Root/
1 | /* memcontrol.c - Memory Controller |
2 | * |
3 | * Copyright IBM Corporation, 2007 |
4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> |
5 | * |
6 | * Copyright 2007 OpenVZ SWsoft Inc |
7 | * Author: Pavel Emelianov <xemul@openvz.org> |
8 | * |
9 | * This program is free software; you can redistribute it and/or modify |
10 | * it under the terms of the GNU General Public License as published by |
11 | * the Free Software Foundation; either version 2 of the License, or |
12 | * (at your option) any later version. |
13 | * |
14 | * This program is distributed in the hope that it will be useful, |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
17 | * GNU General Public License for more details. |
18 | */ |
19 | |
20 | #include <linux/res_counter.h> |
21 | #include <linux/memcontrol.h> |
22 | #include <linux/cgroup.h> |
23 | #include <linux/mm.h> |
24 | #include <linux/pagemap.h> |
25 | #include <linux/smp.h> |
26 | #include <linux/page-flags.h> |
27 | #include <linux/backing-dev.h> |
28 | #include <linux/bit_spinlock.h> |
29 | #include <linux/rcupdate.h> |
30 | #include <linux/limits.h> |
31 | #include <linux/mutex.h> |
32 | #include <linux/rbtree.h> |
33 | #include <linux/slab.h> |
34 | #include <linux/swap.h> |
35 | #include <linux/spinlock.h> |
36 | #include <linux/fs.h> |
37 | #include <linux/seq_file.h> |
38 | #include <linux/vmalloc.h> |
39 | #include <linux/mm_inline.h> |
40 | #include <linux/page_cgroup.h> |
41 | #include "internal.h" |
42 | |
43 | #include <asm/uaccess.h> |
44 | |
45 | struct cgroup_subsys mem_cgroup_subsys __read_mostly; |
46 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
47 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
48 | |
49 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
50 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
51 | int do_swap_account __read_mostly; |
52 | static int really_do_swap_account __initdata = 1; /* for remember boot option*/ |
53 | #else |
54 | #define do_swap_account (0) |
55 | #endif |
56 | |
57 | static DEFINE_MUTEX(memcg_tasklist); /* can be hold under cgroup_mutex */ |
58 | #define SOFTLIMIT_EVENTS_THRESH (1000) |
59 | |
60 | /* |
61 | * Statistics for memory cgroup. |
62 | */ |
63 | enum mem_cgroup_stat_index { |
64 | /* |
65 | * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss. |
66 | */ |
67 | MEM_CGROUP_STAT_CACHE, /* # of pages charged as cache */ |
68 | MEM_CGROUP_STAT_RSS, /* # of pages charged as anon rss */ |
69 | MEM_CGROUP_STAT_MAPPED_FILE, /* # of pages charged as file rss */ |
70 | MEM_CGROUP_STAT_PGPGIN_COUNT, /* # of pages paged in */ |
71 | MEM_CGROUP_STAT_PGPGOUT_COUNT, /* # of pages paged out */ |
72 | MEM_CGROUP_STAT_EVENTS, /* sum of pagein + pageout for internal use */ |
73 | MEM_CGROUP_STAT_SWAPOUT, /* # of pages, swapped out */ |
74 | |
75 | MEM_CGROUP_STAT_NSTATS, |
76 | }; |
77 | |
78 | struct mem_cgroup_stat_cpu { |
79 | s64 count[MEM_CGROUP_STAT_NSTATS]; |
80 | } ____cacheline_aligned_in_smp; |
81 | |
82 | struct mem_cgroup_stat { |
83 | struct mem_cgroup_stat_cpu cpustat[0]; |
84 | }; |
85 | |
86 | static inline void |
87 | __mem_cgroup_stat_reset_safe(struct mem_cgroup_stat_cpu *stat, |
88 | enum mem_cgroup_stat_index idx) |
89 | { |
90 | stat->count[idx] = 0; |
91 | } |
92 | |
93 | static inline s64 |
94 | __mem_cgroup_stat_read_local(struct mem_cgroup_stat_cpu *stat, |
95 | enum mem_cgroup_stat_index idx) |
96 | { |
97 | return stat->count[idx]; |
98 | } |
99 | |
100 | /* |
101 | * For accounting under irq disable, no need for increment preempt count. |
102 | */ |
103 | static inline void __mem_cgroup_stat_add_safe(struct mem_cgroup_stat_cpu *stat, |
104 | enum mem_cgroup_stat_index idx, int val) |
105 | { |
106 | stat->count[idx] += val; |
107 | } |
108 | |
109 | static s64 mem_cgroup_read_stat(struct mem_cgroup_stat *stat, |
110 | enum mem_cgroup_stat_index idx) |
111 | { |
112 | int cpu; |
113 | s64 ret = 0; |
114 | for_each_possible_cpu(cpu) |
115 | ret += stat->cpustat[cpu].count[idx]; |
116 | return ret; |
117 | } |
118 | |
119 | static s64 mem_cgroup_local_usage(struct mem_cgroup_stat *stat) |
120 | { |
121 | s64 ret; |
122 | |
123 | ret = mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_CACHE); |
124 | ret += mem_cgroup_read_stat(stat, MEM_CGROUP_STAT_RSS); |
125 | return ret; |
126 | } |
127 | |
128 | /* |
129 | * per-zone information in memory controller. |
130 | */ |
131 | struct mem_cgroup_per_zone { |
132 | /* |
133 | * spin_lock to protect the per cgroup LRU |
134 | */ |
135 | struct list_head lists[NR_LRU_LISTS]; |
136 | unsigned long count[NR_LRU_LISTS]; |
137 | |
138 | struct zone_reclaim_stat reclaim_stat; |
139 | struct rb_node tree_node; /* RB tree node */ |
140 | unsigned long long usage_in_excess;/* Set to the value by which */ |
141 | /* the soft limit is exceeded*/ |
142 | bool on_tree; |
143 | struct mem_cgroup *mem; /* Back pointer, we cannot */ |
144 | /* use container_of */ |
145 | }; |
146 | /* Macro for accessing counter */ |
147 | #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)]) |
148 | |
149 | struct mem_cgroup_per_node { |
150 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; |
151 | }; |
152 | |
153 | struct mem_cgroup_lru_info { |
154 | struct mem_cgroup_per_node *nodeinfo[MAX_NUMNODES]; |
155 | }; |
156 | |
157 | /* |
158 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
159 | * their hierarchy representation |
160 | */ |
161 | |
162 | struct mem_cgroup_tree_per_zone { |
163 | struct rb_root rb_root; |
164 | spinlock_t lock; |
165 | }; |
166 | |
167 | struct mem_cgroup_tree_per_node { |
168 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; |
169 | }; |
170 | |
171 | struct mem_cgroup_tree { |
172 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
173 | }; |
174 | |
175 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
176 | |
177 | /* |
178 | * The memory controller data structure. The memory controller controls both |
179 | * page cache and RSS per cgroup. We would eventually like to provide |
180 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, |
181 | * to help the administrator determine what knobs to tune. |
182 | * |
183 | * TODO: Add a water mark for the memory controller. Reclaim will begin when |
184 | * we hit the water mark. May be even add a low water mark, such that |
185 | * no reclaim occurs from a cgroup at it's low water mark, this is |
186 | * a feature that will be implemented much later in the future. |
187 | */ |
188 | struct mem_cgroup { |
189 | struct cgroup_subsys_state css; |
190 | /* |
191 | * the counter to account for memory usage |
192 | */ |
193 | struct res_counter res; |
194 | /* |
195 | * the counter to account for mem+swap usage. |
196 | */ |
197 | struct res_counter memsw; |
198 | /* |
199 | * Per cgroup active and inactive list, similar to the |
200 | * per zone LRU lists. |
201 | */ |
202 | struct mem_cgroup_lru_info info; |
203 | |
204 | /* |
205 | protect against reclaim related member. |
206 | */ |
207 | spinlock_t reclaim_param_lock; |
208 | |
209 | int prev_priority; /* for recording reclaim priority */ |
210 | |
211 | /* |
212 | * While reclaiming in a hiearchy, we cache the last child we |
213 | * reclaimed from. |
214 | */ |
215 | int last_scanned_child; |
216 | /* |
217 | * Should the accounting and control be hierarchical, per subtree? |
218 | */ |
219 | bool use_hierarchy; |
220 | unsigned long last_oom_jiffies; |
221 | atomic_t refcnt; |
222 | |
223 | unsigned int swappiness; |
224 | |
225 | /* set when res.limit == memsw.limit */ |
226 | bool memsw_is_minimum; |
227 | |
228 | /* |
229 | * statistics. This must be placed at the end of memcg. |
230 | */ |
231 | struct mem_cgroup_stat stat; |
232 | }; |
233 | |
234 | /* |
235 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
236 | * limit reclaim to prevent infinite loops, if they ever occur. |
237 | */ |
238 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100) |
239 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2) |
240 | |
241 | enum charge_type { |
242 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
243 | MEM_CGROUP_CHARGE_TYPE_MAPPED, |
244 | MEM_CGROUP_CHARGE_TYPE_SHMEM, /* used by page migration of shmem */ |
245 | MEM_CGROUP_CHARGE_TYPE_FORCE, /* used by force_empty */ |
246 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
247 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
248 | NR_CHARGE_TYPE, |
249 | }; |
250 | |
251 | /* only for here (for easy reading.) */ |
252 | #define PCGF_CACHE (1UL << PCG_CACHE) |
253 | #define PCGF_USED (1UL << PCG_USED) |
254 | #define PCGF_LOCK (1UL << PCG_LOCK) |
255 | /* Not used, but added here for completeness */ |
256 | #define PCGF_ACCT (1UL << PCG_ACCT) |
257 | |
258 | /* for encoding cft->private value on file */ |
259 | #define _MEM (0) |
260 | #define _MEMSWAP (1) |
261 | #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val)) |
262 | #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff) |
263 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
264 | |
265 | /* |
266 | * Reclaim flags for mem_cgroup_hierarchical_reclaim |
267 | */ |
268 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 |
269 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) |
270 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 |
271 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) |
272 | #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2 |
273 | #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT) |
274 | |
275 | static void mem_cgroup_get(struct mem_cgroup *mem); |
276 | static void mem_cgroup_put(struct mem_cgroup *mem); |
277 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem); |
278 | |
279 | static struct mem_cgroup_per_zone * |
280 | mem_cgroup_zoneinfo(struct mem_cgroup *mem, int nid, int zid) |
281 | { |
282 | return &mem->info.nodeinfo[nid]->zoneinfo[zid]; |
283 | } |
284 | |
285 | static struct mem_cgroup_per_zone * |
286 | page_cgroup_zoneinfo(struct page_cgroup *pc) |
287 | { |
288 | struct mem_cgroup *mem = pc->mem_cgroup; |
289 | int nid = page_cgroup_nid(pc); |
290 | int zid = page_cgroup_zid(pc); |
291 | |
292 | if (!mem) |
293 | return NULL; |
294 | |
295 | return mem_cgroup_zoneinfo(mem, nid, zid); |
296 | } |
297 | |
298 | static struct mem_cgroup_tree_per_zone * |
299 | soft_limit_tree_node_zone(int nid, int zid) |
300 | { |
301 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
302 | } |
303 | |
304 | static struct mem_cgroup_tree_per_zone * |
305 | soft_limit_tree_from_page(struct page *page) |
306 | { |
307 | int nid = page_to_nid(page); |
308 | int zid = page_zonenum(page); |
309 | |
310 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
311 | } |
312 | |
313 | static void |
314 | __mem_cgroup_insert_exceeded(struct mem_cgroup *mem, |
315 | struct mem_cgroup_per_zone *mz, |
316 | struct mem_cgroup_tree_per_zone *mctz, |
317 | unsigned long long new_usage_in_excess) |
318 | { |
319 | struct rb_node **p = &mctz->rb_root.rb_node; |
320 | struct rb_node *parent = NULL; |
321 | struct mem_cgroup_per_zone *mz_node; |
322 | |
323 | if (mz->on_tree) |
324 | return; |
325 | |
326 | mz->usage_in_excess = new_usage_in_excess; |
327 | if (!mz->usage_in_excess) |
328 | return; |
329 | while (*p) { |
330 | parent = *p; |
331 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
332 | tree_node); |
333 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
334 | p = &(*p)->rb_left; |
335 | /* |
336 | * We can't avoid mem cgroups that are over their soft |
337 | * limit by the same amount |
338 | */ |
339 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
340 | p = &(*p)->rb_right; |
341 | } |
342 | rb_link_node(&mz->tree_node, parent, p); |
343 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
344 | mz->on_tree = true; |
345 | } |
346 | |
347 | static void |
348 | __mem_cgroup_remove_exceeded(struct mem_cgroup *mem, |
349 | struct mem_cgroup_per_zone *mz, |
350 | struct mem_cgroup_tree_per_zone *mctz) |
351 | { |
352 | if (!mz->on_tree) |
353 | return; |
354 | rb_erase(&mz->tree_node, &mctz->rb_root); |
355 | mz->on_tree = false; |
356 | } |
357 | |
358 | static void |
359 | mem_cgroup_remove_exceeded(struct mem_cgroup *mem, |
360 | struct mem_cgroup_per_zone *mz, |
361 | struct mem_cgroup_tree_per_zone *mctz) |
362 | { |
363 | spin_lock(&mctz->lock); |
364 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
365 | spin_unlock(&mctz->lock); |
366 | } |
367 | |
368 | static bool mem_cgroup_soft_limit_check(struct mem_cgroup *mem) |
369 | { |
370 | bool ret = false; |
371 | int cpu; |
372 | s64 val; |
373 | struct mem_cgroup_stat_cpu *cpustat; |
374 | |
375 | cpu = get_cpu(); |
376 | cpustat = &mem->stat.cpustat[cpu]; |
377 | val = __mem_cgroup_stat_read_local(cpustat, MEM_CGROUP_STAT_EVENTS); |
378 | if (unlikely(val > SOFTLIMIT_EVENTS_THRESH)) { |
379 | __mem_cgroup_stat_reset_safe(cpustat, MEM_CGROUP_STAT_EVENTS); |
380 | ret = true; |
381 | } |
382 | put_cpu(); |
383 | return ret; |
384 | } |
385 | |
