]>
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 */ | |
456f998e YH |
97 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
98 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ | |
e9f8974f JW |
99 | MEM_CGROUP_EVENTS_NSTATS, |
100 | }; | |
7a159cc9 JW |
101 | /* |
102 | * Per memcg event counter is incremented at every pagein/pageout. With THP, | |
103 | * it will be incremated by the number of pages. This counter is used for | |
104 | * for trigger some periodic events. This is straightforward and better | |
105 | * than using jiffies etc. to handle periodic memcg event. | |
106 | */ | |
107 | enum mem_cgroup_events_target { | |
108 | MEM_CGROUP_TARGET_THRESH, | |
109 | MEM_CGROUP_TARGET_SOFTLIMIT, | |
453a9bf3 | 110 | MEM_CGROUP_TARGET_NUMAINFO, |
7a159cc9 JW |
111 | MEM_CGROUP_NTARGETS, |
112 | }; | |
113 | #define THRESHOLDS_EVENTS_TARGET (128) | |
114 | #define SOFTLIMIT_EVENTS_TARGET (1024) | |
453a9bf3 | 115 | #define NUMAINFO_EVENTS_TARGET (1024) |
e9f8974f | 116 | |
d52aa412 | 117 | struct mem_cgroup_stat_cpu { |
7a159cc9 | 118 | long count[MEM_CGROUP_STAT_NSTATS]; |
e9f8974f | 119 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
7a159cc9 | 120 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
d52aa412 KH |
121 | }; |
122 | ||
6d12e2d8 KH |
123 | /* |
124 | * per-zone information in memory controller. | |
125 | */ | |
6d12e2d8 | 126 | struct mem_cgroup_per_zone { |
072c56c1 KH |
127 | /* |
128 | * spin_lock to protect the per cgroup LRU | |
129 | */ | |
b69408e8 CL |
130 | struct list_head lists[NR_LRU_LISTS]; |
131 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
132 | |
133 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
134 | struct rb_node tree_node; /* RB tree node */ |
135 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
136 | /* the soft limit is exceeded*/ | |
137 | bool on_tree; | |
4e416953 BS |
138 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
139 | /* use container_of */ | |
6d12e2d8 KH |
140 | }; |
141 | /* Macro for accessing counter */ | |
142 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
143 | ||
144 | struct mem_cgroup_per_node { | |
145 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
146 | }; | |
147 | ||
148 | struct mem_cgroup_lru_info { | |
149 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
150 | }; | |
151 | ||
f64c3f54 BS |
152 | /* |
153 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
154 | * their hierarchy representation | |
155 | */ | |
156 | ||
157 | struct mem_cgroup_tree_per_zone { | |
158 | struct rb_root rb_root; | |
159 | spinlock_t lock; | |
160 | }; | |
161 | ||
162 | struct mem_cgroup_tree_per_node { | |
163 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
164 | }; | |
165 | ||
166 | struct mem_cgroup_tree { | |
167 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
168 | }; | |
169 | ||
170 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
171 | ||
2e72b634 KS |
172 | struct mem_cgroup_threshold { |
173 | struct eventfd_ctx *eventfd; | |
174 | u64 threshold; | |
175 | }; | |
176 | ||
9490ff27 | 177 | /* For threshold */ |
2e72b634 KS |
178 | struct mem_cgroup_threshold_ary { |
179 | /* An array index points to threshold just below usage. */ | |
5407a562 | 180 | int current_threshold; |
2e72b634 KS |
181 | /* Size of entries[] */ |
182 | unsigned int size; | |
183 | /* Array of thresholds */ | |
184 | struct mem_cgroup_threshold entries[0]; | |
185 | }; | |
2c488db2 KS |
186 | |
187 | struct mem_cgroup_thresholds { | |
188 | /* Primary thresholds array */ | |
189 | struct mem_cgroup_threshold_ary *primary; | |
190 | /* | |
191 | * Spare threshold array. | |
192 | * This is needed to make mem_cgroup_unregister_event() "never fail". | |
193 | * It must be able to store at least primary->size - 1 entries. | |
194 | */ | |
195 | struct mem_cgroup_threshold_ary *spare; | |
196 | }; | |
197 | ||
9490ff27 KH |
198 | /* for OOM */ |
199 | struct mem_cgroup_eventfd_list { | |
200 | struct list_head list; | |
201 | struct eventfd_ctx *eventfd; | |
202 | }; | |
2e72b634 | 203 | |
2e72b634 | 204 | static void mem_cgroup_threshold(struct mem_cgroup *mem); |
9490ff27 | 205 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem); |
2e72b634 | 206 | |
8cdea7c0 BS |
207 | /* |
208 | * The memory controller data structure. The memory controller controls both | |
209 | * page cache and RSS per cgroup. We would eventually like to provide | |
210 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
211 | * to help the administrator determine what knobs to tune. | |
212 | * | |
213 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
214 | * we hit the water mark. May be even add a low water mark, such that |
215 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
216 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
217 | */ |
218 | struct mem_cgroup { | |
219 | struct cgroup_subsys_state css; | |
220 | /* | |
221 | * the counter to account for memory usage | |
222 | */ | |
223 | struct res_counter res; | |
8c7c6e34 KH |
224 | /* |
225 | * the counter to account for mem+swap usage. | |
226 | */ | |
227 | struct res_counter memsw; | |
78fb7466 PE |
228 | /* |
229 | * Per cgroup active and inactive list, similar to the | |
230 | * per zone LRU lists. | |
78fb7466 | 231 | */ |
6d12e2d8 | 232 | struct mem_cgroup_lru_info info; |
6d61ef40 | 233 | /* |
af901ca1 | 234 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 235 | * reclaimed from. |
6d61ef40 | 236 | */ |
04046e1a | 237 | int last_scanned_child; |
889976db YH |
238 | int last_scanned_node; |
239 | #if MAX_NUMNODES > 1 | |
240 | nodemask_t scan_nodes; | |
453a9bf3 KH |
241 | atomic_t numainfo_events; |
242 | atomic_t numainfo_updating; | |
889976db | 243 | #endif |
18f59ea7 BS |
244 | /* |
245 | * Should the accounting and control be hierarchical, per subtree? | |
246 | */ | |
247 | bool use_hierarchy; | |
79dfdacc MH |
248 | |
249 | bool oom_lock; | |
250 | atomic_t under_oom; | |
251 | ||
8c7c6e34 | 252 | atomic_t refcnt; |
14797e23 | 253 | |
1f4c025b | 254 | int swappiness; |
3c11ecf4 KH |
255 | /* OOM-Killer disable */ |
256 | int oom_kill_disable; | |
a7885eb8 | 257 | |
22a668d7 KH |
258 | /* set when res.limit == memsw.limit */ |
259 | bool memsw_is_minimum; | |
260 | ||
2e72b634 KS |
261 | /* protect arrays of thresholds */ |
262 | struct mutex thresholds_lock; | |
263 | ||
264 | /* thresholds for memory usage. RCU-protected */ | |
2c488db2 | 265 | struct mem_cgroup_thresholds thresholds; |
907860ed | 266 | |
2e72b634 | 267 | /* thresholds for mem+swap usage. RCU-protected */ |
2c488db2 | 268 | struct mem_cgroup_thresholds memsw_thresholds; |
907860ed | 269 | |
9490ff27 KH |
270 | /* For oom notifier event fd */ |
271 | struct list_head oom_notify; | |
185efc0f | 272 | |
7dc74be0 DN |
273 | /* |
274 | * Should we move charges of a task when a task is moved into this | |
275 | * mem_cgroup ? And what type of charges should we move ? | |
276 | */ | |
277 | unsigned long move_charge_at_immigrate; | |
d52aa412 | 278 | /* |
c62b1a3b | 279 | * percpu counter. |
d52aa412 | 280 | */ |
c62b1a3b | 281 | struct mem_cgroup_stat_cpu *stat; |
711d3d2c KH |
282 | /* |
283 | * used when a cpu is offlined or other synchronizations | |
284 | * See mem_cgroup_read_stat(). | |
285 | */ | |
286 | struct mem_cgroup_stat_cpu nocpu_base; | |
287 | spinlock_t pcp_counter_lock; | |
8cdea7c0 BS |
288 | }; |
289 | ||
7dc74be0 DN |
290 | /* Stuffs for move charges at task migration. */ |
291 | /* | |
292 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
293 | * left-shifted bitmap of these types. | |
294 | */ | |
295 | enum move_type { | |
4ffef5fe | 296 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
87946a72 | 297 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
7dc74be0 DN |
298 | NR_MOVE_TYPE, |
299 | }; | |
300 | ||
4ffef5fe DN |
301 | /* "mc" and its members are protected by cgroup_mutex */ |
302 | static struct move_charge_struct { | |
b1dd693e | 303 | spinlock_t lock; /* for from, to */ |
4ffef5fe DN |
304 | struct mem_cgroup *from; |
305 | struct mem_cgroup *to; | |
306 | unsigned long precharge; | |
854ffa8d | 307 | unsigned long moved_charge; |
483c30b5 | 308 | unsigned long moved_swap; |
8033b97c DN |
309 | struct task_struct *moving_task; /* a task moving charges */ |
310 | wait_queue_head_t waitq; /* a waitq for other context */ | |
311 | } mc = { | |
2bd9bb20 | 312 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
8033b97c DN |
313 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
314 | }; | |
4ffef5fe | 315 | |
90254a65 DN |
316 | static bool move_anon(void) |
317 | { | |
318 | return test_bit(MOVE_CHARGE_TYPE_ANON, | |
319 | &mc.to->move_charge_at_immigrate); | |
320 | } | |
321 | ||
87946a72 DN |
322 | static bool move_file(void) |
323 | { | |
324 | return test_bit(MOVE_CHARGE_TYPE_FILE, | |
325 | &mc.to->move_charge_at_immigrate); | |
326 | } | |
327 | ||
4e416953 BS |
328 | /* |
329 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
330 | * limit reclaim to prevent infinite loops, if they ever occur. | |
331 | */ | |
332 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
333 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
334 | ||
217bc319 KH |
335 | enum charge_type { |
336 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
337 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 338 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 339 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 340 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 341 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
342 | NR_CHARGE_TYPE, |
343 | }; | |
344 | ||
8c7c6e34 KH |
345 | /* for encoding cft->private value on file */ |
346 | #define _MEM (0) | |
347 | #define _MEMSWAP (1) | |
9490ff27 | 348 | #define _OOM_TYPE (2) |
8c7c6e34 KH |
349 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
350 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
351 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
9490ff27 KH |
352 | /* Used for OOM nofiier */ |
353 | #define OOM_CONTROL (0) | |
8c7c6e34 | 354 | |
75822b44 BS |
355 | /* |
356 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
357 | */ | |
358 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
359 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
360 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
361 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
362 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
363 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 364 | |
8c7c6e34 KH |
365 | static void mem_cgroup_get(struct mem_cgroup *mem); |
366 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 367 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
26fe6168 | 368 | static void drain_all_stock_async(struct mem_cgroup *mem); |
8c7c6e34 | 369 | |
f64c3f54 BS |
370 | static struct mem_cgroup_per_zone * |
371 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
372 | { | |
373 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
374 | } | |
375 | ||
d324236b WF |
376 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
377 | { | |
378 | return &mem->css; | |
379 | } | |
380 | ||
f64c3f54 | 381 | static struct mem_cgroup_per_zone * |
97a6c37b | 382 | page_cgroup_zoneinfo(struct mem_cgroup *mem, struct page *page) |
f64c3f54 | 383 | { |
97a6c37b JW |
384 | int nid = page_to_nid(page); |
385 | int zid = page_zonenum(page); | |
f64c3f54 | 386 | |
f64c3f54 BS |
387 | return mem_cgroup_zoneinfo(mem, nid, zid); |
388 | } | |
389 | ||
390 | static struct mem_cgroup_tree_per_zone * | |
391 | soft_limit_tree_node_zone(int nid, int zid) | |
392 | { | |
393 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
394 | } | |
395 | ||
396 | static struct mem_cgroup_tree_per_zone * | |
397 | soft_limit_tree_from_page(struct page *page) | |
398 | { | |
399 | int nid = page_to_nid(page); | |
400 | int zid = page_zonenum(page); | |
401 | ||
402 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
403 | } | |
404 | ||
405 | static void | |
4e416953 | 406 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 407 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
408 | struct mem_cgroup_tree_per_zone *mctz, |
409 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
410 | { |
411 | struct rb_node **p = &mctz->rb_root.rb_node; | |
412 | struct rb_node *parent = NULL; | |
413 | struct mem_cgroup_per_zone *mz_node; | |
414 | ||
415 | if (mz->on_tree) | |
416 | return; | |
417 | ||
ef8745c1 KH |
418 | mz->usage_in_excess = new_usage_in_excess; |
419 | if (!mz->usage_in_excess) | |
420 | return; | |
f64c3f54 BS |
421 | while (*p) { |
422 | parent = *p; | |
423 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
424 | tree_node); | |
425 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
426 | p = &(*p)->rb_left; | |
427 | /* | |
428 | * We can't avoid mem cgroups that are over their soft | |
429 | * limit by the same amount | |
430 | */ | |
431 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
432 | p = &(*p)->rb_right; | |
433 | } | |
434 | rb_link_node(&mz->tree_node, parent, p); | |
435 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
436 | mz->on_tree = true; | |
4e416953 BS |
437 | } |
438 | ||
439 | static void | |
440 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
441 | struct mem_cgroup_per_zone *mz, | |
442 | struct mem_cgroup_tree_per_zone *mctz) | |
443 | { | |
444 | if (!mz->on_tree) | |
445 | return; | |
446 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
447 | mz->on_tree = false; | |
448 | } | |
449 | ||
f64c3f54 BS |
450 | static void |
451 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
452 | struct mem_cgroup_per_zone *mz, | |
453 | struct mem_cgroup_tree_per_zone *mctz) | |
454 | { | |
455 | spin_lock(&mctz->lock); | |
4e416953 | 456 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
457 | spin_unlock(&mctz->lock); |
458 | } | |
459 | ||
f64c3f54 BS |
460 | |
461 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
462 | { | |
ef8745c1 | 463 | unsigned long long excess; |
f64c3f54 BS |
464 | struct mem_cgroup_per_zone *mz; |
465 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
466 | int nid = page_to_nid(page); |
467 | int zid = page_zonenum(page); | |
f64c3f54 BS |
468 | mctz = soft_limit_tree_from_page(page); |
469 | ||
470 | /* | |
4e649152 KH |
471 | * Necessary to update all ancestors when hierarchy is used. |
472 | * because their event counter is not touched. | |
f64c3f54 | 473 | */ |
4e649152 KH |
474 | for (; mem; mem = parent_mem_cgroup(mem)) { |
475 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 476 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
477 | /* |
478 | * We have to update the tree if mz is on RB-tree or | |
479 | * mem is over its softlimit. | |
480 | */ | |
ef8745c1 | 481 | if (excess || mz->on_tree) { |
4e649152 KH |
482 | spin_lock(&mctz->lock); |
483 | /* if on-tree, remove it */ | |
484 | if (mz->on_tree) | |
485 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
486 | /* | |
ef8745c1 KH |
487 | * Insert again. mz->usage_in_excess will be updated. |
488 | * If excess is 0, no tree ops. | |
4e649152 | 489 | */ |
ef8745c1 | 490 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
491 | spin_unlock(&mctz->lock); |
492 | } | |
f64c3f54 BS |
493 | } |
494 | } | |
495 | ||
496 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
497 | { | |
498 | int node, zone; | |
499 | struct mem_cgroup_per_zone *mz; | |
500 | struct mem_cgroup_tree_per_zone *mctz; | |
501 | ||
502 | for_each_node_state(node, N_POSSIBLE) { | |
503 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
504 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
505 | mctz = soft_limit_tree_node_zone(node, zone); | |
506 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
507 | } | |
508 | } | |
509 | } | |
510 | ||
4e416953 BS |
511 | static struct mem_cgroup_per_zone * |
512 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
513 | { | |
514 | struct rb_node *rightmost = NULL; | |
26251eaf | 515 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
516 | |
517 | retry: | |
26251eaf | 518 | mz = NULL; |
4e416953 BS |
519 | rightmost = rb_last(&mctz->rb_root); |
520 | if (!rightmost) | |
521 | goto done; /* Nothing to reclaim from */ | |
522 | ||
523 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
524 | /* | |
525 | * Remove the node now but someone else can add it back, | |
526 | * we will to add it back at the end of reclaim to its correct | |
527 | * position in the tree. | |
528 | */ | |
529 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
530 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
531 | !css_tryget(&mz->mem->css)) | |
532 | goto retry; | |
533 | done: | |
534 | return mz; | |
535 | } | |
536 | ||
537 | static struct mem_cgroup_per_zone * | |
538 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
539 | { | |
540 | struct mem_cgroup_per_zone *mz; | |
541 | ||
542 | spin_lock(&mctz->lock); | |
543 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
544 | spin_unlock(&mctz->lock); | |
545 | return mz; | |
546 | } | |
547 | ||
711d3d2c KH |
548 | /* |
549 | * Implementation Note: reading percpu statistics for memcg. | |
550 | * | |
551 | * Both of vmstat[] and percpu_counter has threshold and do periodic | |
552 | * synchronization to implement "quick" read. There are trade-off between | |
553 | * reading cost and precision of value. Then, we may have a chance to implement | |
554 | * a periodic synchronizion of counter in memcg's counter. | |
555 | * | |
556 | * But this _read() function is used for user interface now. The user accounts | |
557 | * memory usage by memory cgroup and he _always_ requires exact value because | |
558 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always | |
559 | * have to visit all online cpus and make sum. So, for now, unnecessary | |
560 | * synchronization is not implemented. (just implemented for cpu hotplug) | |
561 | * | |
562 | * If there are kernel internal actions which can make use of some not-exact | |
563 | * value, and reading all cpu value can be performance bottleneck in some | |
564 | * common workload, threashold and synchonization as vmstat[] should be | |
565 | * implemented. | |
566 | */ | |
7a159cc9 JW |
567 | static long mem_cgroup_read_stat(struct mem_cgroup *mem, |
568 | enum mem_cgroup_stat_index idx) | |
c62b1a3b | 569 | { |
7a159cc9 | 570 | long val = 0; |
c62b1a3b | 571 | int cpu; |
c62b1a3b | 572 | |
711d3d2c KH |
573 | get_online_cpus(); |
574 | for_each_online_cpu(cpu) | |
c62b1a3b | 575 | val += per_cpu(mem->stat->count[idx], cpu); |
711d3d2c KH |
576 | #ifdef CONFIG_HOTPLUG_CPU |
577 | spin_lock(&mem->pcp_counter_lock); | |
578 | val += mem->nocpu_base.count[idx]; | |
579 | spin_unlock(&mem->pcp_counter_lock); | |
580 | #endif | |
581 | put_online_cpus(); | |
c62b1a3b KH |
582 | return val; |
583 | } | |
584 | ||
0c3e73e8 BS |
585 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
586 | bool charge) | |
587 | { | |
588 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 589 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
590 | } |
591 | ||
456f998e YH |
592 | void mem_cgroup_pgfault(struct mem_cgroup *mem, int val) |
593 | { | |
594 | this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_PGFAULT], val); | |
595 | } | |
596 | ||
597 | void mem_cgroup_pgmajfault(struct mem_cgroup *mem, int val) | |
598 | { | |
599 | this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT], val); | |
600 | } | |
601 | ||
e9f8974f JW |
602 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *mem, |
603 | enum mem_cgroup_events_index idx) | |
604 | { | |
605 | unsigned long val = 0; | |
606 | int cpu; | |
607 | ||
608 | for_each_online_cpu(cpu) | |
609 | val += per_cpu(mem->stat->events[idx], cpu); | |
610 | #ifdef CONFIG_HOTPLUG_CPU | |
611 | spin_lock(&mem->pcp_counter_lock); | |
612 | val += mem->nocpu_base.events[idx]; | |
613 | spin_unlock(&mem->pcp_counter_lock); | |
614 | #endif | |
615 | return val; | |
616 | } | |
617 | ||
c05555b5 | 618 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
e401f176 | 619 | bool file, int nr_pages) |
d52aa412 | 620 | { |
c62b1a3b KH |
621 | preempt_disable(); |
622 | ||
e401f176 KH |
623 | if (file) |
624 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages); | |
d52aa412 | 625 | else |
e401f176 | 626 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages); |
55e462b0 | 627 | |
e401f176 KH |
628 | /* pagein of a big page is an event. So, ignore page size */ |
629 | if (nr_pages > 0) | |
e9f8974f | 630 | __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
3751d604 | 631 | else { |
e9f8974f | 632 | __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
3751d604 KH |
633 | nr_pages = -nr_pages; /* for event */ |
634 | } | |
e401f176 | 635 | |
e9f8974f | 636 | __this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); |
2e72b634 | 637 | |
c62b1a3b | 638 | preempt_enable(); |
6d12e2d8 KH |
639 | } |
640 | ||
bb2a0de9 KH |
641 | unsigned long |
642 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *mem, int nid, int zid, | |
643 | unsigned int lru_mask) | |
889976db YH |
644 | { |
645 | struct mem_cgroup_per_zone *mz; | |
bb2a0de9 KH |
646 | enum lru_list l; |
647 | unsigned long ret = 0; | |
648 | ||
649 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
650 | ||
651 | for_each_lru(l) { | |
652 | if (BIT(l) & lru_mask) | |
653 | ret += MEM_CGROUP_ZSTAT(mz, l); | |
654 | } | |
655 | return ret; | |
656 | } | |
657 | ||
658 | static unsigned long | |
659 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *mem, | |
660 | int nid, unsigned int lru_mask) | |
661 | { | |
889976db YH |
662 | u64 total = 0; |
663 | int zid; | |
664 | ||
bb2a0de9 KH |
665 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
666 | total += mem_cgroup_zone_nr_lru_pages(mem, nid, zid, lru_mask); | |
667 | ||
889976db YH |
668 | return total; |
669 | } | |
bb2a0de9 KH |
670 | |
671 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *mem, | |
672 | unsigned int lru_mask) | |
6d12e2d8 | 673 | { |
889976db | 674 | int nid; |
6d12e2d8 KH |
675 | u64 total = 0; |
676 | ||
bb2a0de9 KH |
677 | for_each_node_state(nid, N_HIGH_MEMORY) |
678 | total += mem_cgroup_node_nr_lru_pages(mem, nid, lru_mask); | |
6d12e2d8 | 679 | return total; |
d52aa412 KH |
680 | } |
681 | ||
7a159cc9 JW |
682 | static bool __memcg_event_check(struct mem_cgroup *mem, int target) |
683 | { | |
684 | unsigned long val, next; | |
685 | ||
686 | val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]); | |
687 | next = this_cpu_read(mem->stat->targets[target]); | |
688 | /* from time_after() in jiffies.h */ | |
689 | return ((long)next - (long)val < 0); | |
690 | } | |
691 | ||
692 | static void __mem_cgroup_target_update(struct mem_cgroup *mem, int target) | |
d2265e6f | 693 | { |
7a159cc9 | 694 | unsigned long val, next; |
d2265e6f | 695 | |
e9f8974f | 696 | val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]); |
d2265e6f | 697 | |
7a159cc9 JW |
698 | switch (target) { |
699 | case MEM_CGROUP_TARGET_THRESH: | |
700 | next = val + THRESHOLDS_EVENTS_TARGET; | |
701 | break; | |
702 | case MEM_CGROUP_TARGET_SOFTLIMIT: | |
703 | next = val + SOFTLIMIT_EVENTS_TARGET; | |
704 | break; | |
453a9bf3 KH |
705 | case MEM_CGROUP_TARGET_NUMAINFO: |
706 | next = val + NUMAINFO_EVENTS_TARGET; | |
707 | break; | |
7a159cc9 JW |
708 | default: |
709 | return; | |
710 | } | |
711 | ||
712 | this_cpu_write(mem->stat->targets[target], next); | |
d2265e6f KH |
713 | } |
714 | ||
715 | /* | |
716 | * Check events in order. | |
717 | * | |
718 | */ | |
719 | static void memcg_check_events(struct mem_cgroup *mem, struct page *page) | |
720 | { | |
721 | /* threshold event is triggered in finer grain than soft limit */ | |
7a159cc9 | 722 | if (unlikely(__memcg_event_check(mem, MEM_CGROUP_TARGET_THRESH))) { |
d2265e6f | 723 | mem_cgroup_threshold(mem); |
7a159cc9 JW |
724 | __mem_cgroup_target_update(mem, MEM_CGROUP_TARGET_THRESH); |
725 | if (unlikely(__memcg_event_check(mem, | |
453a9bf3 | 726 | MEM_CGROUP_TARGET_SOFTLIMIT))) { |
d2265e6f | 727 | mem_cgroup_update_tree(mem, page); |
7a159cc9 | 728 | __mem_cgroup_target_update(mem, |
453a9bf3 KH |
729 | MEM_CGROUP_TARGET_SOFTLIMIT); |
730 | } | |
731 | #if MAX_NUMNODES > 1 | |
732 | if (unlikely(__memcg_event_check(mem, | |
733 | MEM_CGROUP_TARGET_NUMAINFO))) { | |
734 | atomic_inc(&mem->numainfo_events); | |
735 | __mem_cgroup_target_update(mem, | |
736 | MEM_CGROUP_TARGET_NUMAINFO); | |
7a159cc9 | 737 | } |
453a9bf3 | 738 | #endif |
d2265e6f KH |
739 | } |
740 | } | |
741 | ||
d5b69e38 | 742 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
743 | { |
744 | return container_of(cgroup_subsys_state(cont, | |
745 | mem_cgroup_subsys_id), struct mem_cgroup, | |
746 | css); | |
747 | } | |
748 | ||
cf475ad2 | 749 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 750 | { |
31a78f23 BS |
751 | /* |
752 | * mm_update_next_owner() may clear mm->owner to NULL | |
753 | * if it races with swapoff, page migration, etc. | |
754 | * So this can be called with p == NULL. | |
755 | */ | |
756 | if (unlikely(!p)) | |
757 | return NULL; | |
758 | ||
78fb7466 PE |
759 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
760 | struct mem_cgroup, css); | |
761 | } | |
762 | ||
a433658c | 763 | struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
54595fe2 KH |
764 | { |
765 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
