_u64 = mem_cgroup_swappiness_read, | |
4577 | .write_u64 = mem_cgroup_swappiness_write, | |
4578 | }, | |
7dc74be0 DN |
4579 | { |
4580 | .name = "move_charge_at_immigrate", | |
4581 | .read_u64 = mem_cgroup_move_charge_read, | |
4582 | .write_u64 = mem_cgroup_move_charge_write, | |
4583 | }, | |
9490ff27 KH |
4584 | { |
4585 | .name = "oom_control", | |
3c11ecf4 KH |
4586 | .read_map = mem_cgroup_oom_control_read, |
4587 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4588 | .register_event = mem_cgroup_oom_register_event, |
4589 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4590 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4591 | }, | |
406eb0c9 YH |
4592 | #ifdef CONFIG_NUMA |
4593 | { | |
4594 | .name = "numa_stat", | |
4595 | .open = mem_control_numa_stat_open, | |
4596 | }, | |
4597 | #endif | |
8cdea7c0 BS |
4598 | }; |
4599 | ||
8c7c6e34 KH |
4600 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4601 | static struct cftype memsw_cgroup_files[] = { | |
4602 | { | |
4603 | .name = "memsw.usage_in_bytes", | |
4604 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4605 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4606 | .register_event = mem_cgroup_usage_register_event, |
4607 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4608 | }, |
4609 | { | |
4610 | .name = "memsw.max_usage_in_bytes", | |
4611 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4612 | .trigger = mem_cgroup_reset, | |
4613 | .read_u64 = mem_cgroup_read, | |
4614 | }, | |
4615 | { | |
4616 | .name = "memsw.limit_in_bytes", | |
4617 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4618 | .write_string = mem_cgroup_write, | |
4619 | .read_u64 = mem_cgroup_read, | |
4620 | }, | |
4621 | { | |
4622 | .name = "memsw.failcnt", | |
4623 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4624 | .trigger = mem_cgroup_reset, | |
4625 | .read_u64 = mem_cgroup_read, | |
4626 | }, | |
4627 | }; | |
4628 | ||
4629 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4630 | { | |
4631 | if (!do_swap_account) | |
4632 | return 0; | |
4633 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4634 | ARRAY_SIZE(memsw_cgroup_files)); | |
4635 | }; | |
4636 | #else | |
4637 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4638 | { | |
4639 | return 0; | |
4640 | } | |
4641 | #endif | |
4642 | ||
6d12e2d8 KH |
4643 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4644 | { | |
4645 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4646 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 4647 | enum lru_list l; |
41e3355d | 4648 | int zone, tmp = node; |
1ecaab2b KH |
4649 | /* |
4650 | * This routine is called against possible nodes. | |
4651 | * But it's BUG to call kmalloc() against offline node. | |
4652 | * | |
4653 | * TODO: this routine can waste much memory for nodes which will | |
4654 | * never be onlined. It's better to use memory hotplug callback | |
4655 | * function. | |
4656 | */ | |
41e3355d KH |
4657 | if (!node_state(node, N_NORMAL_MEMORY)) |
4658 | tmp = -1; | |
17295c88 | 4659 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4660 | if (!pn) |
4661 | return 1; | |
1ecaab2b | 4662 | |
6d12e2d8 | 4663 | mem->info.nodeinfo[node] = pn; |
1ecaab2b KH |
4664 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4665 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
4666 | for_each_lru(l) |
4667 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 4668 | mz->usage_in_excess = 0; |
4e416953 BS |
4669 | mz->on_tree = false; |
4670 | mz->mem = mem; | |
1ecaab2b | 4671 | } |
6d12e2d8 KH |
4672 | return 0; |
4673 | } | |
4674 | ||
1ecaab2b KH |
4675 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4676 | { | |
4677 | kfree(mem->info.nodeinfo[node]); | |
4678 | } | |
4679 | ||
33327948 KH |
4680 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4681 | { | |
