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> |
08e552c6 | 50 | #include "internal.h" |
8cdea7c0 | 51 | |
8697d331 BS |
52 | #include <asm/uaccess.h> |
53 | ||
a181b0e8 | 54 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
a181b0e8 | 55 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
4b3bde4c | 56 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
8cdea7c0 | 57 | |
c077719b | 58 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
338c8431 | 59 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
c077719b KH |
60 | int do_swap_account __read_mostly; |
61 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ | |
62 | #else | |
63 | #define do_swap_account (0) | |
64 | #endif | |
65 | ||
f64c3f54 | 66 | #define SOFTLIMIT_EVENTS_THRESH (1000) |
2e72b634 | 67 | #define THRESHOLDS_EVENTS_THRESH (100) |
c077719b | 68 | |
d52aa412 KH |
69 | /* |
70 | * Statistics for memory cgroup. | |
71 | */ | |
72 | enum mem_cgroup_stat_index { | |
73 | /* | |
74 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. | |
75 | */ | |
76 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ | |
d69b042f | 77 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
d8046582 | 78 | MEM_CGROUP_STAT_FILE_MAPPED, /* # of pages charged as file rss */ |
55e462b0 BR |
79 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
80 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ | |
0c3e73e8 | 81 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
378ce724 KS |
82 | MEM_CGROUP_STAT_SOFTLIMIT, /* decrements on each page in/out. |
83 | used by soft limit implementation */ | |
2e72b634 KS |
84 | MEM_CGROUP_STAT_THRESHOLDS, /* decrements on each page in/out. |
85 | used by threshold implementation */ | |
d52aa412 KH |
86 | |
87 | MEM_CGROUP_STAT_NSTATS, | |
88 | }; | |
89 | ||
90 | struct mem_cgroup_stat_cpu { | |
91 | s64 count[MEM_CGROUP_STAT_NSTATS]; | |
d52aa412 KH |
92 | }; |
93 | ||
6d12e2d8 KH |
94 | /* |
95 | * per-zone information in memory controller. | |
96 | */ | |
6d12e2d8 | 97 | struct mem_cgroup_per_zone { |
072c56c1 KH |
98 | /* |
99 | * spin_lock to protect the per cgroup LRU | |
100 | */ | |
b69408e8 CL |
101 | struct list_head lists[NR_LRU_LISTS]; |
102 | unsigned long count[NR_LRU_LISTS]; | |
3e2f41f1 KM |
103 | |
104 | struct zone_reclaim_stat reclaim_stat; | |
f64c3f54 BS |
105 | struct rb_node tree_node; /* RB tree node */ |
106 | unsigned long long usage_in_excess;/* Set to the value by which */ | |
107 | /* the soft limit is exceeded*/ | |
108 | bool on_tree; | |
4e416953 BS |
109 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
110 | /* use container_of */ | |
6d12e2d8 KH |
111 | }; |
112 | /* Macro for accessing counter */ | |
113 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) | |
114 | ||
115 | struct mem_cgroup_per_node { | |
116 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; | |
117 | }; | |
118 | ||
119 | struct mem_cgroup_lru_info { | |
120 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; | |
121 | }; | |
122 | ||
f64c3f54 BS |
123 | /* |
124 | * Cgroups above their limits are maintained in a RB-Tree, independent of | |
125 | * their hierarchy representation | |
126 | */ | |
127 | ||
128 | struct mem_cgroup_tree_per_zone { | |
129 | struct rb_root rb_root; | |
130 | spinlock_t lock; | |
131 | }; | |
132 | ||
133 | struct mem_cgroup_tree_per_node { | |
134 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; | |
135 | }; | |
136 | ||
137 | struct mem_cgroup_tree { | |
138 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; | |
139 | }; | |
140 | ||
141 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; | |
142 | ||
2e72b634 KS |
143 | struct mem_cgroup_threshold { |
144 | struct eventfd_ctx *eventfd; | |
145 | u64 threshold; | |
146 | }; | |
147 | ||
148 | struct mem_cgroup_threshold_ary { | |
149 | /* An array index points to threshold just below usage. */ | |
150 | atomic_t current_threshold; | |
151 | /* Size of entries[] */ | |
152 | unsigned int size; | |
153 | /* Array of thresholds */ | |
154 | struct mem_cgroup_threshold entries[0]; | |
155 | }; | |
156 | ||
157 | static bool mem_cgroup_threshold_check(struct mem_cgroup *mem); | |
158 | static void mem_cgroup_threshold(struct mem_cgroup *mem); | |
159 | ||
8cdea7c0 BS |
160 | /* |
161 | * The memory controller data structure. The memory controller controls both | |
162 | * page cache and RSS per cgroup. We would eventually like to provide | |
163 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, | |
164 | * to help the administrator determine what knobs to tune. | |
165 | * | |
166 | * TODO: Add a water mark for the memory controller. Reclaim will begin when | |
8a9f3ccd BS |
167 | * we hit the water mark. May be even add a low water mark, such that |
168 | * no reclaim occurs from a cgroup at it's low water mark, this is | |
169 | * a feature that will be implemented much later in the future. | |
8cdea7c0 BS |
170 | */ |
171 | struct mem_cgroup { | |
172 | struct cgroup_subsys_state css; | |
173 | /* | |
174 | * the counter to account for memory usage | |
175 | */ | |
176 | struct res_counter res; | |
8c7c6e34 KH |
177 | /* |
178 | * the counter to account for mem+swap usage. | |
179 | */ | |
180 | struct res_counter memsw; | |
78fb7466 PE |
181 | /* |
182 | * Per cgroup active and inactive list, similar to the | |
183 | * per zone LRU lists. | |
78fb7466 | 184 | */ |
6d12e2d8 | 185 | struct mem_cgroup_lru_info info; |
072c56c1 | 186 | |
2733c06a KM |
187 | /* |
188 | protect against reclaim related member. | |
189 | */ | |
190 | spinlock_t reclaim_param_lock; | |
191 | ||
6c48a1d0 | 192 | int prev_priority; /* for recording reclaim priority */ |
6d61ef40 BS |
193 | |
194 | /* | |
af901ca1 | 195 | * While reclaiming in a hierarchy, we cache the last child we |
04046e1a | 196 | * reclaimed from. |
6d61ef40 | 197 | */ |
04046e1a | 198 | int last_scanned_child; |
18f59ea7 BS |
199 | /* |
200 | * Should the accounting and control be hierarchical, per subtree? | |
201 | */ | |
202 | bool use_hierarchy; | |
a636b327 | 203 | unsigned long last_oom_jiffies; |
8c7c6e34 | 204 | atomic_t refcnt; |
14797e23 | 205 | |
a7885eb8 KM |
206 | unsigned int swappiness; |
207 | ||
22a668d7 KH |
208 | /* set when res.limit == memsw.limit */ |
209 | bool memsw_is_minimum; | |
210 | ||
2e72b634 KS |
211 | /* protect arrays of thresholds */ |
212 | struct mutex thresholds_lock; | |
213 | ||
214 | /* thresholds for memory usage. RCU-protected */ | |
215 | struct mem_cgroup_threshold_ary *thresholds; | |
216 | ||
217 | /* thresholds for mem+swap usage. RCU-protected */ | |
218 | struct mem_cgroup_threshold_ary *memsw_thresholds; | |
219 | ||
7dc74be0 DN |
220 | /* |
221 | * Should we move charges of a task when a task is moved into this | |
222 | * mem_cgroup ? And what type of charges should we move ? | |
223 | */ | |
224 | unsigned long move_charge_at_immigrate; | |
225 | ||
d52aa412 | 226 | /* |
c62b1a3b | 227 | * percpu counter. |
d52aa412 | 228 | */ |
c62b1a3b | 229 | struct mem_cgroup_stat_cpu *stat; |
8cdea7c0 BS |
230 | }; |
231 | ||
7dc74be0 DN |
232 | /* Stuffs for move charges at task migration. */ |
233 | /* | |
234 | * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a | |
235 | * left-shifted bitmap of these types. | |
236 | */ | |
237 | enum move_type { | |
4ffef5fe | 238 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
7dc74be0 DN |
239 | NR_MOVE_TYPE, |
240 | }; | |
241 | ||
4ffef5fe DN |
242 | /* "mc" and its members are protected by cgroup_mutex */ |
243 | static struct move_charge_struct { | |
244 | struct mem_cgroup *from; | |
245 | struct mem_cgroup *to; | |
246 | unsigned long precharge; | |
854ffa8d | 247 | unsigned long moved_charge; |
483c30b5 | 248 | unsigned long moved_swap; |
8033b97c DN |
249 | struct task_struct *moving_task; /* a task moving charges */ |
250 | wait_queue_head_t waitq; /* a waitq for other context */ | |
251 | } mc = { | |
252 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), | |
253 | }; | |
4ffef5fe | 254 | |
4e416953 BS |
255 | /* |
256 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft | |
257 | * limit reclaim to prevent infinite loops, if they ever occur. | |
258 | */ | |
259 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) | |
260 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) | |
261 | ||
217bc319 KH |
262 | enum charge_type { |
263 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, | |
264 | MEM_CGROUP_CHARGE_TYPE_MAPPED, | |
4f98a2fe | 265 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
c05555b5 | 266 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
d13d1443 | 267 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
8a9478ca | 268 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
c05555b5 KH |
269 | NR_CHARGE_TYPE, |
270 | }; | |
271 | ||
52d4b9ac KH |
272 | /* only for here (for easy reading.) */ |
273 | #define PCGF_CACHE (1UL << PCG_CACHE) | |
274 | #define PCGF_USED (1UL << PCG_USED) | |
52d4b9ac | 275 | #define PCGF_LOCK (1UL << PCG_LOCK) |
4b3bde4c BS |
276 | /* Not used, but added here for completeness */ |
277 | #define PCGF_ACCT (1UL << PCG_ACCT) | |
217bc319 | 278 | |
8c7c6e34 KH |
279 | /* for encoding cft->private value on file */ |
280 | #define _MEM (0) | |
281 | #define _MEMSWAP (1) | |
282 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) | |
283 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) | |
284 | #define MEMFILE_ATTR(val) ((val) & 0xffff) | |
285 | ||
75822b44 BS |
286 | /* |
287 | * Reclaim flags for mem_cgroup_hierarchical_reclaim | |
288 | */ | |
289 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 | |
290 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) | |
291 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 | |
292 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) | |
4e416953 BS |
293 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
294 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) | |
75822b44 | 295 | |
8c7c6e34 KH |
296 | static void mem_cgroup_get(struct mem_cgroup *mem); |
297 | static void mem_cgroup_put(struct mem_cgroup *mem); | |
7bcc1bb1 | 298 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
cdec2e42 | 299 | static void drain_all_stock_async(void); |
8c7c6e34 | 300 | |
f64c3f54 BS |
301 | static struct mem_cgroup_per_zone * |
302 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) | |
303 | { | |
304 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; | |
305 | } | |
306 | ||
d324236b WF |
307 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *mem) |
308 | { | |
309 | return &mem->css; | |
310 | } | |
311 | ||
f64c3f54 BS |
312 | static struct mem_cgroup_per_zone * |
313 | page_cgroup_zoneinfo(struct page_cgroup *pc) | |
314 | { | |
315 | struct mem_cgroup *mem = pc->mem_cgroup; | |
316 | int nid = page_cgroup_nid(pc); | |
317 | int zid = page_cgroup_zid(pc); | |
318 | ||
319 | if (!mem) | |
320 | return NULL; | |
321 | ||
322 | return mem_cgroup_zoneinfo(mem, nid, zid); | |
323 | } | |
324 | ||
325 | static struct mem_cgroup_tree_per_zone * | |
326 | soft_limit_tree_node_zone(int nid, int zid) | |
327 | { | |
328 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
329 | } | |
330 | ||
331 | static struct mem_cgroup_tree_per_zone * | |
332 | soft_limit_tree_from_page(struct page *page) | |
333 | { | |
334 | int nid = page_to_nid(page); | |
335 | int zid = page_zonenum(page); | |
336 | ||
337 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; | |
338 | } | |
339 | ||
340 | static void | |
4e416953 | 341 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
f64c3f54 | 342 | struct mem_cgroup_per_zone *mz, |
ef8745c1 KH |
343 | struct mem_cgroup_tree_per_zone *mctz, |
344 | unsigned long long new_usage_in_excess) | |
f64c3f54 BS |
345 | { |
346 | struct rb_node **p = &mctz->rb_root.rb_node; | |
347 | struct rb_node *parent = NULL; | |
348 | struct mem_cgroup_per_zone *mz_node; | |
349 | ||
350 | if (mz->on_tree) | |
351 | return; | |
352 | ||
ef8745c1 KH |
353 | mz->usage_in_excess = new_usage_in_excess; |
354 | if (!mz->usage_in_excess) | |
355 | return; | |
f64c3f54 BS |
356 | while (*p) { |
357 | parent = *p; | |
358 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, | |
359 | tree_node); | |
360 | if (mz->usage_in_excess < mz_node->usage_in_excess) | |
361 | p = &(*p)->rb_left; | |
362 | /* | |
363 | * We can't avoid mem cgroups that are over their soft | |
364 | * limit by the same amount | |
365 | */ | |
366 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) | |
367 | p = &(*p)->rb_right; | |
368 | } | |
369 | rb_link_node(&mz->tree_node, parent, p); | |
370 | rb_insert_color(&mz->tree_node, &mctz->rb_root); | |
371 | mz->on_tree = true; | |
4e416953 BS |
372 | } |
373 | ||
374 | static void | |
375 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
376 | struct mem_cgroup_per_zone *mz, | |
377 | struct mem_cgroup_tree_per_zone *mctz) | |
378 | { | |
379 | if (!mz->on_tree) | |
380 | return; | |
381 | rb_erase(&mz->tree_node, &mctz->rb_root); | |
382 | mz->on_tree = false; | |
383 | } | |
384 | ||
f64c3f54 BS |
385 | static void |
386 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, | |
387 | struct mem_cgroup_per_zone *mz, | |
388 | struct mem_cgroup_tree_per_zone *mctz) | |
389 | { | |
390 | spin_lock(&mctz->lock); | |
4e416953 | 391 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
f64c3f54 BS |
392 | spin_unlock(&mctz->lock); |
393 | } | |
394 | ||
395 | static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem) | |
396 | { | |
397 | bool ret = false; | |
f64c3f54 | 398 | s64 val; |
f64c3f54 | 399 | |
c62b1a3b | 400 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT]); |
378ce724 | 401 | if (unlikely(val < 0)) { |
c62b1a3b | 402 | this_cpu_write(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT], |
378ce724 | 403 | SOFTLIMIT_EVENTS_THRESH); |
f64c3f54 BS |
404 | ret = true; |
405 | } | |
f64c3f54 BS |
406 | return ret; |
407 | } | |
408 | ||
409 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) | |
410 | { | |
ef8745c1 | 411 | unsigned long long excess; |
f64c3f54 BS |
412 | struct mem_cgroup_per_zone *mz; |
413 | struct mem_cgroup_tree_per_zone *mctz; | |
4e649152 KH |
414 | int nid = page_to_nid(page); |
415 | int zid = page_zonenum(page); | |
f64c3f54 BS |
416 | mctz = soft_limit_tree_from_page(page); |
417 | ||
418 | /* | |
4e649152 KH |
419 | * Necessary to update all ancestors when hierarchy is used. |
420 | * because their event counter is not touched. | |
f64c3f54 | 421 | */ |
4e649152 KH |
422 | for (; mem; mem = parent_mem_cgroup(mem)) { |
423 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
ef8745c1 | 424 | excess = res_counter_soft_limit_excess(&mem->res); |
4e649152 KH |
425 | /* |
426 | * We have to update the tree if mz is on RB-tree or | |
427 | * mem is over its softlimit. | |
428 | */ | |
ef8745c1 | 429 | if (excess || mz->on_tree) { |
4e649152 KH |
430 | spin_lock(&mctz->lock); |
431 | /* if on-tree, remove it */ | |
432 | if (mz->on_tree) | |
433 | __mem_cgroup_remove_exceeded(mem, mz, mctz); | |
434 | /* | |
ef8745c1 KH |
435 | * Insert again. mz->usage_in_excess will be updated. |
436 | * If excess is 0, no tree ops. | |
4e649152 | 437 | */ |
ef8745c1 | 438 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
4e649152 KH |
439 | spin_unlock(&mctz->lock); |
440 | } | |
f64c3f54 BS |
441 | } |
442 | } | |
443 | ||
444 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) | |
445 | { | |
446 | int node, zone; | |
447 | struct mem_cgroup_per_zone *mz; | |
448 | struct mem_cgroup_tree_per_zone *mctz; | |
449 | ||
450 | for_each_node_state(node, N_POSSIBLE) { | |
451 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
452 | mz = mem_cgroup_zoneinfo(mem, node, zone); | |
453 | mctz = soft_limit_tree_node_zone(node, zone); | |
454 | mem_cgroup_remove_exceeded(mem, mz, mctz); | |
455 | } | |
456 | } | |
457 | } | |
458 | ||
4e416953 BS |
459 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
460 | { | |
461 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; | |
462 | } | |
463 | ||
464 | static struct mem_cgroup_per_zone * | |
465 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
466 | { | |
467 | struct rb_node *rightmost = NULL; | |
26251eaf | 468 | struct mem_cgroup_per_zone *mz; |
4e416953 BS |
469 | |
470 | retry: | |
26251eaf | 471 | mz = NULL; |
4e416953 BS |
472 | rightmost = rb_last(&mctz->rb_root); |
473 | if (!rightmost) | |
474 | goto done; /* Nothing to reclaim from */ | |
475 | ||
476 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); | |
477 | /* | |
478 | * Remove the node now but someone else can add it back, | |
479 | * we will to add it back at the end of reclaim to its correct | |
480 | * position in the tree. | |
481 | */ | |
482 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); | |
483 | if (!res_counter_soft_limit_excess(&mz->mem->res) || | |
484 | !css_tryget(&mz->mem->css)) | |
485 | goto retry; | |
486 | done: | |
487 | return mz; | |
488 | } | |
489 | ||
490 | static struct mem_cgroup_per_zone * | |
491 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) | |
492 | { | |
493 | struct mem_cgroup_per_zone *mz; | |
494 | ||
495 | spin_lock(&mctz->lock); | |
496 | mz = __mem_cgroup_largest_soft_limit_node(mctz); | |
497 | spin_unlock(&mctz->lock); | |
498 | return mz; | |
499 | } | |
500 | ||
c62b1a3b KH |
501 | static s64 mem_cgroup_read_stat(struct mem_cgroup *mem, |
502 | enum mem_cgroup_stat_index idx) | |
503 | { | |
504 | int cpu; | |
505 | s64 val = 0; | |
506 | ||
507 | for_each_possible_cpu(cpu) | |
508 | val += per_cpu(mem->stat->count[idx], cpu); | |
509 | return val; | |
510 | } | |
511 | ||
512 | static s64 mem_cgroup_local_usage(struct mem_cgroup *mem) | |
513 | { | |
514 | s64 ret; | |
515 | ||
516 | ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); | |
517 | ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); | |
518 | return ret; | |
519 | } | |
520 | ||
0c3e73e8 BS |
521 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
522 | bool charge) | |
523 | { | |
524 | int val = (charge) ? 1 : -1; | |
c62b1a3b | 525 | this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_SWAPOUT], val); |
0c3e73e8 BS |
526 | } |
527 | ||
c05555b5 KH |
528 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
529 | struct page_cgroup *pc, | |
530 | bool charge) | |
d52aa412 | 531 | { |
0c3e73e8 | 532 | int val = (charge) ? 1 : -1; |
d52aa412 | 533 | |
c62b1a3b KH |
534 | preempt_disable(); |
535 | ||
c05555b5 | 536 | if (PageCgroupCache(pc)) |
c62b1a3b | 537 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val); |
d52aa412 | 538 | else |
c62b1a3b | 539 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val); |
55e462b0 BR |
540 | |
541 | if (charge) | |
c62b1a3b | 542 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]); |
55e462b0 | 543 | else |
c62b1a3b KH |
544 | __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]); |
545 | __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_SOFTLIMIT]); | |
546 | __this_cpu_dec(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS]); | |
2e72b634 | 547 | |
c62b1a3b | 548 | preempt_enable(); |
6d12e2d8 KH |
549 | } |
550 | ||
14067bb3 | 551 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
b69408e8 | 552 | enum lru_list idx) |
6d12e2d8 KH |
553 | { |
554 | int nid, zid; | |
555 | struct mem_cgroup_per_zone *mz; | |
556 | u64 total = 0; | |
557 | ||
558 | for_each_online_node(nid) | |
559 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
560 | mz = mem_cgroup_zoneinfo(mem, nid, zid); | |
561 | total += MEM_CGROUP_ZSTAT(mz, idx); | |
562 | } | |
563 | return total; | |
d52aa412 KH |
564 | } |
565 | ||
d5b69e38 | 566 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
8cdea7c0 BS |
567 | { |