766 | |
767 | if (!mm) | |
768 | return NULL; | |
54595fe2 KH |
769 | /* |
770 | * Because we have no locks, mm->owner's may be being moved to other | |
771 | * cgroup. We use css_tryget() here even if this looks | |
772 | * pessimistic (rather than adding locks here). | |
773 | */ | |
774 | rcu_read_lock(); | |
775 | do { | |
776 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
777 | if (unlikely(!mem)) | |
778 | break; | |
779 | } while (!css_tryget(&mem->css)); | |
780 | rcu_read_unlock(); | |
781 | return mem; | |
782 | } | |
783 | ||
7d74b06f KH |
784 | /* The caller has to guarantee "mem" exists before calling this */ |
785 | static struct mem_cgroup *mem_cgroup_start_loop(struct mem_cgroup *mem) | |
14067bb3 | 786 | { |
711d3d2c KH |
787 | struct cgroup_subsys_state *css; |
788 | int found; | |
789 | ||
790 | if (!mem) /* ROOT cgroup has the smallest ID */ | |
791 | return root_mem_cgroup; /*css_put/get against root is ignored*/ | |
792 | if (!mem->use_hierarchy) { | |
793 | if (css_tryget(&mem->css)) | |
794 | return mem; | |
795 | return NULL; | |
796 | } | |
797 | rcu_read_lock(); | |
798 | /* | |
799 | * searching a memory cgroup which has the smallest ID under given | |
800 | * ROOT cgroup. (ID >= 1) | |
801 | */ | |
802 | css = css_get_next(&mem_cgroup_subsys, 1, &mem->css, &found); | |
803 | if (css && css_tryget(css)) | |
804 | mem = container_of(css, struct mem_cgroup, css); | |
805 | else | |
806 | mem = NULL; | |
807 | rcu_read_unlock(); | |
808 | return mem; | |
7d74b06f KH |
809 | } |
810 | ||
811 | static struct mem_cgroup *mem_cgroup_get_next(struct mem_cgroup *iter, | |
812 | struct mem_cgroup *root, | |
813 | bool cond) | |
814 | { | |
815 | int nextid = css_id(&iter->css) + 1; | |
816 | int found; | |
817 | int hierarchy_used; | |
14067bb3 | 818 | struct cgroup_subsys_state *css; |
14067bb3 | 819 | |
7d74b06f | 820 | hierarchy_used = iter->use_hierarchy; |
14067bb3 | 821 | |
7d74b06f | 822 | css_put(&iter->css); |
711d3d2c KH |
823 | /* If no ROOT, walk all, ignore hierarchy */ |
824 | if (!cond || (root && !hierarchy_used)) | |
7d74b06f | 825 | return NULL; |
14067bb3 | 826 | |
711d3d2c KH |
827 | if (!root) |
828 | root = root_mem_cgroup; | |
829 | ||
7d74b06f KH |
830 | do { |
831 | iter = NULL; | |
14067bb3 | 832 | rcu_read_lock(); |
7d74b06f KH |
833 | |
834 | css = css_get_next(&mem_cgroup_subsys, nextid, | |
835 | &root->css, &found); | |
14067bb3 | 836 | if (css && css_tryget(css)) |
7d74b06f | 837 | iter = container_of(css, struct mem_cgroup, css); |
14067bb3 | 838 | rcu_read_unlock(); |
7d74b06f | 839 | /* If css is NULL, no more cgroups will be found */ |
14067bb3 | 840 | nextid = found + 1; |
7d74b06f | 841 | } while (css && !iter); |
14067bb3 | 842 | |
7d74b06f | 843 | return iter; |
14067bb3 | 844 | } |
7d74b06f KH |
845 | /* |
846 | * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please | |
847 | * be careful that "break" loop is not allowed. We have reference count. | |
848 | * Instead of that modify "cond" to be false and "continue" to exit the loop. | |
849 | */ | |
850 | #define for_each_mem_cgroup_tree_cond(iter, root, cond) \ | |
851 | for (iter = mem_cgroup_start_loop(root);\ | |
852 | iter != NULL;\ | |
853 | iter = mem_cgroup_get_next(iter, root, cond)) | |
854 | ||
855 | #define for_each_mem_cgroup_tree(iter, root) \ | |
856 | for_each_mem_cgroup_tree_cond(iter, root, true) | |
857 | ||
711d3d2c KH |
858 | #define for_each_mem_cgroup_all(iter) \ |
859 | for_each_mem_cgroup_tree_cond(iter, NULL, true) | |
860 | ||
14067bb3 | 861 | |
4b3bde4c BS |
862 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
863 | { | |
864 | return (mem == root_mem_cgroup); | |
865 | } | |
866 | ||
456f998e YH |
867 | void mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
868 | { | |
869 | struct mem_cgroup *mem; | |
870 | ||
871 | if (!mm) | |
872 | return; | |
873 | ||
874 | rcu_read_lock(); | |
875 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
876 | if (unlikely(!mem)) | |
877 | goto out; | |
878 | ||
879 | switch (idx) { | |
880 | case PGMAJFAULT: | |
881 | mem_cgroup_pgmajfault(mem, 1); | |
882 | break; | |
883 | case PGFAULT: | |
884 | mem_cgroup_pgfault(mem, 1); | |
885 | break; | |
886 | default: | |
887 | BUG(); | |
888 | } | |
889 | out: | |
890 | rcu_read_unlock(); | |
891 | } | |
892 | EXPORT_SYMBOL(mem_cgroup_count_vm_event); | |
893 | ||
08e552c6 KH |
894 | /* |
895 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
896 | * Operations are called by routine of global LRU independently from memcg. | |
897 | * What we have to take care of here is validness of pc->mem_cgroup. | |
898 | * | |
899 | * Changes to pc->mem_cgroup happens when | |
900 | * 1. charge | |
901 | * 2. moving account | |
902 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
903 | * It is added to LRU before charge. | |
904 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
905 | * When moving account, the page is not on LRU. It's isolated. | |
906 | */ | |
4f98a2fe | 907 | |
08e552c6 KH |
908 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
909 | { | |
910 | struct page_cgroup *pc; | |
08e552c6 | 911 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 912 | |
f8d66542 | 913 | if (mem_cgroup_disabled()) |
08e552c6 KH |
914 | return; |
915 | pc = lookup_page_cgroup(page); | |
916 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 917 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 918 | return; |
4b3bde4c | 919 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
920 | /* |
921 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
922 | * removed from global LRU. | |
923 | */ | |
97a6c37b | 924 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
925 | /* huge page split is done under lru_lock. so, we have no races. */ |
926 | MEM_CGROUP_ZSTAT(mz, lru) -= 1 << compound_order(page); | |
4b3bde4c BS |
927 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
928 | return; | |
929 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 | 930 | list_del_init(&pc->lru); |
6d12e2d8 KH |
931 | } |
932 | ||
08e552c6 | 933 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 934 | { |
08e552c6 KH |
935 | mem_cgroup_del_lru_list(page, page_lru(page)); |
936 | } | |
b69408e8 | 937 | |
3f58a829 MK |
938 | /* |
939 | * Writeback is about to end against a page which has been marked for immediate | |
940 | * reclaim. If it still appears to be reclaimable, move it to the tail of the | |
941 | * inactive list. | |
942 | */ | |
943 | void mem_cgroup_rotate_reclaimable_page(struct page *page) | |
944 | { | |
945 | struct mem_cgroup_per_zone *mz; | |
946 | struct page_cgroup *pc; | |
947 | enum lru_list lru = page_lru(page); | |
948 | ||
949 | if (mem_cgroup_disabled()) | |
950 | return; | |
951 | ||
952 | pc = lookup_page_cgroup(page); | |
953 | /* unused or root page is not rotated. */ | |
954 | if (!PageCgroupUsed(pc)) | |
955 | return; | |
956 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
957 | smp_rmb(); | |
958 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
959 | return; | |
97a6c37b | 960 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3f58a829 MK |
961 | list_move_tail(&pc->lru, &mz->lists[lru]); |
962 | } | |
963 | ||
08e552c6 KH |
964 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
965 | { | |
966 | struct mem_cgroup_per_zone *mz; | |
967 | struct page_cgroup *pc; | |
b69408e8 | 968 | |
f8d66542 | 969 | if (mem_cgroup_disabled()) |
08e552c6 | 970 | return; |
6d12e2d8 | 971 | |
08e552c6 | 972 | pc = lookup_page_cgroup(page); |
4b3bde4c | 973 | /* unused or root page is not rotated. */ |
713735b4 JW |
974 | if (!PageCgroupUsed(pc)) |
975 | return; | |
976 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ | |
977 | smp_rmb(); | |
978 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 | 979 | return; |
97a6c37b | 980 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
08e552c6 | 981 | list_move(&pc->lru, &mz->lists[lru]); |
6d12e2d8 KH |
982 | } |
983 | ||
08e552c6 | 984 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 985 | { |
08e552c6 KH |
986 | struct page_cgroup *pc; |
987 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 988 | |
f8d66542 | 989 | if (mem_cgroup_disabled()) |
08e552c6 KH |
990 | return; |
991 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 992 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
08e552c6 | 993 | if (!PageCgroupUsed(pc)) |
894bc310 | 994 | return; |
713735b4 JW |
995 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
996 | smp_rmb(); | |
97a6c37b | 997 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
ece35ca8 KH |
998 | /* huge page split is done under lru_lock. so, we have no races. */ |
999 | MEM_CGROUP_ZSTAT(mz, lru) += 1 << compound_order(page); | |
4b3bde4c BS |
1000 | SetPageCgroupAcctLRU(pc); |
1001 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
1002 | return; | |
08e552c6 KH |
1003 | list_add(&pc->lru, &mz->lists[lru]); |
1004 | } | |
544122e5 | 1005 | |
08e552c6 | 1006 | /* |
5a6475a4 KH |
1007 | * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed |
1008 | * while it's linked to lru because the page may be reused after it's fully | |
1009 | * uncharged. To handle that, unlink page_cgroup from LRU when charge it again. | |
1010 | * It's done under lock_page and expected that zone->lru_lock isnever held. | |
08e552c6 | 1011 | */ |
5a6475a4 | 1012 | static void mem_cgroup_lru_del_before_commit(struct page *page) |
08e552c6 | 1013 | { |
544122e5 KH |
1014 | unsigned long flags; |
1015 | struct zone *zone = page_zone(page); | |
1016 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1017 | ||
5a6475a4 KH |
1018 | /* |
1019 | * Doing this check without taking ->lru_lock seems wrong but this | |
1020 | * is safe. Because if page_cgroup's USED bit is unset, the page | |
1021 | * will not be added to any memcg's LRU. If page_cgroup's USED bit is | |
1022 | * set, the commit after this will fail, anyway. | |
1023 | * This all charge/uncharge is done under some mutual execustion. | |
1024 | * So, we don't need to taking care of changes in USED bit. | |
1025 | */ | |
1026 | if (likely(!PageLRU(page))) | |
1027 | return; | |
1028 | ||
544122e5 KH |
1029 | spin_lock_irqsave(&zone->lru_lock, flags); |
1030 | /* | |
1031 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
1032 | * is guarded by lock_page() because the page is SwapCache. | |
1033 | */ | |
1034 | if (!PageCgroupUsed(pc)) | |
1035 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
1036 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
1037 | } |
1038 | ||
5a6475a4 | 1039 | static void mem_cgroup_lru_add_after_commit(struct page *page) |
544122e5 KH |
1040 | { |
1041 | unsigned long flags; | |
1042 | struct zone *zone = page_zone(page); | |
1043 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
1044 | ||
5a6475a4 KH |
1045 | /* taking care of that the page is added to LRU while we commit it */ |
1046 | if (likely(!PageLRU(page))) | |
1047 | return; | |
544122e5 KH |
1048 | spin_lock_irqsave(&zone->lru_lock, flags); |
1049 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 1050 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
1051 | mem_cgroup_add_lru_list(page, page_lru(page)); |
1052 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
1053 | } | |
1054 | ||
1055 | ||
08e552c6 KH |
1056 | void mem_cgroup_move_lists(struct page *page, |
1057 | enum lru_list from, enum lru_list to) | |
1058 | { | |
f8d66542 | 1059 | if (mem_cgroup_disabled()) |
08e552c6 KH |
1060 | return; |
1061 | mem_cgroup_del_lru_list(page, from); | |
1062 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
1063 | } |
1064 | ||
3e92041d MH |
1065 | /* |
1066 | * Checks whether given mem is same or in the root_mem's | |
1067 | * hierarchy subtree | |
1068 | */ | |
1069 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_mem, | |
1070 | struct mem_cgroup *mem) | |
1071 | { | |
1072 | if (root_mem != mem) { | |
1073 | return (root_mem->use_hierarchy && | |
1074 | css_is_ancestor(&mem->css, &root_mem->css)); | |
1075 | } | |
1076 | ||
1077 | return true; | |
1078 | } | |
1079 | ||
4c4a2214 DR |
1080 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
1081 | { | |
1082 | int ret; | |
0b7f569e | 1083 | struct mem_cgroup *curr = NULL; |
158e0a2d | 1084 | struct task_struct *p; |
4c4a2214 | 1085 | |
158e0a2d KH |
1086 | p = find_lock_task_mm(task); |
1087 | if (!p) | |
1088 | return 0; | |
1089 | curr = try_get_mem_cgroup_from_mm(p->mm); | |
1090 | task_unlock(p); | |
0b7f569e KH |
1091 | if (!curr) |
1092 | return 0; | |
d31f56db DN |
1093 | /* |
1094 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
1095 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
1096 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
1097 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
1098 | */ | |
3e92041d | 1099 | ret = mem_cgroup_same_or_subtree(mem, curr); |
0b7f569e | 1100 | css_put(&curr->css); |
4c4a2214 DR |
1101 | return ret; |
1102 | } | |
1103 | ||
c772be93 | 1104 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
1105 | { |
1106 | unsigned long active; | |
1107 | unsigned long inactive; | |
c772be93 KM |
1108 | unsigned long gb; |
1109 | unsigned long inactive_ratio; | |
14797e23 | 1110 | |
bb2a0de9 KH |
1111 | inactive = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_ANON)); |
1112 | active = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_ANON)); | |
14797e23 | 1113 | |
c772be93 KM |
1114 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
1115 | if (gb) | |
1116 | inactive_ratio = int_sqrt(10 * gb); | |
1117 | else | |
1118 | inactive_ratio = 1; | |
1119 | ||
1120 | if (present_pages) { | |
1121 | present_pages[0] = inactive; | |
1122 | present_pages[1] = active; | |
1123 | } | |
1124 | ||
1125 | return inactive_ratio; | |
1126 | } | |
1127 | ||
1128 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
1129 | { | |
1130 | unsigned long active; | |
1131 | unsigned long inactive; | |
1132 | unsigned long present_pages[2]; | |
1133 | unsigned long inactive_ratio; | |
1134 | ||
1135 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
1136 | ||
1137 | inactive = present_pages[0]; | |
1138 | active = present_pages[1]; | |
1139 | ||
1140 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
1141 | return 1; |
1142 | ||
1143 | return 0; | |
1144 | } | |
1145 | ||
56e49d21 RR |
1146 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
1147 | { | |
1148 | unsigned long active; | |
1149 | unsigned long inactive; | |
1150 | ||
bb2a0de9 KH |
1151 | inactive = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_INACTIVE_FILE)); |
1152 | active = mem_cgroup_nr_lru_pages(memcg, BIT(LRU_ACTIVE_FILE)); | |
56e49d21 RR |
1153 | |
1154 | return (active > inactive); | |
1155 | } | |
1156 | ||
3e2f41f1 KM |
1157 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
1158 | struct zone *zone) | |
1159 | { | |
13d7e3a2 | 1160 | int nid = zone_to_nid(zone); |
3e2f41f1 KM |
1161 | int zid = zone_idx(zone); |
1162 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
1163 | ||
1164 | return &mz->reclaim_stat; | |
1165 | } | |
1166 | ||
1167 | struct zone_reclaim_stat * | |
1168 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
1169 | { | |
1170 | struct page_cgroup *pc; | |
1171 | struct mem_cgroup_per_zone *mz; | |
1172 | ||
1173 | if (mem_cgroup_disabled()) | |
1174 | return NULL; | |
1175 | ||
1176 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
1177 | if (!PageCgroupUsed(pc)) |
1178 | return NULL; | |
713735b4 JW |
1179 | /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */ |
1180 | smp_rmb(); | |
97a6c37b | 1181 | mz = page_cgroup_zoneinfo(pc->mem_cgroup, page); |
3e2f41f1 KM |
1182 | return &mz->reclaim_stat; |
1183 | } | |
1184 | ||
66e1707b BS |
1185 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
1186 | struct list_head *dst, | |
1187 | unsigned long *scanned, int order, | |
4356f21d MK |
1188 | isolate_mode_t mode, |
1189 | struct zone *z, | |
66e1707b | 1190 | struct mem_cgroup *mem_cont, |
4f98a2fe | 1191 | int active, int file) |
66e1707b BS |
1192 | { |
1193 | unsigned long nr_taken = 0; | |
1194 | struct page *page; | |
1195 | unsigned long scan; | |
1196 | LIST_HEAD(pc_list); | |
1197 | struct list_head *src; | |
ff7283fa | 1198 | struct page_cgroup *pc, *tmp; |
13d7e3a2 | 1199 | int nid = zone_to_nid(z); |
1ecaab2b KH |
1200 | int zid = zone_idx(z); |
1201 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 1202 | int lru = LRU_FILE * file + active; |
2ffebca6 | 1203 | int ret; |
66e1707b | 1204 | |
cf475ad2 | 1205 | BUG_ON(!mem_cont); |
1ecaab2b | 1206 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 1207 | src = &mz->lists[lru]; |
66e1707b | 1208 | |
ff7283fa KH |
1209 | scan = 0; |
1210 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 1211 | if (scan >= nr_to_scan) |
ff7283fa | 1212 | break; |
08e552c6 | 1213 | |
52d4b9ac KH |
1214 | if (unlikely(!PageCgroupUsed(pc))) |
1215 | continue; | |
5564e88b | 1216 | |
6b3ae58e | 1217 | page = lookup_cgroup_page(pc); |
5564e88b | 1218 | |
436c6541 | 1219 | if (unlikely(!PageLRU(page))) |
ff7283fa | 1220 | continue; |
ff7283fa | 1221 | |
436c6541 | 1222 | scan++; |
2ffebca6 KH |
1223 | ret = __isolate_lru_page(page, mode, file); |
1224 | switch (ret) { | |
1225 | case 0: | |
66e1707b | 1226 | list_move(&page->lru, dst); |
2ffebca6 | 1227 | mem_cgroup_del_lru(page); |
2c888cfb | 1228 | nr_taken += hpage_nr_pages(page); |
2ffebca6 KH |
1229 | break; |
1230 | case -EBUSY: | |
1231 | /* we don't affect global LRU but rotate in our LRU */ | |
1232 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
1233 | break; | |
1234 | default: | |
1235 | break; | |
66e1707b BS |
1236 | } |
1237 | } | |
1238 | ||
66e1707b | 1239 | *scanned = scan; |
cc8e970c KM |
1240 | |
1241 | trace_mm_vmscan_memcg_isolate(0, nr_to_scan, scan, nr_taken, | |
1242 | 0, 0, 0, mode); | |
1243 | ||
66e1707b BS |
1244 | return nr_taken; |
1245 | } | |
1246 | ||
6d61ef40 BS |
1247 | #define mem_cgroup_from_res_counter(counter, member) \ |
1248 | container_of(counter, struct mem_cgroup, member) | |
1249 | ||
19942822 | 1250 | /** |
9d11ea9f JW |
1251 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
1252 | * @mem: the memory cgroup | |
19942822 | 1253 | * |
9d11ea9f | 1254 | * Returns the maximum amount of memory @mem can be charged with, in |
7ec99d62 | 1255 | * pages. |
19942822 | 1256 | */ |
7ec99d62 | 1257 | static unsigned long mem_cgroup_margin(struct mem_cgroup *mem) |
19942822 | 1258 | { |
9d11ea9f JW |
1259 | unsigned long long margin; |
1260 | ||
1261 | margin = res_counter_margin(&mem->res); | |
1262 | if (do_swap_account) | |
1263 | margin = min(margin, res_counter_margin(&mem->memsw)); | |
7ec99d62 | 1264 | return margin >> PAGE_SHIFT; |
19942822 JW |
1265 | } |
1266 | ||
1f4c025b | 1267 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
a7885eb8 KM |
1268 | { |
1269 | struct cgroup *cgrp = memcg->css.cgroup; | |
a7885eb8 KM |
1270 | |
1271 | /* root ? */ | |
1272 | if (cgrp->parent == NULL) | |
1273 | return vm_swappiness; | |
1274 | ||
bf1ff263 | 1275 | return memcg->swappiness; |
a7885eb8 KM |
1276 | } |
1277 | ||
32047e2a KH |
1278 | static void mem_cgroup_start_move(struct mem_cgroup *mem) |
1279 | { | |
1280 | int cpu; | |
1489ebad KH |
1281 | |
1282 | get_online_cpus(); | |
1283 | spin_lock(&mem->pcp_counter_lock); | |
1284 | for_each_online_cpu(cpu) | |
32047e2a | 1285 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1; |
1489ebad KH |
1286 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1; |
1287 | spin_unlock(&mem->pcp_counter_lock); | |
1288 | put_online_cpus(); | |
32047e2a KH |
1289 | |
1290 | synchronize_rcu(); | |
1291 | } | |
1292 | ||
1293 | static void mem_cgroup_end_move(struct mem_cgroup *mem) | |
1294 | { | |
1295 | int cpu; | |
1296 | ||
1297 | if (!mem) | |
1298 | return; | |
1489ebad KH |
1299 | get_online_cpus(); |
1300 | spin_lock(&mem->pcp_counter_lock); | |
1301 | for_each_online_cpu(cpu) | |
32047e2a | 1302 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1; |
1489ebad KH |
1303 | mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1; |
1304 | spin_unlock(&mem->pcp_counter_lock); | |
1305 | put_online_cpus(); | |
32047e2a KH |
1306 | } |
1307 | /* | |
1308 | * 2 routines for checking "mem" is under move_account() or not. | |
1309 | * | |
1310 | * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used | |
1311 | * for avoiding race in accounting. If true, | |
1312 | * pc->mem_cgroup may be overwritten. | |
1313 | * | |
1314 | * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or | |
1315 | * under hierarchy of moving cgroups. This is for | |
1316 | * waiting at hith-memory prressure caused by "move". | |
1317 | */ | |
1318 | ||
1319 | static bool mem_cgroup_stealed(struct mem_cgroup *mem) | |
1320 | { | |
1321 | VM_BUG_ON(!rcu_read_lock_held()); | |
1322 | return this_cpu_read(mem->stat->count[MEM_CGROUP_ON_MOVE]) > 0; | |
1323 | } | |
4b534334 KH |
1324 | |
1325 | static bool mem_cgroup_under_move(struct mem_cgroup *mem) | |
1326 | { | |
2bd9bb20 KH |
1327 | struct mem_cgroup *from; |
1328 | struct mem_cgroup *to; | |
4b534334 | 1329 | bool ret = false; |
2bd9bb20 KH |
1330 | /* |
1331 | * Unlike task_move routines, we access mc.to, mc.from not under | |
1332 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. | |
1333 | */ | |
1334 | spin_lock(&mc.lock); | |
1335 | from = mc.from; | |
1336 | to = mc.to; | |
1337 | if (!from) | |
1338 | goto unlock; | |
3e92041d MH |
1339 | |
1340 | ret = mem_cgroup_same_or_subtree(mem, from) | |
1341 | || mem_cgroup_same_or_subtree(mem, to); | |
2bd9bb20 KH |
1342 | unlock: |
1343 | spin_unlock(&mc.lock); | |
4b534334 KH |
1344 | return ret; |
1345 | } | |
1346 | ||
1347 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *mem) | |
1348 | { | |
1349 | if (mc.moving_task && current != mc.moving_task) { | |
1350 | if (mem_cgroup_under_move(mem)) { | |
1351 | DEFINE_WAIT(wait); | |
1352 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); | |
1353 | /* moving charge context might have finished. */ | |
1354 | if (mc.moving_task) | |
1355 | schedule(); | |
1356 | finish_wait(&mc.waitq, &wait); | |
1357 | return true; | |
1358 | } | |
1359 | } | |
1360 | return false; | |
1361 | } | |
1362 | ||
e222432b | 1363 | /** |
6a6135b6 | 1364 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1365 | * @memcg: The memory cgroup that went over limit |
1366 | * @p: Task that is going to be killed | |
1367 | * | |
1368 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1369 | * enabled | |
1370 | */ | |
1371 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1372 | { | |
1373 | struct cgroup *task_cgrp; | |
1374 | struct cgroup *mem_cgrp; | |
1375 | /* | |
1376 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1377 | * on the assumption that OOM is serialized for memory controller. | |
1378 | * If this assumption is broken, revisit this code. | |
1379 | */ | |
1380 | static char memcg_name[PATH_MAX]; | |
1381 | int ret; | |
1382 | ||
d31f56db | 1383 | if (!memcg || !p) |
e222432b BS |
1384 | return; |
1385 | ||
1386 | ||
1387 | rcu_read_lock(); | |
1388 | ||
1389 | mem_cgrp = memcg->css.cgroup; | |
1390 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1391 | ||
1392 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1393 | if (ret < 0) { | |
1394 | /* | |
1395 | * Unfortunately, we are unable to convert to a useful name | |
1396 | * But we'll still print out the usage information | |
1397 | */ | |
1398 | rcu_read_unlock(); | |
1399 | goto done; | |
1400 | } | |
1401 | rcu_read_unlock(); | |
1402 | ||
1403 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1404 | ||
1405 | rcu_read_lock(); | |
1406 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1407 | if (ret < 0) { | |
1408 | rcu_read_unlock(); | |
1409 | goto done; | |
1410 | } | |
1411 | rcu_read_unlock(); | |
1412 | ||
1413 | /* | |
1414 | * Continues from above, so we don't need an KERN_ level | |
1415 | */ | |
1416 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1417 | done: | |
1418 | ||
1419 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1420 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1421 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1422 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1423 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1424 | "failcnt %llu\n", | |
1425 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1426 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1427 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1428 | } | |
1429 | ||
81d39c20 KH |
1430 | /* |
1431 | * This function returns the number of memcg under hierarchy tree. Returns | |
1432 | * 1(self count) if no children. | |
1433 | */ | |
1434 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1435 | { | |
1436 | int num = 0; | |
7d74b06f KH |
1437 | struct mem_cgroup *iter; |
1438 | ||
1439 | for_each_mem_cgroup_tree(iter, mem) | |
1440 | num++; | |
81d39c20 KH |
1441 | return num; |
1442 | } | |
1443 | ||
a63d83f4 DR |
1444 | /* |
1445 | * Return the memory (and swap, if configured) limit for a memcg. | |
1446 | */ | |
1447 | u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) | |
1448 | { | |
1449 | u64 limit; | |
1450 | u64 memsw; | |
1451 | ||
f3e8eb70 JW |
1452 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
1453 | limit += total_swap_pages << PAGE_SHIFT; | |
1454 | ||
a63d83f4 DR |