4682 | struct mem_cgroup *mem; | |
c62b1a3b | 4683 | int size = sizeof(struct mem_cgroup); |
33327948 | 4684 | |
c62b1a3b | 4685 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4686 | if (size < PAGE_SIZE) |
17295c88 | 4687 | mem = kzalloc(size, GFP_KERNEL); |
33327948 | 4688 | else |
17295c88 | 4689 | mem = vzalloc(size); |
33327948 | 4690 | |
e7bbcdf3 DC |
4691 | if (!mem) |
4692 | return NULL; | |
4693 | ||
c62b1a3b | 4694 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
d2e61b8d DC |
4695 | if (!mem->stat) |
4696 | goto out_free; | |
711d3d2c | 4697 | spin_lock_init(&mem->pcp_counter_lock); |
33327948 | 4698 | return mem; |
d2e61b8d DC |
4699 | |
4700 | out_free: | |
4701 | if (size < PAGE_SIZE) | |
4702 | kfree(mem); | |
4703 | else | |
4704 | vfree(mem); | |
4705 | return NULL; | |
33327948 KH |
4706 | } |
4707 | ||
8c7c6e34 KH |
4708 | /* |
4709 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4710 | * (scanning all at force_empty is too costly...) | |
4711 | * | |
4712 | * Instead of clearing all references at force_empty, we remember | |
4713 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4714 | * it goes down to 0. | |
4715 | * | |
8c7c6e34 KH |
4716 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4717 | */ | |
4718 | ||
a7ba0eef | 4719 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 4720 | { |
08e552c6 KH |
4721 | int node; |
4722 | ||
f64c3f54 | 4723 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
4724 | free_css_id(&mem_cgroup_subsys, &mem->css); |
4725 | ||
08e552c6 KH |
4726 | for_each_node_state(node, N_POSSIBLE) |
4727 | free_mem_cgroup_per_zone_info(mem, node); | |
4728 | ||
c62b1a3b KH |
4729 | free_percpu(mem->stat); |
4730 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
4731 | kfree(mem); |
4732 | else | |
4733 | vfree(mem); | |
4734 | } | |
4735 | ||
8c7c6e34 KH |
4736 | static void mem_cgroup_get(struct mem_cgroup *mem) |
4737 | { | |
4738 | atomic_inc(&mem->refcnt); | |
4739 | } | |
4740 | ||
483c30b5 | 4741 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 4742 | { |
483c30b5 | 4743 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 4744 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 4745 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
4746 | if (parent) |
4747 | mem_cgroup_put(parent); | |
4748 | } | |
8c7c6e34 KH |
4749 | } |
4750 | ||
483c30b5 DN |
4751 | static void mem_cgroup_put(struct mem_cgroup *mem) |
4752 | { | |
4753 | __mem_cgroup_put(mem, 1); | |
4754 | } | |
4755 | ||
7bcc1bb1 DN |
4756 | /* |
4757 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4758 | */ | |
4759 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
4760 | { | |
4761 | if (!mem->res.parent) | |
4762 | return NULL; | |
4763 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
4764 | } | |
33327948 | 4765 | |
c077719b KH |
4766 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4767 | static void __init enable_swap_cgroup(void) | |
4768 | { | |
f8d66542 | 4769 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4770 | do_swap_account = 1; |
4771 | } | |
4772 | #else | |
4773 | static void __init enable_swap_cgroup(void) | |
4774 | { | |
4775 | } | |
4776 | #endif | |
4777 | ||
f64c3f54 BS |
4778 | static int mem_cgroup_soft_limit_tree_init(void) |
4779 | { | |
4780 | struct mem_cgroup_tree_per_node *rtpn; | |
4781 | struct mem_cgroup_tree_per_zone *rtpz; | |
4782 | int tmp, node, zone; | |
4783 | ||
4784 | for_each_node_state(node, N_POSSIBLE) { | |
4785 | tmp = node; | |
4786 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4787 | tmp = -1; | |
4788 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4789 | if (!rtpn) | |
4790 | return 1; | |
4791 | ||
4792 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4793 | ||