568 | return container_of(cgroup_subsys_state(cont, | |
569 | mem_cgroup_subsys_id), struct mem_cgroup, | |
570 | css); | |
571 | } | |
572 | ||
cf475ad2 | 573 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
78fb7466 | 574 | { |
31a78f23 BS |
575 | /* |
576 | * mm_update_next_owner() may clear mm->owner to NULL | |
577 | * if it races with swapoff, page migration, etc. | |
578 | * So this can be called with p == NULL. | |
579 | */ | |
580 | if (unlikely(!p)) | |
581 | return NULL; | |
582 | ||
78fb7466 PE |
583 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
584 | struct mem_cgroup, css); | |
585 | } | |
586 | ||
54595fe2 KH |
587 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
588 | { | |
589 | struct mem_cgroup *mem = NULL; | |
0b7f569e KH |
590 | |
591 | if (!mm) | |
592 | return NULL; | |
54595fe2 KH |
593 | /* |
594 | * Because we have no locks, mm->owner's may be being moved to other | |
595 | * cgroup. We use css_tryget() here even if this looks | |
596 | * pessimistic (rather than adding locks here). | |
597 | */ | |
598 | rcu_read_lock(); | |
599 | do { | |
600 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
601 | if (unlikely(!mem)) | |
602 | break; | |
603 | } while (!css_tryget(&mem->css)); | |
604 | rcu_read_unlock(); | |
605 | return mem; | |
606 | } | |
607 | ||
14067bb3 KH |
608 | /* |
609 | * Call callback function against all cgroup under hierarchy tree. | |
610 | */ | |
611 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, | |
612 | int (*func)(struct mem_cgroup *, void *)) | |
613 | { | |
614 | int found, ret, nextid; | |
615 | struct cgroup_subsys_state *css; | |
616 | struct mem_cgroup *mem; | |
617 | ||
618 | if (!root->use_hierarchy) | |
619 | return (*func)(root, data); | |
620 | ||
621 | nextid = 1; | |
622 | do { | |
623 | ret = 0; | |
624 | mem = NULL; | |
625 | ||
626 | rcu_read_lock(); | |
627 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, | |
628 | &found); | |
629 | if (css && css_tryget(css)) | |
630 | mem = container_of(css, struct mem_cgroup, css); | |
631 | rcu_read_unlock(); | |
632 | ||
633 | if (mem) { | |
634 | ret = (*func)(mem, data); | |
635 | css_put(&mem->css); | |
636 | } | |
637 | nextid = found + 1; | |
638 | } while (!ret && css); | |
639 | ||
640 | return ret; | |
641 | } | |
642 | ||
4b3bde4c BS |
643 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
644 | { | |
645 | return (mem == root_mem_cgroup); | |
646 | } | |
647 | ||
08e552c6 KH |
648 | /* |
649 | * Following LRU functions are allowed to be used without PCG_LOCK. | |
650 | * Operations are called by routine of global LRU independently from memcg. | |
651 | * What we have to take care of here is validness of pc->mem_cgroup. | |
652 | * | |
653 | * Changes to pc->mem_cgroup happens when | |
654 | * 1. charge | |
655 | * 2. moving account | |
656 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. | |
657 | * It is added to LRU before charge. | |
658 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. | |
659 | * When moving account, the page is not on LRU. It's isolated. | |
660 | */ | |
4f98a2fe | 661 | |
08e552c6 KH |
662 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
663 | { | |
664 | struct page_cgroup *pc; | |
08e552c6 | 665 | struct mem_cgroup_per_zone *mz; |
6d12e2d8 | 666 | |
f8d66542 | 667 | if (mem_cgroup_disabled()) |
08e552c6 KH |
668 | return; |
669 | pc = lookup_page_cgroup(page); | |
670 | /* can happen while we handle swapcache. */ | |
4b3bde4c | 671 | if (!TestClearPageCgroupAcctLRU(pc)) |
08e552c6 | 672 | return; |
4b3bde4c | 673 | VM_BUG_ON(!pc->mem_cgroup); |
544122e5 KH |
674 | /* |
675 | * We don't check PCG_USED bit. It's cleared when the "page" is finally | |
676 | * removed from global LRU. | |
677 | */ | |
08e552c6 | 678 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 679 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
4b3bde4c BS |
680 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
681 | return; | |
682 | VM_BUG_ON(list_empty(&pc->lru)); | |
08e552c6 KH |
683 | list_del_init(&pc->lru); |
684 | return; | |
6d12e2d8 KH |
685 | } |
686 | ||
08e552c6 | 687 | void mem_cgroup_del_lru(struct page *page) |
6d12e2d8 | 688 | { |
08e552c6 KH |
689 | mem_cgroup_del_lru_list(page, page_lru(page)); |
690 | } | |
b69408e8 | 691 | |
08e552c6 KH |
692 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
693 | { | |
694 | struct mem_cgroup_per_zone *mz; | |
695 | struct page_cgroup *pc; | |
b69408e8 | 696 | |
f8d66542 | 697 | if (mem_cgroup_disabled()) |
08e552c6 | 698 | return; |
6d12e2d8 | 699 | |
08e552c6 | 700 | pc = lookup_page_cgroup(page); |
bd112db8 DN |
701 | /* |
702 | * Used bit is set without atomic ops but after smp_wmb(). | |
703 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
704 | */ | |
08e552c6 | 705 | smp_rmb(); |
4b3bde4c BS |
706 | /* unused or root page is not rotated. */ |
707 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) | |
08e552c6 KH |
708 | return; |
709 | mz = page_cgroup_zoneinfo(pc); | |
710 | list_move(&pc->lru, &mz->lists[lru]); | |
6d12e2d8 KH |
711 | } |
712 | ||
08e552c6 | 713 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
66e1707b | 714 | { |
08e552c6 KH |
715 | struct page_cgroup *pc; |
716 | struct mem_cgroup_per_zone *mz; | |
6d12e2d8 | 717 | |
f8d66542 | 718 | if (mem_cgroup_disabled()) |
08e552c6 KH |
719 | return; |
720 | pc = lookup_page_cgroup(page); | |
4b3bde4c | 721 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
bd112db8 DN |
722 | /* |
723 | * Used bit is set without atomic ops but after smp_wmb(). | |
724 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
725 | */ | |
08e552c6 KH |
726 | smp_rmb(); |
727 | if (!PageCgroupUsed(pc)) | |
894bc310 | 728 | return; |
b69408e8 | 729 | |
08e552c6 | 730 | mz = page_cgroup_zoneinfo(pc); |
b69408e8 | 731 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
4b3bde4c BS |
732 | SetPageCgroupAcctLRU(pc); |
733 | if (mem_cgroup_is_root(pc->mem_cgroup)) | |
734 | return; | |
08e552c6 KH |
735 | list_add(&pc->lru, &mz->lists[lru]); |
736 | } | |
544122e5 | 737 | |
08e552c6 | 738 | /* |
544122e5 KH |
739 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
740 | * lru because the page may.be reused after it's fully uncharged (because of | |
741 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge | |
742 | * it again. This function is only used to charge SwapCache. It's done under | |
743 | * lock_page and expected that zone->lru_lock is never held. | |
08e552c6 | 744 | */ |
544122e5 | 745 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
08e552c6 | 746 | { |
544122e5 KH |
747 | unsigned long flags; |
748 | struct zone *zone = page_zone(page); | |
749 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
750 | ||
751 | spin_lock_irqsave(&zone->lru_lock, flags); | |
752 | /* | |
753 | * Forget old LRU when this page_cgroup is *not* used. This Used bit | |
754 | * is guarded by lock_page() because the page is SwapCache. | |
755 | */ | |
756 | if (!PageCgroupUsed(pc)) | |
757 | mem_cgroup_del_lru_list(page, page_lru(page)); | |
758 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
08e552c6 KH |
759 | } |
760 | ||
544122e5 KH |
761 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
762 | { | |
763 | unsigned long flags; | |
764 | struct zone *zone = page_zone(page); | |
765 | struct page_cgroup *pc = lookup_page_cgroup(page); | |
766 | ||
767 | spin_lock_irqsave(&zone->lru_lock, flags); | |
768 | /* link when the page is linked to LRU but page_cgroup isn't */ | |
4b3bde4c | 769 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
544122e5 KH |
770 | mem_cgroup_add_lru_list(page, page_lru(page)); |
771 | spin_unlock_irqrestore(&zone->lru_lock, flags); | |
772 | } | |
773 | ||
774 | ||
08e552c6 KH |
775 | void mem_cgroup_move_lists(struct page *page, |
776 | enum lru_list from, enum lru_list to) | |
777 | { | |
f8d66542 | 778 | if (mem_cgroup_disabled()) |
08e552c6 KH |
779 | return; |
780 | mem_cgroup_del_lru_list(page, from); | |
781 | mem_cgroup_add_lru_list(page, to); | |
66e1707b BS |
782 | } |
783 | ||
4c4a2214 DR |
784 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
785 | { | |
786 | int ret; | |
0b7f569e | 787 | struct mem_cgroup *curr = NULL; |
4c4a2214 DR |
788 | |
789 | task_lock(task); | |
0b7f569e KH |
790 | rcu_read_lock(); |
791 | curr = try_get_mem_cgroup_from_mm(task->mm); | |
792 | rcu_read_unlock(); | |
4c4a2214 | 793 | task_unlock(task); |
0b7f569e KH |
794 | if (!curr) |
795 | return 0; | |
d31f56db DN |
796 | /* |
797 | * We should check use_hierarchy of "mem" not "curr". Because checking | |
798 | * use_hierarchy of "curr" here make this function true if hierarchy is | |
799 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* | |
800 | * hierarchy(even if use_hierarchy is disabled in "mem"). | |
801 | */ | |
802 | if (mem->use_hierarchy) | |
0b7f569e KH |
803 | ret = css_is_ancestor(&curr->css, &mem->css); |
804 | else | |
805 | ret = (curr == mem); | |
806 | css_put(&curr->css); | |
4c4a2214 DR |
807 | return ret; |
808 | } | |
809 | ||
6c48a1d0 KH |
810 | /* |
811 | * prev_priority control...this will be used in memory reclaim path. | |
812 | */ | |
813 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) | |
814 | { | |
2733c06a KM |
815 | int prev_priority; |
816 | ||
817 | spin_lock(&mem->reclaim_param_lock); | |
818 | prev_priority = mem->prev_priority; | |
819 | spin_unlock(&mem->reclaim_param_lock); | |
820 | ||
821 | return prev_priority; | |
6c48a1d0 KH |
822 | } |
823 | ||
824 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) | |
825 | { | |
2733c06a | 826 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
827 | if (priority < mem->prev_priority) |
828 | mem->prev_priority = priority; | |
2733c06a | 829 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
830 | } |
831 | ||
832 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) | |
833 | { | |
2733c06a | 834 | spin_lock(&mem->reclaim_param_lock); |
6c48a1d0 | 835 | mem->prev_priority = priority; |
2733c06a | 836 | spin_unlock(&mem->reclaim_param_lock); |
6c48a1d0 KH |
837 | } |
838 | ||
c772be93 | 839 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
14797e23 KM |
840 | { |
841 | unsigned long active; | |
842 | unsigned long inactive; | |
c772be93 KM |
843 | unsigned long gb; |
844 | unsigned long inactive_ratio; | |
14797e23 | 845 | |
14067bb3 KH |
846 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
847 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); | |
14797e23 | 848 | |
c772be93 KM |
849 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
850 | if (gb) | |
851 | inactive_ratio = int_sqrt(10 * gb); | |
852 | else | |
853 | inactive_ratio = 1; | |
854 | ||
855 | if (present_pages) { | |
856 | present_pages[0] = inactive; | |
857 | present_pages[1] = active; | |
858 | } | |
859 | ||
860 | return inactive_ratio; | |
861 | } | |
862 | ||
863 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) | |
864 | { | |
865 | unsigned long active; | |
866 | unsigned long inactive; | |
867 | unsigned long present_pages[2]; | |
868 | unsigned long inactive_ratio; | |
869 | ||
870 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); | |
871 | ||
872 | inactive = present_pages[0]; | |
873 | active = present_pages[1]; | |
874 | ||
875 | if (inactive * inactive_ratio < active) | |
14797e23 KM |
876 | return 1; |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
56e49d21 RR |
881 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
882 | { | |
883 | unsigned long active; | |
884 | unsigned long inactive; | |
885 | ||
886 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); | |
887 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); | |
888 | ||
889 | return (active > inactive); | |
890 | } | |
891 | ||
a3d8e054 KM |
892 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
893 | struct zone *zone, | |
894 | enum lru_list lru) | |
895 | { | |
896 | int nid = zone->zone_pgdat->node_id; | |
897 | int zid = zone_idx(zone); | |
898 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
899 | ||
900 | return MEM_CGROUP_ZSTAT(mz, lru); | |
901 | } | |
902 | ||
3e2f41f1 KM |
903 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
904 | struct zone *zone) | |
905 | { | |
906 | int nid = zone->zone_pgdat->node_id; | |
907 | int zid = zone_idx(zone); | |
908 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); | |
909 | ||
910 | return &mz->reclaim_stat; | |
911 | } | |
912 | ||
913 | struct zone_reclaim_stat * | |
914 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) | |
915 | { | |
916 | struct page_cgroup *pc; | |
917 | struct mem_cgroup_per_zone *mz; | |
918 | ||
919 | if (mem_cgroup_disabled()) | |
920 | return NULL; | |
921 | ||
922 | pc = lookup_page_cgroup(page); | |
bd112db8 DN |
923 | /* |
924 | * Used bit is set without atomic ops but after smp_wmb(). | |
925 | * For making pc->mem_cgroup visible, insert smp_rmb() here. | |
926 | */ | |
927 | smp_rmb(); | |
928 | if (!PageCgroupUsed(pc)) | |
929 | return NULL; | |
930 | ||
3e2f41f1 KM |
931 | mz = page_cgroup_zoneinfo(pc); |
932 | if (!mz) | |
933 | return NULL; | |
934 | ||
935 | return &mz->reclaim_stat; | |
936 | } | |
937 | ||
66e1707b BS |
938 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
939 | struct list_head *dst, | |
940 | unsigned long *scanned, int order, | |
941 | int mode, struct zone *z, | |
942 | struct mem_cgroup *mem_cont, | |
4f98a2fe | 943 | int active, int file) |
66e1707b BS |
944 | { |
945 | unsigned long nr_taken = 0; | |
946 | struct page *page; | |
947 | unsigned long scan; | |
948 | LIST_HEAD(pc_list); | |
949 | struct list_head *src; | |
ff7283fa | 950 | struct page_cgroup *pc, *tmp; |
1ecaab2b KH |
951 | int nid = z->zone_pgdat->node_id; |
952 | int zid = zone_idx(z); | |
953 | struct mem_cgroup_per_zone *mz; | |
b7c46d15 | 954 | int lru = LRU_FILE * file + active; |
2ffebca6 | 955 | int ret; |
66e1707b | 956 | |
cf475ad2 | 957 | BUG_ON(!mem_cont); |
1ecaab2b | 958 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
b69408e8 | 959 | src = &mz->lists[lru]; |
66e1707b | 960 | |
ff7283fa KH |
961 | scan = 0; |
962 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { | |
436c6541 | 963 | if (scan >= nr_to_scan) |
ff7283fa | 964 | break; |
08e552c6 KH |
965 | |
966 | page = pc->page; | |
52d4b9ac KH |
967 | if (unlikely(!PageCgroupUsed(pc))) |
968 | continue; | |
436c6541 | 969 | if (unlikely(!PageLRU(page))) |
ff7283fa | 970 | continue; |
ff7283fa | 971 | |
436c6541 | 972 | scan++; |
2ffebca6 KH |
973 | ret = __isolate_lru_page(page, mode, file); |
974 | switch (ret) { | |
975 | case 0: | |
66e1707b | 976 | list_move(&page->lru, dst); |
2ffebca6 | 977 | mem_cgroup_del_lru(page); |
66e1707b | 978 | nr_taken++; |
2ffebca6 KH |
979 | break; |
980 | case -EBUSY: | |
981 | /* we don't affect global LRU but rotate in our LRU */ | |
982 | mem_cgroup_rotate_lru_list(page, page_lru(page)); | |
983 | break; | |
984 | default: | |
985 | break; | |
66e1707b BS |
986 | } |
987 | } | |
988 | ||
66e1707b BS |
989 | *scanned = scan; |
990 | return nr_taken; | |
991 | } | |
992 | ||
6d61ef40 BS |
993 | #define mem_cgroup_from_res_counter(counter, member) \ |
994 | container_of(counter, struct mem_cgroup, member) | |
995 | ||
b85a96c0 DN |
996 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
997 | { | |
998 | if (do_swap_account) { | |
999 | if (res_counter_check_under_limit(&mem->res) && | |
1000 | res_counter_check_under_limit(&mem->memsw)) | |
1001 | return true; | |
1002 | } else | |
1003 | if (res_counter_check_under_limit(&mem->res)) | |
1004 | return true; | |
1005 | return false; | |
1006 | } | |
1007 | ||
a7885eb8 KM |
1008 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
1009 | { | |
1010 | struct cgroup *cgrp = memcg->css.cgroup; | |
1011 | unsigned int swappiness; | |
1012 | ||
1013 | /* root ? */ | |
1014 | if (cgrp->parent == NULL) | |
1015 | return vm_swappiness; | |
1016 | ||
1017 | spin_lock(&memcg->reclaim_param_lock); | |
1018 | swappiness = memcg->swappiness; | |
1019 | spin_unlock(&memcg->reclaim_param_lock); | |
1020 | ||
1021 | return swappiness; | |
1022 | } | |
1023 | ||
81d39c20 KH |
1024 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
1025 | { | |
1026 | int *val = data; | |
1027 | (*val)++; | |
1028 | return 0; | |
1029 | } | |
e222432b BS |
1030 | |
1031 | /** | |
6a6135b6 | 1032 | * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode. |
e222432b BS |
1033 | * @memcg: The memory cgroup that went over limit |
1034 | * @p: Task that is going to be killed | |
1035 | * | |
1036 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is | |
1037 | * enabled | |
1038 | */ | |
1039 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) | |
1040 | { | |
1041 | struct cgroup *task_cgrp; | |
1042 | struct cgroup *mem_cgrp; | |
1043 | /* | |
1044 | * Need a buffer in BSS, can't rely on allocations. The code relies | |
1045 | * on the assumption that OOM is serialized for memory controller. | |
1046 | * If this assumption is broken, revisit this code. | |
1047 | */ | |
1048 | static char memcg_name[PATH_MAX]; | |
1049 | int ret; | |
1050 | ||
d31f56db | 1051 | if (!memcg || !p) |
e222432b BS |
1052 | return; |
1053 | ||
1054 | ||
1055 | rcu_read_lock(); | |
1056 | ||
1057 | mem_cgrp = memcg->css.cgroup; | |
1058 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); | |
1059 | ||
1060 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); | |
1061 | if (ret < 0) { | |
1062 | /* | |
1063 | * Unfortunately, we are unable to convert to a useful name | |
1064 | * But we'll still print out the usage information | |
1065 | */ | |
1066 | rcu_read_unlock(); | |
1067 | goto done; | |
1068 | } | |
1069 | rcu_read_unlock(); | |
1070 | ||
1071 | printk(KERN_INFO "Task in %s killed", memcg_name); | |
1072 | ||
1073 | rcu_read_lock(); | |
1074 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); | |
1075 | if (ret < 0) { | |
1076 | rcu_read_unlock(); | |
1077 | goto done; | |
1078 | } | |
1079 | rcu_read_unlock(); | |
1080 | ||
1081 | /* | |
1082 | * Continues from above, so we don't need an KERN_ level | |
1083 | */ | |
1084 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); | |
1085 | done: | |
1086 | ||
1087 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", | |
1088 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, | |
1089 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, | |
1090 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); | |
1091 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " | |
1092 | "failcnt %llu\n", | |
1093 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, | |
1094 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, | |
1095 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); | |
1096 | } | |
1097 | ||
81d39c20 KH |
1098 | /* |
1099 | * This function returns the number of memcg under hierarchy tree. Returns | |
1100 | * 1(self count) if no children. | |
1101 | */ | |
1102 | static int mem_cgroup_count_children(struct mem_cgroup *mem) | |
1103 | { | |
1104 | int num = 0; | |
1105 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); | |
1106 | return num; | |
1107 | } | |
1108 | ||
6d61ef40 | 1109 | /* |
04046e1a KH |
1110 | * Visit the first child (need not be the first child as per the ordering |
1111 | * of the cgroup list, since we track last_scanned_child) of @mem and use | |
1112 | * that to reclaim free pages from. | |
1113 | */ | |
1114 | static struct mem_cgroup * | |
1115 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) | |
1116 | { | |
1117 | struct mem_cgroup *ret = NULL; | |
1118 | struct cgroup_subsys_state *css; | |
1119 | int nextid, found; | |
1120 | ||
1121 | if (!root_mem->use_hierarchy) { | |
1122 | css_get(&root_mem->css); | |
1123 | ret = root_mem; | |
1124 | } | |
1125 | ||
1126 | while (!ret) { | |