386 | static void mem_cgroup_update_tree(struct mem_cgroup *mem, struct page *page) |
387 | { |
388 | unsigned long long excess; |
389 | struct mem_cgroup_per_zone *mz; |
390 | struct mem_cgroup_tree_per_zone *mctz; |
391 | int nid = page_to_nid(page); |
392 | int zid = page_zonenum(page); |
393 | mctz = soft_limit_tree_from_page(page); |
394 | |
395 | /* |
396 | * Necessary to update all ancestors when hierarchy is used. |
397 | * because their event counter is not touched. |
398 | */ |
399 | for (; mem; mem = parent_mem_cgroup(mem)) { |
400 | mz = mem_cgroup_zoneinfo(mem, nid, zid); |
401 | excess = res_counter_soft_limit_excess(&mem->res); |
402 | /* |
403 | * We have to update the tree if mz is on RB-tree or |
404 | * mem is over its softlimit. |
405 | */ |
406 | if (excess || mz->on_tree) { |
407 | spin_lock(&mctz->lock); |
408 | /* if on-tree, remove it */ |
409 | if (mz->on_tree) |
410 | __mem_cgroup_remove_exceeded(mem, mz, mctz); |
411 | /* |
412 | * Insert again. mz->usage_in_excess will be updated. |
413 | * If excess is 0, no tree ops. |
414 | */ |
415 | __mem_cgroup_insert_exceeded(mem, mz, mctz, excess); |
416 | spin_unlock(&mctz->lock); |
417 | } |
418 | } |
419 | } |
420 | |
421 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *mem) |
422 | { |
423 | int node, zone; |
424 | struct mem_cgroup_per_zone *mz; |
425 | struct mem_cgroup_tree_per_zone *mctz; |
426 | |
427 | for_each_node_state(node, N_POSSIBLE) { |
428 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
429 | mz = mem_cgroup_zoneinfo(mem, node, zone); |
430 | mctz = soft_limit_tree_node_zone(node, zone); |
431 | mem_cgroup_remove_exceeded(mem, mz, mctz); |
432 | } |
433 | } |
434 | } |
435 | |
436 | static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup *mem) |
437 | { |
438 | return res_counter_soft_limit_excess(&mem->res) >> PAGE_SHIFT; |
439 | } |
440 | |
441 | static struct mem_cgroup_per_zone * |
442 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
443 | { |
444 | struct rb_node *rightmost = NULL; |
445 | struct mem_cgroup_per_zone *mz; |
446 | |
447 | retry: |
448 | mz = NULL; |
449 | rightmost = rb_last(&mctz->rb_root); |
450 | if (!rightmost) |
451 | goto done; /* Nothing to reclaim from */ |
452 | |
453 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
454 | /* |
455 | * Remove the node now but someone else can add it back, |
456 | * we will to add it back at the end of reclaim to its correct |
457 | * position in the tree. |
458 | */ |
459 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
460 | if (!res_counter_soft_limit_excess(&mz->mem->res) || |
461 | !css_tryget(&mz->mem->css)) |
462 | goto retry; |
463 | done: |
464 | return mz; |
465 | } |
466 | |
467 | static struct mem_cgroup_per_zone * |
468 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
469 | { |
470 | struct mem_cgroup_per_zone *mz; |
471 | |
472 | spin_lock(&mctz->lock); |
473 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
474 | spin_unlock(&mctz->lock); |
475 | return mz; |
476 | } |
477 | |
478 | static void mem_cgroup_swap_statistics(struct mem_cgroup *mem, |
479 | bool charge) |
480 | { |
481 | int val = (charge) ? 1 : -1; |
482 | struct mem_cgroup_stat *stat = &mem->stat; |
483 | struct mem_cgroup_stat_cpu *cpustat; |
484 | int cpu = get_cpu(); |
485 | |
486 | cpustat = &stat->cpustat[cpu]; |
487 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_SWAPOUT, val); |
488 | put_cpu(); |
489 | } |
490 | |
491 | static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, |
492 | struct page_cgroup *pc, |
493 | bool charge) |
494 | { |
495 | int val = (charge) ? 1 : -1; |
496 | struct mem_cgroup_stat *stat = &mem->stat; |
497 | struct mem_cgroup_stat_cpu *cpustat; |
498 | int cpu = get_cpu(); |
499 | |
500 | cpustat = &stat->cpustat[cpu]; |
501 | if (PageCgroupCache(pc)) |
502 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_CACHE, val); |
503 | else |
504 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_RSS, val); |
505 | |
506 | if (charge) |
507 | __mem_cgroup_stat_add_safe(cpustat, |
508 | MEM_CGROUP_STAT_PGPGIN_COUNT, 1); |
509 | else |
510 | __mem_cgroup_stat_add_safe(cpustat, |
511 | MEM_CGROUP_STAT_PGPGOUT_COUNT, 1); |
512 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_EVENTS, 1); |
513 | put_cpu(); |
514 | } |
515 | |
516 | static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem, |
517 | enum lru_list idx) |
518 | { |
519 | int nid, zid; |
520 | struct mem_cgroup_per_zone *mz; |
521 | u64 total = 0; |
522 | |
523 | for_each_online_node(nid) |
524 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
525 | mz = mem_cgroup_zoneinfo(mem, nid, zid); |
526 | total += MEM_CGROUP_ZSTAT(mz, idx); |
527 | } |
528 | return total; |
529 | } |
530 | |
531 | static struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont) |
532 | { |
533 | return container_of(cgroup_subsys_state(cont, |
534 | mem_cgroup_subsys_id), struct mem_cgroup, |
535 | css); |
536 | } |
537 | |
538 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
539 | { |
540 | /* |
541 | * mm_update_next_owner() may clear mm->owner to NULL |
542 | * if it races with swapoff, page migration, etc. |
543 | * So this can be called with p == NULL. |
544 | */ |
545 | if (unlikely(!p)) |
546 | return NULL; |
547 | |
548 | return container_of(task_subsys_state(p, mem_cgroup_subsys_id), |
549 | struct mem_cgroup, css); |
550 | } |
551 | |
552 | static struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm) |
553 | { |
554 | struct mem_cgroup *mem = NULL; |
555 | |
556 | if (!mm) |
557 | return NULL; |
558 | /* |
559 | * Because we have no locks, mm->owner's may be being moved to other |
560 | * cgroup. We use css_tryget() here even if this looks |
561 | * pessimistic (rather than adding locks here). |
562 | */ |
563 | rcu_read_lock(); |
564 | do { |
565 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
566 | if (unlikely(!mem)) |
567 | break; |
568 | } while (!css_tryget(&mem->css)); |
569 | rcu_read_unlock(); |
570 | return mem; |
571 | } |
572 | |
573 | /* |
574 | * Call callback function against all cgroup under hierarchy tree. |
575 | */ |
576 | static int mem_cgroup_walk_tree(struct mem_cgroup *root, void *data, |
577 | int (*func)(struct mem_cgroup *, void *)) |
578 | { |
579 | int found, ret, nextid; |
580 | struct cgroup_subsys_state *css; |
581 | struct mem_cgroup *mem; |
582 | |
583 | if (!root->use_hierarchy) |
584 | return (*func)(root, data); |
585 | |
586 | nextid = 1; |
587 | do { |
588 | ret = 0; |
589 | mem = NULL; |
590 | |
591 | rcu_read_lock(); |
592 | css = css_get_next(&mem_cgroup_subsys, nextid, &root->css, |
593 | &found); |
594 | if (css && css_tryget(css)) |
595 | mem = container_of(css, struct mem_cgroup, css); |
596 | rcu_read_unlock(); |
597 | |
598 | if (mem) { |
599 | ret = (*func)(mem, data); |
600 | css_put(&mem->css); |
601 | } |
602 | nextid = found + 1; |
603 | } while (!ret && css); |
604 | |
605 | return ret; |
606 | } |
607 | |
608 | static inline bool mem_cgroup_is_root(struct mem_cgroup *mem) |
609 | { |
610 | return (mem == root_mem_cgroup); |
611 | } |
612 | |
613 | /* |
614 | * Following LRU functions are allowed to be used without PCG_LOCK. |
615 | * Operations are called by routine of global LRU independently from memcg. |
616 | * What we have to take care of here is validness of pc->mem_cgroup. |
617 | * |
618 | * Changes to pc->mem_cgroup happens when |
619 | * 1. charge |
620 | * 2. moving account |
621 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. |
622 | * It is added to LRU before charge. |
623 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. |
624 | * When moving account, the page is not on LRU. It's isolated. |
625 | */ |
626 | |
627 | void mem_cgroup_del_lru_list(struct page *page, enum lru_list lru) |
628 | { |
629 | struct page_cgroup *pc; |
630 | struct mem_cgroup_per_zone *mz; |
631 | |
632 | if (mem_cgroup_disabled()) |
633 | return; |
634 | pc = lookup_page_cgroup(page); |
635 | /* can happen while we handle swapcache. */ |
636 | if (!TestClearPageCgroupAcctLRU(pc)) |
637 | return; |
638 | VM_BUG_ON(!pc->mem_cgroup); |
639 | /* |
640 | * We don't check PCG_USED bit. It's cleared when the "page" is finally |
641 | * removed from global LRU. |
642 | */ |
643 | mz = page_cgroup_zoneinfo(pc); |
644 | MEM_CGROUP_ZSTAT(mz, lru) -= 1; |
645 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
646 | return; |
647 | VM_BUG_ON(list_empty(&pc->lru)); |
648 | list_del_init(&pc->lru); |
649 | return; |
650 | } |
651 | |
652 | void mem_cgroup_del_lru(struct page *page) |
653 | { |
654 | mem_cgroup_del_lru_list(page, page_lru(page)); |
655 | } |
656 | |
657 | void mem_cgroup_rotate_lru_list(struct page *page, enum lru_list lru) |
658 | { |
659 | struct mem_cgroup_per_zone *mz; |
660 | struct page_cgroup *pc; |
661 | |
662 | if (mem_cgroup_disabled()) |
663 | return; |
664 | |
665 | pc = lookup_page_cgroup(page); |
666 | /* |
667 | * Used bit is set without atomic ops but after smp_wmb(). |
668 | * For making pc->mem_cgroup visible, insert smp_rmb() here. |
669 | */ |
670 | smp_rmb(); |
671 | /* unused or root page is not rotated. */ |
672 | if (!PageCgroupUsed(pc) || mem_cgroup_is_root(pc->mem_cgroup)) |
673 | return; |
674 | mz = page_cgroup_zoneinfo(pc); |
675 | list_move(&pc->lru, &mz->lists[lru]); |
676 | } |
677 | |
678 | void mem_cgroup_add_lru_list(struct page *page, enum lru_list lru) |
679 | { |
680 | struct page_cgroup *pc; |
681 | struct mem_cgroup_per_zone *mz; |
682 | |
683 | if (mem_cgroup_disabled()) |
684 | return; |
685 | pc = lookup_page_cgroup(page); |
686 | VM_BUG_ON(PageCgroupAcctLRU(pc)); |
687 | /* |
688 | * Used bit is set without atomic ops but after smp_wmb(). |
689 | * For making pc->mem_cgroup visible, insert smp_rmb() here. |
690 | */ |
691 | smp_rmb(); |
692 | if (!PageCgroupUsed(pc)) |
693 | return; |
694 | |
695 | mz = page_cgroup_zoneinfo(pc); |
696 | MEM_CGROUP_ZSTAT(mz, lru) += 1; |
697 | SetPageCgroupAcctLRU(pc); |
698 | if (mem_cgroup_is_root(pc->mem_cgroup)) |
699 | return; |
700 | list_add(&pc->lru, &mz->lists[lru]); |
701 | } |
702 | |
703 | /* |
704 | * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to |
705 | * lru because the page may.be reused after it's fully uncharged (because of |
706 | * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge |
707 | * it again. This function is only used to charge SwapCache. It's done under |
708 | * lock_page and expected that zone->lru_lock is never held. |
709 | */ |
710 | static void mem_cgroup_lru_del_before_commit_swapcache(struct page *page) |
711 | { |
712 | unsigned long flags; |
713 | struct zone *zone = page_zone(page); |
714 | struct page_cgroup *pc = lookup_page_cgroup(page); |
715 | |
716 | spin_lock_irqsave(&zone->lru_lock, flags); |
717 | /* |
718 | * Forget old LRU when this page_cgroup is *not* used. This Used bit |
719 | * is guarded by lock_page() because the page is SwapCache. |
720 | */ |
721 | if (!PageCgroupUsed(pc)) |
722 | mem_cgroup_del_lru_list(page, page_lru(page)); |
723 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
724 | } |
725 | |
726 | static void mem_cgroup_lru_add_after_commit_swapcache(struct page *page) |
727 | { |
728 | unsigned long flags; |
729 | struct zone *zone = page_zone(page); |
730 | struct page_cgroup *pc = lookup_page_cgroup(page); |
731 | |
732 | spin_lock_irqsave(&zone->lru_lock, flags); |
733 | /* link when the page is linked to LRU but page_cgroup isn't */ |
734 | if (PageLRU(page) && !PageCgroupAcctLRU(pc)) |
735 | mem_cgroup_add_lru_list(page, page_lru(page)); |
736 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
737 | } |
738 | |
739 | |
740 | void mem_cgroup_move_lists(struct page *page, |
741 | enum lru_list from, enum lru_list to) |
742 | { |
743 | if (mem_cgroup_disabled()) |
744 | return; |
745 | mem_cgroup_del_lru_list(page, from); |
746 | mem_cgroup_add_lru_list(page, to); |
747 | } |
748 | |
749 | int task_in_mem_cgroup(struct task_struct *task, const struct mem_cgroup *mem) |
750 | { |
751 | int ret; |
752 | struct mem_cgroup *curr = NULL; |
753 | |
754 | task_lock(task); |
755 | rcu_read_lock(); |
756 | curr = try_get_mem_cgroup_from_mm(task->mm); |
757 | rcu_read_unlock(); |
758 | task_unlock(task); |
759 | if (!curr) |
760 | return 0; |
761 | /* |
762 | * We should check use_hierarchy of "mem" not "curr". Because checking |
763 | * use_hierarchy of "curr" here make this function true if hierarchy is |
764 | * enabled in "curr" and "curr" is a child of "mem" in *cgroup* |
765 | * hierarchy(even if use_hierarchy is disabled in "mem"). |
766 | */ |
767 | if (mem->use_hierarchy) |
768 | ret = css_is_ancestor(&curr->css, &mem->css); |
769 | else |
770 | ret = (curr == mem); |
771 | css_put(&curr->css); |
772 | return ret; |
773 | } |
774 | |
775 | /* |
776 | * prev_priority control...this will be used in memory reclaim path. |
777 | */ |
778 | int mem_cgroup_get_reclaim_priority(struct mem_cgroup *mem) |
779 | { |
780 | int prev_priority; |
781 | |