4794 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4795 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4796 | rtpz->rb_root = RB_ROOT; | |
4797 | spin_lock_init(&rtpz->lock); | |
4798 | } | |
4799 | } | |
4800 | return 0; | |
4801 | } | |
4802 | ||
0eb253e2 | 4803 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4804 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4805 | { | |
28dbc4b6 | 4806 | struct mem_cgroup *mem, *parent; |
04046e1a | 4807 | long error = -ENOMEM; |
6d12e2d8 | 4808 | int node; |
8cdea7c0 | 4809 | |
c8dad2bb JB |
4810 | mem = mem_cgroup_alloc(); |
4811 | if (!mem) | |
04046e1a | 4812 | return ERR_PTR(error); |
78fb7466 | 4813 | |
6d12e2d8 KH |
4814 | for_each_node_state(node, N_POSSIBLE) |
4815 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
4816 | goto free_out; | |
f64c3f54 | 4817 | |
c077719b | 4818 | /* root ? */ |
28dbc4b6 | 4819 | if (cont->parent == NULL) { |
cdec2e42 | 4820 | int cpu; |
c077719b | 4821 | enable_swap_cgroup(); |
28dbc4b6 | 4822 | parent = NULL; |
4b3bde4c | 4823 | root_mem_cgroup = mem; |
f64c3f54 BS |
4824 | if (mem_cgroup_soft_limit_tree_init()) |
4825 | goto free_out; | |
cdec2e42 KH |
4826 | for_each_possible_cpu(cpu) { |
4827 | struct memcg_stock_pcp *stock = | |
4828 | &per_cpu(memcg_stock, cpu); | |
4829 | INIT_WORK(&stock->work, drain_local_stock); | |
4830 | } | |
711d3d2c | 4831 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4832 | } else { |
28dbc4b6 | 4833 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 | 4834 | mem->use_hierarchy = parent->use_hierarchy; |
3c11ecf4 | 4835 | mem->oom_kill_disable = parent->oom_kill_disable; |
18f59ea7 | 4836 | } |
28dbc4b6 | 4837 | |
18f59ea7 BS |
4838 | if (parent && parent->use_hierarchy) { |
4839 | res_counter_init(&mem->res, &parent->res); | |
4840 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4841 | /* |
4842 | * We increment refcnt of the parent to ensure that we can | |
4843 | * safely access it on res_counter_charge/uncharge. | |
4844 | * This refcnt will be decremented when freeing this | |
4845 | * mem_cgroup(see mem_cgroup_put). | |
4846 | */ | |
4847 | mem_cgroup_get(parent); | |
18f59ea7 BS |
4848 | } else { |
4849 | res_counter_init(&mem->res, NULL); | |
4850 | res_counter_init(&mem->memsw, NULL); | |
4851 | } | |
04046e1a | 4852 | mem->last_scanned_child = 0; |
889976db | 4853 | mem->last_scanned_node = MAX_NUMNODES; |
9490ff27 | 4854 | INIT_LIST_HEAD(&mem->oom_notify); |
6d61ef40 | 4855 | |
a7885eb8 KM |
4856 | if (parent) |
4857 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 4858 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 4859 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 4860 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 4861 | return &mem->css; |
6d12e2d8 | 4862 | free_out: |
a7ba0eef | 4863 | __mem_cgroup_free(mem); |
4b3bde4c | 4864 | root_mem_cgroup = NULL; |
04046e1a | 4865 | return ERR_PTR(error); |
8cdea7c0 BS |
4866 | } |
4867 | ||
ec64f515 | 4868 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4869 | struct cgroup *cont) |
4870 | { | |
4871 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
4872 | |
4873 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
4874 | } |
4875 | ||
8cdea7c0 BS |
4876 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4877 | struct cgroup *cont) | |
4878 | { | |
c268e994 | 4879 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 4880 | |
c268e994 | 4881 | mem_cgroup_put(mem); |
8cdea7c0 BS |
4882 | } |
4883 | ||
4884 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4885 | struct cgroup *cont) | |
4886 | { | |
8c7c6e34 KH |
4887 | int ret; |
4888 | ||
4889 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4890 | ARRAY_SIZE(mem_cgroup_files)); | |
4891 | ||
4892 | if (!ret) | |
4893 | ret = register_memsw_files(cont, ss); | |
4894 | return ret; | |