1127 | rcu_read_lock(); | |
1128 | nextid = root_mem->last_scanned_child + 1; | |
1129 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, | |
1130 | &found); | |
1131 | if (css && css_tryget(css)) | |
1132 | ret = container_of(css, struct mem_cgroup, css); | |
1133 | ||
1134 | rcu_read_unlock(); | |
1135 | /* Updates scanning parameter */ | |
1136 | spin_lock(&root_mem->reclaim_param_lock); | |
1137 | if (!css) { | |
1138 | /* this means start scan from ID:1 */ | |
1139 | root_mem->last_scanned_child = 0; | |
1140 | } else | |
1141 | root_mem->last_scanned_child = found; | |
1142 | spin_unlock(&root_mem->reclaim_param_lock); | |
1143 | } | |
1144 | ||
1145 | return ret; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * Scan the hierarchy if needed to reclaim memory. We remember the last child | |
1150 | * we reclaimed from, so that we don't end up penalizing one child extensively | |
1151 | * based on its position in the children list. | |
6d61ef40 BS |
1152 | * |
1153 | * root_mem is the original ancestor that we've been reclaim from. | |
04046e1a KH |
1154 | * |
1155 | * We give up and return to the caller when we visit root_mem twice. | |
1156 | * (other groups can be removed while we're walking....) | |
81d39c20 KH |
1157 | * |
1158 | * If shrink==true, for avoiding to free too much, this returns immedieately. | |
6d61ef40 BS |
1159 | */ |
1160 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, | |
4e416953 | 1161 | struct zone *zone, |
75822b44 BS |
1162 | gfp_t gfp_mask, |
1163 | unsigned long reclaim_options) | |
6d61ef40 | 1164 | { |
04046e1a KH |
1165 | struct mem_cgroup *victim; |
1166 | int ret, total = 0; | |
1167 | int loop = 0; | |
75822b44 BS |
1168 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1169 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; | |
4e416953 BS |
1170 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
1171 | unsigned long excess = mem_cgroup_get_excess(root_mem); | |
04046e1a | 1172 | |
22a668d7 KH |
1173 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1174 | if (root_mem->memsw_is_minimum) | |
1175 | noswap = true; | |
1176 | ||
4e416953 | 1177 | while (1) { |
04046e1a | 1178 | victim = mem_cgroup_select_victim(root_mem); |
4e416953 | 1179 | if (victim == root_mem) { |
04046e1a | 1180 | loop++; |
cdec2e42 KH |
1181 | if (loop >= 1) |
1182 | drain_all_stock_async(); | |
4e416953 BS |
1183 | if (loop >= 2) { |
1184 | /* | |
1185 | * If we have not been able to reclaim | |
1186 | * anything, it might because there are | |
1187 | * no reclaimable pages under this hierarchy | |
1188 | */ | |
1189 | if (!check_soft || !total) { | |
1190 | css_put(&victim->css); | |
1191 | break; | |
1192 | } | |
1193 | /* | |
1194 | * We want to do more targetted reclaim. | |
1195 | * excess >> 2 is not to excessive so as to | |
1196 | * reclaim too much, nor too less that we keep | |
1197 | * coming back to reclaim from this cgroup | |
1198 | */ | |
1199 | if (total >= (excess >> 2) || | |
1200 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { | |
1201 | css_put(&victim->css); | |
1202 | break; | |
1203 | } | |
1204 | } | |
1205 | } | |
c62b1a3b | 1206 | if (!mem_cgroup_local_usage(victim)) { |
04046e1a KH |
1207 | /* this cgroup's local usage == 0 */ |
1208 | css_put(&victim->css); | |
6d61ef40 BS |
1209 | continue; |
1210 | } | |
04046e1a | 1211 | /* we use swappiness of local cgroup */ |
4e416953 BS |
1212 | if (check_soft) |
1213 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, | |
1214 | noswap, get_swappiness(victim), zone, | |
1215 | zone->zone_pgdat->node_id); | |
1216 | else | |
1217 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, | |
1218 | noswap, get_swappiness(victim)); | |
04046e1a | 1219 | css_put(&victim->css); |
81d39c20 KH |
1220 | /* |
1221 | * At shrinking usage, we can't check we should stop here or | |
1222 | * reclaim more. It's depends on callers. last_scanned_child | |
1223 | * will work enough for keeping fairness under tree. | |
1224 | */ | |
1225 | if (shrink) | |
1226 | return ret; | |
04046e1a | 1227 | total += ret; |
4e416953 BS |
1228 | if (check_soft) { |
1229 | if (res_counter_check_under_soft_limit(&root_mem->res)) | |
1230 | return total; | |
1231 | } else if (mem_cgroup_check_under_limit(root_mem)) | |
04046e1a | 1232 | return 1 + total; |
6d61ef40 | 1233 | } |
04046e1a | 1234 | return total; |
6d61ef40 BS |
1235 | } |
1236 | ||
a636b327 KH |
1237 | bool mem_cgroup_oom_called(struct task_struct *task) |
1238 | { | |
1239 | bool ret = false; | |
1240 | struct mem_cgroup *mem; | |
1241 | struct mm_struct *mm; | |
1242 | ||
1243 | rcu_read_lock(); | |
1244 | mm = task->mm; | |
1245 | if (!mm) | |
1246 | mm = &init_mm; | |
1247 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); | |
1248 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) | |
1249 | ret = true; | |
1250 | rcu_read_unlock(); | |
1251 | return ret; | |
1252 | } | |
0b7f569e KH |
1253 | |
1254 | static int record_last_oom_cb(struct mem_cgroup *mem, void *data) | |
1255 | { | |
1256 | mem->last_oom_jiffies = jiffies; | |
1257 | return 0; | |
1258 | } | |
1259 | ||
1260 | static void record_last_oom(struct mem_cgroup *mem) | |
1261 | { | |
1262 | mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb); | |
1263 | } | |
1264 | ||
d69b042f BS |
1265 | /* |
1266 | * Currently used to update mapped file statistics, but the routine can be | |
1267 | * generalized to update other statistics as well. | |
1268 | */ | |
d8046582 | 1269 | void mem_cgroup_update_file_mapped(struct page *page, int val) |
d69b042f BS |
1270 | { |
1271 | struct mem_cgroup *mem; | |
d69b042f BS |
1272 | struct page_cgroup *pc; |
1273 | ||
d69b042f BS |
1274 | pc = lookup_page_cgroup(page); |
1275 | if (unlikely(!pc)) | |
1276 | return; | |
1277 | ||
1278 | lock_page_cgroup(pc); | |
1279 | mem = pc->mem_cgroup; | |
1280 | if (!mem) | |
1281 | goto done; | |
1282 | ||
1283 | if (!PageCgroupUsed(pc)) | |
1284 | goto done; | |
1285 | ||
1286 | /* | |
c62b1a3b | 1287 | * Preemption is already disabled. We can use __this_cpu_xxx |
d69b042f | 1288 | */ |
c62b1a3b | 1289 | __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], val); |
d69b042f | 1290 | |
d69b042f BS |
1291 | done: |
1292 | unlock_page_cgroup(pc); | |
1293 | } | |
0b7f569e | 1294 | |
cdec2e42 KH |
1295 | /* |
1296 | * size of first charge trial. "32" comes from vmscan.c's magic value. | |
1297 | * TODO: maybe necessary to use big numbers in big irons. | |
1298 | */ | |
1299 | #define CHARGE_SIZE (32 * PAGE_SIZE) | |
1300 | struct memcg_stock_pcp { | |
1301 | struct mem_cgroup *cached; /* this never be root cgroup */ | |
1302 | int charge; | |
1303 | struct work_struct work; | |
1304 | }; | |
1305 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); | |
1306 | static atomic_t memcg_drain_count; | |
1307 | ||
1308 | /* | |
1309 | * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed | |
1310 | * from local stock and true is returned. If the stock is 0 or charges from a | |
1311 | * cgroup which is not current target, returns false. This stock will be | |
1312 | * refilled. | |
1313 | */ | |
1314 | static bool consume_stock(struct mem_cgroup *mem) | |
1315 | { | |
1316 | struct memcg_stock_pcp *stock; | |
1317 | bool ret = true; | |
1318 | ||
1319 | stock = &get_cpu_var(memcg_stock); | |
1320 | if (mem == stock->cached && stock->charge) | |
1321 | stock->charge -= PAGE_SIZE; | |
1322 | else /* need to call res_counter_charge */ | |
1323 | ret = false; | |
1324 | put_cpu_var(memcg_stock); | |
1325 | return ret; | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * Returns stocks cached in percpu to res_counter and reset cached information. | |
1330 | */ | |
1331 | static void drain_stock(struct memcg_stock_pcp *stock) | |
1332 | { | |
1333 | struct mem_cgroup *old = stock->cached; | |
1334 | ||
1335 | if (stock->charge) { | |
1336 | res_counter_uncharge(&old->res, stock->charge); | |
1337 | if (do_swap_account) | |
1338 | res_counter_uncharge(&old->memsw, stock->charge); | |
1339 | } | |
1340 | stock->cached = NULL; | |
1341 | stock->charge = 0; | |
1342 | } | |
1343 | ||
1344 | /* | |
1345 | * This must be called under preempt disabled or must be called by | |
1346 | * a thread which is pinned to local cpu. | |
1347 | */ | |
1348 | static void drain_local_stock(struct work_struct *dummy) | |
1349 | { | |
1350 | struct memcg_stock_pcp *stock = &__get_cpu_var(memcg_stock); | |
1351 | drain_stock(stock); | |
1352 | } | |
1353 | ||
1354 | /* | |
1355 | * Cache charges(val) which is from res_counter, to local per_cpu area. | |
1356 | * This will be consumed by consumt_stock() function, later. | |
1357 | */ | |
1358 | static void refill_stock(struct mem_cgroup *mem, int val) | |
1359 | { | |
1360 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); | |
1361 | ||
1362 | if (stock->cached != mem) { /* reset if necessary */ | |
1363 | drain_stock(stock); | |
1364 | stock->cached = mem; | |
1365 | } | |
1366 | stock->charge += val; | |
1367 | put_cpu_var(memcg_stock); | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Tries to drain stocked charges in other cpus. This function is asynchronous | |
1372 | * and just put a work per cpu for draining localy on each cpu. Caller can | |
1373 | * expects some charges will be back to res_counter later but cannot wait for | |
1374 | * it. | |
1375 | */ | |
1376 | static void drain_all_stock_async(void) | |
1377 | { | |
1378 | int cpu; | |
1379 | /* This function is for scheduling "drain" in asynchronous way. | |
1380 | * The result of "drain" is not directly handled by callers. Then, | |
1381 | * if someone is calling drain, we don't have to call drain more. | |
1382 | * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if | |
1383 | * there is a race. We just do loose check here. | |
1384 | */ | |
1385 | if (atomic_read(&memcg_drain_count)) | |
1386 | return; | |
1387 | /* Notify other cpus that system-wide "drain" is running */ | |
1388 | atomic_inc(&memcg_drain_count); | |
1389 | get_online_cpus(); | |
1390 | for_each_online_cpu(cpu) { | |
1391 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); | |
1392 | schedule_work_on(cpu, &stock->work); | |
1393 | } | |
1394 | put_online_cpus(); | |
1395 | atomic_dec(&memcg_drain_count); | |
1396 | /* We don't wait for flush_work */ | |
1397 | } | |
1398 | ||
1399 | /* This is a synchronous drain interface. */ | |
1400 | static void drain_all_stock_sync(void) | |
1401 | { | |
1402 | /* called when force_empty is called */ | |
1403 | atomic_inc(&memcg_drain_count); | |
1404 | schedule_on_each_cpu(drain_local_stock); | |
1405 | atomic_dec(&memcg_drain_count); | |
1406 | } | |
1407 | ||
1408 | static int __cpuinit memcg_stock_cpu_callback(struct notifier_block *nb, | |
1409 | unsigned long action, | |
1410 | void *hcpu) | |
1411 | { | |
1412 | int cpu = (unsigned long)hcpu; | |
1413 | struct memcg_stock_pcp *stock; | |
1414 | ||
1415 | if (action != CPU_DEAD) | |
1416 | return NOTIFY_OK; | |
1417 | stock = &per_cpu(memcg_stock, cpu); | |
1418 | drain_stock(stock); | |
1419 | return NOTIFY_OK; | |
1420 | } | |
1421 | ||
f817ed48 KH |
1422 | /* |
1423 | * Unlike exported interface, "oom" parameter is added. if oom==true, | |
1424 | * oom-killer can be invoked. | |
8a9f3ccd | 1425 | */ |
f817ed48 | 1426 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
430e4863 | 1427 | gfp_t gfp_mask, struct mem_cgroup **memcg, bool oom) |
8a9f3ccd | 1428 | { |
4e649152 | 1429 | struct mem_cgroup *mem, *mem_over_limit; |
7a81b88c | 1430 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
4e649152 | 1431 | struct res_counter *fail_res; |
cdec2e42 | 1432 | int csize = CHARGE_SIZE; |
a636b327 KH |
1433 | |
1434 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { | |
1435 | /* Don't account this! */ | |
1436 | *memcg = NULL; | |
1437 | return 0; | |
1438 | } | |
1439 | ||
8a9f3ccd | 1440 | /* |
3be91277 HD |
1441 | * We always charge the cgroup the mm_struct belongs to. |
1442 | * The mm_struct's mem_cgroup changes on task migration if the | |
8a9f3ccd BS |
1443 | * thread group leader migrates. It's possible that mm is not |
1444 | * set, if so charge the init_mm (happens for pagecache usage). | |
1445 | */ | |
54595fe2 KH |
1446 | mem = *memcg; |
1447 | if (likely(!mem)) { | |
1448 | mem = try_get_mem_cgroup_from_mm(mm); | |
7a81b88c | 1449 | *memcg = mem; |
e8589cc1 | 1450 | } else { |
7a81b88c | 1451 | css_get(&mem->css); |
e8589cc1 | 1452 | } |
54595fe2 KH |
1453 | if (unlikely(!mem)) |
1454 | return 0; | |
1455 | ||
46f7e602 | 1456 | VM_BUG_ON(css_is_removed(&mem->css)); |
cdec2e42 KH |
1457 | if (mem_cgroup_is_root(mem)) |
1458 | goto done; | |
8a9f3ccd | 1459 | |
8c7c6e34 | 1460 | while (1) { |
0c3e73e8 | 1461 | int ret = 0; |
75822b44 | 1462 | unsigned long flags = 0; |
7a81b88c | 1463 | |
cdec2e42 | 1464 | if (consume_stock(mem)) |
430e4863 | 1465 | goto done; |
cdec2e42 KH |
1466 | |
1467 | ret = res_counter_charge(&mem->res, csize, &fail_res); | |
8c7c6e34 KH |
1468 | if (likely(!ret)) { |
1469 | if (!do_swap_account) | |
1470 | break; | |
cdec2e42 | 1471 | ret = res_counter_charge(&mem->memsw, csize, &fail_res); |
8c7c6e34 KH |
1472 | if (likely(!ret)) |
1473 | break; | |
1474 | /* mem+swap counter fails */ | |
cdec2e42 | 1475 | res_counter_uncharge(&mem->res, csize); |
75822b44 | 1476 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
6d61ef40 BS |
1477 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1478 | memsw); | |
1479 | } else | |
1480 | /* mem counter fails */ | |
1481 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, | |
1482 | res); | |
1483 | ||
cdec2e42 KH |
1484 | /* reduce request size and retry */ |
1485 | if (csize > PAGE_SIZE) { | |
1486 | csize = PAGE_SIZE; | |
1487 | continue; | |
1488 | } | |
3be91277 | 1489 | if (!(gfp_mask & __GFP_WAIT)) |
7a81b88c | 1490 | goto nomem; |
e1a1cd59 | 1491 | |
4e416953 BS |
1492 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
1493 | gfp_mask, flags); | |
4d1c6273 DN |
1494 | if (ret) |
1495 | continue; | |
66e1707b BS |
1496 | |
1497 | /* | |
8869b8f6 HD |
1498 | * try_to_free_mem_cgroup_pages() might not give us a full |
1499 | * picture of reclaim. Some pages are reclaimed and might be | |
1500 | * moved to swap cache or just unmapped from the cgroup. | |
1501 | * Check the limit again to see if the reclaim reduced the | |
1502 | * current usage of the cgroup before giving up | |
8c7c6e34 | 1503 | * |
8869b8f6 | 1504 | */ |
b85a96c0 DN |
1505 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
1506 | continue; | |
3be91277 | 1507 | |
8033b97c DN |
1508 | /* try to avoid oom while someone is moving charge */ |
1509 | if (mc.moving_task && current != mc.moving_task) { | |
1510 | struct mem_cgroup *from, *to; | |
1511 | bool do_continue = false; | |
1512 | /* | |
1513 | * There is a small race that "from" or "to" can be | |
1514 | * freed by rmdir, so we use css_tryget(). | |
1515 | */ | |
1516 | rcu_read_lock(); | |
1517 | from = mc.from; | |
1518 | to = mc.to; | |
1519 | if (from && css_tryget(&from->css)) { | |
1520 | if (mem_over_limit->use_hierarchy) | |
1521 | do_continue = css_is_ancestor( | |
1522 | &from->css, | |
1523 | &mem_over_limit->css); | |
1524 | else | |
1525 | do_continue = (from == mem_over_limit); | |
1526 | css_put(&from->css); | |
1527 | } | |
1528 | if (!do_continue && to && css_tryget(&to->css)) { | |
1529 | if (mem_over_limit->use_hierarchy) | |
1530 | do_continue = css_is_ancestor( | |
1531 | &to->css, | |
1532 | &mem_over_limit->css); | |
1533 | else | |
1534 | do_continue = (to == mem_over_limit); | |
1535 | css_put(&to->css); | |
1536 | } | |
1537 | rcu_read_unlock(); | |
1538 | if (do_continue) { | |
1539 | DEFINE_WAIT(wait); | |
1540 | prepare_to_wait(&mc.waitq, &wait, | |
1541 | TASK_INTERRUPTIBLE); | |
1542 | /* moving charge context might have finished. */ | |
1543 | if (mc.moving_task) | |
1544 | schedule(); | |
1545 | finish_wait(&mc.waitq, &wait); | |
1546 | continue; | |
1547 | } | |
1548 | } | |
1549 | ||
3be91277 | 1550 | if (!nr_retries--) { |
a636b327 | 1551 | if (oom) { |
88700756 | 1552 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
0b7f569e | 1553 | record_last_oom(mem_over_limit); |
a636b327 | 1554 | } |
7a81b88c | 1555 | goto nomem; |
66e1707b | 1556 | } |
8a9f3ccd | 1557 | } |
cdec2e42 KH |
1558 | if (csize > PAGE_SIZE) |
1559 | refill_stock(mem, csize - PAGE_SIZE); | |
0c3e73e8 | 1560 | done: |
7a81b88c KH |
1561 | return 0; |
1562 | nomem: | |
1563 | css_put(&mem->css); | |
1564 | return -ENOMEM; | |
1565 | } | |
8a9f3ccd | 1566 | |
a3032a2c DN |
1567 | /* |
1568 | * Somemtimes we have to undo a charge we got by try_charge(). | |
1569 | * This function is for that and do uncharge, put css's refcnt. | |
1570 | * gotten by try_charge(). | |
1571 | */ | |
854ffa8d DN |
1572 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *mem, |
1573 | unsigned long count) | |
a3032a2c DN |
1574 | { |
1575 | if (!mem_cgroup_is_root(mem)) { | |
854ffa8d | 1576 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); |
a3032a2c | 1577 | if (do_swap_account) |
854ffa8d DN |
1578 | res_counter_uncharge(&mem->memsw, PAGE_SIZE * count); |
1579 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
1580 | WARN_ON_ONCE(count > INT_MAX); | |
1581 | __css_put(&mem->css, (int)count); | |
a3032a2c | 1582 | } |
854ffa8d DN |
1583 | /* we don't need css_put for root */ |
1584 | } | |
1585 | ||
1586 | static void mem_cgroup_cancel_charge(struct mem_cgroup *mem) | |
1587 | { | |
1588 | __mem_cgroup_cancel_charge(mem, 1); | |
a3032a2c DN |
1589 | } |
1590 | ||
a3b2d692 KH |
1591 | /* |
1592 | * A helper function to get mem_cgroup from ID. must be called under | |
1593 | * rcu_read_lock(). The caller must check css_is_removed() or some if | |
1594 | * it's concern. (dropping refcnt from swap can be called against removed | |
1595 | * memcg.) | |
1596 | */ | |
1597 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) | |
1598 | { | |
1599 | struct cgroup_subsys_state *css; | |
1600 | ||
1601 | /* ID 0 is unused ID */ | |
1602 | if (!id) | |
1603 | return NULL; | |
1604 | css = css_lookup(&mem_cgroup_subsys, id); | |
1605 | if (!css) | |
1606 | return NULL; | |
1607 | return container_of(css, struct mem_cgroup, css); | |
1608 | } | |
1609 | ||
e42d9d5d | 1610 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
b5a84319 | 1611 | { |
e42d9d5d | 1612 | struct mem_cgroup *mem = NULL; |
3c776e64 | 1613 | struct page_cgroup *pc; |
a3b2d692 | 1614 | unsigned short id; |
b5a84319 KH |
1615 | swp_entry_t ent; |
1616 | ||
3c776e64 DN |
1617 | VM_BUG_ON(!PageLocked(page)); |
1618 | ||
3c776e64 | 1619 | pc = lookup_page_cgroup(page); |
c0bd3f63 | 1620 | lock_page_cgroup(pc); |
a3b2d692 | 1621 | if (PageCgroupUsed(pc)) { |
3c776e64 | 1622 | mem = pc->mem_cgroup; |
a3b2d692 KH |
1623 | if (mem && !css_tryget(&mem->css)) |
1624 | mem = NULL; | |
e42d9d5d | 1625 | } else if (PageSwapCache(page)) { |
3c776e64 | 1626 | ent.val = page_private(page); |
a3b2d692 KH |
1627 | id = lookup_swap_cgroup(ent); |
1628 | rcu_read_lock(); | |
1629 | mem = mem_cgroup_lookup(id); | |
1630 | if (mem && !css_tryget(&mem->css)) | |
1631 | mem = NULL; | |
1632 | rcu_read_unlock(); | |
3c776e64 | 1633 | } |
c0bd3f63 | 1634 | unlock_page_cgroup(pc); |
b5a84319 KH |
1635 | return mem; |
1636 | } | |
1637 | ||
7a81b88c | 1638 | /* |
a5e924f5 | 1639 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
7a81b88c KH |
1640 | * USED state. If already USED, uncharge and return. |
1641 | */ | |
1642 | ||
1643 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, | |
1644 | struct page_cgroup *pc, | |
1645 | enum charge_type ctype) | |
1646 | { | |
7a81b88c KH |
1647 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
1648 | if (!mem) | |
1649 | return; | |
52d4b9ac KH |