1455 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
1456 | /* | |
1457 | * If memsw is finite and limits the amount of swap space available | |
1458 | * to this memcg, return that limit. | |
1459 | */ | |
1460 | return min(limit, memsw); | |
1461 | } | |
1462 | ||
6d61ef40 | 1463 | /* |
04046e1a KH |
1464 | * Visit the first child (need not be the first child as per the ordering |
1465 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1466 | * that to reclaim free pages from. | |
1467 | */ | |
1468 | static struct mem_cgroup * | |
1469 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1470 | { | |
1471 | struct mem_cgroup *ret = NULL; | |
1472 | struct cgroup_subsys_state *css; | |
1473 | int nextid, found; | |
1474 | ||
1475 | if (!root_mem->use_hierarchy) { | |
1476 | css_get(&root_mem->css); | |
1477 | ret = root_mem; | |
1478 | } | |
1479 | ||
1480 | while (!ret) { | |
1481 | rcu_read_lock(); | |
1482 | nextid = root_mem->last_scanned_child + 1; | |
1483 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1484 | &found); | |
1485 | if (css && css_tryget(css)) | |
1486 | ret = container_of(css, struct mem_cgroup, css); | |
1487 | ||
1488 | rcu_read_unlock(); | |
1489 | /* Updates scanning parameter */ | |
04046e1a KH |
1490 | if (!css) { |
1491 | /* this means start scan from ID:1 */ | |
1492 | root_mem->last_scanned_child = 0; | |
1493 | } else | |
1494 | root_mem->last_scanned_child = found; | |
04046e1a KH |
1495 | } |
1496 | ||
1497 | return ret; | |
1498 | } | |
1499 | ||
4d0c066d KH |
1500 | /** |
1501 | * test_mem_cgroup_node_reclaimable | |
1502 | * @mem: the target memcg | |
1503 | * @nid: the node ID to be checked. | |
1504 | * @noswap : specify true here if the user wants flle only information. | |
1505 | * | |
1506 | * This function returns whether the specified memcg contains any | |
1507 | * reclaimable pages on a node. Returns true if there are any reclaimable | |
1508 | * pages in the node. | |
1509 | */ | |
1510 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *mem, | |
1511 | int nid, bool noswap) | |
1512 | { | |
bb2a0de9 | 1513 | if (mem_cgroup_node_nr_lru_pages(mem, nid, LRU_ALL_FILE)) |
4d0c066d KH |
1514 | return true; |
1515 | if (noswap || !total_swap_pages) | |
1516 | return false; | |
bb2a0de9 | 1517 | if (mem_cgroup_node_nr_lru_pages(mem, nid, LRU_ALL_ANON)) |
4d0c066d KH |
1518 | return true; |
1519 | return false; | |
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; | |
453a9bf3 KH |
1533 | /* |
1534 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET | |
1535 | * pagein/pageout changes since the last update. | |
1536 | */ | |
1537 | if (!atomic_read(&mem->numainfo_events)) | |
1538 | return; | |
1539 | if (atomic_inc_return(&mem->numainfo_updating) > 1) | |
889976db YH |
1540 | return; |
1541 | ||
889976db YH |
1542 | /* make a nodemask where this memcg uses memory from */ |
1543 | mem->scan_nodes = node_states[N_HIGH_MEMORY]; | |
1544 | ||
1545 | for_each_node_mask(nid, node_states[N_HIGH_MEMORY]) { | |
1546 | ||
4d0c066d KH |
1547 | if (!test_mem_cgroup_node_reclaimable(mem, nid, false)) |
1548 | node_clear(nid, mem->scan_nodes); | |
889976db | 1549 | } |
453a9bf3 KH |
1550 | |
1551 | atomic_set(&mem->numainfo_events, 0); | |
1552 | atomic_set(&mem->numainfo_updating, 0); | |
889976db YH |
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 | ||
4d0c066d KH |
1590 | /* |
1591 | * Check all nodes whether it contains reclaimable pages or not. | |
1592 | * For quick scan, we make use of scan_nodes. This will allow us to skip | |
1593 | * unused nodes. But scan_nodes is lazily updated and may not cotain | |
1594 | * enough new information. We need to do double check. | |
1595 | */ | |
1596 | bool mem_cgroup_reclaimable(struct mem_cgroup *mem, bool noswap) | |
1597 | { | |
1598 | int nid; | |
1599 | ||
1600 | /* | |
1601 | * quick check...making use of scan_node. | |
1602 | * We can skip unused nodes. | |
1603 | */ | |
1604 | if (!nodes_empty(mem->scan_nodes)) { | |
1605 | for (nid = first_node(mem->scan_nodes); | |
1606 | nid < MAX_NUMNODES; | |
1607 | nid = next_node(nid, mem->scan_nodes)) { | |
1608 | ||
1609 | if (test_mem_cgroup_node_reclaimable(mem, nid, noswap)) | |
1610 | return true; | |
1611 | } | |
1612 | } | |
1613 | /* | |
1614 | * Check rest of nodes. | |
1615 | */ | |
1616 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
1617 | if (node_isset(nid, mem->scan_nodes)) | |
1618 | continue; | |
1619 | if (test_mem_cgroup_node_reclaimable(mem, nid, noswap)) | |
1620 | return true; | |
1621 | } | |
1622 | return false; | |
1623 | } | |
1624 | ||
889976db YH |
1625 | #else |
1626 | int mem_cgroup_select_victim_node(struct mem_cgroup *mem) | |
1627 | { | |
1628 | return 0; | |
1629 | } | |
4d0c066d KH |
1630 | |
1631 | bool mem_cgroup_reclaimable(struct mem_cgroup *mem, bool noswap) | |
1632 | { | |
1633 | return test_mem_cgroup_node_reclaimable(mem, 0, noswap); | |
1634 | } | |
889976db YH |
1635 | #endif |
1636 | ||
04046e1a KH |
1637 | /* |
1638 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1639 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1640 | * based on its position in the children list. | |
6d61ef40 BS |
1641 | * |
1642 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1643 | * |
1644 | * We give up and return to the caller when we visit root_mem twice. | |
1645 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1646 | * |
1647 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1648 | */ |
1649 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1650 | struct zone *zone, |
75822b44 | 1651 | gfp_t gfp_mask, |
0ae5e89c YH |
1652 | unsigned long reclaim_options, |
1653 | unsigned long *total_scanned) | |
6d61ef40 | 1654 | { |
04046e1a KH |
1655 | struct mem_cgroup *victim; |
1656 | int ret, total = 0; | |
1657 | int loop = 0; | |
75822b44 BS |
1658 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1659 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 | 1660 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
9d11ea9f | 1661 | unsigned long excess; |
185efc0f | 1662 | unsigned long nr_scanned; |
9d11ea9f JW |
1663 | |
1664 | excess = res_counter_soft_limit_excess(&root_mem->res) >> PAGE_SHIFT; | |
04046e1a | 1665 | |
22a668d7 | 1666 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
108b6a78 | 1667 | if (!check_soft && !shrink && root_mem->memsw_is_minimum) |
22a668d7 KH |
1668 | noswap = true; |
1669 | ||
4e416953 | 1670 | while (1) { |
04046e1a | 1671 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1672 | if (victim == root_mem) { |
04046e1a | 1673 | loop++; |
fbc29a25 KH |
1674 | /* |
1675 | * We are not draining per cpu cached charges during | |
1676 | * soft limit reclaim because global reclaim doesn't | |
1677 | * care about charges. It tries to free some memory and | |
1678 | * charges will not give any. | |
1679 | */ | |
1680 | if (!check_soft && loop >= 1) | |
26fe6168 | 1681 | drain_all_stock_async(root_mem); |
4e416953 BS |
1682 | if (loop >= 2) { |
1683 | /* | |
1684 | * If we have not been able to reclaim | |
1685 | * anything, it might because there are | |
1686 | * no reclaimable pages under this hierarchy | |
1687 | */ | |
1688 | if (!check_soft || !total) { | |
1689 | css_put(&victim->css); | |
1690 | break; | |
1691 | } | |
1692 | /* | |
25985edc | 1693 | * We want to do more targeted reclaim. |
4e416953 BS |
1694 | * excess >> 2 is not to excessive so as to |
1695 | * reclaim too much, nor too less that we keep | |
1696 | * coming back to reclaim from this cgroup | |
1697 | */ | |
1698 | if (total >= (excess >> 2) || | |
1699 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1700 | css_put(&victim->css); | |
1701 | break; | |
1702 | } | |
1703 | } | |
1704 | } | |
4d0c066d | 1705 | if (!mem_cgroup_reclaimable(victim, noswap)) { |
04046e1a KH |
1706 | /* this cgroup's local usage == 0 */ |
1707 | css_put(&victim->css); | |
6d61ef40 BS |
1708 | continue; |
1709 | } | |
04046e1a | 1710 | /* we use swappiness of local cgroup */ |
0ae5e89c | 1711 | if (check_soft) { |
4e416953 | 1712 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, |
185efc0f JW |
1713 | noswap, zone, &nr_scanned); |
1714 | *total_scanned += nr_scanned; | |
0ae5e89c | 1715 | } else |
4e416953 | 1716 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, |
185efc0f | 1717 | noswap); |
04046e1a | 1718 | css_put(&victim->css); |
81d39c20 KH |
1719 | /* |
1720 | * At shrinking usage, we can't check we should stop here or | |
1721 | * reclaim more. It's depends on callers. last_scanned_child | |
1722 | * will work enough for keeping fairness under tree. | |
1723 | */ | |
1724 | if (shrink) | |
1725 | return ret; | |
04046e1a | 1726 | total += ret; |
4e416953 | 1727 | if (check_soft) { |
9d11ea9f | 1728 | if (!res_counter_soft_limit_excess(&root_mem->res)) |
4e416953 | 1729 | return total; |
9d11ea9f | 1730 | } else if (mem_cgroup_margin(root_mem)) |
4fd14ebf | 1731 | return total; |
6d61ef40 | 1732 | } |
04046e1a | 1733 | return total; |
6d61ef40 BS |
1734 | } |
1735 | ||
867578cb KH |
1736 | /* |
1737 | * Check OOM-Killer is already running under our hierarchy. | |
1738 | * If someone is running, return false. | |
1af8efe9 | 1739 | * Has to be called with memcg_oom_lock |
867578cb KH |
1740 | */ |
1741 | static bool mem_cgroup_oom_lock(struct mem_cgroup *mem) | |
1742 | { | |
79dfdacc MH |
1743 | struct mem_cgroup *iter, *failed = NULL; |
1744 | bool cond = true; | |
a636b327 | 1745 | |
79dfdacc | 1746 | for_each_mem_cgroup_tree_cond(iter, mem, cond) { |
23751be0 | 1747 | if (iter->oom_lock) { |
79dfdacc MH |
1748 | /* |
1749 | * this subtree of our hierarchy is already locked | |
1750 | * so we cannot give a lock. | |
1751 | */ | |
79dfdacc MH |
1752 | failed = iter; |
1753 | cond = false; | |
23751be0 JW |
1754 | } else |
1755 | iter->oom_lock = true; | |
7d74b06f | 1756 | } |
867578cb | 1757 | |
79dfdacc | 1758 | if (!failed) |
23751be0 | 1759 | return true; |
79dfdacc MH |
1760 | |
1761 | /* | |
1762 | * OK, we failed to lock the whole subtree so we have to clean up | |
1763 | * what we set up to the failing subtree | |
1764 | */ | |
1765 | cond = true; | |
1766 | for_each_mem_cgroup_tree_cond(iter, mem, cond) { | |
1767 | if (iter == failed) { | |
1768 | cond = false; | |
1769 | continue; | |
1770 | } | |
1771 | iter->oom_lock = false; | |
1772 | } | |
23751be0 | 1773 | return false; |
a636b327 | 1774 | } |
0b7f569e | 1775 | |
79dfdacc | 1776 | /* |
1af8efe9 | 1777 | * Has to be called with memcg_oom_lock |
79dfdacc | 1778 | */ |
7d74b06f | 1779 | static int mem_cgroup_oom_unlock(struct mem_cgroup *mem) |
0b7f569e | 1780 | { |
7d74b06f KH |
1781 | struct mem_cgroup *iter; |
1782 | ||
79dfdacc MH |
1783 | for_each_mem_cgroup_tree(iter, mem) |
1784 | iter->oom_lock = false; | |
1785 | return 0; | |
1786 | } | |
1787 | ||
1788 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *mem) | |
1789 | { | |
1790 | struct mem_cgroup *iter; | |
1791 | ||
1792 | for_each_mem_cgroup_tree(iter, mem) | |
1793 | atomic_inc(&iter->under_oom); | |
1794 | } | |
1795 | ||
1796 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *mem) | |
1797 | { | |
1798 | struct mem_cgroup *iter; | |
1799 | ||
867578cb KH |
1800 | /* |
1801 | * When a new child is created while the hierarchy is under oom, | |
1802 | * mem_cgroup_oom_lock() may not be called. We have to use | |
1803 | * atomic_add_unless() here. | |
1804 | */ | |
7d74b06f | 1805 | for_each_mem_cgroup_tree(iter, mem) |
79dfdacc | 1806 | atomic_add_unless(&iter->under_oom, -1, 0); |
0b7f569e KH |
1807 | } |
1808 | ||
1af8efe9 | 1809 | static DEFINE_SPINLOCK(memcg_oom_lock); |
867578cb KH |
1810 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
1811 | ||
dc98df5a KH |
1812 | struct oom_wait_info { |
1813 | struct mem_cgroup *mem; | |
1814 | wait_queue_t wait; | |
1815 | }; | |
1816 | ||
1817 | static int memcg_oom_wake_function(wait_queue_t *wait, | |
1818 | unsigned mode, int sync, void *arg) | |
1819 | { | |
3e92041d MH |
1820 | struct mem_cgroup *wake_mem = (struct mem_cgroup *)arg, |
1821 | *oom_wait_mem; | |
dc98df5a KH |
1822 | struct oom_wait_info *oom_wait_info; |
1823 | ||
1824 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); | |
3e92041d | 1825 | oom_wait_mem = oom_wait_info->mem; |
dc98df5a | 1826 | |
dc98df5a KH |
1827 | /* |
1828 | * Both of oom_wait_info->mem and wake_mem are stable under us. | |
1829 | * Then we can use css_is_ancestor without taking care of RCU. | |
1830 | */ | |
3e92041d MH |
1831 | if (!mem_cgroup_same_or_subtree(oom_wait_mem, wake_mem) |
1832 | && !mem_cgroup_same_or_subtree(wake_mem, oom_wait_mem)) | |
dc98df5a | 1833 | return 0; |
dc98df5a KH |
1834 | return autoremove_wake_function(wait, mode, sync, arg); |
1835 | } | |
1836 | ||
1837 | static void memcg_wakeup_oom(struct mem_cgroup *mem) | |
1838 | { | |
1839 | /* for filtering, pass "mem" as argument. */ | |
1840 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, mem); | |
1841 | } | |
1842 | ||
3c11ecf4 KH |
1843 | static void memcg_oom_recover(struct mem_cgroup *mem) |
1844 | { | |
79dfdacc | 1845 | if (mem && atomic_read(&mem->under_oom)) |
3c11ecf4 KH |
1846 | memcg_wakeup_oom(mem); |
1847 | } | |
1848 | ||
867578cb KH |
1849 | /* |
1850 | * try to call OOM killer. returns false if we should exit memory-reclaim loop. | |
1851 | */ | |
1852 | bool mem_cgroup_handle_oom(struct mem_cgroup *mem, gfp_t mask) | |
0b7f569e | 1853 | { |
dc98df5a | 1854 | struct oom_wait_info owait; |
3c11ecf4 | 1855 | bool locked, need_to_kill; |
867578cb | 1856 | |
dc98df5a KH |
1857 | owait.mem = mem; |
1858 | owait.wait.flags = 0; | |
1859 | owait.wait.func = memcg_oom_wake_function; | |
1860 | owait.wait.private = current; | |
1861 | INIT_LIST_HEAD(&owait.wait.task_list); | |
3c11ecf4 | 1862 | need_to_kill = true; |
79dfdacc MH |
1863 | mem_cgroup_mark_under_oom(mem); |
1864 | ||
867578cb | 1865 | /* At first, try to OOM lock hierarchy under mem.*/ |
1af8efe9 | 1866 | spin_lock(&memcg_oom_lock); |
867578cb KH |
1867 | locked = mem_cgroup_oom_lock(mem); |
1868 | /* | |
1869 | * Even if signal_pending(), we can't quit charge() loop without | |
1870 | * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL | |
1871 | * under OOM is always welcomed, use TASK_KILLABLE here. | |
1872 | */ | |
3c11ecf4 KH |
1873 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
1874 | if (!locked || mem->oom_kill_disable) | |
1875 | need_to_kill = false; | |
1876 | if (locked) | |
9490ff27 | 1877 | mem_cgroup_oom_notify(mem); |
1af8efe9 | 1878 | spin_unlock(&memcg_oom_lock); |
867578cb | 1879 | |
3c11ecf4 KH |
1880 | if (need_to_kill) { |
1881 | finish_wait(&memcg_oom_waitq, &owait.wait); | |
867578cb | 1882 | mem_cgroup_out_of_memory(mem, mask); |
3c11ecf4 | 1883 | } else { |
867578cb | 1884 | schedule(); |
dc98df5a | 1885 | finish_wait(&memcg_oom_waitq, &owait.wait); |
867578cb | 1886 | } |
1af8efe9 | 1887 | spin_lock(&memcg_oom_lock); |
79dfdacc MH |
1888 | if (locked) |
1889 | mem_cgroup_oom_unlock(mem); | |
dc98df5a | 1890 | memcg_wakeup_oom(mem); |
1af8efe9 | 1891 | spin_unlock(&memcg_oom_lock); |
867578cb | 1892 | |
79dfdacc MH |
1893 | mem_cgroup_unmark_under_oom(mem); |
1894 | ||
867578cb KH |
1895 | if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) |
1896 | return false; | |
1897 | /* Give chance to dying process */ | |
1898 | schedule_timeout(1); | |
1899 | return true; | |
0b7f569e KH |
1900 | } |
1901 | ||
d69b042f BS |
1902 | /* |
1903 | * Currently used to update mapped file statistics, but the routine can be | |
1904 | * generalized to update other statistics as well. | |
32047e2a KH |
1905 | * |
1906 | * Notes: Race condition | |
1907 | * | |
1908 | * We usually use page_cgroup_lock() for accessing page_cgroup member but | |
1909 | * it tends to be costly. But considering some conditions, we doesn't need | |
1910 | * to do so _always_. | |
1911 | * | |
1912 | * Considering "charge", lock_page_cgroup() is not required because all | |
1913 | * file-stat operations happen after a page is attached to radix-tree. There | |
1914 | * are no race with "charge". | |
1915 | * | |
1916 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup | |
1917 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even | |
1918 | * if there are race with "uncharge". Statistics itself is properly handled | |
1919 | * by flags. | |
1920 | * | |
1921 | * Considering "move", this is an only case we see a race. To make the race | |
1922 | * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are | |
1923 | * possibility of race condition. If there is, we take a lock. | |
d69b042f | 1924 | */ |
26174efd | 1925 | |
2a7106f2 GT |
1926 | void mem_cgroup_update_page_stat(struct page *page, |
1927 | enum mem_cgroup_page_stat_item idx, int val) | |
d69b042f BS |
1928 | { |
1929 | struct mem_cgroup *mem; | |
32047e2a KH |
1930 | struct page_cgroup *pc = lookup_page_cgroup(page); |
1931 | bool need_unlock = false; | |
dbd4ea78 | 1932 | unsigned long uninitialized_var(flags); |
d69b042f | 1933 | |
d69b042f BS |
1934 | if (unlikely(!pc)) |
1935 | return; | |
1936 | ||
32047e2a | 1937 | rcu_read_lock(); |
d69b042f | 1938 | mem = pc->mem_cgroup; |
32047e2a KH |
1939 | if (unlikely(!mem || !PageCgroupUsed(pc))) |
1940 | goto out; | |
1941 | /* pc->mem_cgroup is unstable ? */ | |
ca3e0214 | 1942 | if (unlikely(mem_cgroup_stealed(mem)) || PageTransHuge(page)) { |
32047e2a | 1943 | /* take a lock against to access pc->mem_cgroup */ |
dbd4ea78 | 1944 | move_lock_page_cgroup(pc, &flags); |
32047e2a KH |
1945 | need_unlock = true; |
1946 | mem = pc->mem_cgroup; | |
1947 | if (!mem || !PageCgroupUsed(pc)) | |
1948 | goto out; | |
1949 | } | |
26174efd | 1950 | |
26174efd | 1951 | switch (idx) { |
2a7106f2 | 1952 | case MEMCG_NR_FILE_MAPPED: |
26174efd KH |
1953 | if (val > 0) |
1954 | SetPageCgroupFileMapped(pc); | |
1955 | else if (!page_mapped(page)) | |
0c270f8f | 1956 | ClearPageCgroupFileMapped(pc); |
2a7106f2 | 1957 | idx = MEM_CGROUP_STAT_FILE_MAPPED; |
26174efd KH |
1958 | break; |
1959 | default: | |
1960 | BUG(); | |
8725d541 | 1961 | } |
d69b042f | 1962 | |
2a7106f2 GT |
1963 | this_cpu_add(mem->stat->count[idx], val); |
1964 | ||
32047e2a KH |
1965 | out: |
1966 | if (unlikely(need_unlock)) | |
dbd4ea78 | 1967 | move_unlock_page_cgroup(pc, &flags); |
32047e2a KH |
1968 | rcu_read_unlock(); |
1969 | return; | |
d69b042f | 1970 | } |
2a7106f2 | 1971 | EXPORT_SYMBOL(mem_cgroup_update_page_stat); |
26174efd | 1972 | |
cdec2e42 KH |
1973 | /* |
1974 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1975 | * TODO: maybe necessary to use big numbers in big irons. | |
1976 | */ | |
7ec99d62 | 1977 | #define CHARGE_BATCH 32U |
cdec2e42 KH |
1978 | struct memcg_stock_pcp { |
1979 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
11c9ea4e | 1980 | unsigned int nr_pages; |
cdec2e42 | 1981 | struct work_struct work; |
26fe6168 KH |
1982 | unsigned long flags; |
1983 | #define FLUSHING_CACHED_CHARGE (0) | |
cdec2e42 KH |
1984 | }; |
1985 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
9f50fad6 | 1986 | static DEFINE_MUTEX(percpu_charge_mutex); |
cdec2e42 KH |
1987 | |
1988 | /* | |
11c9ea4e | 1989 | * Try to consume stocked charge on this cpu. If success, one page is consumed |
cdec2e42 KH |
1990 | * from local stock and true is returned. If the stock is 0 or charges from a |
1991 | * cgroup which is not current target, returns false. This stock will be | |
1992 | * refilled. | |
1993 | */ | |
1994 | static bool consume_stock(struct mem_cgroup *mem) | |
1995 | { | |
1996 | struct memcg_stock_pcp *stock; | |
1997 | bool ret = true; | |
1998 | ||
1999 | stock = &get_cpu_var(memcg_stock); | |
11c9ea4e JW |
2000 | if (mem == stock->cached && stock->nr_pages) |
2001 | stock->nr_pages--; | |
cdec2e42 KH |
2002 | else /* need to call res_counter_charge */ |
2003 | ret = false; | |
2004 | put_cpu_var(memcg_stock); | |
2005 | return ret; | |
2006 | } | |
2007 | ||
2008 | /* | |
2009 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
2010 | */ | |
2011 | static void drain_stock(struct memcg_stock_pcp *stock) | |
2012 | { | |
2013 | struct mem_cgroup *old = stock->cached; | |
2014 | ||
11c9ea4e JW |
2015 | if (stock->nr_pages) { |
2016 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; | |
2017 | ||
2018 | res_counter_uncharge(&old->res, bytes); | |
cdec2e42 | 2019 | if (do_swap_account) |
11c9ea4e JW |
2020 | res_counter_uncharge(&old->memsw, bytes); |
2021 | stock->nr_pages = 0; | |
cdec2e42 KH |
2022 | } |
2023 | stock->cached = NULL; | |
cdec2e42 KH |
2024 | } |
2025 | ||
2026 | /* | |
2027 | * This must be called under preempt disabled or must be called by | |
2028 | * a thread which is pinned to local cpu. | |
2029 | */ | |
2030 | static void drain_local_stock(struct work_struct *dummy) | |
2031 | { | |
2032 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
2033 | drain_stock(stock); | |
26fe6168 | 2034 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
cdec2e42 KH |
2035 | } |
2036 | ||
2037 | /* | |
2038 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
320cc51d | 2039 | * This will be consumed by consume_stock() function, later. |
cdec2e42 | 2040 | */ |
11c9ea4e | 2041 | static void refill_stock(struct mem_cgroup *mem, unsigned int nr_pages) |
cdec2e42 KH |
2042 | { |
2043 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
2044 | ||
2045 | if (stock->cached != mem) { /* reset if necessary */ | |
2046 | drain_stock(stock); | |
2047 | stock->cached = mem; | |
2048 | } | |
11c9ea4e | 2049 | stock->nr_pages += nr_pages; |
cdec2e42 KH |
2050 | put_cpu_var(memcg_stock); |
2051 | } | |
2052 | ||
2053 | /* | |
d38144b7 MH |
2054 | * Drains all per-CPU charge caches for given root_mem resp. subtree |
2055 | * of the hierarchy under it. sync flag says whether we should block | |
2056 | * until the work is done. | |
cdec2e42 | 2057 | */ |
d38144b7 | 2058 | static void drain_all_stock(struct mem_cgroup *root_mem, bool sync) |
cdec2e42 | 2059 | { |
26fe6168 | 2060 | int cpu, curcpu; |
d38144b7 | 2061 | |
cdec2e42 | 2062 | /* Notify other cpus that system-wide "drain" is running */ |
cdec2e42 | 2063 | get_online_cpus(); |
5af12d0e | 2064 | curcpu = get_cpu(); |
cdec2e42 KH |
2065 | for_each_online_cpu(cpu) { |
2066 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
26fe6168 KH |
2067 | struct mem_cgroup *mem; |
2068 | ||
26fe6168 | 2069 | mem = stock->cached; |
d1a05b69 | 2070 | if (!mem || !stock->nr_pages) |
26fe6168 | 2071 | continue; |
3e92041d MH |
2072 | if (!mem_cgroup_same_or_subtree(root_mem, mem)) |
2073 | continue; | |
d1a05b69 MH |
2074 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
2075 | if (cpu == curcpu) | |
2076 | drain_local_stock(&stock->work); | |
2077 | else | |
2078 | schedule_work_on(cpu, &stock->work); | |
2079 | } | |
cdec2e42 | 2080 | } |
5af12d0e | 2081 | put_cpu(); |
d38144b7 MH |
2082 | |
2083 | if (!sync) | |
2084 | goto out; | |
2085 | ||
2086 | for_each_online_cpu(cpu) { | |
2087 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
9f50fad6 | 2088 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
d38144b7 MH |
2089 | flush_work(&stock->work); |
2090 | } | |
2091 | out: | |
cdec2e42 | 2092 | put_online_cpus(); |
d38144b7 MH |
2093 | } |
2094 | ||
2095 | /* | |
2096 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
2097 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
2098 | * expects some charges will be back to res_counter later but cannot wait for | |
2099 | * it. | |
2100 | */ | |
2101 | static void drain_all_stock_async(struct mem_cgroup *root_mem) | |
2102 | { | |
9f50fad6 MH |
2103 | /* |
2104 | * If someone calls draining, avoid adding more kworker runs. | |
2105 | */ | |
2106 | if (!mutex_trylock(&percpu_charge_mutex)) | |
2107 | return; | |
d38144b7 | 2108 | drain_all_stock(root_mem, false); |
9f50fad6 | 2109 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2110 | } |
2111 | ||
2112 | /* This is a synchronous drain interface. */ | |
d38144b7 | 2113 | static void drain_all_stock_sync(struct mem_cgroup *root_mem) |
cdec2e42 KH |
2114 | { |
2115 | /* called when force_empty is called */ | |
9f50fad6 | 2116 | mutex_lock(&percpu_charge_mutex); |
d38144b7 | 2117 | drain_all_stock(root_mem, true); |
9f50fad6 | 2118 | mutex_unlock(&percpu_charge_mutex); |
cdec2e42 KH |
2119 | } |
2120 | ||
711d3d2c KH |
2121 | /* |
2122 | * This function drains percpu counter value from DEAD cpu and | |
2123 | * move it to local cpu. Note that this function can be preempted. | |
2124 | */ | |
2125 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu) | |
2126 | { | |
2127 | int i; | |
2128 | ||
2129 | spin_lock(&mem->pcp_counter_lock); | |
2130 | for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) { | |
7a159cc9 | 2131 | long x = per_cpu(mem->stat->count[i], cpu); |
711d3d2c KH |
2132 | |
2133 | per_cpu(mem->stat->count[i], cpu) = 0; | |
2134 | mem->nocpu_base.count[i] += x; | |
2135 | } | |
e9f8974f JW |
2136 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
2137 | unsigned long x = per_cpu(mem->stat->events[i], cpu); | |
2138 | ||
2139 | per_cpu(mem->stat->events[i], cpu) = 0; | |
2140 | mem->nocpu_base.events[i] += x; | |
2141 | } | |
1489ebad KH |
2142 | /* need to clear ON_MOVE value, works as a kind of lock. */ |
2143 | per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0; | |
2144 | spin_unlock(&mem->pcp_counter_lock); | |
2145 | } | |
2146 | ||
2147 | static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu) | |
2148 | { | |