8cdea7c0 BS |
4895 | } |
4896 | ||
02491447 | 4897 | #ifdef CONFIG_MMU |
7dc74be0 | 4898 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4899 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4900 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4901 | { |
854ffa8d DN |
4902 | int ret = 0; |
4903 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
4904 | struct mem_cgroup *mem = mc.to; |
4905 | ||
854ffa8d DN |
4906 | if (mem_cgroup_is_root(mem)) { |
4907 | mc.precharge += count; | |
4908 | /* we don't need css_get for root */ | |
4909 | return ret; | |
4910 | } | |
4911 | /* try to charge at once */ | |
4912 | if (count > 1) { | |
4913 | struct res_counter *dummy; | |
4914 | /* | |
4915 | * "mem" cannot be under rmdir() because we've already checked | |
4916 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
4917 | * are still under the same cgroup_mutex. So we can postpone | |
4918 | * css_get(). | |
4919 | */ | |
4920 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
4921 | goto one_by_one; | |
4922 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
4923 | PAGE_SIZE * count, &dummy)) { | |
4924 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
4925 | goto one_by_one; | |
4926 | } | |
4927 | mc.precharge += count; | |
854ffa8d DN |
4928 | return ret; |
4929 | } | |
4930 | one_by_one: | |
4931 | /* fall back to one by one charge */ | |
4932 | while (count--) { | |
4933 | if (signal_pending(current)) { | |
4934 | ret = -EINTR; | |
4935 | break; | |
4936 | } | |
4937 | if (!batch_count--) { | |
4938 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4939 | cond_resched(); | |
4940 | } | |
7ec99d62 | 4941 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, 1, &mem, false); |
854ffa8d DN |
4942 | if (ret || !mem) |
4943 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
4944 | return -ENOMEM; | |
4945 | mc.precharge++; | |
4946 | } | |
4ffef5fe DN |
4947 | return ret; |
4948 | } | |
4949 | ||
4950 | /** | |
4951 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
4952 | * @vma: the vma the pte to be checked belongs | |
4953 | * @addr: the address corresponding to the pte to be checked | |
4954 | * @ptent: the pte to be checked | |
02491447 | 4955 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
4956 | * |
4957 | * Returns | |
4958 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
4959 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
4960 | * move charge. if @target is not NULL, the page is stored in target->page | |
4961 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
4962 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
4963 | * target for charge migration. if @target is not NULL, the entry is stored | |
4964 | * in target->ent. | |
4ffef5fe DN |
4965 | * |
4966 | * Called with pte lock held. | |
4967 | */ | |
4ffef5fe DN |
4968 | union mc_target { |
4969 | struct page *page; | |
02491447 | 4970 | swp_entry_t ent; |
4ffef5fe DN |
4971 | }; |
4972 | ||
4ffef5fe DN |
4973 | enum mc_target_type { |
4974 | MC_TARGET_NONE, /* not used */ | |
4975 | MC_TARGET_PAGE, | |
02491447 | 4976 | MC_TARGET_SWAP, |
4ffef5fe DN |
4977 | }; |
4978 | ||
90254a65 DN |
4979 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
4980 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 4981 | { |
90254a65 | 4982 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 4983 | |
90254a65 DN |
4984 | if (!page || !page_mapped(page)) |
4985 | return NULL; | |
4986 | if (PageAnon(page)) { | |
4987 | /* we don't move shared anon */ | |
4988 | if (!move_anon() || page_mapcount(page) > 2) | |
4989 | return NULL; | |
87946a72 DN |
4990 | } else if (!move_file()) |
4991 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
4992 | return NULL; |
4993 | if (!get_page_unless_zero(page)) | |
4994 | return NULL; | |
4995 | ||