1650 | |
1651 | lock_page_cgroup(pc); | |
1652 | if (unlikely(PageCgroupUsed(pc))) { | |
1653 | unlock_page_cgroup(pc); | |
a3032a2c | 1654 | mem_cgroup_cancel_charge(mem); |
7a81b88c | 1655 | return; |
52d4b9ac | 1656 | } |
4b3bde4c | 1657 | |
8a9f3ccd | 1658 | pc->mem_cgroup = mem; |
261fb61a KH |
1659 | /* |
1660 | * We access a page_cgroup asynchronously without lock_page_cgroup(). | |
1661 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup | |
1662 | * is accessed after testing USED bit. To make pc->mem_cgroup visible | |
1663 | * before USED bit, we need memory barrier here. | |
1664 | * See mem_cgroup_add_lru_list(), etc. | |
1665 | */ | |
08e552c6 | 1666 | smp_wmb(); |
4b3bde4c BS |
1667 | switch (ctype) { |
1668 | case MEM_CGROUP_CHARGE_TYPE_CACHE: | |
1669 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: | |
1670 | SetPageCgroupCache(pc); | |
1671 | SetPageCgroupUsed(pc); | |
1672 | break; | |
1673 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
1674 | ClearPageCgroupCache(pc); | |
1675 | SetPageCgroupUsed(pc); | |
1676 | break; | |
1677 | default: | |
1678 | break; | |
1679 | } | |
3be91277 | 1680 | |
08e552c6 | 1681 | mem_cgroup_charge_statistics(mem, pc, true); |
52d4b9ac | 1682 | |
52d4b9ac | 1683 | unlock_page_cgroup(pc); |
430e4863 KH |
1684 | /* |
1685 | * "charge_statistics" updated event counter. Then, check it. | |
1686 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. | |
1687 | * if they exceeds softlimit. | |
1688 | */ | |
1689 | if (mem_cgroup_soft_limit_check(mem)) | |
1690 | mem_cgroup_update_tree(mem, pc->page); | |
1691 | if (mem_cgroup_threshold_check(mem)) | |
1692 | mem_cgroup_threshold(mem); | |
1693 | ||
7a81b88c | 1694 | } |
66e1707b | 1695 | |
f817ed48 | 1696 | /** |
57f9fd7d | 1697 | * __mem_cgroup_move_account - move account of the page |
f817ed48 KH |
1698 | * @pc: page_cgroup of the page. |
1699 | * @from: mem_cgroup which the page is moved from. | |
1700 | * @to: mem_cgroup which the page is moved to. @from != @to. | |
854ffa8d | 1701 | * @uncharge: whether we should call uncharge and css_put against @from. |
f817ed48 KH |
1702 | * |
1703 | * The caller must confirm following. | |
08e552c6 | 1704 | * - page is not on LRU (isolate_page() is useful.) |
57f9fd7d | 1705 | * - the pc is locked, used, and ->mem_cgroup points to @from. |
f817ed48 | 1706 | * |
854ffa8d DN |
1707 | * This function doesn't do "charge" nor css_get to new cgroup. It should be |
1708 | * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is | |
1709 | * true, this function does "uncharge" from old cgroup, but it doesn't if | |
1710 | * @uncharge is false, so a caller should do "uncharge". | |
f817ed48 KH |
1711 | */ |
1712 | ||
57f9fd7d | 1713 | static void __mem_cgroup_move_account(struct page_cgroup *pc, |
854ffa8d | 1714 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
f817ed48 | 1715 | { |
d69b042f | 1716 | struct page *page; |
f817ed48 | 1717 | |
f817ed48 | 1718 | VM_BUG_ON(from == to); |
08e552c6 | 1719 | VM_BUG_ON(PageLRU(pc->page)); |
57f9fd7d DN |
1720 | VM_BUG_ON(!PageCgroupLocked(pc)); |
1721 | VM_BUG_ON(!PageCgroupUsed(pc)); | |
1722 | VM_BUG_ON(pc->mem_cgroup != from); | |
f817ed48 | 1723 | |
d69b042f | 1724 | page = pc->page; |
d8046582 | 1725 | if (page_mapped(page) && !PageAnon(page)) { |
c62b1a3b KH |
1726 | /* Update mapped_file data for mem_cgroup */ |
1727 | preempt_disable(); | |
1728 | __this_cpu_dec(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1729 | __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]); | |
1730 | preempt_enable(); | |
d69b042f | 1731 | } |
854ffa8d DN |
1732 | mem_cgroup_charge_statistics(from, pc, false); |
1733 | if (uncharge) | |
1734 | /* This is not "cancel", but cancel_charge does all we need. */ | |
1735 | mem_cgroup_cancel_charge(from); | |
d69b042f | 1736 | |
854ffa8d | 1737 | /* caller should have done css_get */ |
08e552c6 KH |
1738 | pc->mem_cgroup = to; |
1739 | mem_cgroup_charge_statistics(to, pc, true); | |
88703267 KH |
1740 | /* |
1741 | * We charges against "to" which may not have any tasks. Then, "to" | |
1742 | * can be under rmdir(). But in current implementation, caller of | |
4ffef5fe DN |
1743 | * this function is just force_empty() and move charge, so it's |
1744 | * garanteed that "to" is never removed. So, we don't check rmdir | |
1745 | * status here. | |
88703267 | 1746 | */ |
57f9fd7d DN |
1747 | } |
1748 | ||
1749 | /* | |
1750 | * check whether the @pc is valid for moving account and call | |
1751 | * __mem_cgroup_move_account() | |
1752 | */ | |
1753 | static int mem_cgroup_move_account(struct page_cgroup *pc, | |
854ffa8d | 1754 | struct mem_cgroup *from, struct mem_cgroup *to, bool uncharge) |
57f9fd7d DN |
1755 | { |
1756 | int ret = -EINVAL; | |
1757 | lock_page_cgroup(pc); | |
1758 | if (PageCgroupUsed(pc) && pc->mem_cgroup == from) { | |
854ffa8d | 1759 | __mem_cgroup_move_account(pc, from, to, uncharge); |
57f9fd7d DN |
1760 | ret = 0; |
1761 | } | |
1762 | unlock_page_cgroup(pc); | |
f817ed48 KH |
1763 | return ret; |
1764 | } | |
1765 | ||
1766 | /* | |
1767 | * move charges to its parent. | |
1768 | */ | |
1769 | ||
1770 | static int mem_cgroup_move_parent(struct page_cgroup *pc, | |
1771 | struct mem_cgroup *child, | |
1772 | gfp_t gfp_mask) | |
1773 | { | |
08e552c6 | 1774 | struct page *page = pc->page; |
f817ed48 KH |
1775 | struct cgroup *cg = child->css.cgroup; |
1776 | struct cgroup *pcg = cg->parent; | |
1777 | struct mem_cgroup *parent; | |
f817ed48 KH |
1778 | int ret; |
1779 | ||
1780 | /* Is ROOT ? */ | |
1781 | if (!pcg) | |
1782 | return -EINVAL; | |
1783 | ||
57f9fd7d DN |
1784 | ret = -EBUSY; |
1785 | if (!get_page_unless_zero(page)) | |
1786 | goto out; | |
1787 | if (isolate_lru_page(page)) | |
1788 | goto put; | |
08e552c6 | 1789 | |
f817ed48 | 1790 | parent = mem_cgroup_from_cont(pcg); |
430e4863 | 1791 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false); |
a636b327 | 1792 | if (ret || !parent) |
57f9fd7d | 1793 | goto put_back; |
f817ed48 | 1794 | |
854ffa8d DN |
1795 | ret = mem_cgroup_move_account(pc, child, parent, true); |
1796 | if (ret) | |
1797 | mem_cgroup_cancel_charge(parent); | |
57f9fd7d | 1798 | put_back: |
08e552c6 | 1799 | putback_lru_page(page); |
57f9fd7d | 1800 | put: |
40d58138 | 1801 | put_page(page); |
57f9fd7d | 1802 | out: |
f817ed48 KH |
1803 | return ret; |
1804 | } | |
1805 | ||
7a81b88c KH |
1806 | /* |
1807 | * Charge the memory controller for page usage. | |
1808 | * Return | |
1809 | * 0 if the charge was successful | |
1810 | * < 0 if the cgroup is over its limit | |
1811 | */ | |
1812 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, | |
1813 | gfp_t gfp_mask, enum charge_type ctype, | |
1814 | struct mem_cgroup *memcg) | |
1815 | { | |
1816 | struct mem_cgroup *mem; | |
1817 | struct page_cgroup *pc; | |
1818 | int ret; | |
1819 | ||
1820 | pc = lookup_page_cgroup(page); | |
1821 | /* can happen at boot */ | |
1822 | if (unlikely(!pc)) | |
1823 | return 0; | |
1824 | prefetchw(pc); | |
1825 | ||
1826 | mem = memcg; | |
430e4863 | 1827 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true); |
a636b327 | 1828 | if (ret || !mem) |
7a81b88c KH |
1829 | return ret; |
1830 | ||
1831 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
8a9f3ccd | 1832 | return 0; |
8a9f3ccd BS |
1833 | } |
1834 | ||
7a81b88c KH |
1835 | int mem_cgroup_newpage_charge(struct page *page, |
1836 | struct mm_struct *mm, gfp_t gfp_mask) | |
217bc319 | 1837 | { |
f8d66542 | 1838 | if (mem_cgroup_disabled()) |
cede86ac | 1839 | return 0; |
52d4b9ac KH |
1840 | if (PageCompound(page)) |
1841 | return 0; | |
69029cd5 KH |
1842 | /* |
1843 | * If already mapped, we don't have to account. | |
1844 | * If page cache, page->mapping has address_space. | |
1845 | * But page->mapping may have out-of-use anon_vma pointer, | |
1846 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping | |
1847 | * is NULL. | |
1848 | */ | |
1849 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) | |
1850 | return 0; | |
1851 | if (unlikely(!mm)) | |
1852 | mm = &init_mm; | |
217bc319 | 1853 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
e8589cc1 | 1854 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
217bc319 KH |
1855 | } |
1856 | ||
83aae4c7 DN |
1857 | static void |
1858 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1859 | enum charge_type ctype); | |
1860 | ||
e1a1cd59 BS |
1861 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
1862 | gfp_t gfp_mask) | |
8697d331 | 1863 | { |
b5a84319 KH |
1864 | struct mem_cgroup *mem = NULL; |
1865 | int ret; | |
1866 | ||
f8d66542 | 1867 | if (mem_cgroup_disabled()) |
cede86ac | 1868 | return 0; |
52d4b9ac KH |
1869 | if (PageCompound(page)) |
1870 | return 0; | |
accf163e KH |
1871 | /* |
1872 | * Corner case handling. This is called from add_to_page_cache() | |
1873 | * in usual. But some FS (shmem) precharges this page before calling it | |
1874 | * and call add_to_page_cache() with GFP_NOWAIT. | |
1875 | * | |
1876 | * For GFP_NOWAIT case, the page may be pre-charged before calling | |
1877 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call | |
1878 | * charge twice. (It works but has to pay a bit larger cost.) | |
b5a84319 KH |
1879 | * And when the page is SwapCache, it should take swap information |
1880 | * into account. This is under lock_page() now. | |
accf163e KH |
1881 | */ |
1882 | if (!(gfp_mask & __GFP_WAIT)) { | |
1883 | struct page_cgroup *pc; | |
1884 | ||
52d4b9ac KH |
1885 | |
1886 | pc = lookup_page_cgroup(page); | |
1887 | if (!pc) | |
1888 | return 0; | |
1889 | lock_page_cgroup(pc); | |
1890 | if (PageCgroupUsed(pc)) { | |
1891 | unlock_page_cgroup(pc); | |
accf163e KH |
1892 | return 0; |
1893 | } | |
52d4b9ac | 1894 | unlock_page_cgroup(pc); |
accf163e KH |
1895 | } |
1896 | ||
b5a84319 | 1897 | if (unlikely(!mm && !mem)) |
8697d331 | 1898 | mm = &init_mm; |
accf163e | 1899 | |
c05555b5 KH |
1900 | if (page_is_file_cache(page)) |
1901 | return mem_cgroup_charge_common(page, mm, gfp_mask, | |
e8589cc1 | 1902 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
b5a84319 | 1903 | |
83aae4c7 DN |
1904 | /* shmem */ |
1905 | if (PageSwapCache(page)) { | |
1906 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); | |
1907 | if (!ret) | |
1908 | __mem_cgroup_commit_charge_swapin(page, mem, | |
1909 | MEM_CGROUP_CHARGE_TYPE_SHMEM); | |
1910 | } else | |
1911 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, | |
1912 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); | |
b5a84319 | 1913 | |
b5a84319 | 1914 | return ret; |
e8589cc1 KH |
1915 | } |
1916 | ||
54595fe2 KH |
1917 | /* |
1918 | * While swap-in, try_charge -> commit or cancel, the page is locked. | |
1919 | * And when try_charge() successfully returns, one refcnt to memcg without | |
21ae2956 | 1920 | * struct page_cgroup is acquired. This refcnt will be consumed by |
54595fe2 KH |
1921 | * "commit()" or removed by "cancel()" |
1922 | */ | |
8c7c6e34 KH |
1923 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1924 | struct page *page, | |
1925 | gfp_t mask, struct mem_cgroup **ptr) | |
1926 | { | |
1927 | struct mem_cgroup *mem; | |
54595fe2 | 1928 | int ret; |
8c7c6e34 | 1929 | |
f8d66542 | 1930 | if (mem_cgroup_disabled()) |
8c7c6e34 KH |
1931 | return 0; |
1932 | ||
1933 | if (!do_swap_account) | |
1934 | goto charge_cur_mm; | |
8c7c6e34 KH |
1935 | /* |
1936 | * A racing thread's fault, or swapoff, may have already updated | |
407f9c8b HD |
1937 | * the pte, and even removed page from swap cache: in those cases |
1938 | * do_swap_page()'s pte_same() test will fail; but there's also a | |
1939 | * KSM case which does need to charge the page. | |
8c7c6e34 KH |
1940 | */ |
1941 | if (!PageSwapCache(page)) | |
407f9c8b | 1942 | goto charge_cur_mm; |
e42d9d5d | 1943 | mem = try_get_mem_cgroup_from_page(page); |
54595fe2 KH |
1944 | if (!mem) |
1945 | goto charge_cur_mm; | |
8c7c6e34 | 1946 | *ptr = mem; |
430e4863 | 1947 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true); |
54595fe2 KH |
1948 | /* drop extra refcnt from tryget */ |
1949 | css_put(&mem->css); | |
1950 | return ret; | |
8c7c6e34 KH |
1951 | charge_cur_mm: |
1952 | if (unlikely(!mm)) | |
1953 | mm = &init_mm; | |
430e4863 | 1954 | return __mem_cgroup_try_charge(mm, mask, ptr, true); |
8c7c6e34 KH |
1955 | } |
1956 | ||
83aae4c7 DN |
1957 | static void |
1958 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, | |
1959 | enum charge_type ctype) | |
7a81b88c KH |
1960 | { |
1961 | struct page_cgroup *pc; | |
1962 | ||
f8d66542 | 1963 | if (mem_cgroup_disabled()) |
7a81b88c KH |
1964 | return; |
1965 | if (!ptr) | |
1966 | return; | |
88703267 | 1967 | cgroup_exclude_rmdir(&ptr->css); |
7a81b88c | 1968 | pc = lookup_page_cgroup(page); |
544122e5 | 1969 | mem_cgroup_lru_del_before_commit_swapcache(page); |
83aae4c7 | 1970 | __mem_cgroup_commit_charge(ptr, pc, ctype); |
544122e5 | 1971 | mem_cgroup_lru_add_after_commit_swapcache(page); |
8c7c6e34 KH |
1972 | /* |
1973 | * Now swap is on-memory. This means this page may be | |
1974 | * counted both as mem and swap....double count. | |
03f3c433 KH |
1975 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
1976 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() | |
1977 | * may call delete_from_swap_cache() before reach here. | |
8c7c6e34 | 1978 | */ |
03f3c433 | 1979 | if (do_swap_account && PageSwapCache(page)) { |
8c7c6e34 | 1980 | swp_entry_t ent = {.val = page_private(page)}; |
a3b2d692 | 1981 | unsigned short id; |
8c7c6e34 | 1982 | struct mem_cgroup *memcg; |
a3b2d692 KH |
1983 | |
1984 | id = swap_cgroup_record(ent, 0); | |
1985 | rcu_read_lock(); | |
1986 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 1987 | if (memcg) { |
a3b2d692 KH |
1988 | /* |
1989 | * This recorded memcg can be obsolete one. So, avoid | |
1990 | * calling css_tryget | |
1991 | */ | |
0c3e73e8 | 1992 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 1993 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 1994 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
1995 | mem_cgroup_put(memcg); |
1996 | } | |
a3b2d692 | 1997 | rcu_read_unlock(); |
8c7c6e34 | 1998 | } |
88703267 KH |
1999 | /* |
2000 | * At swapin, we may charge account against cgroup which has no tasks. | |
2001 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2002 | * In that case, we need to call pre_destroy() again. check it here. | |
2003 | */ | |
2004 | cgroup_release_and_wakeup_rmdir(&ptr->css); | |
7a81b88c KH |
2005 | } |
2006 | ||
83aae4c7 DN |
2007 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
2008 | { | |
2009 | __mem_cgroup_commit_charge_swapin(page, ptr, | |
2010 | MEM_CGROUP_CHARGE_TYPE_MAPPED); | |
2011 | } | |
2012 | ||
7a81b88c KH |
2013 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
2014 | { | |
f8d66542 | 2015 | if (mem_cgroup_disabled()) |
7a81b88c KH |
2016 | return; |
2017 | if (!mem) | |
2018 | return; | |
a3032a2c | 2019 | mem_cgroup_cancel_charge(mem); |
7a81b88c KH |
2020 | } |
2021 | ||
569b846d KH |
2022 | static void |
2023 | __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype) | |
2024 | { | |
2025 | struct memcg_batch_info *batch = NULL; | |
2026 | bool uncharge_memsw = true; | |
2027 | /* If swapout, usage of swap doesn't decrease */ | |
2028 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2029 | uncharge_memsw = false; | |
2030 | /* | |
2031 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. | |
2032 | * In those cases, all pages freed continously can be expected to be in | |
2033 | * the same cgroup and we have chance to coalesce uncharges. | |
2034 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) | |
2035 | * because we want to do uncharge as soon as possible. | |
2036 | */ | |
2037 | if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE)) | |
2038 | goto direct_uncharge; | |
2039 | ||
2040 | batch = ¤t->memcg_batch; | |
2041 | /* | |
2042 | * In usual, we do css_get() when we remember memcg pointer. | |
2043 | * But in this case, we keep res->usage until end of a series of | |
2044 | * uncharges. Then, it's ok to ignore memcg's refcnt. | |
2045 | */ | |
2046 | if (!batch->memcg) | |
2047 | batch->memcg = mem; | |
2048 | /* | |
2049 | * In typical case, batch->memcg == mem. This means we can | |
2050 | * merge a series of uncharges to an uncharge of res_counter. | |
2051 | * If not, we uncharge res_counter ony by one. | |
2052 | */ | |
2053 | if (batch->memcg != mem) | |
2054 | goto direct_uncharge; | |
2055 | /* remember freed charge and uncharge it later */ | |
2056 | batch->bytes += PAGE_SIZE; | |
2057 | if (uncharge_memsw) | |
2058 | batch->memsw_bytes += PAGE_SIZE; | |
2059 | return; | |
2060 | direct_uncharge: | |
2061 | res_counter_uncharge(&mem->res, PAGE_SIZE); | |
2062 | if (uncharge_memsw) | |
2063 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); | |
2064 | return; | |
2065 | } | |
7a81b88c | 2066 | |
8a9f3ccd | 2067 | /* |
69029cd5 | 2068 | * uncharge if !page_mapped(page) |
8a9f3ccd | 2069 | */ |
8c7c6e34 | 2070 | static struct mem_cgroup * |
69029cd5 | 2071 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
8a9f3ccd | 2072 | { |
8289546e | 2073 | struct page_cgroup *pc; |
8c7c6e34 | 2074 | struct mem_cgroup *mem = NULL; |
072c56c1 | 2075 | struct mem_cgroup_per_zone *mz; |
8a9f3ccd | 2076 | |
f8d66542 | 2077 | if (mem_cgroup_disabled()) |
8c7c6e34 | 2078 | return NULL; |
4077960e | 2079 | |
d13d1443 | 2080 | if (PageSwapCache(page)) |
8c7c6e34 | 2081 | return NULL; |
d13d1443 | 2082 | |
8697d331 | 2083 | /* |
3c541e14 | 2084 | * Check if our page_cgroup is valid |
8697d331 | 2085 | */ |
52d4b9ac KH |
2086 | pc = lookup_page_cgroup(page); |
2087 | if (unlikely(!pc || !PageCgroupUsed(pc))) | |
8c7c6e34 | 2088 | return NULL; |
b9c565d5 | 2089 | |
52d4b9ac | 2090 | lock_page_cgroup(pc); |
d13d1443 | 2091 | |
8c7c6e34 KH |
2092 | mem = pc->mem_cgroup; |
2093 | ||
d13d1443 KH |
2094 | if (!PageCgroupUsed(pc)) |
2095 | goto unlock_out; | |
2096 | ||
2097 | switch (ctype) { | |
2098 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: | |
8a9478ca | 2099 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
d13d1443 KH |
2100 | if (page_mapped(page)) |
2101 | goto unlock_out; | |
2102 | break; | |
2103 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: | |
2104 | if (!PageAnon(page)) { /* Shared memory */ | |
2105 | if (page->mapping && !page_is_file_cache(page)) | |
2106 | goto unlock_out; | |
2107 | } else if (page_mapped(page)) /* Anon */ | |
2108 | goto unlock_out; | |
2109 | break; | |
2110 | default: | |
2111 | break; | |
52d4b9ac | 2112 | } |
d13d1443 | 2113 | |
569b846d KH |
2114 | if (!mem_cgroup_is_root(mem)) |
2115 | __do_uncharge(mem, ctype); | |
0c3e73e8 BS |
2116 | if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
2117 | mem_cgroup_swap_statistics(mem, true); | |
08e552c6 | 2118 | mem_cgroup_charge_statistics(mem, pc, false); |
04046e1a | 2119 | |
52d4b9ac | 2120 | ClearPageCgroupUsed(pc); |
544122e5 KH |
2121 | /* |
2122 | * pc->mem_cgroup is not cleared here. It will be accessed when it's | |
2123 | * freed from LRU. This is safe because uncharged page is expected not | |
2124 | * to be reused (freed soon). Exception is SwapCache, it's handled by | |
2125 | * special functions. | |
2126 | */ | |
b9c565d5 | 2127 | |
69029cd5 | 2128 | mz = page_cgroup_zoneinfo(pc); |
52d4b9ac | 2129 | unlock_page_cgroup(pc); |
fb59e9f1 | 2130 | |
4e649152 | 2131 | if (mem_cgroup_soft_limit_check(mem)) |
f64c3f54 | 2132 | mem_cgroup_update_tree(mem, page); |
2e72b634 KS |
2133 | if (mem_cgroup_threshold_check(mem)) |
2134 | mem_cgroup_threshold(mem); | |
a7fe942e KH |
2135 | /* at swapout, this memcg will be accessed to record to swap */ |
2136 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) | |