782 | spin_lock(&mem->reclaim_param_lock); |
783 | prev_priority = mem->prev_priority; |
784 | spin_unlock(&mem->reclaim_param_lock); |
785 | |
786 | return prev_priority; |
787 | } |
788 | |
789 | void mem_cgroup_note_reclaim_priority(struct mem_cgroup *mem, int priority) |
790 | { |
791 | spin_lock(&mem->reclaim_param_lock); |
792 | if (priority < mem->prev_priority) |
793 | mem->prev_priority = priority; |
794 | spin_unlock(&mem->reclaim_param_lock); |
795 | } |
796 | |
797 | void mem_cgroup_record_reclaim_priority(struct mem_cgroup *mem, int priority) |
798 | { |
799 | spin_lock(&mem->reclaim_param_lock); |
800 | mem->prev_priority = priority; |
801 | spin_unlock(&mem->reclaim_param_lock); |
802 | } |
803 | |
804 | static int calc_inactive_ratio(struct mem_cgroup *memcg, unsigned long *present_pages) |
805 | { |
806 | unsigned long active; |
807 | unsigned long inactive; |
808 | unsigned long gb; |
809 | unsigned long inactive_ratio; |
810 | |
811 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_ANON); |
812 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_ANON); |
813 | |
814 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
815 | if (gb) |
816 | inactive_ratio = int_sqrt(10 * gb); |
817 | else |
818 | inactive_ratio = 1; |
819 | |
820 | if (present_pages) { |
821 | present_pages[0] = inactive; |
822 | present_pages[1] = active; |
823 | } |
824 | |
825 | return inactive_ratio; |
826 | } |
827 | |
828 | int mem_cgroup_inactive_anon_is_low(struct mem_cgroup *memcg) |
829 | { |
830 | unsigned long active; |
831 | unsigned long inactive; |
832 | unsigned long present_pages[2]; |
833 | unsigned long inactive_ratio; |
834 | |
835 | inactive_ratio = calc_inactive_ratio(memcg, present_pages); |
836 | |
837 | inactive = present_pages[0]; |
838 | active = present_pages[1]; |
839 | |
840 | if (inactive * inactive_ratio < active) |
841 | return 1; |
842 | |
843 | return 0; |
844 | } |
845 | |
846 | int mem_cgroup_inactive_file_is_low(struct mem_cgroup *memcg) |
847 | { |
848 | unsigned long active; |
849 | unsigned long inactive; |
850 | |
851 | inactive = mem_cgroup_get_local_zonestat(memcg, LRU_INACTIVE_FILE); |
852 | active = mem_cgroup_get_local_zonestat(memcg, LRU_ACTIVE_FILE); |
853 | |
854 | return (active > inactive); |
855 | } |
856 | |
857 | unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup *memcg, |
858 | struct zone *zone, |
859 | enum lru_list lru) |
860 | { |
861 | int nid = zone->zone_pgdat->node_id; |
862 | int zid = zone_idx(zone); |
863 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
864 | |
865 | return MEM_CGROUP_ZSTAT(mz, lru); |
866 | } |
867 | |
868 | struct zone_reclaim_stat *mem_cgroup_get_reclaim_stat(struct mem_cgroup *memcg, |
869 | struct zone *zone) |
870 | { |
871 | int nid = zone->zone_pgdat->node_id; |
872 | int zid = zone_idx(zone); |
873 | struct mem_cgroup_per_zone *mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
874 | |
875 | return &mz->reclaim_stat; |
876 | } |
877 | |
878 | struct zone_reclaim_stat * |
879 | mem_cgroup_get_reclaim_stat_from_page(struct page *page) |
880 | { |
881 | struct page_cgroup *pc; |
882 | struct mem_cgroup_per_zone *mz; |
883 | |
884 | if (mem_cgroup_disabled()) |
885 | return NULL; |
886 | |
887 | pc = lookup_page_cgroup(page); |
888 | /* |
889 | * Used bit is set without atomic ops but after smp_wmb(). |
890 | * For making pc->mem_cgroup visible, insert smp_rmb() here. |
891 | */ |
892 | smp_rmb(); |
893 | if (!PageCgroupUsed(pc)) |
894 | return NULL; |
895 | |
896 | mz = page_cgroup_zoneinfo(pc); |
897 | if (!mz) |
898 | return NULL; |
899 | |
900 | return &mz->reclaim_stat; |
901 | } |
902 | |
903 | unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan, |
904 | struct list_head *dst, |
905 | unsigned long *scanned, int order, |
906 | int mode, struct zone *z, |
907 | struct mem_cgroup *mem_cont, |
908 | int active, int file) |
909 | { |
910 | unsigned long nr_taken = 0; |
911 | struct page *page; |
912 | unsigned long scan; |
913 | LIST_HEAD(pc_list); |
914 | struct list_head *src; |
915 | struct page_cgroup *pc, *tmp; |
916 | int nid = z->zone_pgdat->node_id; |
917 | int zid = zone_idx(z); |
918 | struct mem_cgroup_per_zone *mz; |
919 | int lru = LRU_FILE * file + active; |
920 | int ret; |
921 | |
922 | BUG_ON(!mem_cont); |
923 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
924 | src = &mz->lists[lru]; |
925 | |
926 | scan = 0; |
927 | list_for_each_entry_safe_reverse(pc, tmp, src, lru) { |
928 | if (scan >= nr_to_scan) |
929 | break; |
930 | |
931 | page = pc->page; |
932 | if (unlikely(!PageCgroupUsed(pc))) |
933 | continue; |
934 | if (unlikely(!PageLRU(page))) |
935 | continue; |
936 | |
937 | scan++; |
938 | ret = __isolate_lru_page(page, mode, file); |
939 | switch (ret) { |
940 | case 0: |
941 | list_move(&page->lru, dst); |
942 | mem_cgroup_del_lru(page); |
943 | nr_taken++; |
944 | break; |
945 | case -EBUSY: |
946 | /* we don't affect global LRU but rotate in our LRU */ |
947 | mem_cgroup_rotate_lru_list(page, page_lru(page)); |
948 | break; |
949 | default: |
950 | break; |
951 | } |
952 | } |
953 | |
954 | *scanned = scan; |
955 | return nr_taken; |
956 | } |
957 | |
958 | #define mem_cgroup_from_res_counter(counter, member) \ |
959 | container_of(counter, struct mem_cgroup, member) |
960 | |
961 | static bool mem_cgroup_check_under_limit(struct mem_cgroup *mem) |
962 | { |
963 | if (do_swap_account) { |
964 | if (res_counter_check_under_limit(&mem->res) && |
965 | res_counter_check_under_limit(&mem->memsw)) |
966 | return true; |
967 | } else |
968 | if (res_counter_check_under_limit(&mem->res)) |
969 | return true; |
970 | return false; |
971 | } |
972 | |
973 | static unsigned int get_swappiness(struct mem_cgroup *memcg) |
974 | { |
975 | struct cgroup *cgrp = memcg->css.cgroup; |
976 | unsigned int swappiness; |
977 | |
978 | /* root ? */ |
979 | if (cgrp->parent == NULL) |
980 | return vm_swappiness; |
981 | |
982 | spin_lock(&memcg->reclaim_param_lock); |
983 | swappiness = memcg->swappiness; |
984 | spin_unlock(&memcg->reclaim_param_lock); |
985 | |
986 | return swappiness; |
987 | } |
988 | |
989 | static int mem_cgroup_count_children_cb(struct mem_cgroup *mem, void *data) |
990 | { |
991 | int *val = data; |
992 | (*val)++; |
993 | return 0; |
994 | } |
995 | |
996 | /** |
997 | * mem_cgroup_print_mem_info: Called from OOM with tasklist_lock held in read mode. |
998 | * @memcg: The memory cgroup that went over limit |
999 | * @p: Task that is going to be killed |
1000 | * |
1001 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
1002 | * enabled |
1003 | */ |
1004 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
1005 | { |
1006 | struct cgroup *task_cgrp; |
1007 | struct cgroup *mem_cgrp; |
1008 | /* |
1009 | * Need a buffer in BSS, can't rely on allocations. The code relies |
1010 | * on the assumption that OOM is serialized for memory controller. |
1011 | * If this assumption is broken, revisit this code. |
1012 | */ |
1013 | static char memcg_name[PATH_MAX]; |
1014 | int ret; |
1015 | |
1016 | if (!memcg) |
1017 | return; |
1018 | |
1019 | |
1020 | rcu_read_lock(); |
1021 | |
1022 | mem_cgrp = memcg->css.cgroup; |
1023 | task_cgrp = task_cgroup(p, mem_cgroup_subsys_id); |
1024 | |
1025 | ret = cgroup_path(task_cgrp, memcg_name, PATH_MAX); |
1026 | if (ret < 0) { |
1027 | /* |
1028 | * Unfortunately, we are unable to convert to a useful name |
1029 | * But we'll still print out the usage information |
1030 | */ |
1031 | rcu_read_unlock(); |
1032 | goto done; |
1033 | } |
1034 | rcu_read_unlock(); |
1035 | |
1036 | printk(KERN_INFO "Task in %s killed", memcg_name); |
1037 | |
1038 | rcu_read_lock(); |
1039 | ret = cgroup_path(mem_cgrp, memcg_name, PATH_MAX); |
1040 | if (ret < 0) { |
1041 | rcu_read_unlock(); |
1042 | goto done; |
1043 | } |
1044 | rcu_read_unlock(); |
1045 | |
1046 | /* |
1047 | * Continues from above, so we don't need an KERN_ level |
1048 | */ |
1049 | printk(KERN_CONT " as a result of limit of %s\n", memcg_name); |
1050 | done: |
1051 | |
1052 | printk(KERN_INFO "memory: usage %llukB, limit %llukB, failcnt %llu\n", |
1053 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
1054 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, |
1055 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); |
1056 | printk(KERN_INFO "memory+swap: usage %llukB, limit %llukB, " |
1057 | "failcnt %llu\n", |
1058 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
1059 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, |
1060 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); |
1061 | } |
1062 | |
1063 | /* |
1064 | * This function returns the number of memcg under hierarchy tree. Returns |
1065 | * 1(self count) if no children. |
1066 | */ |
1067 | static int mem_cgroup_count_children(struct mem_cgroup *mem) |
1068 | { |
1069 | int num = 0; |
1070 | mem_cgroup_walk_tree(mem, &num, mem_cgroup_count_children_cb); |
1071 | return num; |
1072 | } |
1073 | |
1074 | /* |
1075 | * Visit the first child (need not be the first child as per the ordering |
1076 | * of the cgroup list, since we track last_scanned_child) of @mem and use |
1077 | * that to reclaim free pages from. |
1078 | */ |
1079 | static struct mem_cgroup * |
1080 | mem_cgroup_select_victim(struct mem_cgroup *root_mem) |
1081 | { |
1082 | struct mem_cgroup *ret = NULL; |
1083 | struct cgroup_subsys_state *css; |
1084 | int nextid, found; |
1085 | |
1086 | if (!root_mem->use_hierarchy) { |
1087 | css_get(&root_mem->css); |
1088 | ret = root_mem; |
1089 | } |
1090 | |
1091 | while (!ret) { |
1092 | rcu_read_lock(); |
1093 | nextid = root_mem->last_scanned_child + 1; |
1094 | css = css_get_next(&mem_cgroup_subsys, nextid, &root_mem->css, |
1095 | &found); |
1096 | if (css && css_tryget(css)) |
1097 | ret = container_of(css, struct mem_cgroup, css); |
1098 | |
1099 | rcu_read_unlock(); |
1100 | /* Updates scanning parameter */ |
1101 | spin_lock(&root_mem->reclaim_param_lock); |
1102 | if (!css) { |
1103 | /* this means start scan from ID:1 */ |
1104 | root_mem->last_scanned_child = 0; |
1105 | } else |
1106 | root_mem->last_scanned_child = found; |
1107 | spin_unlock(&root_mem->reclaim_param_lock); |
1108 | } |
1109 | |
1110 | return ret; |
1111 | } |
1112 | |
1113 | /* |
1114 | * Scan the hierarchy if needed to reclaim memory. We remember the last child |
1115 | * we reclaimed from, so that we don't end up penalizing one child extensively |
1116 | * based on its position in the children list. |
1117 | * |
1118 | * root_mem is the original ancestor that we've been reclaim from. |
1119 | * |
1120 | * We give up and return to the caller when we visit root_mem twice. |
1121 | * (other groups can be removed while we're walking....) |
1122 | * |
1123 | * If shrink==true, for avoiding to free too much, this returns immedieately. |
1124 | */ |
1125 | static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup *root_mem, |
1126 | struct zone *zone, |
1127 | gfp_t gfp_mask, |
1128 | unsigned long reclaim_options) |
1129 | { |
1130 | struct mem_cgroup *victim; |
1131 | int ret, total = 0; |
1132 | int loop = 0; |
1133 | bool noswap = reclaim_options & MEM_CGROUP_RECLAIM_NOSWAP; |
1134 | bool shrink = reclaim_options & MEM_CGROUP_RECLAIM_SHRINK; |
1135 | bool check_soft = reclaim_options & MEM_CGROUP_RECLAIM_SOFT; |
1136 | unsigned long excess = mem_cgroup_get_excess(root_mem); |
1137 | |
1138 | /* If memsw_is_minimum==1, swap-out is of-no-use. */ |
1139 | if (root_mem->memsw_is_minimum) |
1140 | noswap = true; |
1141 | |
1142 | while (1) { |
1143 | victim = mem_cgroup_select_victim(root_mem); |
1144 | if (victim == root_mem) { |
1145 | loop++; |
1146 | if (loop >= 2) { |
1147 | /* |
1148 | * If we have not been able to reclaim |
1149 | * anything, it might because there are |
1150 | * no reclaimable pages under this hierarchy |
1151 | */ |
1152 | if (!check_soft || !total) { |
1153 | css_put(&victim->css); |
1154 | break; |
1155 | } |
1156 | /* |
1157 | * We want to do more targetted reclaim. |
1158 | * excess >> 2 is not to excessive so as to |
1159 | * reclaim too much, nor too less that we keep |
1160 | * coming back to reclaim from this cgroup |
1161 | */ |
1162 | if (total >= (excess >> 2) || |
1163 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) { |
1164 | css_put(&victim->css); |
1165 | break; |
1166 | } |
1167 | } |
1168 | } |
1169 | if (!mem_cgroup_local_usage(&victim->stat)) { |
1170 | /* this cgroup's local usage == 0 */ |
1171 | css_put(&victim->css); |
1172 | continue; |
1173 | } |
1174 | /* we use swappiness of local cgroup */ |
1175 | if (check_soft) |
1176 | ret = mem_cgroup_shrink_node_zone(victim, gfp_mask, |
1177 | noswap, get_swappiness(victim), zone, |
1178 | zone->zone_pgdat->node_id); |
1179 | else |
1180 | ret = try_to_free_mem_cgroup_pages(victim, gfp_mask, |
1181 | noswap, get_swappiness(victim)); |
1182 | css_put(&victim->css); |
1183 | /* |
1184 | * At shrinking usage, we can't check we should stop here or |