4996 | return page; | |
4997 | } | |
4998 | ||
4999 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5000 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5001 | { | |
5002 | int usage_count; | |
5003 | struct page *page = NULL; | |
5004 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5005 | ||
5006 | if (!move_anon() || non_swap_entry(ent)) | |
5007 | return NULL; | |
5008 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
5009 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
5010 | if (page) |
5011 | put_page(page); | |
90254a65 | 5012 | return NULL; |
02491447 | 5013 | } |
90254a65 DN |
5014 | if (do_swap_account) |
5015 | entry->val = ent.val; | |
5016 | ||
5017 | return page; | |
5018 | } | |
5019 | ||
87946a72 DN |
5020 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5021 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5022 | { | |
5023 | struct page *page = NULL; | |
5024 | struct inode *inode; | |
5025 | struct address_space *mapping; | |
5026 | pgoff_t pgoff; | |
5027 | ||
5028 | if (!vma->vm_file) /* anonymous vma */ | |
5029 | return NULL; | |
5030 | if (!move_file()) | |
5031 | return NULL; | |
5032 | ||
5033 | inode = vma->vm_file->f_path.dentry->d_inode; | |
5034 | mapping = vma->vm_file->f_mapping; | |
5035 | if (pte_none(ptent)) | |
5036 | pgoff = linear_page_index(vma, addr); | |
5037 | else /* pte_file(ptent) is true */ | |
5038 | pgoff = pte_to_pgoff(ptent); | |
5039 | ||
5040 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
5041 | if (!mapping_cap_swap_backed(mapping)) { /* normal file */ | |
5042 | page = find_get_page(mapping, pgoff); | |
5043 | } else { /* shmem/tmpfs file. we should take account of swap too. */ | |
5044 | swp_entry_t ent; | |
5045 | mem_cgroup_get_shmem_target(inode, pgoff, &page, &ent); | |
5046 | if (do_swap_account) | |
5047 | entry->val = ent.val; | |
5048 | } | |
5049 | ||
5050 | return page; | |
5051 | } | |
5052 | ||
90254a65 DN |
5053 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
5054 | unsigned long addr, pte_t ptent, union mc_target *target) | |
5055 | { | |
5056 | struct page *page = NULL; | |
5057 | struct page_cgroup *pc; | |
5058 | int ret = 0; | |
5059 | swp_entry_t ent = { .val = 0 }; | |
5060 | ||
5061 | if (pte_present(ptent)) | |
5062 | page = mc_handle_present_pte(vma, addr, ptent); | |
5063 | else if (is_swap_pte(ptent)) | |
5064 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5065 | else if (pte_none(ptent) || pte_file(ptent)) |
5066 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5067 | |
5068 | if (!page && !ent.val) | |
5069 | return 0; | |
02491447 DN |
5070 | if (page) { |
5071 | pc = lookup_page_cgroup(page); | |
5072 | /* | |
5073 | * Do only loose check w/o page_cgroup lock. | |
5074 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5075 | * the lock. | |
5076 | */ | |
5077 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5078 | ret = MC_TARGET_PAGE; | |
5079 | if (target) | |
5080 | target->page = page; | |
5081 | } | |
5082 | if (!ret || !target) | |
5083 | put_page(page); | |
5084 | } | |
90254a65 DN |
5085 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5086 | if (ent.val && !ret && | |
7f0f1546 KH |
5087 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
5088 | ret = MC_TARGET_SWAP; | |
5089 | if (target) | |
5090 | target->ent = ent; | |
4ffef5fe | 5091 | } |
4ffef5fe DN |
5092 | return ret; |
5093 | } | |
5094 | ||
5095 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
5096 | unsigned long addr, unsigned long end, | |
5097 | struct mm_walk *walk) | |
5098 | { | |
5099 | struct vm_area_struct *vma = walk->private; | |
5100 | pte_t *pte; | |
5101 | spinlock_t *ptl; | |
5102 | ||
03319327 DH |
5103 | split_huge_page_pmd(walk->mm, pmd); |
5104 | ||
4ffef5fe DN |
5105 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5106 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