2137 | css_put(&mem->css); | |
6d12e2d8 | 2138 | |
8c7c6e34 | 2139 | return mem; |
d13d1443 KH |
2140 | |
2141 | unlock_out: | |
2142 | unlock_page_cgroup(pc); | |
8c7c6e34 | 2143 | return NULL; |
3c541e14 BS |
2144 | } |
2145 | ||
69029cd5 KH |
2146 | void mem_cgroup_uncharge_page(struct page *page) |
2147 | { | |
52d4b9ac KH |
2148 | /* early check. */ |
2149 | if (page_mapped(page)) | |
2150 | return; | |
2151 | if (page->mapping && !PageAnon(page)) | |
2152 | return; | |
69029cd5 KH |
2153 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
2154 | } | |
2155 | ||
2156 | void mem_cgroup_uncharge_cache_page(struct page *page) | |
2157 | { | |
2158 | VM_BUG_ON(page_mapped(page)); | |
b7abea96 | 2159 | VM_BUG_ON(page->mapping); |
69029cd5 KH |
2160 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
2161 | } | |
2162 | ||
569b846d KH |
2163 | /* |
2164 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. | |
2165 | * In that cases, pages are freed continuously and we can expect pages | |
2166 | * are in the same memcg. All these calls itself limits the number of | |
2167 | * pages freed at once, then uncharge_start/end() is called properly. | |
2168 | * This may be called prural(2) times in a context, | |
2169 | */ | |
2170 | ||
2171 | void mem_cgroup_uncharge_start(void) | |
2172 | { | |
2173 | current->memcg_batch.do_batch++; | |
2174 | /* We can do nest. */ | |
2175 | if (current->memcg_batch.do_batch == 1) { | |
2176 | current->memcg_batch.memcg = NULL; | |
2177 | current->memcg_batch.bytes = 0; | |
2178 | current->memcg_batch.memsw_bytes = 0; | |
2179 | } | |
2180 | } | |
2181 | ||
2182 | void mem_cgroup_uncharge_end(void) | |
2183 | { | |
2184 | struct memcg_batch_info *batch = ¤t->memcg_batch; | |
2185 | ||
2186 | if (!batch->do_batch) | |
2187 | return; | |
2188 | ||
2189 | batch->do_batch--; | |
2190 | if (batch->do_batch) /* If stacked, do nothing. */ | |
2191 | return; | |
2192 | ||
2193 | if (!batch->memcg) | |
2194 | return; | |
2195 | /* | |
2196 | * This "batch->memcg" is valid without any css_get/put etc... | |
2197 | * bacause we hide charges behind us. | |
2198 | */ | |
2199 | if (batch->bytes) | |
2200 | res_counter_uncharge(&batch->memcg->res, batch->bytes); | |
2201 | if (batch->memsw_bytes) | |
2202 | res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes); | |
2203 | /* forget this pointer (for sanity check) */ | |
2204 | batch->memcg = NULL; | |
2205 | } | |
2206 | ||
e767e056 | 2207 | #ifdef CONFIG_SWAP |
8c7c6e34 | 2208 | /* |
e767e056 | 2209 | * called after __delete_from_swap_cache() and drop "page" account. |
8c7c6e34 KH |
2210 | * memcg information is recorded to swap_cgroup of "ent" |
2211 | */ | |
8a9478ca KH |
2212 | void |
2213 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) | |
8c7c6e34 KH |
2214 | { |
2215 | struct mem_cgroup *memcg; | |
8a9478ca KH |
2216 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
2217 | ||
2218 | if (!swapout) /* this was a swap cache but the swap is unused ! */ | |
2219 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; | |
2220 | ||
2221 | memcg = __mem_cgroup_uncharge_common(page, ctype); | |
8c7c6e34 | 2222 | |
8c7c6e34 | 2223 | /* record memcg information */ |
8a9478ca | 2224 | if (do_swap_account && swapout && memcg) { |
a3b2d692 | 2225 | swap_cgroup_record(ent, css_id(&memcg->css)); |
8c7c6e34 KH |
2226 | mem_cgroup_get(memcg); |
2227 | } | |
8a9478ca | 2228 | if (swapout && memcg) |
a7fe942e | 2229 | css_put(&memcg->css); |
8c7c6e34 | 2230 | } |
e767e056 | 2231 | #endif |
8c7c6e34 KH |
2232 | |
2233 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
2234 | /* | |
2235 | * called from swap_entry_free(). remove record in swap_cgroup and | |
2236 | * uncharge "memsw" account. | |
2237 | */ | |
2238 | void mem_cgroup_uncharge_swap(swp_entry_t ent) | |
d13d1443 | 2239 | { |
8c7c6e34 | 2240 | struct mem_cgroup *memcg; |
a3b2d692 | 2241 | unsigned short id; |
8c7c6e34 KH |
2242 | |
2243 | if (!do_swap_account) | |
2244 | return; | |
2245 | ||
a3b2d692 KH |
2246 | id = swap_cgroup_record(ent, 0); |
2247 | rcu_read_lock(); | |
2248 | memcg = mem_cgroup_lookup(id); | |
8c7c6e34 | 2249 | if (memcg) { |
a3b2d692 KH |
2250 | /* |
2251 | * We uncharge this because swap is freed. | |
2252 | * This memcg can be obsolete one. We avoid calling css_tryget | |
2253 | */ | |
0c3e73e8 | 2254 | if (!mem_cgroup_is_root(memcg)) |
4e649152 | 2255 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
0c3e73e8 | 2256 | mem_cgroup_swap_statistics(memcg, false); |
8c7c6e34 KH |
2257 | mem_cgroup_put(memcg); |
2258 | } | |
a3b2d692 | 2259 | rcu_read_unlock(); |
d13d1443 | 2260 | } |
02491447 DN |
2261 | |
2262 | /** | |
2263 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. | |
2264 | * @entry: swap entry to be moved | |
2265 | * @from: mem_cgroup which the entry is moved from | |
2266 | * @to: mem_cgroup which the entry is moved to | |
483c30b5 | 2267 | * @need_fixup: whether we should fixup res_counters and refcounts. |
02491447 DN |
2268 | * |
2269 | * It succeeds only when the swap_cgroup's record for this entry is the same | |
2270 | * as the mem_cgroup's id of @from. | |
2271 | * | |
2272 | * Returns 0 on success, -EINVAL on failure. | |
2273 | * | |
2274 | * The caller must have charged to @to, IOW, called res_counter_charge() about | |
2275 | * both res and memsw, and called css_get(). | |
2276 | */ | |
2277 | static int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2278 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2279 | { |
2280 | unsigned short old_id, new_id; | |
2281 | ||
2282 | old_id = css_id(&from->css); | |
2283 | new_id = css_id(&to->css); | |
2284 | ||
2285 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { | |
02491447 | 2286 | mem_cgroup_swap_statistics(from, false); |
483c30b5 | 2287 | mem_cgroup_swap_statistics(to, true); |
02491447 | 2288 | /* |
483c30b5 DN |
2289 | * This function is only called from task migration context now. |
2290 | * It postpones res_counter and refcount handling till the end | |
2291 | * of task migration(mem_cgroup_clear_mc()) for performance | |
2292 | * improvement. But we cannot postpone mem_cgroup_get(to) | |
2293 | * because if the process that has been moved to @to does | |
2294 | * swap-in, the refcount of @to might be decreased to 0. | |
02491447 | 2295 | */ |
02491447 | 2296 | mem_cgroup_get(to); |
483c30b5 DN |
2297 | if (need_fixup) { |
2298 | if (!mem_cgroup_is_root(from)) | |
2299 | res_counter_uncharge(&from->memsw, PAGE_SIZE); | |
2300 | mem_cgroup_put(from); | |
2301 | /* | |
2302 | * we charged both to->res and to->memsw, so we should | |
2303 | * uncharge to->res. | |
2304 | */ | |
2305 | if (!mem_cgroup_is_root(to)) | |
2306 | res_counter_uncharge(&to->res, PAGE_SIZE); | |
2307 | css_put(&to->css); | |
2308 | } | |
02491447 DN |
2309 | return 0; |
2310 | } | |
2311 | return -EINVAL; | |
2312 | } | |
2313 | #else | |
2314 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, | |
483c30b5 | 2315 | struct mem_cgroup *from, struct mem_cgroup *to, bool need_fixup) |
02491447 DN |
2316 | { |
2317 | return -EINVAL; | |
2318 | } | |
8c7c6e34 | 2319 | #endif |
d13d1443 | 2320 | |
ae41be37 | 2321 | /* |
01b1ae63 KH |
2322 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
2323 | * page belongs to. | |
ae41be37 | 2324 | */ |
01b1ae63 | 2325 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
ae41be37 KH |
2326 | { |
2327 | struct page_cgroup *pc; | |
e8589cc1 | 2328 | struct mem_cgroup *mem = NULL; |
e8589cc1 | 2329 | int ret = 0; |
8869b8f6 | 2330 | |
f8d66542 | 2331 | if (mem_cgroup_disabled()) |
4077960e BS |
2332 | return 0; |
2333 | ||
52d4b9ac KH |
2334 | pc = lookup_page_cgroup(page); |
2335 | lock_page_cgroup(pc); | |
2336 | if (PageCgroupUsed(pc)) { | |
e8589cc1 KH |
2337 | mem = pc->mem_cgroup; |
2338 | css_get(&mem->css); | |
e8589cc1 | 2339 | } |
52d4b9ac | 2340 | unlock_page_cgroup(pc); |
01b1ae63 | 2341 | |
e8589cc1 | 2342 | if (mem) { |
430e4863 | 2343 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
e8589cc1 KH |
2344 | css_put(&mem->css); |
2345 | } | |
01b1ae63 | 2346 | *ptr = mem; |
e8589cc1 | 2347 | return ret; |
ae41be37 | 2348 | } |
8869b8f6 | 2349 | |
69029cd5 | 2350 | /* remove redundant charge if migration failed*/ |
01b1ae63 KH |
2351 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
2352 | struct page *oldpage, struct page *newpage) | |
ae41be37 | 2353 | { |
01b1ae63 KH |
2354 | struct page *target, *unused; |
2355 | struct page_cgroup *pc; | |
2356 | enum charge_type ctype; | |
2357 | ||
2358 | if (!mem) | |
2359 | return; | |
88703267 | 2360 | cgroup_exclude_rmdir(&mem->css); |
01b1ae63 KH |
2361 | /* at migration success, oldpage->mapping is NULL. */ |
2362 | if (oldpage->mapping) { | |
2363 | target = oldpage; | |
2364 | unused = NULL; | |
2365 | } else { | |
2366 | target = newpage; | |
2367 | unused = oldpage; | |
2368 | } | |
2369 | ||
2370 | if (PageAnon(target)) | |
2371 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; | |
2372 | else if (page_is_file_cache(target)) | |
2373 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; | |
2374 | else | |
2375 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; | |
2376 | ||
2377 | /* unused page is not on radix-tree now. */ | |
d13d1443 | 2378 | if (unused) |
01b1ae63 KH |
2379 | __mem_cgroup_uncharge_common(unused, ctype); |
2380 | ||
2381 | pc = lookup_page_cgroup(target); | |
69029cd5 | 2382 | /* |
01b1ae63 KH |
2383 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
2384 | * So, double-counting is effectively avoided. | |
2385 | */ | |
2386 | __mem_cgroup_commit_charge(mem, pc, ctype); | |
2387 | ||
2388 | /* | |
2389 | * Both of oldpage and newpage are still under lock_page(). | |
2390 | * Then, we don't have to care about race in radix-tree. | |
2391 | * But we have to be careful that this page is unmapped or not. | |
2392 | * | |
2393 | * There is a case for !page_mapped(). At the start of | |
2394 | * migration, oldpage was mapped. But now, it's zapped. | |
2395 | * But we know *target* page is not freed/reused under us. | |
2396 | * mem_cgroup_uncharge_page() does all necessary checks. | |
69029cd5 | 2397 | */ |
01b1ae63 KH |
2398 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
2399 | mem_cgroup_uncharge_page(target); | |
88703267 KH |
2400 | /* |
2401 | * At migration, we may charge account against cgroup which has no tasks | |
2402 | * So, rmdir()->pre_destroy() can be called while we do this charge. | |
2403 | * In that case, we need to call pre_destroy() again. check it here. | |
2404 | */ | |
2405 | cgroup_release_and_wakeup_rmdir(&mem->css); | |
ae41be37 | 2406 | } |
78fb7466 | 2407 | |
c9b0ed51 | 2408 | /* |
ae3abae6 DN |
2409 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
2410 | * Calling hierarchical_reclaim is not enough because we should update | |
2411 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. | |
2412 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, | |
2413 | * not from the memcg which this page would be charged to. | |
2414 | * try_charge_swapin does all of these works properly. | |
c9b0ed51 | 2415 | */ |
ae3abae6 | 2416 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
b5a84319 KH |
2417 | struct mm_struct *mm, |
2418 | gfp_t gfp_mask) | |
c9b0ed51 | 2419 | { |
b5a84319 | 2420 | struct mem_cgroup *mem = NULL; |
ae3abae6 | 2421 | int ret; |
c9b0ed51 | 2422 | |
f8d66542 | 2423 | if (mem_cgroup_disabled()) |
cede86ac | 2424 | return 0; |
c9b0ed51 | 2425 | |
ae3abae6 DN |
2426 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2427 | if (!ret) | |
2428 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ | |
c9b0ed51 | 2429 | |
ae3abae6 | 2430 | return ret; |
c9b0ed51 KH |
2431 | } |
2432 | ||
8c7c6e34 KH |
2433 | static DEFINE_MUTEX(set_limit_mutex); |
2434 | ||
d38d2a75 | 2435 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
8c7c6e34 | 2436 | unsigned long long val) |
628f4235 | 2437 | { |
81d39c20 | 2438 | int retry_count; |
8c7c6e34 | 2439 | u64 memswlimit; |
628f4235 | 2440 | int ret = 0; |
81d39c20 KH |
2441 | int children = mem_cgroup_count_children(memcg); |
2442 | u64 curusage, oldusage; | |
2443 | ||
2444 | /* | |
2445 | * For keeping hierarchical_reclaim simple, how long we should retry | |
2446 | * is depends on callers. We set our retry-count to be function | |
2447 | * of # of children which we should visit in this loop. | |
2448 | */ | |
2449 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; | |
2450 | ||
2451 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); | |
628f4235 | 2452 | |
8c7c6e34 | 2453 | while (retry_count) { |
628f4235 KH |
2454 | if (signal_pending(current)) { |
2455 | ret = -EINTR; | |
2456 | break; | |
2457 | } | |
8c7c6e34 KH |
2458 | /* |
2459 | * Rather than hide all in some function, I do this in | |
2460 | * open coded manner. You see what this really does. | |
2461 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2462 | */ | |
2463 | mutex_lock(&set_limit_mutex); | |
2464 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2465 | if (memswlimit < val) { | |
2466 | ret = -EINVAL; | |
2467 | mutex_unlock(&set_limit_mutex); | |
628f4235 KH |
2468 | break; |
2469 | } | |
8c7c6e34 | 2470 | ret = res_counter_set_limit(&memcg->res, val); |
22a668d7 KH |
2471 | if (!ret) { |
2472 | if (memswlimit == val) | |
2473 | memcg->memsw_is_minimum = true; | |
2474 | else | |
2475 | memcg->memsw_is_minimum = false; | |
2476 | } | |
8c7c6e34 KH |
2477 | mutex_unlock(&set_limit_mutex); |
2478 | ||
2479 | if (!ret) | |
2480 | break; | |
2481 | ||
aa20d489 | 2482 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
4e416953 | 2483 | MEM_CGROUP_RECLAIM_SHRINK); |
81d39c20 KH |
2484 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
2485 | /* Usage is reduced ? */ | |
2486 | if (curusage >= oldusage) | |
2487 | retry_count--; | |
2488 | else | |
2489 | oldusage = curusage; | |
8c7c6e34 | 2490 | } |
14797e23 | 2491 | |
8c7c6e34 KH |
2492 | return ret; |
2493 | } | |
2494 | ||
338c8431 LZ |
2495 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
2496 | unsigned long long val) | |
8c7c6e34 | 2497 | { |
81d39c20 | 2498 | int retry_count; |
8c7c6e34 | 2499 | u64 memlimit, oldusage, curusage; |
81d39c20 KH |
2500 | int children = mem_cgroup_count_children(memcg); |
2501 | int ret = -EBUSY; | |
8c7c6e34 | 2502 | |
81d39c20 KH |
2503 | /* see mem_cgroup_resize_res_limit */ |
2504 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; | |
2505 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); | |
8c7c6e34 KH |
2506 | while (retry_count) { |
2507 | if (signal_pending(current)) { | |
2508 | ret = -EINTR; | |
2509 | break; | |
2510 | } | |
2511 | /* | |
2512 | * Rather than hide all in some function, I do this in | |
2513 | * open coded manner. You see what this really does. | |
2514 | * We have to guarantee mem->res.limit < mem->memsw.limit. | |
2515 | */ | |
2516 | mutex_lock(&set_limit_mutex); | |
2517 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2518 | if (memlimit > val) { | |
2519 | ret = -EINVAL; | |
2520 | mutex_unlock(&set_limit_mutex); | |
2521 | break; | |
2522 | } | |
2523 | ret = res_counter_set_limit(&memcg->memsw, val); | |
22a668d7 KH |
2524 | if (!ret) { |
2525 | if (memlimit == val) | |
2526 | memcg->memsw_is_minimum = true; | |
2527 | else | |
2528 | memcg->memsw_is_minimum = false; | |
2529 | } | |
8c7c6e34 KH |
2530 | mutex_unlock(&set_limit_mutex); |
2531 | ||
2532 | if (!ret) | |
2533 | break; | |
2534 | ||
4e416953 | 2535 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
75822b44 BS |
2536 | MEM_CGROUP_RECLAIM_NOSWAP | |
2537 | MEM_CGROUP_RECLAIM_SHRINK); | |
8c7c6e34 | 2538 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
81d39c20 | 2539 | /* Usage is reduced ? */ |
8c7c6e34 | 2540 | if (curusage >= oldusage) |
628f4235 | 2541 | retry_count--; |
81d39c20 KH |
2542 | else |
2543 | oldusage = curusage; | |
628f4235 KH |
2544 | } |
2545 | return ret; | |
2546 | } | |
2547 | ||
4e416953 BS |
2548 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2549 | gfp_t gfp_mask, int nid, | |
2550 | int zid) | |
2551 | { | |
2552 | unsigned long nr_reclaimed = 0; | |
2553 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; | |
2554 | unsigned long reclaimed; | |
2555 | int loop = 0; | |
2556 | struct mem_cgroup_tree_per_zone *mctz; | |
ef8745c1 | 2557 | unsigned long long excess; |
4e416953 BS |
2558 | |
2559 | if (order > 0) | |
2560 | return 0; | |
2561 | ||
2562 | mctz = soft_limit_tree_node_zone(nid, zid); | |
2563 | /* | |
2564 | * This loop can run a while, specially if mem_cgroup's continuously | |
2565 | * keep exceeding their soft limit and putting the system under | |
2566 | * pressure | |
2567 | */ | |
2568 | do { | |
2569 | if (next_mz) | |
2570 | mz = next_mz; | |
2571 | else | |
2572 | mz = mem_cgroup_largest_soft_limit_node(mctz); | |
2573 | if (!mz) | |
2574 | break; | |
2575 | ||
2576 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, | |
2577 | gfp_mask, | |
2578 | MEM_CGROUP_RECLAIM_SOFT); | |
2579 | nr_reclaimed += reclaimed; | |
2580 | spin_lock(&mctz->lock); | |
2581 | ||
2582 | /* | |
2583 | * If we failed to reclaim anything from this memory cgroup | |
2584 | * it is time to move on to the next cgroup | |
2585 | */ | |
2586 | next_mz = NULL; | |
2587 | if (!reclaimed) { | |
2588 | do { | |
2589 | /* | |
2590 | * Loop until we find yet another one. | |
2591 | * | |
2592 | * By the time we get the soft_limit lock | |
2593 | * again, someone might have aded the | |
2594 | * group back on the RB tree. Iterate to | |
2595 | * make sure we get a different mem. | |
2596 | * mem_cgroup_largest_soft_limit_node returns | |
2597 | * NULL if no other cgroup is present on | |
2598 | * the tree | |
2599 | */ | |
2600 | next_mz = | |
2601 | __mem_cgroup_largest_soft_limit_node(mctz); | |
2602 | if (next_mz == mz) { | |
2603 | css_put(&next_mz->mem->css); | |
2604 | next_mz = NULL; | |
2605 | } else /* next_mz == NULL or other memcg */ | |
2606 | break; | |
2607 | } while (1); | |
2608 | } | |
4e416953 | 2609 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
ef8745c1 | 2610 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
4e416953 BS |
2611 | /* |
2612 | * One school of thought says that we should not add | |
2613 | * back the node to the tree if reclaim returns 0. | |
2614 | * But our reclaim could return 0, simply because due | |
2615 | * to priority we are exposing a smaller subset of | |
2616 | * memory to reclaim from. Consider this as a longer | |
2617 | * term TODO. | |
2618 | */ | |
ef8745c1 KH |
2619 | /* If excess == 0, no tree ops */ |
2620 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); | |
4e416953 BS |
2621 | spin_unlock(&mctz->lock); |
2622 | css_put(&mz->mem->css); | |
2623 | loop++; | |
2624 | /* | |
2625 | * Could not reclaim anything and there are no more | |
2626 | * mem cgroups to try or we seem to be looping without | |
2627 | * reclaiming anything. | |
2628 | */ | |
2629 | if (!nr_reclaimed && | |
2630 | (next_mz == NULL || | |
2631 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) | |
2632 | break; | |
2633 | } while (!nr_reclaimed); | |
2634 | if (next_mz) | |
2635 | css_put(&next_mz->mem->css); | |
2636 | return nr_reclaimed; | |
2637 | } | |
2638 | ||
cc847582 KH |
2639 | /* |
2640 | * This routine traverse page_cgroup in given list and drop them all. | |
cc847582 KH |
2641 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
2642 | */ | |
f817ed48 | 2643 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
08e552c6 | 2644 | int node, int zid, enum lru_list lru) |
cc847582 | 2645 | { |
08e552c6 KH |
2646 | struct zone *zone; |
2647 | struct mem_cgroup_per_zone *mz; | |
f817ed48 | 2648 | struct page_cgroup *pc, *busy; |
08e552c6 | 2649 | unsigned long flags, loop; |