2149 | int idx = MEM_CGROUP_ON_MOVE; | |
2150 | ||
2151 | spin_lock(&mem->pcp_counter_lock); | |
2152 | per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx]; | |
711d3d2c KH |
2153 | spin_unlock(&mem->pcp_counter_lock); |
2154 | } | |
2155 | ||
2156 | static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb, | |
cdec2e42 KH |
2157 | unsigned long action, |
2158 | void *hcpu) | |
2159 | { | |
2160 | int cpu = (unsigned long)hcpu; | |
2161 | struct memcg_stock_pcp *stock; | |
711d3d2c | 2162 | struct mem_cgroup *iter; |
cdec2e42 | 2163 | |
1489ebad KH |
2164 | if ((action == CPU_ONLINE)) { |
2165 | for_each_mem_cgroup_all(iter) | |
2166 | synchronize_mem_cgroup_on_move(iter, cpu); | |
2167 | return NOTIFY_OK; | |
2168 | } | |
2169 | ||
711d3d2c | 2170 | if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN) |
cdec2e42 | 2171 | return NOTIFY_OK; |
711d3d2c KH |
2172 | |
2173 | for_each_mem_cgroup_all(iter) | |
2174 | mem_cgroup_drain_pcp_counter(iter, cpu); | |
2175 | ||
cdec2e42 KH |
2176 | stock = &per_cpu(memcg_stock, cpu); |
2177 | drain_stock(stock); | |
2178 | return NOTIFY_OK; | |
2179 | } | |
2180 | ||
4b534334 KH |
2181 | |
2182 | /* See __mem_cgroup_try_charge() for details */ | |
2183 | enum { | |
2184 | CHARGE_OK, /* success */ | |
2185 | CHARGE_RETRY, /* need to retry but retry is not bad */ | |
2186 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ | |
2187 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ | |
2188 | CHARGE_OOM_DIE, /* the current is killed because of OOM */ | |
2189 | }; | |
2190 | ||
7ec99d62 JW |
2191 | static int mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask, |
2192 | unsigned int nr_pages, bool oom_check) | |
4b534334 | 2193 | { |
7ec99d62 | 2194 | unsigned long csize = nr_pages * PAGE_SIZE; |
4b534334 KH |
2195 | struct mem_cgroup *mem_over_limit; |
2196 | struct res_counter *fail_res; | |
2197 | unsigned long flags = 0; | |
2198 | int ret; | |
2199 | ||
2200 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
2201 | ||
2202 | if (likely(!ret)) { | |
2203 | if (!do_swap_account) | |
2204 | return CHARGE_OK; | |
2205 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); | |
2206 | if (likely(!ret)) | |
2207 | return CHARGE_OK; | |
2208 | ||
01c88e2d | 2209 | res_counter_uncharge(&mem->res, csize); |
4b534334 KH |
2210 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
2211 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; | |
2212 | } else | |
2213 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); | |
9221edb7 | 2214 | /* |
7ec99d62 JW |
2215 | * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch |
2216 | * of regular pages (CHARGE_BATCH), or a single regular page (1). | |
9221edb7 JW |
2217 | * |
2218 | * Never reclaim on behalf of optional batching, retry with a | |
2219 | * single page instead. | |
2220 | */ | |
7ec99d62 | 2221 | if (nr_pages == CHARGE_BATCH) |
4b534334 KH |
2222 | return CHARGE_RETRY; |
2223 | ||
2224 | if (!(gfp_mask & __GFP_WAIT)) | |
2225 | return CHARGE_WOULDBLOCK; | |
2226 | ||
2227 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, | |
0ae5e89c | 2228 | gfp_mask, flags, NULL); |
7ec99d62 | 2229 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
19942822 | 2230 | return CHARGE_RETRY; |
4b534334 | 2231 | /* |
19942822 JW |
2232 | * Even though the limit is exceeded at this point, reclaim |
2233 | * may have been able to free some pages. Retry the charge | |
2234 | * before killing the task. | |
2235 | * | |
2236 | * Only for regular pages, though: huge pages are rather | |
2237 | * unlikely to succeed so close to the limit, and we fall back | |
2238 | * to regular pages anyway in case of failure. | |
4b534334 | 2239 | */ |
7ec99d62 | 2240 | if (nr_pages == 1 && ret) |
4b534334 KH |
2241 | return CHARGE_RETRY; |
2242 | ||
2243 | /* | |
2244 | * At task move, charge accounts can be doubly counted. So, it's | |
2245 | * better to wait until the end of task_move if something is going on. | |
2246 | */ | |
2247 | if (mem_cgroup_wait_acct_move(mem_over_limit)) | |
2248 | return CHARGE_RETRY; | |
2249 | ||
2250 | /* If we don't need to call oom-killer at el, return immediately */ | |
2251 | if (!oom_check) | |
2252 | return CHARGE_NOMEM; | |
2253 | /* check OOM */ | |
2254 | if (!mem_cgroup_handle_oom(mem_over_limit, gfp_mask)) | |
2255 | return CHARGE_OOM_DIE; | |
2256 | ||
2257 | return CHARGE_RETRY; | |
2258 | } | |
2259 | ||
f817ed48 KH |
2260 | /* |
2261 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
2262 | * oom-killer can be invoked. | |
8a9f3ccd | 2263 | */ |
f817ed48 | 2264 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
ec168510 | 2265 | gfp_t gfp_mask, |
7ec99d62 JW |
2266 | unsigned int nr_pages, |
2267 | struct mem_cgroup **memcg, | |
2268 | bool oom) | |
8a9f3ccd | 2269 | { |
7ec99d62 | 2270 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
4b534334 KH |
2271 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
2272 | struct mem_cgroup *mem = NULL; | |
2273 | int ret; | |
a636b327 | 2274 | |
867578cb KH |
2275 | /* |
2276 | * Unlike gloval-vm's OOM-kill, we're not in memory shortage | |
2277 | * in system level. So, allow to go ahead dying process in addition to | |
2278 | * MEMDIE process. | |
2279 | */ | |
2280 | if (unlikely(test_thread_flag(TIF_MEMDIE) | |
2281 | || fatal_signal_pending(current))) | |
2282 | goto bypass; | |
a636b327 | 2283 | |
8a9f3ccd | 2284 | /* |
3be91277 HD |
2285 | * We always charge the cgroup the mm_struct belongs to. |
2286 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
2287 | * thread group leader migrates. It's possible that mm is not |
2288 | * set, if so charge the init_mm (happens for pagecache usage). | |
2289 | */ | |
f75ca962 KH |
2290 | if (!*memcg && !mm) |
2291 | goto bypass; | |
2292 | again: | |
2293 | if (*memcg) { /* css should be a valid one */ | |
4b534334 | 2294 | mem = *memcg; |
f75ca962 KH |
2295 | VM_BUG_ON(css_is_removed(&mem->css)); |
2296 | if (mem_cgroup_is_root(mem)) | |
2297 | goto done; | |
7ec99d62 | 2298 | if (nr_pages == 1 && consume_stock(mem)) |
f75ca962 | 2299 | goto done; |
4b534334 KH |
2300 | css_get(&mem->css); |
2301 | } else { | |
f75ca962 | 2302 | struct task_struct *p; |
54595fe2 | 2303 | |
f75ca962 KH |
2304 | rcu_read_lock(); |
2305 | p = rcu_dereference(mm->owner); | |
f75ca962 | 2306 | /* |
ebb76ce1 KH |
2307 | * Because we don't have task_lock(), "p" can exit. |
2308 | * In that case, "mem" can point to root or p can be NULL with | |
2309 | * race with swapoff. Then, we have small risk of mis-accouning. | |
2310 | * But such kind of mis-account by race always happens because | |
2311 | * we don't have cgroup_mutex(). It's overkill and we allo that | |
2312 | * small race, here. | |
2313 | * (*) swapoff at el will charge against mm-struct not against | |
2314 | * task-struct. So, mm->owner can be NULL. | |
f75ca962 KH |
2315 | */ |
2316 | mem = mem_cgroup_from_task(p); | |
ebb76ce1 | 2317 | if (!mem || mem_cgroup_is_root(mem)) { |
f75ca962 KH |
2318 | rcu_read_unlock(); |
2319 | goto done; | |
2320 | } | |
7ec99d62 | 2321 | if (nr_pages == 1 && consume_stock(mem)) { |
f75ca962 KH |
2322 | /* |
2323 | * It seems dagerous to access memcg without css_get(). | |
2324 | * But considering how consume_stok works, it's not | |
2325 | * necessary. If consume_stock success, some charges | |
2326 | * from this memcg are cached on this cpu. So, we | |
2327 | * don't need to call css_get()/css_tryget() before | |
2328 | * calling consume_stock(). | |
2329 | */ | |
2330 | rcu_read_unlock(); | |
2331 | goto done; | |
2332 | } | |
2333 | /* after here, we may be blocked. we need to get refcnt */ | |
2334 | if (!css_tryget(&mem->css)) { | |
2335 | rcu_read_unlock(); | |
2336 | goto again; | |
2337 | } | |
2338 | rcu_read_unlock(); | |
2339 | } | |
8a9f3ccd | 2340 | |
4b534334 KH |
2341 | do { |
2342 | bool oom_check; | |
7a81b88c | 2343 | |
4b534334 | 2344 | /* If killed, bypass charge */ |
f75ca962 KH |
2345 | if (fatal_signal_pending(current)) { |
2346 | css_put(&mem->css); | |
4b534334 | 2347 | goto bypass; |
f75ca962 | 2348 | } |
6d61ef40 | 2349 | |
4b534334 KH |
2350 | oom_check = false; |
2351 | if (oom && !nr_oom_retries) { | |
2352 | oom_check = true; | |
2353 | nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
cdec2e42 | 2354 | } |
66e1707b | 2355 | |
7ec99d62 | 2356 | ret = mem_cgroup_do_charge(mem, gfp_mask, batch, oom_check); |
4b534334 KH |
2357 | switch (ret) { |
2358 | case CHARGE_OK: | |
2359 | break; | |
2360 | case CHARGE_RETRY: /* not in OOM situation but retry */ | |
7ec99d62 | 2361 | batch = nr_pages; |
f75ca962 KH |
2362 | css_put(&mem->css); |
2363 | mem = NULL; | |
2364 | goto again; | |
4b534334 | 2365 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
f75ca962 | 2366 | css_put(&mem->css); |
4b534334 KH |
2367 | goto nomem; |
2368 | case CHARGE_NOMEM: /* OOM routine works */ | |
f75ca962 KH |
2369 | if (!oom) { |
2370 | css_put(&mem->css); | |
867578cb | 2371 | goto nomem; |
f75ca962 | 2372 | } |
4b534334 KH |
2373 | /* If oom, we never return -ENOMEM */ |
2374 | nr_oom_retries--; | |
2375 | break; | |
2376 | case CHARGE_OOM_DIE: /* Killed by OOM Killer */ | |
f75ca962 | 2377 | css_put(&mem->css); |
867578cb | 2378 | goto bypass; |
66e1707b | 2379 | } |
4b534334 KH |
2380 | } while (ret != CHARGE_OK); |
2381 | ||
7ec99d62 JW |
2382 | if (batch > nr_pages) |
2383 | refill_stock(mem, batch - nr_pages); | |
f75ca962 | 2384 | css_put(&mem->css); |
0c3e73e8 | 2385 | done: |
f75ca962 | 2386 | *memcg = mem; |
7a81b88c KH |
2387 | return 0; |
2388 | nomem: | |
f75ca962 | 2389 | *memcg = NULL; |
7a81b88c | 2390 | return -ENOMEM; |
867578cb KH |
2391 | bypass: |
2392 | *memcg = NULL; | |
2393 | return 0; | |
7a81b88c | 2394 | } |
8a9f3ccd | 2395 | |
a3032a2c DN |
2396 | /* |
2397 | * Somemtimes we have to undo a charge we got by try_charge(). | |
2398 | * This function is for that and do uncharge, put css's refcnt. | |
2399 | * gotten by try_charge(). | |
2400 | */ | |
854ffa8d | 2401 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
e7018b8d | 2402 | unsigned int nr_pages) |
a3032a2c DN |
2403 | { |
2404 | if (!mem_cgroup_is_root(mem)) { | |
e7018b8d JW |
2405 | unsigned long bytes = nr_pages * PAGE_SIZE; |
2406 | ||
2407 | res_counter_uncharge(&mem->res, bytes); | |
a3032a2c | 2408 | if (do_swap_account) |
e7018b8d | 2409 | res_counter_uncharge(&mem->memsw, bytes); |
a3032a2c | 2410 | } |
854ffa8d DN |
2411 | } |
2412 | ||
a3b2d692 KH |
2413 | /* |
2414 | * A helper function to get mem_cgroup from ID. must be called under | |
2415 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
2416 | * it's concern. (dropping refcnt from swap can be called against removed | |
2417 | * memcg.) | |
2418 | */ | |
2419 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
2420 | { | |
2421 | struct cgroup_subsys_state *css; | |
2422 | ||
2423 | /* ID 0 is unused ID */ | |
2424 | if (!id) | |
2425 | return NULL; | |
2426 | css = css_lookup(&mem_cgroup_subsys, id); | |
2427 | if (!css) | |
2428 | return NULL; | |
2429 | return container_of(css, struct mem_cgroup, css); | |
2430 | } | |
2431 | ||
e42d9d5d | 2432 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 2433 | { |
e42d9d5d | 2434 | struct mem_cgroup *mem = NULL; |
3c776e64 | 2435 | struct page_cgroup *pc; |
a3b2d692 | 2436 | unsigned short id; |
b5a84319 KH |
2437 | swp_entry_t ent; |
2438 | ||
3c776e64 DN |
2439 | VM_BUG_ON(!PageLocked(page)); |
2440 | ||
3c776e64 | 2441 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 2442 | lock_page_cgroup(pc); |
a3b2d692 | 2443 | if (PageCgroupUsed(pc)) { |
3c776e64 | 2444 | mem = pc->mem_cgroup; |
a3b2d692 KH |
2445 | if (mem && !css_tryget(&mem->css)) |
2446 | mem = NULL; | |
e42d9d5d | 2447 | } else if (PageSwapCache(page)) { |
3c776e64 | 2448 | ent.val = page_private(page); |
a3b2d692 KH |
2449 | id = lookup_swap_cgroup(ent); |
2450 | rcu_read_lock(); | |
2451 | mem = mem_cgroup_lookup(id); | |
2452 | if (mem && !css_tryget(&mem->css)) | |
2453 | mem = NULL; | |
2454 | rcu_read_unlock(); | |
3c776e64 | 2455 | } |
c0bd3f63 | 2456 | unlock_page_cgroup(pc); |
b5a84319 KH |
2457 | return mem; |
2458 | } | |
2459 | ||
ca3e0214 | 2460 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, |
5564e88b | 2461 | struct page *page, |
7ec99d62 | 2462 | unsigned int nr_pages, |
ca3e0214 | 2463 | struct page_cgroup *pc, |
7ec99d62 | 2464 | enum charge_type ctype) |
7a81b88c | 2465 | { |
ca3e0214 KH |
2466 | lock_page_cgroup(pc); |
2467 | if (unlikely(PageCgroupUsed(pc))) { | |
2468 | unlock_page_cgroup(pc); | |
e7018b8d | 2469 | __mem_cgroup_cancel_charge(mem, nr_pages); |
ca3e0214 KH |
2470 | return; |
2471 | } | |
2472 | /* | |
2473 | * we don't need page_cgroup_lock about tail pages, becase they are not | |
2474 | * accessed by any other context at this point. | |
2475 | */ | |
8a9f3ccd | 2476 | pc->mem_cgroup = mem; |
261fb61a KH |
2477 | /* |
2478 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
2479 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
2480 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
2481 | * before USED bit, we need memory barrier here. | |
2482 | * See mem_cgroup_add_lru_list(), etc. | |
2483 | */ | |
08e552c6 | 2484 | smp_wmb(); |
4b3bde4c BS |
2485 | switch (ctype) { |
2486 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
2487 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
2488 | SetPageCgroupCache(pc); | |
2489 | SetPageCgroupUsed(pc); | |
2490 | break; | |
2491 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
2492 | ClearPageCgroupCache(pc); | |
2493 | SetPageCgroupUsed(pc); | |
2494 | break; | |
2495 | default: | |
2496 | break; | |
2497 | } | |
3be91277 | 2498 | |
ca3e0214 | 2499 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), nr_pages); |
52d4b9ac | 2500 | unlock_page_cgroup(pc); |
430e4863 KH |
2501 | /* |
2502 | * "charge_statistics" updated event counter. Then, check it. | |
2503 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
2504 | * if they exceeds softlimit. | |
2505 | */ | |
5564e88b | 2506 | memcg_check_events(mem, page); |
7a81b88c | 2507 | } |
66e1707b | 2508 | |
ca3e0214 KH |
2509 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
2510 | ||
2511 | #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\ | |
2512 | (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION)) | |
2513 | /* | |
2514 | * Because tail pages are not marked as "used", set it. We're under | |
2515 | * zone->lru_lock, 'splitting on pmd' and compund_lock. | |
2516 | */ | |
2517 | void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail) | |
2518 | { | |
2519 | struct page_cgroup *head_pc = lookup_page_cgroup(head); | |
2520 | struct page_cgroup *tail_pc = lookup_page_cgroup(tail); | |
2521 | unsigned long flags; | |
2522 | ||
3d37c4a9 KH |
2523 | if (mem_cgroup_disabled()) |
2524 | return; | |
ca3e0214 | 2525 | /* |
ece35ca8 | 2526 | * We have no races with charge/uncharge but will have races with |
ca3e0214 KH |
2527 | * page state accounting. |
2528 | */ | |
2529 | move_lock_page_cgroup(head_pc, &flags); | |
2530 | ||
2531 | tail_pc->mem_cgroup = head_pc->mem_cgroup; | |
2532 | smp_wmb(); /* see __commit_charge() */ | |
ece35ca8 KH |
2533 | if (PageCgroupAcctLRU(head_pc)) { |
2534 | enum lru_list lru; | |
2535 | struct mem_cgroup_per_zone *mz; | |
2536 | ||
2537 | /* | |
2538 | * LRU flags cannot be copied because we need to add tail | |
2539 | *.page to LRU by generic call and our hook will be called. | |
2540 | * We hold lru_lock, then, reduce counter directly. | |
2541 | */ | |
2542 | lru = page_lru(head); | |
97a6c37b | 2543 | mz = page_cgroup_zoneinfo(head_pc->mem_cgroup, head); |
ece35ca8 KH |
2544 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
2545 | } | |
ca3e0214 KH |
2546 | tail_pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
2547 | move_unlock_page_cgroup(head_pc, &flags); | |
2548 | } | |
2549 | #endif | |
2550 | ||
f817ed48 | 2551 | /** |
de3638d9 | 2552 | * mem_cgroup_move_account - move account of the page |
5564e88b | 2553 | * @page: the page |
7ec99d62 | 2554 | * @nr_pages: number of regular pages (>1 for huge pages) |
f817ed48 KH |
2555 | * @pc: page_cgroup of the page. |
2556 | * @from: mem_cgroup which the page is moved from. | |
2557 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 2558 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
2559 | * |
2560 | * The caller must confirm following. | |
08e552c6 | 2561 | * - page is not on LRU (isolate_page() is useful.) |
7ec99d62 | 2562 | * - compound_lock is held when nr_pages > 1 |
f817ed48 | 2563 | * |
854ffa8d | 2564 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
25985edc | 2565 | * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is |
854ffa8d DN |
2566 | * true, this function does "uncharge" from old cgroup, but it doesn't if |
2567 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 | 2568 | */ |
7ec99d62 JW |
2569 | static int mem_cgroup_move_account(struct page *page, |
2570 | unsigned int nr_pages, | |
2571 | struct page_cgroup *pc, | |
2572 | struct mem_cgroup *from, | |
2573 | struct mem_cgroup *to, | |
2574 | bool uncharge) | |
f817ed48 | 2575 | { |
de3638d9 JW |
2576 | unsigned long flags; |
2577 | int ret; | |
987eba66 | 2578 | |
f817ed48 | 2579 | VM_BUG_ON(from == to); |
5564e88b | 2580 | VM_BUG_ON(PageLRU(page)); |
de3638d9 JW |
2581 | /* |
2582 | * The page is isolated from LRU. So, collapse function | |
2583 | * will not handle this page. But page splitting can happen. | |
2584 | * Do this check under compound_page_lock(). The caller should | |
2585 | * hold it. | |
2586 | */ | |
2587 | ret = -EBUSY; | |
7ec99d62 | 2588 | if (nr_pages > 1 && !PageTransHuge(page)) |
de3638d9 JW |
2589 | goto out; |
2590 | ||
2591 | lock_page_cgroup(pc); | |
2592 | ||
2593 | ret = -EINVAL; | |
2594 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) | |
2595 | goto unlock; | |
2596 | ||
2597 | move_lock_page_cgroup(pc, &flags); | |
f817ed48 | 2598 | |
8725d541 | 2599 | if (PageCgroupFileMapped(pc)) { |
c62b1a3b KH |
2600 | /* Update mapped_file data for mem_cgroup */ |
2601 | preempt_disable(); | |
2602 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2603 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
2604 | preempt_enable(); | |
d69b042f | 2605 | } |
987eba66 | 2606 | mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -nr_pages); |
854ffa8d DN |
2607 | if (uncharge) |
2608 | /* This is not "cancel", but cancel_charge does all we need. */ | |
e7018b8d | 2609 | __mem_cgroup_cancel_charge(from, nr_pages); |
d69b042f | 2610 | |
854ffa8d | 2611 | /* caller should have done css_get */ |
08e552c6 | 2612 | pc->mem_cgroup = to; |
987eba66 | 2613 | mem_cgroup_charge_statistics(to, PageCgroupCache(pc), nr_pages); |
88703267 KH |
2614 | /* |
2615 | * We charges against "to" which may not have any tasks. Then, "to" | |
2616 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe | 2617 | * this function is just force_empty() and move charge, so it's |
25985edc | 2618 | * guaranteed that "to" is never removed. So, we don't check rmdir |
4ffef5fe | 2619 | * status here. |
88703267 | 2620 | */ |
de3638d9 JW |
2621 | move_unlock_page_cgroup(pc, &flags); |
2622 | ret = 0; | |
2623 | unlock: | |
57f9fd7d | 2624 | unlock_page_cgroup(pc); |
d2265e6f KH |
2625 | /* |
2626 | * check events | |
2627 | */ | |
5564e88b JW |
2628 | memcg_check_events(to, page); |
2629 | memcg_check_events(from, page); | |
de3638d9 | 2630 | out: |
f817ed48 KH |
2631 | return ret; |
2632 | } | |
2633 | ||
2634 | /* | |
2635 | * move charges to its parent. | |
2636 | */ | |
2637 | ||
5564e88b JW |
2638 | static int mem_cgroup_move_parent(struct page *page, |
2639 | struct page_cgroup *pc, | |
f817ed48 KH |
2640 | struct mem_cgroup *child, |
2641 | gfp_t gfp_mask) | |
2642 | { | |
2643 | struct cgroup *cg = child->css.cgroup; | |
2644 | struct cgroup *pcg = cg->parent; | |
2645 | struct mem_cgroup *parent; | |
7ec99d62 | 2646 | unsigned int nr_pages; |
4be4489f | 2647 | unsigned long uninitialized_var(flags); |
f817ed48 KH |
2648 | int ret; |
2649 | ||
2650 | /* Is ROOT ? */ | |
2651 | if (!pcg) | |
2652 | return -EINVAL; | |
2653 | ||
57f9fd7d DN |
2654 | ret = -EBUSY; |
2655 | if (!get_page_unless_zero(page)) | |
2656 | goto out; | |
2657 | if (isolate_lru_page(page)) | |
2658 | goto put; | |
52dbb905 | 2659 | |
7ec99d62 | 2660 | nr_pages = hpage_nr_pages(page); |
08e552c6 | 2661 | |
f817ed48 | 2662 | parent = mem_cgroup_from_cont(pcg); |
7ec99d62 | 2663 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false); |
a636b327 | 2664 | if (ret || !parent) |
57f9fd7d | 2665 | goto put_back; |
f817ed48 | 2666 | |
7ec99d62 | 2667 | if (nr_pages > 1) |
987eba66 KH |
2668 | flags = compound_lock_irqsave(page); |
2669 | ||
7ec99d62 | 2670 | ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true); |
854ffa8d | 2671 | if (ret) |
7ec99d62 | 2672 | __mem_cgroup_cancel_charge(parent, nr_pages); |
8dba474f | 2673 | |
7ec99d62 | 2674 | if (nr_pages > 1) |
987eba66 | 2675 | compound_unlock_irqrestore(page, flags); |
8dba474f | 2676 | put_back: |
08e552c6 | 2677 | putback_lru_page(page); |
57f9fd7d | 2678 | put: |
40d58138 | 2679 | put_page(page); |
57f9fd7d | 2680 | out: |
f817ed48 KH |
2681 | return ret; |
2682 | } | |
2683 | ||
7a81b88c KH |
2684 | /* |
2685 | * Charge the memory controller for page usage. | |
2686 | * Return | |
2687 | * 0 if the charge was successful | |
2688 | * < 0 if the cgroup is over its limit | |
2689 | */ | |
2690 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
73045c47 | 2691 | gfp_t gfp_mask, enum charge_type ctype) |
7a81b88c | 2692 | { |
73045c47 | 2693 | struct mem_cgroup *mem = NULL; |
7ec99d62 | 2694 | unsigned int nr_pages = 1; |
7a81b88c | 2695 | struct page_cgroup *pc; |
8493ae43 | 2696 | bool oom = true; |
7a81b88c | 2697 | int ret; |
ec168510 | 2698 | |
37c2ac78 | 2699 | if (PageTransHuge(page)) { |
7ec99d62 | 2700 | nr_pages <<= compound_order(page); |
37c2ac78 | 2701 | VM_BUG_ON(!PageTransHuge(page)); |
8493ae43 JW |
2702 | /* |
2703 | * Never OOM-kill a process for a huge page. The | |
2704 | * fault handler will fall back to regular pages. | |
2705 | */ | |
2706 | oom = false; | |
37c2ac78 | 2707 | } |
7a81b88c KH |
2708 | |
2709 | pc = lookup_page_cgroup(page); | |
af4a6621 | 2710 | BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */ |
7a81b88c | 2711 | |
7ec99d62 | 2712 | ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &mem, oom); |
a636b327 | 2713 | if (ret || !mem) |
7a81b88c KH |
2714 | return ret; |
2715 | ||
7ec99d62 | 2716 | __mem_cgroup_commit_charge(mem, page, nr_pages, pc, ctype); |
8a9f3ccd | 2717 | return 0; |
8a9f3ccd BS |
2718 | } |
2719 | ||
7a81b88c KH |
2720 | int mem_cgroup_newpage_charge(struct page *page, |
2721 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 2722 | { |
f8d66542 | 2723 | if (mem_cgroup_disabled()) |
cede86ac | 2724 | return 0; |
69029cd5 KH |
2725 | /* |
2726 | * If already mapped, we don't have to account. | |
2727 | * If page cache, page->mapping has address_space. | |
2728 | * But page->mapping may have out-of-use anon_vma pointer, | |
2729 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
2730 | * is NULL. | |
2731 | */ | |
2732 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
2733 | return 0; | |
2734 | if (unlikely(!mm)) | |
2735 | mm = &init_mm; | |
217bc319 | 2736 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
73045c47 | 2737 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
217bc319 KH |
2738 | } |
2739 | ||
83aae4c7 DN |
2740 | static void |
2741 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2742 | enum charge_type ctype); | |
2743 | ||
5a6475a4 KH |
2744 | static void |
2745 | __mem_cgroup_commit_charge_lrucare(struct page *page, struct mem_cgroup *mem, | |
2746 | enum charge_type ctype) | |
2747 | { | |
2748 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
2749 | /* | |
2750 | * In some case, SwapCache, FUSE(splice_buf->radixtree), the page | |
2751 | * is already on LRU. It means the page may on some other page_cgroup's | |
2752 | * LRU. Take care of it. | |
2753 | */ | |
2754 | mem_cgroup_lru_del_before_commit(page); | |
2755 | __mem_cgroup_commit_charge(mem, page, 1, pc, ctype); | |
2756 | mem_cgroup_lru_add_after_commit(page); | |
2757 | return; | |
2758 | } | |
2759 | ||
e1a1cd59 BS |
2760 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
2761 | gfp_t gfp_mask) | |
8697d331 | 2762 | { |
5a6475a4 | 2763 | struct mem_cgroup *mem = NULL; |
b5a84319 KH |
2764 | int ret; |
2765 | ||
f8d66542 | 2766 | if (mem_cgroup_disabled()) |
cede86ac | 2767 | return 0; |
52d4b9ac KH |
2768 | if (PageCompound(page)) |
2769 | return 0; | |
accf163e | 2770 | |
73045c47 | 2771 | if (unlikely(!mm)) |
8697d331 | 2772 | mm = &init_mm; |
accf163e | 2773 | |
5a6475a4 KH |
2774 | if (page_is_file_cache(page)) { |
2775 | ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, &mem, true); | |
2776 | if (ret || !mem) | |
2777 | return ret; | |
b5a84319 | 2778 | |
5a6475a4 KH |
2779 | /* |
2780 | * FUSE reuses pages without going through the final | |
2781 | * put that would remove them from the LRU list, make | |
2782 | * sure that they get relinked properly. | |
2783 | */ | |
2784 | __mem_cgroup_commit_charge_lrucare(page, mem, | |
2785 | MEM_CGROUP_CHARGE_TYPE_CACHE); | |
2786 | return ret; | |
2787 | } | |
83aae4c7 DN |
2788 | /* shmem */ |
2789 | if (PageSwapCache(page)) { | |
2790 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