5107 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
5108 | mc.precharge++; /* increment precharge temporarily */ | |
5109 | pte_unmap_unlock(pte - 1, ptl); | |
5110 | cond_resched(); | |
5111 | ||
7dc74be0 DN |
5112 | return 0; |
5113 | } | |
5114 | ||
4ffef5fe DN |
5115 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5116 | { | |
5117 | unsigned long precharge; | |
5118 | struct vm_area_struct *vma; | |
5119 | ||
dfe076b0 | 5120 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5121 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5122 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5123 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5124 | .mm = mm, | |
5125 | .private = vma, | |
5126 | }; | |
5127 | if (is_vm_hugetlb_page(vma)) | |
5128 | continue; | |
4ffef5fe DN |
5129 | walk_page_range(vma->vm_start, vma->vm_end, |
5130 | &mem_cgroup_count_precharge_walk); | |
5131 | } | |
dfe076b0 | 5132 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5133 | |
5134 | precharge = mc.precharge; | |
5135 | mc.precharge = 0; | |
5136 | ||
5137 | return precharge; | |
5138 | } | |
5139 | ||
4ffef5fe DN |
5140 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5141 | { | |
dfe076b0 DN |
5142 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5143 | ||
5144 | VM_BUG_ON(mc.moving_task); | |
5145 | mc.moving_task = current; | |
5146 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5147 | } |
5148 | ||
dfe076b0 DN |
5149 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5150 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5151 | { |
2bd9bb20 KH |
5152 | struct mem_cgroup *from = mc.from; |
5153 | struct mem_cgroup *to = mc.to; | |
5154 | ||
4ffef5fe | 5155 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5156 | if (mc.precharge) { |
5157 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5158 | mc.precharge = 0; | |
5159 | } | |
5160 | /* | |
5161 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5162 | * we must uncharge here. | |
5163 | */ | |
5164 | if (mc.moved_charge) { | |
5165 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5166 | mc.moved_charge = 0; | |
4ffef5fe | 5167 | } |
483c30b5 DN |
5168 | /* we must fixup refcnts and charges */ |
5169 | if (mc.moved_swap) { | |
483c30b5 DN |
5170 | /* uncharge swap account from the old cgroup */ |
5171 | if (!mem_cgroup_is_root(mc.from)) | |
5172 | res_counter_uncharge(&mc.from->memsw, | |
5173 | PAGE_SIZE * mc.moved_swap); | |
5174 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5175 | ||
5176 | if (!mem_cgroup_is_root(mc.to)) { | |
5177 | /* | |
5178 | * we charged both to->res and to->memsw, so we should | |
5179 | * uncharge to->res. | |
5180 | */ | |
5181 | res_counter_uncharge(&mc.to->res, | |
5182 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5183 | } |
5184 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5185 | mc.moved_swap = 0; |
5186 | } | |
dfe076b0 DN |
5187 | memcg_oom_recover(from); |
5188 | memcg_oom_recover(to); | |
5189 | wake_up_all(&mc.waitq); | |
5190 | } | |
5191 | ||
5192 | static void mem_cgroup_clear_mc(void) | |
5193 | { | |
5194 | struct mem_cgroup *from = mc.from; | |
5195 | ||
5196 | /* | |
5197 | * we must clear moving_task before waking up waiters at the end of | |
5198 | * task migration. | |
5199 | */ | |
5200 | mc.moving_task = NULL; | |
5201 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5202 | spin_lock(&mc.lock); |
4ffef5fe DN |
5203 | mc.from = NULL; |
5204 | mc.to = NULL; | |
2bd9bb20 | 5205 | spin_unlock(&mc.lock); |
32047e2a | 5206 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5207 | } |
5208 | ||
7dc74be0 DN |
5209 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
5210 | struct cgroup *cgroup, | |
f780bdb7 | 5211 | struct task_struct *p) |