1185 | * reclaim more. It's depends on callers. last_scanned_child |
1186 | * will work enough for keeping fairness under tree. |
1187 | */ |
1188 | if (shrink) |
1189 | return ret; |
1190 | total += ret; |
1191 | if (check_soft) { |
1192 | if (res_counter_check_under_soft_limit(&root_mem->res)) |
1193 | return total; |
1194 | } else if (mem_cgroup_check_under_limit(root_mem)) |
1195 | return 1 + total; |
1196 | } |
1197 | return total; |
1198 | } |
1199 | |
1200 | bool mem_cgroup_oom_called(struct task_struct *task) |
1201 | { |
1202 | bool ret = false; |
1203 | struct mem_cgroup *mem; |
1204 | struct mm_struct *mm; |
1205 | |
1206 | rcu_read_lock(); |
1207 | mm = task->mm; |
1208 | if (!mm) |
1209 | mm = &init_mm; |
1210 | mem = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
1211 | if (mem && time_before(jiffies, mem->last_oom_jiffies + HZ/10)) |
1212 | ret = true; |
1213 | rcu_read_unlock(); |
1214 | return ret; |
1215 | } |
1216 | |
1217 | static int record_last_oom_cb(struct mem_cgroup *mem, void *data) |
1218 | { |
1219 | mem->last_oom_jiffies = jiffies; |
1220 | return 0; |
1221 | } |
1222 | |
1223 | static void record_last_oom(struct mem_cgroup *mem) |
1224 | { |
1225 | mem_cgroup_walk_tree(mem, NULL, record_last_oom_cb); |
1226 | } |
1227 | |
1228 | /* |
1229 | * Currently used to update mapped file statistics, but the routine can be |
1230 | * generalized to update other statistics as well. |
1231 | */ |
1232 | void mem_cgroup_update_mapped_file_stat(struct page *page, int val) |
1233 | { |
1234 | struct mem_cgroup *mem; |
1235 | struct mem_cgroup_stat *stat; |
1236 | struct mem_cgroup_stat_cpu *cpustat; |
1237 | int cpu; |
1238 | struct page_cgroup *pc; |
1239 | |
1240 | if (!page_is_file_cache(page)) |
1241 | return; |
1242 | |
1243 | pc = lookup_page_cgroup(page); |
1244 | if (unlikely(!pc)) |
1245 | return; |
1246 | |
1247 | lock_page_cgroup(pc); |
1248 | mem = pc->mem_cgroup; |
1249 | if (!mem) |
1250 | goto done; |
1251 | |
1252 | if (!PageCgroupUsed(pc)) |
1253 | goto done; |
1254 | |
1255 | /* |
1256 | * Preemption is already disabled, we don't need get_cpu() |
1257 | */ |
1258 | cpu = smp_processor_id(); |
1259 | stat = &mem->stat; |
1260 | cpustat = &stat->cpustat[cpu]; |
1261 | |
1262 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, val); |
1263 | done: |
1264 | unlock_page_cgroup(pc); |
1265 | } |
1266 | |
1267 | /* |
1268 | * Unlike exported interface, "oom" parameter is added. if oom==true, |
1269 | * oom-killer can be invoked. |
1270 | */ |
1271 | static int __mem_cgroup_try_charge(struct mm_struct *mm, |
1272 | gfp_t gfp_mask, struct mem_cgroup **memcg, |
1273 | bool oom, struct page *page) |
1274 | { |
1275 | struct mem_cgroup *mem, *mem_over_limit; |
1276 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
1277 | struct res_counter *fail_res; |
1278 | |
1279 | if (unlikely(test_thread_flag(TIF_MEMDIE))) { |
1280 | /* Don't account this! */ |
1281 | *memcg = NULL; |
1282 | return 0; |
1283 | } |
1284 | |
1285 | /* |
1286 | * We always charge the cgroup the mm_struct belongs to. |
1287 | * The mm_struct's mem_cgroup changes on task migration if the |
1288 | * thread group leader migrates. It's possible that mm is not |
1289 | * set, if so charge the init_mm (happens for pagecache usage). |
1290 | */ |
1291 | mem = *memcg; |
1292 | if (likely(!mem)) { |
1293 | mem = try_get_mem_cgroup_from_mm(mm); |
1294 | *memcg = mem; |
1295 | } else { |
1296 | css_get(&mem->css); |
1297 | } |
1298 | if (unlikely(!mem)) |
1299 | return 0; |
1300 | |
1301 | VM_BUG_ON(css_is_removed(&mem->css)); |
1302 | |
1303 | while (1) { |
1304 | int ret = 0; |
1305 | unsigned long flags = 0; |
1306 | |
1307 | if (mem_cgroup_is_root(mem)) |
1308 | goto done; |
1309 | ret = res_counter_charge(&mem->res, PAGE_SIZE, &fail_res); |
1310 | if (likely(!ret)) { |
1311 | if (!do_swap_account) |
1312 | break; |
1313 | ret = res_counter_charge(&mem->memsw, PAGE_SIZE, |
1314 | &fail_res); |
1315 | if (likely(!ret)) |
1316 | break; |
1317 | /* mem+swap counter fails */ |
1318 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1319 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
1320 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1321 | memsw); |
1322 | } else |
1323 | /* mem counter fails */ |
1324 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, |
1325 | res); |
1326 | |
1327 | if (!(gfp_mask & __GFP_WAIT)) |
1328 | goto nomem; |
1329 | |
1330 | ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL, |
1331 | gfp_mask, flags); |
1332 | if (ret) |
1333 | continue; |
1334 | |
1335 | /* |
1336 | * try_to_free_mem_cgroup_pages() might not give us a full |
1337 | * picture of reclaim. Some pages are reclaimed and might be |
1338 | * moved to swap cache or just unmapped from the cgroup. |
1339 | * Check the limit again to see if the reclaim reduced the |
1340 | * current usage of the cgroup before giving up |
1341 | * |
1342 | */ |
1343 | if (mem_cgroup_check_under_limit(mem_over_limit)) |
1344 | continue; |
1345 | |
1346 | if (!nr_retries--) { |
1347 | if (oom) { |
1348 | mutex_lock(&memcg_tasklist); |
1349 | mem_cgroup_out_of_memory(mem_over_limit, gfp_mask); |
1350 | mutex_unlock(&memcg_tasklist); |
1351 | record_last_oom(mem_over_limit); |
1352 | } |
1353 | goto nomem; |
1354 | } |
1355 | } |
1356 | /* |
1357 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. |
1358 | * if they exceeds softlimit. |
1359 | */ |
1360 | if (mem_cgroup_soft_limit_check(mem)) |
1361 | mem_cgroup_update_tree(mem, page); |
1362 | done: |
1363 | return 0; |
1364 | nomem: |
1365 | css_put(&mem->css); |
1366 | return -ENOMEM; |
1367 | } |
1368 | |
1369 | /* |
1370 | * A helper function to get mem_cgroup from ID. must be called under |
1371 | * rcu_read_lock(). The caller must check css_is_removed() or some if |
1372 | * it's concern. (dropping refcnt from swap can be called against removed |
1373 | * memcg.) |
1374 | */ |
1375 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) |
1376 | { |
1377 | struct cgroup_subsys_state *css; |
1378 | |
1379 | /* ID 0 is unused ID */ |
1380 | if (!id) |
1381 | return NULL; |
1382 | css = css_lookup(&mem_cgroup_subsys, id); |
1383 | if (!css) |
1384 | return NULL; |
1385 | return container_of(css, struct mem_cgroup, css); |
1386 | } |
1387 | |
1388 | static struct mem_cgroup *try_get_mem_cgroup_from_swapcache(struct page *page) |
1389 | { |
1390 | struct mem_cgroup *mem; |
1391 | struct page_cgroup *pc; |
1392 | unsigned short id; |
1393 | swp_entry_t ent; |
1394 | |
1395 | VM_BUG_ON(!PageLocked(page)); |
1396 | |
1397 | if (!PageSwapCache(page)) |
1398 | return NULL; |
1399 | |
1400 | pc = lookup_page_cgroup(page); |
1401 | lock_page_cgroup(pc); |
1402 | if (PageCgroupUsed(pc)) { |
1403 | mem = pc->mem_cgroup; |
1404 | if (mem && !css_tryget(&mem->css)) |
1405 | mem = NULL; |
1406 | } else { |
1407 | ent.val = page_private(page); |
1408 | id = lookup_swap_cgroup(ent); |
1409 | rcu_read_lock(); |
1410 | mem = mem_cgroup_lookup(id); |
1411 | if (mem && !css_tryget(&mem->css)) |
1412 | mem = NULL; |
1413 | rcu_read_unlock(); |
1414 | } |
1415 | unlock_page_cgroup(pc); |
1416 | return mem; |
1417 | } |
1418 | |
1419 | /* |
1420 | * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be |
1421 | * USED state. If already USED, uncharge and return. |
1422 | */ |
1423 | |
1424 | static void __mem_cgroup_commit_charge(struct mem_cgroup *mem, |
1425 | struct page_cgroup *pc, |
1426 | enum charge_type ctype) |
1427 | { |
1428 | /* try_charge() can return NULL to *memcg, taking care of it. */ |
1429 | if (!mem) |
1430 | return; |
1431 | |
1432 | lock_page_cgroup(pc); |
1433 | if (unlikely(PageCgroupUsed(pc))) { |
1434 | unlock_page_cgroup(pc); |
1435 | if (!mem_cgroup_is_root(mem)) { |
1436 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1437 | if (do_swap_account) |
1438 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); |
1439 | } |
1440 | css_put(&mem->css); |
1441 | return; |
1442 | } |
1443 | |
1444 | pc->mem_cgroup = mem; |
1445 | /* |
1446 | * We access a page_cgroup asynchronously without lock_page_cgroup(). |
1447 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup |
1448 | * is accessed after testing USED bit. To make pc->mem_cgroup visible |
1449 | * before USED bit, we need memory barrier here. |
1450 | * See mem_cgroup_add_lru_list(), etc. |
1451 | */ |
1452 | smp_wmb(); |
1453 | switch (ctype) { |
1454 | case MEM_CGROUP_CHARGE_TYPE_CACHE: |
1455 | case MEM_CGROUP_CHARGE_TYPE_SHMEM: |
1456 | SetPageCgroupCache(pc); |
1457 | SetPageCgroupUsed(pc); |
1458 | break; |
1459 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: |
1460 | ClearPageCgroupCache(pc); |
1461 | SetPageCgroupUsed(pc); |
1462 | break; |
1463 | default: |
1464 | break; |
1465 | } |
1466 | |
1467 | mem_cgroup_charge_statistics(mem, pc, true); |
1468 | |
1469 | unlock_page_cgroup(pc); |
1470 | } |
1471 | |
1472 | /** |
1473 | * mem_cgroup_move_account - move account of the page |
1474 | * @pc: page_cgroup of the page. |
1475 | * @from: mem_cgroup which the page is moved from. |
1476 | * @to: mem_cgroup which the page is moved to. @from != @to. |
1477 | * |
1478 | * The caller must confirm following. |
1479 | * - page is not on LRU (isolate_page() is useful.) |
1480 | * |
1481 | * returns 0 at success, |
1482 | * returns -EBUSY when lock is busy or "pc" is unstable. |
1483 | * |
1484 | * This function does "uncharge" from old cgroup but doesn't do "charge" to |
1485 | * new cgroup. It should be done by a caller. |
1486 | */ |
1487 | |
1488 | static int mem_cgroup_move_account(struct page_cgroup *pc, |
1489 | struct mem_cgroup *from, struct mem_cgroup *to) |
1490 | { |
1491 | struct mem_cgroup_per_zone *from_mz, *to_mz; |
1492 | int nid, zid; |
1493 | int ret = -EBUSY; |
1494 | struct page *page; |
1495 | int cpu; |
1496 | struct mem_cgroup_stat *stat; |
1497 | struct mem_cgroup_stat_cpu *cpustat; |
1498 | |
1499 | VM_BUG_ON(from == to); |
1500 | VM_BUG_ON(PageLRU(pc->page)); |
1501 | |
1502 | nid = page_cgroup_nid(pc); |
1503 | zid = page_cgroup_zid(pc); |
1504 | from_mz = mem_cgroup_zoneinfo(from, nid, zid); |
1505 | to_mz = mem_cgroup_zoneinfo(to, nid, zid); |
1506 | |
1507 | if (!trylock_page_cgroup(pc)) |
1508 | return ret; |
1509 | |
1510 | if (!PageCgroupUsed(pc)) |
1511 | goto out; |
1512 | |
1513 | if (pc->mem_cgroup != from) |
1514 | goto out; |
1515 | |
1516 | if (!mem_cgroup_is_root(from)) |
1517 | res_counter_uncharge(&from->res, PAGE_SIZE); |
1518 | mem_cgroup_charge_statistics(from, pc, false); |
1519 | |
1520 | page = pc->page; |
1521 | if (page_is_file_cache(page) && page_mapped(page)) { |
1522 | cpu = smp_processor_id(); |
1523 | /* Update mapped_file data for mem_cgroup "from" */ |
1524 | stat = &from->stat; |
1525 | cpustat = &stat->cpustat[cpu]; |
1526 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, |
1527 | -1); |
1528 | |
1529 | /* Update mapped_file data for mem_cgroup "to" */ |
1530 | stat = &to->stat; |
1531 | cpustat = &stat->cpustat[cpu]; |
1532 | __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, |
1533 | 1); |
1534 | } |
1535 | |
1536 | if (do_swap_account && !mem_cgroup_is_root(from)) |
1537 | res_counter_uncharge(&from->memsw, PAGE_SIZE); |
1538 | css_put(&from->css); |
1539 | |
1540 | css_get(&to->css); |
1541 | pc->mem_cgroup = to; |
1542 | mem_cgroup_charge_statistics(to, pc, true); |
1543 | ret = 0; |
1544 | out: |
1545 | unlock_page_cgroup(pc); |
1546 | /* |
1547 | * We charges against "to" which may not have any tasks. Then, "to" |
1548 | * can be under rmdir(). But in current implementation, caller of |
1549 | * this function is just force_empty() and it's garanteed that |
1550 | * "to" is never removed. So, we don't check rmdir status here. |
1551 | */ |
1552 | return ret; |
1553 | } |
1554 | |
1555 | /* |
1556 | * move charges to its parent. |
1557 | */ |
1558 | |
1559 | static int mem_cgroup_move_parent(struct page_cgroup *pc, |
1560 | struct mem_cgroup *child, |
1561 | gfp_t gfp_mask) |
1562 | { |
1563 | struct page *page = pc->page; |
1564 | struct cgroup *cg = child->css.cgroup; |
1565 | struct cgroup *pcg = cg->parent; |
1566 | struct mem_cgroup *parent; |
1567 | int ret; |
1568 | |
1569 | /* Is ROOT ? */ |
1570 | if (!pcg) |
1571 | return -EINVAL; |
1572 | |
1573 | |
1574 | parent = mem_cgroup_from_cont(pcg); |
1575 | |
1576 | |
1577 | ret = __mem_cgroup_try_charge(NULL, gfp_mask, &parent, false, page); |
1578 | if (ret || !parent) |
1579 | return ret; |
1580 | |
1581 | if (!get_page_unless_zero(page)) { |
1582 | ret = -EBUSY; |
1583 | goto uncharge; |
1584 | } |
1585 | |
1586 | ret = isolate_lru_page(page); |
1587 | |
1588 | if (ret) |
1589 | goto cancel; |
1590 | |
1591 | ret = mem_cgroup_move_account(pc, child, parent); |
1592 | |
1593 | putback_lru_page(page); |
1594 | if (!ret) { |
1595 | put_page(page); |
1596 | /* drop extra refcnt by try_charge() */ |
1597 | css_put(&parent->css); |
1598 | return 0; |
1599 | } |
1600 | |
1601 | cancel: |
1602 | put_page(page); |
1603 | uncharge: |