2791 | if (!ret) | |
2792 | __mem_cgroup_commit_charge_swapin(page, mem, | |
2793 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
2794 | } else | |
2795 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
73045c47 | 2796 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
b5a84319 | 2797 | |
b5a84319 | 2798 | return ret; |
e8589cc1 KH |
2799 | } |
2800 | ||
54595fe2 KH |
2801 | /* |
2802 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
2803 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 2804 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
2805 | * "commit()" or removed by "cancel()" |
2806 | */ | |
8c7c6e34 KH |
2807 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
2808 | struct page *page, | |
2809 | gfp_t mask, struct mem_cgroup **ptr) | |
2810 | { | |
2811 | struct mem_cgroup *mem; | |
54595fe2 | 2812 | int ret; |
8c7c6e34 | 2813 | |
56039efa KH |
2814 | *ptr = NULL; |
2815 | ||
f8d66542 | 2816 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
2817 | return 0; |
2818 | ||
2819 | if (!do_swap_account) | |
2820 | goto charge_cur_mm; | |
8c7c6e34 KH |
2821 | /* |
2822 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
2823 | * the pte, and even removed page from swap cache: in those cases |
2824 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
2825 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
2826 | */ |
2827 | if (!PageSwapCache(page)) | |
407f9c8b | 2828 | goto charge_cur_mm; |
e42d9d5d | 2829 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
2830 | if (!mem) |
2831 | goto charge_cur_mm; | |
8c7c6e34 | 2832 | *ptr = mem; |
7ec99d62 | 2833 | ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true); |
54595fe2 KH |
2834 | css_put(&mem->css); |
2835 | return ret; | |
8c7c6e34 KH |
2836 | charge_cur_mm: |
2837 | if (unlikely(!mm)) | |
2838 | mm = &init_mm; | |
7ec99d62 | 2839 | return __mem_cgroup_try_charge(mm, mask, 1, ptr, true); |
8c7c6e34 KH |
2840 | } |
2841 | ||
83aae4c7 DN |
2842 | static void |
2843 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
2844 | enum charge_type ctype) | |
7a81b88c | 2845 | { |
f8d66542 | 2846 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2847 | return; |
2848 | if (!ptr) | |
2849 | return; | |
88703267 | 2850 | cgroup_exclude_rmdir(&ptr->css); |
5a6475a4 KH |
2851 | |
2852 | __mem_cgroup_commit_charge_lrucare(page, ptr, ctype); | |
8c7c6e34 KH |
2853 | /* |
2854 | * Now swap is on-memory. This means this page may be | |
2855 | * counted both as mem and swap....double count. | |
03f3c433 KH |
2856 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
2857 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
2858 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 2859 | */ |
03f3c433 | 2860 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 2861 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 2862 | unsigned short id; |
8c7c6e34 | 2863 | struct mem_cgroup *memcg; |
a3b2d692 KH |
2864 | |
2865 | id = swap_cgroup_record(ent, 0); | |
2866 | rcu_read_lock(); | |
2867 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2868 | if (memcg) { |
a3b2d692 KH |
2869 | /* |
2870 | * This recorded memcg can be obsolete one. So, avoid | |
2871 | * calling css_tryget | |
2872 | */ | |
0c3e73e8 | 2873 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2874 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2875 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2876 | mem_cgroup_put(memcg); |
2877 | } | |
a3b2d692 | 2878 | rcu_read_unlock(); |
8c7c6e34 | 2879 | } |
88703267 KH |
2880 | /* |
2881 | * At swapin, we may charge account against cgroup which has no tasks. | |
2882 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2883 | * In that case, we need to call pre_destroy() again. check it here. | |
2884 | */ | |
2885 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2886 | } |
2887 | ||
83aae4c7 DN |
2888 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2889 | { | |
2890 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2891 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2892 | } | |
2893 | ||
7a81b88c KH |
2894 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2895 | { | |
f8d66542 | 2896 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2897 | return; |
2898 | if (!mem) | |
2899 | return; | |
e7018b8d | 2900 | __mem_cgroup_cancel_charge(mem, 1); |
7a81b88c KH |
2901 | } |
2902 | ||
7ec99d62 JW |
2903 | static void mem_cgroup_do_uncharge(struct mem_cgroup *mem, |
2904 | unsigned int nr_pages, | |
2905 | const enum charge_type ctype) | |
569b846d KH |
2906 | { |
2907 | struct memcg_batch_info *batch = NULL; | |
2908 | bool uncharge_memsw = true; | |
7ec99d62 | 2909 | |
569b846d KH |
2910 | /* If swapout, usage of swap doesn't decrease */ |
2911 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2912 | uncharge_memsw = false; | |
569b846d KH |
2913 | |
2914 | batch = ¤t->memcg_batch; | |
2915 | /* | |
2916 | * In usual, we do css_get() when we remember memcg pointer. | |
2917 | * But in this case, we keep res->usage until end of a series of | |
2918 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2919 | */ | |
2920 | if (!batch->memcg) | |
2921 | batch->memcg = mem; | |
3c11ecf4 KH |
2922 | /* |
2923 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
25985edc | 2924 | * In those cases, all pages freed continuously can be expected to be in |
3c11ecf4 KH |
2925 | * the same cgroup and we have chance to coalesce uncharges. |
2926 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2927 | * because we want to do uncharge as soon as possible. | |
2928 | */ | |
2929 | ||
2930 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) | |
2931 | goto direct_uncharge; | |
2932 | ||
7ec99d62 | 2933 | if (nr_pages > 1) |
ec168510 AA |
2934 | goto direct_uncharge; |
2935 | ||
569b846d KH |
2936 | /* |
2937 | * In typical case, batch->memcg == mem. This means we can | |
2938 | * merge a series of uncharges to an uncharge of res_counter. | |
2939 | * If not, we uncharge res_counter ony by one. | |
2940 | */ | |
2941 | if (batch->memcg != mem) | |
2942 | goto direct_uncharge; | |
2943 | /* remember freed charge and uncharge it later */ | |
7ffd4ca7 | 2944 | batch->nr_pages++; |
569b846d | 2945 | if (uncharge_memsw) |
7ffd4ca7 | 2946 | batch->memsw_nr_pages++; |
569b846d KH |
2947 | return; |
2948 | direct_uncharge: | |
7ec99d62 | 2949 | res_counter_uncharge(&mem->res, nr_pages * PAGE_SIZE); |
569b846d | 2950 | if (uncharge_memsw) |
7ec99d62 | 2951 | res_counter_uncharge(&mem->memsw, nr_pages * PAGE_SIZE); |
3c11ecf4 KH |
2952 | if (unlikely(batch->memcg != mem)) |
2953 | memcg_oom_recover(mem); | |
569b846d KH |
2954 | return; |
2955 | } | |
7a81b88c | 2956 | |
8a9f3ccd | 2957 | /* |
69029cd5 | 2958 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2959 | */ |
8c7c6e34 | 2960 | static struct mem_cgroup * |
69029cd5 | 2961 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2962 | { |
8c7c6e34 | 2963 | struct mem_cgroup *mem = NULL; |
7ec99d62 JW |
2964 | unsigned int nr_pages = 1; |
2965 | struct page_cgroup *pc; | |
8a9f3ccd | 2966 | |
f8d66542 | 2967 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2968 | return NULL; |
4077960e | 2969 | |
d13d1443 | 2970 | if (PageSwapCache(page)) |
8c7c6e34 | 2971 | return NULL; |
d13d1443 | 2972 | |
37c2ac78 | 2973 | if (PageTransHuge(page)) { |
7ec99d62 | 2974 | nr_pages <<= compound_order(page); |
37c2ac78 AA |
2975 | VM_BUG_ON(!PageTransHuge(page)); |
2976 | } | |
8697d331 | 2977 | /* |
3c541e14 | 2978 | * Check if our page_cgroup is valid |
8697d331 | 2979 | */ |
52d4b9ac KH |
2980 | pc = lookup_page_cgroup(page); |
2981 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2982 | return NULL; |
b9c565d5 | 2983 | |
52d4b9ac | 2984 | lock_page_cgroup(pc); |
d13d1443 | 2985 | |
8c7c6e34 KH |
2986 | mem = pc->mem_cgroup; |
2987 | ||
d13d1443 KH |
2988 | if (!PageCgroupUsed(pc)) |
2989 | goto unlock_out; | |
2990 | ||
2991 | switch (ctype) { | |
2992 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2993 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
ac39cf8c | 2994 | /* See mem_cgroup_prepare_migration() */ |
2995 | if (page_mapped(page) || PageCgroupMigration(pc)) | |
d13d1443 KH |
2996 | goto unlock_out; |
2997 | break; | |
2998 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2999 | if (!PageAnon(page)) { /* Shared memory */ | |
3000 | if (page->mapping && !page_is_file_cache(page)) | |
3001 | goto unlock_out; | |
3002 | } else if (page_mapped(page)) /* Anon */ | |
3003 | goto unlock_out; | |
3004 | break; | |
3005 | default: | |
3006 | break; | |
52d4b9ac | 3007 | } |
d13d1443 | 3008 | |
7ec99d62 | 3009 | mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -nr_pages); |
04046e1a | 3010 | |
52d4b9ac | 3011 | ClearPageCgroupUsed(pc); |
544122e5 KH |
3012 | /* |
3013 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
3014 | * freed from LRU. This is safe because uncharged page is expected not | |
3015 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
3016 | * special functions. | |
3017 | */ | |
b9c565d5 | 3018 | |
52d4b9ac | 3019 | unlock_page_cgroup(pc); |
f75ca962 KH |
3020 | /* |
3021 | * even after unlock, we have mem->res.usage here and this memcg | |
3022 | * will never be freed. | |
3023 | */ | |
d2265e6f | 3024 | memcg_check_events(mem, page); |
f75ca962 KH |
3025 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
3026 | mem_cgroup_swap_statistics(mem, true); | |
3027 | mem_cgroup_get(mem); | |
3028 | } | |
3029 | if (!mem_cgroup_is_root(mem)) | |
7ec99d62 | 3030 | mem_cgroup_do_uncharge(mem, nr_pages, ctype); |
6d12e2d8 | 3031 | |
8c7c6e34 | 3032 | return mem; |
d13d1443 KH |
3033 | |
3034 | unlock_out: | |
3035 | unlock_page_cgroup(pc); | |
8c7c6e34 | 3036 | return NULL; |
3c541e14 BS |
3037 | } |
3038 | ||
69029cd5 KH |
3039 | void mem_cgroup_uncharge_page(struct page *page) |
3040 | { | |
52d4b9ac KH |
3041 | /* early check. */ |
3042 | if (page_mapped(page)) | |
3043 | return; | |
3044 | if (page->mapping && !PageAnon(page)) | |
3045 | return; | |
69029cd5 KH |
3046 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
3047 | } | |
3048 | ||
3049 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
3050 | { | |
3051 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 3052 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
3053 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
3054 | } | |
3055 | ||
569b846d KH |
3056 | /* |
3057 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
3058 | * In that cases, pages are freed continuously and we can expect pages | |
3059 | * are in the same memcg. All these calls itself limits the number of | |
3060 | * pages freed at once, then uncharge_start/end() is called properly. | |
3061 | * This may be called prural(2) times in a context, | |
3062 | */ | |
3063 | ||
3064 | void mem_cgroup_uncharge_start(void) | |
3065 | { | |
3066 | current->memcg_batch.do_batch++; | |
3067 | /* We can do nest. */ | |
3068 | if (current->memcg_batch.do_batch == 1) { | |
3069 | current->memcg_batch.memcg = NULL; | |
7ffd4ca7 JW |
3070 | current->memcg_batch.nr_pages = 0; |
3071 | current->memcg_batch.memsw_nr_pages = 0; | |
569b846d KH |
3072 | } |
3073 | } | |
3074 | ||
3075 | void mem_cgroup_uncharge_end(void) | |
3076 | { | |
3077 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
3078 | ||
3079 | if (!batch->do_batch) | |
3080 | return; | |
3081 | ||
3082 | batch->do_batch--; | |
3083 | if (batch->do_batch) /* If stacked, do nothing. */ | |
3084 | return; | |
3085 | ||
3086 | if (!batch->memcg) | |
3087 | return; | |
3088 | /* | |
3089 | * This "batch->memcg" is valid without any css_get/put etc... | |
3090 | * bacause we hide charges behind us. | |
3091 | */ | |
7ffd4ca7 JW |
3092 | if (batch->nr_pages) |
3093 | res_counter_uncharge(&batch->memcg->res, | |
3094 | batch->nr_pages * PAGE_SIZE); | |
3095 | if (batch->memsw_nr_pages) | |
3096 | res_counter_uncharge(&batch->memcg->memsw, | |
3097 | batch->memsw_nr_pages * PAGE_SIZE); | |
3c11ecf4 | 3098 | memcg_oom_recover(batch->memcg); |
569b846d KH |
3099 | /* forget this pointer (for sanity check) */ |
3100 | batch->memcg = NULL; | |
3101 | } | |
3102 | ||
e767e056 | 3103 | #ifdef CONFIG_SWAP |
8c7c6e34 | 3104 | /* |
e767e056 | 3105 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
3106 | * memcg information is recorded to swap_cgroup of "ent" |
3107 | */ | |
8a9478ca KH |
3108 | void |
3109 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
3110 | { |
3111 | struct mem_cgroup *memcg; | |
8a9478ca KH |
3112 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
3113 | ||
3114 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
3115 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
3116 | ||
3117 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 3118 | |
f75ca962 KH |
3119 | /* |
3120 | * record memcg information, if swapout && memcg != NULL, | |
3121 | * mem_cgroup_get() was called in uncharge(). | |
3122 | */ | |
3123 | if (do_swap_account && swapout && memcg) | |
a3b2d692 | 3124 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 | 3125 | } |
e767e056 | 3126 | #endif |
8c7c6e34 KH |
3127 | |
3128 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
3129 | /* | |
3130 | * called from swap_entry_free(). remove record in swap_cgroup and | |
3131 | * uncharge "memsw" account. | |
3132 | */ | |
3133 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 3134 | { |
8c7c6e34 | 3135 | struct mem_cgroup *memcg; |
a3b2d692 | 3136 | unsigned short id; |
8c7c6e34 KH |
3137 | |
3138 | if (!do_swap_account) | |
3139 | return; | |
3140 | ||
a3b2d692 KH |
3141 | id = swap_cgroup_record(ent, 0); |
3142 | rcu_read_lock(); | |
3143 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 3144 | if (memcg) { |
a3b2d692 KH |
3145 | /* |
3146 | * We uncharge this because swap is freed. | |
3147 | * This memcg can be obsolete one. We avoid calling css_tryget | |
3148 | */ | |
0c3e73e8 | 3149 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 3150 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 3151 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
3152 | mem_cgroup_put(memcg); |
3153 | } | |
a3b2d692 | 3154 | rcu_read_unlock(); |
d13d1443 | 3155 | } |
02491447 DN |
3156 | |
3157 | /** | |
3158 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
3159 | * @entry: swap entry to be moved | |
3160 | * @from: mem_cgroup which the entry is moved from | |
3161 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 3162 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
3163 | * |
3164 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
3165 | * as the mem_cgroup's id of @from. | |
3166 | * | |
3167 | * Returns 0 on success, -EINVAL on failure. | |
3168 | * | |
3169 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
3170 | * both res and memsw, and called css_get(). | |
3171 | */ | |
3172 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3173 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3174 | { |
3175 | unsigned short old_id, new_id; | |
3176 | ||
3177 | old_id = css_id(&from->css); | |
3178 | new_id = css_id(&to->css); | |
3179 | ||
3180 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 3181 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 3182 | mem_cgroup_swap_statistics(to, true); |
02491447 | 3183 | /* |
483c30b5 DN |
3184 | * This function is only called from task migration context now. |
3185 | * It postpones res_counter and refcount handling till the end | |
3186 | * of task migration(mem_cgroup_clear_mc()) for performance | |
3187 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
3188 | * because if the process that has been moved to @to does | |
3189 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 3190 | */ |
02491447 | 3191 | mem_cgroup_get(to); |
483c30b5 DN |
3192 | if (need_fixup) { |
3193 | if (!mem_cgroup_is_root(from)) | |
3194 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
3195 | mem_cgroup_put(from); | |
3196 | /* | |
3197 | * we charged both to->res and to->memsw, so we should | |
3198 | * uncharge to->res. | |
3199 | */ | |
3200 | if (!mem_cgroup_is_root(to)) | |
3201 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
483c30b5 | 3202 | } |
02491447 DN |
3203 | return 0; |
3204 | } | |
3205 | return -EINVAL; | |
3206 | } | |
3207 | #else | |
3208 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 3209 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
3210 | { |
3211 | return -EINVAL; | |
3212 | } | |
8c7c6e34 | 3213 | #endif |
d13d1443 | 3214 | |
ae41be37 | 3215 | /* |
01b1ae63 KH |
3216 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
3217 | * page belongs to. | |
ae41be37 | 3218 | */ |
ac39cf8c | 3219 | int mem_cgroup_prepare_migration(struct page *page, |
ef6a3c63 | 3220 | struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask) |
ae41be37 | 3221 | { |
e8589cc1 | 3222 | struct mem_cgroup *mem = NULL; |
7ec99d62 | 3223 | struct page_cgroup *pc; |
ac39cf8c | 3224 | enum charge_type ctype; |
e8589cc1 | 3225 | int ret = 0; |
8869b8f6 | 3226 | |
56039efa KH |
3227 | *ptr = NULL; |
3228 | ||
ec168510 | 3229 | VM_BUG_ON(PageTransHuge(page)); |
f8d66542 | 3230 | if (mem_cgroup_disabled()) |
4077960e BS |
3231 | return 0; |
3232 | ||
52d4b9ac KH |
3233 | pc = lookup_page_cgroup(page); |
3234 | lock_page_cgroup(pc); | |
3235 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
3236 | mem = pc->mem_cgroup; |
3237 | css_get(&mem->css); | |
ac39cf8c | 3238 | /* |
3239 | * At migrating an anonymous page, its mapcount goes down | |
3240 | * to 0 and uncharge() will be called. But, even if it's fully | |
3241 | * unmapped, migration may fail and this page has to be | |
3242 | * charged again. We set MIGRATION flag here and delay uncharge | |
3243 | * until end_migration() is called | |
3244 | * | |
3245 | * Corner Case Thinking | |
3246 | * A) | |
3247 | * When the old page was mapped as Anon and it's unmap-and-freed | |
3248 | * while migration was ongoing. | |
3249 | * If unmap finds the old page, uncharge() of it will be delayed | |
3250 | * until end_migration(). If unmap finds a new page, it's | |
3251 | * uncharged when it make mapcount to be 1->0. If unmap code | |
3252 | * finds swap_migration_entry, the new page will not be mapped | |
3253 | * and end_migration() will find it(mapcount==0). | |
3254 | * | |
3255 | * B) | |
3256 | * When the old page was mapped but migraion fails, the kernel | |
3257 | * remaps it. A charge for it is kept by MIGRATION flag even | |
3258 | * if mapcount goes down to 0. We can do remap successfully | |
3259 | * without charging it again. | |
3260 | * | |
3261 | * C) | |
3262 | * The "old" page is under lock_page() until the end of | |
3263 | * migration, so, the old page itself will not be swapped-out. | |
3264 | * If the new page is swapped out before end_migraton, our | |
3265 | * hook to usual swap-out path will catch the event. | |
3266 | */ | |
3267 | if (PageAnon(page)) | |
3268 | SetPageCgroupMigration(pc); | |
e8589cc1 | 3269 | } |
52d4b9ac | 3270 | unlock_page_cgroup(pc); |
ac39cf8c | 3271 | /* |
3272 | * If the page is not charged at this point, | |
3273 | * we return here. | |
3274 | */ | |
3275 | if (!mem) | |
3276 | return 0; | |
01b1ae63 | 3277 | |
93d5c9be | 3278 | *ptr = mem; |
7ec99d62 | 3279 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false); |
ac39cf8c | 3280 | css_put(&mem->css);/* drop extra refcnt */ |
3281 | if (ret || *ptr == NULL) { | |
3282 | if (PageAnon(page)) { | |
3283 | lock_page_cgroup(pc); | |
3284 | ClearPageCgroupMigration(pc); | |
3285 | unlock_page_cgroup(pc); | |
3286 | /* | |
3287 | * The old page may be fully unmapped while we kept it. | |
3288 | */ | |
3289 | mem_cgroup_uncharge_page(page); | |
3290 | } | |
3291 | return -ENOMEM; | |
e8589cc1 | 3292 | } |
ac39cf8c | 3293 | /* |
3294 | * We charge new page before it's used/mapped. So, even if unlock_page() | |
3295 | * is called before end_migration, we can catch all events on this new | |
3296 | * page. In the case new page is migrated but not remapped, new page's | |
3297 | * mapcount will be finally 0 and we call uncharge in end_migration(). | |
3298 | */ | |
3299 | pc = lookup_page_cgroup(newpage); | |
3300 | if (PageAnon(page)) | |
3301 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
3302 | else if (page_is_file_cache(page)) | |
3303 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
3304 | else | |
3305 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
7ec99d62 | 3306 | __mem_cgroup_commit_charge(mem, page, 1, pc, ctype); |
e8589cc1 | 3307 | return ret; |
ae41be37 | 3308 | } |
8869b8f6 | 3309 | |
69029cd5 | 3310 | /* remove redundant charge if migration failed*/ |
01b1ae63 | 3311 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
50de1dd9 | 3312 | struct page *oldpage, struct page *newpage, bool migration_ok) |
ae41be37 | 3313 | { |
ac39cf8c | 3314 | struct page *used, *unused; |
01b1ae63 | 3315 | struct page_cgroup *pc; |
01b1ae63 KH |
3316 | |
3317 | if (!mem) | |
3318 | return; | |
ac39cf8c | 3319 | /* blocks rmdir() */ |
88703267 | 3320 | cgroup_exclude_rmdir(&mem->css); |
50de1dd9 | 3321 | if (!migration_ok) { |
ac39cf8c | 3322 | used = oldpage; |
3323 | unused = newpage; | |
01b1ae63 | 3324 | } else { |
ac39cf8c | 3325 | used = newpage; |
01b1ae63 KH |
3326 | unused = oldpage; |
3327 | } | |
69029cd5 | 3328 | /* |
ac39cf8c | 3329 | * We disallowed uncharge of pages under migration because mapcount |
3330 | * of the page goes down to zero, temporarly. | |
3331 | * Clear the flag and check the page should be charged. | |
01b1ae63 | 3332 | */ |
ac39cf8c | 3333 | pc = lookup_page_cgroup(oldpage); |
3334 | lock_page_cgroup(pc); | |
3335 | ClearPageCgroupMigration(pc); | |
3336 | unlock_page_cgroup(pc); | |
01b1ae63 | 3337 | |
ac39cf8c | 3338 | __mem_cgroup_uncharge_common(unused, MEM_CGROUP_CHARGE_TYPE_FORCE); |
3339 | ||
01b1ae63 | 3340 | /* |
ac39cf8c | 3341 | * If a page is a file cache, radix-tree replacement is very atomic |
3342 | * and we can skip this check. When it was an Anon page, its mapcount | |
3343 | * goes down to 0. But because we added MIGRATION flage, it's not | |
3344 | * uncharged yet. There are several case but page->mapcount check | |
3345 | * and USED bit check in mem_cgroup_uncharge_page() will do enough | |
3346 | * check. (see prepare_charge() also) | |
69029cd5 | 3347 | */ |
ac39cf8c | 3348 | if (PageAnon(used)) |
3349 | mem_cgroup_uncharge_page(used); | |
88703267 | 3350 | /* |
ac39cf8c | 3351 | * At migration, we may charge account against cgroup which has no |
3352 | * tasks. | |
88703267 KH |
3353 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
3354 | * In that case, we need to call pre_destroy() again. check it here. | |
3355 | */ | |
3356 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 3357 | } |
78fb7466 | 3358 | |
f212ad7c DN |
3359 | #ifdef CONFIG_DEBUG_VM |
3360 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) | |
3361 | { | |
3362 | struct page_cgroup *pc; | |
3363 | ||
3364 | pc = lookup_page_cgroup(page); | |
3365 | if (likely(pc) && PageCgroupUsed(pc)) | |
3366 | return pc; | |
3367 | return NULL; | |
3368 | } | |
3369 | ||
3370 | bool mem_cgroup_bad_page_check(struct page *page) | |
3371 | { | |
3372 | if (mem_cgroup_disabled()) | |
3373 | return false; | |
3374 | ||
3375 | return lookup_page_cgroup_used(page) != NULL; | |
3376 | } | |
3377 | ||
3378 | void mem_cgroup_print_bad_page(struct page *page) | |
3379 | { | |
3380 | struct page_cgroup *pc; | |
3381 | ||
3382 | pc = lookup_page_cgroup_used(page); | |
3383 | if (pc) { | |
3384 | int ret = -1; | |
3385 | char *path; | |
3386 | ||
3387 | printk(KERN_ALERT "pc:%p pc->flags:%lx pc->mem_cgroup:%p", | |
3388 | pc, pc->flags, pc->mem_cgroup); | |
3389 | ||
3390 | path = kmalloc(PATH_MAX, GFP_KERNEL); | |
3391 | if (path) { | |
3392 | rcu_read_lock(); | |
3393 | ret = cgroup_path(pc->mem_cgroup->css.cgroup, | |
3394 | path, PATH_MAX); | |
3395 | rcu_read_unlock(); | |
3396 | } | |
3397 | ||
3398 | printk(KERN_CONT "(%s)\n", | |
3399 | (ret < 0) ? "cannot get the path" : path); | |
3400 | kfree(path); | |
3401 | } | |
3402 | } | |
3403 | #endif | |
3404 | ||
8c7c6e34 KH |
3405 | static DEFINE_MUTEX(set_limit_mutex); |
3406 | ||
d38d2a75 | 3407 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 3408 | unsigned long long val) |
628f4235 | 3409 | { |
81d39c20 | 3410 | int retry_count; |
3c11ecf4 | 3411 | u64 memswlimit, memlimit; |
628f4235 | 3412 | int ret = 0; |
81d39c20 KH |
3413 | int children = mem_cgroup_count_children(memcg); |
3414 | u64 curusage, oldusage; | |
3c11ecf4 | 3415 | int enlarge; |
81d39c20 KH |
3416 | |
3417 | /* | |
3418 | * For keeping hierarchical_reclaim simple, how long we should retry | |
3419 | * is depends on callers. We set our retry-count to be function | |
3420 | * of # of children which we should visit in this loop. | |
3421 | */ | |
3422 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
3423 | ||
3424 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 3425 | |
3c11ecf4 | 3426 | enlarge = 0; |
8c7c6e34 | 3427 | while (retry_count) { |
628f4235 KH |
3428 | if (signal_pending(current)) { |
3429 | ret = -EINTR; | |
3430 | break; | |
3431 | } | |
8c7c6e34 KH |
3432 | /* |
3433 | * Rather than hide all in some function, I do this in | |
3434 | * open coded manner. You see what this really does. | |
3435 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3436 | */ | |