7dc74be0 DN |
5212 | { |
5213 | int ret = 0; | |
5214 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
5215 | ||
5216 | if (mem->move_charge_at_immigrate) { | |
5217 | struct mm_struct *mm; | |
5218 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5219 | ||
5220 | VM_BUG_ON(from == mem); | |
5221 | ||
5222 | mm = get_task_mm(p); | |
5223 | if (!mm) | |
5224 | return 0; | |
7dc74be0 | 5225 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5226 | if (mm->owner == p) { |
5227 | VM_BUG_ON(mc.from); | |
5228 | VM_BUG_ON(mc.to); | |
5229 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5230 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5231 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5232 | mem_cgroup_start_move(from); |
2bd9bb20 | 5233 | spin_lock(&mc.lock); |
4ffef5fe DN |
5234 | mc.from = from; |
5235 | mc.to = mem; | |
2bd9bb20 | 5236 | spin_unlock(&mc.lock); |
dfe076b0 | 5237 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5238 | |
5239 | ret = mem_cgroup_precharge_mc(mm); | |
5240 | if (ret) | |
5241 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5242 | } |
5243 | mmput(mm); | |
7dc74be0 DN |
5244 | } |
5245 | return ret; | |
5246 | } | |
5247 | ||
5248 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5249 | struct cgroup *cgroup, | |
f780bdb7 | 5250 | struct task_struct *p) |
7dc74be0 | 5251 | { |
4ffef5fe | 5252 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5253 | } |
5254 | ||
4ffef5fe DN |
5255 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5256 | unsigned long addr, unsigned long end, | |
5257 | struct mm_walk *walk) | |
7dc74be0 | 5258 | { |
4ffef5fe DN |
5259 | int ret = 0; |
5260 | struct vm_area_struct *vma = walk->private; | |
5261 | pte_t *pte; | |
5262 | spinlock_t *ptl; | |
5263 | ||
03319327 | 5264 | split_huge_page_pmd(walk->mm, pmd); |
4ffef5fe DN |
5265 | retry: |
5266 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5267 | for (; addr != end; addr += PAGE_SIZE) { | |
5268 | pte_t ptent = *(pte++); | |
5269 | union mc_target target; | |
5270 | int type; | |
5271 | struct page *page; | |
5272 | struct page_cgroup *pc; | |
02491447 | 5273 | swp_entry_t ent; |
4ffef5fe DN |
5274 | |
5275 | if (!mc.precharge) | |
5276 | break; | |
5277 | ||
5278 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
5279 | switch (type) { | |
5280 | case MC_TARGET_PAGE: | |
5281 | page = target.page; | |
5282 | if (isolate_lru_page(page)) | |
5283 | goto put; | |
5284 | pc = lookup_page_cgroup(page); | |
7ec99d62 JW |
5285 | if (!mem_cgroup_move_account(page, 1, pc, |
5286 | mc.from, mc.to, false)) { | |
4ffef5fe | 5287 | mc.precharge--; |
854ffa8d DN |
5288 | /* we uncharge from mc.from later. */ |
5289 | mc.moved_charge++; | |
4ffef5fe DN |
5290 | } |
5291 | putback_lru_page(page); | |
5292 | put: /* is_target_pte_for_mc() gets the page */ | |
5293 | put_page(page); | |
5294 | break; | |
02491447 DN |
5295 | case MC_TARGET_SWAP: |
5296 | ent = target.ent; | |
483c30b5 DN |
5297 | if (!mem_cgroup_move_swap_account(ent, |
5298 | mc.from, mc.to, false)) { | |
02491447 | 5299 | mc.precharge--; |
483c30b5 DN |
5300 | /* we fixup refcnts and charges later. */ |
5301 | mc.moved_swap++; | |
5302 | } | |
02491447 | 5303 | break; |
4ffef5fe DN |
5304 | default: |
5305 | break; | |
5306 | } | |
5307 | } | |
5308 | pte_unmap_unlock(pte - 1, ptl); | |
5309 | cond_resched(); | |
5310 | ||
5311 | if (addr != end) { | |
5312 | /* | |
5313 | * We have consumed all precharges we got in can_attach(). | |
5314 | * We try charge one by one, but don't do any additional | |
5315 | * charges to mc.to if we have failed in charge once in attach() | |
5316 | * phase. | |
5317 | */ | |
854ffa8d | 5318 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5319 | if (!ret) |
5320 | goto retry; | |
5321 | } | |
5322 | ||