072c56c1 | 2650 | struct list_head *list; |
f817ed48 | 2651 | int ret = 0; |
072c56c1 | 2652 | |
08e552c6 KH |
2653 | zone = &NODE_DATA(node)->node_zones[zid]; |
2654 | mz = mem_cgroup_zoneinfo(mem, node, zid); | |
b69408e8 | 2655 | list = &mz->lists[lru]; |
cc847582 | 2656 | |
f817ed48 KH |
2657 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
2658 | /* give some margin against EBUSY etc...*/ | |
2659 | loop += 256; | |
2660 | busy = NULL; | |
2661 | while (loop--) { | |
2662 | ret = 0; | |
08e552c6 | 2663 | spin_lock_irqsave(&zone->lru_lock, flags); |
f817ed48 | 2664 | if (list_empty(list)) { |
08e552c6 | 2665 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
52d4b9ac | 2666 | break; |
f817ed48 KH |
2667 | } |
2668 | pc = list_entry(list->prev, struct page_cgroup, lru); | |
2669 | if (busy == pc) { | |
2670 | list_move(&pc->lru, list); | |
648bcc77 | 2671 | busy = NULL; |
08e552c6 | 2672 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 KH |
2673 | continue; |
2674 | } | |
08e552c6 | 2675 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
f817ed48 | 2676 | |
2c26fdd7 | 2677 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
f817ed48 | 2678 | if (ret == -ENOMEM) |
52d4b9ac | 2679 | break; |
f817ed48 KH |
2680 | |
2681 | if (ret == -EBUSY || ret == -EINVAL) { | |
2682 | /* found lock contention or "pc" is obsolete. */ | |
2683 | busy = pc; | |
2684 | cond_resched(); | |
2685 | } else | |
2686 | busy = NULL; | |
cc847582 | 2687 | } |
08e552c6 | 2688 | |
f817ed48 KH |
2689 | if (!ret && !list_empty(list)) |
2690 | return -EBUSY; | |
2691 | return ret; | |
cc847582 KH |
2692 | } |
2693 | ||
2694 | /* | |
2695 | * make mem_cgroup's charge to be 0 if there is no task. | |
2696 | * This enables deleting this mem_cgroup. | |
2697 | */ | |
c1e862c1 | 2698 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
cc847582 | 2699 | { |
f817ed48 KH |
2700 | int ret; |
2701 | int node, zid, shrink; | |
2702 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; | |
c1e862c1 | 2703 | struct cgroup *cgrp = mem->css.cgroup; |
8869b8f6 | 2704 | |
cc847582 | 2705 | css_get(&mem->css); |
f817ed48 KH |
2706 | |
2707 | shrink = 0; | |
c1e862c1 KH |
2708 | /* should free all ? */ |
2709 | if (free_all) | |
2710 | goto try_to_free; | |
f817ed48 | 2711 | move_account: |
fce66477 | 2712 | do { |
f817ed48 | 2713 | ret = -EBUSY; |
c1e862c1 KH |
2714 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
2715 | goto out; | |
2716 | ret = -EINTR; | |
2717 | if (signal_pending(current)) | |
cc847582 | 2718 | goto out; |
52d4b9ac KH |
2719 | /* This is for making all *used* pages to be on LRU. */ |
2720 | lru_add_drain_all(); | |
cdec2e42 | 2721 | drain_all_stock_sync(); |
f817ed48 | 2722 | ret = 0; |
299b4eaa | 2723 | for_each_node_state(node, N_HIGH_MEMORY) { |
f817ed48 | 2724 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
b69408e8 | 2725 | enum lru_list l; |
f817ed48 KH |
2726 | for_each_lru(l) { |
2727 | ret = mem_cgroup_force_empty_list(mem, | |
08e552c6 | 2728 | node, zid, l); |
f817ed48 KH |
2729 | if (ret) |
2730 | break; | |
2731 | } | |
1ecaab2b | 2732 | } |
f817ed48 KH |
2733 | if (ret) |
2734 | break; | |
2735 | } | |
2736 | /* it seems parent cgroup doesn't have enough mem */ | |
2737 | if (ret == -ENOMEM) | |
2738 | goto try_to_free; | |
52d4b9ac | 2739 | cond_resched(); |
fce66477 DN |
2740 | /* "ret" should also be checked to ensure all lists are empty. */ |
2741 | } while (mem->res.usage > 0 || ret); | |
cc847582 KH |
2742 | out: |
2743 | css_put(&mem->css); | |
2744 | return ret; | |
f817ed48 KH |
2745 | |
2746 | try_to_free: | |
c1e862c1 KH |
2747 | /* returns EBUSY if there is a task or if we come here twice. */ |
2748 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { | |
f817ed48 KH |
2749 | ret = -EBUSY; |
2750 | goto out; | |
2751 | } | |
c1e862c1 KH |
2752 | /* we call try-to-free pages for make this cgroup empty */ |
2753 | lru_add_drain_all(); | |
f817ed48 KH |
2754 | /* try to free all pages in this cgroup */ |
2755 | shrink = 1; | |
2756 | while (nr_retries && mem->res.usage > 0) { | |
2757 | int progress; | |
c1e862c1 KH |
2758 | |
2759 | if (signal_pending(current)) { | |
2760 | ret = -EINTR; | |
2761 | goto out; | |
2762 | } | |
a7885eb8 KM |
2763 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
2764 | false, get_swappiness(mem)); | |
c1e862c1 | 2765 | if (!progress) { |
f817ed48 | 2766 | nr_retries--; |
c1e862c1 | 2767 | /* maybe some writeback is necessary */ |
8aa7e847 | 2768 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
c1e862c1 | 2769 | } |
f817ed48 KH |
2770 | |
2771 | } | |
08e552c6 | 2772 | lru_add_drain(); |
f817ed48 | 2773 | /* try move_account...there may be some *locked* pages. */ |
fce66477 | 2774 | goto move_account; |
cc847582 KH |
2775 | } |
2776 | ||
c1e862c1 KH |
2777 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
2778 | { | |
2779 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); | |
2780 | } | |
2781 | ||
2782 | ||
18f59ea7 BS |
2783 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
2784 | { | |
2785 | return mem_cgroup_from_cont(cont)->use_hierarchy; | |
2786 | } | |
2787 | ||
2788 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, | |
2789 | u64 val) | |
2790 | { | |
2791 | int retval = 0; | |
2792 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
2793 | struct cgroup *parent = cont->parent; | |
2794 | struct mem_cgroup *parent_mem = NULL; | |
2795 | ||
2796 | if (parent) | |
2797 | parent_mem = mem_cgroup_from_cont(parent); | |
2798 | ||
2799 | cgroup_lock(); | |
2800 | /* | |
af901ca1 | 2801 | * If parent's use_hierarchy is set, we can't make any modifications |
18f59ea7 BS |
2802 | * in the child subtrees. If it is unset, then the change can |
2803 | * occur, provided the current cgroup has no children. | |
2804 | * | |
2805 | * For the root cgroup, parent_mem is NULL, we allow value to be | |
2806 | * set if there are no children. | |
2807 | */ | |
2808 | if ((!parent_mem || !parent_mem->use_hierarchy) && | |
2809 | (val == 1 || val == 0)) { | |
2810 | if (list_empty(&cont->children)) | |
2811 | mem->use_hierarchy = val; | |
2812 | else | |
2813 | retval = -EBUSY; | |
2814 | } else | |
2815 | retval = -EINVAL; | |
2816 | cgroup_unlock(); | |
2817 | ||
2818 | return retval; | |
2819 | } | |
2820 | ||
0c3e73e8 BS |
2821 | struct mem_cgroup_idx_data { |
2822 | s64 val; | |
2823 | enum mem_cgroup_stat_index idx; | |
2824 | }; | |
2825 | ||
2826 | static int | |
2827 | mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) | |
2828 | { | |
2829 | struct mem_cgroup_idx_data *d = data; | |
c62b1a3b | 2830 | d->val += mem_cgroup_read_stat(mem, d->idx); |
0c3e73e8 BS |
2831 | return 0; |
2832 | } | |
2833 | ||
2834 | static void | |
2835 | mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, | |
2836 | enum mem_cgroup_stat_index idx, s64 *val) | |
2837 | { | |
2838 | struct mem_cgroup_idx_data d; | |
2839 | d.idx = idx; | |
2840 | d.val = 0; | |
2841 | mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat); | |
2842 | *val = d.val; | |
2843 | } | |
2844 | ||
104f3928 KS |
2845 | static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap) |
2846 | { | |
2847 | u64 idx_val, val; | |
2848 | ||
2849 | if (!mem_cgroup_is_root(mem)) { | |
2850 | if (!swap) | |
2851 | return res_counter_read_u64(&mem->res, RES_USAGE); | |
2852 | else | |
2853 | return res_counter_read_u64(&mem->memsw, RES_USAGE); | |
2854 | } | |
2855 | ||
2856 | mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE, &idx_val); | |
2857 | val = idx_val; | |
2858 | mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS, &idx_val); | |
2859 | val += idx_val; | |
2860 | ||
2861 | if (swap) { | |
2862 | mem_cgroup_get_recursive_idx_stat(mem, | |
2863 | MEM_CGROUP_STAT_SWAPOUT, &idx_val); | |
2864 | val += idx_val; | |
2865 | } | |
2866 | ||
2867 | return val << PAGE_SHIFT; | |
2868 | } | |
2869 | ||
2c3daa72 | 2870 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
8cdea7c0 | 2871 | { |
8c7c6e34 | 2872 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
104f3928 | 2873 | u64 val; |
8c7c6e34 KH |
2874 | int type, name; |
2875 | ||
2876 | type = MEMFILE_TYPE(cft->private); | |
2877 | name = MEMFILE_ATTR(cft->private); | |
2878 | switch (type) { | |
2879 | case _MEM: | |
104f3928 KS |
2880 | if (name == RES_USAGE) |
2881 | val = mem_cgroup_usage(mem, false); | |
2882 | else | |
0c3e73e8 | 2883 | val = res_counter_read_u64(&mem->res, name); |
8c7c6e34 KH |
2884 | break; |
2885 | case _MEMSWAP: | |
104f3928 KS |
2886 | if (name == RES_USAGE) |
2887 | val = mem_cgroup_usage(mem, true); | |
2888 | else | |
0c3e73e8 | 2889 | val = res_counter_read_u64(&mem->memsw, name); |
8c7c6e34 KH |
2890 | break; |
2891 | default: | |
2892 | BUG(); | |
2893 | break; | |
2894 | } | |
2895 | return val; | |
8cdea7c0 | 2896 | } |
628f4235 KH |
2897 | /* |
2898 | * The user of this function is... | |
2899 | * RES_LIMIT. | |
2900 | */ | |
856c13aa PM |
2901 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
2902 | const char *buffer) | |
8cdea7c0 | 2903 | { |
628f4235 | 2904 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
8c7c6e34 | 2905 | int type, name; |
628f4235 KH |
2906 | unsigned long long val; |
2907 | int ret; | |
2908 | ||
8c7c6e34 KH |
2909 | type = MEMFILE_TYPE(cft->private); |
2910 | name = MEMFILE_ATTR(cft->private); | |
2911 | switch (name) { | |
628f4235 | 2912 | case RES_LIMIT: |
4b3bde4c BS |
2913 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2914 | ret = -EINVAL; | |
2915 | break; | |
2916 | } | |
628f4235 KH |
2917 | /* This function does all necessary parse...reuse it */ |
2918 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
8c7c6e34 KH |
2919 | if (ret) |
2920 | break; | |
2921 | if (type == _MEM) | |
628f4235 | 2922 | ret = mem_cgroup_resize_limit(memcg, val); |
8c7c6e34 KH |
2923 | else |
2924 | ret = mem_cgroup_resize_memsw_limit(memcg, val); | |
628f4235 | 2925 | break; |
296c81d8 BS |
2926 | case RES_SOFT_LIMIT: |
2927 | ret = res_counter_memparse_write_strategy(buffer, &val); | |
2928 | if (ret) | |
2929 | break; | |
2930 | /* | |
2931 | * For memsw, soft limits are hard to implement in terms | |
2932 | * of semantics, for now, we support soft limits for | |
2933 | * control without swap | |
2934 | */ | |
2935 | if (type == _MEM) | |
2936 | ret = res_counter_set_soft_limit(&memcg->res, val); | |
2937 | else | |
2938 | ret = -EINVAL; | |
2939 | break; | |
628f4235 KH |
2940 | default: |
2941 | ret = -EINVAL; /* should be BUG() ? */ | |
2942 | break; | |
2943 | } | |
2944 | return ret; | |
8cdea7c0 BS |
2945 | } |
2946 | ||
fee7b548 KH |
2947 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
2948 | unsigned long long *mem_limit, unsigned long long *memsw_limit) | |
2949 | { | |
2950 | struct cgroup *cgroup; | |
2951 | unsigned long long min_limit, min_memsw_limit, tmp; | |
2952 | ||
2953 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2954 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2955 | cgroup = memcg->css.cgroup; | |
2956 | if (!memcg->use_hierarchy) | |
2957 | goto out; | |
2958 | ||
2959 | while (cgroup->parent) { | |
2960 | cgroup = cgroup->parent; | |
2961 | memcg = mem_cgroup_from_cont(cgroup); | |
2962 | if (!memcg->use_hierarchy) | |
2963 | break; | |
2964 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); | |
2965 | min_limit = min(min_limit, tmp); | |
2966 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); | |
2967 | min_memsw_limit = min(min_memsw_limit, tmp); | |
2968 | } | |
2969 | out: | |
2970 | *mem_limit = min_limit; | |
2971 | *memsw_limit = min_memsw_limit; | |
2972 | return; | |
2973 | } | |
2974 | ||
29f2a4da | 2975 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
c84872e1 PE |
2976 | { |
2977 | struct mem_cgroup *mem; | |
8c7c6e34 | 2978 | int type, name; |
c84872e1 PE |
2979 | |
2980 | mem = mem_cgroup_from_cont(cont); | |
8c7c6e34 KH |
2981 | type = MEMFILE_TYPE(event); |
2982 | name = MEMFILE_ATTR(event); | |
2983 | switch (name) { | |
29f2a4da | 2984 | case RES_MAX_USAGE: |
8c7c6e34 KH |
2985 | if (type == _MEM) |
2986 | res_counter_reset_max(&mem->res); | |
2987 | else | |
2988 | res_counter_reset_max(&mem->memsw); | |
29f2a4da PE |
2989 | break; |
2990 | case RES_FAILCNT: | |
8c7c6e34 KH |
2991 | if (type == _MEM) |
2992 | res_counter_reset_failcnt(&mem->res); | |
2993 | else | |
2994 | res_counter_reset_failcnt(&mem->memsw); | |
29f2a4da PE |
2995 | break; |
2996 | } | |
f64c3f54 | 2997 | |
85cc59db | 2998 | return 0; |
c84872e1 PE |
2999 | } |
3000 | ||
7dc74be0 DN |
3001 | static u64 mem_cgroup_move_charge_read(struct cgroup *cgrp, |
3002 | struct cftype *cft) | |
3003 | { | |
3004 | return mem_cgroup_from_cont(cgrp)->move_charge_at_immigrate; | |
3005 | } | |
3006 | ||
02491447 | 3007 | #ifdef CONFIG_MMU |
7dc74be0 DN |
3008 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, |
3009 | struct cftype *cft, u64 val) | |
3010 | { | |
3011 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgrp); | |
3012 | ||
3013 | if (val >= (1 << NR_MOVE_TYPE)) | |
3014 | return -EINVAL; | |
3015 | /* | |
3016 | * We check this value several times in both in can_attach() and | |
3017 | * attach(), so we need cgroup lock to prevent this value from being | |
3018 | * inconsistent. | |
3019 | */ | |
3020 | cgroup_lock(); | |
3021 | mem->move_charge_at_immigrate = val; | |
3022 | cgroup_unlock(); | |
3023 | ||
3024 | return 0; | |
3025 | } | |
02491447 DN |
3026 | #else |
3027 | static int mem_cgroup_move_charge_write(struct cgroup *cgrp, | |
3028 | struct cftype *cft, u64 val) | |
3029 | { | |
3030 | return -ENOSYS; | |
3031 | } | |
3032 | #endif | |
7dc74be0 | 3033 | |
14067bb3 KH |
3034 | |
3035 | /* For read statistics */ | |
3036 | enum { | |
3037 | MCS_CACHE, | |
3038 | MCS_RSS, | |
d8046582 | 3039 | MCS_FILE_MAPPED, |
14067bb3 KH |
3040 | MCS_PGPGIN, |
3041 | MCS_PGPGOUT, | |
1dd3a273 | 3042 | MCS_SWAP, |
14067bb3 KH |
3043 | MCS_INACTIVE_ANON, |
3044 | MCS_ACTIVE_ANON, | |
3045 | MCS_INACTIVE_FILE, | |
3046 | MCS_ACTIVE_FILE, | |
3047 | MCS_UNEVICTABLE, | |
3048 | NR_MCS_STAT, | |
3049 | }; | |
3050 | ||
3051 | struct mcs_total_stat { | |
3052 | s64 stat[NR_MCS_STAT]; | |
d2ceb9b7 KH |
3053 | }; |
3054 | ||
14067bb3 KH |
3055 | struct { |
3056 | char *local_name; | |
3057 | char *total_name; | |
3058 | } memcg_stat_strings[NR_MCS_STAT] = { | |
3059 | {"cache", "total_cache"}, | |
3060 | {"rss", "total_rss"}, | |
d69b042f | 3061 | {"mapped_file", "total_mapped_file"}, |
14067bb3 KH |
3062 | {"pgpgin", "total_pgpgin"}, |
3063 | {"pgpgout", "total_pgpgout"}, | |
1dd3a273 | 3064 | {"swap", "total_swap"}, |
14067bb3 KH |
3065 | {"inactive_anon", "total_inactive_anon"}, |
3066 | {"active_anon", "total_active_anon"}, | |
3067 | {"inactive_file", "total_inactive_file"}, | |
3068 | {"active_file", "total_active_file"}, | |
3069 | {"unevictable", "total_unevictable"} | |
3070 | }; | |
3071 | ||
3072 | ||
3073 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) | |
3074 | { | |
3075 | struct mcs_total_stat *s = data; | |
3076 | s64 val; | |
3077 | ||
3078 | /* per cpu stat */ | |
c62b1a3b | 3079 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE); |
14067bb3 | 3080 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
c62b1a3b | 3081 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS); |
14067bb3 | 3082 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
c62b1a3b | 3083 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_FILE_MAPPED); |
d8046582 | 3084 | s->stat[MCS_FILE_MAPPED] += val * PAGE_SIZE; |
c62b1a3b | 3085 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGIN_COUNT); |
14067bb3 | 3086 | s->stat[MCS_PGPGIN] += val; |
c62b1a3b | 3087 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
14067bb3 | 3088 | s->stat[MCS_PGPGOUT] += val; |
1dd3a273 | 3089 | if (do_swap_account) { |
c62b1a3b | 3090 | val = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_SWAPOUT); |
1dd3a273 DN |
3091 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
3092 | } | |
14067bb3 KH |
3093 | |
3094 | /* per zone stat */ | |
3095 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); | |
3096 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; | |
3097 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); | |
3098 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; | |
3099 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); | |
3100 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; | |
3101 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); | |
3102 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; | |
3103 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); | |
3104 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; | |
3105 | return 0; | |
3106 | } | |
3107 | ||
3108 | static void | |
3109 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) | |
3110 | { | |
3111 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); | |
3112 | } | |
3113 | ||
c64745cf PM |
3114 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
3115 | struct cgroup_map_cb *cb) | |
d2ceb9b7 | 3116 | { |
d2ceb9b7 | 3117 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
14067bb3 | 3118 | struct mcs_total_stat mystat; |
d2ceb9b7 KH |
3119 | int i; |
3120 | ||
14067bb3 KH |
3121 | memset(&mystat, 0, sizeof(mystat)); |
3122 | mem_cgroup_get_local_stat(mem_cont, &mystat); | |
d2ceb9b7 | 3123 | |
1dd3a273 DN |
3124 | for (i = 0; i < NR_MCS_STAT; i++) { |
3125 | if (i == MCS_SWAP && !do_swap_account) | |
3126 | continue; | |
14067bb3 | 3127 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
1dd3a273 | 3128 | } |
7b854121 | 3129 | |
14067bb3 | 3130 | /* Hierarchical information */ |
fee7b548 KH |
3131 | { |
3132 | unsigned long long limit, memsw_limit; | |
3133 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); | |
3134 | cb->fill(cb, "hierarchical_memory_limit", limit); | |
3135 | if (do_swap_account) | |
3136 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); | |
3137 | } | |
7f016ee8 | 3138 | |
14067bb3 KH |
3139 | memset(&mystat, 0, sizeof(mystat)); |
3140 | mem_cgroup_get_total_stat(mem_cont, &mystat); | |
1dd3a273 DN |
3141 | for (i = 0; i < NR_MCS_STAT; i++) { |
3142 | if (i == MCS_SWAP && !do_swap_account) | |
3143 | continue; | |
14067bb3 | 3144 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
1dd3a273 | 3145 | } |
14067bb3 | 3146 | |
7f016ee8 | 3147 | #ifdef CONFIG_DEBUG_VM |
c772be93 | 3148 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
7f016ee8 KM |
3149 | |
3150 | { | |
3151 | int nid, zid; | |
3152 | struct mem_cgroup_per_zone *mz; | |
3153 | unsigned long recent_rotated[2] = {0, 0}; | |
3154 | unsigned long recent_scanned[2] = {0, 0}; | |
3155 | ||
3156 | for_each_online_node(nid) | |
3157 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { | |
3158 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); | |
3159 | ||
3160 | recent_rotated[0] += | |
3161 | mz->reclaim_stat.recent_rotated[0]; | |
3162 | recent_rotated[1] += | |
3163 | mz->reclaim_stat.recent_rotated[1]; | |
3164 | recent_scanned[0] += | |
3165 | mz->reclaim_stat.recent_scanned[0]; | |
3166 | recent_scanned[1] += | |
3167 | mz->reclaim_stat.recent_scanned[1]; | |
3168 | } | |
3169 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); | |
3170 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); | |
3171 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); | |
3172 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); | |
3173 | } | |
3174 | #endif | |
3175 | ||
d2ceb9b7 KH |
3176 | return 0; |
3177 | } | |
3178 | ||
a7885eb8 KM |
3179 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
3180 | { | |
3181 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3182 | ||
3183 | return get_swappiness(memcg); | |
3184 | } | |
3185 | ||
3186 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, | |