3437 | mutex_lock(&set_limit_mutex); | |
3438 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3439 | if (memswlimit < val) { | |
3440 | ret = -EINVAL; | |
3441 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
3442 | break; |
3443 | } | |
3c11ecf4 KH |
3444 | |
3445 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3446 | if (memlimit < val) | |
3447 | enlarge = 1; | |
3448 | ||
8c7c6e34 | 3449 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
3450 | if (!ret) { |
3451 | if (memswlimit == val) | |
3452 | memcg->memsw_is_minimum = true; | |
3453 | else | |
3454 | memcg->memsw_is_minimum = false; | |
3455 | } | |
8c7c6e34 KH |
3456 | mutex_unlock(&set_limit_mutex); |
3457 | ||
3458 | if (!ret) | |
3459 | break; | |
3460 | ||
aa20d489 | 3461 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
0ae5e89c YH |
3462 | MEM_CGROUP_RECLAIM_SHRINK, |
3463 | NULL); | |
81d39c20 KH |
3464 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
3465 | /* Usage is reduced ? */ | |
3466 | if (curusage >= oldusage) | |
3467 | retry_count--; | |
3468 | else | |
3469 | oldusage = curusage; | |
8c7c6e34 | 3470 | } |
3c11ecf4 KH |
3471 | if (!ret && enlarge) |
3472 | memcg_oom_recover(memcg); | |
14797e23 | 3473 | |
8c7c6e34 KH |
3474 | return ret; |
3475 | } | |
3476 | ||
338c8431 LZ |
3477 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
3478 | unsigned long long val) | |
8c7c6e34 | 3479 | { |
81d39c20 | 3480 | int retry_count; |
3c11ecf4 | 3481 | u64 memlimit, memswlimit, oldusage, curusage; |
81d39c20 KH |
3482 | int children = mem_cgroup_count_children(memcg); |
3483 | int ret = -EBUSY; | |
3c11ecf4 | 3484 | int enlarge = 0; |
8c7c6e34 | 3485 | |
81d39c20 KH |
3486 | /* see mem_cgroup_resize_res_limit */ |
3487 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
3488 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
3489 | while (retry_count) { |
3490 | if (signal_pending(current)) { | |
3491 | ret = -EINTR; | |
3492 | break; | |
3493 | } | |
3494 | /* | |
3495 | * Rather than hide all in some function, I do this in | |
3496 | * open coded manner. You see what this really does. | |
3497 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
3498 | */ | |
3499 | mutex_lock(&set_limit_mutex); | |
3500 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3501 | if (memlimit > val) { | |
3502 | ret = -EINVAL; | |
3503 | mutex_unlock(&set_limit_mutex); | |
3504 | break; | |
3505 | } | |
3c11ecf4 KH |
3506 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
3507 | if (memswlimit < val) | |
3508 | enlarge = 1; | |
8c7c6e34 | 3509 | ret = res_counter_set_limit(&memcg->memsw, val); |
22a668d7 KH |
3510 | if (!ret) { |
3511 | if (memlimit == val) | |
3512 | memcg->memsw_is_minimum = true; | |
3513 | else | |
3514 | memcg->memsw_is_minimum = false; | |
3515 | } | |
8c7c6e34 KH |
3516 | mutex_unlock(&set_limit_mutex); |
3517 | ||
3518 | if (!ret) | |
3519 | break; | |
3520 | ||
4e416953 | 3521 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 | 3522 | MEM_CGROUP_RECLAIM_NOSWAP | |
0ae5e89c YH |
3523 | MEM_CGROUP_RECLAIM_SHRINK, |
3524 | NULL); | |
8c7c6e34 | 3525 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 3526 | /* Usage is reduced ? */ |
8c7c6e34 | 3527 | if (curusage >= oldusage) |
628f4235 | 3528 | retry_count--; |
81d39c20 KH |
3529 | else |
3530 | oldusage = curusage; | |
628f4235 | 3531 | } |
3c11ecf4 KH |
3532 | if (!ret && enlarge) |
3533 | memcg_oom_recover(memcg); | |
628f4235 KH |
3534 | return ret; |
3535 | } | |
3536 | ||
4e416953 | 3537 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
0ae5e89c YH |
3538 | gfp_t gfp_mask, |
3539 | unsigned long *total_scanned) | |
4e416953 BS |
3540 | { |
3541 | unsigned long nr_reclaimed = 0; | |
3542 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
3543 | unsigned long reclaimed; | |
3544 | int loop = 0; | |
3545 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 3546 | unsigned long long excess; |
0ae5e89c | 3547 | unsigned long nr_scanned; |
4e416953 BS |
3548 | |
3549 | if (order > 0) | |
3550 | return 0; | |
3551 | ||
00918b6a | 3552 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
4e416953 BS |
3553 | /* |
3554 | * This loop can run a while, specially if mem_cgroup's continuously | |
3555 | * keep exceeding their soft limit and putting the system under | |
3556 | * pressure | |
3557 | */ | |
3558 | do { | |
3559 | if (next_mz) | |
3560 | mz = next_mz; | |
3561 | else | |
3562 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
3563 | if (!mz) | |
3564 | break; | |
3565 | ||
0ae5e89c | 3566 | nr_scanned = 0; |
4e416953 BS |
3567 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, |
3568 | gfp_mask, | |
0ae5e89c YH |
3569 | MEM_CGROUP_RECLAIM_SOFT, |
3570 | &nr_scanned); | |
4e416953 | 3571 | nr_reclaimed += reclaimed; |
0ae5e89c | 3572 | *total_scanned += nr_scanned; |
4e416953 BS |
3573 | spin_lock(&mctz->lock); |
3574 | ||
3575 | /* | |
3576 | * If we failed to reclaim anything from this memory cgroup | |
3577 | * it is time to move on to the next cgroup | |
3578 | */ | |
3579 | next_mz = NULL; | |
3580 | if (!reclaimed) { | |
3581 | do { | |
3582 | /* | |
3583 | * Loop until we find yet another one. | |
3584 | * | |
3585 | * By the time we get the soft_limit lock | |
3586 | * again, someone might have aded the | |
3587 | * group back on the RB tree. Iterate to | |
3588 | * make sure we get a different mem. | |
3589 | * mem_cgroup_largest_soft_limit_node returns | |
3590 | * NULL if no other cgroup is present on | |
3591 | * the tree | |
3592 | */ | |
3593 | next_mz = | |
3594 | __mem_cgroup_largest_soft_limit_node(mctz); | |
39cc98f1 | 3595 | if (next_mz == mz) |
4e416953 | 3596 | css_put(&next_mz->mem->css); |
39cc98f1 | 3597 | else /* next_mz == NULL or other memcg */ |
4e416953 BS |
3598 | break; |
3599 | } while (1); | |
3600 | } | |
4e416953 | 3601 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 3602 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
3603 | /* |
3604 | * One school of thought says that we should not add | |
3605 | * back the node to the tree if reclaim returns 0. | |
3606 | * But our reclaim could return 0, simply because due | |
3607 | * to priority we are exposing a smaller subset of | |
3608 | * memory to reclaim from. Consider this as a longer | |
3609 | * term TODO. | |
3610 | */ | |
ef8745c1 KH |
3611 | /* If excess == 0, no tree ops */ |
3612 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
3613 | spin_unlock(&mctz->lock); |
3614 | css_put(&mz->mem->css); | |
3615 | loop++; | |
3616 | /* | |
3617 | * Could not reclaim anything and there are no more | |
3618 | * mem cgroups to try or we seem to be looping without | |
3619 | * reclaiming anything. | |
3620 | */ | |
3621 | if (!nr_reclaimed && | |
3622 | (next_mz == NULL || | |
3623 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
3624 | break; | |
3625 | } while (!nr_reclaimed); | |
3626 | if (next_mz) | |
3627 | css_put(&next_mz->mem->css); | |
3628 | return nr_reclaimed; | |
3629 | } | |
3630 | ||
cc847582 KH |
3631 | /* |
3632 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
3633 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
3634 | */ | |
f817ed48 | 3635 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 3636 | int node, int zid, enum lru_list lru) |
cc847582 | 3637 | { |
08e552c6 KH |
3638 | struct zone *zone; |
3639 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 3640 | struct page_cgroup *pc, *busy; |
08e552c6 | 3641 | unsigned long flags, loop; |
072c56c1 | 3642 | struct list_head *list; |
f817ed48 | 3643 | int ret = 0; |
072c56c1 | 3644 | |
08e552c6 KH |
3645 | zone = &NODE_DATA(node)->node_zones[zid]; |
3646 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 3647 | list = &mz->lists[lru]; |
cc847582 | 3648 | |
f817ed48 KH |
3649 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
3650 | /* give some margin against EBUSY etc...*/ | |
3651 | loop += 256; | |
3652 | busy = NULL; | |
3653 | while (loop--) { | |
5564e88b JW |
3654 | struct page *page; |
3655 | ||
f817ed48 | 3656 | ret = 0; |
08e552c6 | 3657 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 3658 | if (list_empty(list)) { |
08e552c6 | 3659 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 3660 | break; |
f817ed48 KH |
3661 | } |
3662 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
3663 | if (busy == pc) { | |
3664 | list_move(&pc->lru, list); | |
648bcc77 | 3665 | busy = NULL; |
08e552c6 | 3666 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
3667 | continue; |
3668 | } | |
08e552c6 | 3669 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 3670 | |
6b3ae58e | 3671 | page = lookup_cgroup_page(pc); |
5564e88b JW |
3672 | |
3673 | ret = mem_cgroup_move_parent(page, pc, mem, GFP_KERNEL); | |
f817ed48 | 3674 | if (ret == -ENOMEM) |
52d4b9ac | 3675 | break; |
f817ed48 KH |
3676 | |
3677 | if (ret == -EBUSY || ret == -EINVAL) { | |
3678 | /* found lock contention or "pc" is obsolete. */ | |
3679 | busy = pc; | |
3680 | cond_resched(); | |
3681 | } else | |
3682 | busy = NULL; | |
cc847582 | 3683 | } |
08e552c6 | 3684 | |
f817ed48 KH |
3685 | if (!ret && !list_empty(list)) |
3686 | return -EBUSY; | |
3687 | return ret; | |
cc847582 KH |
3688 | } |
3689 | ||
3690 | /* | |
3691 | * make mem_cgroup's charge to be 0 if there is no task. | |
3692 | * This enables deleting this mem_cgroup. | |
3693 | */ | |
c1e862c1 | 3694 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 3695 | { |
f817ed48 KH |
3696 | int ret; |
3697 | int node, zid, shrink; | |
3698 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 3699 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 3700 | |
cc847582 | 3701 | css_get(&mem->css); |
f817ed48 KH |
3702 | |
3703 | shrink = 0; | |
c1e862c1 KH |
3704 | /* should free all ? */ |
3705 | if (free_all) | |
3706 | goto try_to_free; | |
f817ed48 | 3707 | move_account: |
fce66477 | 3708 | do { |
f817ed48 | 3709 | ret = -EBUSY; |
c1e862c1 KH |
3710 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
3711 | goto out; | |
3712 | ret = -EINTR; | |
3713 | if (signal_pending(current)) | |
cc847582 | 3714 | goto out; |
52d4b9ac KH |
3715 | /* This is for making all *used* pages to be on LRU. */ |
3716 | lru_add_drain_all(); | |
d38144b7 | 3717 | drain_all_stock_sync(mem); |
f817ed48 | 3718 | ret = 0; |
32047e2a | 3719 | mem_cgroup_start_move(mem); |
299b4eaa | 3720 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 3721 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 3722 | enum lru_list l; |
f817ed48 KH |
3723 | for_each_lru(l) { |
3724 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 3725 | node, zid, l); |
f817ed48 KH |
3726 | if (ret) |
3727 | break; | |
3728 | } | |
1ecaab2b | 3729 | } |
f817ed48 KH |
3730 | if (ret) |
3731 | break; | |
3732 | } | |
32047e2a | 3733 | mem_cgroup_end_move(mem); |
3c11ecf4 | 3734 | memcg_oom_recover(mem); |
f817ed48 KH |
3735 | /* it seems parent cgroup doesn't have enough mem */ |
3736 | if (ret == -ENOMEM) | |
3737 | goto try_to_free; | |
52d4b9ac | 3738 | cond_resched(); |
fce66477 DN |
3739 | /* "ret" should also be checked to ensure all lists are empty. */ |
3740 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
3741 | out: |
3742 | css_put(&mem->css); | |
3743 | return ret; | |
f817ed48 KH |
3744 | |
3745 | try_to_free: | |
c1e862c1 KH |
3746 | /* returns EBUSY if there is a task or if we come here twice. */ |
3747 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
3748 | ret = -EBUSY; |
3749 | goto out; | |
3750 | } | |
c1e862c1 KH |
3751 | /* we call try-to-free pages for make this cgroup empty */ |
3752 | lru_add_drain_all(); | |
f817ed48 KH |
3753 | /* try to free all pages in this cgroup */ |
3754 | shrink = 1; | |
3755 | while (nr_retries && mem->res.usage > 0) { | |
3756 | int progress; | |
c1e862c1 KH |
3757 | |
3758 | if (signal_pending(current)) { | |
3759 | ret = -EINTR; | |
3760 | goto out; | |
3761 | } | |
a7885eb8 | 3762 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
185efc0f | 3763 | false); |
c1e862c1 | 3764 | if (!progress) { |
f817ed48 | 3765 | nr_retries--; |
c1e862c1 | 3766 | /* maybe some writeback is necessary */ |
8aa7e847 | 3767 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 3768 | } |
f817ed48 KH |
3769 | |
3770 | } | |
08e552c6 | 3771 | lru_add_drain(); |
f817ed48 | 3772 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 3773 | goto move_account; |
cc847582 KH |
3774 | } |
3775 | ||
c1e862c1 KH |
3776 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
3777 | { | |
3778 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
3779 | } | |
3780 | ||
3781 | ||
18f59ea7 BS |
3782 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
3783 | { | |
3784 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
3785 | } | |
3786 | ||
3787 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
3788 | u64 val) | |
3789 | { | |
3790 | int retval = 0; | |
3791 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
3792 | struct cgroup *parent = cont->parent; | |
3793 | struct mem_cgroup *parent_mem = NULL; | |
3794 | ||
3795 | if (parent) | |
3796 | parent_mem = mem_cgroup_from_cont(parent); | |
3797 | ||
3798 | cgroup_lock(); | |
3799 | /* | |
af901ca1 | 3800 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
3801 | * in the child subtrees. If it is unset, then the change can |
3802 | * occur, provided the current cgroup has no children. | |
3803 | * | |
3804 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
3805 | * set if there are no children. | |
3806 | */ | |
3807 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
3808 | (val == 1 || val == 0)) { | |
3809 | if (list_empty(&cont->children)) | |
3810 | mem->use_hierarchy = val; | |
3811 | else | |
3812 | retval = -EBUSY; | |
3813 | } else | |
3814 | retval = -EINVAL; | |
3815 | cgroup_unlock(); | |
3816 | ||
3817 | return retval; | |
3818 | } | |
3819 | ||
0c3e73e8 | 3820 | |
7a159cc9 JW |
3821 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *mem, |
3822 | enum mem_cgroup_stat_index idx) | |
0c3e73e8 | 3823 | { |
7d74b06f | 3824 | struct mem_cgroup *iter; |
7a159cc9 | 3825 | long val = 0; |
0c3e73e8 | 3826 | |
7a159cc9 | 3827 | /* Per-cpu values can be negative, use a signed accumulator */ |
7d74b06f KH |
3828 | for_each_mem_cgroup_tree(iter, mem) |
3829 | val += mem_cgroup_read_stat(iter, idx); | |
3830 | ||
3831 | if (val < 0) /* race ? */ | |
3832 | val = 0; | |
3833 | return val; | |
0c3e73e8 BS |
3834 | } |
3835 | ||
104f3928 KS |
3836 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
3837 | { | |
7d74b06f | 3838 | u64 val; |
104f3928 KS |
3839 | |
3840 | if (!mem_cgroup_is_root(mem)) { | |
3841 | if (!swap) | |
3842 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
3843 | else | |
3844 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
3845 | } | |
3846 | ||
7a159cc9 JW |
3847 | val = mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_CACHE); |
3848 | val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_RSS); | |
104f3928 | 3849 | |
7d74b06f | 3850 | if (swap) |
7a159cc9 | 3851 | val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
104f3928 KS |
3852 | |
3853 | return val << PAGE_SHIFT; | |
3854 | } | |
3855 | ||
2c3daa72 | 3856 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 3857 | { |
8c7c6e34 | 3858 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 3859 | u64 val; |
8c7c6e34 KH |
3860 | int type, name; |
3861 | ||
3862 | type = MEMFILE_TYPE(cft->private); | |
3863 | name = MEMFILE_ATTR(cft->private); | |
3864 | switch (type) { | |
3865 | case _MEM: | |
104f3928 KS |
3866 | if (name == RES_USAGE) |
3867 | val = mem_cgroup_usage(mem, false); | |
3868 | else | |
0c3e73e8 | 3869 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
3870 | break; |
3871 | case _MEMSWAP: | |
104f3928 KS |
3872 | if (name == RES_USAGE) |
3873 | val = mem_cgroup_usage(mem, true); | |
3874 | else | |
0c3e73e8 | 3875 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
3876 | break; |
3877 | default: | |
3878 | BUG(); | |
3879 | break; | |
3880 | } | |
3881 | return val; | |
8cdea7c0 | 3882 | } |
628f4235 KH |
3883 | /* |
3884 | * The user of this function is... | |
3885 | * RES_LIMIT. | |
3886 | */ | |
856c13aa PM |
3887 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
3888 | const char *buffer) | |
8cdea7c0 | 3889 | { |
628f4235 | 3890 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 3891 | int type, name; |
628f4235 KH |
3892 | unsigned long long val; |
3893 | int ret; | |
3894 | ||
8c7c6e34 KH |
3895 | type = MEMFILE_TYPE(cft->private); |
3896 | name = MEMFILE_ATTR(cft->private); | |
3897 | switch (name) { | |
628f4235 | 3898 | case RES_LIMIT: |
4b3bde4c BS |
3899 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
3900 | ret = -EINVAL; | |
3901 | break; | |
3902 | } | |
628f4235 KH |
3903 | /* This function does all necessary parse...reuse it */ |
3904 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
3905 | if (ret) |
3906 | break; | |
3907 | if (type == _MEM) | |
628f4235 | 3908 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
3909 | else |
3910 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 3911 | break; |
296c81d8 BS |
3912 | case RES_SOFT_LIMIT: |
3913 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
3914 | if (ret) | |
3915 | break; | |
3916 | /* | |
3917 | * For memsw, soft limits are hard to implement in terms | |
3918 | * of semantics, for now, we support soft limits for | |
3919 | * control without swap | |
3920 | */ | |
3921 | if (type == _MEM) | |
3922 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
3923 | else | |
3924 | ret = -EINVAL; | |
3925 | break; | |
628f4235 KH |
3926 | default: |
3927 | ret = -EINVAL; /* should be BUG() ? */ | |
3928 | break; | |
3929 | } | |
3930 | return ret; | |
8cdea7c0 BS |
3931 | } |
3932 | ||
fee7b548 KH |
3933 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
3934 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
3935 | { | |
3936 | struct cgroup *cgroup; | |
3937 | unsigned long long min_limit, min_memsw_limit, tmp; | |
3938 | ||
3939 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3940 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3941 | cgroup = memcg->css.cgroup; | |
3942 | if (!memcg->use_hierarchy) | |
3943 | goto out; | |
3944 | ||
3945 | while (cgroup->parent) { | |
3946 | cgroup = cgroup->parent; | |
3947 | memcg = mem_cgroup_from_cont(cgroup); | |
3948 | if (!memcg->use_hierarchy) | |
3949 | break; | |
3950 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
3951 | min_limit = min(min_limit, tmp); | |
3952 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
3953 | min_memsw_limit = min(min_memsw_limit, tmp); | |
3954 | } | |
3955 | out: | |
3956 | *mem_limit = min_limit; | |
3957 | *memsw_limit = min_memsw_limit; | |
3958 | return; | |
3959 | } | |
3960 | ||
29f2a4da | 3961 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
3962 | { |
3963 | struct mem_cgroup *mem; | |
8c7c6e34 | 3964 | int type, name; |
c84872e1 PE |
3965 | |
3966 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
3967 | type = MEMFILE_TYPE(event); |
3968 | name = MEMFILE_ATTR(event); | |
3969 | switch (name) { | |
29f2a4da | 3970 | case RES_MAX_USAGE: |
8c7c6e34 KH |
3971 | if (type == _MEM) |
3972 | res_counter_reset_max(&mem->res); | |
3973 | else | |
3974 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
3975 | break; |
3976 | case RES_FAILCNT: | |
8c7c6e34 KH |
3977 | if (type == _MEM) |
3978 | res_counter_reset_failcnt(&mem->res); | |
3979 | else | |
3980 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
3981 | break; |
3982 | } | |
f64c3f54 | 3983 | |
85cc59db | 3984 | return 0; |
c84872e1 PE |
3985 | } |
3986 | ||
7dc74be0 DN |
3987 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3988 | struct cftype *cft) | |
3989 | { | |
3990 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3991 | } | |
3992 | ||
02491447 | 3993 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3994 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3995 | struct cftype *cft, u64 val) | |
3996 | { | |
3997 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3998 | ||
3999 | if (val >= (1 << NR_MOVE_TYPE)) | |
4000 | return -EINVAL; | |
4001 | /* | |
4002 | * We check this value several times in both in can_attach() and | |
4003 | * attach(), so we need cgroup lock to prevent this value from being | |
4004 | * inconsistent. | |
4005 | */ | |
4006 | cgroup_lock(); | |
4007 | mem->move_charge_at_immigrate = val; | |
4008 | cgroup_unlock(); | |
4009 | ||
4010 | return 0; | |
4011 | } | |
02491447 DN |
4012 | #else |
4013 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
4014 | struct cftype *cft, u64 val) | |
4015 | { | |
4016 | return -ENOSYS; | |
4017 | } | |
4018 | #endif | |
7dc74be0 | 4019 | |
14067bb3 KH |
4020 | |
4021 | /* For read statistics */ | |
4022 | enum { | |
4023 | MCS_CACHE, | |
4024 | MCS_RSS, | |
d8046582 | 4025 | MCS_FILE_MAPPED, |
14067bb3 KH |
4026 | MCS_PGPGIN, |
4027 | MCS_PGPGOUT, | |
1dd3a273 | 4028 | MCS_SWAP, |
456f998e YH |
4029 | MCS_PGFAULT, |
4030 | MCS_PGMAJFAULT, | |
14067bb3 KH |
4031 | MCS_INACTIVE_ANON, |
4032 | MCS_ACTIVE_ANON, | |
4033 | MCS_INACTIVE_FILE, | |
4034 | MCS_ACTIVE_FILE, | |
4035 | MCS_UNEVICTABLE, | |
4036 | NR_MCS_STAT, | |
4037 | }; | |
4038 | ||
4039 | struct mcs_total_stat { | |
4040 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
4041 | }; |
4042 | ||
14067bb3 KH |
4043 | struct { |
4044 | char *local_name; | |
4045 | char *total_name; | |
4046 | } memcg_stat_strings[NR_MCS_STAT] = { | |
4047 | {"cache", "total_cache"}, | |
4048 | {"rss", "total_rss"}, | |
d69b042f | 4049 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
4050 | {"pgpgin", "total_pgpgin"}, |
4051 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 4052 | {"swap", "total_swap"}, |
456f998e YH |
4053 | {"pgfault", "total_pgfault"}, |
4054 | {"pgmajfault", "total_pgmajfault"}, | |
14067bb3 KH |
4055 | {"inactive_anon", "total_inactive_anon"}, |
4056 | {"active_anon", "total_active_anon"}, | |
4057 | {"inactive_file", "total_inactive_file"}, | |
4058 | {"active_file", "total_active_file"}, | |
4059 | {"unevictable", "total_unevictable"} | |
4060 | }; | |
4061 | ||
4062 | ||
7d74b06f KH |
4063 | static void |
4064 | mem_cgroup_get_local_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
14067bb3 | 4065 | { |
14067bb3 KH |
4066 | s64 val; |
4067 | ||
4068 | /* per cpu stat */ | |
c62b1a3b | 4069 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 4070 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 4071 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 4072 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 4073 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 4074 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
e9f8974f | 4075 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGPGIN); |
14067bb3 | 4076 | s->stat[MCS_PGPGIN] += val; |
e9f8974f | 4077 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGPGOUT); |
14067bb3 | 4078 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 4079 | if (do_swap_account) { |
c62b1a3b | 4080 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
4081 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
4082 | } | |
456f998e YH |
4083 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGFAULT); |
4084 | s->stat[MCS_PGFAULT] += val; | |
4085 | val = mem_cgroup_read_events(mem, MEM_CGROUP_EVENTS_PGMAJFAULT); | |
4086 | s->stat[MCS_PGMAJFAULT] += val; | |
14067bb3 KH |
4087 | |
4088 | /* per zone stat */ | |
bb2a0de9 | 4089 | val = mem_cgroup_nr_lru_pages(mem, BIT(LRU_INACTIVE_ANON)); |
14067bb3 | 4090 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
bb2a0de9 | 4091 | val = mem_cgroup_nr_lru_pages(mem, BIT(LRU_ACTIVE_ANON)); |
14067bb3 | 4092 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
bb2a0de9 | 4093 | val = mem_cgroup_nr_lru_pages(mem, BIT(LRU_INACTIVE_FILE)); |
14067bb3 | 4094 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
bb2a0de9 | 4095 | val = mem_cgroup_nr_lru_pages(mem, BIT(LRU_ACTIVE_FILE)); |
14067bb3 | 4096 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
bb2a0de9 | 4097 | val = mem_cgroup_nr_lru_pages(mem, BIT(LRU_UNEVICTABLE)); |
14067bb3 | 4098 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
14067bb3 KH |
4099 | } |
4100 | ||
4101 | static void | |
4102 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
4103 | { | |
7d74b06f KH |
4104 | struct mem_cgroup *iter; |
4105 | ||
4106 | for_each_mem_cgroup_tree(iter, mem) | |
4107 | mem_cgroup_get_local_stat(iter, s); | |
14067bb3 KH |
4108 | } |
4109 | ||
406eb0c9 YH |
4110 | #ifdef CONFIG_NUMA |
4111 | static int mem_control_numa_stat_show(struct seq_file *m, void *arg) | |
4112 | { | |
4113 | int nid; | |
4114 | unsigned long total_nr, file_nr, anon_nr, unevictable_nr; | |
4115 | unsigned long node_nr; | |
4116 | struct cgroup *cont = m->private; | |
4117 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); | |
4118 | ||
bb2a0de9 | 4119 | total_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL); |
406eb0c9 YH |
4120 | seq_printf(m, "total=%lu", total_nr); |
4121 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 | 4122 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, LRU_ALL); |
406eb0c9 YH |
4123 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4124 | } | |
4125 | seq_putc(m, '\n'); | |
4126 | ||
bb2a0de9 | 4127 | file_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_FILE); |
406eb0c9 YH |