1604 | /* drop extra refcnt by try_charge() */ |
1605 | css_put(&parent->css); |
1606 | /* uncharge if move fails */ |
1607 | if (!mem_cgroup_is_root(parent)) { |
1608 | res_counter_uncharge(&parent->res, PAGE_SIZE); |
1609 | if (do_swap_account) |
1610 | res_counter_uncharge(&parent->memsw, PAGE_SIZE); |
1611 | } |
1612 | return ret; |
1613 | } |
1614 | |
1615 | /* |
1616 | * Charge the memory controller for page usage. |
1617 | * Return |
1618 | * 0 if the charge was successful |
1619 | * < 0 if the cgroup is over its limit |
1620 | */ |
1621 | static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm, |
1622 | gfp_t gfp_mask, enum charge_type ctype, |
1623 | struct mem_cgroup *memcg) |
1624 | { |
1625 | struct mem_cgroup *mem; |
1626 | struct page_cgroup *pc; |
1627 | int ret; |
1628 | |
1629 | pc = lookup_page_cgroup(page); |
1630 | /* can happen at boot */ |
1631 | if (unlikely(!pc)) |
1632 | return 0; |
1633 | prefetchw(pc); |
1634 | |
1635 | mem = memcg; |
1636 | ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, true, page); |
1637 | if (ret || !mem) |
1638 | return ret; |
1639 | |
1640 | __mem_cgroup_commit_charge(mem, pc, ctype); |
1641 | return 0; |
1642 | } |
1643 | |
1644 | int mem_cgroup_newpage_charge(struct page *page, |
1645 | struct mm_struct *mm, gfp_t gfp_mask) |
1646 | { |
1647 | if (mem_cgroup_disabled()) |
1648 | return 0; |
1649 | if (PageCompound(page)) |
1650 | return 0; |
1651 | /* |
1652 | * If already mapped, we don't have to account. |
1653 | * If page cache, page->mapping has address_space. |
1654 | * But page->mapping may have out-of-use anon_vma pointer, |
1655 | * detecit it by PageAnon() check. newly-mapped-anon's page->mapping |
1656 | * is NULL. |
1657 | */ |
1658 | if (page_mapped(page) || (page->mapping && !PageAnon(page))) |
1659 | return 0; |
1660 | if (unlikely(!mm)) |
1661 | mm = &init_mm; |
1662 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
1663 | MEM_CGROUP_CHARGE_TYPE_MAPPED, NULL); |
1664 | } |
1665 | |
1666 | static void |
1667 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, |
1668 | enum charge_type ctype); |
1669 | |
1670 | int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm, |
1671 | gfp_t gfp_mask) |
1672 | { |
1673 | struct mem_cgroup *mem = NULL; |
1674 | int ret; |
1675 | |
1676 | if (mem_cgroup_disabled()) |
1677 | return 0; |
1678 | if (PageCompound(page)) |
1679 | return 0; |
1680 | /* |
1681 | * Corner case handling. This is called from add_to_page_cache() |
1682 | * in usual. But some FS (shmem) precharges this page before calling it |
1683 | * and call add_to_page_cache() with GFP_NOWAIT. |
1684 | * |
1685 | * For GFP_NOWAIT case, the page may be pre-charged before calling |
1686 | * add_to_page_cache(). (See shmem.c) check it here and avoid to call |
1687 | * charge twice. (It works but has to pay a bit larger cost.) |
1688 | * And when the page is SwapCache, it should take swap information |
1689 | * into account. This is under lock_page() now. |
1690 | */ |
1691 | if (!(gfp_mask & __GFP_WAIT)) { |
1692 | struct page_cgroup *pc; |
1693 | |
1694 | |
1695 | pc = lookup_page_cgroup(page); |
1696 | if (!pc) |
1697 | return 0; |
1698 | lock_page_cgroup(pc); |
1699 | if (PageCgroupUsed(pc)) { |
1700 | unlock_page_cgroup(pc); |
1701 | return 0; |
1702 | } |
1703 | unlock_page_cgroup(pc); |
1704 | } |
1705 | |
1706 | if (unlikely(!mm && !mem)) |
1707 | mm = &init_mm; |
1708 | |
1709 | if (page_is_file_cache(page)) |
1710 | return mem_cgroup_charge_common(page, mm, gfp_mask, |
1711 | MEM_CGROUP_CHARGE_TYPE_CACHE, NULL); |
1712 | |
1713 | /* shmem */ |
1714 | if (PageSwapCache(page)) { |
1715 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
1716 | if (!ret) |
1717 | __mem_cgroup_commit_charge_swapin(page, mem, |
1718 | MEM_CGROUP_CHARGE_TYPE_SHMEM); |
1719 | } else |
1720 | ret = mem_cgroup_charge_common(page, mm, gfp_mask, |
1721 | MEM_CGROUP_CHARGE_TYPE_SHMEM, mem); |
1722 | |
1723 | return ret; |
1724 | } |
1725 | |
1726 | /* |
1727 | * While swap-in, try_charge -> commit or cancel, the page is locked. |
1728 | * And when try_charge() successfully returns, one refcnt to memcg without |
1729 | * struct page_cgroup is aquired. This refcnt will be cumsumed by |
1730 | * "commit()" or removed by "cancel()" |
1731 | */ |
1732 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
1733 | struct page *page, |
1734 | gfp_t mask, struct mem_cgroup **ptr) |
1735 | { |
1736 | struct mem_cgroup *mem; |
1737 | int ret; |
1738 | |
1739 | if (mem_cgroup_disabled()) |
1740 | return 0; |
1741 | |
1742 | if (!do_swap_account) |
1743 | goto charge_cur_mm; |
1744 | /* |
1745 | * A racing thread's fault, or swapoff, may have already updated |
1746 | * the pte, and even removed page from swap cache: return success |
1747 | * to go on to do_swap_page()'s pte_same() test, which should fail. |
1748 | */ |
1749 | if (!PageSwapCache(page)) |
1750 | return 0; |
1751 | mem = try_get_mem_cgroup_from_swapcache(page); |
1752 | if (!mem) |
1753 | goto charge_cur_mm; |
1754 | *ptr = mem; |
1755 | ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, page); |
1756 | /* drop extra refcnt from tryget */ |
1757 | css_put(&mem->css); |
1758 | return ret; |
1759 | charge_cur_mm: |
1760 | if (unlikely(!mm)) |
1761 | mm = &init_mm; |
1762 | return __mem_cgroup_try_charge(mm, mask, ptr, true, page); |
1763 | } |
1764 | |
1765 | static void |
1766 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr, |
1767 | enum charge_type ctype) |
1768 | { |
1769 | struct page_cgroup *pc; |
1770 | |
1771 | if (mem_cgroup_disabled()) |
1772 | return; |
1773 | if (!ptr) |
1774 | return; |
1775 | cgroup_exclude_rmdir(&ptr->css); |
1776 | pc = lookup_page_cgroup(page); |
1777 | mem_cgroup_lru_del_before_commit_swapcache(page); |
1778 | __mem_cgroup_commit_charge(ptr, pc, ctype); |
1779 | mem_cgroup_lru_add_after_commit_swapcache(page); |
1780 | /* |
1781 | * Now swap is on-memory. This means this page may be |
1782 | * counted both as mem and swap....double count. |
1783 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
1784 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() |
1785 | * may call delete_from_swap_cache() before reach here. |
1786 | */ |
1787 | if (do_swap_account && PageSwapCache(page)) { |
1788 | swp_entry_t ent = {.val = page_private(page)}; |
1789 | unsigned short id; |
1790 | struct mem_cgroup *memcg; |
1791 | |
1792 | id = swap_cgroup_record(ent, 0); |
1793 | rcu_read_lock(); |
1794 | memcg = mem_cgroup_lookup(id); |
1795 | if (memcg) { |
1796 | /* |
1797 | * This recorded memcg can be obsolete one. So, avoid |
1798 | * calling css_tryget |
1799 | */ |
1800 | if (!mem_cgroup_is_root(memcg)) |
1801 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
1802 | mem_cgroup_swap_statistics(memcg, false); |
1803 | mem_cgroup_put(memcg); |
1804 | } |
1805 | rcu_read_unlock(); |
1806 | } |
1807 | /* |
1808 | * At swapin, we may charge account against cgroup which has no tasks. |
1809 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
1810 | * In that case, we need to call pre_destroy() again. check it here. |
1811 | */ |
1812 | cgroup_release_and_wakeup_rmdir(&ptr->css); |
1813 | } |
1814 | |
1815 | void mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr) |
1816 | { |
1817 | __mem_cgroup_commit_charge_swapin(page, ptr, |
1818 | MEM_CGROUP_CHARGE_TYPE_MAPPED); |
1819 | } |
1820 | |
1821 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem) |
1822 | { |
1823 | if (mem_cgroup_disabled()) |
1824 | return; |
1825 | if (!mem) |
1826 | return; |
1827 | if (!mem_cgroup_is_root(mem)) { |
1828 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1829 | if (do_swap_account) |
1830 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); |
1831 | } |
1832 | css_put(&mem->css); |
1833 | } |
1834 | |
1835 | |
1836 | /* |
1837 | * uncharge if !page_mapped(page) |
1838 | */ |
1839 | static struct mem_cgroup * |
1840 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype) |
1841 | { |
1842 | struct page_cgroup *pc; |
1843 | struct mem_cgroup *mem = NULL; |
1844 | struct mem_cgroup_per_zone *mz; |
1845 | |
1846 | if (mem_cgroup_disabled()) |
1847 | return NULL; |
1848 | |
1849 | if (PageSwapCache(page)) |
1850 | return NULL; |
1851 | |
1852 | /* |
1853 | * Check if our page_cgroup is valid |
1854 | */ |
1855 | pc = lookup_page_cgroup(page); |
1856 | if (unlikely(!pc || !PageCgroupUsed(pc))) |
1857 | return NULL; |
1858 | |
1859 | lock_page_cgroup(pc); |
1860 | |
1861 | mem = pc->mem_cgroup; |
1862 | |
1863 | if (!PageCgroupUsed(pc)) |
1864 | goto unlock_out; |
1865 | |
1866 | switch (ctype) { |
1867 | case MEM_CGROUP_CHARGE_TYPE_MAPPED: |
1868 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
1869 | if (page_mapped(page)) |
1870 | goto unlock_out; |
1871 | break; |
1872 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: |
1873 | if (!PageAnon(page)) { /* Shared memory */ |
1874 | if (page->mapping && !page_is_file_cache(page)) |
1875 | goto unlock_out; |
1876 | } else if (page_mapped(page)) /* Anon */ |
1877 | goto unlock_out; |
1878 | break; |
1879 | default: |
1880 | break; |
1881 | } |
1882 | |
1883 | if (!mem_cgroup_is_root(mem)) { |
1884 | res_counter_uncharge(&mem->res, PAGE_SIZE); |
1885 | if (do_swap_account && |
1886 | (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT)) |
1887 | res_counter_uncharge(&mem->memsw, PAGE_SIZE); |
1888 | } |
1889 | if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
1890 | mem_cgroup_swap_statistics(mem, true); |
1891 | mem_cgroup_charge_statistics(mem, pc, false); |
1892 | |
1893 | ClearPageCgroupUsed(pc); |
1894 | /* |
1895 | * pc->mem_cgroup is not cleared here. It will be accessed when it's |
1896 | * freed from LRU. This is safe because uncharged page is expected not |
1897 | * to be reused (freed soon). Exception is SwapCache, it's handled by |
1898 | * special functions. |
1899 | */ |
1900 | |
1901 | mz = page_cgroup_zoneinfo(pc); |
1902 | unlock_page_cgroup(pc); |
1903 | |
1904 | if (mem_cgroup_soft_limit_check(mem)) |
1905 | mem_cgroup_update_tree(mem, page); |
1906 | /* at swapout, this memcg will be accessed to record to swap */ |
1907 | if (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
1908 | css_put(&mem->css); |
1909 | |
1910 | return mem; |
1911 | |
1912 | unlock_out: |
1913 | unlock_page_cgroup(pc); |
1914 | return NULL; |
1915 | } |
1916 | |
1917 | void mem_cgroup_uncharge_page(struct page *page) |
1918 | { |
1919 | /* early check. */ |
1920 | if (page_mapped(page)) |
1921 | return; |
1922 | if (page->mapping && !PageAnon(page)) |
1923 | return; |
1924 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_MAPPED); |
1925 | } |
1926 | |
1927 | void mem_cgroup_uncharge_cache_page(struct page *page) |
1928 | { |
1929 | VM_BUG_ON(page_mapped(page)); |
1930 | VM_BUG_ON(page->mapping); |
1931 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE); |
1932 | } |
1933 | |
1934 | #ifdef CONFIG_SWAP |
1935 | /* |
1936 | * called after __delete_from_swap_cache() and drop "page" account. |
1937 | * memcg information is recorded to swap_cgroup of "ent" |
1938 | */ |
1939 | void |
1940 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) |
1941 | { |
1942 | struct mem_cgroup *memcg; |
1943 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
1944 | |
1945 | if (!swapout) /* this was a swap cache but the swap is unused ! */ |
1946 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; |
1947 | |
1948 | memcg = __mem_cgroup_uncharge_common(page, ctype); |
1949 | |
1950 | /* record memcg information */ |
1951 | if (do_swap_account && swapout && memcg) { |
1952 | swap_cgroup_record(ent, css_id(&memcg->css)); |
1953 | mem_cgroup_get(memcg); |
1954 | } |
1955 | if (swapout && memcg) |
1956 | css_put(&memcg->css); |
1957 | } |
1958 | #endif |
1959 | |
1960 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
1961 | /* |
1962 | * called from swap_entry_free(). remove record in swap_cgroup and |
1963 | * uncharge "memsw" account. |
1964 | */ |
1965 | void mem_cgroup_uncharge_swap(swp_entry_t ent) |
1966 | { |
1967 | struct mem_cgroup *memcg; |
1968 | unsigned short id; |
1969 | |
1970 | if (!do_swap_account) |
1971 | return; |
1972 | |
1973 | id = swap_cgroup_record(ent, 0); |
1974 | rcu_read_lock(); |
1975 | memcg = mem_cgroup_lookup(id); |
1976 | if (memcg) { |
1977 | /* |
1978 | * We uncharge this because swap is freed. |
1979 | * This memcg can be obsolete one. We avoid calling css_tryget |
1980 | */ |
1981 | if (!mem_cgroup_is_root(memcg)) |
1982 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
1983 | mem_cgroup_swap_statistics(memcg, false); |
1984 | mem_cgroup_put(memcg); |
1985 | } |
1986 | rcu_read_unlock(); |
1987 | } |
1988 | #endif |
1989 | |
1990 | /* |
1991 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
1992 | * page belongs to. |
1993 | */ |
1994 | int mem_cgroup_prepare_migration(struct page *page, struct mem_cgroup **ptr) |
1995 | { |
1996 | struct page_cgroup *pc; |
1997 | struct mem_cgroup *mem = NULL; |
1998 | int ret = 0; |