3187 | u64 val) | |
3188 | { | |
3189 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3190 | struct mem_cgroup *parent; | |
068b38c1 | 3191 | |
a7885eb8 KM |
3192 | if (val > 100) |
3193 | return -EINVAL; | |
3194 | ||
3195 | if (cgrp->parent == NULL) | |
3196 | return -EINVAL; | |
3197 | ||
3198 | parent = mem_cgroup_from_cont(cgrp->parent); | |
068b38c1 LZ |
3199 | |
3200 | cgroup_lock(); | |
3201 | ||
a7885eb8 KM |
3202 | /* If under hierarchy, only empty-root can set this value */ |
3203 | if ((parent->use_hierarchy) || | |
068b38c1 LZ |
3204 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
3205 | cgroup_unlock(); | |
a7885eb8 | 3206 | return -EINVAL; |
068b38c1 | 3207 | } |
a7885eb8 KM |
3208 | |
3209 | spin_lock(&memcg->reclaim_param_lock); | |
3210 | memcg->swappiness = val; | |
3211 | spin_unlock(&memcg->reclaim_param_lock); | |
3212 | ||
068b38c1 LZ |
3213 | cgroup_unlock(); |
3214 | ||
a7885eb8 KM |
3215 | return 0; |
3216 | } | |
3217 | ||
2e72b634 KS |
3218 | static bool mem_cgroup_threshold_check(struct mem_cgroup *mem) |
3219 | { | |
3220 | bool ret = false; | |
2e72b634 | 3221 | s64 val; |
2e72b634 | 3222 | |
c62b1a3b | 3223 | val = this_cpu_read(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS]); |
2e72b634 | 3224 | if (unlikely(val < 0)) { |
c62b1a3b | 3225 | this_cpu_write(mem->stat->count[MEM_CGROUP_STAT_THRESHOLDS], |
2e72b634 KS |
3226 | THRESHOLDS_EVENTS_THRESH); |
3227 | ret = true; | |
3228 | } | |
2e72b634 KS |
3229 | return ret; |
3230 | } | |
3231 | ||
3232 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) | |
3233 | { | |
3234 | struct mem_cgroup_threshold_ary *t; | |
3235 | u64 usage; | |
3236 | int i; | |
3237 | ||
3238 | rcu_read_lock(); | |
3239 | if (!swap) | |
3240 | t = rcu_dereference(memcg->thresholds); | |
3241 | else | |
3242 | t = rcu_dereference(memcg->memsw_thresholds); | |
3243 | ||
3244 | if (!t) | |
3245 | goto unlock; | |
3246 | ||
3247 | usage = mem_cgroup_usage(memcg, swap); | |
3248 | ||
3249 | /* | |
3250 | * current_threshold points to threshold just below usage. | |
3251 | * If it's not true, a threshold was crossed after last | |
3252 | * call of __mem_cgroup_threshold(). | |
3253 | */ | |
3254 | i = atomic_read(&t->current_threshold); | |
3255 | ||
3256 | /* | |
3257 | * Iterate backward over array of thresholds starting from | |
3258 | * current_threshold and check if a threshold is crossed. | |
3259 | * If none of thresholds below usage is crossed, we read | |
3260 | * only one element of the array here. | |
3261 | */ | |
3262 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) | |
3263 | eventfd_signal(t->entries[i].eventfd, 1); | |
3264 | ||
3265 | /* i = current_threshold + 1 */ | |
3266 | i++; | |
3267 | ||
3268 | /* | |
3269 | * Iterate forward over array of thresholds starting from | |
3270 | * current_threshold+1 and check if a threshold is crossed. | |
3271 | * If none of thresholds above usage is crossed, we read | |
3272 | * only one element of the array here. | |
3273 | */ | |
3274 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) | |
3275 | eventfd_signal(t->entries[i].eventfd, 1); | |
3276 | ||
3277 | /* Update current_threshold */ | |
3278 | atomic_set(&t->current_threshold, i - 1); | |
3279 | unlock: | |
3280 | rcu_read_unlock(); | |
3281 | } | |
3282 | ||
3283 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) | |
3284 | { | |
3285 | __mem_cgroup_threshold(memcg, false); | |
3286 | if (do_swap_account) | |
3287 | __mem_cgroup_threshold(memcg, true); | |
3288 | } | |
3289 | ||
3290 | static int compare_thresholds(const void *a, const void *b) | |
3291 | { | |
3292 | const struct mem_cgroup_threshold *_a = a; | |
3293 | const struct mem_cgroup_threshold *_b = b; | |
3294 | ||
3295 | return _a->threshold - _b->threshold; | |
3296 | } | |
3297 | ||
3298 | static int mem_cgroup_register_event(struct cgroup *cgrp, struct cftype *cft, | |
3299 | struct eventfd_ctx *eventfd, const char *args) | |
3300 | { | |
3301 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3302 | struct mem_cgroup_threshold_ary *thresholds, *thresholds_new; | |
3303 | int type = MEMFILE_TYPE(cft->private); | |
3304 | u64 threshold, usage; | |
3305 | int size; | |
3306 | int i, ret; | |
3307 | ||
3308 | ret = res_counter_memparse_write_strategy(args, &threshold); | |
3309 | if (ret) | |
3310 | return ret; | |
3311 | ||
3312 | mutex_lock(&memcg->thresholds_lock); | |
3313 | if (type == _MEM) | |
3314 | thresholds = memcg->thresholds; | |
3315 | else if (type == _MEMSWAP) | |
3316 | thresholds = memcg->memsw_thresholds; | |
3317 | else | |
3318 | BUG(); | |
3319 | ||
3320 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3321 | ||
3322 | /* Check if a threshold crossed before adding a new one */ | |
3323 | if (thresholds) | |
3324 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3325 | ||
3326 | if (thresholds) | |
3327 | size = thresholds->size + 1; | |
3328 | else | |
3329 | size = 1; | |
3330 | ||
3331 | /* Allocate memory for new array of thresholds */ | |
3332 | thresholds_new = kmalloc(sizeof(*thresholds_new) + | |
3333 | size * sizeof(struct mem_cgroup_threshold), | |
3334 | GFP_KERNEL); | |
3335 | if (!thresholds_new) { | |
3336 | ret = -ENOMEM; | |
3337 | goto unlock; | |
3338 | } | |
3339 | thresholds_new->size = size; | |
3340 | ||
3341 | /* Copy thresholds (if any) to new array */ | |
3342 | if (thresholds) | |
3343 | memcpy(thresholds_new->entries, thresholds->entries, | |
3344 | thresholds->size * | |
3345 | sizeof(struct mem_cgroup_threshold)); | |
3346 | /* Add new threshold */ | |
3347 | thresholds_new->entries[size - 1].eventfd = eventfd; | |
3348 | thresholds_new->entries[size - 1].threshold = threshold; | |
3349 | ||
3350 | /* Sort thresholds. Registering of new threshold isn't time-critical */ | |
3351 | sort(thresholds_new->entries, size, | |
3352 | sizeof(struct mem_cgroup_threshold), | |
3353 | compare_thresholds, NULL); | |
3354 | ||
3355 | /* Find current threshold */ | |
3356 | atomic_set(&thresholds_new->current_threshold, -1); | |
3357 | for (i = 0; i < size; i++) { | |
3358 | if (thresholds_new->entries[i].threshold < usage) { | |
3359 | /* | |
3360 | * thresholds_new->current_threshold will not be used | |
3361 | * until rcu_assign_pointer(), so it's safe to increment | |
3362 | * it here. | |
3363 | */ | |
3364 | atomic_inc(&thresholds_new->current_threshold); | |
3365 | } | |
3366 | } | |
3367 | ||
3368 | /* | |
3369 | * We need to increment refcnt to be sure that all thresholds | |
3370 | * will be unregistered before calling __mem_cgroup_free() | |
3371 | */ | |
3372 | mem_cgroup_get(memcg); | |
3373 | ||
3374 | if (type == _MEM) | |
3375 | rcu_assign_pointer(memcg->thresholds, thresholds_new); | |
3376 | else | |
3377 | rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new); | |
3378 | ||
3379 | /* To be sure that nobody uses thresholds before freeing it */ | |
3380 | synchronize_rcu(); | |
3381 | ||
3382 | kfree(thresholds); | |
3383 | unlock: | |
3384 | mutex_unlock(&memcg->thresholds_lock); | |
3385 | ||
3386 | return ret; | |
3387 | } | |
3388 | ||
3389 | static int mem_cgroup_unregister_event(struct cgroup *cgrp, struct cftype *cft, | |
3390 | struct eventfd_ctx *eventfd) | |
3391 | { | |
3392 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); | |
3393 | struct mem_cgroup_threshold_ary *thresholds, *thresholds_new; | |
3394 | int type = MEMFILE_TYPE(cft->private); | |
3395 | u64 usage; | |
3396 | int size = 0; | |
3397 | int i, j, ret; | |
3398 | ||
3399 | mutex_lock(&memcg->thresholds_lock); | |
3400 | if (type == _MEM) | |
3401 | thresholds = memcg->thresholds; | |
3402 | else if (type == _MEMSWAP) | |
3403 | thresholds = memcg->memsw_thresholds; | |
3404 | else | |
3405 | BUG(); | |
3406 | ||
3407 | /* | |
3408 | * Something went wrong if we trying to unregister a threshold | |
3409 | * if we don't have thresholds | |
3410 | */ | |
3411 | BUG_ON(!thresholds); | |
3412 | ||
3413 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); | |
3414 | ||
3415 | /* Check if a threshold crossed before removing */ | |
3416 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); | |
3417 | ||
3418 | /* Calculate new number of threshold */ | |
3419 | for (i = 0; i < thresholds->size; i++) { | |
3420 | if (thresholds->entries[i].eventfd != eventfd) | |
3421 | size++; | |
3422 | } | |
3423 | ||
3424 | /* Set thresholds array to NULL if we don't have thresholds */ | |
3425 | if (!size) { | |
3426 | thresholds_new = NULL; | |
3427 | goto assign; | |
3428 | } | |
3429 | ||
3430 | /* Allocate memory for new array of thresholds */ | |
3431 | thresholds_new = kmalloc(sizeof(*thresholds_new) + | |
3432 | size * sizeof(struct mem_cgroup_threshold), | |
3433 | GFP_KERNEL); | |
3434 | if (!thresholds_new) { | |
3435 | ret = -ENOMEM; | |
3436 | goto unlock; | |
3437 | } | |
3438 | thresholds_new->size = size; | |
3439 | ||
3440 | /* Copy thresholds and find current threshold */ | |
3441 | atomic_set(&thresholds_new->current_threshold, -1); | |
3442 | for (i = 0, j = 0; i < thresholds->size; i++) { | |
3443 | if (thresholds->entries[i].eventfd == eventfd) | |
3444 | continue; | |
3445 | ||
3446 | thresholds_new->entries[j] = thresholds->entries[i]; | |
3447 | if (thresholds_new->entries[j].threshold < usage) { | |
3448 | /* | |
3449 | * thresholds_new->current_threshold will not be used | |
3450 | * until rcu_assign_pointer(), so it's safe to increment | |
3451 | * it here. | |
3452 | */ | |
3453 | atomic_inc(&thresholds_new->current_threshold); | |
3454 | } | |
3455 | j++; | |
3456 | } | |
3457 | ||
3458 | assign: | |
3459 | if (type == _MEM) | |
3460 | rcu_assign_pointer(memcg->thresholds, thresholds_new); | |
3461 | else | |
3462 | rcu_assign_pointer(memcg->memsw_thresholds, thresholds_new); | |
3463 | ||
3464 | /* To be sure that nobody uses thresholds before freeing it */ | |
3465 | synchronize_rcu(); | |
3466 | ||
3467 | for (i = 0; i < thresholds->size - size; i++) | |
3468 | mem_cgroup_put(memcg); | |
3469 | ||
3470 | kfree(thresholds); | |
3471 | unlock: | |
3472 | mutex_unlock(&memcg->thresholds_lock); | |
3473 | ||
3474 | return ret; | |
3475 | } | |
c1e862c1 | 3476 | |
8cdea7c0 BS |
3477 | static struct cftype mem_cgroup_files[] = { |
3478 | { | |
0eea1030 | 3479 | .name = "usage_in_bytes", |
8c7c6e34 | 3480 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2c3daa72 | 3481 | .read_u64 = mem_cgroup_read, |
2e72b634 KS |
3482 | .register_event = mem_cgroup_register_event, |
3483 | .unregister_event = mem_cgroup_unregister_event, | |
8cdea7c0 | 3484 | }, |
c84872e1 PE |
3485 | { |
3486 | .name = "max_usage_in_bytes", | |
8c7c6e34 | 3487 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
29f2a4da | 3488 | .trigger = mem_cgroup_reset, |
c84872e1 PE |
3489 | .read_u64 = mem_cgroup_read, |
3490 | }, | |
8cdea7c0 | 3491 | { |
0eea1030 | 3492 | .name = "limit_in_bytes", |
8c7c6e34 | 3493 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
856c13aa | 3494 | .write_string = mem_cgroup_write, |
2c3daa72 | 3495 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3496 | }, |
296c81d8 BS |
3497 | { |
3498 | .name = "soft_limit_in_bytes", | |
3499 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), | |
3500 | .write_string = mem_cgroup_write, | |
3501 | .read_u64 = mem_cgroup_read, | |
3502 | }, | |
8cdea7c0 BS |
3503 | { |
3504 | .name = "failcnt", | |
8c7c6e34 | 3505 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
29f2a4da | 3506 | .trigger = mem_cgroup_reset, |
2c3daa72 | 3507 | .read_u64 = mem_cgroup_read, |
8cdea7c0 | 3508 | }, |
d2ceb9b7 KH |
3509 | { |
3510 | .name = "stat", | |
c64745cf | 3511 | .read_map = mem_control_stat_show, |
d2ceb9b7 | 3512 | }, |
c1e862c1 KH |
3513 | { |
3514 | .name = "force_empty", | |
3515 | .trigger = mem_cgroup_force_empty_write, | |
3516 | }, | |
18f59ea7 BS |
3517 | { |
3518 | .name = "use_hierarchy", | |
3519 | .write_u64 = mem_cgroup_hierarchy_write, | |
3520 | .read_u64 = mem_cgroup_hierarchy_read, | |
3521 | }, | |
a7885eb8 KM |
3522 | { |
3523 | .name = "swappiness", | |
3524 | .read_u64 = mem_cgroup_swappiness_read, | |
3525 | .write_u64 = mem_cgroup_swappiness_write, | |
3526 | }, | |
7dc74be0 DN |
3527 | { |
3528 | .name = "move_charge_at_immigrate", | |
3529 | .read_u64 = mem_cgroup_move_charge_read, | |
3530 | .write_u64 = mem_cgroup_move_charge_write, | |
3531 | }, | |
8cdea7c0 BS |
3532 | }; |
3533 | ||
8c7c6e34 KH |
3534 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3535 | static struct cftype memsw_cgroup_files[] = { | |
3536 | { | |
3537 | .name = "memsw.usage_in_bytes", | |
3538 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), | |
3539 | .read_u64 = mem_cgroup_read, | |
2e72b634 KS |
3540 | .register_event = mem_cgroup_register_event, |
3541 | .unregister_event = mem_cgroup_unregister_event, | |
8c7c6e34 KH |
3542 | }, |
3543 | { | |
3544 | .name = "memsw.max_usage_in_bytes", | |
3545 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), | |
3546 | .trigger = mem_cgroup_reset, | |
3547 | .read_u64 = mem_cgroup_read, | |
3548 | }, | |
3549 | { | |
3550 | .name = "memsw.limit_in_bytes", | |
3551 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), | |
3552 | .write_string = mem_cgroup_write, | |
3553 | .read_u64 = mem_cgroup_read, | |
3554 | }, | |
3555 | { | |
3556 | .name = "memsw.failcnt", | |
3557 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), | |
3558 | .trigger = mem_cgroup_reset, | |
3559 | .read_u64 = mem_cgroup_read, | |
3560 | }, | |
3561 | }; | |
3562 | ||
3563 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3564 | { | |
3565 | if (!do_swap_account) | |
3566 | return 0; | |
3567 | return cgroup_add_files(cont, ss, memsw_cgroup_files, | |
3568 | ARRAY_SIZE(memsw_cgroup_files)); | |
3569 | }; | |
3570 | #else | |
3571 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) | |
3572 | { | |
3573 | return 0; | |
3574 | } | |
3575 | #endif | |
3576 | ||
6d12e2d8 KH |
3577 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3578 | { | |
3579 | struct mem_cgroup_per_node *pn; | |
1ecaab2b | 3580 | struct mem_cgroup_per_zone *mz; |
b69408e8 | 3581 | enum lru_list l; |
41e3355d | 3582 | int zone, tmp = node; |
1ecaab2b KH |
3583 | /* |
3584 | * This routine is called against possible nodes. | |
3585 | * But it's BUG to call kmalloc() against offline node. | |
3586 | * | |
3587 | * TODO: this routine can waste much memory for nodes which will | |
3588 | * never be onlined. It's better to use memory hotplug callback | |
3589 | * function. | |
3590 | */ | |
41e3355d KH |
3591 | if (!node_state(node, N_NORMAL_MEMORY)) |
3592 | tmp = -1; | |
3593 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); | |
6d12e2d8 KH |
3594 | if (!pn) |
3595 | return 1; | |
1ecaab2b | 3596 | |
6d12e2d8 KH |
3597 | mem->info.nodeinfo[node] = pn; |
3598 | memset(pn, 0, sizeof(*pn)); | |
1ecaab2b KH |
3599 | |
3600 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3601 | mz = &pn->zoneinfo[zone]; | |
b69408e8 CL |
3602 | for_each_lru(l) |
3603 | INIT_LIST_HEAD(&mz->lists[l]); | |
f64c3f54 | 3604 | mz->usage_in_excess = 0; |
4e416953 BS |
3605 | mz->on_tree = false; |
3606 | mz->mem = mem; | |
1ecaab2b | 3607 | } |
6d12e2d8 KH |
3608 | return 0; |
3609 | } | |
3610 | ||
1ecaab2b KH |
3611 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
3612 | { | |
3613 | kfree(mem->info.nodeinfo[node]); | |
3614 | } | |
3615 | ||
33327948 KH |
3616 | static struct mem_cgroup *mem_cgroup_alloc(void) |
3617 | { | |
3618 | struct mem_cgroup *mem; | |
c62b1a3b | 3619 | int size = sizeof(struct mem_cgroup); |
33327948 | 3620 | |
c62b1a3b | 3621 | /* Can be very big if MAX_NUMNODES is very big */ |
c8dad2bb JB |
3622 | if (size < PAGE_SIZE) |
3623 | mem = kmalloc(size, GFP_KERNEL); | |
33327948 | 3624 | else |
c8dad2bb | 3625 | mem = vmalloc(size); |
33327948 KH |
3626 | |
3627 | if (mem) | |
c8dad2bb | 3628 | memset(mem, 0, size); |
c62b1a3b KH |
3629 | mem->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
3630 | if (!mem->stat) { | |
3631 | if (size < PAGE_SIZE) | |
3632 | kfree(mem); | |
3633 | else | |
3634 | vfree(mem); | |
3635 | mem = NULL; | |
3636 | } | |
33327948 KH |
3637 | return mem; |
3638 | } | |
3639 | ||
8c7c6e34 KH |
3640 | /* |
3641 | * At destroying mem_cgroup, references from swap_cgroup can remain. | |
3642 | * (scanning all at force_empty is too costly...) | |
3643 | * | |
3644 | * Instead of clearing all references at force_empty, we remember | |
3645 | * the number of reference from swap_cgroup and free mem_cgroup when | |
3646 | * it goes down to 0. | |
3647 | * | |
8c7c6e34 KH |
3648 | * Removal of cgroup itself succeeds regardless of refs from swap. |
3649 | */ | |
3650 | ||
a7ba0eef | 3651 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
33327948 | 3652 | { |
08e552c6 KH |
3653 | int node; |
3654 | ||
f64c3f54 | 3655 | mem_cgroup_remove_from_trees(mem); |
04046e1a KH |
3656 | free_css_id(&mem_cgroup_subsys, &mem->css); |
3657 | ||
08e552c6 KH |
3658 | for_each_node_state(node, N_POSSIBLE) |
3659 | free_mem_cgroup_per_zone_info(mem, node); | |
3660 | ||
c62b1a3b KH |
3661 | free_percpu(mem->stat); |
3662 | if (sizeof(struct mem_cgroup) < PAGE_SIZE) | |
33327948 KH |
3663 | kfree(mem); |
3664 | else | |
3665 | vfree(mem); | |
3666 | } | |
3667 | ||
8c7c6e34 KH |
3668 | static void mem_cgroup_get(struct mem_cgroup *mem) |
3669 | { | |
3670 | atomic_inc(&mem->refcnt); | |
3671 | } | |
3672 | ||
483c30b5 | 3673 | static void __mem_cgroup_put(struct mem_cgroup *mem, int count) |
8c7c6e34 | 3674 | { |
483c30b5 | 3675 | if (atomic_sub_and_test(count, &mem->refcnt)) { |
7bcc1bb1 | 3676 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
a7ba0eef | 3677 | __mem_cgroup_free(mem); |
7bcc1bb1 DN |
3678 | if (parent) |
3679 | mem_cgroup_put(parent); | |
3680 | } | |
8c7c6e34 KH |
3681 | } |
3682 | ||
483c30b5 DN |
3683 | static void mem_cgroup_put(struct mem_cgroup *mem) |
3684 | { | |
3685 | __mem_cgroup_put(mem, 1); | |
3686 | } | |
3687 | ||
7bcc1bb1 DN |
3688 | /* |
3689 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. | |
3690 | */ | |
3691 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) | |
3692 | { | |
3693 | if (!mem->res.parent) | |
3694 | return NULL; | |
3695 | return mem_cgroup_from_res_counter(mem->res.parent, res); | |
3696 | } | |
33327948 | 3697 | |
c077719b KH |
3698 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3699 | static void __init enable_swap_cgroup(void) | |
3700 | { | |
f8d66542 | 3701 | if (!mem_cgroup_disabled() && really_do_swap_account) |
c077719b KH |
3702 | do_swap_account = 1; |
3703 | } | |
3704 | #else | |
3705 | static void __init enable_swap_cgroup(void) | |
3706 | { | |
3707 | } | |
3708 | #endif | |
3709 | ||
f64c3f54 BS |
3710 | static int mem_cgroup_soft_limit_tree_init(void) |
3711 | { | |
3712 | struct mem_cgroup_tree_per_node *rtpn; | |
3713 | struct mem_cgroup_tree_per_zone *rtpz; | |
3714 | int tmp, node, zone; | |
3715 | ||
3716 | for_each_node_state(node, N_POSSIBLE) { | |
3717 | tmp = node; | |
3718 | if (!node_state(node, N_NORMAL_MEMORY)) | |
3719 | tmp = -1; | |
3720 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); | |
3721 | if (!rtpn) | |
3722 | return 1; | |
3723 | ||
3724 | soft_limit_tree.rb_tree_per_node[node] = rtpn; | |
3725 | ||
3726 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { | |
3727 | rtpz = &rtpn->rb_tree_per_zone[zone]; | |
3728 | rtpz->rb_root = RB_ROOT; | |
3729 | spin_lock_init(&rtpz->lock); | |
3730 | } | |
3731 | } | |
3732 | return 0; | |
3733 | } | |
3734 | ||
0eb253e2 | 3735 | static struct cgroup_subsys_state * __ref |
8cdea7c0 BS |
3736 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
3737 | { | |
28dbc4b6 | 3738 | struct mem_cgroup *mem, *parent; |