4128 | seq_printf(m, "file=%lu", file_nr); |
4129 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4130 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4131 | LRU_ALL_FILE); | |
406eb0c9 YH |
4132 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4133 | } | |
4134 | seq_putc(m, '\n'); | |
4135 | ||
bb2a0de9 | 4136 | anon_nr = mem_cgroup_nr_lru_pages(mem_cont, LRU_ALL_ANON); |
406eb0c9 YH |
4137 | seq_printf(m, "anon=%lu", anon_nr); |
4138 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4139 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4140 | LRU_ALL_ANON); | |
406eb0c9 YH |
4141 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4142 | } | |
4143 | seq_putc(m, '\n'); | |
4144 | ||
bb2a0de9 | 4145 | unevictable_nr = mem_cgroup_nr_lru_pages(mem_cont, BIT(LRU_UNEVICTABLE)); |
406eb0c9 YH |
4146 | seq_printf(m, "unevictable=%lu", unevictable_nr); |
4147 | for_each_node_state(nid, N_HIGH_MEMORY) { | |
bb2a0de9 KH |
4148 | node_nr = mem_cgroup_node_nr_lru_pages(mem_cont, nid, |
4149 | BIT(LRU_UNEVICTABLE)); | |
406eb0c9 YH |
4150 | seq_printf(m, " N%d=%lu", nid, node_nr); |
4151 | } | |
4152 | seq_putc(m, '\n'); | |
4153 | return 0; | |
4154 | } | |
4155 | #endif /* CONFIG_NUMA */ | |
4156 | ||
c64745cf PM |
4157 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
4158 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 4159 | { |
d2ceb9b7 | 4160 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 4161 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
4162 | int i; |
4163 | ||
14067bb3 KH |
4164 | memset(&mystat, 0, sizeof(mystat)); |
4165 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 4166 | |
406eb0c9 | 4167 | |
1dd3a273 DN |
4168 | for (i = 0; i < NR_MCS_STAT; i++) { |
4169 | if (i == MCS_SWAP && !do_swap_account) | |
4170 | continue; | |
14067bb3 | 4171 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 4172 | } |
7b854121 | 4173 | |
14067bb3 | 4174 | /* Hierarchical information */ |
fee7b548 KH |
4175 | { |
4176 | unsigned long long limit, memsw_limit; | |
4177 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
4178 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
4179 | if (do_swap_account) | |
4180 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
4181 | } | |
7f016ee8 | 4182 | |
14067bb3 KH |
4183 | memset(&mystat, 0, sizeof(mystat)); |
4184 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
4185 | for (i = 0; i < NR_MCS_STAT; i++) { |
4186 | if (i == MCS_SWAP && !do_swap_account) | |
4187 | continue; | |
14067bb3 | 4188 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 4189 | } |
14067bb3 | 4190 | |
7f016ee8 | 4191 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 4192 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
4193 | |
4194 | { | |
4195 | int nid, zid; | |
4196 | struct mem_cgroup_per_zone *mz; | |
4197 | unsigned long recent_rotated[2] = {0, 0}; | |
4198 | unsigned long recent_scanned[2] = {0, 0}; | |
4199 | ||
4200 | for_each_online_node(nid) | |
4201 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
4202 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
4203 | ||
4204 | recent_rotated[0] += | |
4205 | mz->reclaim_stat.recent_rotated[0]; | |
4206 | recent_rotated[1] += | |
4207 | mz->reclaim_stat.recent_rotated[1]; | |
4208 | recent_scanned[0] += | |
4209 | mz->reclaim_stat.recent_scanned[0]; | |
4210 | recent_scanned[1] += | |
4211 | mz->reclaim_stat.recent_scanned[1]; | |
4212 | } | |
4213 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
4214 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
4215 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
4216 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
4217 | } | |
4218 | #endif | |
4219 | ||
d2ceb9b7 KH |
4220 | return 0; |
4221 | } | |
4222 | ||
a7885eb8 KM |
4223 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
4224 | { | |
4225 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4226 | ||
1f4c025b | 4227 | return mem_cgroup_swappiness(memcg); |
a7885eb8 KM |
4228 | } |
4229 | ||
4230 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
4231 | u64 val) | |
4232 | { | |
4233 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4234 | struct mem_cgroup *parent; | |
068b38c1 | 4235 | |
a7885eb8 KM |
4236 | if (val > 100) |
4237 | return -EINVAL; | |
4238 | ||
4239 | if (cgrp->parent == NULL) | |
4240 | return -EINVAL; | |
4241 | ||
4242 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
4243 | |
4244 | cgroup_lock(); | |
4245 | ||
a7885eb8 KM |
4246 | /* If under hierarchy, only empty-root can set this value */ |
4247 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
4248 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
4249 | cgroup_unlock(); | |
a7885eb8 | 4250 | return -EINVAL; |
068b38c1 | 4251 | } |
a7885eb8 | 4252 | |
a7885eb8 | 4253 | memcg->swappiness = val; |
a7885eb8 | 4254 | |
068b38c1 LZ |
4255 | cgroup_unlock(); |
4256 | ||
a7885eb8 KM |
4257 | return 0; |
4258 | } | |
4259 | ||
2e72b634 KS |
4260 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
4261 | { | |
4262 | struct mem_cgroup_threshold_ary *t; | |
4263 | u64 usage; | |
4264 | int i; | |
4265 | ||
4266 | rcu_read_lock(); | |
4267 | if (!swap) | |
2c488db2 | 4268 | t = rcu_dereference(memcg->thresholds.primary); |
2e72b634 | 4269 | else |
2c488db2 | 4270 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
2e72b634 KS |
4271 | |
4272 | if (!t) | |
4273 | goto unlock; | |
4274 | ||
4275 | usage = mem_cgroup_usage(memcg, swap); | |
4276 | ||
4277 | /* | |
4278 | * current_threshold points to threshold just below usage. | |
4279 | * If it's not true, a threshold was crossed after last | |
4280 | * call of __mem_cgroup_threshold(). | |
4281 | */ | |
5407a562 | 4282 | i = t->current_threshold; |
2e72b634 KS |
4283 | |
4284 | /* | |
4285 | * Iterate backward over array of thresholds starting from | |
4286 | * current_threshold and check if a threshold is crossed. | |
4287 | * If none of thresholds below usage is crossed, we read | |
4288 | * only one element of the array here. | |
4289 | */ | |
4290 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
4291 | eventfd_signal(t->entries[i].eventfd, 1); | |
4292 | ||
4293 | /* i = current_threshold + 1 */ | |
4294 | i++; | |
4295 | ||
4296 | /* | |
4297 | * Iterate forward over array of thresholds starting from | |
4298 | * current_threshold+1 and check if a threshold is crossed. | |
4299 | * If none of thresholds above usage is crossed, we read | |
4300 | * only one element of the array here. | |
4301 | */ | |
4302 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
4303 | eventfd_signal(t->entries[i].eventfd, 1); | |
4304 | ||
4305 | /* Update current_threshold */ | |
5407a562 | 4306 | t->current_threshold = i - 1; |
2e72b634 KS |
4307 | unlock: |
4308 | rcu_read_unlock(); | |
4309 | } | |
4310 | ||
4311 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
4312 | { | |
ad4ca5f4 KS |
4313 | while (memcg) { |
4314 | __mem_cgroup_threshold(memcg, false); | |
4315 | if (do_swap_account) | |
4316 | __mem_cgroup_threshold(memcg, true); | |
4317 | ||
4318 | memcg = parent_mem_cgroup(memcg); | |
4319 | } | |
2e72b634 KS |
4320 | } |
4321 | ||
4322 | static int compare_thresholds(const void *a, const void *b) | |
4323 | { | |
4324 | const struct mem_cgroup_threshold *_a = a; | |
4325 | const struct mem_cgroup_threshold *_b = b; | |
4326 | ||
4327 | return _a->threshold - _b->threshold; | |
4328 | } | |
4329 | ||
7d74b06f | 4330 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *mem) |
9490ff27 KH |
4331 | { |
4332 | struct mem_cgroup_eventfd_list *ev; | |
4333 | ||
4334 | list_for_each_entry(ev, &mem->oom_notify, list) | |
4335 | eventfd_signal(ev->eventfd, 1); | |
4336 | return 0; | |
4337 | } | |
4338 | ||
4339 | static void mem_cgroup_oom_notify(struct mem_cgroup *mem) | |
4340 | { | |
7d74b06f KH |
4341 | struct mem_cgroup *iter; |
4342 | ||
4343 | for_each_mem_cgroup_tree(iter, mem) | |
4344 | mem_cgroup_oom_notify_cb(iter); | |
9490ff27 KH |
4345 | } |
4346 | ||
4347 | static int mem_cgroup_usage_register_event(struct cgroup *cgrp, | |
4348 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
2e72b634 KS |
4349 | { |
4350 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4351 | struct mem_cgroup_thresholds *thresholds; |
4352 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4353 | int type = MEMFILE_TYPE(cft->private); |
4354 | u64 threshold, usage; | |
2c488db2 | 4355 | int i, size, ret; |
2e72b634 KS |
4356 | |
4357 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
4358 | if (ret) | |
4359 | return ret; | |
4360 | ||
4361 | mutex_lock(&memcg->thresholds_lock); | |
2c488db2 | 4362 | |
2e72b634 | 4363 | if (type == _MEM) |
2c488db2 | 4364 | thresholds = &memcg->thresholds; |
2e72b634 | 4365 | else if (type == _MEMSWAP) |
2c488db2 | 4366 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4367 | else |
4368 | BUG(); | |
4369 | ||
4370 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4371 | ||
4372 | /* Check if a threshold crossed before adding a new one */ | |
2c488db2 | 4373 | if (thresholds->primary) |
2e72b634 KS |
4374 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
4375 | ||
2c488db2 | 4376 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
2e72b634 KS |
4377 | |
4378 | /* Allocate memory for new array of thresholds */ | |
2c488db2 | 4379 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
2e72b634 | 4380 | GFP_KERNEL); |
2c488db2 | 4381 | if (!new) { |
2e72b634 KS |
4382 | ret = -ENOMEM; |
4383 | goto unlock; | |
4384 | } | |
2c488db2 | 4385 | new->size = size; |
2e72b634 KS |
4386 | |
4387 | /* Copy thresholds (if any) to new array */ | |
2c488db2 KS |
4388 | if (thresholds->primary) { |
4389 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * | |
2e72b634 | 4390 | sizeof(struct mem_cgroup_threshold)); |
2c488db2 KS |
4391 | } |
4392 | ||
2e72b634 | 4393 | /* Add new threshold */ |
2c488db2 KS |
4394 | new->entries[size - 1].eventfd = eventfd; |
4395 | new->entries[size - 1].threshold = threshold; | |
2e72b634 KS |
4396 | |
4397 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
2c488db2 | 4398 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
2e72b634 KS |
4399 | compare_thresholds, NULL); |
4400 | ||
4401 | /* Find current threshold */ | |
2c488db2 | 4402 | new->current_threshold = -1; |
2e72b634 | 4403 | for (i = 0; i < size; i++) { |
2c488db2 | 4404 | if (new->entries[i].threshold < usage) { |
2e72b634 | 4405 | /* |
2c488db2 KS |
4406 | * new->current_threshold will not be used until |
4407 | * rcu_assign_pointer(), so it's safe to increment | |
2e72b634 KS |
4408 | * it here. |
4409 | */ | |
2c488db2 | 4410 | ++new->current_threshold; |
2e72b634 KS |
4411 | } |
4412 | } | |
4413 | ||
2c488db2 KS |
4414 | /* Free old spare buffer and save old primary buffer as spare */ |
4415 | kfree(thresholds->spare); | |
4416 | thresholds->spare = thresholds->primary; | |
4417 | ||
4418 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4419 | |
907860ed | 4420 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4421 | synchronize_rcu(); |
4422 | ||
2e72b634 KS |
4423 | unlock: |
4424 | mutex_unlock(&memcg->thresholds_lock); | |
4425 | ||
4426 | return ret; | |
4427 | } | |
4428 | ||
907860ed | 4429 | static void mem_cgroup_usage_unregister_event(struct cgroup *cgrp, |
9490ff27 | 4430 | struct cftype *cft, struct eventfd_ctx *eventfd) |
2e72b634 KS |
4431 | { |
4432 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
2c488db2 KS |
4433 | struct mem_cgroup_thresholds *thresholds; |
4434 | struct mem_cgroup_threshold_ary *new; | |
2e72b634 KS |
4435 | int type = MEMFILE_TYPE(cft->private); |
4436 | u64 usage; | |
2c488db2 | 4437 | int i, j, size; |
2e72b634 KS |
4438 | |
4439 | mutex_lock(&memcg->thresholds_lock); | |
4440 | if (type == _MEM) | |
2c488db2 | 4441 | thresholds = &memcg->thresholds; |
2e72b634 | 4442 | else if (type == _MEMSWAP) |
2c488db2 | 4443 | thresholds = &memcg->memsw_thresholds; |
2e72b634 KS |
4444 | else |
4445 | BUG(); | |
4446 | ||
4447 | /* | |
4448 | * Something went wrong if we trying to unregister a threshold | |
4449 | * if we don't have thresholds | |
4450 | */ | |
4451 | BUG_ON(!thresholds); | |
4452 | ||
4453 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
4454 | ||
4455 | /* Check if a threshold crossed before removing */ | |
4456 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
4457 | ||
4458 | /* Calculate new number of threshold */ | |
2c488db2 KS |
4459 | size = 0; |
4460 | for (i = 0; i < thresholds->primary->size; i++) { | |
4461 | if (thresholds->primary->entries[i].eventfd != eventfd) | |
2e72b634 KS |
4462 | size++; |
4463 | } | |
4464 | ||
2c488db2 | 4465 | new = thresholds->spare; |
907860ed | 4466 | |
2e72b634 KS |
4467 | /* Set thresholds array to NULL if we don't have thresholds */ |
4468 | if (!size) { | |
2c488db2 KS |
4469 | kfree(new); |
4470 | new = NULL; | |
907860ed | 4471 | goto swap_buffers; |
2e72b634 KS |
4472 | } |
4473 | ||
2c488db2 | 4474 | new->size = size; |
2e72b634 KS |
4475 | |
4476 | /* Copy thresholds and find current threshold */ | |
2c488db2 KS |
4477 | new->current_threshold = -1; |
4478 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { | |
4479 | if (thresholds->primary->entries[i].eventfd == eventfd) | |
2e72b634 KS |
4480 | continue; |
4481 | ||
2c488db2 KS |
4482 | new->entries[j] = thresholds->primary->entries[i]; |
4483 | if (new->entries[j].threshold < usage) { | |
2e72b634 | 4484 | /* |
2c488db2 | 4485 | * new->current_threshold will not be used |
2e72b634 KS |
4486 | * until rcu_assign_pointer(), so it's safe to increment |
4487 | * it here. | |
4488 | */ | |
2c488db2 | 4489 | ++new->current_threshold; |
2e72b634 KS |
4490 | } |
4491 | j++; | |
4492 | } | |
4493 | ||
907860ed | 4494 | swap_buffers: |
2c488db2 KS |
4495 | /* Swap primary and spare array */ |
4496 | thresholds->spare = thresholds->primary; | |
4497 | rcu_assign_pointer(thresholds->primary, new); | |
2e72b634 | 4498 | |
907860ed | 4499 | /* To be sure that nobody uses thresholds */ |
2e72b634 KS |
4500 | synchronize_rcu(); |
4501 | ||
2e72b634 | 4502 | mutex_unlock(&memcg->thresholds_lock); |
2e72b634 | 4503 | } |
c1e862c1 | 4504 | |
9490ff27 KH |
4505 | static int mem_cgroup_oom_register_event(struct cgroup *cgrp, |
4506 | struct cftype *cft, struct eventfd_ctx *eventfd, const char *args) | |
4507 | { | |
4508 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
4509 | struct mem_cgroup_eventfd_list *event; | |
4510 | int type = MEMFILE_TYPE(cft->private); | |
4511 | ||
4512 | BUG_ON(type != _OOM_TYPE); | |
4513 | event = kmalloc(sizeof(*event), GFP_KERNEL); | |
4514 | if (!event) | |
4515 | return -ENOMEM; | |
4516 | ||
1af8efe9 | 4517 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4518 | |
4519 | event->eventfd = eventfd; | |
4520 | list_add(&event->list, &memcg->oom_notify); | |
4521 | ||
4522 | /* already in OOM ? */ | |
79dfdacc | 4523 | if (atomic_read(&memcg->under_oom)) |
9490ff27 | 4524 | eventfd_signal(eventfd, 1); |
1af8efe9 | 4525 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4526 | |
4527 | return 0; | |
4528 | } | |
4529 | ||
907860ed | 4530 | static void mem_cgroup_oom_unregister_event(struct cgroup *cgrp, |
9490ff27 KH |
4531 | struct cftype *cft, struct eventfd_ctx *eventfd) |
4532 | { | |
4533 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4534 | struct mem_cgroup_eventfd_list *ev, *tmp; | |
4535 | int type = MEMFILE_TYPE(cft->private); | |
4536 | ||
4537 | BUG_ON(type != _OOM_TYPE); | |
4538 | ||
1af8efe9 | 4539 | spin_lock(&memcg_oom_lock); |
9490ff27 KH |
4540 | |
4541 | list_for_each_entry_safe(ev, tmp, &mem->oom_notify, list) { | |
4542 | if (ev->eventfd == eventfd) { | |
4543 | list_del(&ev->list); | |
4544 | kfree(ev); | |
4545 | } | |
4546 | } | |
4547 | ||
1af8efe9 | 4548 | spin_unlock(&memcg_oom_lock); |
9490ff27 KH |
4549 | } |
4550 | ||
3c11ecf4 KH |
4551 | static int mem_cgroup_oom_control_read(struct cgroup *cgrp, |
4552 | struct cftype *cft, struct cgroup_map_cb *cb) | |
4553 | { | |
4554 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4555 | ||
4556 | cb->fill(cb, "oom_kill_disable", mem->oom_kill_disable); | |
4557 | ||
79dfdacc | 4558 | if (atomic_read(&mem->under_oom)) |
3c11ecf4 KH |
4559 | cb->fill(cb, "under_oom", 1); |
4560 | else | |
4561 | cb->fill(cb, "under_oom", 0); | |
4562 | return 0; | |
4563 | } | |
4564 | ||
3c11ecf4 KH |
4565 | static int mem_cgroup_oom_control_write(struct cgroup *cgrp, |
4566 | struct cftype *cft, u64 val) | |
4567 | { | |
4568 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
4569 | struct mem_cgroup *parent; | |
4570 | ||
4571 | /* cannot set to root cgroup and only 0 and 1 are allowed */ | |
4572 | if (!cgrp->parent || !((val == 0) || (val == 1))) | |
4573 | return -EINVAL; | |
4574 | ||
4575 | parent = mem_cgroup_from_cont(cgrp->parent); | |
4576 | ||
4577 | cgroup_lock(); | |
4578 | /* oom-kill-disable is a flag for subhierarchy. */ | |
4579 | if ((parent->use_hierarchy) || | |
4580 | (mem->use_hierarchy && !list_empty(&cgrp->children))) { | |
4581 | cgroup_unlock(); | |
4582 | return -EINVAL; | |
4583 | } | |
4584 | mem->oom_kill_disable = val; | |
4d845ebf KH |
4585 | if (!val) |
4586 | memcg_oom_recover(mem); | |
3c11ecf4 KH |
4587 | cgroup_unlock(); |
4588 | return 0; | |
4589 | } | |
4590 | ||
406eb0c9 YH |
4591 | #ifdef CONFIG_NUMA |
4592 | static const struct file_operations mem_control_numa_stat_file_operations = { | |
4593 | .read = seq_read, | |
4594 | .llseek = seq_lseek, | |
4595 | .release = single_release, | |
4596 | }; | |
4597 | ||
4598 | static int mem_control_numa_stat_open(struct inode *unused, struct file *file) | |
4599 | { | |
4600 | struct cgroup *cont = file->f_dentry->d_parent->d_fsdata; | |
4601 | ||
4602 | file->f_op = &mem_control_numa_stat_file_operations; | |
4603 | return single_open(file, mem_control_numa_stat_show, cont); | |
4604 | } | |
4605 | #endif /* CONFIG_NUMA */ | |
4606 | ||
8cdea7c0 BS |
4607 | static struct cftype mem_cgroup_files[] = { |
4608 | { | |
0eea1030 | 4609 | .name = "usage_in_bytes", |
8c7c6e34 | 4610 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 4611 | .read_u64 = mem_cgroup_read, |
9490ff27 KH |
4612 | .register_event = mem_cgroup_usage_register_event, |
4613 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8cdea7c0 | 4614 | }, |
c84872e1 PE |
4615 | { |
4616 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 4617 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 4618 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
4619 | .read_u64 = mem_cgroup_read, |
4620 | }, | |
8cdea7c0 | 4621 | { |
0eea1030 | 4622 | .name = "limit_in_bytes", |
8c7c6e34 | 4623 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 4624 | .write_string = mem_cgroup_write, |
2c3daa72 | 4625 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4626 | }, |
296c81d8 BS |
4627 | { |
4628 | .name = "soft_limit_in_bytes", | |
4629 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
4630 | .write_string = mem_cgroup_write, | |
4631 | .read_u64 = mem_cgroup_read, | |
4632 | }, | |
8cdea7c0 BS |
4633 | { |
4634 | .name = "failcnt", | |
8c7c6e34 | 4635 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 4636 | .trigger = mem_cgroup_reset, |
2c3daa72 | 4637 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 4638 | }, |
d2ceb9b7 KH |
4639 | { |
4640 | .name = "stat", | |
c64745cf | 4641 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 4642 | }, |
c1e862c1 KH |
4643 | { |
4644 | .name = "force_empty", | |
4645 | .trigger = mem_cgroup_force_empty_write, | |
4646 | }, | |
18f59ea7 BS |
4647 | { |
4648 | .name = "use_hierarchy", | |
4649 | .write_u64 = mem_cgroup_hierarchy_write, | |
4650 | .read_u64 = mem_cgroup_hierarchy_read, | |
4651 | }, | |
a7885eb8 KM |
4652 | { |
4653 | .name = "swappiness", | |
4654 | .read_u64 = mem_cgroup_swappiness_read, | |
4655 | .write_u64 = mem_cgroup_swappiness_write, | |
4656 | }, | |
7dc74be0 DN |
4657 | { |
4658 | .name = "move_charge_at_immigrate", | |
4659 | .read_u64 = mem_cgroup_move_charge_read, | |
4660 | .write_u64 = mem_cgroup_move_charge_write, | |
4661 | }, | |
9490ff27 KH |
4662 | { |
4663 | .name = "oom_control", | |
3c11ecf4 KH |
4664 | .read_map = mem_cgroup_oom_control_read, |
4665 | .write_u64 = mem_cgroup_oom_control_write, | |
9490ff27 KH |
4666 | .register_event = mem_cgroup_oom_register_event, |
4667 | .unregister_event = mem_cgroup_oom_unregister_event, | |
4668 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), | |
4669 | }, | |
406eb0c9 YH |
4670 | #ifdef CONFIG_NUMA |
4671 | { | |
4672 | .name = "numa_stat", | |
4673 | .open = mem_control_numa_stat_open, | |
89577127 | 4674 | .mode = S_IRUGO, |
406eb0c9 YH |
4675 | }, |
4676 | #endif | |
8cdea7c0 BS |
4677 | }; |
4678 | ||
8c7c6e34 KH |
4679 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4680 | static struct cftype memsw_cgroup_files[] = { | |
4681 | { | |
4682 | .name = "memsw.usage_in_bytes", | |
4683 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
4684 | .read_u64 = mem_cgroup_read, | |
9490ff27 KH |
4685 | .register_event = mem_cgroup_usage_register_event, |
4686 | .unregister_event = mem_cgroup_usage_unregister_event, | |
8c7c6e34 KH |
4687 | }, |
4688 | { | |
4689 | .name = "memsw.max_usage_in_bytes", | |
4690 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
4691 | .trigger = mem_cgroup_reset, | |
4692 | .read_u64 = mem_cgroup_read, | |
4693 | }, | |
4694 | { | |
4695 | .name = "memsw.limit_in_bytes", | |
4696 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
4697 | .write_string = mem_cgroup_write, | |
4698 | .read_u64 = mem_cgroup_read, | |
4699 | }, | |
4700 | { | |
4701 | .name = "memsw.failcnt", | |
4702 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
4703 | .trigger = mem_cgroup_reset, | |
4704 | .read_u64 = mem_cgroup_read, | |
4705 | }, | |
4706 | }; | |
4707 | ||
4708 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4709 | { | |
4710 | if (!do_swap_account) | |
4711 | return 0; | |
4712 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
4713 | ARRAY_SIZE(memsw_cgroup_files)); | |
4714 | }; | |
4715 | #else | |
4716 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
4717 | { | |
4718 | return 0; | |
4719 | } | |
4720 | #endif | |
4721 | ||
6d12e2d8 KH |
4722 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4723 | { | |
4724 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 4725 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 4726 | enum lru_list l; |
41e3355d | 4727 | int zone, tmp = node; |
1ecaab2b KH |
4728 | /* |
4729 | * This routine is called against possible nodes. | |
4730 | * But it's BUG to call kmalloc() against offline node. | |
4731 | * | |
4732 | * TODO: this routine can waste much memory for nodes which will | |
4733 | * never be onlined. It's better to use memory hotplug callback | |
4734 | * function. | |
4735 | */ | |
41e3355d KH |
4736 | if (!node_state(node, N_NORMAL_MEMORY)) |
4737 | tmp = -1; | |
17295c88 | 4738 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
6d12e2d8 KH |
4739 | if (!pn) |
4740 | return 1; | |
1ecaab2b | 4741 | |
6d12e2d8 | 4742 | mem->info.nodeinfo[node] = pn; |
1ecaab2b KH |
4743 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
4744 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
4745 | for_each_lru(l) |
4746 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 4747 | mz->usage_in_excess = 0; |
4e416953 BS |
4748 | mz->on_tree = false; |
4749 | mz->mem = mem; | |
1ecaab2b | 4750 | } |
6d12e2d8 KH |
4751 | return 0; |
4752 | } | |
4753 | ||
1ecaab2b KH |
4754 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
4755 | { | |
4756 | kfree(mem->info.nodeinfo[node]); | |
4757 | } | |
4758 | ||
33327948 KH |
4759 | static struct mem_cgroup *mem_cgroup_alloc(void) |
4760 | { | |
4761 | struct mem_cgroup *mem; | |
c62b1a3b | 4762 | int size = sizeof(struct mem_cgroup); |
33327948 | 4763 | |
c62b1a3b | 4764 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb | 4765 | if (size < PAGE_SIZE) |
17295c88 | 4766 | mem = kzalloc(size, GFP_KERNEL); |
33327948 | 4767 | else |
17295c88 | 4768 | mem = vzalloc(size); |
33327948 | 4769 | |
e7bbcdf3 DC |
4770 | if (!mem) |
4771 | return NULL; | |
4772 | ||
c62b1a3b | 4773 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
d2e61b8d DC |
4774 | if (!mem->stat) |
4775 | goto out_free; | |
711d3d2c | 4776 | spin_lock_init(&mem->pcp_counter_lock); |
33327948 | 4777 | return mem; |
d2e61b8d DC |
4778 | |
4779 | out_free: | |
4780 | if (size < PAGE_SIZE) | |
4781 | kfree(mem); | |
4782 | else | |
4783 | vfree(mem); | |
4784 | return NULL; | |
33327948 KH |
4785 | } |
4786 | ||
8c7c6e34 KH |
4787 | /* |
4788 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
4789 | * (scanning all at force_empty is too costly...) | |
4790 | * | |
4791 | * Instead of clearing all references at force_empty, we remember | |
4792 | * the number of reference from swap_cgroup and free mem_cgroup when | |
4793 | * it goes down to 0. | |
4794 | * | |
8c7c6e34 KH |
4795 | * Removal of cgroup itself succeeds regardless of refs from swap. |
4796 | */ | |
4797 | ||
a7ba0eef | 4798 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 4799 | { |
08e552c6 KH |
4800 | int node; |
4801 | ||
f64c3f54 | 4802 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
4803 | free_css_id(&mem_cgroup_subsys, &mem->css); |
4804 | ||
08e552c6 KH |
4805 | for_each_node_state(node, N_POSSIBLE) |