1999 | |
2000 | if (mem_cgroup_disabled()) |
2001 | return 0; |
2002 | |
2003 | pc = lookup_page_cgroup(page); |
2004 | lock_page_cgroup(pc); |
2005 | if (PageCgroupUsed(pc)) { |
2006 | mem = pc->mem_cgroup; |
2007 | css_get(&mem->css); |
2008 | } |
2009 | unlock_page_cgroup(pc); |
2010 | |
2011 | if (mem) { |
2012 | ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false, |
2013 | page); |
2014 | css_put(&mem->css); |
2015 | } |
2016 | *ptr = mem; |
2017 | return ret; |
2018 | } |
2019 | |
2020 | /* remove redundant charge if migration failed*/ |
2021 | void mem_cgroup_end_migration(struct mem_cgroup *mem, |
2022 | struct page *oldpage, struct page *newpage) |
2023 | { |
2024 | struct page *target, *unused; |
2025 | struct page_cgroup *pc; |
2026 | enum charge_type ctype; |
2027 | |
2028 | if (!mem) |
2029 | return; |
2030 | cgroup_exclude_rmdir(&mem->css); |
2031 | /* at migration success, oldpage->mapping is NULL. */ |
2032 | if (oldpage->mapping) { |
2033 | target = oldpage; |
2034 | unused = NULL; |
2035 | } else { |
2036 | target = newpage; |
2037 | unused = oldpage; |
2038 | } |
2039 | |
2040 | if (PageAnon(target)) |
2041 | ctype = MEM_CGROUP_CHARGE_TYPE_MAPPED; |
2042 | else if (page_is_file_cache(target)) |
2043 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
2044 | else |
2045 | ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM; |
2046 | |
2047 | /* unused page is not on radix-tree now. */ |
2048 | if (unused) |
2049 | __mem_cgroup_uncharge_common(unused, ctype); |
2050 | |
2051 | pc = lookup_page_cgroup(target); |
2052 | /* |
2053 | * __mem_cgroup_commit_charge() check PCG_USED bit of page_cgroup. |
2054 | * So, double-counting is effectively avoided. |
2055 | */ |
2056 | __mem_cgroup_commit_charge(mem, pc, ctype); |
2057 | |
2058 | /* |
2059 | * Both of oldpage and newpage are still under lock_page(). |
2060 | * Then, we don't have to care about race in radix-tree. |
2061 | * But we have to be careful that this page is unmapped or not. |
2062 | * |
2063 | * There is a case for !page_mapped(). At the start of |
2064 | * migration, oldpage was mapped. But now, it's zapped. |
2065 | * But we know *target* page is not freed/reused under us. |
2066 | * mem_cgroup_uncharge_page() does all necessary checks. |
2067 | */ |
2068 | if (ctype == MEM_CGROUP_CHARGE_TYPE_MAPPED) |
2069 | mem_cgroup_uncharge_page(target); |
2070 | /* |
2071 | * At migration, we may charge account against cgroup which has no tasks |
2072 | * So, rmdir()->pre_destroy() can be called while we do this charge. |
2073 | * In that case, we need to call pre_destroy() again. check it here. |
2074 | */ |
2075 | cgroup_release_and_wakeup_rmdir(&mem->css); |
2076 | } |
2077 | |
2078 | /* |
2079 | * A call to try to shrink memory usage on charge failure at shmem's swapin. |
2080 | * Calling hierarchical_reclaim is not enough because we should update |
2081 | * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM. |
2082 | * Moreover considering hierarchy, we should reclaim from the mem_over_limit, |
2083 | * not from the memcg which this page would be charged to. |
2084 | * try_charge_swapin does all of these works properly. |
2085 | */ |
2086 | int mem_cgroup_shmem_charge_fallback(struct page *page, |
2087 | struct mm_struct *mm, |
2088 | gfp_t gfp_mask) |
2089 | { |
2090 | struct mem_cgroup *mem = NULL; |
2091 | int ret; |
2092 | |
2093 | if (mem_cgroup_disabled()) |
2094 | return 0; |
2095 | |
2096 | ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &mem); |
2097 | if (!ret) |
2098 | mem_cgroup_cancel_charge_swapin(mem); /* it does !mem check */ |
2099 | |
2100 | return ret; |
2101 | } |
2102 | |
2103 | static DEFINE_MUTEX(set_limit_mutex); |
2104 | |
2105 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
2106 | unsigned long long val) |
2107 | { |
2108 | int retry_count; |
2109 | int progress; |
2110 | u64 memswlimit; |
2111 | int ret = 0; |
2112 | int children = mem_cgroup_count_children(memcg); |
2113 | u64 curusage, oldusage; |
2114 | |
2115 | /* |
2116 | * For keeping hierarchical_reclaim simple, how long we should retry |
2117 | * is depends on callers. We set our retry-count to be function |
2118 | * of # of children which we should visit in this loop. |
2119 | */ |
2120 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; |
2121 | |
2122 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
2123 | |
2124 | while (retry_count) { |
2125 | if (signal_pending(current)) { |
2126 | ret = -EINTR; |
2127 | break; |
2128 | } |
2129 | /* |
2130 | * Rather than hide all in some function, I do this in |
2131 | * open coded manner. You see what this really does. |
2132 | * We have to guarantee mem->res.limit < mem->memsw.limit. |
2133 | */ |
2134 | mutex_lock(&set_limit_mutex); |
2135 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
2136 | if (memswlimit < val) { |
2137 | ret = -EINVAL; |
2138 | mutex_unlock(&set_limit_mutex); |
2139 | break; |
2140 | } |
2141 | ret = res_counter_set_limit(&memcg->res, val); |
2142 | if (!ret) { |
2143 | if (memswlimit == val) |
2144 | memcg->memsw_is_minimum = true; |
2145 | else |
2146 | memcg->memsw_is_minimum = false; |
2147 | } |
2148 | mutex_unlock(&set_limit_mutex); |
2149 | |
2150 | if (!ret) |
2151 | break; |
2152 | |
2153 | progress = mem_cgroup_hierarchical_reclaim(memcg, NULL, |
2154 | GFP_KERNEL, |
2155 | MEM_CGROUP_RECLAIM_SHRINK); |
2156 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
2157 | /* Usage is reduced ? */ |
2158 | if (curusage >= oldusage) |
2159 | retry_count--; |
2160 | else |
2161 | oldusage = curusage; |
2162 | } |
2163 | |
2164 | return ret; |
2165 | } |
2166 | |
2167 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
2168 | unsigned long long val) |
2169 | { |
2170 | int retry_count; |
2171 | u64 memlimit, oldusage, curusage; |
2172 | int children = mem_cgroup_count_children(memcg); |
2173 | int ret = -EBUSY; |
2174 | |
2175 | /* see mem_cgroup_resize_res_limit */ |
2176 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
2177 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
2178 | while (retry_count) { |
2179 | if (signal_pending(current)) { |
2180 | ret = -EINTR; |
2181 | break; |
2182 | } |
2183 | /* |
2184 | * Rather than hide all in some function, I do this in |
2185 | * open coded manner. You see what this really does. |
2186 | * We have to guarantee mem->res.limit < mem->memsw.limit. |
2187 | */ |
2188 | mutex_lock(&set_limit_mutex); |
2189 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
2190 | if (memlimit > val) { |
2191 | ret = -EINVAL; |
2192 | mutex_unlock(&set_limit_mutex); |
2193 | break; |
2194 | } |
2195 | ret = res_counter_set_limit(&memcg->memsw, val); |
2196 | if (!ret) { |
2197 | if (memlimit == val) |
2198 | memcg->memsw_is_minimum = true; |
2199 | else |
2200 | memcg->memsw_is_minimum = false; |
2201 | } |
2202 | mutex_unlock(&set_limit_mutex); |
2203 | |
2204 | if (!ret) |
2205 | break; |
2206 | |
2207 | mem_cgroup_hierarchical_reclaim(memcg, NULL, GFP_KERNEL, |
2208 | MEM_CGROUP_RECLAIM_NOSWAP | |
2209 | MEM_CGROUP_RECLAIM_SHRINK); |
2210 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
2211 | /* Usage is reduced ? */ |
2212 | if (curusage >= oldusage) |
2213 | retry_count--; |
2214 | else |
2215 | oldusage = curusage; |
2216 | } |
2217 | return ret; |
2218 | } |
2219 | |
2220 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
2221 | gfp_t gfp_mask, int nid, |
2222 | int zid) |
2223 | { |
2224 | unsigned long nr_reclaimed = 0; |
2225 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
2226 | unsigned long reclaimed; |
2227 | int loop = 0; |
2228 | struct mem_cgroup_tree_per_zone *mctz; |
2229 | unsigned long long excess; |
2230 | |
2231 | if (order > 0) |
2232 | return 0; |
2233 | |
2234 | mctz = soft_limit_tree_node_zone(nid, zid); |
2235 | /* |
2236 | * This loop can run a while, specially if mem_cgroup's continuously |
2237 | * keep exceeding their soft limit and putting the system under |
2238 | * pressure |
2239 | */ |
2240 | do { |
2241 | if (next_mz) |
2242 | mz = next_mz; |
2243 | else |
2244 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
2245 | if (!mz) |
2246 | break; |
2247 | |
2248 | reclaimed = mem_cgroup_hierarchical_reclaim(mz->mem, zone, |
2249 | gfp_mask, |
2250 | MEM_CGROUP_RECLAIM_SOFT); |
2251 | nr_reclaimed += reclaimed; |
2252 | spin_lock(&mctz->lock); |
2253 | |
2254 | /* |
2255 | * If we failed to reclaim anything from this memory cgroup |
2256 | * it is time to move on to the next cgroup |
2257 | */ |
2258 | next_mz = NULL; |
2259 | if (!reclaimed) { |
2260 | do { |
2261 | /* |
2262 | * Loop until we find yet another one. |
2263 | * |
2264 | * By the time we get the soft_limit lock |
2265 | * again, someone might have aded the |
2266 | * group back on the RB tree. Iterate to |
2267 | * make sure we get a different mem. |
2268 | * mem_cgroup_largest_soft_limit_node returns |
2269 | * NULL if no other cgroup is present on |
2270 | * the tree |
2271 | */ |
2272 | next_mz = |
2273 | __mem_cgroup_largest_soft_limit_node(mctz); |
2274 | if (next_mz == mz) { |
2275 | css_put(&next_mz->mem->css); |
2276 | next_mz = NULL; |
2277 | } else /* next_mz == NULL or other memcg */ |
2278 | break; |
2279 | } while (1); |
2280 | } |
2281 | __mem_cgroup_remove_exceeded(mz->mem, mz, mctz); |
2282 | excess = res_counter_soft_limit_excess(&mz->mem->res); |
2283 | /* |
2284 | * One school of thought says that we should not add |
2285 | * back the node to the tree if reclaim returns 0. |
2286 | * But our reclaim could return 0, simply because due |
2287 | * to priority we are exposing a smaller subset of |
2288 | * memory to reclaim from. Consider this as a longer |
2289 | * term TODO. |
2290 | */ |
2291 | /* If excess == 0, no tree ops */ |
2292 | __mem_cgroup_insert_exceeded(mz->mem, mz, mctz, excess); |
2293 | spin_unlock(&mctz->lock); |
2294 | css_put(&mz->mem->css); |
2295 | loop++; |
2296 | /* |
2297 | * Could not reclaim anything and there are no more |
2298 | * mem cgroups to try or we seem to be looping without |
2299 | * reclaiming anything. |
2300 | */ |
2301 | if (!nr_reclaimed && |
2302 | (next_mz == NULL || |
2303 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
2304 | break; |
2305 | } while (!nr_reclaimed); |
2306 | if (next_mz) |
2307 | css_put(&next_mz->mem->css); |
2308 | return nr_reclaimed; |
2309 | } |
2310 | |
2311 | /* |
2312 | * This routine traverse page_cgroup in given list and drop them all. |
2313 | * *And* this routine doesn't reclaim page itself, just removes page_cgroup. |
2314 | */ |
2315 | static int mem_cgroup_force_empty_list(struct mem_cgroup *mem, |
2316 | int node, int zid, enum lru_list lru) |
2317 | { |
2318 | struct zone *zone; |
2319 | struct mem_cgroup_per_zone *mz; |
2320 | struct page_cgroup *pc, *busy; |
2321 | unsigned long flags, loop; |
2322 | struct list_head *list; |
2323 | int ret = 0; |
2324 | |
2325 | zone = &NODE_DATA(node)->node_zones[zid]; |
2326 | mz = mem_cgroup_zoneinfo(mem, node, zid); |
2327 | list = &mz->lists[lru]; |
2328 | |
2329 | loop = MEM_CGROUP_ZSTAT(mz, lru); |
2330 | /* give some margin against EBUSY etc...*/ |
2331 | loop += 256; |
2332 | busy = NULL; |
2333 | while (loop--) { |
2334 | ret = 0; |
2335 | spin_lock_irqsave(&zone->lru_lock, flags); |
2336 | if (list_empty(list)) { |
2337 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2338 | break; |
2339 | } |
2340 | pc = list_entry(list->prev, struct page_cgroup, lru); |
2341 | if (busy == pc) { |
2342 | list_move(&pc->lru, list); |
2343 | busy = 0; |
2344 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2345 | continue; |
2346 | } |
2347 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
2348 | |
2349 | ret = mem_cgroup_move_parent(pc, mem, GFP_KERNEL); |
2350 | if (ret == -ENOMEM) |
2351 | break; |
2352 | |
2353 | if (ret == -EBUSY || ret == -EINVAL) { |
2354 | /* found lock contention or "pc" is obsolete. */ |
2355 | busy = pc; |
2356 | cond_resched(); |
2357 | } else |
2358 | busy = NULL; |
2359 | } |
2360 | |
2361 | if (!ret && !list_empty(list)) |
2362 | return -EBUSY; |
2363 | return ret; |
2364 | } |
2365 | |
2366 | /* |
2367 | * make mem_cgroup's charge to be 0 if there is no task. |
2368 | * This enables deleting this mem_cgroup. |
2369 | */ |
2370 | static int mem_cgroup_force_empty(struct mem_cgroup *mem, bool free_all) |
2371 | { |
2372 | int ret; |
2373 | int node, zid, shrink; |
2374 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
2375 | struct cgroup *cgrp = mem->css.cgroup; |
2376 | |
2377 | css_get(&mem->css); |
2378 | |
2379 | shrink = 0; |
2380 | /* should free all ? */ |
2381 | if (free_all) |
2382 | goto try_to_free; |
2383 | move_account: |
2384 | do { |
2385 | ret = -EBUSY; |
2386 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children)) |
2387 | goto out; |
2388 | ret = -EINTR; |
2389 | if (signal_pending(current)) |
2390 | goto out; |
2391 | /* This is for making all *used* pages to be on LRU. */ |
2392 | lru_add_drain_all(); |
2393 | ret = 0; |