04046e1a | 3739 | long error = -ENOMEM; |
6d12e2d8 | 3740 | int node; |
8cdea7c0 | 3741 | |
c8dad2bb JB |
3742 | mem = mem_cgroup_alloc(); |
3743 | if (!mem) | |
04046e1a | 3744 | return ERR_PTR(error); |
78fb7466 | 3745 | |
6d12e2d8 KH |
3746 | for_each_node_state(node, N_POSSIBLE) |
3747 | if (alloc_mem_cgroup_per_zone_info(mem, node)) | |
3748 | goto free_out; | |
f64c3f54 | 3749 | |
c077719b | 3750 | /* root ? */ |
28dbc4b6 | 3751 | if (cont->parent == NULL) { |
cdec2e42 | 3752 | int cpu; |
c077719b | 3753 | enable_swap_cgroup(); |
28dbc4b6 | 3754 | parent = NULL; |
4b3bde4c | 3755 | root_mem_cgroup = mem; |
f64c3f54 BS |
3756 | if (mem_cgroup_soft_limit_tree_init()) |
3757 | goto free_out; | |
cdec2e42 KH |
3758 | for_each_possible_cpu(cpu) { |
3759 | struct memcg_stock_pcp *stock = | |
3760 | &per_cpu(memcg_stock, cpu); | |
3761 | INIT_WORK(&stock->work, drain_local_stock); | |
3762 | } | |
3763 | hotcpu_notifier(memcg_stock_cpu_callback, 0); | |
18f59ea7 | 3764 | } else { |
28dbc4b6 | 3765 | parent = mem_cgroup_from_cont(cont->parent); |
18f59ea7 BS |
3766 | mem->use_hierarchy = parent->use_hierarchy; |
3767 | } | |
28dbc4b6 | 3768 | |
18f59ea7 BS |
3769 | if (parent && parent->use_hierarchy) { |
3770 | res_counter_init(&mem->res, &parent->res); | |
3771 | res_counter_init(&mem->memsw, &parent->memsw); | |
7bcc1bb1 DN |
3772 | /* |
3773 | * We increment refcnt of the parent to ensure that we can | |
3774 | * safely access it on res_counter_charge/uncharge. | |
3775 | * This refcnt will be decremented when freeing this | |
3776 | * mem_cgroup(see mem_cgroup_put). | |
3777 | */ | |
3778 | mem_cgroup_get(parent); | |
18f59ea7 BS |
3779 | } else { |
3780 | res_counter_init(&mem->res, NULL); | |
3781 | res_counter_init(&mem->memsw, NULL); | |
3782 | } | |
04046e1a | 3783 | mem->last_scanned_child = 0; |
2733c06a | 3784 | spin_lock_init(&mem->reclaim_param_lock); |
6d61ef40 | 3785 | |
a7885eb8 KM |
3786 | if (parent) |
3787 | mem->swappiness = get_swappiness(parent); | |
a7ba0eef | 3788 | atomic_set(&mem->refcnt, 1); |
7dc74be0 | 3789 | mem->move_charge_at_immigrate = 0; |
2e72b634 | 3790 | mutex_init(&mem->thresholds_lock); |
8cdea7c0 | 3791 | return &mem->css; |
6d12e2d8 | 3792 | free_out: |
a7ba0eef | 3793 | __mem_cgroup_free(mem); |
4b3bde4c | 3794 | root_mem_cgroup = NULL; |
04046e1a | 3795 | return ERR_PTR(error); |
8cdea7c0 BS |
3796 | } |
3797 | ||
ec64f515 | 3798 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
df878fb0 KH |
3799 | struct cgroup *cont) |
3800 | { | |
3801 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); | |
ec64f515 KH |
3802 | |
3803 | return mem_cgroup_force_empty(mem, false); | |
df878fb0 KH |
3804 | } |
3805 | ||
8cdea7c0 BS |
3806 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
3807 | struct cgroup *cont) | |
3808 | { | |
c268e994 | 3809 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
c268e994 | 3810 | |
c268e994 | 3811 | mem_cgroup_put(mem); |
8cdea7c0 BS |
3812 | } |
3813 | ||
3814 | static int mem_cgroup_populate(struct cgroup_subsys *ss, | |
3815 | struct cgroup *cont) | |
3816 | { | |
8c7c6e34 KH |
3817 | int ret; |
3818 | ||
3819 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, | |
3820 | ARRAY_SIZE(mem_cgroup_files)); | |
3821 | ||
3822 | if (!ret) | |
3823 | ret = register_memsw_files(cont, ss); | |
3824 | return ret; | |
8cdea7c0 BS |
3825 | } |
3826 | ||
02491447 | 3827 | #ifdef CONFIG_MMU |
7dc74be0 | 3828 | /* Handlers for move charge at task migration. */ |
854ffa8d DN |
3829 | #define PRECHARGE_COUNT_AT_ONCE 256 |
3830 | static int mem_cgroup_do_precharge(unsigned long count) | |
7dc74be0 | 3831 | { |
854ffa8d DN |
3832 | int ret = 0; |
3833 | int batch_count = PRECHARGE_COUNT_AT_ONCE; | |
4ffef5fe DN |
3834 | struct mem_cgroup *mem = mc.to; |
3835 | ||
854ffa8d DN |
3836 | if (mem_cgroup_is_root(mem)) { |
3837 | mc.precharge += count; | |
3838 | /* we don't need css_get for root */ | |
3839 | return ret; | |
3840 | } | |
3841 | /* try to charge at once */ | |
3842 | if (count > 1) { | |
3843 | struct res_counter *dummy; | |
3844 | /* | |
3845 | * "mem" cannot be under rmdir() because we've already checked | |
3846 | * by cgroup_lock_live_cgroup() that it is not removed and we | |
3847 | * are still under the same cgroup_mutex. So we can postpone | |
3848 | * css_get(). | |
3849 | */ | |
3850 | if (res_counter_charge(&mem->res, PAGE_SIZE * count, &dummy)) | |
3851 | goto one_by_one; | |
3852 | if (do_swap_account && res_counter_charge(&mem->memsw, | |
3853 | PAGE_SIZE * count, &dummy)) { | |
3854 | res_counter_uncharge(&mem->res, PAGE_SIZE * count); | |
3855 | goto one_by_one; | |
3856 | } | |
3857 | mc.precharge += count; | |
3858 | VM_BUG_ON(test_bit(CSS_ROOT, &mem->css.flags)); | |
3859 | WARN_ON_ONCE(count > INT_MAX); | |
3860 | __css_get(&mem->css, (int)count); | |
3861 | return ret; | |
3862 | } | |
3863 | one_by_one: | |
3864 | /* fall back to one by one charge */ | |
3865 | while (count--) { | |
3866 | if (signal_pending(current)) { | |
3867 | ret = -EINTR; | |
3868 | break; | |
3869 | } | |
3870 | if (!batch_count--) { | |
3871 | batch_count = PRECHARGE_COUNT_AT_ONCE; | |
3872 | cond_resched(); | |
3873 | } | |
430e4863 | 3874 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false); |
854ffa8d DN |
3875 | if (ret || !mem) |
3876 | /* mem_cgroup_clear_mc() will do uncharge later */ | |
3877 | return -ENOMEM; | |
3878 | mc.precharge++; | |
3879 | } | |
4ffef5fe DN |
3880 | return ret; |
3881 | } | |
02491447 DN |
3882 | #else /* !CONFIG_MMU */ |
3883 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, | |
3884 | struct cgroup *cgroup, | |
3885 | struct task_struct *p, | |
3886 | bool threadgroup) | |
3887 | { | |
3888 | return 0; | |
3889 | } | |
3890 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
3891 | struct cgroup *cgroup, | |
3892 | struct task_struct *p, | |
3893 | bool threadgroup) | |
3894 | { | |
3895 | } | |
3896 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, | |
3897 | struct cgroup *cont, | |
3898 | struct cgroup *old_cont, | |
3899 | struct task_struct *p, | |
3900 | bool threadgroup) | |
3901 | { | |
3902 | } | |
3903 | #endif | |
4ffef5fe DN |
3904 | |
3905 | /** | |
3906 | * is_target_pte_for_mc - check a pte whether it is valid for move charge | |
3907 | * @vma: the vma the pte to be checked belongs | |
3908 | * @addr: the address corresponding to the pte to be checked | |
3909 | * @ptent: the pte to be checked | |
02491447 | 3910 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
4ffef5fe DN |
3911 | * |
3912 | * Returns | |
3913 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. | |
3914 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for | |
3915 | * move charge. if @target is not NULL, the page is stored in target->page | |
3916 | * with extra refcnt got(Callers should handle it). | |
02491447 DN |
3917 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
3918 | * target for charge migration. if @target is not NULL, the entry is stored | |
3919 | * in target->ent. | |
4ffef5fe DN |
3920 | * |
3921 | * Called with pte lock held. | |
3922 | */ | |
4ffef5fe DN |
3923 | union mc_target { |
3924 | struct page *page; | |
02491447 | 3925 | swp_entry_t ent; |
4ffef5fe DN |
3926 | }; |
3927 | ||
4ffef5fe DN |
3928 | enum mc_target_type { |
3929 | MC_TARGET_NONE, /* not used */ | |
3930 | MC_TARGET_PAGE, | |
02491447 | 3931 | MC_TARGET_SWAP, |
4ffef5fe DN |
3932 | }; |
3933 | ||
3934 | static int is_target_pte_for_mc(struct vm_area_struct *vma, | |
3935 | unsigned long addr, pte_t ptent, union mc_target *target) | |
3936 | { | |
02491447 | 3937 | struct page *page = NULL; |
4ffef5fe DN |
3938 | struct page_cgroup *pc; |
3939 | int ret = 0; | |
02491447 DN |
3940 | swp_entry_t ent = { .val = 0 }; |
3941 | int usage_count = 0; | |
4ffef5fe DN |
3942 | bool move_anon = test_bit(MOVE_CHARGE_TYPE_ANON, |
3943 | &mc.to->move_charge_at_immigrate); | |
3944 | ||
02491447 DN |
3945 | if (!pte_present(ptent)) { |
3946 | /* TODO: handle swap of shmes/tmpfs */ | |
3947 | if (pte_none(ptent) || pte_file(ptent)) | |
3948 | return 0; | |
3949 | else if (is_swap_pte(ptent)) { | |
3950 | ent = pte_to_swp_entry(ptent); | |
3951 | if (!move_anon || non_swap_entry(ent)) | |
3952 | return 0; | |
3953 | usage_count = mem_cgroup_count_swap_user(ent, &page); | |
3954 | } | |
3955 | } else { | |
3956 | page = vm_normal_page(vma, addr, ptent); | |
3957 | if (!page || !page_mapped(page)) | |
3958 | return 0; | |
3959 | /* | |
3960 | * TODO: We don't move charges of file(including shmem/tmpfs) | |
3961 | * pages for now. | |
3962 | */ | |
3963 | if (!move_anon || !PageAnon(page)) | |
3964 | return 0; | |
3965 | if (!get_page_unless_zero(page)) | |
3966 | return 0; | |
3967 | usage_count = page_mapcount(page); | |
3968 | } | |
3969 | if (usage_count > 1) { | |
3970 | /* | |
3971 | * TODO: We don't move charges of shared(used by multiple | |
3972 | * processes) pages for now. | |
3973 | */ | |
3974 | if (page) | |
3975 | put_page(page); | |
4ffef5fe | 3976 | return 0; |
02491447 DN |
3977 | } |
3978 | if (page) { | |
3979 | pc = lookup_page_cgroup(page); | |
3980 | /* | |
3981 | * Do only loose check w/o page_cgroup lock. | |
3982 | * mem_cgroup_move_account() checks the pc is valid or not under | |
3983 | * the lock. | |
3984 | */ | |
3985 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { | |
3986 | ret = MC_TARGET_PAGE; | |
3987 | if (target) | |
3988 | target->page = page; | |
3989 | } | |
3990 | if (!ret || !target) | |
3991 | put_page(page); | |
3992 | } | |
3993 | /* throught */ | |
3994 | if (ent.val && do_swap_account && !ret && | |
3995 | css_id(&mc.from->css) == lookup_swap_cgroup(ent)) { | |
3996 | ret = MC_TARGET_SWAP; | |
4ffef5fe | 3997 | if (target) |
02491447 | 3998 | target->ent = ent; |
4ffef5fe | 3999 | } |
4ffef5fe DN |
4000 | return ret; |
4001 | } | |
4002 | ||
4003 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, | |
4004 | unsigned long addr, unsigned long end, | |
4005 | struct mm_walk *walk) | |
4006 | { | |
4007 | struct vm_area_struct *vma = walk->private; | |
4008 | pte_t *pte; | |
4009 | spinlock_t *ptl; | |
4010 | ||
4011 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4012 | for (; addr != end; pte++, addr += PAGE_SIZE) | |
4013 | if (is_target_pte_for_mc(vma, addr, *pte, NULL)) | |
4014 | mc.precharge++; /* increment precharge temporarily */ | |
4015 | pte_unmap_unlock(pte - 1, ptl); | |
4016 | cond_resched(); | |
4017 | ||
7dc74be0 DN |
4018 | return 0; |
4019 | } | |
4020 | ||
4ffef5fe DN |
4021 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
4022 | { | |
4023 | unsigned long precharge; | |
4024 | struct vm_area_struct *vma; | |
4025 | ||
4026 | down_read(&mm->mmap_sem); | |
4027 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
4028 | struct mm_walk mem_cgroup_count_precharge_walk = { | |
4029 | .pmd_entry = mem_cgroup_count_precharge_pte_range, | |
4030 | .mm = mm, | |
4031 | .private = vma, | |
4032 | }; | |
4033 | if (is_vm_hugetlb_page(vma)) | |
4034 | continue; | |
4035 | /* TODO: We don't move charges of shmem/tmpfs pages for now. */ | |
4036 | if (vma->vm_flags & VM_SHARED) | |
4037 | continue; | |
4038 | walk_page_range(vma->vm_start, vma->vm_end, | |
4039 | &mem_cgroup_count_precharge_walk); | |
4040 | } | |
4041 | up_read(&mm->mmap_sem); | |
4042 | ||
4043 | precharge = mc.precharge; | |
4044 | mc.precharge = 0; | |
4045 | ||
4046 | return precharge; | |
4047 | } | |
4048 | ||
4ffef5fe DN |
4049 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
4050 | { | |
854ffa8d | 4051 | return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm)); |
4ffef5fe DN |
4052 | } |
4053 | ||
4054 | static void mem_cgroup_clear_mc(void) | |
4055 | { | |
4056 | /* we must uncharge all the leftover precharges from mc.to */ | |
854ffa8d DN |
4057 | if (mc.precharge) { |
4058 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); | |
4059 | mc.precharge = 0; | |
4060 | } | |
4061 | /* | |
4062 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so | |
4063 | * we must uncharge here. | |
4064 | */ | |
4065 | if (mc.moved_charge) { | |
4066 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); | |
4067 | mc.moved_charge = 0; | |
4ffef5fe | 4068 | } |
483c30b5 DN |
4069 | /* we must fixup refcnts and charges */ |
4070 | if (mc.moved_swap) { | |
4071 | WARN_ON_ONCE(mc.moved_swap > INT_MAX); | |
4072 | /* uncharge swap account from the old cgroup */ | |
4073 | if (!mem_cgroup_is_root(mc.from)) | |
4074 | res_counter_uncharge(&mc.from->memsw, | |
4075 | PAGE_SIZE * mc.moved_swap); | |
4076 | __mem_cgroup_put(mc.from, mc.moved_swap); | |
4077 | ||
4078 | if (!mem_cgroup_is_root(mc.to)) { | |
4079 | /* | |
4080 | * we charged both to->res and to->memsw, so we should | |
4081 | * uncharge to->res. | |
4082 | */ | |
4083 | res_counter_uncharge(&mc.to->res, | |
4084 | PAGE_SIZE * mc.moved_swap); | |
4085 | VM_BUG_ON(test_bit(CSS_ROOT, &mc.to->css.flags)); | |
4086 | __css_put(&mc.to->css, mc.moved_swap); | |
4087 | } | |
4088 | /* we've already done mem_cgroup_get(mc.to) */ | |
4089 | ||
4090 | mc.moved_swap = 0; | |
4091 | } | |
4ffef5fe DN |
4092 | mc.from = NULL; |
4093 | mc.to = NULL; | |
8033b97c DN |
4094 | mc.moving_task = NULL; |
4095 | wake_up_all(&mc.waitq); | |
4ffef5fe DN |
4096 | } |
4097 | ||
7dc74be0 DN |
4098 | static int mem_cgroup_can_attach(struct cgroup_subsys *ss, |
4099 | struct cgroup *cgroup, | |
4100 | struct task_struct *p, | |
4101 | bool threadgroup) | |
4102 | { | |
4103 | int ret = 0; | |
4104 | struct mem_cgroup *mem = mem_cgroup_from_cont(cgroup); | |
4105 | ||
4106 | if (mem->move_charge_at_immigrate) { | |
4107 | struct mm_struct *mm; | |
4108 | struct mem_cgroup *from = mem_cgroup_from_task(p); | |
4109 | ||
4110 | VM_BUG_ON(from == mem); | |
4111 | ||
4112 | mm = get_task_mm(p); | |
4113 | if (!mm) | |
4114 | return 0; | |
7dc74be0 | 4115 | /* We move charges only when we move a owner of the mm */ |
4ffef5fe DN |
4116 | if (mm->owner == p) { |
4117 | VM_BUG_ON(mc.from); | |
4118 | VM_BUG_ON(mc.to); | |
4119 | VM_BUG_ON(mc.precharge); | |
854ffa8d | 4120 | VM_BUG_ON(mc.moved_charge); |
483c30b5 | 4121 | VM_BUG_ON(mc.moved_swap); |
8033b97c | 4122 | VM_BUG_ON(mc.moving_task); |
4ffef5fe DN |
4123 | mc.from = from; |
4124 | mc.to = mem; | |
4125 | mc.precharge = 0; | |
854ffa8d | 4126 | mc.moved_charge = 0; |
483c30b5 | 4127 | mc.moved_swap = 0; |
8033b97c | 4128 | mc.moving_task = current; |
4ffef5fe DN |
4129 | |
4130 | ret = mem_cgroup_precharge_mc(mm); | |
4131 | if (ret) | |
4132 | mem_cgroup_clear_mc(); | |
4133 | } | |
7dc74be0 DN |
4134 | mmput(mm); |
4135 | } | |
4136 | return ret; | |
4137 | } | |
4138 | ||
4139 | static void mem_cgroup_cancel_attach(struct cgroup_subsys *ss, | |
4140 | struct cgroup *cgroup, | |
4141 | struct task_struct *p, | |
4142 | bool threadgroup) | |
4143 | { | |
4ffef5fe | 4144 | mem_cgroup_clear_mc(); |
7dc74be0 DN |
4145 | } |
4146 | ||
4ffef5fe DN |
4147 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
4148 | unsigned long addr, unsigned long end, | |
4149 | struct mm_walk *walk) | |
7dc74be0 | 4150 | { |
4ffef5fe DN |
4151 | int ret = 0; |
4152 | struct vm_area_struct *vma = walk->private; | |
4153 | pte_t *pte; | |
4154 | spinlock_t *ptl; | |
4155 | ||
4156 | retry: | |
4157 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); | |
4158 | for (; addr != end; addr += PAGE_SIZE) { | |
4159 | pte_t ptent = *(pte++); | |
4160 | union mc_target target; | |
4161 | int type; | |
4162 | struct page *page; | |
4163 | struct page_cgroup *pc; | |
02491447 | 4164 | swp_entry_t ent; |
4ffef5fe DN |
4165 | |
4166 | if (!mc.precharge) | |
4167 | break; | |
4168 | ||
4169 | type = is_target_pte_for_mc(vma, addr, ptent, &target); | |
4170 | switch (type) { | |
4171 | case MC_TARGET_PAGE: | |
4172 | page = target.page; | |
4173 | if (isolate_lru_page(page)) | |
4174 | goto put; | |
4175 | pc = lookup_page_cgroup(page); | |
854ffa8d DN |
4176 | if (!mem_cgroup_move_account(pc, |
4177 | mc.from, mc.to, false)) { | |
4ffef5fe | 4178 | mc.precharge--; |
854ffa8d DN |
4179 | /* we uncharge from mc.from later. */ |
4180 | mc.moved_charge++; | |
4ffef5fe DN |
4181 | } |
4182 | putback_lru_page(page); | |
4183 | put: /* is_target_pte_for_mc() gets the page */ | |
4184 | put_page(page); | |
4185 | break; | |
02491447 DN |
4186 | case MC_TARGET_SWAP: |
4187 | ent = target.ent; | |
483c30b5 DN |
4188 | if (!mem_cgroup_move_swap_account(ent, |
4189 | mc.from, mc.to, false)) { | |
02491447 | 4190 | mc.precharge--; |
483c30b5 DN |
4191 | /* we fixup refcnts and charges later. */ |
4192 | mc.moved_swap++; | |
4193 | } | |
02491447 | 4194 | break; |
4ffef5fe DN |
4195 | default: |
4196 | break; | |
4197 | } | |
4198 | } | |
4199 | pte_unmap_unlock(pte - 1, ptl); | |
4200 | cond_resched(); | |
4201 | ||
4202 | if (addr != end) { | |
4203 | /* | |
4204 | * We have consumed all precharges we got in can_attach(). | |
4205 | * We try charge one by one, but don't do any additional | |
4206 | * charges to mc.to if we have failed in charge once in attach() | |
4207 | * phase. | |
4208 | */ | |
854ffa8d | 4209 | ret = mem_cgroup_do_precharge(1); |
4ffef5fe DN |
4210 | if (!ret) |
4211 | goto retry; | |
4212 | } | |
4213 | ||
4214 | return ret; | |
4215 | } | |
4216 | ||
4217 | static void mem_cgroup_move_charge(struct mm_struct *mm) | |
4218 | { | |
4219 | struct vm_area_struct *vma; | |
4220 | ||
4221 | lru_add_drain_all(); | |
4222 | down_read(&mm->mmap_sem); | |
4223 | for (vma = mm->mmap; vma; vma = vma->vm_next) { | |
4224 | int ret; | |
4225 | struct mm_walk mem_cgroup_move_charge_walk = { | |
4226 | .pmd_entry = mem_cgroup_move_charge_pte_range, | |
4227 | .mm = mm, | |
4228 | .private = vma, | |
4229 | }; | |
4230 | if (is_vm_hugetlb_page(vma)) | |
4231 | continue; | |
4232 | /* TODO: We don't move charges of shmem/tmpfs pages for now. */ | |
4233 | if (vma->vm_flags & VM_SHARED) | |
4234 | continue; | |
4235 | ret = walk_page_range(vma->vm_start, vma->vm_end, | |
4236 | &mem_cgroup_move_charge_walk); | |
4237 | if (ret) | |
4238 | /* | |
4239 | * means we have consumed all precharges and failed in | |
4240 | * doing additional charge. Just abandon here. | |
4241 | */ | |
4242 | break; | |
4243 | } | |
4244 | up_read(&mm->mmap_sem); | |
7dc74be0 DN |
4245 | } |
4246 | ||
67e465a7 BS |
4247 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
4248 | struct cgroup *cont, | |
4249 | struct cgroup *old_cont, | |
be367d09 BB |
4250 | struct task_struct *p, |
4251 | bool threadgroup) | |
67e465a7 | 4252 | { |
4ffef5fe DN |
4253 | struct mm_struct *mm; |
4254 | ||
4255 | if (!mc.to) | |
4256 | /* no need to move charge */ | |
4257 | return; | |
4258 | ||
4259 | mm = get_task_mm(p); | |
4260 | if (mm) { | |
4261 | mem_cgroup_move_charge(mm); | |
4262 | mmput(mm); | |
4263 | } | |
4264 | mem_cgroup_clear_mc(); | |
67e465a7 BS |
4265 | } |
4266 | ||
8cdea7c0 BS |
4267 | struct cgroup_subsys mem_cgroup_subsys = { |
4268 | .name = "memory", | |
4269 | .subsys_id = mem_cgroup_subsys_id, | |
4270 | .create = mem_cgroup_create, | |
df878fb0 | 4271 | .pre_destroy = mem_cgroup_pre_destroy, |
8cdea7c0 BS |
4272 | .destroy = mem_cgroup_destroy, |
4273 | .populate = mem_cgroup_populate, | |
7dc74be0 DN |
4274 | .can_attach = mem_cgroup_can_attach, |
4275 | .cancel_attach = mem_cgroup_cancel_attach, | |
67e465a7 | 4276 | .attach = mem_cgroup_move_task, |
6d12e2d8 | 4277 | .early_init = 0, |
04046e1a | 4278 | .use_id = 1, |
8cdea7c0 | 4279 | }; |
c077719b KH |
4280 | |
4281 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP | |
4282 | ||
4283 | static int __init disable_swap_account(char *s) | |
4284 | { | |
4285 | really_do_swap_account = 0; | |
4286 | return 1; | |
4287 | } | |
4288 | __setup("noswapaccount", disable_swap_account); | |
4289 | #endif |