4806 | free_mem_cgroup_per_zone_info(mem, node); | |
4807 | ||
c62b1a3b KH |
4808 | free_percpu(mem->stat); |
4809 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
4810 | kfree(mem); |
4811 | else | |
4812 | vfree(mem); | |
4813 | } | |
4814 | ||
8c7c6e34 KH |
4815 | static void mem_cgroup_get(struct mem_cgroup *mem) |
4816 | { | |
4817 | atomic_inc(&mem->refcnt); | |
4818 | } | |
4819 | ||
483c30b5 | 4820 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 4821 | { |
483c30b5 | 4822 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 4823 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 4824 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
4825 | if (parent) |
4826 | mem_cgroup_put(parent); | |
4827 | } | |
8c7c6e34 KH |
4828 | } |
4829 | ||
483c30b5 DN |
4830 | static void mem_cgroup_put(struct mem_cgroup *mem) |
4831 | { | |
4832 | __mem_cgroup_put(mem, 1); | |
4833 | } | |
4834 | ||
7bcc1bb1 DN |
4835 | /* |
4836 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
4837 | */ | |
4838 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
4839 | { | |
4840 | if (!mem->res.parent) | |
4841 | return NULL; | |
4842 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
4843 | } | |
33327948 | 4844 | |
c077719b KH |
4845 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
4846 | static void __init enable_swap_cgroup(void) | |
4847 | { | |
f8d66542 | 4848 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
4849 | do_swap_account = 1; |
4850 | } | |
4851 | #else | |
4852 | static void __init enable_swap_cgroup(void) | |
4853 | { | |
4854 | } | |
4855 | #endif | |
4856 | ||
f64c3f54 BS |
4857 | static int mem_cgroup_soft_limit_tree_init(void) |
4858 | { | |
4859 | struct mem_cgroup_tree_per_node *rtpn; | |
4860 | struct mem_cgroup_tree_per_zone *rtpz; | |
4861 | int tmp, node, zone; | |
4862 | ||
4863 | for_each_node_state(node, N_POSSIBLE) { | |
4864 | tmp = node; | |
4865 | if (!node_state(node, N_NORMAL_MEMORY)) | |
4866 | tmp = -1; | |
4867 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
4868 | if (!rtpn) | |
4869 | return 1; | |
4870 | ||
4871 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
4872 | ||
4873 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
4874 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
4875 | rtpz->rb_root = RB_ROOT; | |
4876 | spin_lock_init(&rtpz->lock); | |
4877 | } | |
4878 | } | |
4879 | return 0; | |
4880 | } | |
4881 | ||
0eb253e2 | 4882 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
4883 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
4884 | { | |
28dbc4b6 | 4885 | struct mem_cgroup *mem, *parent; |
04046e1a | 4886 | long error = -ENOMEM; |
6d12e2d8 | 4887 | int node; |
8cdea7c0 | 4888 | |
c8dad2bb JB |
4889 | mem = mem_cgroup_alloc(); |
4890 | if (!mem) | |
04046e1a | 4891 | return ERR_PTR(error); |
78fb7466 | 4892 | |
6d12e2d8 KH |
4893 | for_each_node_state(node, N_POSSIBLE) |
4894 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
4895 | goto free_out; | |
f64c3f54 | 4896 | |
c077719b | 4897 | /* root ? */ |
28dbc4b6 | 4898 | if (cont->parent == NULL) { |
cdec2e42 | 4899 | int cpu; |
c077719b | 4900 | enable_swap_cgroup(); |
28dbc4b6 | 4901 | parent = NULL; |
4b3bde4c | 4902 | root_mem_cgroup = mem; |
f64c3f54 BS |
4903 | if (mem_cgroup_soft_limit_tree_init()) |
4904 | goto free_out; | |
cdec2e42 KH |
4905 | for_each_possible_cpu(cpu) { |
4906 | struct memcg_stock_pcp *stock = | |
4907 | &per_cpu(memcg_stock, cpu); | |
4908 | INIT_WORK(&stock->work, drain_local_stock); | |
4909 | } | |
711d3d2c | 4910 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
18f59ea7 | 4911 | } else { |
28dbc4b6 | 4912 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 | 4913 | mem->use_hierarchy = parent->use_hierarchy; |
3c11ecf4 | 4914 | mem->oom_kill_disable = parent->oom_kill_disable; |
18f59ea7 | 4915 | } |
28dbc4b6 | 4916 | |
18f59ea7 BS |
4917 | if (parent && parent->use_hierarchy) { |
4918 | res_counter_init(&mem->res, &parent->res); | |
4919 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
4920 | /* |
4921 | * We increment refcnt of the parent to ensure that we can | |
4922 | * safely access it on res_counter_charge/uncharge. | |
4923 | * This refcnt will be decremented when freeing this | |
4924 | * mem_cgroup(see mem_cgroup_put). | |
4925 | */ | |
4926 | mem_cgroup_get(parent); | |
18f59ea7 BS |
4927 | } else { |
4928 | res_counter_init(&mem->res, NULL); | |
4929 | res_counter_init(&mem->memsw, NULL); | |
4930 | } | |
04046e1a | 4931 | mem->last_scanned_child = 0; |
889976db | 4932 | mem->last_scanned_node = MAX_NUMNODES; |
9490ff27 | 4933 | INIT_LIST_HEAD(&mem->oom_notify); |
6d61ef40 | 4934 | |
a7885eb8 | 4935 | if (parent) |
1f4c025b | 4936 | mem->swappiness = mem_cgroup_swappiness(parent); |
a7ba0eef | 4937 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 4938 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 4939 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 4940 | return &mem->css; |
6d12e2d8 | 4941 | free_out: |
a7ba0eef | 4942 | __mem_cgroup_free(mem); |
4b3bde4c | 4943 | root_mem_cgroup = NULL; |
04046e1a | 4944 | return ERR_PTR(error); |
8cdea7c0 BS |
4945 | } |
4946 | ||
ec64f515 | 4947 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
4948 | struct cgroup *cont) |
4949 | { | |
4950 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
4951 | |
4952 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
4953 | } |
4954 | ||
8cdea7c0 BS |
4955 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
4956 | struct cgroup *cont) | |
4957 | { | |
c268e994 | 4958 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 4959 | |
c268e994 | 4960 | mem_cgroup_put(mem); |
8cdea7c0 BS |
4961 | } |
4962 | ||
4963 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
4964 | struct cgroup *cont) | |
4965 | { | |
8c7c6e34 KH |
4966 | int ret; |
4967 | ||
4968 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
4969 | ARRAY_SIZE(mem_cgroup_files)); | |
4970 | ||
4971 | if (!ret) | |
4972 | ret = register_memsw_files(cont, ss); | |
4973 | return ret; | |
8cdea7c0 BS |
4974 | } |
4975 | ||
02491447 | 4976 | #ifdef CONFIG_MMU |
7dc74be0 | 4977 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
4978 | #define PRECHARGE_COUNT_AT_ONCE 256 |
4979 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 4980 | { |
854ffa8d DN |
4981 | int ret = 0; |
4982 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
4983 | struct mem_cgroup *mem = mc.to; |
4984 | ||
854ffa8d DN |
4985 | if (mem_cgroup_is_root(mem)) { |
4986 | mc.precharge += count; | |
4987 | /* we don't need css_get for root */ | |
4988 | return ret; | |
4989 | } | |
4990 | /* try to charge at once */ | |
4991 | if (count > 1) { | |
4992 | struct res_counter *dummy; | |
4993 | /* | |
4994 | * "mem" cannot be under rmdir() because we've already checked | |
4995 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
4996 | * are still under the same cgroup_mutex. So we can postpone | |
4997 | * css_get(). | |
4998 | */ | |
4999 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
5000 | goto one_by_one; | |
5001 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
5002 | PAGE_SIZE * count, &dummy)) { | |
5003 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
5004 | goto one_by_one; | |
5005 | } | |
5006 | mc.precharge += count; | |
854ffa8d DN |
5007 | return ret; |
5008 | } | |
5009 | one_by_one: | |
5010 | /* fall back to one by one charge */ | |
5011 | while (count--) { | |
5012 | if (signal_pending(current)) { | |
5013 | ret = -EINTR; | |
5014 | break; | |
5015 | } | |
5016 | if (!batch_count--) { | |
5017 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
5018 | cond_resched(); | |
5019 | } | |
7ec99d62 | 5020 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, 1, &mem, false); |
854ffa8d DN |
5021 | if (ret || !mem) |
5022 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
5023 | return -ENOMEM; | |
5024 | mc.precharge++; | |
5025 | } | |
4ffef5fe DN |
5026 | return ret; |
5027 | } | |
5028 | ||
5029 | /** | |
5030 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
5031 | * @vma: the vma the pte to be checked belongs | |
5032 | * @addr: the address corresponding to the pte to be checked | |
5033 | * @ptent: the pte to be checked | |
02491447 | 5034 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
5035 | * |
5036 | * Returns | |
5037 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
5038 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
5039 | * move charge. if @target is not NULL, the page is stored in target->page | |
5040 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
5041 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
5042 | * target for charge migration. if @target is not NULL, the entry is stored | |
5043 | * in target->ent. | |
4ffef5fe DN |
5044 | * |
5045 | * Called with pte lock held. | |
5046 | */ | |
4ffef5fe DN |
5047 | union mc_target { |
5048 | struct page *page; | |
02491447 | 5049 | swp_entry_t ent; |
4ffef5fe DN |
5050 | }; |
5051 | ||
4ffef5fe DN |
5052 | enum mc_target_type { |
5053 | MC_TARGET_NONE, /* not used */ | |
5054 | MC_TARGET_PAGE, | |
02491447 | 5055 | MC_TARGET_SWAP, |
4ffef5fe DN |
5056 | }; |
5057 | ||
90254a65 DN |
5058 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
5059 | unsigned long addr, pte_t ptent) | |
4ffef5fe | 5060 | { |
90254a65 | 5061 | struct page *page = vm_normal_page(vma, addr, ptent); |
4ffef5fe | 5062 | |
90254a65 DN |
5063 | if (!page || !page_mapped(page)) |
5064 | return NULL; | |
5065 | if (PageAnon(page)) { | |
5066 | /* we don't move shared anon */ | |
5067 | if (!move_anon() || page_mapcount(page) > 2) | |
5068 | return NULL; | |
87946a72 DN |
5069 | } else if (!move_file()) |
5070 | /* we ignore mapcount for file pages */ | |
90254a65 DN |
5071 | return NULL; |
5072 | if (!get_page_unless_zero(page)) | |
5073 | return NULL; | |
5074 | ||
5075 | return page; | |
5076 | } | |
5077 | ||
5078 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, | |
5079 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5080 | { | |
5081 | int usage_count; | |
5082 | struct page *page = NULL; | |
5083 | swp_entry_t ent = pte_to_swp_entry(ptent); | |
5084 | ||
5085 | if (!move_anon() || non_swap_entry(ent)) | |
5086 | return NULL; | |
5087 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
5088 | if (usage_count > 1) { /* we don't move shared anon */ | |
02491447 DN |
5089 | if (page) |
5090 | put_page(page); | |
90254a65 | 5091 | return NULL; |
02491447 | 5092 | } |
90254a65 DN |
5093 | if (do_swap_account) |
5094 | entry->val = ent.val; | |
5095 | ||
5096 | return page; | |
5097 | } | |
5098 | ||
87946a72 DN |
5099 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
5100 | unsigned long addr, pte_t ptent, swp_entry_t *entry) | |
5101 | { | |
5102 | struct page *page = NULL; | |
5103 | struct inode *inode; | |
5104 | struct address_space *mapping; | |
5105 | pgoff_t pgoff; | |
5106 | ||
5107 | if (!vma->vm_file) /* anonymous vma */ | |
5108 | return NULL; | |
5109 | if (!move_file()) | |
5110 | return NULL; | |
5111 | ||
5112 | inode = vma->vm_file->f_path.dentry->d_inode; | |
5113 | mapping = vma->vm_file->f_mapping; | |
5114 | if (pte_none(ptent)) | |
5115 | pgoff = linear_page_index(vma, addr); | |
5116 | else /* pte_file(ptent) is true */ | |
5117 | pgoff = pte_to_pgoff(ptent); | |
5118 | ||
5119 | /* page is moved even if it's not RSS of this task(page-faulted). */ | |
aa3b1895 HD |
5120 | page = find_get_page(mapping, pgoff); |
5121 | ||
5122 | #ifdef CONFIG_SWAP | |
5123 | /* shmem/tmpfs may report page out on swap: account for that too. */ | |
5124 | if (radix_tree_exceptional_entry(page)) { | |
5125 | swp_entry_t swap = radix_to_swp_entry(page); | |
87946a72 | 5126 | if (do_swap_account) |
aa3b1895 HD |
5127 | *entry = swap; |
5128 | page = find_get_page(&swapper_space, swap.val); | |
87946a72 | 5129 | } |
aa3b1895 | 5130 | #endif |
87946a72 DN |
5131 | return page; |
5132 | } | |
5133 | ||
90254a65 DN |
5134 | static int is_target_pte_for_mc(struct vm_area_struct *vma, |
5135 | unsigned long addr, pte_t ptent, union mc_target *target) | |
5136 | { | |
5137 | struct page *page = NULL; | |
5138 | struct page_cgroup *pc; | |
5139 | int ret = 0; | |
5140 | swp_entry_t ent = { .val = 0 }; | |
5141 | ||
5142 | if (pte_present(ptent)) | |
5143 | page = mc_handle_present_pte(vma, addr, ptent); | |
5144 | else if (is_swap_pte(ptent)) | |
5145 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); | |
87946a72 DN |
5146 | else if (pte_none(ptent) || pte_file(ptent)) |
5147 | page = mc_handle_file_pte(vma, addr, ptent, &ent); | |
90254a65 DN |
5148 | |
5149 | if (!page && !ent.val) | |
5150 | return 0; | |
02491447 DN |
5151 | if (page) { |
5152 | pc = lookup_page_cgroup(page); | |
5153 | /* | |
5154 | * Do only loose check w/o page_cgroup lock. | |
5155 | * mem_cgroup_move_account() checks the pc is valid or not under | |
5156 | * the lock. | |
5157 | */ | |
5158 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
5159 | ret = MC_TARGET_PAGE; | |
5160 | if (target) | |
5161 | target->page = page; | |
5162 | } | |
5163 | if (!ret || !target) | |
5164 | put_page(page); | |
5165 | } | |
90254a65 DN |
5166 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
5167 | if (ent.val && !ret && | |
7f0f1546 KH |
5168 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { |
5169 | ret = MC_TARGET_SWAP; | |
5170 | if (target) | |
5171 | target->ent = ent; | |
4ffef5fe | 5172 | } |
4ffef5fe DN |
5173 | return ret; |
5174 | } | |
5175 | ||
5176 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
5177 | unsigned long addr, unsigned long end, | |
5178 | struct mm_walk *walk) | |
5179 | { | |
5180 | struct vm_area_struct *vma = walk->private; | |
5181 | pte_t *pte; | |
5182 | spinlock_t *ptl; | |
5183 | ||
03319327 DH |
5184 | split_huge_page_pmd(walk->mm, pmd); |
5185 | ||
4ffef5fe DN |
5186 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
5187 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
5188 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
5189 | mc.precharge++; /* increment precharge temporarily */ | |
5190 | pte_unmap_unlock(pte - 1, ptl); | |
5191 | cond_resched(); | |
5192 | ||
7dc74be0 DN |
5193 | return 0; |
5194 | } | |
5195 | ||
4ffef5fe DN |
5196 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
5197 | { | |
5198 | unsigned long precharge; | |
5199 | struct vm_area_struct *vma; | |
5200 | ||
dfe076b0 | 5201 | down_read(&mm->mmap_sem); |
4ffef5fe DN |
5202 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5203 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
5204 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
5205 | .mm = mm, | |
5206 | .private = vma, | |
5207 | }; | |
5208 | if (is_vm_hugetlb_page(vma)) | |
5209 | continue; | |
4ffef5fe DN |
5210 | walk_page_range(vma->vm_start, vma->vm_end, |
5211 | &mem_cgroup_count_precharge_walk); | |
5212 | } | |
dfe076b0 | 5213 | up_read(&mm->mmap_sem); |
4ffef5fe DN |
5214 | |
5215 | precharge = mc.precharge; | |
5216 | mc.precharge = 0; | |
5217 | ||
5218 | return precharge; | |
5219 | } | |
5220 | ||
4ffef5fe DN |
5221 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
5222 | { | |
dfe076b0 DN |
5223 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
5224 | ||
5225 | VM_BUG_ON(mc.moving_task); | |
5226 | mc.moving_task = current; | |
5227 | return mem_cgroup_do_precharge(precharge); | |
4ffef5fe DN |
5228 | } |
5229 | ||
dfe076b0 DN |
5230 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
5231 | static void __mem_cgroup_clear_mc(void) | |
4ffef5fe | 5232 | { |
2bd9bb20 KH |
5233 | struct mem_cgroup *from = mc.from; |
5234 | struct mem_cgroup *to = mc.to; | |
5235 | ||
4ffef5fe | 5236 | /* we must uncharge all the leftover precharges from mc.to */ |
854ffa8d DN |
5237 | if (mc.precharge) { |
5238 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
5239 | mc.precharge = 0; | |
5240 | } | |
5241 | /* | |
5242 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
5243 | * we must uncharge here. | |
5244 | */ | |
5245 | if (mc.moved_charge) { | |
5246 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
5247 | mc.moved_charge = 0; | |
4ffef5fe | 5248 | } |
483c30b5 DN |
5249 | /* we must fixup refcnts and charges */ |
5250 | if (mc.moved_swap) { | |
483c30b5 DN |
5251 | /* uncharge swap account from the old cgroup */ |
5252 | if (!mem_cgroup_is_root(mc.from)) | |
5253 | res_counter_uncharge(&mc.from->memsw, | |
5254 | PAGE_SIZE * mc.moved_swap); | |
5255 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
5256 | ||
5257 | if (!mem_cgroup_is_root(mc.to)) { | |
5258 | /* | |
5259 | * we charged both to->res and to->memsw, so we should | |
5260 | * uncharge to->res. | |
5261 | */ | |
5262 | res_counter_uncharge(&mc.to->res, | |
5263 | PAGE_SIZE * mc.moved_swap); | |
483c30b5 DN |
5264 | } |
5265 | /* we've already done mem_cgroup_get(mc.to) */ | |
483c30b5 DN |
5266 | mc.moved_swap = 0; |
5267 | } | |
dfe076b0 DN |
5268 | memcg_oom_recover(from); |
5269 | memcg_oom_recover(to); | |
5270 | wake_up_all(&mc.waitq); | |
5271 | } | |
5272 | ||
5273 | static void mem_cgroup_clear_mc(void) | |
5274 | { | |
5275 | struct mem_cgroup *from = mc.from; | |
5276 | ||
5277 | /* | |
5278 | * we must clear moving_task before waking up waiters at the end of | |
5279 | * task migration. | |
5280 | */ | |
5281 | mc.moving_task = NULL; | |
5282 | __mem_cgroup_clear_mc(); | |
2bd9bb20 | 5283 | spin_lock(&mc.lock); |
4ffef5fe DN |
5284 | mc.from = NULL; |
5285 | mc.to = NULL; | |
2bd9bb20 | 5286 | spin_unlock(&mc.lock); |
32047e2a | 5287 | mem_cgroup_end_move(from); |
4ffef5fe DN |
5288 | } |
5289 | ||
7dc74be0 DN |
5290 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
5291 | struct cgroup *cgroup, | |
f780bdb7 | 5292 | struct task_struct *p) |
7dc74be0 DN |
5293 | { |
5294 | int ret = 0; | |
5295 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
5296 | ||
5297 | if (mem->move_charge_at_immigrate) { | |
5298 | struct mm_struct *mm; | |
5299 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
5300 | ||
5301 | VM_BUG_ON(from == mem); | |
5302 | ||
5303 | mm = get_task_mm(p); | |
5304 | if (!mm) | |
5305 | return 0; | |
7dc74be0 | 5306 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
5307 | if (mm->owner == p) { |
5308 | VM_BUG_ON(mc.from); | |
5309 | VM_BUG_ON(mc.to); | |
5310 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 5311 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 5312 | VM_BUG_ON(mc.moved_swap); |
32047e2a | 5313 | mem_cgroup_start_move(from); |
2bd9bb20 | 5314 | spin_lock(&mc.lock); |
4ffef5fe DN |
5315 | mc.from = from; |
5316 | mc.to = mem; | |
2bd9bb20 | 5317 | spin_unlock(&mc.lock); |
dfe076b0 | 5318 | /* We set mc.moving_task later */ |
4ffef5fe DN |
5319 | |
5320 | ret = mem_cgroup_precharge_mc(mm); | |
5321 | if (ret) | |
5322 | mem_cgroup_clear_mc(); | |
dfe076b0 DN |
5323 | } |
5324 | mmput(mm); | |
7dc74be0 DN |
5325 | } |
5326 | return ret; | |
5327 | } | |
5328 | ||
5329 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5330 | struct cgroup *cgroup, | |
f780bdb7 | 5331 | struct task_struct *p) |
7dc74be0 | 5332 | { |
4ffef5fe | 5333 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
5334 | } |
5335 | ||
4ffef5fe DN |
5336 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
5337 | unsigned long addr, unsigned long end, | |
5338 | struct mm_walk *walk) | |
7dc74be0 | 5339 | { |
4ffef5fe DN |
5340 | int ret = 0; |
5341 | struct vm_area_struct *vma = walk->private; | |
5342 | pte_t *pte; | |
5343 | spinlock_t *ptl; | |
5344 | ||
03319327 | 5345 | split_huge_page_pmd(walk->mm, pmd); |
4ffef5fe DN |
5346 | retry: |
5347 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
5348 | for (; addr != end; addr += PAGE_SIZE) { | |
5349 | pte_t ptent = *(pte++); | |
5350 | union mc_target target; | |
5351 | int type; | |
5352 | struct page *page; | |
5353 | struct page_cgroup *pc; | |
02491447 | 5354 | swp_entry_t ent; |
4ffef5fe DN |
5355 | |
5356 | if (!mc.precharge) | |
5357 | break; | |
5358 | ||
5359 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
5360 | switch (type) { | |
5361 | case MC_TARGET_PAGE: | |
5362 | page = target.page; | |
5363 | if (isolate_lru_page(page)) | |
5364 | goto put; | |
5365 | pc = lookup_page_cgroup(page); | |
7ec99d62 JW |
5366 | if (!mem_cgroup_move_account(page, 1, pc, |
5367 | mc.from, mc.to, false)) { | |
4ffef5fe | 5368 | mc.precharge--; |
854ffa8d DN |
5369 | /* we uncharge from mc.from later. */ |
5370 | mc.moved_charge++; | |
4ffef5fe DN |
5371 | } |
5372 | putback_lru_page(page); | |
5373 | put: /* is_target_pte_for_mc() gets the page */ | |
5374 | put_page(page); | |
5375 | break; | |
02491447 DN |
5376 | case MC_TARGET_SWAP: |
5377 | ent = target.ent; | |
483c30b5 DN |
5378 | if (!mem_cgroup_move_swap_account(ent, |
5379 | mc.from, mc.to, false)) { | |
02491447 | 5380 | mc.precharge--; |
483c30b5 DN |
5381 | /* we fixup refcnts and charges later. */ |
5382 | mc.moved_swap++; | |
5383 | } | |
02491447 | 5384 | break; |
4ffef5fe DN |
5385 | default: |
5386 | break; | |
5387 | } | |
5388 | } | |
5389 | pte_unmap_unlock(pte - 1, ptl); | |
5390 | cond_resched(); | |
5391 | ||
5392 | if (addr != end) { | |
5393 | /* | |
5394 | * We have consumed all precharges we got in can_attach(). | |
5395 | * We try charge one by one, but don't do any additional | |
5396 | * charges to mc.to if we have failed in charge once in attach() | |
5397 | * phase. | |
5398 | */ | |
854ffa8d | 5399 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
5400 | if (!ret) |
5401 | goto retry; | |
5402 | } | |
5403 | ||
5404 | return ret; | |
5405 | } | |
5406 | ||
5407 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
5408 | { | |
5409 | struct vm_area_struct *vma; | |
5410 | ||
5411 | lru_add_drain_all(); | |
dfe076b0 DN |
5412 | retry: |
5413 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { | |
5414 | /* | |
5415 | * Someone who are holding the mmap_sem might be waiting in | |
5416 | * waitq. So we cancel all extra charges, wake up all waiters, | |
5417 | * and retry. Because we cancel precharges, we might not be able | |
5418 | * to move enough charges, but moving charge is a best-effort | |
5419 | * feature anyway, so it wouldn't be a big problem. | |
5420 | */ | |
5421 | __mem_cgroup_clear_mc(); | |
5422 | cond_resched(); | |
5423 | goto retry; | |
5424 | } | |
4ffef5fe DN |
5425 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
5426 | int ret; | |
5427 | struct mm_walk mem_cgroup_move_charge_walk = { | |
5428 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
5429 | .mm = mm, | |
5430 | .private = vma, | |
5431 | }; | |
5432 | if (is_vm_hugetlb_page(vma)) | |
5433 | continue; | |
4ffef5fe DN |
5434 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
5435 | &mem_cgroup_move_charge_walk); | |
5436 | if (ret) | |
5437 | /* | |
5438 | * means we have consumed all precharges and failed in | |
5439 | * doing additional charge. Just abandon here. | |
5440 | */ | |
5441 | break; | |
5442 | } | |
dfe076b0 | 5443 | up_read(&mm->mmap_sem); |
7dc74be0 DN |
5444 | } |
5445 | ||
67e465a7 BS |
5446 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
5447 | struct cgroup *cont, | |
5448 | struct cgroup *old_cont, | |
f780bdb7 | 5449 | struct task_struct *p) |
67e465a7 | 5450 | { |
a433658c | 5451 | struct mm_struct *mm = get_task_mm(p); |
dfe076b0 | 5452 | |
dfe076b0 | 5453 | if (mm) { |
a433658c KM |
5454 | if (mc.to) |
5455 | mem_cgroup_move_charge(mm); | |
5456 | put_swap_token(mm); | |
dfe076b0 DN |
5457 | mmput(mm); |
5458 | } | |
a433658c KM |
5459 | if (mc.to) |
5460 | mem_cgroup_clear_mc(); | |
67e465a7 | 5461 | } |
5cfb80a7 DN |
5462 | #else /* !CONFIG_MMU */ |
5463 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
5464 | struct cgroup *cgroup, | |
f780bdb7 | 5465 | struct task_struct *p) |
5cfb80a7 DN |
5466 | { |
5467 | return 0; | |
5468 | } | |
5469 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
5470 | struct cgroup *cgroup, | |
f780bdb7 | 5471 | struct task_struct *p) |
5cfb80a7 DN |
5472 | { |
5473 | } | |
5474 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
5475 | struct cgroup *cont, | |
5476 | struct cgroup *old_cont, | |
f780bdb7 | 5477 | struct task_struct *p) |
5cfb80a7 DN |
5478 | { |
5479 | } | |
5480 | #endif | |
67e465a7 | 5481 | |
8cdea7c0 BS |
5482 | struct cgroup_subsys mem_cgroup_subsys = { |
5483 | .name = "memory", | |
5484 | .subsys_id = mem_cgroup_subsys_id, | |
5485 | .create = mem_cgroup_create, | |
df878fb0 | 5486 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
5487 | .destroy = mem_cgroup_destroy, |
5488 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
5489 | .can_attach = mem_cgroup_can_attach, |
5490 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 5491 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 5492 | .early_init = 0, |
04046e1a | 5493 | .use_id = 1, |
8cdea7c0 | 5494 | }; |
c077719b KH |
5495 | |
5496 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
a42c390c MH |
5497 | static int __init enable_swap_account(char *s) |
5498 | { | |
5499 | /* consider enabled if no parameter or 1 is given */ | |
a2c8990a | 5500 | if (!strcmp(s, "1")) |
a42c390c | 5501 | really_do_swap_account = 1; |
a2c8990a | 5502 | else if (!strcmp(s, "0")) |
a42c390c MH |
5503 | really_do_swap_account = 0; |
5504 | return 1; | |
5505 | } | |
a2c8990a | 5506 | __setup("swapaccount=", enable_swap_account); |
c077719b | 5507 | |
c077719b | 5508 | #endif |