5323 | return ret; | |
5324 | } | |
5325 | ||
5326 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5327 | { | |
5328 | struct vm_area_struct *vma; | |
5329 | ||
5330 | lru_add_drain_all(); | |
dfe076b0 DN |
5331 | retry: |
5332 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5333 | /* | |
5334 | * Someone who are holding the mmap_sem might be waiting in | |
5335 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5336 | * and retry. Because we cancel precharges, we might not be able | |
5337 | * to move enough charges, but moving charge is a best-effort | |
5338 | * feature anyway, so it wouldn't be a big problem. | |
5339 | */ | |
5340 | __mem_cgroup_clear_mc(); | |
5341 | cond_resched(); | |
5342 | goto retry; | |
5343 | } | |
4ffef5fe DN |
5344 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5345 | int ret; | |
5346 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5347 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5348 | .mm = mm, | |
5349 | .private = vma, | |
5350 | }; | |
5351 | if (is_vm_hugetlb_page(vma)) | |
5352 | continue; | |
4ffef5fe DN |
5353 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5354 | &mem_cgroup_move_charge_walk); | |
5355 | if (ret) | |
5356 | /* | |
5357 | * means we have consumed all precharges and failed in | |
5358 | * doing additional charge. Just abandon here. | |
5359 | */ | |
5360 | break; | |
5361 | } | |
dfe076b0 | 5362 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5363 | } |
5364 | ||
67e465a7 BS |
5365 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
5366 | struct cgroup *cont, | |
5367 | struct cgroup *old_cont, | |
f780bdb7 | 5368 | struct task_struct *p) |
67e465a7 | 5369 | { |
dfe076b0 DN |
5370 | struct mm_struct *mm; |
5371 | ||
5372 | if (!mc.to) | |
4ffef5fe DN |
5373 | /* no need to move charge */ |
5374 | return; | |
5375 | ||
dfe076b0 DN |
5376 | mm = get_task_mm(p); |
5377 | if (mm) { | |
5378 | mem_cgroup_move_charge(mm); | |
5379 | mmput(mm); | |
5380 | } | |
4ffef5fe | 5381 | mem_cgroup_clear_mc(); |
67e465a7 | 5382 | } |
5cfb80a7 DN |
5383 | #else /* !CONFIG_MMU */ |
5384 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
5385 | struct cgroup *cgroup, | |
f780bdb7 | 5386 | struct task_struct *p) |
5cfb80a7 DN |
5387 | { |
5388 | return 0; | |
5389 | } | |
5390 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5391 | struct cgroup *cgroup, | |
f780bdb7 | 5392 | struct task_struct *p) |
5cfb80a7 DN |
5393 | { |
5394 | } | |
5395 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
5396 | struct cgroup *cont, | |
5397 | struct cgroup *old_cont, | |
f780bdb7 | 5398 | struct task_struct *p) |
5cfb80a7 DN |
5399 | { |
5400 | } | |
5401 | #endif | |
67e465a7 | 5402 | |
8cdea7c0 BS |
5403 | struct cgroup_subsys mem_cgroup_subsys = { |
5404 | .name = "memory", | |
5405 | .subsys_id = mem_cgroup_subsys_id, | |
5406 | .create = mem_cgroup_create, | |
df878fb0 | 5407 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
5408 | .destroy = mem_cgroup_destroy, |
5409 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
5410 | .can_attach = mem_cgroup_can_attach, |
5411 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5412 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5413 | .early_init = 0, |
04046e1a | 5414 | .use_id = 1, |
8cdea7c0 | 5415 | }; |
c077719b KH |
5416 | |
5417 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5418 | static int __init enable_swap_account(char *s) |
5419 | { | |
5420 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5421 | if (!strcmp(s, "1")) |
a42c390c | 5422 | really_do_swap_account = 1; |
a2c8990a | 5423 | else if (!strcmp(s, "0")) |
a42c390c MH |
5424 | really_do_swap_account = 0; |
5425 | return 1; | |
5426 | } | |
a2c8990a | 5427 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5428 | |
c077719b | 5429 | #endif |