2394 | for_each_node_state(node, N_HIGH_MEMORY) { |
2395 | for (zid = 0; !ret && zid < MAX_NR_ZONES; zid++) { |
2396 | enum lru_list l; |
2397 | for_each_lru(l) { |
2398 | ret = mem_cgroup_force_empty_list(mem, |
2399 | node, zid, l); |
2400 | if (ret) |
2401 | break; |
2402 | } |
2403 | } |
2404 | if (ret) |
2405 | break; |
2406 | } |
2407 | /* it seems parent cgroup doesn't have enough mem */ |
2408 | if (ret == -ENOMEM) |
2409 | goto try_to_free; |
2410 | cond_resched(); |
2411 | /* "ret" should also be checked to ensure all lists are empty. */ |
2412 | } while (mem->res.usage > 0 || ret); |
2413 | out: |
2414 | css_put(&mem->css); |
2415 | return ret; |
2416 | |
2417 | try_to_free: |
2418 | /* returns EBUSY if there is a task or if we come here twice. */ |
2419 | if (cgroup_task_count(cgrp) || !list_empty(&cgrp->children) || shrink) { |
2420 | ret = -EBUSY; |
2421 | goto out; |
2422 | } |
2423 | /* we call try-to-free pages for make this cgroup empty */ |
2424 | lru_add_drain_all(); |
2425 | /* try to free all pages in this cgroup */ |
2426 | shrink = 1; |
2427 | while (nr_retries && mem->res.usage > 0) { |
2428 | int progress; |
2429 | |
2430 | if (signal_pending(current)) { |
2431 | ret = -EINTR; |
2432 | goto out; |
2433 | } |
2434 | progress = try_to_free_mem_cgroup_pages(mem, GFP_KERNEL, |
2435 | false, get_swappiness(mem)); |
2436 | if (!progress) { |
2437 | nr_retries--; |
2438 | /* maybe some writeback is necessary */ |
2439 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
2440 | } |
2441 | |
2442 | } |
2443 | lru_add_drain(); |
2444 | /* try move_account...there may be some *locked* pages. */ |
2445 | goto move_account; |
2446 | } |
2447 | |
2448 | int mem_cgroup_force_empty_write(struct cgroup *cont, unsigned int event) |
2449 | { |
2450 | return mem_cgroup_force_empty(mem_cgroup_from_cont(cont), true); |
2451 | } |
2452 | |
2453 | |
2454 | static u64 mem_cgroup_hierarchy_read(struct cgroup *cont, struct cftype *cft) |
2455 | { |
2456 | return mem_cgroup_from_cont(cont)->use_hierarchy; |
2457 | } |
2458 | |
2459 | static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft, |
2460 | u64 val) |
2461 | { |
2462 | int retval = 0; |
2463 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
2464 | struct cgroup *parent = cont->parent; |
2465 | struct mem_cgroup *parent_mem = NULL; |
2466 | |
2467 | if (parent) |
2468 | parent_mem = mem_cgroup_from_cont(parent); |
2469 | |
2470 | cgroup_lock(); |
2471 | /* |
2472 | * If parent's use_hiearchy is set, we can't make any modifications |
2473 | * in the child subtrees. If it is unset, then the change can |
2474 | * occur, provided the current cgroup has no children. |
2475 | * |
2476 | * For the root cgroup, parent_mem is NULL, we allow value to be |
2477 | * set if there are no children. |
2478 | */ |
2479 | if ((!parent_mem || !parent_mem->use_hierarchy) && |
2480 | (val == 1 || val == 0)) { |
2481 | if (list_empty(&cont->children)) |
2482 | mem->use_hierarchy = val; |
2483 | else |
2484 | retval = -EBUSY; |
2485 | } else |
2486 | retval = -EINVAL; |
2487 | cgroup_unlock(); |
2488 | |
2489 | return retval; |
2490 | } |
2491 | |
2492 | struct mem_cgroup_idx_data { |
2493 | s64 val; |
2494 | enum mem_cgroup_stat_index idx; |
2495 | }; |
2496 | |
2497 | static int |
2498 | mem_cgroup_get_idx_stat(struct mem_cgroup *mem, void *data) |
2499 | { |
2500 | struct mem_cgroup_idx_data *d = data; |
2501 | d->val += mem_cgroup_read_stat(&mem->stat, d->idx); |
2502 | return 0; |
2503 | } |
2504 | |
2505 | static void |
2506 | mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem, |
2507 | enum mem_cgroup_stat_index idx, s64 *val) |
2508 | { |
2509 | struct mem_cgroup_idx_data d; |
2510 | d.idx = idx; |
2511 | d.val = 0; |
2512 | mem_cgroup_walk_tree(mem, &d, mem_cgroup_get_idx_stat); |
2513 | *val = d.val; |
2514 | } |
2515 | |
2516 | static u64 mem_cgroup_read(struct cgroup *cont, struct cftype *cft) |
2517 | { |
2518 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
2519 | u64 idx_val, val; |
2520 | int type, name; |
2521 | |
2522 | type = MEMFILE_TYPE(cft->private); |
2523 | name = MEMFILE_ATTR(cft->private); |
2524 | switch (type) { |
2525 | case _MEM: |
2526 | if (name == RES_USAGE && mem_cgroup_is_root(mem)) { |
2527 | mem_cgroup_get_recursive_idx_stat(mem, |
2528 | MEM_CGROUP_STAT_CACHE, &idx_val); |
2529 | val = idx_val; |
2530 | mem_cgroup_get_recursive_idx_stat(mem, |
2531 | MEM_CGROUP_STAT_RSS, &idx_val); |
2532 | val += idx_val; |
2533 | val <<= PAGE_SHIFT; |
2534 | } else |
2535 | val = res_counter_read_u64(&mem->res, name); |
2536 | break; |
2537 | case _MEMSWAP: |
2538 | if (name == RES_USAGE && mem_cgroup_is_root(mem)) { |
2539 | mem_cgroup_get_recursive_idx_stat(mem, |
2540 | MEM_CGROUP_STAT_CACHE, &idx_val); |
2541 | val = idx_val; |
2542 | mem_cgroup_get_recursive_idx_stat(mem, |
2543 | MEM_CGROUP_STAT_RSS, &idx_val); |
2544 | val += idx_val; |
2545 | mem_cgroup_get_recursive_idx_stat(mem, |
2546 | MEM_CGROUP_STAT_SWAPOUT, &idx_val); |
2547 | val += idx_val; |
2548 | val <<= PAGE_SHIFT; |
2549 | } else |
2550 | val = res_counter_read_u64(&mem->memsw, name); |
2551 | break; |
2552 | default: |
2553 | BUG(); |
2554 | break; |
2555 | } |
2556 | return val; |
2557 | } |
2558 | /* |
2559 | * The user of this function is... |
2560 | * RES_LIMIT. |
2561 | */ |
2562 | static int mem_cgroup_write(struct cgroup *cont, struct cftype *cft, |
2563 | const char *buffer) |
2564 | { |
2565 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cont); |
2566 | int type, name; |
2567 | unsigned long long val; |
2568 | int ret; |
2569 | |
2570 | type = MEMFILE_TYPE(cft->private); |
2571 | name = MEMFILE_ATTR(cft->private); |
2572 | switch (name) { |
2573 | case RES_LIMIT: |
2574 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
2575 | ret = -EINVAL; |
2576 | break; |
2577 | } |
2578 | /* This function does all necessary parse...reuse it */ |
2579 | ret = res_counter_memparse_write_strategy(buffer, &val); |
2580 | if (ret) |
2581 | break; |
2582 | if (type == _MEM) |
2583 | ret = mem_cgroup_resize_limit(memcg, val); |
2584 | else |
2585 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
2586 | break; |
2587 | case RES_SOFT_LIMIT: |
2588 | ret = res_counter_memparse_write_strategy(buffer, &val); |
2589 | if (ret) |
2590 | break; |
2591 | /* |
2592 | * For memsw, soft limits are hard to implement in terms |
2593 | * of semantics, for now, we support soft limits for |
2594 | * control without swap |
2595 | */ |
2596 | if (type == _MEM) |
2597 | ret = res_counter_set_soft_limit(&memcg->res, val); |
2598 | else |
2599 | ret = -EINVAL; |
2600 | break; |
2601 | default: |
2602 | ret = -EINVAL; /* should be BUG() ? */ |
2603 | break; |
2604 | } |
2605 | return ret; |
2606 | } |
2607 | |
2608 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
2609 | unsigned long long *mem_limit, unsigned long long *memsw_limit) |
2610 | { |
2611 | struct cgroup *cgroup; |
2612 | unsigned long long min_limit, min_memsw_limit, tmp; |
2613 | |
2614 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
2615 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
2616 | cgroup = memcg->css.cgroup; |
2617 | if (!memcg->use_hierarchy) |
2618 | goto out; |
2619 | |
2620 | while (cgroup->parent) { |
2621 | cgroup = cgroup->parent; |
2622 | memcg = mem_cgroup_from_cont(cgroup); |
2623 | if (!memcg->use_hierarchy) |
2624 | break; |
2625 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); |
2626 | min_limit = min(min_limit, tmp); |
2627 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
2628 | min_memsw_limit = min(min_memsw_limit, tmp); |
2629 | } |
2630 | out: |
2631 | *mem_limit = min_limit; |
2632 | *memsw_limit = min_memsw_limit; |
2633 | return; |
2634 | } |
2635 | |
2636 | static int mem_cgroup_reset(struct cgroup *cont, unsigned int event) |
2637 | { |
2638 | struct mem_cgroup *mem; |
2639 | int type, name; |
2640 | |
2641 | mem = mem_cgroup_from_cont(cont); |
2642 | type = MEMFILE_TYPE(event); |
2643 | name = MEMFILE_ATTR(event); |
2644 | switch (name) { |
2645 | case RES_MAX_USAGE: |
2646 | if (type == _MEM) |
2647 | res_counter_reset_max(&mem->res); |
2648 | else |
2649 | res_counter_reset_max(&mem->memsw); |
2650 | break; |
2651 | case RES_FAILCNT: |
2652 | if (type == _MEM) |
2653 | res_counter_reset_failcnt(&mem->res); |
2654 | else |
2655 | res_counter_reset_failcnt(&mem->memsw); |
2656 | break; |
2657 | } |
2658 | |
2659 | return 0; |
2660 | } |
2661 | |
2662 | |
2663 | /* For read statistics */ |
2664 | enum { |
2665 | MCS_CACHE, |
2666 | MCS_RSS, |
2667 | MCS_MAPPED_FILE, |
2668 | MCS_PGPGIN, |
2669 | MCS_PGPGOUT, |
2670 | MCS_SWAP, |
2671 | MCS_INACTIVE_ANON, |
2672 | MCS_ACTIVE_ANON, |
2673 | MCS_INACTIVE_FILE, |
2674 | MCS_ACTIVE_FILE, |
2675 | MCS_UNEVICTABLE, |
2676 | NR_MCS_STAT, |
2677 | }; |
2678 | |
2679 | struct mcs_total_stat { |
2680 | s64 stat[NR_MCS_STAT]; |
2681 | }; |
2682 | |
2683 | struct { |
2684 | char *local_name; |
2685 | char *total_name; |
2686 | } memcg_stat_strings[NR_MCS_STAT] = { |
2687 | {"cache", "total_cache"}, |
2688 | {"rss", "total_rss"}, |
2689 | {"mapped_file", "total_mapped_file"}, |
2690 | {"pgpgin", "total_pgpgin"}, |
2691 | {"pgpgout", "total_pgpgout"}, |
2692 | {"swap", "total_swap"}, |
2693 | {"inactive_anon", "total_inactive_anon"}, |
2694 | {"active_anon", "total_active_anon"}, |
2695 | {"inactive_file", "total_inactive_file"}, |
2696 | {"active_file", "total_active_file"}, |
2697 | {"unevictable", "total_unevictable"} |
2698 | }; |
2699 | |
2700 | |
2701 | static int mem_cgroup_get_local_stat(struct mem_cgroup *mem, void *data) |
2702 | { |
2703 | struct mcs_total_stat *s = data; |
2704 | s64 val; |
2705 | |
2706 | /* per cpu stat */ |
2707 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_CACHE); |
2708 | s->stat[MCS_CACHE] += val * PAGE_SIZE; |
2709 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_RSS); |
2710 | s->stat[MCS_RSS] += val * PAGE_SIZE; |
2711 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_MAPPED_FILE); |
2712 | s->stat[MCS_MAPPED_FILE] += val * PAGE_SIZE; |
2713 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGIN_COUNT); |
2714 | s->stat[MCS_PGPGIN] += val; |
2715 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_PGPGOUT_COUNT); |
2716 | s->stat[MCS_PGPGOUT] += val; |
2717 | if (do_swap_account) { |
2718 | val = mem_cgroup_read_stat(&mem->stat, MEM_CGROUP_STAT_SWAPOUT); |
2719 | s->stat[MCS_SWAP] += val * PAGE_SIZE; |
2720 | } |
2721 | |
2722 | /* per zone stat */ |
2723 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_ANON); |
2724 | s->stat[MCS_INACTIVE_ANON] += val * PAGE_SIZE; |
2725 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_ANON); |
2726 | s->stat[MCS_ACTIVE_ANON] += val * PAGE_SIZE; |
2727 | val = mem_cgroup_get_local_zonestat(mem, LRU_INACTIVE_FILE); |
2728 | s->stat[MCS_INACTIVE_FILE] += val * PAGE_SIZE; |
2729 | val = mem_cgroup_get_local_zonestat(mem, LRU_ACTIVE_FILE); |
2730 | s->stat[MCS_ACTIVE_FILE] += val * PAGE_SIZE; |
2731 | val = mem_cgroup_get_local_zonestat(mem, LRU_UNEVICTABLE); |
2732 | s->stat[MCS_UNEVICTABLE] += val * PAGE_SIZE; |
2733 | return 0; |
2734 | } |
2735 | |
2736 | static void |
2737 | mem_cgroup_get_total_stat(struct mem_cgroup *mem, struct mcs_total_stat *s) |
2738 | { |
2739 | mem_cgroup_walk_tree(mem, s, mem_cgroup_get_local_stat); |
2740 | } |
2741 | |
2742 | static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft, |
2743 | struct cgroup_map_cb *cb) |
2744 | { |
2745 | struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont); |
2746 | struct mcs_total_stat mystat; |
2747 | int i; |
2748 | |
2749 | memset(&mystat, 0, sizeof(mystat)); |
2750 | mem_cgroup_get_local_stat(mem_cont, &mystat); |
2751 | |
2752 | for (i = 0; i < NR_MCS_STAT; i++) { |
2753 | if (i == MCS_SWAP && !do_swap_account) |
2754 | continue; |
2755 | cb->fill(cb, memcg_stat_strings[i].local_name, mystat.stat[i]); |
2756 | } |
2757 | |
2758 | /* Hierarchical information */ |
2759 | { |
2760 | unsigned long long limit, memsw_limit; |
2761 | memcg_get_hierarchical_limit(mem_cont, &limit, &memsw_limit); |
2762 | cb->fill(cb, "hierarchical_memory_limit", limit); |
2763 | if (do_swap_account) |
2764 | cb->fill(cb, "hierarchical_memsw_limit", memsw_limit); |
2765 | } |
2766 | |
2767 | memset(&mystat, 0, sizeof(mystat)); |
2768 | mem_cgroup_get_total_stat(mem_cont, &mystat); |
2769 | for (i = 0; i < NR_MCS_STAT; i++) { |
2770 | if (i == MCS_SWAP && !do_swap_account) |
2771 | continue; |
2772 | cb->fill(cb, memcg_stat_strings[i].total_name, mystat.stat[i]); |
2773 | } |
2774 | |
2775 | #ifdef CONFIG_DEBUG_VM |
2776 | cb->fill(cb, "inactive_ratio", calc_inactive_ratio(mem_cont, NULL)); |
2777 | |
2778 | { |
2779 | int nid, zid; |
2780 | struct mem_cgroup_per_zone *mz; |
2781 | unsigned long recent_rotated[2] = {0, 0}; |
2782 | unsigned long recent_scanned[2] = {0, 0}; |
2783 | |
2784 | for_each_online_node(nid) |
2785 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
2786 | mz = mem_cgroup_zoneinfo(mem_cont, nid, zid); |
2787 | |
2788 | recent_rotated[0] += |
2789 | mz->reclaim_stat.recent_rotated[0]; |
2790 | recent_rotated[1] += |
2791 | mz->reclaim_stat.recent_rotated[1]; |
2792 | recent_scanned[0] += |
2793 | mz->reclaim_stat.recent_scanned[0]; |
2794 | recent_scanned[1] += |
2795 | mz->reclaim_stat.recent_scanned[1]; |
2796 | } |
2797 | cb->fill(cb, "recent_rotated_anon", recent_rotated[0]); |
2798 | cb->fill(cb, "recent_rotated_file", recent_rotated[1]); |
2799 | cb->fill(cb, "recent_scanned_anon", recent_scanned[0]); |
2800 | cb->fill(cb, "recent_scanned_file", recent_scanned[1]); |
2801 | } |
2802 | #endif |
2803 | |
2804 | return 0; |
2805 | } |
2806 | |
2807 | static u64 mem_cgroup_swappiness_read(struct cgroup *cgrp, struct cftype *cft) |
2808 | { |
2809 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
2810 | |
2811 | return get_swappiness(memcg); |
2812 | } |
2813 | |
2814 | static int mem_cgroup_swappiness_write(struct cgroup *cgrp, struct cftype *cft, |
2815 | u64 val) |
2816 | { |
2817 | struct mem_cgroup *memcg = mem_cgroup_from_cont(cgrp); |
2818 | struct mem_cgroup *parent; |
2819 | |
2820 | if (val > 100) |
2821 | return -EINVAL; |
2822 | |
2823 | if (cgrp->parent == NULL) |
2824 | return -EINVAL; |
2825 | |
2826 | parent = mem_cgroup_from_cont(cgrp->parent); |
2827 | |
2828 | cgroup_lock(); |
2829 | |
2830 | /* If under hierarchy, only empty-root can set this value */ |
2831 | if ((parent->use_hierarchy) || |
2832 | (memcg->use_hierarchy && !list_empty(&cgrp->children))) { |
2833 | cgroup_unlock(); |
2834 | return -EINVAL; |
2835 | } |
2836 | |
2837 | spin_lock(&memcg->reclaim_param_lock); |
2838 | memcg->swappiness = val; |
2839 | spin_unlock(&memcg->reclaim_param_lock); |
2840 | |
2841 | cgroup_unlock(); |
2842 | |
2843 | return 0; |
2844 | } |
2845 | |
2846 | |
2847 | static struct cftype mem_cgroup_files[] = { |
2848 | { |
2849 | .name = "usage_in_bytes", |
2850 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
2851 | .read_u64 = mem_cgroup_read, |
2852 | }, |
2853 | { |
2854 | .name = "max_usage_in_bytes", |
2855 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
2856 | .trigger = mem_cgroup_reset, |
2857 | .read_u64 = mem_cgroup_read, |
2858 | }, |
2859 | { |
2860 | .name = "limit_in_bytes", |
2861 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
2862 | .write_string = mem_cgroup_write, |
2863 | .read_u64 = mem_cgroup_read, |
2864 | }, |
2865 | { |
2866 | .name = "soft_limit_in_bytes", |
2867 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
2868 | .write_string = mem_cgroup_write, |
2869 | .read_u64 = mem_cgroup_read, |
2870 | }, |
2871 | { |
2872 | .name = "failcnt", |
2873 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
2874 | .trigger = mem_cgroup_reset, |
2875 | .read_u64 = mem_cgroup_read, |
2876 | }, |
2877 | { |
2878 | .name = "stat", |
2879 | .read_map = mem_control_stat_show, |
2880 | }, |
2881 | { |
2882 | .name = "force_empty", |
2883 | .trigger = mem_cgroup_force_empty_write, |
2884 | }, |
2885 | { |
2886 | .name = "use_hierarchy", |
2887 | .write_u64 = mem_cgroup_hierarchy_write, |
2888 | .read_u64 = mem_cgroup_hierarchy_read, |
2889 | }, |
2890 | { |
2891 | .name = "swappiness", |
2892 | .read_u64 = mem_cgroup_swappiness_read, |
2893 | .write_u64 = mem_cgroup_swappiness_write, |
2894 | }, |
2895 | }; |
2896 | |
2897 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
2898 | static struct cftype memsw_cgroup_files[] = { |
2899 | { |
2900 | .name = "memsw.usage_in_bytes", |
2901 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
2902 | .read_u64 = mem_cgroup_read, |
2903 | }, |
2904 | { |
2905 | .name = "memsw.max_usage_in_bytes", |
2906 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
2907 | .trigger = mem_cgroup_reset, |
2908 | .read_u64 = mem_cgroup_read, |
2909 | }, |
2910 | { |
2911 | .name = "memsw.limit_in_bytes", |
2912 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
2913 | .write_string = mem_cgroup_write, |
2914 | .read_u64 = mem_cgroup_read, |
2915 | }, |
2916 | { |
2917 | .name = "memsw.failcnt", |
2918 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
2919 | .trigger = mem_cgroup_reset, |
2920 | .read_u64 = mem_cgroup_read, |
2921 | }, |
2922 | }; |
2923 | |
2924 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) |
2925 | { |
2926 | if (!do_swap_account) |
2927 | return 0; |
2928 | return cgroup_add_files(cont, ss, memsw_cgroup_files, |
2929 | ARRAY_SIZE(memsw_cgroup_files)); |
2930 | }; |
2931 | #else |
2932 | static int register_memsw_files(struct cgroup *cont, struct cgroup_subsys *ss) |
2933 | { |
2934 | return 0; |
2935 | } |
2936 | #endif |
2937 | |
2938 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2939 | { |
2940 | struct mem_cgroup_per_node *pn; |
2941 | struct mem_cgroup_per_zone *mz; |
2942 | enum lru_list l; |
2943 | int zone, tmp = node; |
2944 | /* |
2945 | * This routine is called against possible nodes. |
2946 | * But it's BUG to call kmalloc() against offline node. |
2947 | * |
2948 | * TODO: this routine can waste much memory for nodes which will |
2949 | * never be onlined. It's better to use memory hotplug callback |
2950 | * function. |
2951 | */ |
2952 | if (!node_state(node, N_NORMAL_MEMORY)) |
2953 | tmp = -1; |
2954 | pn = kmalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
2955 | if (!pn) |
2956 | return 1; |
2957 | |
2958 | mem->info.nodeinfo[node] = pn; |
2959 | memset(pn, 0, sizeof(*pn)); |
2960 | |
2961 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
2962 | mz = &pn->zoneinfo[zone]; |
2963 | for_each_lru(l) |
2964 | INIT_LIST_HEAD(&mz->lists[l]); |
2965 | mz->usage_in_excess = 0; |
2966 | mz->on_tree = false; |
2967 | mz->mem = mem; |
2968 | } |
2969 | return 0; |
2970 | } |
2971 | |
2972 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *mem, int node) |
2973 | { |
2974 | kfree(mem->info.nodeinfo[node]); |
2975 | } |
2976 | |
2977 | static int mem_cgroup_size(void) |
2978 | { |
2979 | int cpustat_size = nr_cpu_ids * sizeof(struct mem_cgroup_stat_cpu); |
2980 | return sizeof(struct mem_cgroup) + cpustat_size; |
2981 | } |
2982 | |
2983 | static struct mem_cgroup *mem_cgroup_alloc(void) |
2984 | { |
2985 | struct mem_cgroup *mem; |
2986 | int size = mem_cgroup_size(); |
2987 | |
2988 | if (size < PAGE_SIZE) |
2989 | mem = kmalloc(size, GFP_KERNEL); |
2990 | else |
2991 | mem = vmalloc(size); |
2992 | |
2993 | if (mem) |
2994 | memset(mem, 0, size); |
2995 | return mem; |
2996 | } |
2997 | |
2998 | /* |
2999 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
3000 | * (scanning all at force_empty is too costly...) |
3001 | * |
3002 | * Instead of clearing all references at force_empty, we remember |
3003 | * the number of reference from swap_cgroup and free mem_cgroup when |
3004 | * it goes down to 0. |
3005 | * |
3006 | * Removal of cgroup itself succeeds regardless of refs from swap. |
3007 | */ |
3008 | |
3009 | static void __mem_cgroup_free(struct mem_cgroup *mem) |
3010 | { |
3011 | int node; |
3012 | |
3013 | mem_cgroup_remove_from_trees(mem); |
3014 | free_css_id(&mem_cgroup_subsys, &mem->css); |
3015 | |
3016 | for_each_node_state(node, N_POSSIBLE) |
3017 | free_mem_cgroup_per_zone_info(mem, node); |
3018 | |
3019 | if (mem_cgroup_size() < PAGE_SIZE) |
3020 | kfree(mem); |
3021 | else |
3022 | vfree(mem); |
3023 | } |
3024 | |
3025 | static void mem_cgroup_get(struct mem_cgroup *mem) |
3026 | { |
3027 | atomic_inc(&mem->refcnt); |
3028 | } |
3029 | |
3030 | static void mem_cgroup_put(struct mem_cgroup *mem) |
3031 | { |
3032 | if (atomic_dec_and_test(&mem->refcnt)) { |
3033 | struct mem_cgroup *parent = parent_mem_cgroup(mem); |
3034 | __mem_cgroup_free(mem); |
3035 | if (parent) |
3036 | mem_cgroup_put(parent); |
3037 | } |
3038 | } |
3039 | |
3040 | /* |
3041 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. |
3042 | */ |
3043 | static struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *mem) |
3044 | { |
3045 | if (!mem->res.parent) |
3046 | return NULL; |
3047 | return mem_cgroup_from_res_counter(mem->res.parent, res); |
3048 | } |
3049 | |
3050 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3051 | static void __init enable_swap_cgroup(void) |
3052 | { |
3053 | if (!mem_cgroup_disabled() && really_do_swap_account) |
3054 | do_swap_account = 1; |
3055 | } |
3056 | #else |
3057 | static void __init enable_swap_cgroup(void) |
3058 | { |
3059 | } |
3060 | #endif |
3061 | |
3062 | static int mem_cgroup_soft_limit_tree_init(void) |
3063 | { |
3064 | struct mem_cgroup_tree_per_node *rtpn; |
3065 | struct mem_cgroup_tree_per_zone *rtpz; |
3066 | int tmp, node, zone; |
3067 | |
3068 | for_each_node_state(node, N_POSSIBLE) { |
3069 | tmp = node; |
3070 | if (!node_state(node, N_NORMAL_MEMORY)) |
3071 | tmp = -1; |
3072 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); |
3073 | if (!rtpn) |
3074 | return 1; |
3075 | |
3076 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
3077 | |
3078 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
3079 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
3080 | rtpz->rb_root = RB_ROOT; |
3081 | spin_lock_init(&rtpz->lock); |
3082 | } |
3083 | } |
3084 | return 0; |
3085 | } |
3086 | |
3087 | static struct cgroup_subsys_state * __ref |
3088 | mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont) |
3089 | { |
3090 | struct mem_cgroup *mem, *parent; |
3091 | long error = -ENOMEM; |
3092 | int node; |
3093 | |
3094 | mem = mem_cgroup_alloc(); |
3095 | if (!mem) |
3096 | return ERR_PTR(error); |
3097 | |
3098 | for_each_node_state(node, N_POSSIBLE) |
3099 | if (alloc_mem_cgroup_per_zone_info(mem, node)) |
3100 | goto free_out; |
3101 | |
3102 | /* root ? */ |
3103 | if (cont->parent == NULL) { |
3104 | enable_swap_cgroup(); |
3105 | parent = NULL; |
3106 | root_mem_cgroup = mem; |
3107 | if (mem_cgroup_soft_limit_tree_init()) |
3108 | goto free_out; |
3109 | |
3110 | } else { |
3111 | parent = mem_cgroup_from_cont(cont->parent); |
3112 | mem->use_hierarchy = parent->use_hierarchy; |
3113 | } |
3114 | |
3115 | if (parent && parent->use_hierarchy) { |
3116 | res_counter_init(&mem->res, &parent->res); |
3117 | res_counter_init(&mem->memsw, &parent->memsw); |
3118 | /* |
3119 | * We increment refcnt of the parent to ensure that we can |
3120 | * safely access it on res_counter_charge/uncharge. |
3121 | * This refcnt will be decremented when freeing this |
3122 | * mem_cgroup(see mem_cgroup_put). |
3123 | */ |
3124 | mem_cgroup_get(parent); |
3125 | } else { |
3126 | res_counter_init(&mem->res, NULL); |
3127 | res_counter_init(&mem->memsw, NULL); |
3128 | } |
3129 | mem->last_scanned_child = 0; |
3130 | spin_lock_init(&mem->reclaim_param_lock); |
3131 | |
3132 | if (parent) |
3133 | mem->swappiness = get_swappiness(parent); |
3134 | atomic_set(&mem->refcnt, 1); |
3135 | return &mem->css; |
3136 | free_out: |
3137 | __mem_cgroup_free(mem); |
3138 | root_mem_cgroup = NULL; |
3139 | return ERR_PTR(error); |
3140 | } |
3141 | |
3142 | static int mem_cgroup_pre_destroy(struct cgroup_subsys *ss, |
3143 | struct cgroup *cont) |
3144 | { |
3145 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
3146 | |
3147 | return mem_cgroup_force_empty(mem, false); |
3148 | } |
3149 | |
3150 | static void mem_cgroup_destroy(struct cgroup_subsys *ss, |
3151 | struct cgroup *cont) |
3152 | { |
3153 | struct mem_cgroup *mem = mem_cgroup_from_cont(cont); |
3154 | |
3155 | mem_cgroup_put(mem); |
3156 | } |
3157 | |
3158 | static int mem_cgroup_populate(struct cgroup_subsys *ss, |
3159 | struct cgroup *cont) |
3160 | { |
3161 | int ret; |
3162 | |
3163 | ret = cgroup_add_files(cont, ss, mem_cgroup_files, |
3164 | ARRAY_SIZE(mem_cgroup_files)); |
3165 | |
3166 | if (!ret) |
3167 | ret = register_memsw_files(cont, ss); |
3168 | return ret; |
3169 | } |
3170 | |
3171 | static void mem_cgroup_move_task(struct cgroup_subsys *ss, |
3172 | struct cgroup *cont, |
3173 | struct cgroup *old_cont, |
3174 | struct task_struct *p, |
3175 | bool threadgroup) |
3176 | { |
3177 | mutex_lock(&memcg_tasklist); |
3178 | /* |
3179 | * FIXME: It's better to move charges of this process from old |
3180 | * memcg to new memcg. But it's just on TODO-List now. |
3181 | */ |
3182 | mutex_unlock(&memcg_tasklist); |
3183 | } |
3184 | |
3185 | struct cgroup_subsys mem_cgroup_subsys = { |
3186 | .name = "memory", |
3187 | .subsys_id = mem_cgroup_subsys_id, |
3188 | .create = mem_cgroup_create, |
3189 | .pre_destroy = mem_cgroup_pre_destroy, |
3190 | .destroy = mem_cgroup_destroy, |
3191 | .populate = mem_cgroup_populate, |
3192 | .attach = mem_cgroup_move_task, |
3193 | .early_init = 0, |
3194 | .use_id = 1, |
3195 | }; |
3196 | |
3197 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP |
3198 | |
3199 | static int __init disable_swap_account(char *s) |
3200 | { |
3201 | really_do_swap_account = 0; |
3202 | return 1; |
3203 | } |
3204 | __setup("noswapaccount", disable_swap_account); |
3205 | #endif |
3206 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9